the impact of land use change on greenhouse gas … · 5 executive summary the modelling of the...

232
1 THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS EMISSIONS FROM BIOFUELS AND BIOLIQUIDS Literature review July 2010

Upload: others

Post on 08-May-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

1

THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS EMISSIONS FROM

BIOFUELS AND BIOLIQUIDS

Literature review

July 2010

Page 2: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

2

An in-house review conducted for DG Energy as part of the European Commission's analytical work on indirect land use change

Page 3: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

3

Contents

Executive summary ............................................................................................................... 5 Summary of chapters ............................................................................................................. 9

Purpose of the literature review (chapter 1) ........................................................................ 9 Outline of the paper (chapter 2)........................................................................................ 10 Data (chapter 3) ............................................................................................................... 10 Baseline scenarios (chapter 4.2) ....................................................................................... 11 Baseline assumptions (chapter 4.3 - 4.9)........................................................................... 11 Policy scenarios (chapter 5).............................................................................................. 13 Models (chapter 6) ........................................................................................................... 13 Co-products (chapter 7).................................................................................................... 14 Yields and demand (chapter 8.1 to 8.4) ............................................................................ 15 Impact of public policy (chapter 8.5)................................................................................ 18 Determining the type of land that is converted (chapter 9)................................................ 19 Evaluating carbon stock changes (chapter 10) .................................................................. 21 Results for the land use change impact of biofuels (chapter 11)........................................ 24 Comparing land use effects with GHG savings from biofuels (chapter 12)....................... 26

1. Purpose of the literature review.................................................................................... 28

1.1. The task................................................................................................................ 28 1.2. Main policy questions........................................................................................... 28 1.3. "Land use change" vs. "indirect land use change" ................................................. 28

2. Outline of the paper...................................................................................................... 29 2.1. Why model? ......................................................................................................... 29 2.2. Analytical options................................................................................................. 29 2.3. Steps in the (scenario analysis) modelling process ................................................ 29

3. Data ............................................................................................................................. 30 3.1. The role of data in the modelling process.............................................................. 30 3.2. Land use and crop production data........................................................................ 31 3.3. Trade data............................................................................................................. 36 3.4. Biofuel data.......................................................................................................... 37

4. Baseline ....................................................................................................................... 39 4.1. Introduction.......................................................................................................... 39 4.2. Biofuels in the baseline scenario........................................................................... 39 4.3. Baseline assumptions: crop yields......................................................................... 43 4.4. Baseline assumptions: food/crop demand.............................................................. 47 4.5. Baseline assumptions: oil price............................................................................. 47 4.6. Baseline assumptions: petrol and diesel consumption ........................................... 48 4.7. Baseline assumptions: land use............................................................................. 48 4.8. Baseline assumptions: agriculture policy .............................................................. 50 4.9. Baseline assumptions: tariff policy ....................................................................... 51 4.10. Conclusion ....................................................................................................... 52

5. Policy scenarios ........................................................................................................... 52 5.1. Introduction.......................................................................................................... 52 5.2. Anticipated biofuel/bioliquid consumption under the EU policies whose impact is to be assessed....................................................................................................................... 53 5.3. Policy scenarios in the modelling work reported on.............................................. 61

6. Models......................................................................................................................... 64

Page 4: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

4

6.1. Choices to be made............................................................................................... 64 6.2. General and partial equilibrium models ................................................................ 65 6.3. Allocation models................................................................................................. 69

7. Model calculations: crop output ................................................................................... 70 7.1. Introduction.......................................................................................................... 70 7.2. Co-products.......................................................................................................... 70

8. Model calculations: from tons to hectares..................................................................... 93 8.1. Yields: introduction .............................................................................................. 93 8.2. Factors affecting yields: analysis and empirical work ........................................... 93 8.3. Yield response to demand in modelling exercises ............................................... 105 8.4. Commentary....................................................................................................... 115 8.5. Impact of public policy....................................................................................... 123

9. Model calculations: from hectares to land types ......................................................... 129 9.1. Introduction........................................................................................................ 129 9.2. The historical approach....................................................................................... 129 9.3. The suitability approach ..................................................................................... 135 9.4. Other approaches................................................................................................ 140 9.5. Summary for both historical and suitability approaches ...................................... 141 9.6. Commentary on land types considered................................................................ 144 9.7. Commentary on other issues............................................................................... 151 9.8. Drivers of deforestation ...................................................................................... 157

10. Model calculations: carbon stock changes .............................................................. 169 10.1. Data and assumptions: general........................................................................ 169 10.2. Data and assumptions: carbon stocks in cropland............................................ 172 10.3. Data and assumptions: conversion of wetland/peatlands ................................. 173 10.4. Empirical evidence on foregone carbon sequestration..................................... 174 10.5. Assumptions on foregone carbon sequestration............................................... 175 10.6. Results............................................................................................................ 176 10.7. Commentary: foregone sequestration.............................................................. 184 10.8. Commentary: carbon stock ............................................................................. 185

11. Results for the land use change impact of biofuels.................................................. 188 11.1. Estimated impact of biofuel promotion ........................................................... 188 11.2. Variation of impacts among biofuels............................................................... 190

12. Model calculations: comparing land use effects with GHG savings from biofuels .. 198 12.1. GHG emissions from biofuel production......................................................... 198 12.2. The fossil fuel comparator .............................................................................. 205 12.3. Comparing flows with stock changes.............................................................. 210

Acronyms and abbreviations.............................................................................................. 220 References ......................................................................................................................... 221

Page 5: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

5

Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared only in 2007. A great deal of scientific progress has been made since then. However, it became apparent in the course of this literature review that consensus is far from being reached among scientists on many key aspects of methodology and data; there are still aspects that no studies have addressed; and these issues have a significant impact on the studies’ results. The main issues identified are summarised below: 1. Land use data play an essential part in land use change modelling. The available

datasets give markedly different results – for example, estimates of global cultivated land in 2000 ranged from 1.2 billion to 2.0 billion hectares. There is no consensus about which data-set is best for work of the type covered here.

2. None of the identified modelling exercises explores the option of biofuels not being

used at all. Thus, none evaluates the land use change impact of biofuels per se, as opposed to that of particular biofuel-promoting policies.

3. The modelling exercises either do not report clearly the way in which the assumed quantity of biofuels from crops in the policy scenario is calculated, or else – at least in the case of the EU – calculate it in a way that gives too high a result.

4. If crop yields increase, less land will be needed. Most of the modelling reviewed assumes a yield increase in the baseline. Its size is rarely clear. High assumptions could reduce the amount of land converted by 15% compared with low assumptions. There is reason to believe that in the underlying data, increases in cropping intensity (such as multiple crops per year) are mis-classified as increases in land use. If so, studies that rely on historic figures for their yield assumptions will tend to use a lower value than they should.

5. If crop yields grow faster in response to the extra demand created by biofuels, the

land needed will be even less. It makes sense from a theoretical point of view to believe this will happen (because price increases will make investments viable). However, the relationship has proved difficult to quantify empirically. Early modelling exercises assumed that the effect was zero or negligible. More recent work allows for yield to respond to demand through changes in quantities of inputs, albeit in ways that are rather inexplicit and therefore difficult to assess. Only one study has been identified that also allows yield to respond through changes in cropping intensity, and no study allows yield to respond through faster technological development, even though it is common for such a relationship to be taken into account in other modelling. Sensitivity exercises showed that different assumptions about the response of yields to demand have big impacts on the results, with higher-response assumptions leading to reductions of 27-80% in land conversion or carbon stock loss as compared to the results with studies' central assumptions.

6. In the EU and other regions, the cropped area has been shrinking for some time and

is still doing so. Extra demand for biofuels will slow this down. Some modellers treat this as a missed opportunity to allow the land to revert to grassland or forest, absorbing carbon from the atmosphere as it does so. This missed opportunity is a greenhouse gas

Page 6: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

6

“cost” of the policy. This seems a reasonable approach. However, some of the land will not “revert” in this way (for example, because it is land that rotates regularly into and out of crop production). This is not taken into account in the modelling.

7. It is not clear in any of the modelling exercises how the biodiesel or ethanol

feedstock used is determined. 8. The production of most biofuel crops necessarily entails the production of co-

products, many of which – used as animal feed – replace crops that would otherwise need to be grown. When this is taken into account, the estimated land use change impact of biofuel promotion is reduced. Studies suggest that this reduction is by between 8 and 64% (median 36%) for the policy as a whole, and by 35-94% for individual crops such as maize, sugar beet, wheat and rapeseed. There is significant divergence between studies concerning the rate at which co-products are assumed to substitute for other types of animal feed and for the types of animal feed they are assumed to replace. (The type matters because the production of soya meal is more land intensive than the production of cereals, and an increase in soya production is more likely to trigger conversion of forest land than is an increase in cereal production.) For maize, for example, estimates of the quantity of animal feed substituted per ton of co-product vary from 0.89 to 1.25 tons. Estimates of the share of soya in the substituted animal feed vary from 0 to 50%. The ratios chosen will have a significant impact on the result of modelling exercises.

9. Some studies assume that converted land will have a lower yield than already-

cultivated land. Depending on the study, the removal of this assumption would reduce the amount of land needing to be converted by approximately 17% to 67%. The literature contained no empirical data in support of this assumption, and there is good reason for thinking that even the smallest of the estimated effects is too large.

10. The EU legislative framework includes measures that require biofuels used for EU

targets to achieve a minimum greenhouse gas saving and not use raw material from certain types of land. None of the modelling reviewed here takes account of these limits. In addition, public authorities impose more general limits on the use of land, e.g. for nature protection purposes. The modelling reviewed here takes account of these limits as they exist today, but not of the fact that they can be expected to expand in future.

11. Once modellers have estimated how much land needs to be converted to meet the extra

demand for biofuels, they need to determine which type of land is converted. This matters because different types of land have different carbon stocks. Some studies use a “historical” approach, under which the proportions of different land types in land converted in the future are assumed to be the same as they were during some period in the past. These studies give sharply different results at the level of regions. For example, three studies put the historical share of forest/woodland in land converted to cropland in the US at 0%, 7% and 38% respectively. Other studies use a “suitability” approach to determine which land is converted. The sets of suitability criteria that they use vary substantially. It has not been possible to assess how these differences affect the studies’ results.

Page 7: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

7

12. Most models do not allow for the possibility of the converted land being peatland or wetland. Because these types of land have a very high carbon stock, this leads to underestimation of the carbon stock loss caused by crop expansion.

13. Crops may move onto land that has been deforested for another reason; or an act of

deforestation may be driven simultaneously by crop expansion and logging. (That is, the deforestation would not have happened if both drivers had not been present.) All the studies (with one possible exception) treat this as if the crop expansion were the sole cause of the deforestation. This leads to overestimation of the carbon stock loss caused by crop expansion.

14. Modelling of the input that pasture provides to livestock production needs to be more

sophisticated. It is not obvious how this would affect the results. 15. There are numerous differences in how the studies calculate changes in carbon

stocks. For example, the carbon stocks attributed to particular land types vary by factors of between 2 and 15 from one study to another. This must have a significant impact on the results.

16. The modelling exercises compare the land use impact of demand for biofuels with the

emissions avoided by biofuels replacing fossil fuel. For this comparison they use an average value for present-day emissions from conventional sources of crude oil. This is not correct. The comparison should be with the long-term marginal source of fossil fuel (that is, the marginal source in a perspective under which investment decisions play a role). In the literature on that topic there is consensus that this will be a non-conventional source with higher emissions. Using such a comparator would be equivalent to reducing the estimated greenhouse gas impact from land use change by about 30%.

17. None of the studies takes into account the likelihood that the greenhouse gas

performance of biofuels will improve between now and 2020. 18. Finally, the reports assessed in this review are generally not explicit enough for the

purposes of this review as regards data and methodology. It has not been possible to make a full comparison of the modelling choices made because in many cases, the reports do not reveal what these choices are.

19. In terms of results, the estimated impact of the land use change attributed to biofuels has fallen over time, presumably as study methods have become more refined. While the original work of Searchinger et al. suggested that the greenhouse gas impact of biofuels' land use change was twice as great as that of the fossil fuel consumption avoided, three of the four most recent studies estimating greenhouse gas impacts – including the only one dealing with the EU – have concluded that biofuels are beneficial in greenhouse gas terms even when their land use impact, as well as a full life cycle analysis, is taken into account.

20. Studies that look at the relative impact of different types of biofuel give widely varying results. Most commonly, they suggest that one or another type of biodiesel – most frequently, soya – performs worse than ethanol. Only one study looks at whether the same feedstock has a different land use change impact if produced in a different

Page 8: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

8

location. Since this study concludes that the effect of location is large, it is a pity that this issue has not been explored further.

Page 9: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

9

Summary of chapters

Purpose of the literature review (chapter 1) The Renewable Energy Directive (as far as biofuels and bioliquids are concerned) and the Fuel Quality Directive (as far as biofuels are concerned) require the Commission to submit to Parliament and Council a report with two components: (a) "reviewing the impact of indirect land-use change on greenhouse gas emissions"; (b) "addressing ways to minimise that impact". This paper is part of the work of the Commission services in preparing the first of these components. It draws on more than 150 contributions related to the topic. It is hereby made available for comments from interested parties. The literature review concentrates on comparing studies’ methodological and data choices rather than their results. The land use change modelling exercises reviewed are the following: - AGLINK-COSIMO (for the European Commission) - AGLINK/OECD - BLUM - CAPRI (for the European Commission) - CARB - Dumortier et al. - EPA1 - ESIM (for the European Commission) - GLOBIOM - Hertel et al. - IFPRI (for the European Commission) - IIASA (for OFID) - Keeney and Hertel - Kim et al. - Kløverpris et al. - LEITAP/ALTERRA (for the European Commission) - LEITAP/Banse et al. - Lywood/Ensus - O'Hare et al. - Searchinger et al. - Taheripour et al. - Tyner et al.

1 Note: references to the EPA study do not take fully into account the changes in the final version released in January 2010.

Page 10: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

10

Outline of the paper (chapter 2) It will never be possible to physically observe indirect land use change. It will never be possible, looking forward, to say that the introduction of a biofuel policy will lead to the conversion of a particular, identified piece of land. It will never be possible, looking back, to say that the introduction of a biofuel policy was the cause of a particular, identified piece of land being converted. It follows that the assessment of the impact of land use change requires the use of modelling. Different modelling approaches can be used. This paper focusses on "scenario analysis", which asks what the land use change impact of biofuels is "likely" to be under different scenarios. Drawing on a foundation of data, scenario analyses compare a baseline (including a baseline scenario with limited biofuel promotion/use, and a set of baseline assumptions) and a policy scenario (with promotion of biofuels). They use a model to do this. The model is used to calculate how the scenarios differ in terms of: - quantity of crops produced; - quantity of land converted for crop production; - types of land converted; - carbon stock changes from land conversion; - overall greenhouse gas impact. Each of the underlined terms is the subject of a chapter in the paper.

Data (chapter 3) Land use data play an essential part in land use change modelling. There are a variety of data-sets of global land use, derived from satellites or agricultural inventories. These data-sets give markedly different results – for example, five estimates of cultivated land in 2000 ranged from 1.2 to 2.0 bn ha (Ramos et al., 2009). While there are doubts about the reliability of different data-sets (ADAS UK Limited, 2008; Lywood, 2009; Ramos et al., 2009), there appears to be no consensus about which is best to use for work of the type covered here. FAO's ResourceSTAT data, and the Agro-Ecological Zone data derived from them, are the only data-sets of global crop production so far identified. Again, their reliability has been questioned (ADAS UK Limited, 2008). Trade data and elasticities also play an essential part in land use change modelling, including for assessing whether the impact of biofuels is heterogeneous between countries. According to the founder of the GTAP project, which is the data source for all the CGE modelling reported here, “[the] economics profession and policy makers under-invest in estimation of … parameters” (Hertel, n.d.). The EU data for biofuel consumption in the base year in AGLINK-COSIMO appear low compared with data from other sources (Eur'Observer; Smeets, 2009). This suggests that the model is not set up correctly.

Page 11: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

11

Baseline scenarios (chapter 4.2) In IFPRI the baseline scenario includes policies/measures to promote biofuels in the jurisdiction of interest – but at a lower level than in the policy scenario. In GLOBIOM and IIASA the baseline scenario includes a fixed level of biofuel use – at a lower level than in the policy scenario. In AGLINK-COSIMO, AGLINK/OECD, ESIM and apparently also in LEITAP/ALTERRA and LEITAP/Banse et al. the baseline scenario is defined by the absence of policies or measures to promote biofuels in the jurisdiction of interest. “Commercially viable” biofuel use remains. None of the identified modelling exercises explores the option of biofuels not being used at all. Thus, none can be said to evaluate the land use change impact of “biofuels” per se. As would be expected, predicted biofuel consumption seems to be higher in baseline scenarios with biofuel promotion than in those without. Thus, EU consumption is predicted at 3-4% in IFPRI’s baseline scenario compared with 1.6% in AGLINK-COSIMO. The results of modelling exercises are typically expressed as the difference between the land use change impacts of the policy and baseline scenarios divided by the difference between the volumes of biofuel consumption they depict. In models that depict the land use change impact as non-linear and rising2, the choice of a baseline scenario that includes a relatively high level of biofuel consumption will then result in a higher estimated unit land use change impact than the choice of a baseline scenario with less biofuel consumption.

Baseline assumptions (chapter 4.3 - 4.9) Modelling exercises generally assume that crop yields will increase in the baseline (that is to say, in both the baseline and the policy scenarios). GLOBIOM appears to be an exception. The modelling exercises identified are not, however, generally transparent as to the values used for yield increases in the baseline. Two estimates were found in the literature on biofuels and land use change of the impact of different baseline yield assumptions on the amount of land that would need to be converted for biofuels: 1-2% (Fernandez-Cornejo et al., 2008, cited in Bouët et al., 2009) and about 10% (RFA, 2008). However, the gap between high- and low-yield scenarios in the literature on future crop yields implies higher figures: 15%, for example, if the values in ADAS UK Limited, 2008 are used. If the land use change impact of demand is non-linear, any given gap between high- and low-yield assumptions would have a higher impact. Crop production can increase in three ways:

2 Henceforth in this summary the term ‘non-linear’ is used as shorthand for ‘non-linear in such a way that the unit land use change impact of biofuels is an increasing function of the volume of biofuel consumption’.

Page 12: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

12

- higher yield per harvest; - more harvests per hectare (increases in cropping intensity: multiple crops in a year, or

a reduction in the frequency of fallow years in a crop rotation); - bringing new land into use for crop production. The focus of this work is on land use change. The third way of increasing crop production constitutes land use change while the first two do not. There is a suggestion, however, that the literature on yields generally expresses land use in terms of “area harvested” (Biomass Research and Development Initiative, n.d.), as also appears to be the case with FAO crop production data. If this is the case, an increase in cropping intensity would incorrectly be depicted as a land use change. This would risk distorting the analysis and exaggerating the amount of land use change. This distortion could be quite significant since, according to Millennium Ecosystem Assessment, 2005, increases in cropping intensity accounted for about a third of the global increase in area harvested between 1961 and 1999. If the land use change impact of biofuels is modelled as non-linear, the volume of biofuel consumption implied by the policy scenario will affect the model’s estimate of unit greenhouse gas emissions attributable to land use change. The EU target that is modelled in policy scenarios is expressed as a proportion of transport petrol, diesel and electricity consumption in 2020. The baseline assumption about transport fuel consumption in 2020 matters, therefore. The latest Commission modelling estimates this figure at 316.3 Mtoe, without taking into account the reductions likely to follow from the recently adopted Cars and CO2 regulation and other EU legislation. Even this figure is therefore to be seen as an overestimate. AGLINK, however, puts EU petrol and diesel consumption in 2020 at 341 Mtoe; in other modelling exercises the figure is not transparent. In the EU and other regions, the arable area is currently shrinking. Modellers who use the concept of “foregone sequestration” (see chapters 9 and 10) need to make an assumption about what will happen to this land in the baseline (so that this can be contrasted with the land’s fate in the policy scenario). Searchinger et al., Dumortier et al. and (seemingly) Tyner et al. assume that it will “revert” to forest or grassland. It is not obvious that such an assumption, implying that all the land would pass into a higher-carbon-stock use, should be made. Another possibility is that at least part of the land can be expected to rotate regularly into and out of crop production, never having the opportunity to accumulate significant additional carbon stocks, and that additional demand (under the policy scenario) affects only the proportion of this rotation time that is spent in rather than out of production. Nor is it obvious that, to the extent that the land would otherwise have passed to a higher carbon stock use, it is correct to assume that the proportion of forest and grassland growth opportunities foregone is equal to their proportion in historic land use change. There is no obvious reason to exclude the possibility that the rate of afforestation is at least partly policy-driven, with cropland-to-grassland conversion as a residual category. If that were true, then the foregone higher carbon stock land would be grassland rather than forest.

Page 13: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

13

Policy scenarios (chapter 5) In the modelling reported here, the way in which the policy scenario(s) differ from the baseline scenario is in respect of the assumed level of biofuel consumption. As far as the EU in 2020 is concerned, the Renewable Energy Directive and Fuel Quality Directive should be used to determine an appropriate assumption. These Directives contain requirements that will lead to the consumption of biofuels from crops. The quantity of biofuels from crops that will be consumed to meet these requirements will depend, in particular, on the contribution made by electric road vehicles and by second-generation biofuels (which, even if they are made from crops, count double towards the targets in the Renewable Energy Directive). With transport petrol, diesel and electricity consumption set at 316.3 Mtoe (see section 4.6), a scenario with no contribution from electric road transport and second-generation biofuels would require 27 Mtoe of biofuels from crops, equivalent to a 8.6% share (the rest coming from a 0.3% share of first-generation biofuels from wastes and residues – counting double – and a 0.8% share of electricity from renewable sources in rail transport). A scenario with an optimistic assumption for electric road transport (with electric vehicles accounting for 20% of car and light van sales by 2020) and a 1.5% share of second-generation biofuels would require 18 Mtoe of biofuels from crops. A scenario with a 3% share of second-generation biofuels would require 13 Mtoe of biofuels from crops. In the first two cases it would be the constraint in the Renewable Energy Directive that was binding; in the third case, the constraint in the Fuel Quality Directive. The modelling exercises reported here either do not report clearly the way in which the assumed quantity of biofuels from crops in the policy scenario is calculated, or else – at least in the case of the EU – calculate it in a way which gives too high a result (e.g. 10% of all transport fuel use). It seems that the types of biofuels used are generally determined exogenously rather than endogenously. Again, little detail is given on this in the model descriptions.

Models (chapter 6) Some writers draw a distinction between economic models on the one hand and agricultural or environmental models on the other (e.g. REFUEL, n.d., Sands et al., n.d., Valin, 2009). In the present context these approaches are not real alternatives, since both aspects need to be covered. Drawing on Lywood, 2009b, a distinction can be drawn between scenario analysis on the one hand and 'allocation methods' on the other. Scenario analysis answers the question, "What is likely to be the land use change impact attributable to a policy of promoting the consumption of biofuels"; allocation models answer the question, what is the land use change impact of the consumption of agricultural commodities (or biofuels) in general. While the work reported here is scenario analysis, it can be argued that the question answered by allocation models is at least as relevant for the task the Commission must perform.

Page 14: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

14

The scenario analysis modelling reported here includes partial equilibrium work (e.g. AGLINK, EPA, GLOBIOM, IIASA and Searchinger et al.) and general equilibrium work (e.g. CARB, IFPRI, Tyner et al. and LEITAP). Advocates of CGE models point to their ability to model "inter-sectoral and international interaction" (Beckman and Hertel, n.d.). In practice, while the treatment of international interaction certainly differs between PE and CGE models (they model trade in different ways), it has not been possible to determine whether the treatment is systematically more satisfactory in the CGE cases. More work is needed to understand the importance or otherwise (in the final results) of the inter-sectoral interactions depicted in CGE models.

Co-products (chapter 7) The first step in the modelling process is to use the model to calculate the difference between the baseline and the policy scenario in terms of the quantities of crops produced. The key factor is obviously the quantity of crops needed to make the extra biofuels that are produced in the policy scenario. But two other factors have to be taken into account. First, the production of most biofuel crops necessarily entails the accompanying production of co-products, many of which – used as animal feed – replace crops that would otherwise have been grown. Second, the increased demand for crops caused by the production of biofuels is likely to lead to an increase in their price, which is likely to lead to a reduction in demand for crops for non-energy purposes. Both these factors mean that the difference in crop output between the policy and baseline scenarios will be less than would otherwise have been expected. This literature review concentrates on the question of co-products. Six studies have been identified that compare the estimated land requirement for biofuels (a) when co-products are not taken into account and (b) when they are. The results are shown in the tables. Tables – Estimated reduction in required quantity of land when co-products are taken into account:

individual crops

maize sugar beet wheat rape seed palm oil CE Delft, 2008 EU 47%

US 61% EU 42%

US 35% EU 38% US 61%

Lywood, 2009 EU 74% EU 65% EU 94% EU 61% SE Asia 7% overall policy Gallagher review, 2008 62-64% GLOBIOM (Havlík, 2009) 8-16%

(result for "net deforestation" Ozdemir et al., 2009 23-37% Tyner et al., 2009 43% Source: calculations of the Commission services from the sources shown. In the first table, locations are both for the production of the feedstock and the consumption of the co-product.

Page 15: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

15

It can be seen that except for one crop (palm oil, which comes with only a small quantity of co-product) and one modelling exercise (GLOBIOM), the studies are consistent in stating that the taking into account of co-products reduces the estimated land requirement by significant quantities - between 23% and 94%. The median crop-specific reduction is 61%; the median overall reduction is 36%. (It makes sense for the overall reduction to be less because as well as crops with high shares of co-products, scenarios are likely to include crops with little or no co-products such as sugar cane, palm oil and grasses or woody crops for second-generation biofuels.) It is clearly important, therefore, for modelling exercises to incorporate a correct treatment of co-product questions. The importance of this is reinforced by the fact that at least in the literature surveyed, there is consensus that the markets' absorptive capacity is not a constraint on the use of biofuel co-products to substitute for other animal feeds (CE Delft, 2008; Lywood et al., 2009; Taheripour et al., 2009). A number of studies (JEC well-to-wheel study; Lywood et al., 2009; Özdemir et al., 2009; CE Delft, 2008; Taheripour et al., 2008) give data for the ratio they assume between biofuel production and co-product production. In general, the numbers used in these studies are reasonably similar; the exceptions are rape seed and soya, to which JEC attributes about 50% more meal production than do Lywood et al. and Özdemir et al.) As the table shows, there is much greater divergence between studies in the rate at which biofuel co-products are assumed to substitute for other types of animal feed, and for the types of animal feed they are assumed to replace. (The type matters because the production of soya meal is more land intensive than the production of cereals, and an increase in soya production is more likely to trigger conversion of forest land in Brazil than is an increase in cereal production.) Table – Amount of cereal (C) and soya meal (S) that each unit of co-product can be expected to replace (mass basis: t/t) Air

Improvement Resource, 2008

CARB CE Delft, 2008

Lywood, 2009

report for EPA, 2007

Kim et al., 2008

Searchinger et al., 2008

Tyner et al., 2009

maize C 0.95 S 0.28

C 1.00 S 0.00

C 0.69 S 0.45

C 0.49 S 0.40

C 0.50 S 0.50

C 0.95 S 0.30

C 1.00 S 0.00

C ? S 0.00

wheat C 0.66 S 0.50

C 0.39 S 0.59

rape seed

C 0.26 S 0.66

C 0.15 S 0.61

It can be seen that for maize, for example, estimates of the mass of animal feed substituted per ton of DDGS vary from 0.89 to 1.25 tons; estimates of the share of soya in the substituted animal feed range from 0 to 50%; and there is no correlation between the two estimates.

Yields and demand (chapter 8.1 to 8.4)

Page 16: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

16

The term “yield” refers to the quantity of crop production per hectare. Yields can be expected to change exogenously in the baseline – this is addressed in chapter 4. If agriculture is a normal economic sector, they can also be expected to change in response to changes in demand/price. This is the subject of sections 8.1 to 8.4. They deal with four possible yield effects:

- changes in average yields caused by changes in the quantity of inputs in response to changes in demand;

- changes in average yields caused by changes in technological development in

response to changes in demand;

- changes in average yields caused by changes in cropping intensity in response to changes in demand;

- changes in average yields caused by lower productivity of land that is converted

in response to changes in demand. The first three yield effects, if real, would lead to a positive correlation between demand and yields. That is, increases in demand for crops – for example caused by demand for biofuels – would lead to an increase in yields. The last effect, if real, could be expected to lead to a negative correlation. The first three are considered together, followed by the last one. Yield effects through inputs, technological development and cropping intensity Increases in demand lead to increases in price. Price increases make investments (in additional inputs, technological progress or increased cropping intensity) viable for economic actors when they would otherwise have been so. Therefore, it makes little sense from the point of view of economic theory to argue that yields are independent of demand. Quantifying this relationship empirically is a different matter. The problem is that both yields and price are strongly correlated with climate. When the weather is good, yields rise and prices fall – and vice versa. Econometricians have not yet found a way to strip out the resulting non-causative negative correlation between yields and demand to reveal the remaining positive, causative relationship. From the literature surveyed here (ADAS UK Ltd, 2008, Biomass Research and Development Initiative, n.d., Keeney and Hertel, 2008, Kløverpris et al. 2008, Kløverpris, 2009, Lywood, 2009 and 2009d, Millennium Ecosystem Assessment, 2005, Sanz Labrador, 2009, Searchinger and Heimlich, 2008, Thirtle et al., 2003, Woods and Murphy, 2009 and the literature review in ECOFYS, 2009), the only reasonable conclusion that can be drawn is that there has not yet been a successful attempt to quantify how demand effects yields through changes in inputs, technological development or cropping intensity. Early modelling work (FAPRI for EPA, Searchinger et al.) assumed that the yield effect was zero or negligible. All new production had to come from new land. More recent work allows for yield to respond to demand through changes in quantities of inputs. This includes work using the GTAP database (CARB, Hertel et al., Tyner et al., Kløverpris et al., and IFPRI) as well as some partial equilibrium work (IIASA, GLOBIOM and AGLINK). This work is rarely

Page 17: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

17

presented in a way that permits a cross-model comparison of the assumed responsiveness of yields to demand (Keeney and Hertel, 2008 is an exception). BLUM, covering Brazil, is the only piece of modelling work identified that allows for the possibility of increased cropping intensity. None of the pieces of modelling work identified allows for the possibility of increased yields through technological development in response to demand, even though it is common for such a relationship to be taken into account in other types of modelling (e.g. the Green-X energy model). In summary, the literature does not take into account two of the three ways in which yields could respond to demand (technological development and – with the exception of BLUM – increased cropping intensity); the way in which the third possible response (increased inputs) is taken into account is not transparent; and the empirical basis for these choices is obscure. This is unfortunate because it is clear, from the sensitivity tests carried out with some models, that this factor makes a great deal of difference to the modelling exercises’ results, as the table shows. Table – Sensitivity exercises illustrating the impact of different assumptions about the responsiveness of yields to demand Modelling exercise

Sensitivity test Impact

IFPRI (Al-Riffai et al., 2009)

Doubling of the land/fertilizer substitution elasticity

GHG impact: -28%

Keeney and Hertel, 2008

Permitting the amount of labour and capital used in agriculture in the US to change between the baseline and policy scenario

Land conversion: -43%

Kløverpris et al., 2009

Technological development: a 2% increase in the price of land causes a 1% increase in its productivity

Land conversion: Brazil -27% China: -80% US: -57% Denmark: -80%

Dumortier et al., 2009

Recalculation of results of Searchinger et al., 2008, assuming that, instead of no yield response, yields grow faster (by a little less than 0.1% p.a.) in the policy scenario than in the baseline scenario

GHG impact: -72%

Yield effects through lower productivity of converted land If converted land has a lower yield than “existing” (already cultivated) land, more land will need to be converted. As the table shows, most of the work reported here assumes that this is indeed the case.

Page 18: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

18

Table – yield attributed to converted land as a proportion of the yield of already-cultivated land Modelling exercise Assumption EPA 100% (?) Tyner et al. 67% IFPRI Brazil – 75%; elsewhere – 50% CARB 50% Searchinger et al. Low enough to cancel out a 12.5% increase in US corn yields (for

example, if the converted land area is 1/5 of the existing land affected by the yield increase, the assumption would be 33%)

These assumptions increase the amount of land needed by 1/(the percentage shown). For example, the assumption of Tyner et al. makes the amount of land needed 50% higher than it would otherwise be. The literature reviewed here did not include any assessment of whether converted land does indeed have lower yields than already-cultivated land. In all cases, the inclusion of such a factor in the modelling is simply an assumption. The same is true of the magnitude of the assumed difference in yields.

Impact of public policy (chapter 8.5) The EU legislative framework includes measures to require biofuels used for EU targets to achieve a minimum greenhouse gas saving and not to use raw material from certain types of land (including land with high carbon stocks). No studies evaluating the potential impact of such limits were identified in the literature review. None of the modelling reported here takes any account of these limits. Only one author – Searchinger, 2009 – gives reasons why they should be treated as having no impact. His argument ignores the possibility that if a premium is available for crops produced in compliance with particular requirements – as it is, for example, under the Round Table for Sustainable Palm Oil – this could tip the balance between investing in production pathways that comply with these requirements as opposed to investing in those that do not. It also ignores the potential cost to producers of the bad publicity that could come from evading sustainability requirements in the way that Searchinger describes. Public authorities impose more general limits on the use of land, e.g. for nature protection purposes. The literature review did not identify studies that formally evaluate the impact of such limits, but those authors who commented on the topic all considered that such measures affect, in the direction intended, the process and nature of land conversion. (See Tabeau et al., n.d., Zanchi et al., 2007, Greenpeace Brazil, 2009 and Lambin, n.d.) The modelling reported here takes into account these limits as they exist today, but does not take into account the fact that they can be expected to expand in future. (For example, data from the World Database on Protected Areas and the Millennium development goals indicators show that the area covered by nature protection areas has increased steadily since 1872 and at a rate of 1.8% a year since 1990.) Hertel et al., 2009, Babcock, 2009, IFPRI and Kløverpris et al. (in press), all note as a weakness their own modelling tools’ inability to model restrictions relating to the use of land. This point is also made by critics including Lambin et al., 2000 and New Fuels Alliance, 2008.

Page 19: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

19

LEITAP/ALTERRA and GLOBIOM estimate the effect of a (rather far reaching) restriction on the character of land use change – an effective policy preventing forest conversion. LEITAP estimate that such a policy would make the increase in the agricultural area 35% less than it would otherwise have been, and the greenhouse gas impact 52% less. In GLOBIOM’s study, the average “payback period” would fall by more than 90%.

Determining the type of land that is converted (chapter 9) Chapter 8 analyses how models estimate how much land will need to be converted from non-cropland to fulfil demand for crops. Since different land types have different carbon stocks, it matters which type of land is considered to be converted. That is the subject of chapter 9. Ensus, Kløverpris et al. and Searchinger et al. use a “historical” method, under which the proportions of different land types in land converted in the future are assumed to be the same as they were during some period in the past. EPA also use a historical approach outside the United States and IFPRI use a historical method for conversion of land from non-commercial to commercial use. As the table shows, practitioners of the historical approach use different time periods. Table – Time periods used by practitioners of the historical approach Study Time period Searchinger et al. 1990-1999 Ensus not stated IFPRI (for exogenous land use change in baseline): 2000-2004

Kløverpris et al. 1970-2000 or later? EPA 2001-2004 They all apply the approach at a regional level, but use different regions. It seems likely that the choice of regions has a big impact on the results. (If agriculture is growing in one region, and deforestation is occurring in a neighbouring region, then the merging of the two regions would lead a practitioner of the historical approach to conclude that a higher share of new land comes from forest.) No example was identified in which practitioners justified the regional breakdown chosen. Nasser et al. have shown, in relation to Brazil, the difference that this choice makes. As the tables shows, different applications of the historical method give sharply different results at the level of individual regions. Table – share of forest/woodland in land conversion to cropland Searchinger et al.

(1990-99) Ensus EPA

(2001-04)

Page 20: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

20

US 38% 0% 7% Latin America 75% 0-50% 4-8% South and South East Asia 100% 97-100% 34-74% India and China 0% 0-40% 7-17% Europe (conversion from cropland) 75% 0% 27% BLUM, CARD, IIASA and LEITAP use a “suitability” method, under which the land that is assumed to be converted is the land that is considered most suitable according to biophysical criteria. The sets of suitability criteria used vary substantially, but do not in any example include the infrastructure and social aspects highlighted in the methodological review in Kløverpris et al., 2008a. Models that use GTAP (including CARB and IFPRI, and perhaps also Kløverpris et al. in the case of conversion from managed grassland) use economic suitability criteria to model conversion between different types of “commercial” or “managed” land. A general criticism of the modelling exercises that use the biophysical suitability method is that they are not transparent. It is not clear exactly what suitability data are used, how they are weighted or what results they give. In the case of IFPRI and Kløverpris et al. it is not clear which allocations are made on the basis of historical data and which on the basis of economic suitability. It seems reasonable to consider that the 3 and 4 year time intervals used in some applications of the historical method are too short, given the deficiencies of the underlying land data (see chapter 13) and the extent of short-term ebbs and flows in land uses, notably between cropland and grassland. (This point is made in ICF International, 2009a.) Beyond this there appears to be no a priori reason to prefer one method to the other, or one way of applying a method to another. It is clear, nevertheless, that this is a choice that makes a real difference. For example, Searchinger et al.’s assumption on forest to cropland conversion in the US is criticised on empirical and methodological grounds by Dumortier et al., 2009. These authors conclude that the carbon payback time of US corn ethanol is less when US deforestation is excluded than when it is not by a factor variously stated as equivalent to about 1/3 and to 23%. The purpose of the scenario modelling reported here is to forecast what would happen under various conditions. This means that proponents of the historical approach need to argue why what happened in the past (at the level of geographical assessment they have chosen) is a good guide to what will happen in the future. Proponents of the suitability approach need to argue that the factors they have chosen to model are good predictors of the location of crop conversion. Little of this type of argument has been found in the literature. Most important, perhaps, is the need for empirical work to compare the predictions made by each approach with what actually happened. All the exercises require a decision about the set of land types that is to be modelled. All those mentioned here take into account forests and grassland (though CARB does not take into account unmanaged or non-commercial forests and grassland). Only Searchinger et al. and IFPRI take into account wetland/peatland. Only Ensus take into account recently abandoned cropland. None of the exercises takes into account recently deforested land.

Page 21: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

21

The gaps in this list are important. Wetlands and peatlands have high carbon stocks and can be converted to cropland (Gibbs in ICF International, 2009b; Searchinger, 2009a; Swallow and van Noordwijk, 2009). A failure to take this into account leads to an understatement of the carbon stock loss. Recently abandoned agricultural land has a lower carbon stock than grassland or forest. All the studies except Ensus treat the former as if it automatically becomes the latter. Crops may move onto land that has been deforested for another reason (e.g. logging). All the studies treat this as if the crop expansion were the cause of the deforestation. As a third case, an act of deforestation may be driven simultaneously by crop expansion and logging. (That is, the deforestation would not have happened if both drivers had not been present.) All the studies treat this as if the crop expansion were the sole cause of the deforestation. In all these cases, the result is an exaggeration of the carbon stock loss caused by crop expansion. (Tyner et al., 2009, may be an exception to this. Where other studies assume that 100% of forest carbon is lost on conversion to cropland, they assume that only 75% is lost. However, no explanation for this choice is given, so it has not been possible to check whether it is intended to take into account the factors mentioned here.) The need for a more nuanced approach to deforestation is underlined by the clear evidence that deforestation is usually the result of the operation of multiple factors. (See in particular Geist and Lambin, 2001, Geist and Lambin, 2002 and Geist et al., 2003.) In the first and second cases, it could be argued that the procedure used is correct because the land would eventually revert to grassland/forest (in the first case) or forest (in the second). But there are good reasons for thinking that this is not automatically so. Searchinger and his co-authors (Searchinger et al., 2008; Searchinger and Heimlich, 2008; Searchinger, 2009) suggest that there are two problems in the standard approach to modelling pasture: (a) In the policy scenario, there is a reduction in livestock nourishment from pasture

(because some is converted to cropland) that ought to be compensated (with conversion of other land to pasture or increased production of animal feed) and is not.

(b) In the policy scenario, there is a reduction in livestock nourishment from cereal feed

(because some is diverted to biofuels). This is modelled as being compensated by increased production of animal feed. Properly, a part ought to be compensated (more land-intensively) by the conversion of other land to pasture.

It has not been possible to verify whether (a) correctly describes the models reviewed here. If it does, it seems a reasonable point to raise. For the EU, the models reviewed here depict an increase rather than a decrease in the availability of animal feed (reflected in a decrease in its price). This means that if Searchinger et al.’s are correct to state that the models function in the way described in (b) and that this means an understatement of the carbon stock effect when animal feed becomes scarcer, the opposite effect could be expected in the EU case where animal feed availability is increasing. At any rate, both arguments point to a need for realistic modelling of the input that pasture provides to livestock production.

Evaluating carbon stock changes (chapter 10)

Page 22: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

22

The means by which indirect land use change has a greenhouse gas impact is through changes in carbon stocks. This calculation is at the centre of the work reviewed here. There are numerous significant differences in how the studies calculate this. With the partial exception of Searchinger et al., the literature contains little explicit discussion of the methodological and data issues involved. Each study tends to use its own assumptions without mentioning, let alone critically evaluating, others. A first point on which studies differ is on the proportion of carbon stocks lost when land is converted to cropland. For example, in the case of the conversion of a forest to cropland: - Searchinger et al., 2008, assume that all the carbon in vegetation (above and below

ground) and 25% of the soil carbon will be lost; - O'Hare et al., 2009, assume that all the above ground carbon and 25% of the below

ground carbon will be lost; - Mortimer et al., 2008 assume that all the above ground carbon and 10% of the below

ground "biomass" will be lost. - Tyner et al., 2009 assume that 75% of the carbon in vegetation and 25% of the carbon

in soil will be lost. Only Searchinger et al. provide data to justify (in part) their assumption. Among the studies mentioned, Searchinger et al.'s assumption gives the highest result for the quantity of carbon lost. O'Hare et al.'s assumption gives a lower result in the sense that only 25%, rather than 100% of below-ground vegetation is assumed to be lost. The assumptions of Mortimer et al. and Tyner et al. give results that are lower still. In Mortimer et al., 15% less below-ground carbon is lost3; in Tyner et al., 25% less above-ground carbon is lost. A second point on which studies differ is on the carbon stock attributed to vegetation on cropland (i.e., to crops). According to Nasser et al., 2009, for CARB, the figure is zero; for EPA it is 5 tCO2eq/ha; for BLUM it is 5 tCO2eq/ha for most crops and 17 for sugar cane. A third point on which studies differ is the treatment of foregone sequestration: the take-up of carbon from the atmosphere by forests that are present in the baseline scenario but not in the policy scenario. These may be (i) existing forests that are converted to cropland only in the policy scenario or (ii) existing cropland that remains cropland in the policy scenario, but is converted to forest in the baseline scenario. With some oversimplification, existing forests can be divided into those that are mature and those that are still growing. It can then be said that (i) should only be taken into account in relation to forests that are still growing. It seems that CARB do not take (ii) into account. Dumortier et al. take (ii) into account and attribute 20 years of sequestration to the forest.

3 If the term "biomass" is meant to refer only to vegetation and to exclude soil carbon, then Mortimer et al.'s result would be even lower.

Page 23: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

23

Searchinger et al. take both (i) and (ii) into account. The method to determine the share of growing forests assumed for (i) is not explained. For both (i) and (ii), 30 years of sequestration is attributed to the forest. (In Searchinger et al.'s paper, carbon stock changes are divided over 30 years.) Tyner et al. take (i) into account, assuming for each region that the proportions of growing and mature forest converted are the same as their proportions in that region's forest stock. They present analyses in which carbon stock changes are divided over either 30 or 100 years; this determines the amount of sequestration attributed to the forest. It seems likely that they also take (ii) into account. A fourth point on which studies differ is the carbon stock values attributed to different land uses. The table shows some examples for land types which appear in several different studies. Table – examples of differences in carbon stock values used

tCO2/ha

Tropical forest/rain forest, Brazil/SE Asia

Temperate forest, Europe

Boreal forest, Canada

Cerrado grassland, Brazil

Temperate grassland

Temperate cropland

Meizlish et al., 2007

549

Fritsche, 2007

539 172

Searchinger et al.

Brazil: 872-1092 SEA: 843

931-1077 1085 191 US: 330 Europe: 718

IFPRI 359-1506 339-983 71-238 139-220 139-348 111-240 Fargione et al., 2008

Brazil: 2700 SEA: 2572

311 US: 605

"German SBO draft"

971 256 202

Amaral et al., 2008

Brazil: 993 262

Tyner et al. 1085 492-515 WWF/ Cooperative Bank, n.d.

601

Mortimer et al., 2008

409

Some of the disparities may be due to differences in the precise pieces of land each value is meant to cover. Nevertheless, they appear rather significant, with the highest values of those found in the literature being between 2.2 and 15.3 times the lowest (median 5.7). Even when the source of the carbon stock values is supposed to be the same, these differences remain. Thus, Tipper et al., 2009 state that the carbon stock loss when forest is converted to non-forest use, calculated using IPCC data, is 322 tCO2/ha. However, Searchinger et al. state that IPCC data for forest to cropland conversion give losses of 299-627 tCO2/ha (temperate), and 553-824 tCO2/ha (tropical). Finally, Tyner et al. state that IPCC data give a loss of 1374 tCO2/ha for conversion of forest to cropland.

Page 24: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

24

Many studies state that the underlying source of carbon stock data is the IPCC. According to Koeble et al., 2009, the IPCC estimate of global soil carbon (0-100 cm), at 1500 Pg, is significantly higher than that obtainable from other sources (Harmonised World Soil Database: 1273 Pg; National Resources Conservation Service: 1376 Pg). This suggests that modelling exercises using IPCC carbon stock values may tend to overestimate soil carbon. Instead of IPCC, Searchinger et al. and Tyner et al. use the "Woods Hole" data-set. It can be seen that the carbon stock values used in these studies are generally at the high end of those in the literature. If studies using IPCC values do indeed tend to overestimate soil carbon, this suggests that studies using Woods Hole data may tend to do so to a greater degree (if indeed the difference is attributable to soil rather than vegetation values). A fifth point on which studies differ is the values attributed to foregone sequestration. Two studies go into detail on this: Tyner et al. and Searchinger et al. Both studies agree that the highest loss (around 5 tCO2/year per hectare of forest converted to cropland) occurs in the EU, and that the loss is also high (3-4 tCO2/ha/year) in South and South East Asia. However, for Tyner et al. there is also a significant loss (1.5 tCO2/ha/year) in the US, while the cost in Canada is close to zero. For Searchinger et al., the opposite is the case, with 2 tCO2/ha/year in Canada and close to zero in the US. None of the studies appears to take into account foregone sequestration from converted grassland. This could be because the amount of sequestration foregone is low. However, the value given in Mortimer et al., 2008 (about 1 tCO2/ha/year) does not appear insignificant. While the differences revealed in this analysis would seem significant, it has not in general been possible to quantify their impact on the results. An example of the quantification of one specific factor is found in Kim et al., 2008. These authors state that Searchinger et al. assume that all converted maize fields use conventional tillage. The current average is that 60% use conventional tillage while 40% use conservation tillage, which reduces the loss of soil carbon. If Searchinger et al. had used that ratio in their work, the calculated greenhouse gas benefits over 100 years would have been 8-9% higher.

Results for the land use change impact of biofuels (chapter 11) A common way to measure the greenhouse gas impact of fuels is "gCO2eq/MJ". This measure relates greenhouse gas emissions, measured in grammes, and with emissions of other gases (such as N2O) translated into equivalent quantities of CO2, to the energy value of the fuel, measured in megajoules. The table shows estimates of the greenhouse gas impact of land use change associated with biofuels that have been calculated from a range of studies in the literature.4 The "net impact" is calculated by adding the land use change impact of biofuels to the greenhouse gas emissions incurred in their production, and subtracting the greenhouse gas emissions of the avoided fossil fuel consumption (for this, the values used vary between 84 and 104 g/MJ).

4 Searchinger et al. (2008), Searchinger (2009), EPA XXX, CARB (2009), Tyner et al. (XXX), Al-Riffai et al. (2010), AGLINK (XXX)

Page 25: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

25

Table – Estimates of the greenhouse gas impact of land use change associated with biofuels, gCO2eq/MJ5 date of study land use change impact

(gCO2eq/MJ) net impact of biofuel consumption when indirect land use change is taken into account (a negative figure indicates greenhouse gas savings)

Searchinger et al.

2008 156 to 270 127 to 232

EPA (draft data for use in US Renewable Fuel Standard)

106 to 130 41 to 52

CARB (data for use in California's Low Carbon Fuel Standard)

44 to 68 15 to -13

Tyner et al.

2009

36 8 EPA (final data) 8 to 54 -4 to -69 Hertel et al. 40 10 Tyner et al. 21 to 32 1 to -9 IFPRI (study for European Commission)

2010

17 -43

It can be seen that with time, presumably as study methods have become more refined, the estimated impact of the land use change attributable to biofuels has fallen. For early studies, the estimated overall impact made biofuel use clearly undesirable (in greenhouse gas terms). In recent studies, the overall impact is usually estimated at values which make biofuel use beneficial in greenhouse gas terms. It should be borne in mind that these studies vary in numerous ways, including the policy scenario evaluated. In IFPRI, the additional biofuel consumption in the policy scenario is in the EU; in all the others, it is in the US. Some studies present results that can be used to compare the land use change impact of different types of biofuel. Their results vary widely. Most often, these suggest that one or another type of biodiesel – most frequently, soya - performs worse than ethanol – although the results of the model comparison exercise coordinated by the JRC-IE tend to point in the opposite direction. The literature reviewed does not permit a reliable check on whether this result – if believed – should be taken as applying to all vegetable oils or only to some. Only one study (Kløverpris et al., 2008b) looks into whether demand for a given feedstock has a different impact depending on the geographical location of the demand. This calculates that the land use change effect is 12 times as high when consuming a product from one place

5 Emissions from land use change have been divided over 20 years. See section [ ]. The cited results are all for first-generation biofuels.

Page 26: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

26

(China) as from another (the US). Given the magnitude of this difference, and its important potential policy implications, it is unfortunate that it has not been further explored in the literature.

Comparing land use effects with GHG savings from biofuels (chapter 12) In order to compare the land use effects calculated by the models with the GHG savings from biofuels, three further steps are necessary: - calculate the emissions from the production of the biofuels - calculate the emissions avoided from an appropriate fossil fuel comparator - decide how the carbon stock loss (a stock change) should be compared with the annual

greenhouse gas savings from using biofuels rather than fossil fuels (a flow) These are the topics covered in chapter 12. Concerning emissions from the production of biofuels, a variety of different values for emissions from different production pathways has been found in the literature. However, it has not been possible to identify which values are used in the land use change studies reviewed here, and thus to assess to what extent the choice of values has affected the results. None of the studies takes into account the likelihood that the greenhouse gas performance of biofuels will improve between now and 2020. Concerning the fossil fuel comparator, the key question is, if biofuels were not to be used, what type of fossil fuel would be added to the mix to fill the gap? This means that the fossil fuel comparator should be the marginal source of fossil fuel in 2020, not the average. It should be the long term rather than the short term marginal, because the legislation that promotes the use of biofuels is far reaching in both time and volume, and can therefore be expected to influence decision making concerning investments. In practice, this is not the case in the studies reviewed here. EPA, CARB and IFPRI all use as their fossil fuel comparator an estimate based on average, present-day emissions of conventional crude (102-103, 95-96 and 92 gCO2/MJ, respectively). No study has been identified which takes a different approach. By contrast, while studies in the literature that attempt to identify the long term marginal fossil fuel source have concluded in different ways, none seems to assume that conventional crude is the marginal source. For Persson et al., 2007, synthetic fuels from coal and oil shale are on the margin. This implies emissions of at least 120 g/MJ. For Renewable Fuel Association, 2009, the margin will contain a mix of conventional and unconventional fuels, with emissions of 101-109 g/MJ in the best case and 104-115 g/MJ in the worst case. Johnston, 2009, also sees the more carbon-intensive fuels as being on the margin. Charts from BP, 2009 and from Brandt and Farrell, 2007 both suggest that at higher oil prices it is unconventional fuels that are on the margin. This difference can have a significant impact on studies' results. For example, if CARB and EPA used a fossil fuel comparator that was 15g higher than the one actually chosen (which would be consistent with RFA's worst case assumption and less than the values implied by

Page 27: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

27

Persson et al. and by Johnston), this would be equivalent to reducing the estimated greenhouse gas impact from land use change by about 30%. Concerning the comparison of flows with stock changes, the method required by the Renewable Energy Directive (in the case of "direct" land use change) and used in the research of IFPRI is to divide the carbon stock change by 20 (thus spreading its cost across 20 years, without discounting). This assumption, while simple, is relatively conservative. (American work has tended to use a 30 year period, which decreases the calculated greenhouse gas impact by 33%.) Any more sophisticated choice would raise unresolved scientific issues and ought to be considered in the framework of the IPCC alongside the issue of the method for calculating global warming potentials for different gases.

Page 28: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

28

1. Purpose of the literature review

1.1. The task The Renewable Energy Directive (as far as biofuels and bioliquids are concerned)6 and the Fuel Quality Directive (as far as biofuels are concerned)7 require the Commission to submit to Parliament and Council a report with two components: (a) "reviewing the impact of indirect land-use change on greenhouse gas emissions"; (b) "addressing ways to minimise that impact". This paper is part of the work of the Commission services in preparing the first of these components. It draws on more than 150 contributions related to the topic. It is hereby made available for comments from interested parties.8

1.2. Main policy questions In June and July 2009 the Commission carried out a pre-consultation exercise on elements for addressing the issue of indirect land use change. Reflection on the comments made during that exercise has led the Commission services to focus on two particular aspects of the impact of biofuels and bioliquids9: - Is the land use change impact small or large relative to the greenhouse gas savings

delivered by biofuels and bioliquids (when their emissions from cultivation, processing, transport/distribution and use are compared with those of an appropriate fossil fuel comparator)?10

- Is the land use change impact homogenous (of similar size for all types of biofuels and

bioliquids that are produced using land) or heterogeneous (varying between types of biofuel/bioliquid, types of feedstock and/or location of production).

The analytical work needs to address these issues.

1.3. "Land use change" vs. "indirect land use change" In their work on this topic, including this literature review, the Commission services are addressing "land use change" in general rather than limiting the assessment to "indirect land use change" in particular. This is for two reasons. First, all the work reported here makes an estimate of the total land use change attributable to biofuels and bioliquids or to policies that

6 Directive 2009/28/EC 7 Directive 2009/30/EC 8 [It is assumed that another paper will include a simple-language explanation of what the issue actually is, including the terms "direct land use change" and "indirect land use change", the term "conversion", and the way in which C stock change causes an increased concentration of GHGs in the atmosphere. Therefore, this is not done here.] 9 Note: at many points in the paper the term "biofuels" alone is used. 10 These are the steps that are included in the method for biofuels, bioliquids and their fossil fuel comparators set out in Annex V of the Renewable Energy Directive.

Page 29: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

29

promote their consumption.11 None of the work attempts to estimate how much of the biofuels and bioliquids that are consumed will actually come from converted land, and which from non-converted land. Without such an estimate of the volume of "direct" land use change it is impossible to derive an estimate of the volume of "indirect" land use change. Second, this distinction has no relevance for the purpose of the present paper (that is, policy analysis). What matters is the total impact of the policy, not its breakdown between "direct" and "indirect" effects.

2. Outline of the paper

2.1. Why model? That indirect land use change occurs is intuitively plausible, because it seems plausible that agricultural commodities that would otherwise have served non-energy markets play a significant part in serving the new markets created by policies to promote biofuels and bioliquids. If this is so, the land use change impact of those policies cannot be properly estimated without taking indirect land use change into account. However, an inherent characteristic of indirect land use change is that it cannot be observed. It will never be possible, looking forward, to say that the introduction of a biofuel/bioliquid policy will lead to the conversion of a particular, identified piece of land. It will never be possible, looking back, to say that the introduction of a biofuel/bioliquid policy was the cause of a particular, identified piece of land being converted. It follows that the assessment of the impact of land use change requires the use of modelling.

2.2. Analytical options Three approaches can be identified to reviewing land use change impacts: (i) What has been the land use change impact of the consumption of biofuels and

bioliquids in the EU, Brazil and the US in the past? (historical analysis) (ii) What is the expected future land use change impact of biofuels and bioliquids under

different scenarios? (scenario analysis) (iii) Within a defined framework of restraints or limits on land use change impacts, what

would be the maximum possible consumption of biofuels and bioliquids/bioenergy in the future (potential analysis)?

This paper concentrates on exercises that have conducted scenario analysis. It attempts to understand each of the analytical steps taken in these exercises, and to comment on the choices made in relation to data and modelling approaches.

2.3. Steps in the (scenario analysis) modelling process The modelling exercises described here rest on a foundation of data describing the world (assessed in chapter 3).

11 Section [4.4.4] discusses the distinction between these two types of assessment.

Page 30: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

30

The exercises compare a baseline scenario with no or limited biofuel/bioliquid promotion/use (chapter 4) with one or more policy scenarios that include biofuel/bioliquid promotion/use (chapter 5).12 They use a model to do this (chapter 6). The model is used to calculate how the scenarios differ in terms of the quantity of crops produced (chapter 7). The model translates this difference into a difference in hectares of land used for crops (chapter 8), which must then be broken down by types of land converted (chapter 9) and translated into carbon stock changes (chapter 10) to give results for the land use change impact of biofuels or biofuel promotion (chapter 11). The result of this calculation must be compared to the greenhouse gas savings attributable to biofuels (chapter 12).

3. Data

3.1. The role of data in the modelling process All the “scenario analysis” modelling reported here has the following steps: (i) Describe relevant aspects of the world as it was in a given past or present year (termed

the base year in this paper); (ii) Estimate how these aspects of the world would change, between the base year and a

given future year (termed the modelled year in this paper) under "business-as-usual" conditions (the "baseline scenario" with no biofuel/bioliquid consumption or with no policies, or no new policies, that promote biofuels and bioliquids);

(iii) Estimate how the world would instead change under a policy that promotes biofuels

and bioliquids (the "policy scenario"); (iv) Subtract (ii) from (iii) to give the impact of the policy. Data are needed to describe the world in the base year; to predict how important explanatory factors (such as population and GDP) will change between the base year and the modelled year; and to calculate elasticities that can be used to estimate the change in one quantity that will be caused by change in another (for example, what change in meat consumption will be caused by a given increase in GDP). This chapter concentrates on data concerning four items at the centre of the land use change discussion: land use, crop production, trade and biofuels. Data on land carbon stocks and

12 Other terms are sometimes used – e.g., "reference scenario" instead of "baseline scenario". The AGLINK modelling exercise uses the term "baseline scenario" for what is here called the "policy scenario", and "counter-factual scenario" for what is here called the "baseline scenario". For the sake of clarity, the particular terms used in different studies are ignored here: this common vocabulary is used instead.

Page 31: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

31

greenhouse gas savings from biofuels are looked at in chapters 10 and 11. The values used for elasticities will be mentioned at various points but will not be comprehensively examined.

3.2. Land use and crop production data

3.2.1. Global datasets Ramos et al. (2009)13 offer an inventory of datasets of global land use: - FAO statistics (FAOSTAT) - Global land use maps:

- Global Land Cover 2000 - GlobCover 2005 - M3 Datasets

- Map of forest areas: - MODIS VCF14

- Map of zones suitable for agriculture: - GAEZ 2002 dataset (also known as AEZ)

Ramos et al.15 describe the data sources used to compile these datasets as follows: Satellite datasets – Global Land Cover, GlobCover, one M3 dataset Agricultural inventories – FAOSTAT, another M3 dataset. Although not stated in Ramos et al., MODIS is also a satellite dataset.16 The AEZ dataset is built up from FAOSTAT data (see below) and so can be classified as an agricultural inventory. As well as data on land use, the FAOSTAT dataset (and the AEZ dataset built on it) give a spatially explicit breakdown of crop production. They appear to be the only datasets that do this.17

13 p. 9 14 It appears from Lywood (2009) (p. 17) that this dataset is not limited to forests – Lywood quotes MODIS data on cropland. 15 p. 12 16 Lywood (2009) (p. 17) 17 [To check. According to Hertel (n.d) (p. 7), “[there is] only one published global data base on spatially explicit yields and harvested area” – this presumably means that FAOSTAT is unique in including crop production data. Sands et al. (n.d.) (p. 3) are less categoric, referring to “Spatially aggregated information about area harvested for specific crops at the national or sub-national level often from FAOSTAT” – but they do not mention the other sources they have in mind. If FAOSTAT is the only source, where do the data in the M3 dataset that is an ‘agricultural inventory’ come from?]

Page 32: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

32

3.2.2. Commentary - FAOSTAT The FAOSTAT database [ProdSTAT]18 contains annual data, by country, on crops produced and areas harvested. According to Bouët et al. (2009)19, the FAOSTAT database ResourceSTAT20 contains a description of global land use categorised as arable land; permanent meadows and pasture; permanent crops; forest area (plantation and natural forest); and other land. Finer categories are available for some regions but not all. These datasets are updated annually. They are based on reporting by states. They also contain a significant amount of estimation by FAO staff. Both could result in unreliability or bias. ADAS UK Ltd (2008)21 comment, “As Smil (2005) points out, the accuracy of some FAO-published data is open to question. Whilst concerns relate primarily to small developing countries without local auditing, there are significant concerns over large countries, such as China, where crop production data are treated with more than the usual level of secrecy, and crop areas are widely regarded as being understated.”22

3.2.3. Commentary - the AEZ dataset The AEZ or GAEZ (global agro-ecological zone) dataset is a sophistication of the FAOSTAT land and crop dataset. Instead of being presented by countries, as in the FAOSTAT database, land and crop data are presented by AEZs. All crop-growing land is allocated to one of 18 AEZs depending on climate (tropical, temperate or boreal) and growing period (6 tranches ranging from ‘0-59 days’ to ‘more than 300 days’). The land is classified according to its degree of suitability for growing each of the major crops. This work of allocating land to AEZs was done by the International Institute for Applied Systems Analysis (IIASA) in collaboration with FAO. The data year for the dataset that includes crop production data is 2004.23 The agricultural production data contained in GTAP, previously presented by countries, have presumably similarly been broken down by AEZ.24 25 ADAS UK Ltd (2008)26 raise queries about the AEZ data, stating that “there are many instances where GAEZ estimates of attainable yield, appear very different from local knowledge… e.g. they have little or no land as very suitable for sugar beet in the UK, and attainable yields are said to be 3 t/ha or so, whereas current yields are of the order of … 8 t/ha sugar”. They add that the GAEZ dataset was due to be updated in 2008. 18 [reference] 19 p. 66 20 http://faostate.fao.org/site/377/default.aspx [check; look directly at the data] 21 p. 7 22 [More comments needed to form a firm view. The Smil reference is missing from the list of references in the ADAS report but presumably would not be difficult to find.] 23 ADAS UK Ltd (2008) (pp. 6-7); Bouët et al. (2009) (pp. 66-70); Sands et al. (n.d.) (p. 4) 24 Sands et al. (n.d.) (p. 5) 25 [This work is reported in C.Monfreda et al., “Global agricultural land use data for climate change analysis”, GTAP technical paper, 2007 and A. Goleb et al., “Land use modeling in recursively-dynamic GTAP framework”, GTAP technical report, 2008. It would be good to read these papers to check that the description given in the text is accurate.] 26 pp. 6-7

Page 33: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

33

The fact that the AEZ dataset has a single data year (2004) for crops could be one source of bias, since crop yields vary a good deal from year to year and these variations are spatially differentiated.

3.2.4. Commentary – satellite datasets Smukler and Palm (2009) state that freely available Landsat satellite images are adequate for monitoring land use change at the regional level but not yet at the farm level. The situation is expected to improve over the next decade.27

3.2.5. Commentary – general Britz (2009) states that the type of modelling reported here does not cover animals and does not deal with farming practice or forestry practice.28 He argues that more finely grained data are needed for this.

3.2.6. Comparison of datasets The table gives examples of comparisons of the estimates given by different global datasets for the same quantity. Table: Comparisons of the estimates given by different datasets for the same quantity

27 pp. 17-18 28 p. 9

Page 34: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

34

Cultivated

land 2000 (Mha)

Cultivated land 2005 (Mha)

Cultivated land 2007 (Mha)

Change in cropland area 2001-2004

Soil organic carbon, 0-100 cm (Pg)

Global Land Cover 2000 GlobCover 1740 FAOSTAT (ProdSTAT)

1180 1260

FAOSTAT (ResourceSTAT)

1520 1540 EU: -1.1% Brazil: +20.7% US: +1.1%

M3 (satellite dataset)

1490

M3 (agricultural inventory)

1250

MODIS EU: +17.9% Brazil: -2.0% US: +7.1%

Harmonised World Soil Database

1273

National Resources Conservation Service

1376

IPCC 1500 Source for the first three columns: Ramos et al. (2009)29, approximate figures obtained from measuring on a graph, rounded to 10 Mha. Source for the fourth column: Lywood (2009).30 Source for the fifth column: Koeble et al. (2009)31; the paper states that all three estimates used FAO as the data source. It can be seen that there are large differences in the estimated cropped area (estimates for 2000 range from 1180 to 2000 Mha) and rate of land use change (ProdSTAT shows an increase in cropland of 80 Mha between 2000 and 2007 while ResourceSTAT shows an increase of 20 Mha; the fourth column shows bigger differences at regional level.)

3.2.7. Datasets used in modelling work IFPRI use a model based on GTAP to determine the country in which crops will be grown. They then use the AEZ dataset in translating this demand for crops into the employment of particular types of land as cropland. The authors calculate the amount of “marginal land that could be used for complementary production” by subtracting existing cropland, pastureland and forest from the total land area. This is done for each country and for each AEZ region within each country. 32 Cropland expands through two processes termed “substitution” and “extension”. Under both processes, land of other types can be converted to cropland.33 Both

29 p. 12 30 p. 17 31 p. 13 32 Valin et al. (2009) (p. 7); Valin (2009) (p. 15); Bouët et al. (2009) (pp. 70-71) 33 For more details see [chapter 9].

Page 35: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

35

processes use the AEZ dataset. The second process also uses the FAO ResourceSTAT dataset.34 EPA use MODIS for a process analogous to IFPRI’s “extension” process.35

Tyner et al.’s modelling exercise “incorporates disaggregated Agro-ecological zones (AEZs) (Lee et al., 2005) for each of the land using sectors”.36 LEITAP use data from the IMAGE model for non-European regions and from the CLUE model for European regions.37

3.2.8. Conclusion While there are several different data-sets concerning land use, it seems that the only global dataset of crop production is the FAOSTAT database. A check should be made of whether systematic biases can be expected in the FAO data. Differences in land use, and the associated differences in carbon stocks, are the main subject of the Commission assessment to which this paper contributes. It seems from section 3.2.6 that datasets often differ in the land use category to which land is assigned. It is necessary to form a better understanding of these differences and the reason for them; and to decide which dataset(s) offer a reliable basis for the Commission’s assessment. If the only dataset of crop production is the FAOSTAT database, and this database is found not to offer sufficiently reliable data on land use, it will be necessary to consider how to link FAOSTAT data on crop production to other data on land use. Apart from their use in modelling, historic data on land use change have important things to tell us about the scale of the phenomenon (land conversion for crops) that we face. Data on the rate of change of (a) the cropped area and (b) the forested area are therefore particularly important. In assessing the FAO ProdSTAT data it would be helpful to understand better how much of the reported “harvested area” represents multi-cropping and, if this is significant, how fast multi-cropping is growing. In assessing FAO and satellite data it is important to obtain data for land that was cropped recently but not in the data year (because it has been abandoned or is fallow as part of a rotation). It would be helpful to better understand the scale and character of this category of land, its productivity, and what typically happens to abandoned land in the medium term. Modellers do not seem to have systematically considered these issues.

34 Bouët et al. (2009) (pp. 73-5) 35 Lywood (2009) (p. 17) 36 Taheripour et al. (2008) (p. 7). [The reference is to Lee et al., “Towards an integrated land use data base for assessing the potential for greenhouse gas mitigation”, GTAP technical paper #25 (2005) 37 ALTERRA (2009) (p. 7) The CLUE model uses the CLC/CORINE land cover database (ALTERRA (2009) (p. 8)). This is not examined further because it is not a global dataset.

Page 36: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

36

3.3. Trade data

3.3.1. Introduction If the land use change impact of biofuels and bioliquids is heterogeneous38, it will make a difference where the raw materials to make them are sourced from and where the production increases to replace these diverted commodities occur. The modelling reported here uses trade data (including data on elasticities) to determine this. The models are divided into computable general equilibrium (CGE) models and partial equilibrium (PE) models.39 The CGE models all use variants of the GTAP database. The source of the data used in the PE models is not clear.

3.3.2. Commentary – the GTAP database CGE models need a great deal of data. The GTAP database has developed, especially since the 1990s, as a common source of these data.40 It is “first and foremost a global data base describing bilateral international trade, production/consumption for [the] entire economy” and is “a peer-reviewed, fully-vetted, published and documented data base”.41 42 43 It is possible for GTAP users to add data for a specific purpose. Examples have been the creation of the GTAP-E model, including a more sophisticated treatment of the energy sector44, and of “GTAP-AEZ”, described in section 3.2.3. Up to now, a weakness of the GTAP database has been that there is a single category of “oil seeds”.45 GTAP has therefore not been an appropriate tool to analyse land use change associated with biodiesel. However, IFPRI's work breaks the category of “oil seeds” down into its constituent parts. Sugar cane and sugar beet are also separated. Hertel (n.d.) comments that “[the] economics profession and policy makers under-invest in estimation of … parameters”, implying that this is a weakness of the database.46

3.3.3. Datasets used in the modelling work

38 See section [1.2] 39 See chapter 6 40 Decreux and Valin (2007) (p. 9) 41 Hertel (n.d.) (p. 2) 42 This description makes the data sound transparent. Lywood (2009) (p.7) suggests that they are not: “Although the database and source code for the GTAP model is available on the internet, this does not provide the data needed to validate the [CGE] models.” 43 Kløverpris et al. (2008) state, “GTAP … maintains a database representing the global economy… Primary production factors (capital, labour and land) are constrained in the model, and the interaction between sectors and regions is based on economic input-output databases, elasticities of supply and demand (empirically estimated or calibrated by the model), international trade regulations, and trade agreements (bilateral and multilateral). The economic consequences of a change (in demand, supply, policy, etc.) can be studies by introducing a so-called ‘shock’, e.g. a region-specific change in crop demand. The result is a new economic equilibrium with all changes expressed in relative terms.” (p. 19) 44 [generally referenced to Burniaux, J. and T. Truong, “GTAP-E: An Energy-Environmental Version of the GTAP Model”, GTAP Technical Paper No. 16, Purdue University, 2002] 45 Tyner et al. (2009) (p. 8) 46 p. 6

Page 37: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

37

CARB use GTAP. IFPRI use GTAP-E with additional refinements concerning the energy side.47 They consider two types of biofuel (bioethanol and biodiesel). Up to now they have modelled four biofuel feedstocks: maize, wheat, “sugar crops” and “oilseeds”. They are working to add more (see above). The GTAP data year is 2004. Tyner et al.48 and LEITAP consider two types of biofuel but do not have such a detailed breakdown of biofuel feedstocks.49

3.4. Biofuel data

3.4.1. Introduction Some modelling exercises include “existing policies in support of biofuels” in the baseline.50 For these models it is important to correctly depict the biofuel consumption that is taking place today. More generally, a significant deviation between the depicted and real quantities and types of biofuels consumed today would suggest that a model is not correctly set up.

3.4.2. Historical production/consumption data Smeets (2009) gives the following data for biofuel production in 2006-2008, giving the source as the International Centre for Trade and Sustainable Development.51 Table – EU27 biofuel production, imports and consumption, 2006-2008 (Mtoe) – data from Smeets (2009) 2006 2007

(estimate) 2008 (estimate)

47 GTAP-E has a more fully specified energy sector than the ordinary GTAP database (Al-Riffai et al. (2010) (p. 32)) 48 Modelling done at Purdue University in 2008 49 According to the authors of the IFPRI work (Bouët et al. (2009) (p. 24)) 50 See chapter 4 51 [Full reference: ICTSD, “EU support for Biofuels and Bioenergy, Environmental Sustainability Criteria, and Trade Policy”, 2009, p. 56. The source also gives projections for 2009 and 2010. It is stated that the projections are based on projections by the US Department of Agriculture; it is not clear if this is also the case for the 2006-2008 data.]

Page 38: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

38

biodiesel rapeseed oil 2.8 3.1 3.3 soybean oil 0.7 0.8 0.8 palm oil 0.1 0.4 0.4 sunflower oil 0.2 0.2 0.3 other and not attributed 0.1 0.1 0.1 subtotal vegetable oils 3.9 4.6 4.8 recycled vegetable oil 0.1 0.1 0.2 animal fats 0.0 0.0 0.1 grand total 4.0 4.7 5.1 imports 0.1 0.7 0.9 consumption 4.1 5.4 6.052 biodiesel share of diesel market 2.3% 3.0% 3.2%

bioethanol production 1.0 1.1 1.4 imports 0.2 0.6 0.8 consumption 1.2 1.7 2.2 bioethanol share of petrol market 0.8% 1.2% 1.5% Source: Smeets (2009)53, sourced from International Centre for Trade and Sustainable Development. Net ethanol imports calculated, data converted from tonnes to Mtoe and rounded to nearest 0.1 Mtoe by Commission services. IIASA (2009), citing the IEA’s 2008 World Energy Outlook as the source, put the global market share of biofuels in transport energy demand at about 1.5% “today”.54

3.4.3. Depiction of historical production/consumption data in modelling exercises

AGLINK depicts a biofuel market share in the EU of 1.6% in 2008. This is below the correct figure of 3.3%.55 56 EU transport fuel consumption is depicted as 145214 Ml petrol and 179193 Ml diesel in that year. Ethanol consumption is depicted as 6698 Ml (“1.9% energy share”) made up of 5021 Ml domestic production and 1677 Ml imports.57 Biodiesel consumption is depicted as 9200 Ml (“3.5% energy share”) made up of 8064 Ml domestic production and 1136 Ml imports.58 GLOBIOM uses estimates from POLES, depicting global biofuel consumption of 10 Mtoe in 2000.59

52 The source gives 6.7 Mt, equivalent to 5.9 Mtoe, but this appears to be an error of arithmetic or transcription. The figure given here is based on a total, derived from adding constituent elements, of 6.76 Mt. 53 p. 7 54 pp. 15, 17 55 See [previous] section. [The 3.3% figure comes from Eur’Observer] 56 The figure also looks odd in relation to the energy shares attributed individually to ethanol (1.9%) and biodiesel (3.5%). 57 Gay and Kavallari (2009) (p. 9) 58 Gay and Kavallari (2009) (p. 10) 59 ECORYS (2009) (p. 41)

Page 39: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

39

3.4.4. Conversion efficiency UNICA (2009)60 state that from the point of view of ethanol production, Brazilian sugar cane yields increased by more than 16.6% between 2001 and 2006-2008, derived from an 8.2% increase in tons of sugar cane per hectare (between 2001 and 2006-2008) and an 8.3% increase in Total Recoverable Sugars per ton of sugar cane (between 2001 and 2006)61.

4. Baseline

4.1. Introduction Modellers define a baseline scenario and define a set of baseline assumptions. The baseline scenario constitutes some kind of "business as usual" story about the level of biofuel consumption.62 It is compared with one or more policy scenarios (see chapter 5). The baseline assumptions concern variables that are not themselves the phenomenon of primary interest for the modelling exercise, but which affect model outcomes. They may be set at the same values in the baseline and policy scenarios, or may differ between the two. This chapter looks first at baseline scenarios in modelling exercises, and then at baseline assumptions about: - crop yields and demand - the oil price and petrol and diesel consumption - land use - agriculture and tariff policy It is assumed that other factors, such as population and GDP, are of less importance for the purposes of this paper (because differences in assumptions on these points are unlikely to be correlated with differences in the size or character of the difference in impact between the baseline and policy scenarios).

4.2. Biofuels in the baseline scenario

4.2.1. Second generation biofuels In Searchinger et al. it is assumed that second generation biofuels will play no significant part up to 2016. In IFPRI it is assumed that second generation biofuels will play no significant part up to 2020. The same appears to be true of LEITAP/ALTERRA and LEITAP/Banse et al. In AGLINK, CAPRI, ESIM, GLOBIOM and IIASA63 second generation biofuels play no role in the baseline, but make an exogenously defined contribution in the policy scenario.64 60 p. 19 61 Data for 2007 and 2008 were not available 62 None of the work reported here covers bioliquids. 63 All these are partial equilibrium models. 64 See section [5.3.1]

Page 40: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

40

4.2.2. Biofuel policy in jurisdictions other than the one of interest65 In IFPRI, the jurisdiction of interest is the EU. As for other jurisdictions, “The US mandate will lead to the consumption of 40 Mtoe of ethanol by 2020 … while… the US biodiesel sectors … will represent … 12% of the ethanol sector… [T]he Brazilian blending target [is] fixed at 24.4% over the period… We also include a 5% mandate for Indonesia, Malaysia, Rest of OECD and China."66 The US and Brazil levels are achieved through mandates and fiscal measures.67 It appears that no countries other than the EU, US and Brazil are considered to adopt biofuel policies in the baseline, since it is stated that “The third country [after the EU and US] that we consider in the baseline is Brazil”.68 In ESIM, the jurisdiction of interest is the EU. Countries other than the European Union do not have a biofuel policy.69 In LEITAP/ALTERRA, the jurisdictions of interest can be seen as (a) the EU and (b) the world. In assessing the impact of biofuel policy in the European Union, a scenario with a “biofuel mandate” in other OECD countries but not the EU can be used as a baseline.70

4.2.3. Biofuel policy in the baseline in the jurisdiction of interest Jurisdiction of interest is the European Union In AGLINK the baseline scenario is “without EU renewable energy directive”. Tax rebates for biofuels are eliminated and there is no blending obligation.71 In CAPRI the baseline scenario is “without European Union Biofuel Directive in year 2020”.72 In IFPRI, “In the baseline … [t]he status-quo is assumed to prevail until 2020, with biofuel blending levels not exceeding the 3.3% level in 2008."73 In LEITAP/ALTERRA, when the jurisdiction of interest is the EU, the baseline is no biofuel mandate in the EU.74 Jurisdiction(s) of interest is the world In AGLINK’s 2008 study, the baseline is “no support”.75 This is global.

65 The expression ‘the jurisdiction of interest’ is used because some of the modelling work reported focusses on the US or the world rather than the EU . 66 Al-Riffai et al. (2010) (p. 44) 67 Bouët et al. (2009) (pp. 91-2) 68 Bouët et al. (2009) (p. 92; see also p. 25) 69 Tonini and Henseler (2009) (p. 2) 70 ALTERRA (2009) (p. 14). There are two reference scenarios: one without biofuel mandates, and one with biofuel mandates in the rest of the OECD but not the EU. A particularity of LEITAP/ALTERRA is its assessment of the impact of public policies restricting undesirable land use change (see chapter 8). This assessment is only made in the case where the jurisdiction of interest is the world. 71 Gay and Kavallari (2009) (p. 7) 72 Blanco Fonseca and Pérez Domínguez (2009) (p. 4) 73 Al-Riffai et al. (2010) (p. 43) 74 ALTERRA (2009) (p. 14)

Page 41: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

41

GLOBIOM use a baseline scenario in which global biofuel consumption is fixed at 2005 levels until 2030.76 IIASA use a baseline scenario in which global biofuel consumption is fixed at 2008 levels.77 In LEITAP/ALTERRA, when the jurisdiction of interest is the world, the baseline is no biofuel mandate in any OECD country.78 LEITAP/Banse et al. use a baseline scenario in which there are no biofuel blending obligations (and, presumably, no other types of biofuel support) in any jurisdiction.79

4.2.4. Modelled consumption of biofuels in the baseline In AGLINK/OECD’s 2008 study, biofuel consumption in the EU, US and Canada reaches approximately 32.4 Mtoe in the baseline (29.8 Mtoe ethanol and 2.6 Mtoe biodiesel).80 In AGLINK-COSIMO EU ethanol consumption in the baseline is 3.4 Mtoe (“2% energy share”) in 2020. This is made up of 3.2 Mtoe domestic production and 0.2 Mtoe imports. EU biodiesel consumption in the baseline is 3.4 Mtoe (“1.3% energy share”), made up of 2.8 Mtoe domestic production and 0.6 Mtoe imports.81 Although AGLINK-COSIMO includes second generation biofuels, none are produced in the baseline scenario.82 In IFPRI global ethanol production in the baseline is 84.4 Mtoe in 2020 (US: 29.1 Mtoe, Brazil: 28.5 Mtoe). Global biodiesel production is 19.5 Mtoe (EU: 8.1 Mtoe).83 EU consumption is 3.3% of petrol and diesel (9.75 Mtoe). In ESIM biofuels achieve a 2% share of the EU transport fuel market in 2020 in the baseline.84 This is 8.552 Mt of biofuel, made up of 7.817 Mt of domestic production and 0.734 Mt of imports.85 It is not clear if the term ‘imports’ refers only to finished biofuels or also to raw material used to make biofuel in the European Union. In LEITAP/Banse et al., the share of biofuels in Brazil increases from 14% in 2001 to 21% in 2010 in the baseline. The share in France and Germany increases to 1.3% and 2.3% respectively over the same period.86 75 von Lampe (2009) (p. 11) 76 ECORYS (2009) (p. 41). Formally this is the “No biofuels” scenario in the vocabulary of the GLOBIOM study, while the “Baseline” scenario is one of three that include policies favouring biofuels. 77 IIASA (2009) (p. 15) 78 ALTERRA (2009) (p. 14) 79 Banse et al. (2007) (p. 8). It is not clear if this describes the European Union only or all countries. Banse (2008) (p. 11) would seem to imply that the statement describes the baseline in all countries, stating that “implementation of biofuel initiatives outside Europe” occurs in the policy scenarios. 80 von Lampe (2009) (p.11); back-calculated from percentages and converted to Mtoe by Commission services. 81 Gay and Kavalleri (2009) (pp. 9-10); conversion to Mtoe by Commission services. 82 Strictly speaking, this can only be said of the domestic production, since for imports no distinction is made between first and second generation biofuels. 83 Al-Riffai et al. (2010) (p. 50) 84 Tonini and Henseler (2009) (p. 7); this is an approximate value that has been measured off a graph. 85 Tonini and Henseler (2009) (p. 9) 86 Banse et al. (2007) (p. 10)

Page 42: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

42

In PEEP biofuels achieve a 1.7% share of the EU transport fuel market in 2020 in the baseline.87 In Searchinger et al. US consumption of ethanol (all from maize) in the baseline is 28.0 Mtoe in 2016.88 These results are summarised in the table. Table – Biofuel consumption in the baseline in the jurisdiction of interest Modelling exercise

Jurisdiction of interest

Modelled year

Biofuel promotion in baseline scenario?

Modelled biofuel consumption

AGLINK/OECD EU+US+Canada 2020? x 32.4 Mtoe Searchinger et al.

US 2016 ü 28.0 Mtoe

AGLINK-COSIMO

EU 2020 x 6.8 Mtoe – 1.6%89

IFPRI EU 2020 ü 9.75 Mtoe - 3.3% of petrol+diesel

ESIM EU 2020 x 2% PEEP EU 2020 ? 1.7% of transport fuel Sources – see sections 4.2.3 and 4.2.4

4.2.5. Commentary In all the modelling work, it is assumed in the policy scenario that measures are introduced in the jurisdiction of interest to promote the consumption of biofuels90 (see chapter 5). Some modelling work also assumes the introduction (or retention) of a more limited set of measures in the baseline. Other modelling work assumes no measures in favour of biofuels in the baseline. This matters if the model is of a type that depicts the land use change impact as a non-linear function of demand.91 For example, if the unit land use change impact is modelled as an increasing function of demand, the average land use change impact per unit of biofuel will appear to be greater in a model whose baseline includes measures in favour of biofuels than in a model whose baseline does not. As would be expected, predicted biofuel consumption seems to be higher in baseline scenarios with biofuel promotion than in those without. Thus, EU consumption is predicted at 3.3% in IFPRI’s baseline scenario compared with 1.6 in AGLINK-COSIMO.

87 Berndes and Hansson (2007) (p. 5972) 88 Searchinger et al. (2008) (p. 5); conversion to Mtoe by Commission services. 89 Calculation of the Commission services 90 None of the work reported here includes bioliquids 91 See chapter 14

Page 43: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

43

The results of modelling exercises are typically expressed as the difference between the land use change impacts of the policy and baseline scenarios divided by the difference between the volumes of biofuel consumption they depict. In models that depict the land use change impact as non-linear and rising, the choice of a baseline scenario that includes a relatively high level of biofuel consumption will then result in a higher estimated unit land use change impact than the choice of a baseline scenario with less biofuel consumption. Similarly, the assumed level of biofuel consumption elsewhere than in the jurisdiction of interest matters if the model is of a type that depicts the land use change impact as a non-linear function of demand – and otherwise, not. The models with no measures in favour of biofuels in the baseline (AGLINK, ESIM and LEITAP/Banse et al.) still predict a certain level of biofuel consumption as a result of market forces. This means that none of the scenario analysis modelling (calculating the difference between the policy scenario and the baseline) can be said to estimate the land use change impact of “biofuels” (that is, of all those consumed, not merely those that are brought to market as a result of a particular policy). In one way or another, all the exercises depict the impact of “policies to promote the consumption of biofuels.

4.3. Baseline assumptions: crop yields

4.3.1. Empirical data ADAS UK Ltd (2008) assesses past trends in crop yields and forecasts future changes. The table shows their data for the major crop groups. Table – yield improvement rates

Improvement rate 1960-1991

Improvement rate 1991-2006

Improvement rate 2006-2020 (kg/ha/year and CAGR92)

Current average yield (2002- (kg/ha/ (kg/ha/ (kg/ha/year Business as Maximum

92 Compound Annual Growth Rate

Page 44: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

44

2006, kg/ha)

year) year) as % of current yield)

usual improvement

All cereals

3232 45.6 42.3 1.3%

Wheat 2788 46.0 25.7 0.9% 30.3 (1.0%) 73.4 (2.3%) Maize 4716 62.5 79.5 1.7% 67.5 (1.3%) 117.5 (2.2%) Rice 4008 55.9 35.7 0.9% 56.4 (1.3%) 97.9 (2.1%) Barley 2538 27.5 26.8 1.1% 25.0 (0.9%) 74.0 (2.5%) Palm oil 1291993 217.9 227.2 1.8% 73.5 (0.5%) 449.1 (2.9%) Oilseed rape

1696 24.6 31.6 1.9% 18.6 (1.0%) 37.3 (1.9%)

Soya 2305 25.1 24.0 1.0% 29.4 (1.2%) 52.6 (2.0%) Sunflower 1261 8.8 8.3 0.7% 16.1 (1.2%) 30.9 (2.1%) Sugar cane

65733 415.2 386.0 0.6% 597.6 (0.9%) 1230.9 (1.7%)

Sugar beet

44383 361.0 1023.5 2.3%

Source: ADAS UK Ltd (2008); last two columns with calculations of Commission services, treating the 2002-2006 average yield as the yield for 200694 The authors analyse the historical data as follows: “Despite substantial variation in yield trends between countries, the world trends for most of the major crops show a remarkably consistent linear trend. Despite yield plateaus in some crops in some regions, plateaus are not generally strikingly evident in world yield trends, though reductions in the improvement rate are apparent in some crops. Overall yields have tended to increase in a linear, arithmetic, Malthusian fashion (Hafner 2003)95. As world yields increase it appears inevitable that the yearly increase will decrease in percentage terms. It is therefore sensible to assess change by comparing rates in kg/ha/year… For cereals as a whole, yields have continued to increase at a steady rate, comparable across the species. However, the yield improvement rate has slowed for wheat and rice since around 1990. The yield improvement rate for maize, however, seems to have increased since 1990… On a global basis there has been no observable slowdown in yield improvement in the oil crops …Global yields of sugar cane and sugar beet have increased similarly over the past 40 years.”96 By contrast, Searchinger and Heimlich (2008) state that “One recent study has calculated that improvements in corn yields in the U.S. are actually a response to improved weather (Tannura 2008), and another study calculates that corn yields in the U.S. are falling behind recent trends (Basse 2008)”.97 UNICA (2009)98 state that from the point of view of ethanol production, Brazilian sugar cane yields increased by more than 16.6% between 2001 and 2006-2008, derived from an 8.2%

93 The table on p. 9 gives 2538. This looks like mistaken data entry. The figure of 12919 is given on p. 36. 94 p. 9 (historic data); pp. 18-36 (projections) 95 The reference is to S. Hafner, "Trends in maize, rice and wheat yields for 188 nations over the past 40 years: a prevalence of linear growth”, Agriculture Ecosystems and Environment 97, 2003 96 pp. 9-12 97 p. 14 (referenced as Tannura, M. et al., “Are Corn Yields Increasing at a Faster Rate” (2008) and Basse, D., Presentation on Biotech Farm Yields to the Farm Foundation (March 8, 2008)) 98 p. 19

Page 45: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

45

increase in tons of sugar cane per hectare (between 2001 and 2006-2008) and an 8.3% increase in Total Recoverable Sugars per ton of sugar cane (between 2001 and 2006)99. A gain of 16.6% between 2001 and 2007 would equate to a gain of 2.6% per year. A gain of 8.2% would equate to a gain of 1.3% per year. It seems likely that such changes in yields of Total Recoverable Sugars per ton of sugar cane are not taken into account in ADAS UK Ltd, which states “Sugar cane yields have increased by about 1% per year in Brazil”.100 Biomass Research and Development Initiative (n.d.) state that the literature on yield trends usually considers yields per harvested area rather than per planted area.101 This would seem to imply that the scope for increases in cropping intensity is not usually taken into account. According to Millennium Ecosystem Assessment (2005), “Globally, 78% of the increase in crop output between 1961 and 1999 was attributable to yield increases and 22% to expansion of harvested area. Of the expansion in area harvested, roughly two thirds was accounted for by physical expansion of arable land and the remainder was due to increases in cropping intensity.”102 Similarly, according to RFA (2008), FAO estimate that in 2030 about 20% of extra food production “is [sic]” the result of expansion of arable land, 70% of increasing yields and 10% of increased cropping intensity.103 This suggests that this (possibly omitted) factor is reasonably important.

4.3.2. Assessments of future prospects ADAS UK Limited's assessment of future prospects is in the table in the previous section. Searchinger (2009) reports on a study by Trostle104 according to which the U.S. Department of Agriculture predicts, based on recent trends, that global cereal and oilseed yields will grow at 0.8% per year. Smeets (2009)105 reports on a study by Bindraban et al.106 according to which the FAO has used values for global yield increases in the future ranging from 1.1% (for rice) to 1.6% (for horticulture) per year. Bouët et al. (2009) point out that “one likely impact of increased crop prices [in general, rather than specifically as a result of the use of biofuels] is an increase in yields”.107

4.3.3. Yield assumptions in modelling work It was concluded in section 3.2.8 that all the work reported here probably uses FAOSTAT as the ultimate source of empirical data on global crop production. Since yields are a function of crop yields and areas cropped, it is assumed that FAOSTAT is also the only source of data for yields. It therefore appear likely that the models do not differ in the account they give of yields in the data year, but only in the rate of yield change that is assumed between the data year/base year and the modelled year. 99 Data for 2007 and 2008 were not available 100 p. 25 101 p. 49 102 p. 775, sourced to Bruinsma (2003) 103 p. 29; referenced to ADAS UK Ltd (2008) 104 Trostle, R., “Global agricultural supply and demand: factors contributing to the recent increase in food commodity prices”, Economic Research Service, U.S. Department of Agriculture, July 2008 105 p. 12; the study also compares yield increase values used in EU-only studies. 106 Bindraban et al., “Can biofuels be sustainable by 2020?”, Wageningen University Research Centre, 2009 107 p. 17

Page 46: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

46

According to UNICA (2009)108, CARB updated the GTAP values on Brazilian sugar cane yields to reflect the 8.2% increase observed between 2001 and the average for the 2006-2008 time period.109 In GLOBIOM it is intended to use “zero “autonomous” technological progress in crop improvement” as the central case.110 LEITAP/Banse et al. use a scenario in which “technological change is high”. It is not stated what this means for agriculture.111

4.3.4. Commentary Differences in the baseline rate of increase of yields matter for two reasons: their effects on the land requirements of additional crop demand due to biofuels and on the land requirements due to the use of crops for other purposes. In terms of the land requirements of additional demand due to biofuels, less hectares of land will need to be converted if the yield from each converted hectare is higher. In the modelling work reported here, the median interval between the yield data year and the modelled year is around 15 years. The lowest estimate of the rate of increase of yields in section 4.3.2 is 0.8% p.a. (Trostle as cited by Searchinger).112 The median “maximum improvement” estimated by ADAS UK Ltd is 2.1%. The former rate of increase implies as 12.7% increase in yields over 15 years while the latter rate of increase implies a 36.6% increase. 21% more land would need to be converted with the lower yield increase estimate than with the higher. This is the minimum effect that such a difference in yield improvement estimates would have on the results of the modelling work. If the relationship between demand and its land use change impact is modelled as non-linear, and if, in particular the unit land use change impact is an increasing function of demand113, then the effect on the results of a difference in yield improvement estimates would be greater because its effect on land requirements due to demand for crops for other purposes would also have to be taken into account. This is because a higher yield means that non-biofuel demand will be satisfied using less land. The land that is then modelled as being used to satisfy the biofuel demand will be different (in quantity and/or character) from the land that would be modelled as satisfying the biofuel demand under a low-yield hypothesis. If the land use change impact of demand is modelled as a linear function of demand, it does not make any difference which land is used to satisfy the biofuel demand. If the land use change impact is modelled as an increasing function of demand, a high-yield hypothesis would mean that the land converted to satisfy the biofuel demand would lead to a lower land use change impact than the land converted under a low-yield hypothesis. This suggests that assumptions about yield improvements matter to the results.

108 p. 19 109 For UNICA’s comment on this see section [4.3.1]. 110 ECORYS (2009) (p. 40) 111 Banse et al. (2007) (p. 7) 112 GLOBIOM’s central-case modelling assumption of “zero “autonomous” technological progress in crop improvement” would seem to denote an even lower rate of 0.0% p.a. (see section [4.2.3]) 113 See chapter 13

Page 47: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

47

Estimates in the literature of the effect of different yield improvement scenarios are, however, lower. RFA (2008) state that “High and low yield improvement scenarios result in approximately +/- 10% influence on total land demand for biofuels”114 – a figure that is lower than the figure (at least 15%) that would be expected from comparing the “business as usual” and “maximum improvement” forecasts in the work from ADAS UK Ltd (cited above) that was commissioned for the RFA report. According to Bouët et al. (2009), “Fernandez-Cornejo et al. (2008)115 incorporated the potential increase in yields into their GTAP-based analysis by running two scenarios, one with baseline projection increases in yields and one with additional productivity gains. The increases in yields did not have major impacts on land use. Corn acreage in the US increased by 18% in the baseline scenario compared to 16% in the increased yields scenario, while sugar acreage in Brazil increased by 52% and 51% respectively.” The implication here is that differences in assumptions about yields affect the estimated demand for land by 2% or less. If it is correct to conclude that yield data do not generally take into account the scope for changes in cropping intensity, this should be addressed.

4.4. Baseline assumptions: food/crop demand IFPRI project an increase of 27% in world crop production between 2010 and 2020.116 A lower crop demand for non-biofuel purposes would have the same effect as higher yields: non-biofuel crop demand could be met from less land. As analysed in the previous section, this would make a difference to the results if the relationship between demand and its land use change impact is modelled as non-linear.

4.5. Baseline assumptions: oil price In AGLINK, oil prices up to 2017 are in the range $90-104/bbl. IFPRI use an IEA projection that shows the oil price reaching $109/bbl in 2020 ($2004).117 ESIM use an oil price of $76/bbl in the base year118, rising to about $80/bbl in 2020.119 A higher oil price means that more biofuels will be consumed, for market reasons, in the absence of measures in favour of biofuels. In modelling work that assumes no measures in favour of biofuels in the baseline, this should make a difference to the modelled quantity of biofuel consumption. It might also make a difference to the modelled quantity of biofuel consumption in modelling work that includes in the baseline some measures in favour of biofuels – depending on the detailed specification of those measures.

114 p. 29 115 Fernandez-Cornejo, J. et al., “Modelling the Global Economic and Environmental implications of biofuels production”, GTAP conference paper, June 2008 [weblink given] 116 Al-Riffai et al. (2010) (p. 45) 117 Al-Riffai et al. et al. (2010) (p. 44) 118 The graph on p. 7 would suggest that the base year is 2005. 119 Tonini and Henseler (2009) (p. 6). It could be that AGLINK and CAPRI use the same assumption.

Page 48: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

48

As analysed in the previous section, this difference in the volume of biofuel consumption in the baseline scenario matters if the model is of a type that depicts the land use change impact as a non-linear function of demand.

4.6. Baseline assumptions: petrol and diesel consumption AGLINK estimate EU consumption at 112.8 Mtoe petrol and 228.2 Mtoe diesel in 2020 (total 341.0 Mtoe).120 This estimate is sourced to PRIMES. These values appear to be exogenous and not to vary between the baseline and policy scenario. IIASA obtain values for transport energy demand from the IEA’s “WEO 2008 Reference Scenario”.121 The latest Commission modelling estimates EU consumption of petrol, diesel and electricity in transport (the denominator for the 10% calculation) in 2020 at 316.3 Mtoe, without taking into account the reductions likely to follow from the recently adopted Cars and CO2 regulation and other EU legislation.122 Differences in transport fuel consumption should make a difference to the modelled volume of biofuel consumption in scenarios where measures in favour of biofuels (in the baseline and/or the policy scenario) are specified in a way that is designed to achieve a given percentage share of biofuel consumption. This difference in the volume of biofuel consumption matters if the model is of a type that depicts the land use change impact as a non-linear function of demand.

4.7. Baseline assumptions: land use

4.7.1. Land use in the baseline in modelling exercises IFPRI state that in the baseline, cropland is expected to increase by about 100 Mha (9%) between 2008 and 2020 (equivalent to about 8.5 Mha per year).123 IFPRI estimate land use in the base year as follows124: (i) Use FAO and AEZ data to determine a quantity of land in each country in the base

year by type (forest/permanent meadows & pasture/arable & permanent crops/other;) and AEZ;

(ii) Distribute GTAP country-level data for quantities of crop production across the

cropland in the country. (Presumably, this step includes two tasks: (a) using yield data to translate production quantities into hectares; (b) using a rule to distribute production of each individual crop across the AEZ types in the country. The source does not fully

120 Gay and Kavallari (2009) (p. 7); [diesel figure corrected – give new source] conversion to Mtoe by Commission services 121 IIASA (2009) (p. 15) 122 PRIMES 123 Al-Riffai et al. (2010) (p. 45) 124 Valin et al. (2009) (pp. 6-7)

Page 49: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

49

explain how either of these two tasks is conducted. It does not explain, for example how yield differences between AEZs are taken into account.)

(iii) “Pastureland is … distributed among different uses using GTAP information

[presumably in relation to quantities of meat and dairy production] assuming that rents are the same for all cattle lands”.

(iv) The area of forest is divided into managed and unmanaged forest. The procedure here

is described as follows: “FAO data relative to forest areas are distributed between managed and unmanaged using data from Sohngen et al. (2007)125 on forest management practice. Tropical and forests with limited accessibility are considered to be unmanaged whereas temperate mixed forests with accessibility and forest plantations are considered to be managed forests. This distinction will be useful for assessing lands’ economic values (unmanaged lands have no economic values at the base year) and to take rents into account in the model. Unmanaged forests also contain more carbon stock that can be released in the case of their destruction.”

(v) For each AEZ in each country, a calculation is made of the amount of land that is not

managed forest, is not cropland, and is suitable (according to a classification made by IIASA) for rainfed crop production. This is defined as “land available for cropland expansion” or “marginal land”.

In earlier work, IFPRI estimated exogenous land use change in the baseline “by considering that land use change for the main land categories (cropland/meadows-pasture/forest) follows the patterns reported in FAO time series. Variation rates are computed using observed variation from 2000 to 2004.”126 (Such a short period does not seem to provide a particularly secure basis for this part of the analysis.) It has not been possible to identify how this is handled in the work reported in Al-Riffai et al. (2010). The baseline scenario, like the policy scenario, also includes endogenous land use change caused by changes in demand for crops.127 In LEITAP/ALTERRA there is substantial abandonment of agricultural land. It is not clear if this refers to the European Union or the world.128 (See also section 9.8.) The effect of these assumptions will depend on how the model in question translates demand into land and land into land types (see chapters 8 and 9). Searchinger et al. assume that unused cropland, if not brought back into use for crop production, would "revert" to forest or grassland.129 They justify this assumption on the basis that "Much of what is considered surplus cropland consists of land brought into crop

125 Cited as Sohngen, B. et al., “Global Forestry Data for the Economic Modelling of Land Use”, GTAP technical paper (2007) 126 Valin et al. (2009) (p. 12) 127 Valin et al. (2009) (p. 12) 128 ALTERRA (2009) (p. 12). A factor in this outcome is likely to be the fact that the baseline scenario envisages stabilisation of world population at 8 billion in 2030 and a shift away from meat eating (p. 12). 129 Searchinger et al. (2008) (p. 14)

Page 50: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

50

production in a minority of years when prices are high. This land comes into and out of production because of fluctuations in price. Although biofuel expansion will increase the average price of crops, prices will continue to fluctuate around a new average equilibrium price… There will remain an "extant margin" of cropland that will come in and out of crop production with price surpluses even as the amount of cropland in production on average rises." Banerjee and Alavalapati (2009) conduct a CGE modelling exercise in which, “In the light of the economic importance of forest sector illegality and in contrast to previous forest sector applications of CGE models, our analysis explicitly considers both legal and illegal forestry and deforestation sectors… [including] the role of deforestation as a supplier of cleared land.”130

4.7.2. Commentary: abandoned agricultural land It seems reasonable for the baseline to make assumptions about the future employment of abandoned agricultural land, rather than automatically treating it as cropland. However, it is not obvious that the assumption should be made, as by Searchinger et al. [and others], that all the land would pass into a higher-carbon-stock use. Another possibility is that at least part of the land can be expected to rotate regularly into and out of crop production, never having the opportunity to accumulate significant additional carbon stocks, and that additional demand (under the policy scenario) affects only the proportion of this rotation time that is spent in rather than out of production. Nor is it obvious that, to the extent that the land would otherwise have passed to a higher carbon stock use, it is correct to assume that the proportion of forest and grassland growth opportunities foregone is equal to their proportion in historic land use change. There is no obvious reason to exclude the possibility that the rate of afforestation is at least partly policy-driven, with cropland-to-grassland conversion as a residual category. If that were true, then the foregone higher carbon stock land would be grassland rather than forest. It certainly does not seem appropriate to automatically discount these possibilities, as Searchinger et al. do.

4.8. Baseline assumptions: agriculture policy CAPRI use a scenario in which the European Union CAP ‘health check’ is not implemented (milk quota remains).131 ESIM use a scenario in which the CAP continues and the ‘health check’ is implemented.132

130 p. 3 131 Blanco-Fonseca and Pérez Domínguez (2009) (p. 11)

Page 51: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

51

LEITAP/ALTERRA assume that “The current CAP export subsidies are abolished… Border support is also phased out. Agricultural income support is reduced to 33%, mainly aiming at maintaining environmental services”. Product quotas are abolished by 2020. Farm payments are fully decoupled and are gradually reduced (by 50% by 2030). Intervention prices are phased out and are abolished by 2030. Compulsory set-aside is abolished by 2020, with “energy crops allowed” (presumably on set-aside land before 2020).133 LEITAP/Banse et al. 134 (with 2001 as the base year) assume that CAP reform is implemented in 2003 with full decoupling. In 2010 there is a 25% reduction of domestic agricultural support135; in 2020 a 50% reduction compared to 2010; and in 2030 a complete elimination of support. Production quotas are abolished in 2020.136 Set-aside is abolished in the EU15 by 2010 and is never introduced in the new Member States. 137 The specification of agriculture policy should, in the models, affect the propensity of increased demand for biofuels in Europe to be met through the use of domestic rather than imported feedstocks, and the propensity of domestic production increases to be achieved through yield increases as opposed to the conversion of one or another type of land. It is difficult to make a prima facie statement on the likely importance of these effects.

4.9. Baseline assumptions: tariff policy In AGLINK import tariffs for ethanol and biodiesel are “in the price transmission function” in the policy scenario, while they are “unchanged” in the baseline scenario.138 LEITAP/ALTERRA use a baseline scenario in which “Tariff barriers restricting market access are gradually removed but international food safety standards are raised and new mechanisms are introduced to ensure high social and environmental production standards of traded goods.” In particular, a slight increase in non-tariff barriers to trade in agricultural products is envisaged.139 It is stated that “Trade restrictions are … considered, leading to higher costs for imports of ethanol from, for example, Brazil”.140 LEITAP/Banse et al. use a baseline scenario in which “the WTO negotiations are successful, global trade fully liberalised”. Trade barriers are reduced by 25% by 2010 compared with 2001; reduced by 50% by 2020 compared with 2010; and abolished by 2030. 141 The specification of tariff policy should, in the models, affect the propensity of increased demand for biofuels in Europe to be met through domestic rather than imported feedstocks. It is difficult to say prima facie whether or not this should be expected to have an important influence on the results.

132 Tonini and Henseler (2009) (p. 6) 133 ALTERRA (2009) (p. 12) 134 Banse et al. (2007) (p. 8) 135 It is not clear if this describes the European Union only or all countries. 136 It is not clear if this describes the European Union only or all countries. 137 p. 12 138 Gay and Kavallari (2009) (pp. 6-7) 139 ALTERRA (2009) (p. 12) 140 ALTERRA (2009) (p. 15) 141 Banse et al. (2007) (pp. 7-8)

Page 52: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

52

4.10. Conclusion Differences in the baseline matter if they are likely to lead to different conclusions about the land use impact of biofuels and bioliquids. It seems reasonable to believe that differences in assumptions about yields would fall into this category and could have a significant effect on estimates of the amount of land that needs to be converted as a result of demand for biofuels. However, the two estimates found in the literature suggest a smaller effect; the reason for this needs to be examined. Modellers may model the relationship between demand and its land use change impact as linear or non-linear (see chapter 13). Where the relationship is modelled as non-linear impacts on the results can be expected from differences in assumptions about: - trends in demand for crops for non-biofuel purposes; - the inclusion in, or exclusion from the baseline of measures favouring the

consumption of biofuels; - trends in the oil price; and - the volume of petrol and diesel consumption. In addition, differences in assumptions about yields can be expected to have a greater impact on the results. The specification of agriculture and tariff policies in the baseline should, in the models, affect the propensity of increased demand for biofuels in Europe to be met through domestic as opposed to imported biofuels/feedstocks, and (in the case of agriculture policy) the propensity of domestic production to be met through yield increases as opposed to the conversion of one or another type of land. It would seem appropriate for the importance of these effects to be assessed through sensitivity analysis. It is clearly necessary to obtain fuller information about the baseline assumptions used in the reported modelling work.

5. Policy scenarios

5.1. Introduction In the modelling work reported here, the policy scenario(s) differ from the baseline scenario in respect of the measures favouring the consumption of biofuels142 that are included, or the level of biofuel consumption that is assumed. The requirement for the Commission to report on land use change impacts is laid down in the Renewable Energy Directive and the Fuel Quality Directive. It would therefore be helpful for modelling to assess the measures in favour of biofuels and bioliquids that are laid down in those Directives. These are evaluated in section 5.2. Section 5.3 summarises the measures in the modelling exercises reported here.

142 None of the modelling reported here covers bioliquids

Page 53: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

53

5.2. Anticipated biofuel/bioliquid consumption under the EU policies whose impact is to be assessed

5.2.1. Renewable Energy Directive The Renewable Energy Directive requires each Member State to achieve a 10% share of renewable energy in transport petrol, diesel and electricity consumption by 2020.143 EU consumption of these transport fuels is estimated at 316.3 Mtoe in 2020 (see section 4.6). This therefore requires a contribution of 31.6 Mtoe from renewable sources. Renewable energy in transport could come from hydrogen (from renewable sources); electricity (from renewable sources) in rail or road transport; or biofuels. The following assumptions are made: - Hydrogen will not make a significant contribution by 2020144; - Electricity in rail transport will account for 7.3 Mtoe in 2020145, of which 2.6 Mtoe

will be counted as coming from renewable sources146; - On a pessimistic assumption, electricity in road transport will not make a significant

contribution in 2020. On an optimistic assumption, it will contribute 0.8 Mtoe.147 The Directive provides that this should be multiplied by 2.5 in assessing whether the 10% target is met; on this basis, electricity in road transport will, on an optimistic assumption, contribute 2.1 Mtoe towards the target.

143 Article 3(4). To be precise, the denominator for the calculation is defined as "petrol, diesel, biofuels consumed in road and rail transport, and electricity [consumed in transport])". The figure of 316.3 Mtoe does not include biogas in road transport (which would make it slightly higher) but does include biofuel use in aviation and maritime transport (which, if excluded, would make it slightly lower). It should be noted that renewable energy consumption in the aviation and maritime sectors counts towards the numerator in evaluating compliance with this target, even though renewable energy and conventional fuel consumption in those sectors does not count towards the denominator. 144 PRIMES gives 0.02 Mtoe for the use of hydrogen from all sources in transport in 2020. 145 From PRIMES, assuming that in the PRIMES baseline all electricity use in transport is in rail. 146 Member States can use either the national or the EU ratio in determining the proportion of electricity to be counted as coming from renewable sources. The EU proportion is expected to be around 35% in 2020. The Member States with most electricity use in rail (France and Germany) are expected to have a share of renewable energy in electricity in 2020 below that figure. As a first approximation, the EU-wide figure is therefore retained for this calculation. 147 Assumptions: cars and light vans account for 50% of vehicle mileage; they last for 10 years; sales of electric cars and light vans account for 1% of vehicle sales in 2011, rising linearly to 20% in 2020 (source: GM/Toyota estimates cited in the Financial Times); annual sales volumes of cars and light vans will be constant; in 2020 electric vehicles will thus account for 10.5% of cars and light vans on the road; electric vehicles will continue to have a constrained range, with the effect that average km driven per electric vehicle in 2002 will be half the figure for other vehicles; electric vehicles will drive 3.5 times as many km per unit of energy as other vehicles; 35% of the electricity they use will be counted as coming from renewable sources. On these assumptions, energy from renewable sources used by road vehicles will be 0.83 Mtoe in 2020. The Directive provides that this quantity should be multiplied by 2.5 in counting towards the 10% target, giving the result of 2.08 Mtoe. While optimistic, the assumption of a 20% of car/light van sales in 2020 is not outside the range of that found in the literature. Kampman et al. (2010) (pp. 22-23), for example, explore three scenarios. In one, electric vehicles would account for 0.4% of car sales in 2020 while plug-in hybrid electric vehicles would account for a further 1.3%. In a second, their shares would be 11% and 24% respectively; and in a third, 40% and 7% respectively.

Page 54: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

54

It is common to distinguish between “first generation” and “second generation” biofuels. The former are made from starches, sugars and vegetable oils. The latter are made from ligno-cellulosic and non-food cellulosic materials. Second generation biofuels count double towards the 10% target, as do first generation biofuels from wastes and residues. On a pessimistic assumption, second generation biofuels will not make a significant contribution in 2020. On an intermediate assumption they will contribute 1.5% of total petrol and diesel consumption, counting as 3 percentage points towards the 10% target. On an optimistic assumption they will contribute 3% of total petrol and diesel consumption, counting as 6 percentage points towards the target.148 It is necessary to assess how much of the raw material for second generation biofuels will be land-using crops (such as farmed grass, farmed wood or forestry wood) and how much will be non-land-using residues (such as straw and sawdust). The second generation biofuel technologies that are closest to market are cellulosic ethanol and BTL (“biomass to liquid” using the Fischer-Tropsch process). Cellulosic ethanol technology uses enzymes to break down the raw material. It is likely that enzymes that can handle ‘soft’ material such as straw and grass will be available before those that can handle ‘harder’ material such as sawdust and, a fortiori, wood. BTL technology can handle any raw material. However, it is further from market than cellulosic ethanol technology. Residues, particularly straw, much of which is currently unused, are likely to be cheaper than crops. It has been estimated that 15.5 Mtoe of straw in terms of primary energy are currently available in the EU25 for making biofuels.149 This could be used to make approximately 5.5 Mtoe of cellulosic ethanol or 7.5 Mtoe of BTL. In the light of the preceding discussion of technological development and price, it is assumed that if second generation biofuels enter the market by 2020, half (up to a ceiling of 5.5 Mtoe) will come from straw and half from crops. The main wastes and residues from which first generation biofuels are made are used cooking oil and animal residues.150 It has been estimated that the potentially available quantities of these resources are 0.95 Mtoe151 and 2.25 Mtoe152 respectively. Since they count double, there will be an incentive to use them. It is assumed that about 1/3 of the available resource, or 1 Mtoe will be used in transport in 2020, counting as 2 Mtoe for the purposes of the 10% target. 153

148 According to IIASA (2009) (p. 17), it is estimated in the reference scenario of the IEA’s World Energy Outlook that second generation biofuels will account for 4% of developed countries’ biofuel consumption in 2020. Under the pessimistic, intermediate and optimistic assumptions presented in the present paper, second generation biofuels would account respectively for 0%, 22% and 55% of EU biofuel consumption in 2020. 149 European Commission (2007) ; conversion ratios back-calculated from the figures given in the source (pp. 8-9). This estimate takes into account the need to leave sufficient straw on the land to regenerate the soil. 150 [where does biogas fit in?] 151 [check reference in IEA report from 2004; 1 Mt converted to Mtoe at ratio of 0.95 – check] 152 According to Coelenbier (2009) [get full reference], [EU27] animal fat production was 2.25 Mtoe in 2008 (2.367 Mt at ratio of 0.95 [check]. [insert reference; also useful to give present-day shares] 153 Smeets (2009) (p. 7) cites data from ICTSD (2009) according to which estimated EU consumption of EU-produced biodiesel from recycled vegetable oil and animal fats in 2008 was 0.20 Mtoe and 0.11 Mtoe respectively (total 0.32 Mtoe). These figures are forecast to grow by 2010 to 0.43 Mtoe and 0.18 Mtoe respectively (total 0.61 Mtoe). The forecast for 2010 (and possibly also the estimate for 2008) is attributed to the US Department of Agriculture. These figures suggest that the assumed contribution of 1 Mtoe in 2020 is, if anything, conservative. This is reinforced by the fact that Coelenbeir (2009) [get full reference] gives a higher figure, (0.34 Mtoe) for animal fats (0.367 Mt at a ratio of 0.95 [check]).

Page 55: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

55

The remainder of the 10% target will have to be fulfilled through the use of first generation biofuels from crops, counting single rather than double. The table on the next page sets out the scenarios that result from these figures. It can be seen that the quantity of crop-using biofuels154 required to meet the 10% target laid down in the Renewable Energy Directive can be estimated to fall somewhere between 10.8 Mtoe (6.0 Mtoe from first generation crops, 4.7 Mtoe from second generation crops) and 27.1 Mtoe (all from first generation crops). Crop-using biofuels would account for between 3.4% and 8.6% of petrol, diesel and electricity consumption in road transport.

154 The word ‘crops’ here includes farmed and forestry wood.

Page 56: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

56

Table – Scenarios for fulfilling the Renewable Energy Directive target for 10% transport fuel from renewable sources Mtoe

Pessimistic assumptions for electric road transport and second generation biofuels

Optimistic assumption for electric road transport, intermediate assumption for second generation biofuels

Optimistic assumptions for electric road transport and second generation biofuels

Consumption of petrol and diesel [in road transport] 316 316 316 Without

multipliers With

multipliers Without

multipliers With

multipliers Without

multipliers With

multipliersTotal required contribution from renewable sources 31.6 31.6 31.6 Hydrogen from renewable sources 0.0 0.0 0.0 0.0 0.0 0.0 Electricity from renewable sources in rail 2.6 2.6 2.6 2.6 2.6 2.6 Electricity from renewable sources in road (multiplier: 2.5)

0.0 0.0 0.8 2.1 0.8 2.1

Second generation biofuels from straw (multiplier: 2) 0.0 0.0 2.4 4.7 4.7 9.5 First generation biofuels from wastes and residues (multiplier: 2)

1.0 2.0 1.0 2.0 1.0 2.0

Total contribution from sources other than crops 4.6 11.4 16.1 Second generation biofuels from crops (multiplier: 2) 0.0 0.0 2.4 4.7 4.7 9.5 First generation biofuels from crops 27.1 27.1 15.5 15.5 6.0 6.0 Total contribution 31.6 31.6 31.6 Total biofuels from crops 27.1 (8.6%) 17.9 (5.7%) 10.8 (3.4%) Source: estimates of the Commission services (see text)

Page 57: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

57

5.2.2. Fuel Quality Directive The Fuel Quality Directive requires fuel suppliers to (i) calculate the greenhouse gas emissions from the fossil fuels they supply in 2010 and (ii) reduce this by 6% by 2020, taking into account all the road transport fuels they supply. The Commission has previously forecast that [half of one of the percentage points] of reductions in greenhouse gas emissions required by the Fuel Quality Directive will be achieved through the use of alternative fuels other than biofuels (compressed natural gas and liquefied petroleum gas).155 In order to calculate the net requirement for biofuels from crops created by the two pieces of legislation, it is necessary to calculate whether the remaining 5.5% reduction will be achieved in full as a result of compliance with the requirements of the Renewable Energy Directive, or whether, after those requirements have been fulfilled, there will remain further work for fuel suppliers to do in order to fulfil the requirements of the Fuel Quality Directive. The following assumptions are made: - The baseline for calculating the required savings is road transport fuel consumption in

2020 (forecast, in the PRIMES scenario used here, at 307.5 Mtoe) - Average unit emissions from the fossil fuels supplied by fuel suppliers in 2010 will be

83.8 gCO2eq/MJ, the initial value assigned to the fossil fuel comparator in the Renewable Energy Directive;

- This figure will not change, as far as petrol and diesel are concerned, between 2010

and 2020 – increased emissions from the use of oil from non-conventional sources will be offset by reduced emissions from refining;

- The same multiplier (2.5) will be used in calculating the contribution of electricity in

road transport for the purposes of the Fuel Quality Directive as for the purposes of the Renewable Energy Directive; each unit of electricity used in road transport will be assessed as delivering a 50% saving relative to petrol and diesel in 2020; and all electricity used in road transport will be taken into account for the purposes of calculating compliance with the requirements of the Fuel Quality Directive.

- Biofuels used in the rail, aviation and maritime sectors are produced by fuel suppliers,

but do not count towards their targets under the Fuel Quality Directive. Fuel suppliers will therefore have a relatively low incentive to develop these markets. It is assumed that these markets will account for no more than 2.5% of the biofuel consumption to fulfil the requirements of the Renewable Energy Directive, and that this will come equally from each type of biofuel.

- First generation biofuels from wastes and residues will deliver average savings,

relative to petrol and diesel, of 88% in 2020; second generation biofuels from straw

155 [reference] [Commission non-paper of September 2008]

Page 58: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

58

will deliver average savings of 87%; and second generation biofuels from crops will deliver average savings of 84.5%156;

- In fulfilling their obligations under the Fuel Quality Directive, fuel suppliers will first

calculate the contribution made by biofuel and electric road vehicle use required for the purposes of the Renewable Energy Directive (since these are contributions that will have to be made in any case) before introducing other measures (for which they would incur additional costs).

The table on the next page shows, for each of the scenarios from section 5.2.1, the results of these assumptions and in particular the average rate of greenhouse gas savings from first generation biofuels from crops that would be sufficient to ensure that no additional measures (potentially including additional biofuel use) would be required to fulfil the requirements of the Fuel Quality Directive.

156 Typical values given in the annexes to the Directives for waste vegetable and animal oil biodiesel; wheat straw ethanol; and farmed wood ethanol/farmed wood Fischer-Tropsch diesel (average of the two); assumed unchanged in 2020.

Page 59: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

59

Table – Scenarios for fulfilling the Fuel Quality Directive target for 6% reduction in unit greenhouse gas emissions from road transport fuels

Mtoe

Pessimistic assumptions for electric road transport and second generation biofuels

Optimistic assumption for electric road transport, intermediate assumption for second generation biofuels

Optimistic assumptions for electric road transport and second generation biofuels

Road transport energy consumption 307.5 307.5 307.5 Required emission reduction (6%)157 18.5 18.5 18.5 Consumption to fulfil Renewable Energy Directive targets (values derived from section 5.2.1; 2.5% of each type of biofuel deducted for consumption in rail, aviation and maritime transport)

road transport electricity 0.0 first generation residues 1.0 second generation straw 0.0 second generation crops 0.0 first generation crops 26.4

road transport electricity (with multiplier) 6.0 first generation residues 1.0 second generation straw 2.3 second generation crops 2.3 first generation crops 15.1

road transport electricity (with multiplier) 6.0 first generation residues 1.0 second generation straw 4.6 second generation crops 4.6 first generation crops 5.9

Savings from fuels other than first generation biofuels from crops, Mtoe equivalent

road transport electricity (50% saving) 0.0 first generation residues (88%) 0.9 second generation straw (87%) 0.0 second generation crops (84.5%) 0.0 other alternative fuels (0.5 percentage points) 1.5 total 2.4

road transport electricity (50% saving) 3.0 first generation residues (88%) 0.9 second generation straw (87%) 2.0 second generation crops (84.5%) 2.0 other alternative fuels (0.5 percentage points) 1.5 total 9.3

road transport electricity (50% saving) 3.0 first generation residues (88%) 0.9 second generation straw (87%) 4.0 second generation crops (84.5%) 3.9 other alternative fuels (0.5 percentage points) 1.5 total 13.3

Remaining savings requirement (18.5 - 2.4) = 16.1 (18.5 – 9.3) = 9.1 (18.5 – 13.3) = 5.2 Required savings rate from first generation biofuels from crops (if Renewable Energy Directive requirements are enough to also fulfil Fuel Quality Directive)

(16.1/26.4) = 61% (9.1/15.1) = 60% (5.2/5.9) = 88%

Extra first-generation biofuels from crops if saving can be no more than in pessimistic scenario

reference case 0 2.7 (total biofuels from crops: 13.4 Mtoe/4.3% of

transport petrol/diesel/electricity) Source: estimates of the Commission services (see text). Arithmetic differences are due to rounding

157 In this table, GHG savings are expressed in terms of the equivalent quantity of emission-free fuel.

Page 60: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

60

It can be seen that under the pessimistic scenario, first generation biofuels from crops would have to achieve an average saving of 61%. Under the second scenario (optimistic for electricity road transport, intermediate for second generation biofuels), the necessary average saving would be about the same: 60%. It is expected that this saving would be achieved158, and the requirements of the Fuel Quality Directive would therefore be fully met through (a) the expected use of alternative fuels other than biofuels and electricity in road transport and (b) compliance with the requirements of the Renewable Energy Directive. However, it can be seen that under the optimistic scenario, first generation biofuels from crops would have to achieve an average saving of 88%. It is not considered that such a saving could be achieved, since this figure exceeds the typical savings currently attributed to all first generation biofuels from crops. Fulfilment of the target laid down by the Fuel Quality Directive would therefore savings to be found over and above those expected from the fulfilment of the requirements of the Renewable Energy Directive and from other alternative fuels. Since the contributions available from other sources have already been fully taken into account, it is considered that this contribution would have to come from first generation biofuels from crops. At a savings rate of 61% (the rate required under the first scenario), this would require an extra 2.7 Mtoe of biofuels from crops.

5.2.3. Conclusion In order to assess the land use change impact of the biofuels that will be used to fulfil the requirements of the Renewable Energy Directive and the Fuel Quality Directive it is necessary to calculate the volume of those biofuels that will come from crops. This has been analysed in sections 5.2.1 and 5.2.2. The table summarises the results.

158 This assessment is based on the fact that by 2020 all biofuels will be required to achieve a saving of 50%; biofuels from installations in which production started in or after 2017 will be required to achieve a saving of 60%; typical savings for several first generation pathways, as laid down in the Directives, are already close to or above 60% (sugar beet ethanol: 61%; maize ethanol: 56%; sugar cane ethanol: 71%; sunflower biodiesel: 58%; palm oil biodiesel with methane capture at oil mill: 62%; pure vegetable oil from rapeseed: 58%); and the Commission has identified significant scope for improvement in the greenhouse gas savings of first generation biofuels. [reference: non-paper of autumn 2008]

Page 61: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

61

Table – Biofuels from crops (Mtoe 2020): requirements of the Renewable Energy Directive and Fuel Quality Directive

Mtoe

Pessimistic assumptions for electric road transport and second generation biofuels

Optimistic assumption for electric road transport, intermediate assumption for second generation biofuels

Optimistic assumptions for electric road transport and second generation biofuels

Requirements of Renewable Energy Directive first generation biofuels from crops

27.1 15.5 6.0

second generation biofuels from crops

0.0 2.4 4.7

total 27.1 17.9 10.8 Additional requirements of Fuel Quality Directive

first generation biofuels from crops

0.0 0.0 2.7

Requirements of the Directives in combination first generation biofuels from crops

(27.1 + 0.0) = 27.1 (17.9 + 0.0) = 17.9 (6.0 + 2.7) = 8.7

second generation biofuels from crops

0.0 2.4 4.7

total 27.1 17.9 13.4 share of road transport fuel

8.6% 5.7% 4.2%

It can be seen that the total amount of biofuels from crops required to fulfil the requirements of the Directives in 2020 is estimated to lie the range of 13.4 – 27.1 Mtoe, equivalent to between 4.2% and 8.6% of road transport fuel consumption. The contribution required from crops is lower where electricity in road transport and/or second generation biofuels play a significant part. Even where they do not, electricity in rail transport and first generation biofuels from wastes and residues ensure that the required share is less than 10%. It should be noted that neither Directive lays down intermediate targets for the transport sector (though the Renewable Energy Directive lays down a trajectory for renewable energy use in general).

5.3. Policy scenarios in the modelling work reported on

5.3.1. Policy scenarios concerning EU biofuel use In AGLINK import tariffs for ethanol and biodiesel are “in the price transmission function” in the policy scenario, while they are “unchanged” in the baseline scenario.159

159 Gay and Kavallari (2009) (pp. 6-7)

Page 62: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

62

The policy scenario in CAPRI is “with European Union Biofuel Directive in year 2020”. The demand for biofuels, and its split between ethanol and biodiesel, is exogenous.160 The central policy scenario in IFPRI is the middle one of the three shown in the table in section 5.2.3. The other two scenarios in that table are also modelled in IFPRI's work.The split between bioethanol and biodiesel is fixed at 45%:55%. 161 The policy scenario in ESIM is “a 7% first generation biofuel target by 2020”.162 When the jurisdiction of interest for LEITAP/ALTERRA is the EU, the policy scenario is an EU “biofuel mandate”. This is set at the level of 10% of final energy consumption in transport.163 The policy scenario in a recent LEITAP/Banse et al. modelling exercise is described below under the heading ‘global biofuel use’. An earlier LEITAP/Banse et al. exercise had two policy scenarios with “5.75% obligatory blending” and “11.5% obligatory blending” in each Member State in 2010. The denominator was not stated in the reference, nor was the quantity of biofuels to which these percentages were assumed to equate.164 It appears that these policy scenarios did not differ from the baseline in relation to biofuel policy elsewhere in the world: “Under the policy scenarios only the mandatory blending obligation within the European Union are changed. All other policy instruments remain unchanged compared to the reference scenario.”165

5.3.2. Policy scenarios concerning US biofuel use In a separate exercise from the main IFPRI results reported here, Valin (2009) reports on “the assessment of a biofuels mandate for transportation in the United States”.166 The paper does not give further details of this policy scenario. The policy scenario in Searchinger et al. is US consumption of 56.1 Mtoe of ethanol in 2016. The use of a single feedstock (maize), produced in the US, is assumed.167

5.3.3. Policy scenarios concerning global biofuel use AGLINK sets biofuel policy exogenously and determines biofuel consumption endogenously. In AGLINK’s 2008 study, the policy scenario is “a continuation of the 2007 policies” (in favour of biofuels).168 This scenario is modelled as leading to biofuel consumption in the EU, US and Canada of about 48.2 Mtoe in 2013-17 (35.3 Mtoe ethanol and 12.8 Mtoe biodiesel).169 160 Blanco Fonseca and Pérez Domínguez (2009) (p. 4). [It is understood that the same exogenous split between ethanol and biodiesel is used in AGLINK and CAPRI.] 161 Al-Riffai et al. (2010) (pp. 45, 67). 162 Tonini and Henseler (2009) (p. 2) 163 ALTERRA (2009) (pp. 14, 15) 164 Banse et al. (2007) (p. 8) 165 Banse et al. (2007) (p. 8) 166 p. 1 167 Searchinger et al. (2008) (p. 5), conversion to Mtoe by Commission services 168 von Lampe (2009) (p. 11) 169 von Lampe (2009) (p. 11); back-calculated from percentages and converted to Mtoe by Commission services.

Page 63: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

63

GLOBIOM has three policy scenarios. In all three, global biofuel consumption is fixed exogenously (based on a POLES model result) at 104 Mtoe in 2010, 153 Mtoe in 2020 and 277 Mtoe in 2030. In the central scenario170, these figures include respectively 3, 12 and 112 Mtoe of second generation biofuels. In the “first generation” scenario, all new production above 2005 values is first generation biofuels. In the “second generation” scenario, all new production above 2005 values is second generation biofuels. IIASA has four policy scenarios. In the first and second, biofuel use grows in accordance with the IEA’s “WEO 2008 reference scenario”. Depending on the scenario, second generation biofuels either become gradually available after 2015, or are not available up to 2030. In the third and fourth scenarios, it is assumed that “mandatory, voluntary or indicative targets for biofuels use announced by major developed and developing countries will be implemented by 2020, resulting in about twice the biofuels consumption compared to WEO 2008”. Depending on the scenario, second generation biofuels either become gradually available after 2015 (accounting for 4% of biofuel use in developed countries and none in developing countries in 2020), or are deployed rapidly (accounting for 33% of biofuel use in developed countries and 3% in developing countries in 2020).171 When the jurisdiction of interest is the world, the LEITAP/ALTERRA policy scenario “includes biofuel policy in OECD and EU” in the form of mandates.172 The policy scenario in LEITAP/Banse et al. is described as “a mandatory minimum share of biofuels in total fuel consumption in the transport sector of 10 per cent per Member State by 2020”173 and as “10% share of biofuel consumption in transportation by 2020”.174 The policy scenario also includes “implementation of biofuel initiatives outside Europe”.175

5.3.4. Commentary176 Most of the modelling exercises reported here, namely AGLINK, CAPRI, IFPRI, ESIM, GLOBIOM, IIASA, LEITAP/ALTERRA, LEITAP/Banse et al. and Searchinger et al., exogenously fix the consumption of biofuels in the policy scenario. By contrast, AGLINK exogenously fixes measures favouring biofuels (biofuel consumption itself is endogenous). In practice, this distinction is probably not important, because in order to determine the types of biofuel consumed, exercises in the first group must also fix the measures that will be used to achieve the requirement. It is not clear whether the choice of measure (notably as between blending obligations placed on fuel suppliers on the one hand and subsidies on the other) matters. It seems likely that it does matter in the case of CGE models, where obligations will tend to increase fuel prices and therefore reduce consumption. Some modelling exercises exogenously fix the types of available biofuels that will be used.177 CAPRI, AGLINK, ESIM and IFPRI fix the split between ethanol and biodiesel. Searchinger 170 In the vocabulary of the study, this is described as the “Baseline” scenario, while the baseline scenario in the sense of the present paper is called the “No Biofuels” scenario. 171 IIASA (2009) (pp. 15, 17) 172 ALTERRA (2009) (pp. 7, 11, 14). For the EU component, see section [5.3.2]. 173 Banse et al. (2008) (p. 119) 174 Banse (2008) (p. 11) 175 Banse (2008) (p. 11) 176 [NB since this was drafted, new information has been added on AGLINK and Tyner et al.] 177 This is a different matter from estimating to what extent second generation biofuels will be commercially available in the modelled year.

Page 64: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

64

et al. determine that the stated biofuel requirement will be met entirely from US-produced maize ethanol. If biofuels/bioliquids are heterogeneous in their land use change impact, this fixing is liable to make a difference. It should be justified and should preferably be in a sensitivity test, not the central policy scenario. In general, descriptions of modelling exercises do not make it clear how the contributions of different types of biofuel are determined. It seems probable that this is fully endogenous in few if any cases. If the land use change impact of biofuels does not vary linearly with demand, the absolute size of the modelled biofuel consumption in the policy scenario matters. This information is often not given. If size matters in this way, it is particularly important in the current context to depict accurately the level of biofuel consumption required (in 2020) by the recently adopted EU legislation. This required level was evaluated in section 5.2. In all the modelling exercises where the choice made is clear, the policy scenario exaggerates the amount of biofuels from crops that will be needed. This exaggeration has two origins. First, the EU’s 10% target should be calculated as a share of road transport fuel - but LEITAP/ALTERRA and LEITAP/Banse et al. calculate it as a share of fuel used in all transport sectors. (IFPRI, CAPRI, AGLINK and ESIM calculate it correctly.) Second, some types of biofuel count twice towards the target; non-crop-using biofuels from wastes and residues are certain to make a contribution178; and electricity used in rail and road transport also counts towards the target. This means that the total share of biofuels from crops will be no more than 8.6%. However, most of the models set the requirement for biofuels from crops at 10% (this is true of AGLINK, CAPRI, ESIM, LEITAP/ALTERRA and LEITAP/Banse et al.).179

6. Models

6.1. Choices to be made In the literature it is commonly suggested that an important dichotomy to be addressed is that between economic models on the one hand and agricultural or environmental models on the other. For instance, the authors of the REFUEL study distinguish between “agri-economic” and “agro-technical” approaches to analysing the relationship between biofuels and agriculture; Sands et al. distinguish between the use of “economic models” and “biophysical models” in the analysis of alternative climate policies; and Valin distinguishes between “economic models” and “agronomic and ecological models” as approaches to modelling land use change. The distinctions drawn have something in common with that made here between scenario analysis on the one hand and potential analysis on the other – except that these authors present the approaches as alternatives that must be chosen between or reconciled. Thus the REFUEL authors conclude that potential analysis gives optimistic results, scenario analysis gives pessimistic results, and “as often, reality will probably lie somewhere in-between”. Sands et al. argue that ideally, economic models should be linked to biophysical

178 Although it is never stated explicitly, it seems clear that all the modelling work reported here treats “biofuels”, certainly “first generation biofuels” as a synonym for “crop-using biofuels”. 179 It is assumed here that AGLINK and ESIM share CAPRI’s assumption that second generation biofuels will contribute 3% of road transport energy in 2020.

Page 65: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

65

models. Valin states that there are many initiatives under way to link the two types of model, but highlights the risk of inconsistencies in doing this.180 In the present context, however, these approaches are not real alternatives. All scenario analysis in this field needs some kind of model of land, its potential uses, its expected performance if devoted to the production of different crops, and its carbon stock when used in different ways (alongside, of course, economic analysis to predict the choices that resource-constrained actors will make). Similarly, all potential analysis has, and has to have, an economic dimension – none is simply an estimate of maximum technical yields in a world without resource constraints. Lywood (2009b) distinguishes between ‘macro-economic models’, ‘spreadsheet-based models’ and ‘allocation methods’ for determining greenhouse gas emissions from indirect land use change.181 The distinction Lywood draws between macro-economic and spreadsheet-based models appears to be one of degree (of complexity) rather than of kind. In the language of this literature review, they are both forms of scenario analysis models. Indirect land use change effects, to the extent that they actually occur, naturally involve processes having a fair number of steps and potential feedback loops (for example, biofuel demand à crop prices à demand for land à land prices à crop yields). A thorough investigation of the phenomenon needs to be open to such links. It does not therefore appear appropriate to opt against complexity in principle. It is unlikely that any commentator would advocate opting against simplicity in principle. Therefore, this particular distinction does not appear to pose a fundamental choice to be addressed in this chapter. Complex scenario analysis models can be divided into general equilibrium (CGE) models and partial equilibrium (PE) models. They are discussed in section 6.2, which looks into This section looks into (a) the factors that are taken into account in CGE and PE models and (b) the models’ relationship to reality. Allocation models answer a different question from scenario analysis models. While scenario analysis models answer the question "what is likely to be the land use change impact attributable to a policy of promoting the consumption of biofuels", allocation models answer the question, what is the land use change impact of the consumption of agricultural commodities (or biofuels) in general. Allocation models are explored in section 6.3.

6.2. General and partial equilibrium models

6.2.1. Types of model used in the work reported here Among the modelling exercises reported here, AGLINK, CAPRI, EPA, ESIM, GLOBIOM, IIASA and Searchinger et al. use partial equilibrium models. CARB, IFPRI, Tyner et al., LEITAP/ALTERRA and LEITAP/Banse et al. use computable general equilibrium models.

180 REFUEL (n.d.) (p. 3); Sands et al. (n.d.) (p. 1); Valin (2009) (p. 3) 181 p. 2

Page 66: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

66

6.2.2. Factors only taken into account in CGE models For Beckman and Hertel (n.d.) the advantage of CGE models is “their ability to simulate potential impacts of prospective economic policies taking into account inter-sectoral and international interaction”.182 Valin et al. (2009) state that “Researchers [on biofuels] require integrated framework to take into account both agricultural and energy markets and their interactions, as well as emissions impact and climate change feedback. For this purpose, computable general equilibrium models are particularly appropriate as they explicitly incorporate the linkages between sectors… partial equilibrium studies tempted more precise assessment but they lack important substitution and revenue effects that play a role for this type of assessments.”183

6.2.3. Factors only taken into account in PE models These are considered to give a more complex and sophisticated depiction of the agricultural sector.

6.2.4. Factors not taken into account in either family of models Demand for raw materials for biofuels and bioliquids can be expected to alter land prices and the demand for labour in rural areas. It would seem reasonable to expect a relationship between the state of the rural economy (including land prices and demand for labour) and the rate of urbanisation in developing countries/periurbanisation in developed countries. Urbanisation and periurbanisation are known to have consequences for greenhouse gas emissions.184 This is not taken into account in the models. The models calculate demand by country for quantities of agricultural commodities. It is not clear whether the translation of this demand into hectares of land employed for crop production is endogenous. It is clear that the translation of hectares of land employed into particular types of land converted – a key issue in the modelling of land use change – is not endogenous. Instead, it is handled through an independent modelling process without feedback loops to the main model of the economy. The point of using complex models is precisely to be able to take into account such feedback loops. Econometrica (2009) argue that CGE and PE models “do not take account of multiple factors in land use change decisions (e.g. land tenure, land price speculation, market information, proximity of roads)”.185

6.2.5. Relationship between results and reality: CGE models Beckman and Hertel (n.d.) state that “CGE models … are often criticized as being insufficiently validated. Key parameters are often not econometrically estimated, and the performance of the model as a whole is rarely checked against historical outcomes”. Their paper conducts such a validation of one particular result – the estimated cost of climate change mitigation. They conclude that it has been underestimated by 36% because the price

182 p. 1 183 pp. 2-3 184 [reference] 185 p. 4

Page 67: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

67

elasticity of demand for energy has been overstated in the GTAP-E186 database. They call for “further rigorous examination of key parameters in other CGE models”.187 In similar vein, Hertel (n.d.) states, “There are many parameters in the GTAP model, and they are all uncertain to some degree. Economics profession and policy-makers under-invest in estimation of these parameters; GTAP can’t change this.”188 Hertel (n.d.) goes on to state, “Virtue of GTAP model is that it is easy to see which parameters are most important to results”.189 Decreux and Valin (2007) appear to make a similar point where they suggest that the merit of CGE modelling is that it attributes numerical values to each of the main impacts of policy. They do not make strong claims about the accuracy of these values.190 Elsewhere, Valin (2009) underlines that the linkage made between land and economic modelling in the IFPRI work is a “very flexible and fully consistent tool BUT more simplistic representation”. These are therefore “Exploratory tools”.191 Al-Riffai et al. (2010) et al. highlight "important uncertainties with regard to a number of behavioural parameters in the model", the critical role of yield response and land elasticities, and the poor quality of the EU27 data available for the modelling work.192 Lywood (2009) states that “Although the database and source code for the GTAP model is available on the internet, this does not provide the data needed to validate the models.”193 While Kløverpris et al. (2008) imply that GTAP models are equally appropriate for the assessment of changes in “demand, supply, policy, etc.”194, UNICA (2009) state that “CGE models are used to perform analysis of policy instruments (e.g., taxes and subsidies), technological changes, and changes in resources supply. It is uncommon to find in the CGE literature demand shocks, as implemented by CARB.”195

186 see section [3.3.2] 187 pp. 1, 4 and 20; “The elasticities of substitution between petroleum and other fuels are too high, as is the consumer demand elasticity for petroleum products in many countries. In addition, supply response in the petroleum sector appears to be too large.” (p. 25) 188 p. 6 189 p. 6 190 “A model is no more than the quantified expression of a number of well-identified, robust mechanisms. The relevant point is about the way it is used. CGE models simulations are not an ending point, that would give a definitive answer to the question of the impact of a given trade policy decision. It is on the contrary a starting point making it possible, based on (often complex) protection scheme changes [i.e. changes in the specification of trade policy instruments], to deliver a synthetic numbering of their main impacts. The interpretation then requires a well-suited analysis, taking into account the problems tackled, and the important mechanisms not included in the model.” (p. 21) The authors further state that “Such an analysis makes it possible to put forward the main mechanisms, to give their sign and their order of magnitude”. (p.8) – although if only the ordinary of magnitude of quantities is known, it is not clear how modellers can be confident of the sign of results that derive from the subtraction of one quantity from another, as is the case in scenario modelling in general and a fortiori when weighing the greenhouse gas savings from the use of biofuels and bioliquids against carbon stock losses from land use change. (Emphasis added.) 191 p. 3 192 pp. 12-13 193 p. 7 194 p. 19 195 p. 13; this point, if justified, is of course equally applicable to all uses of CGE models for the evaluation of biofuel policies.

Page 68: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

68

It may be that construction of a functioning model is such a difficult task that there is little scope to adapt a model, once it functions, to meet additional demands such as calibration. The same appears to be true for the establishment of confidence intervals. No information appears to be available that would allow an assessment to be made of whether apparent differences in land use change impacts between crops or locations of production are significant.

6.2.6. Relationship between results and developing country reality: CGE models

Decreux and Valin (2007) state that “similar structural models are applied to different economies. Doing otherwise would be difficult, in a world-wide model devoted to various applications. Nonetheless, this is a strong assumption, and it could be useful to implement more flexible options to describe specific economic mechanisms, in particular for developing countries.”196 New Fuels Alliance (2008) state that “Lands in developing countries without clear rents (economic values in a marketplace) cannot be analysed in GTAP. This includes much one-time cropland that is not accounted for or included in the GTAP estimates of effects.”197 However, while this is appears to be a correct statement about the CARB modelling using GTAP, it is overcome to some extent by the strand of GTAP work, including IFPRI, that uses the AEZ approach. As described in section 3.2, these models rely on FAO data on crop production and – when different land uses are included, as required when assessing land use change – on land use. ADAS UK Ltd (2008) suggest that the questionability of the accuracy of these data is greatest in relation to developing countries.198

6.2.7. Relationship between results and reality: PE models In the context of AGLINK, Gay and Kavallari (2009) state that “Historic information on biofuels is not sufficiently available to allow calibrating market behaviour”.199 By contrast, von Lampe (2009) expresses confidence in the biofuel elements of the AGLINK model (pointing to “detailed biofuel policies” and “different biofuel chains in various countries”) while highlighting the “approximative” character of its results on indirect land use change and, even more so, the related greenhouse gas emissions.200

6.2.8. Conclusion This literature review suggests the following conclusions:

196 p. 21 197 p. 3 198 see section [3.2.2] 199 p. 14 200 p. 18

Page 69: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

69

(1) Among the modelling exercises reported here, inter-sectoral factors that could reasonably be thought relevant to an assessment of the land use change impact of policies favouring the use of biofuels and bioliquids are only taken into account in CGE models;

(2) Other relevant factors, however, have been taken into account in some PE models and

have not yet been systematically included in CGE models; (3) A further set of relevant factors have not yet been taken into account in either family

of models; (4) There is no particular reason to believe that CGE models’ depiction of the

relationships between the factors they include is especially compatible with the relationships that obtain in reality;

(5) There is reason to believe that the depictions provided by CGE models are closer to

the reality of developed countries than to that of developing countries. Proponents seem to suggest that CGE models offer two contributions that PE models do not: “inter-sectoral and international interaction”, in the words of Beckman and Hertel (n.d.)201. In assessing the phenomenon of indirect land use change it is essential to investigate the extent to which changes in demand in one country lead to changes in land use in another. If PE models do not and cannot handle international interaction properly, it is difficult to see it being possible for the Commission to dispense with appropriately specified CGE work in fulfilling its task. However, it is not apparent that the PE work reviewed here performs systematically less well on this dimension than the CGE work: the results (e.g. of FAPRI and GTAP, in the US context) are simply different (notably, they model trade in different ways). It is less obvious how important it is to this work to model interactions between agriculture markets and other sectors of the economy in a complex way, although this question too would bear further investigation.

6.3. Allocation models Lywood (2009b) distinguishes between ‘macro-economic models’, ‘spreadsheet-based models’ and ‘allocation methods’ for determining greenhouse gas emissions from indirect land use change.202 Allocation methods, of which the work reported in Econometrica (2009) is an example, calculate the total amount of land use change or carbon stock change attributable to agriculture and divide it between agricultural commodities in proportion to some characteristic (economic value, energy value or – as in the Econometrica study – mass). They provide an answer to a different question (what is the land use change impact of the consumption of agricultural commodities (or biofuels and bioliquids) in general?) rather than

201 p. 1 202 p. 2

Page 70: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

70

the question addressed by scenario analysis (what is likely to be the land use change impact attributable to a policy of promoting the consumption of biofuels/bioliquids?).203

7. Model calculations: crop output

7.1. Introduction The last chapter described the selection of appropriate models. The previous two chapters described the establishment of baseline and policy scenarios. These tasks having been done, scenario analysts’ next step is to use the model to calculate the difference between the two scenarios in terms of the quantities of crops produced. This is not just a matter of calculating the amount of crops needed to make the additional quantities of biofuel required by the policy scenario, for two reasons. First, the production of most biofuel crops necessarily entails the accompanying production of co-products, many of which – notably DDGS (Distillers Dried Grains and Stillage) from ethanol crops, pressed beet slices from sugar beet ethanol and meal from biodiesel crops - are usually used as animal feed. These co-products substitute for other crops which would otherwise have been grown to make this animal feed. These substituted crops need to be deducted in calculating the difference between the scenarios in terms of quantity of crops produced. Given that soya production in Brazil is growing rapidly, partly on land that was previously forest, a particularly important question is whether biofuel co-products substitute for soya meal from Brazil. Second, the increased demand for crops caused by the promotion of biofuels and bioliquids is liable to lead to an increase in their price. This is liable to lead to a reduction in demand for crops for non-energy purposes. In principle this, too, needs to be deducted in calculating the difference between the scenarios in quantity of crops produced (and should be reported as a modelling result). This chapter focusses on the first of these issues: co-products.

7.2. Co-products

7.2.1. Empirical evidence: quantity of co-products per unit of biofuel/bioliquid

Energy basis The table shows the quantities and proportions (in energy terms) of biofuels/bioliquids and land-saving co-products that can be calculated from data in the JEC well to wheels study 204 as being produced per ton of feedstock and – for comparison – those that can be calculated from data in Ensus (2008) and Lywood et al. (2009). Only co-products that are generally used as animal feed are included, because – by substituting for the dedicated production of animal 203 As pointed out in section [4.4.5], none of the scenario analysis modelling reported here estimates the land use change impact of biofuels per se. All the exercises depict the impact of policies to promote the consumption of biofuels. 204 http://ies.jrc.ec.europa.eu/WTW

Page 71: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

71

feed – these have the potential to save land. Co-products that are used for energy purposes (such as bagasse from sugar cane), usually not used (such as straw from wheat) or used for industrial purposes (such as glycerine from biodiesel) are not included. Co-products used in these ways affect the economics of biofuel/bioliquid production and the greenhouse gas savings from biofuels and bioliquids (they are taken into account in the discussion of these savings in chapter 11) but do not save land. Table – Production of biofuels and co-products suitable for use as animal feed: tons of oil equivalent per ton of fresh feedstock205 (Note: calculation on energy basis) J: calculations of the Commission services from data in JEC well to wheels study206 E1: calculations of the Commission services from data in Ensus (2008)207 E2: calculations of the Commission services from data in Lywood et al. (2009)208

toe

biofuel co-products

total co-product share

ethanol feedstocks sugar beet

J: 0.046 E1: 0.050 E2: 0.050

J: 0.026 E1: 0.016 E2: 0.028

J: 0.072 E1: 0.065

E2: 0.078

J: 36% E1: 24% E2: 36%

wheat

J: 0.183 E1: 0.206 E2: 0.205

J: 0.117 E1: 0.087 E2: 0.098

J: 0.300 E1: 0.293 E2: 0.303

J: 39% E1: 30% E2: 32%

sugar cane

J: 0.046 E2: 0.041

-

J: 0.046 E2: 0.041

0%

maize

E1: 0.211 E2: 0.214

E1: 0.085 E2: 0.092

E1: 0.296 E2: 0.306

E1: 29% E2: 30%

biodiesel feedstocks rape seed

J: 0.342 E1: 0.376 E2: 0.345

J: 0.233 E1: 0.135 E2: 0.141

J: 0.575 E1: 0.511 E2: 0.486

J: 40% E1: 26% E2: 29%

sunflower J: 0.367 J: 0.191 J: 0.558 J: 34% palm oil fruit bunches

J: 0.193 E2: 0.177

J: 0.010 E2: 0.008

J: 0.203 E2: 0.185

J: 5% E2: 4%

soya

J: 0.161 E1: 0.143 E2: 0.148

J: 0.308 E1: 0.203 E2: 0.230

J: 0.469 E1: 0.347 E2: 0.378

J: 66% E1: 59% E2: 61%

It can be seen that for ethanol feedstocks, while the results of Ensus' original work (E1/Ensus (2008)) diverged substantially from those of JEC, the company's more recent values (E2/Lywood et al. (2009) are close to those of JEC, except that a higher proportion of the energy contained in wheat is considered to be capable of conversion to biofuel.209 By contrast, 205 (expression used by Ensus: “te raw grain”. 206 http://ies.jrc.ec.europa.eu/WTW 207 p. 44. Data for co-products are for the energy value metabolisable by ruminants. Energy value metabolisable by non-ruminants is between 10 and 75% less (Lywood et al. (2009), p. 19). 208 pp. 17 and 19; data for co-products are for the energy value metabolisable by ruminants. Energy value metabolisable by non-ruminants is between 10 and 75% less. 209 This could be because Ensus's data are for feed wheat rather than milling wheat. [Check whether that is also the case with JEC]

Page 72: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

72

both sets of Ensus results show yields per ton of crops for biodiesel feedstocks that are lower than those given by JEC, but with higher proportions of the yield (and, except for soya, higher absolute amounts) being usable in the form of biofuel. In accordance with the Commission’s established practice, the JEC figures will be used in calculations elsewhere in this paper. Working on an energy basis, Özdemir et al. (2009) give the same figure as that calculated from the JEC data for soya oil from soya: 0.161 toe/t.210 However they give a different figure for the co-product (soya oil cake or meal): 0.211 toe/t, implying a yield of 0.372 toe/t and a co-product share of 57% rather than 66%.211 Mass basis CE Delft (2008) give figures for “amount of by-products per unit of crop”.212 The figures are shown in the table. The authors add that “Jatropha was ignored because the varieties currently considered for biodiesel production produce a press cake toxic for livestock. It is in practice often returned to the plantation for utilization as green manure.”213 Table – Share of co-products per unit of crop feedstock for biofuel production (figures from CE Delft) (Note: calculation on mass basis) crop share of co-products

ethanol feedstocks wheat 32.3% maize 30% cassava 2% sorghum 2%

biodiesel feedstocks soya 83% palm 2% sunflower 60% rapeseed 57% Source: CE Delft (2008)214 Özdemir et al. (2009) state that on a mass basis, 81% of soya goes to meal and 19% to oil.215 Basis of economic value Taheripour et al. (2008) state, “According to our calculation about 16 percent of a corn based dry milling ethanol plant’s revenue comes from DDGS sales. Corresponding shares for typical rapeseed and soybean based biodiesel producers are about 23% and 53%, respectively.” Özdemir et al. (2009) state that on the basis of economic value, 60% of soya is attributable to meal and 40% to oil.216

210 Vegetable oil and biofuel content are generally taken to be identical [source] 211 p. 2944 (conversion from GJ to toe by Commission services) 212 p. 3 213 p. 6 214 p. 3; p. 18 makes it clear that the figures are expressed on a mass basis 215 p. 2994 216 p. 2994

Page 73: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

73

7.2.2. Empirical evidence – capacity of co-products to substitute for animal feed

Total amount of co-products that the market could absorb According to CE Delft (2008), “DG’s [distillers grains] were until recently viewed as a less suitable feed for non-ruminants. However, new dry milling plants seem to produce a (far?) more digestible product that yields comparable digestive energy compared with corn.”217 A detailed estimation of incorporation limits of different co-products for different animals is given.218 The authors estimate the global potential for the use of DDGS in cattle diets in 2015 at 156 Mt, comparing this with a maximum likely global production [in 2020] of 80 Mt. They estimate the global potential for the use of rape seed meal in pig and poultry diets in 2016 at 75-105 Mt, comparing this with a maximum likely global production [in 2020] of 50 Mt. As a check, they compare current consumption of protein meals for feed and conclude that in all scenarios examined, the total amount of co-products is still less than 50% of total protein demand.219 Lywood et al. (2009) state that “advised inclusion limits for DDGS in EU animal feed allow a large increase in DDGS use”220 and, reviewing the literature, that “The potential market for DDGS produced in the EU is therefore of the order of 21 million tonnes at current advised incorporation rates, with potential incorporation volume up to 38 million tonnes in the current compound feed market”.221 Taheripour et al. (2008) state, “Of course, as with any feedstuff, there are limits to the amount of DDGS that can be fed to livestock. However, Cooper (2005) and Dhuyvetter (2005) have reported two estimates: 42 million tons and 52 million tons, respectively, of the potential demand for DDGS within the US. These numbers are significantly larger than the current production of DDGS within the US – suggesting that the maximum ration may not be an issue in the near future. In addition, the potential market overseas is even further from satiation.”222 According to Tyner et al. (2009), [commenting on US maize co-products], “some papers have shown that biofuels byproducts can be used in livestock industry more extensively [than previously thought]”.223 Substitution rates Air Improvement Resource (2008) state that “Recent Argonne detailed analysis shows 1 lb [pound] of DG [DDGS] replaces 1.28 lbs of feed… The meal replaced consist of 0.95 lbs of corn and 0.28 lbs of soy meal … Using most recent Argonne analysis, DG land use credit increases from 33% to 71%”.224 217 p. 10 218 pp. 10-11 219 pp. 13-14 220 p. 8 221 p. 10 222 p. 13 [The references are G. Cooper, “An update on foreign and domestic dry-grind ethanol co-products markets”, National Corn Growers Association (2005) and K. Dhuyvetter et al., “The U.S. ethanol industry: where will it be located in the future?”, Agricultural Issues Centre, University of California (2005) 223 p. 5. [Citations: Erin Daly 2007, Arora 2008.] 224 p. 8

Page 74: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

74

The table shows the substitution ratios given by CE Delft (2008). Table – Substitution rates between animal feeds – data from CE Delft (2008)

mass basis (t/t)

substitute for soya meal

for wheat for maize

consumption by pigs rape meal 0.72 0.23 maize DDGS 0.28 0.97 wheat DDGS 0.28 1.05

consumption by cattle rape meal 0.60225 0.28 maize DDGS 0.61226 0.41227 wheat DDGS 0.72 0.27 rates used for analysis (the authors assume 50% of co-products go to cattle and 50% to pigs) rape meal 0.66 0.26 maize DDGS 0.45 0.69 wheat DDGS 0.50 0.66 Source: CE Delft (2008)228 Analysis of the data229 shows that these figures are cumulative – that is, that one ton of rape meal, for example, if consumed by pigs, substitutes for 0.72 tons of soya and for 0.23 tons of maize. Ensus assume that the protein value of biofuel co-products is 80% of that of soya meal.230 They state, “The effectiveness of the DDGS protein is dependent on the amount of degradation that occurs in the fermentation and DDGS drying process. With improvements in technology, it is likely that the digestibility of the DDGS protein can be increased to be similar to that of soy meal.” As a result, “the DDGS would displace more soy meal”.231 Lywood et al. (2009) state that “DDGS protein has a lower and more variable nutritional value than soy meal protein due to a higher fibre content, lower levels of some essential amino acids (EEAs) … and due to amino acid degradation, particularly lysine, in the bioethanol conversion process. This makes little difference when DDGS is fed to ruminants, but limits the effectiveness of DDGS protein in feeds for mono-gastric animals such as pigs and poultry … EU animal feed compounders overcome the major limitations of DDGS protein by addition of synthetic essential amino acids… and overcome digestibility limitations by using higher crude protein levels than when using soy meal. EEAs are manufactured by biosynthesis and do not require any land area… In the US where localized feed protein surpluses depress protein market value, animal feed co-products are often used directly in local feed lots rather than compounded and EEAs are not routinely added.”232 They add, “Feed compounders … will use the protein and energy in biofuel co-products to minimize

225 Source cited by the authors for comparison: USB, 0.60-0.70 (p. 12) 226 Source cited by the authors for comparison: GHGenius, 0.60 (p. 12) 227 Source cited by the authors for comparison: GHGenius, 0.68 (p. 12) 228 pp. 11 and 18 229 tables on p. 18 230 Ensus (2008) (p. 13) 231 p. 29 232 pp. 10-11

Page 75: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

75

nutritionally equivalent amounts of protein and energy in relatively more expensive raw materials such as soy meal and cereals, the marginal sources of protein and energy in current compound feed formulations. In this way, although DDGS may not fully replace soy meal in any particular animal feed formulation, it will on aggregate replace a mix of soy meal and cereals across all compound animal feed formulations where it is used”.233 The substitution rates used in Lywood (2009) are shown in the table. Table – Substitution rates between animal feeds – data from Lywood (2009)

mass basis (t/t) substitute for soya meal for wheat consumption by pigs

rape meal 0.56 0.18 maize DDGS 0.38 0.44 wheat DDGS 0.62 0.41 oil palm expeller 0.11 0.45 sugar beet pulp - 0.60

consumption by poultry rape meal 0.61 0.06 maize DDGS 0.38 0.50 wheat DDGS 0.58 0.40 oil palm expeller - - sugar beet pulp - -

consumption by ruminants rape meal 0.66 0.21 maize DDGS 0.43 0.56 wheat DDGS 0.62 0.41 oil palm expeller 0.16 0.81 sugar beet pulp - 1.01

rates used for analysis (weighted in proportion to the livestocks' shares of the [feed? compound feed?] market)

rape meal 0.61 0.15 maize DDGS 0.40 0.49 wheat DDGS 0.59 0.39 oil palm expeller 0.13 0.61 sugar beet pulp - 0.78 Source: Lywood (2009)234, rounded to two decimal points It can be seen that the data given in Lywood et al. (2009) differ in several ways from those given in CE Delft (2008). For Lywood et al., all 5 types of co-product substitute for soya meal and wheat; for CE Delft, 2 of the 3 considered (rape meal and maize DDGS) substitute for soya meal and maize. For Lywood et al., substitution rates for rape meal, maize DDGS and wheat DDGS are broadly similar for the 3 livestock types considered. CE Delft give similar figures for rape meal; but they see the balance of substitution from cereal DDGS in pig feed tipping much more towards wheat, while in cow feed the opposite is true; poultry are not considered. The rates that the two teams use for analysis, constructed by averaging

233 pp. 11-12 234 p. 19

Page 76: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

76

substitution rates for different livestock types, are closer together. In 5 of the 6 cases where both teams give a figure, the figure given by CE Delft is higher (the exception is soya substitution by wheat, where the figure of Lywood et al. is higher). Taheripour et al. (2008) state that “DDGS substitutes for corn and soybean meal in livestock rations but mainly for corn”.235 They do not give a source for this assessment.

7.2.3. Empirical evidence – what source of animal feed is substituted? Ensus state, “When crops such as wheat, maize and rape are used to make biofuels, only part of the crop is used. The co-product, containing all the crop protein is critical for the animal feed supply chain … Indeed, as the co-product protein will be in a more concentrated form as a result of the manufacture of the biofuel, it can and will displace soy meal…The optimum level of protein in animal feed e.g. for cattle is around 20%236, whereas cereal grains have a level of protein of 8-13%237 and oilseed meals such as soy and rape have protein concentrations of 35-45%238. Currently therefore cereals need to be blended with imported oil seed meals to give the optimum protein concentration. However when biofuels are produced from cereal crops, such as wheat and maize, the non starch part of the grain is concentrated into a co-product – distillers dried grain and solubles (DDGS), at a concentration of 30%-38%239. Therefore, DDGS can be used to replace some of the soy meal for blending, to lift the protein concentration to that required for animal feed.”240 Lywood (2009) states, "In the US, the marginal animal feed components are corn exports and soy meal exports. In the EU, the marginal animal feed components are wheat exports and soy meal imports… In the US substantial quantities of corn DDGS are used as liquid feeds or other direct feeds in local feedlots and so mainly displace corn feed. In the EU most of the DDGS is dried and used in formulated animal feed. The costs of high protein animal i.e. imported soy meal In the EU are substantially higher than energy feeds such as wheat. Therefore animal feed compounders maximise the use of DDGS to displace soy meal to minimise the overall cost of the feed. The limitations of DDGS protein quality due to low levels of some essential amino acids (EEAs) are overcome by addition of synthetic EEAs. The price of DDGS as with other co-products used in EU animal feed adjusts so that they will be fully utilized for animal feed. It may therefore be assumed that in the EU biofuel co-products will displace a mixture of soy meal and wheat to give the same digestible energy and digestible protein level of the animal feed."241 The table shows data given in Lywood (2009c) for trends in trade in meals for animal feed. Table - Trends in trade in meals for animal feed (Mt/year)

Mt/year

Global exports: average 2006-2008

Global exports: growth 1998-2008

235 p. 6 236 Lywood et al. (2009) state, “Target protein levels in compound feeds are typically between 15-30% depending on the animal and feeding regime (CWG, 2009)”. (p. 7) [reference is CWG, “Compound feed specifications”, 2009] 237 Lywood et al. (2009) give 9-13% (p. 7) 238 Lywood et al. (2009) give 35-55% (p. 7) 239 Lywood et al. (2009) give 25-35%. 240 p. 12 241 p. 8

Page 77: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

77

soya 102.1 50.1 rapeseed 7.1 3.0 sunflower seed 4.0 0.1 fish meal 2.7 0.6 total 115.9 (soya share: 88%) 53.8 (soya share: 93%) Source: Lywood (2009c)242, where the source is cited as the online database of the US Department of Agriculture’s Foreign Agricultural Service It can be seen that between 2006 and 2008 soya meal accounted for 88% of exports of the 4 meals for which data are given, and for 93% of the growth in exports of these 4 meals between 1998 and 2008. Lywood states, “it may therefore be concluded that soy meal is the marginal global source of protein meals to meet the demand for animal feed”. Lywood et al. (2009) state that Brazil and Argentina account for almost 90% of EU soya imports243 and add that “the increase in global soy bean land area and exports of soy meal over the last ten years has been predominantly from South America”.244 Similarly, CE Delft (2008) state, “given - The amounts of by-products produced according to the … scenarios [for global biofuel

consumption in 2020] - Current and prognosed amounts of different protein sources applied in feeds there is actually little other possibility than that the larger part of the by-products volumes compete with soy meal. All other protein sources are much smaller in volume than the considered volumes of by-products… It seems reasonable to assume that DG’s [Distiller’s Grains] and RSM [Rape Seed Meal] will primarily substitute soy meal.”245 It should be noted that these statements of Lywood (2009c) and CE Delft (2008) apply to the protein component of feed. It is clear that CE Delft consider that co-products also substitute for non-protein components (embodied in cereals)246; the opinion of Lywood on this point is not clear.

7.2.4. Treatment of co-products in modelling exercises Tyner et al. list some PE modelling exercises in which the “importance of incorporating byproducts is well recognized”.247 AGLINK does not take into account sugar beet co-products.248

242 pp. 16-17 243 p. 8 244 (p. 9); see also Lywood (2009) (p. 15) 245 pp. 14-15 246 see table above 247 p. 5 [Citations: Tokgoz et al. 2007, Tyner and Taheripour 2008, Babcock 2008] 248 [comment from M. Poinelli; reference]

Page 78: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

78

In CAPRI, “By-products (oil cakes, gluten) are used as feed, their prices are endogenous, feed composition can change according to cost minimisation while protein/energy requirements are met”.249 Air Improvement Resource (2008) states that CARB, in its model as it stood on 16 October 2008, “assumes DGs [DDGS] replace only corn meal, and on a lb [pound] for lb basis. This results in 33% land use credit for DGs”.250 Lywood (2009) state similarly that "CARB do not account for the higher protein value of the biofuel co-products and simply substitute them for cereals on a weight basis". CE Delft (2008) assess how taking into account co-products will reduce the estimated land requirements for biofuels.251 The following assumptions are made252: - DDGS from US-produced wheat substitutes for US-produced wheat and soya; - DDGS from US-produced maize and rapeseed meal from US-produced rapeseed

substitute for US-produced maize and soya; - DDGS from EU-produced wheat and wheat produced in the rest of the world

substitute for locally produced wheat and for soya imported from Latin America; - DDGS from EU-produced maize and maize produced in the rest of the world, and

rapeseed meal from EU-produced rapeseed and rapeseed produced in the rest of the world, substitute for locally produced maize and for soya imported from Latin America.

The authors explain that “It was assumed local cereal production will be substituted, because both the EU and North America – the two area’s where the bulk of the considered amounts by-products would be produced – are (almost) self supporting in terms of cereal cultivation for feed application.”253 The rates of substitution used are those calculated by the authors.254 They state, “soy bean figures have been corrected for the fact that 1 tonne of bean yields approximately 0,83 tonne of meal (and hull)”. Study of the data suggests that this correction has been carried out by multiplying the amount of substituted crop production by (1/0.83).255 In other words, if biofuel co-products replace 1 ton of soya meal, they are assumed to replace (1 x 1/0.83) = 1.2 tons of soya beans. Forecast soya bean yields for 2020 are given as 3.3-3.5 t/ha256, thus the replacement of 1 ton of soya meal with biofuel co-products appears to be assumed, by CE Delft, to avoid the use of about (1.2/3.4) = 0.35 ha of land. This procedure does not appear to be correct. If biofuel co-products replace 1 ton of soya meal, it is true that it can be argued that a certain amount of land would then no longer need to be cultivated for soya. However, as well as the soya meal production from that land (replaced 249 Blanco Fonseca and Pérez Domínguez (2009) (p. 3) 250 p. 8 251 [need to identify the policy scenarios – source E4Tech] 252 p. 18 253 p. 17 254 see section [7.2.2] 255 calculations of the Commission services from the data on p. 18 256 p. 19

Page 79: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

79

by the biofuel co-product), the associated production of soya oil would be lost. The oil would not be replaced by the biofuel co-product and would need to be produced in some other way. The purpose of the correction factor, as used for example by Özdemir et al. (2009)257, is to reduce the calculated land benefit to take into account this requirement for some compensating production of vegetable oil. Thus, it would appear more correct to multiply the amount of substituted crop production by 0.83 rather than 1/0.83. 1 ton of replaced soya meal would then equate to the replacement of 0.83 tons of soya beans, and would, under CE Delft’s method, avoid the use of (0.83/3.4) = 0.24 ha of land. If this analysis is correct, it would appear that CE Delft’s figures for the avoided requirement for soya bean land are overstated by about 45%. IFPRI take co-products into account.258 Ensus calculate the land area required for biofuel crops, taking into account co-products. In Ensus (2008) the authors assume that the protein value of biofuel co-products is 80% of that of soya meal.259 Although numerous input data are given260, it is a little difficult to discern the steps in the calculation. The basis for dividing the substituted products between soya and cereal feed is also not clear. It appears that land used for producing the substituted product is divided between animal feed and biofuel components on a mass basis. Lywood et al. (2009) carries out a similar procedure using new input data, and gives a clearer explanation of the method as follows261: - the amount of biofuel and animal feed produced per hectare of feed crop is determined

using biofuel process yield data; - substitution ratios are calculated for each co-product and for each animal group based

on digestible energy availability and protein digestibility values; - substitution ratios for each animal group are weighted by EU compound feed mass use

per animal group, giving an average substitution ratio for cereal and soy meal replacement by co-products in EU animal feed;

- “net area required per hectare biofuel feedstock” is calculated by subtracting land area

which would otherwise be needed to produce soy beans and wheat for animal feed, using the calculated substitution ratios.

The substitution rates used are shown in section 7.2.2. The “allocation problem” for soya is dealt with by calculating the amount of biofuel that could have been produced from the soya oil whose production is “lost” alongside the substituted soya meal, and subtracting this quantity of biofuel from the amount considered to have been produced.

257 [check that Ö et al. do in fact use it in this way] 258 Al-Riffai et al. (2010) (pp. 20, 33, 56) 259 p. 13 260 pp. 44-45 261 pp. 12-13

Page 80: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

80

The authors criticise "the assumption [in several macro-economic studies] that wheat DDGS replaces wheat in EU compound animal feeds on a mass basis" as "ignoring its valuable protein content". It is implicitly assumed in this Ensus work that yields do not change in response to demand, so that the increased demand for crops for biofuels must entirely be met by land conversion. Lywood et al. (2009)262 state that this is because the analysis of co-product effects is intended as a contribution to broader work rather than being intended to give definitive effects in its own right. According to Lywood (2009), the EPA models "can account for biofuel co-products, but appear to use exogenous displacement ratios, instead of using economics to determine displacement ratios."263 GLOBIOM264 includes a “livestock production system approach” with “14 systems”. Inputs and outputs are a function of species, production system and region, calculated by the “RUMINANT” model.265 Feed ratios are “defined in terms of” grazing, stover, cut & carry, occasional and grains. The reference does not give more detail. Tyner et al. include by-products in the GTAP model.266 They claim to be the first authors to have done this in a CGE model.267 At least in the sensitivity test reported in Taheripour et al. (2008), this is done through the incorporation in version 6 of the GTAP database of two new databases, GTAP-BIOA and GTAP-BIOB. The first includes three new commodities (in addition to the 57 in the basic database): “Ethanol 1” from coarse grains; “Ethanol 2” from sugar cane; and “Biodiesel” from oilseeds. The second includes two further commodities: “DDGS” (a byproduct of ethanol 1) and “Meals” (a byproduct of biodiesel).268 In the authors’ “earlier model”, [biofuel co-products were used as inputs by] as single “livestock” sector.269 In their “new model”, this is broken down into three livestock industries: “Dairy Farms”, “Other Ruminant” and “Non-Ruminant”.270 DDGS is modelled as a substitute for “cereal grains (mainly corn)”. Meals are modelled as a substitute for “feedstuffs produced by the food industry (OthFoodPdt)”.271 It can be seen that the work of Tyner et al. does not break down GTAP’s single category of ‘oilseeds’.272 Since the share of animal feed co-products (meal) produced from biodiesel crops varies from 5% for palm oil to 66% for soya273, this means that its depiction of co-products in the biodiesel sector is not likely to be adequate. (It matters which oilseed crop is used.) No

262 p. 6 263 p. 8 264 Havlík et al. (2009) (p. 18) 265 [sourced to “Herrero”] 266 Tyner et al. (2009) (p. 5). [Also cited are Taheripour et al. 2008 (treated here as a sensitivity test of Tyner et al.) and Birur et al. 2008] 267 Taheripour et al. (2008) (p. 2) 268 Tyner et al. (2009) (p. 8); Taheripour et al. (2008) use the term “biodiesel by-product” (BDBP) rather than “meals”. 269 p. 15 270 p. 17 271 Taheripour et al. (2008) (p. 11) 272 See section [3.3.2] 273 Measured in terms of energy value; see table in section [7.2.1]

Page 81: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

81

data are available documenting Tyner et al.’s assumption for the amount of meal produced per unit of oilseeds.274 Kim et al. “adopt the assumption from the GREET model that 1 kg of DDGS displaces 0.95 kg of dry corn grain, 0.3 kg of dry soybean meal, and 0.03 kg of nitrogen in urea”.275 However they note that “A report published by the U.S. EPA … shows that 1 kg of DDGS displaces 0.5 kg of dry corn grain and 0.5 kg of dry soybean meal”. This ratio is used in a sensitivity test.276 The effect is to reduce the estimated greenhouse gas benefit over 100 years by about 4%.277 LEITAP/Banse et al. do not take co-products into account, according to Tyner et al. (2009).278 Özdemir et al. use standard values for yield per hectare to calculate the land requirements attributable to EU-consumed biofuels. (This equates to assuming that changes in demand have no effect on yields: see chapter 8). They deduct standard values for the yield per hectare of substituted animal feed products to calculate the effect on the total land requirement of taking co-products into account. They assume that oil meal co-products replace soya meal from Brazil, and that DDGS and pressed beet slices replace maize from the EU.279 In Searchinger et al. (2008) it is assumed that each unit of maize used to make ethanol is accompanied by the production of DDGS which replaces 1/3 of a unit of maize. Thus, they assume that DDGS from ethanol made in the US from US maize replaces US-produced maize used as animal feed.280 ADAS UK Ltd (2008a) comment, “Searchinger’s ‘pound for pound’ substitution of feed corn for corn diverted to bioethanol appears to be an overestimate (ES15, Wang and Haq, ECCM). It should be recognised that the co-product, dried distillers grains (DDGS), has ~30% protein and ~5% fibre. If heat damage is avoided, maximum inclusion in diets can be ~400 g/kg for cattle, 200 g/kg for sheep, and 100-250 g/kg for non-ruminants (Cottrill et al. 2007). Thus the displacement value of DDGS is at least 23% higher than that assumed by Searchinger et al. (Klopfenstein et al. 2008) … Allowing for the higher protein of DDGS, and also for land to replace the oil foregone (we assumed palm oil); we calculate that Searchinger’s assumption about doubles the land required to substitute for US corn-ethanol. Ensus (ES15) conclude the assumption trebles the results, but they do not account for the ‘lost’ oil from the displaced soya.”281

274 [need to read Taheripour et al. (2008) thoroughly to see what it adds to the account of Tyner et al. given in this and subsequent sections on co-products] 275 Kim et al. (2008) (p. 8) 276 Kim et al. (2008) (p. 11). [The reference is EPA, “Regulatory Impact Analysis: Renewable Fuel Standard Program”, EPA420-R-07-004, 2007.] 277 Kim et al. (2008a) (p. 19); comparison of Scenario G and Reference. 278 p. 5 279 p. 2988 280 Searchinger et al. (2008) (p. 5) 281 pp. 4-5. [The sources referred to in this passage are ES (Evidence Submission) 15 to the Gallagher review, by Ensus; Wang, M. and Z. Haq, letter to Science, 14 March 2008; ES 13 by ECCM; Cottrill, B. et al., “Opportunities and implications of using the co-products from biofuel production as feeds for livestock”, HGCA Review 66 (2007); and Klopfenstein, T. et al., “Use of Distillers’ By-Products in the Beef Cattle Feeding Industry”, forthcoming in Journal of Animal Science (2008)]

Page 82: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

82

Searchinger (2009) gives a different figure from Searchinger et al. (2008): "[A] hectare's worth of ethanol saves 1.8 tons of carbon dioxide, but the process generates distillers grains that provide an amount of animal feed equivalent to four tenths of a hectare of corn according to some recent analysis."282 This implies that taking co-products into account would reduce the amount of land needed by 40%, rather than by 1/3 as in Searchinger et al. (2008).

7.2.5. Modelling results – what source of animal feed is substituted?283 From the data given in CE Delft (2008), the estimates shown in the table can be calculated: Table - Reduction in land requirement for biofuels when co-products are taken into account - data from CE Delft

offsetting reduction in land requirement due to use of co-products

biofuel feedstock

production location

: in EU (cereals)

: in US (cereals)

: in Latin America (soya)

: total

wheat EU 21% 21% 42% maize EU 21% 26% 47% rapeseed EU 6% 31% 38% wheat US 35% 35% maize US 61% 61% rapeseed US 35% 35% Source: CE Delft (2008) and calculations of Commission services284 It can be seen, inter alia, that CE Delft’s prediction of the offset associated with the co-products from US maize is 61%; and that while they predict that land savings from US biofuel production will be located in the US, they predict that most of the land savings from EU biofuel production will take the form of reduced soya production in Latin America. Ensus estimate the present-day impact of taking co-products into account when EU-grown crops are used to make biofuels for consumption in the EU. The results are shown in the tables. Table – Reduction in land requirement for biofuels when co-products are taken into account – data from Ensus (2008)

282 [The reference given is Klopfenstein, T. et al., "Board Invited Review: Use of Distillers Byproducts in the Beef Cattle Feeding Industry, Journal of Animal Science (2008)] 283 This section describes endogenous results. Exogenous assumptions on the topic are documented in section [7.2.4]. 284 Data used: co-product share (by mass) in amount produced (p. 3); t of substituted product per t of co-product (p. 18); forecast yields/ha in 2020 (p. 19). NB the cereals sections of the yields table are headed “wheat DDGS” and “corn DDGS”. It has been assumed that these are full-crop yields per ha and not co-product-only yields per ha. Land divided between soya oil and meal on a mass balance basis (as in the source) but with error corrected so that total land replaced is multiplied by 0.83, not divided by 0.83 as in the source (see section [7.2.4]). Rapeseed yields (not given in the source) were assumed to be the same as those of soya.

Page 83: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

83

offsetting reduction in land requirement due to use of co-products

biofuel feedstock

production location

: cereals : soya : total wheat EU 14% 69% 84% maize EU 17% 61% 79% rapeseed EU 10% 53% 63% Source: Ensus (2008)285 Table - Reduction in land requirement for biofuels when co-products are taken into account – data from Lywood et al. (2009)

offsetting reduction in land requirement due to use of co-products

biofuel feedstock

production location

: cereals : soya : total wheat EU 13% 81% 94% maize EU 18% 56% 74% rapeseed EU 5% 56% 61% sugar beet EU 65% - 65% palm oil S E Asia 4% 3% 7% Source: calculations of Commission services from Lywood et al. (2009)286 For the products and production locations considered by both teams (EU wheat, maize and rape seed), it can be seen that the land savings from co-products substituting for cereal feed anticipated by Ensus (2008) are of similar size to those anticipated by CE Delft, while those from substituting for soya are much greater, especially for wheat. Tyner et al.’s policy scenario is a 6.25% biofuel market share in the EU and 28.5 Mtoe of ethanol consumption in the US in 2015.287 Data in Taheripour et al. (2008), as analysed in section 7.2.6, show that the inclusion of co-products in the modelling exercise (seemingly using Tyner et al.’s “earlier model”, with a less sophisticated depiction of the livestock sector) reduces the modelled required output increase by 50% in the EU, 27% in the US and 12% in Brazil. This suggests (tentatively) that the substituted animal feed is modelled as being locally produced and (at any rate) that the marginal animal feed consumed in the EU and US is not seen as coming from Brazil. This reading of the data is reinforced by the fact that the modelled reduction in the increase in Brazilian oilseed production – used for animal feed - is less, at 10% (0.5 Mtoe) than the reduction in the increase in Brazilian agricultural production as a whole.288 However, this reading is not borne out by the description of the results in Taheripour et al. (2008). Here it is explained that “In the presence of by-products, EU uses its own DDGS and BDBP [biodiesel by-products] and imports some by-products to from the US to support its own livestock industry. As a result, it does not need to allocate more land to meet the demand for grains used in its livestock industry. Instead, it allocates additional land to produce more oilseeds to support its biodiesel production… [T]he model with biofuel by-products predicts higher growth rates for oilseeds outputs in both US and EU and a lower growth rate in Brazil.” 285 Ensus (2008) (p. 13). NB these figures were obtained by measuring off a graph; those for wheat are confirmed by the text. 286 pp. 17 and 21 287 See section [5.3.3] 288 Analysis by the Commission services of data from Taheripour et al. (p. 19). For details see section [7.2.6]

Page 84: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

84

Tyner et al. (2009)289 gives results from the same authors’ “new model” with increased sophistication in depiction of the livestock sector.290 These results show how livestock’s diet changes between the baseline and policy scenarios.291 They are summarised in the table. Table – Changes in livestock diet between baseline and policy scenarios – results from Tyner et al. (share of diet, units not stated)

DDGS oilseed meals coarse grain total EU – dairy farms 1% à 5% (+4) 3% à 7% (+3) 12% à 8% (-4) 16% à 19% (+3)EU – other ruminants 1% à 10% (+9) 6% à 11% (+6) 11% à 3% (-8) 17% à 24% (+7)EU – non-ruminant 0% à 1% (+1) 3% à 7% (+4) 17% à 17% (-1) 21% à 25% (+4)US – dairy farms 3% à 9% (+6) 5% à 5% (0) 16% à 10% (-6) 25% à 24% (0) US – other ruminants 5% à 14% (+9) 3% à 3% (0) 13% à 3% (-9) 21% à 21% (0) US – non-ruminants 1% à 2% (+1) 6% à 6% (0) 39% à 36% (-3) 46% à 44% (-1) Source: Calculations of the Commission services (NB measured off graphs) from Tyner et al. (2009)292. Apparent arithmetic errors are attributable to rounding. As in the simpler model, the effects of availability of biofuel co-products as animal feed appear to be “local”. Increased availability of oilseed meals as a co-product of biodiesel consumption in Europe leads to increased consumption of these meals by animals in Europe, but does not affect their consumption in the US. It can also be seen that DDGS and coarse grain seem to be modelled as substitutes – an increase in consumption of the one is accompanied by a similar-sized reduction in consumption of the other. By contrast, increased consumption of oilseed meals is depicted as replacing some other part of the animals’ diet. It is not stated what this is. It should be pointed out that this modelling exercise does not include the main crops used in the EU to make ethanol (wheat and sugar beet).293 While these data do not completely rule out the possibility that biofuel co-products replace, in part, soya from Brazil, they contain nothing to suggest that this is in fact what is depicted.

7.2.6. Modelling results – what difference does it make to include co-products?

Some studies calculate the land use impact of taking biofuels into account on the basis of individual crops; others make the calculation for the overall difference between the policy scenario and the baseline scenario. Impact for individual crops 289 p. 18 290 see section [7.2.4] 291 It can be assumed that these results are for the scenario “with byproducts”. 292 p. 18 293 see section [7.2.4]

Page 85: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

85

According to data provided in CE Delft (2008), the estimated land requirement for EU-produced biofuel feedstocks should be reduced by 42% for wheat feedstocks, 47% for maize and 38% for rapeseed to take into account land savings associated with co-products. The equivalent figures for US-produced feedstocks are 35% for wheat, 61% for maize and 35% for rapeseed.294 Ensus estimate the present-day impact of taking co-products into account when EU-grown crops are used to make biofuels for consumption in the EU. For wheat ethanol, the amount of land conversion required is estimated to be reduced by about 84%; for maize ethanol, by 79%; and for rapeseed biodiesel, by 63%.295 They explain this result as follows: “The total protein yield per hectare of wheat and maize is comparable with that of soy, and the fermentation process produces additional protein by growing yeast. Therefore little or no additional land is required to produce biofuels from wheat and maize, after taking into account the land saved by not having to grow soy.”296 Using a similar method with slightly different data, the following estimates can be derived from Lywood et al. (2009): Table – Reduction in land requirement for biofuels when co-products are taken into account – data from Lywood (2009) biofuel feedstock

production location offsetting reduction in land requirement due to use of co-products

wheat EU 94% maize EU 74% rapeseed EU 61% sugar beet EU 65% palm oil S E Asia 7% Source: calculations of Commission services from Lywood et al. (2009)297 Sensitivity tests are carried out on the estimate for wheat ethanol. The figure of 94% assumes that DDGS is consumed by all types of livestock (ruminants, pigs and poultry). Assuming that DDGS is consumed by only one type of livestock would reduce the saving, at the worst, to 85%. However, "if wheat DDGS is assumed to replace wheat on a mass basis, as is assumed in several macro-economic models", the reduction in the land area would fall to 33%.298 Impacts for overall policy The Gallagher review concludes that “The EU 10% target contributes a gross land requirement of between 22 million hectares and 31.5 million hectares. The lower and higher estimates relate to assumptions about yield and 2nd generation fuels. These are reduced to a net land requirement of between 8 and 12 million hectares when the potential avoided land

294 For details of the calculation see section [7.2.5]. Note that an apparent mistake in the source, which would exaggerate the land savings when soya meal is the substituted product, has been corrected. 295 Ensus (2008) (p. 13). NB these figures were obtained by measuring off a graph; those for wheat are confirmed by the text. 296 Ensus (2008) (pp. 12-13) 297 pp. 17 and 21 298 p. 24

Page 86: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

86

use benefits of co-products are taken into account.”299 According to these estimates, taking co-products into account thus reduces the estimated land requirement by 62-64%. However, since these results are based on those of CE Delft reported above, it seems better to rely on that underlying work. Havlík et al. (2009) have a graph showing the result of including co-products in the GLOBIOM modelling exercise. The table shows the results. Table – “Net deforestation due to biofuel expansion” (Mha, approx.), with and without the taking into account of biofuel co-products, in GLOBIOM

Mha

co-products not taken into account

co-products taken into account

reduction when co-products taken into account

100% of “2020 liquid biofuel projection”

15.6 14.3 -8.1%

150% 29.2 25.2 -13.7% 200% 45.0 37.9 -15.9% Source: Calculations of the Commission services (including measuring off a graph) from Havlík et al. (2009).300 It is not clear what the units in the first column mean – perhaps biofuel consumption as a proportion of the level in the baseline scenario. It can be seen that this study suggests that the taking into account of co-products reduces the land use change impact (as far as it takes the form of deforestation) by 8-16%, and that the reduction rises as the modelled level of biofuel consumption rises. Özdemir et al. (2009) calculate the land needed for a 10% share of biofuel in European Union road transport fuel consumption in 2020 with and without taking co-products into account. The table shows the results. Table – Land requirements for a 10% share of biofuel in EU road transport fuel consumption in 2020 as calculated by Özdemir et al. (2009)

Mha Co-products ignored

Co-products taken into account

Reduction in required land when co-products are taken into account

Share of domestic production in meeting target: 100% 18.5 14.3 23% 75% 21.1 15.2 28% 40% 20.4 12.9 37% Source: Özdemir et al. (2009)301; calculations of Commission services The authors explain that for soya meal (assumed to be substituted by biodiesel co-products), land was allocated between the oil and the meal on a mass basis. The use of economic or energy allocation would have pointed to a smaller reduction in the required land, because the 299 p. 32 [source is Ecofys 2008, probably “Land use requirements of different EU biofuel scenarios in 2020”]. CE Delft’s work (reported in sections [7.2.4 and 7.2.5] is an input to this exercise. It is not clear how this result relates to lower figures reported in CE Delft (2008a) (p. 29), which was an input to the work of the Gallagher review. 300 p. 36 301 p. 2990

Page 87: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

87

proportion of the land used for soya production allocated to the meal would then be 60% or 57% (respectively) rather than 81%.302 Taheripour et al. (2008)303 and Tyner et al. (2009)304 give results from Tyner et al.’s policy scenario of a 6.25% biofuel market share in the EU and 28.5 Mtoe of ethanol consumption in the US in 2015.305 The impact of model runs with and without co-products taken into account is shown. From the placement of these results in the presentation by Tyner et al., it appears that the “with byproducts” model run was made using the authors’ “earlier model” and does not make use of the more sophisticated depiction of the livestock sector in the “new model” that this presentation also describes.306 The results are expressed in the form of percentage changes in output of four crop groups (coarse grains, other grains, oilseeds and sugar cane) in three jurisdictions (EU, US and Brazil). Changes are modelled relative to a base year (2006). Data for the jurisdictions’ production of the crop groups in question in 2006 were not given in the sources and were therefore obtained from FAOSTAT. They were translated into Mtoe using the conversion factors in the table in section 7.2.2.307 The results are shown in the table. Table – Modelled output of crops with and without co-products taken into account in Tyner et al. 302 p. 2994; it should however be noted that calculations from the JEC well to wheels study suggest that soya meal accounts for 66%, rather than 57%, of the energy content of soya (see section [7.2.1]. CE Delft (2008) also use a mass-based ratio in this context (p. 18). 303 p. 3 304 p. 16 305 see section [5.3.3] 306 see section [7.2.4] 307 The values in the ‘total’ column were used. It was assumed that the value given in the table for wheat could appropriately be used for all cereals. For more on this method see chapter 13 [section on historical analysis]. The values used were those derived from the well to wheels study, not those derived from Ensus (2008)

Page 88: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

88

Mtoe (approx.)

coarse grains

other grains oilseeds sugar cane total

2006 308 EU 42 39 14 0 95 US 84 17 42 1 145 Brazil 14 4 25 21 64 total 140 60 81 22 303 2015: Policy scenario without taking co-products into account EU 43 35 21 0 99 US 98 15 45 1 159 Brazil 13 4 30 23 71 total 155 55 96 24 329 2015: Policy scenario: co-products taken into account EU 41 35 21 0 97 US 93 15 46 1 155 Brazil 13 4 30 23 70 total 147 55 96 24 346 Source: Calculations of the Commission services from data in Taheripour et al. (2008)309 If co-products are ignored, the model suggests that the policy scenario will require an increase of crop production of about 26 Mtoe in the three jurisdictions relative to 2006. If co-products are taken into account, this requirement falls by 27% to 19 Mtoe. In proportionate terms, the biggest effect of not taking co-products into account is in the European Union, where the modelled output increase is then twice as big it is when co-products are taken into account (the production increase required falls by 50%). In the US, the required production increase falls by 27% when co-products are taken into account; in Brazil, by 12%.310 Taheripour et al. (2008) also gives results for land use change in the three jurisdictions.311 Again, these are expressed in percentage terms and therefore had to be complemented with data on absolute levels of land use (area harvested) in the base year (2006), which were obtained from FAOSTAT. If co-products are ignored, the model suggests that land for cereals, oilseeds and sugar cane in the three jurisdictions will increase from 198.8 to 202.7 Mha, an increase of 3.9 Mha. If co-products are taken into account, this requirement falls by 43% to 2.2 Mha. The biggest effects are in the EU (where the land requirement falls by 1.3 Mha) and to a lesser extent Brazil (where it falls by 0.6 Mha). In the US, the land requirement is depicted as actually being higher (by 0.2 Mha) in the scenario where co-products are taken into account. It can be seen that in this model, taking co-products into account has a bigger effect on land use change (reducing the impact by 43%) than on output (where the impact is reduced by 27%).

308 Source: FAOSTAT; calculations of Commission services 309 p. 19 310 In interpreting this description it should be borne in mind that not all countries and crop groups are included. However, according to data given in Taheripour (2008) (p. 19), percentage changes in output in other crop groups are smaller. 311 p. 21; data are also given for a group of Latin American energy exporting countries (Argentina, Colombia, Mexico and Venezuela) but were not analysed.

Page 89: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

89

7.2.7. Conclusion The production of most biofuels is accompanied by the production of co-products that are generally used as animal feed. When used in this way, they replace dedicated animal feed crops that would otherwise be grown on land. It is clear that this should be taken into account in estimating the land use impact of the promotion of biofuels. To do this, modellers need data on - the quantity of co-product produced; - the quantity and type of animal feed that each unit of co-product can be expected to

replace (the "substituted product"); - the relative land requirements of the biofuel crop and the substituted product. As the tables below show, different empirical studies and models differ in their data/assumptions on all these points, and – unsurprisingly – also in their results. The first table shows data on the volume of co-product production. It can be seen that for most parameters the highest estimate given in these sources is no more than 25% greater than the lowest estimate. The exceptions are rape seed and soya, to which JEC attributes around 50% more meal production than do Ensus and Özdemir et al. Table – Volume of co-product production312 co-product quantity (toe/t

feedstock) co-product share313

312 Land-saving co-products only

Page 90: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

90

Energy basis JEC Ensus Özdemir

et al. JEC Ensus Özdemir

et al. sugar beet

0.026 0.028 36% 36%

wheat 0.117 0.098 39% 32% rape seed 0.233 0.141 40% 29% palm oil 0.010 0.008 5% 4% soya 0.308 0.230 0.211 66% 61% 57%

Mass basis CE Delft Özdemir

et al. soya 83% 81%

Economic basis Taheripour

et al. Özdemir

et al. soya 53% 60% Source: calculations of the Commission services from JEC well-to-wheel study314, Lywood et al. (2009)315, Özdemir et al. (2009), CE Delft (2008), Taheripour et al. (2008) The next topic – the animal feed that each co-product could, technically, replace – can be divided into (a) the total amount of co-products that the market could absorb and (b) the amount of different animal feeds that each unit of co-product could replace. On the first point, studies were reviewed by CE Delft (2008), Lywood et al. (2009) and Taheripour et al. (2009). There was consensus among these authors that the markets' absorptive capacity was not a constraint on the use of biofuel co-products to substitute for other animal feeds. The table shows data on the second point. The distinction between replacement of cereals and of soya meal is important because the production of soya is more land intensive.316 Table – Amount of cereal (C) and soya meal (S) that each unit of co-product can be expected to replace (mass basis: t/t) Air CARB317 CE Ensus report Kim et Searchinger Tyner

313 Proportion of combined production of biofuel and co-product 314 [reference] 315 Data from Ensus (2008) have been omitted on the assumption that Lywood et al. (2009) updates them 316 In dealing with the replacement of cereals, studies also differ on whether wheat and rape seed co-products replace wheat or maize. It is assumed here that this distinction is of less significance.

Page 91: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

91

Improvement Resource

Delft318 for EPA319

al. et al. et al.

maize C 0.95 S 0.28

C 1.00 S 0.00

C 0.69 S 0.45

C 0.49 S 0.40

C 0.50 S 0.50

C 0.95 S 0.30

C 1.00 S 0.00

C ? S 0.00

wheat C 0.66 S 0.50

C 0.39 S 0.59

rape seed

C 0.26 S 0.66

C 0.15 S 0.61

Source: Commission calculations from Air Improvement Resource (2008), CE Delft (2008), Lywood (2009), Kim et al. (2008), Searchinger et al. (2008), Tyner et al. (2009, data for "new model". Note: these data embody assumptions about the share of different livestock types in consumption, since livestock types differ in their consumption of different animal feeds. These data show more variation than the data for quantities of co-product per unit of biofuel. For maize, for example, estimates of the mass of animal feed substituted per ton of DDGS vary from 0.89 to 1.25 tons; estimates of the share of soya in the substituted animal feed range from 0 to 50%; and there is no correlation between the two estimates. Using assumptions about co-product production per unit of biofuel (as in the first table above), assumptions about cereal and soya substitution per unit of co-product (as in the second table above), and techniques to depict how shifts in demand affect output, modellers can estimate the change in crop output due to the promotion of biofuels (a) when co-products are ignored and (b) when they are taken into account. Using further assumptions about crop yields per hectare and techniques to depict how shifts in output affect land use, modellers can estimate the change in cropland due to the promotion of biofuels under the same two hypotheses. The tables shows the impact of taking biofuels into account in modelling exercises. (The results from Ensus (2008) are not included in the first table, since Lywood is the main author of that study, and it is assumed that Lywood (2009) updates the results in Ensus (2008).) Tables – Estimated reduction in required quantity of land when co-products are taken into account:

individual crops

317 According to Air Resources Board (2008) and Lywood (2009) 318 These figures are for co-products consumed in the EU. CE Delft consider that co-products consumed in the US replace only maize. 319 According to Kim et al. (2008) [The reference is EPA, “Regulatory Impact Analysis: Renewable Fuel Standard Program”, EPA420-R-07-004, 2007.]

Page 92: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

92

maize sugar beet wheat rape seed palm oil CE Delft (2008)320 EU 47%

US 61% EU 42%

US 35% EU 38% US 61%

Lywood (2009)321 EU 74% EU 65% EU 94% EU 61% SE Asia 7% overall policy Gallagher review (2008) 322 62-64% GLOBIOM323 8-16%

(result for "net deforestation" Ozdemir et al. (2009) 324 23-37% Tyner et al. (2009)325 43% Source: calculations of the Commission services from the sources shown. In the first table, locations are both for the production of the feedstock and the consumption of the co-product. Given this evidence on the large impact that can be expected when co-products are included, the different ways in which this is done in the different modelling exercises can be expected to have a significant impact on their results.

320 For details of the calculation see section [7.2.5]. Note that an apparent mistake in the source, which would exaggerate the land savings when soya meal is the substituted product, has been corrected. 321 pp. 17 and 21 322 p. 32 [source is Ecofys 2008, probably “Land use requirements of different EU biofuel scenarios in 2020”]. CE Delft’s work (reported in sections [7.2.4 and 7.2.5] is an input to this exercise. It is not clear how this result relates to lower figures reported in CE Delft (2008a) (p. 29), which was an input to the work of the Gallagher review. 323 Havlík et al. (2009) (p. 36) 324 p. 2990 325 p. 16; also Taheripour et al. (2008) (p. 3)

Page 93: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

93

8. Model calculations: from tons to hectares

8.1. Yields: introduction Once modelling exercises have established the additional quantity of agricultural commodities that needs to be produced as a result of demand for biofuels, the next step is to translate this demand into a quantity of land use change (measured in hectares). The quantity of land needed will depend on the values used in the model for yields (crop production per ha). Yields feature in the calculation in two ways. First, the model will include baseline assumptions for yields. These are considered in section 4.3. Second, some models depict yields as varying in response to changes in demand and land use. That is the subject of this chapter. Yields can be increased by using more non-land inputs on a given piece of land or by technological development that gives more production per unit of input. (ECOFYS (2009) express the distinction as one between short term and long term yield developments.326) One particular way to use more inputs is to increase the frequency of cropping (by reducing the incidence of fallow years in a rotation, or growing multiple crops on the same land in a single year). In principle, all these types of yield improvement could be accelerated in response to increased demand (or price). These would be reasons for average yields to be higher in the policy scenario than in the baseline scenario. Even if such a response to increased demand occurs, it may well be that it does not satisfy the whole of the demand increase and that a part remains to be satisfied by the conversion of land from other crops or from non-crop uses. It is possible that this converted land will have a lower yield than the land previously used for the cultivation of the crop in question. This would be a reason for average yields to be lower in the policy scenario (with more land use) than in the baseline scenario. (Lywood (2009) says that this effect is sometimes termed “slippage”.327) The treatment of these yield issues in analytical work is described in section 8.2; the treatment of yields in modelling exercises is described in section 8.3.

8.2. Factors affecting yields: analysis and empirical work

8.2.1. Increased inputs in response to demand Kløverpris et al. (2008) use the term “optimisation of production” where the present paper uses the term “increase in inputs”. They identify the following ways to achieve this: fertiliser application; pesticide application; irrigation level; and cropping intensity (see section 8.2.2). They state that these options are all subject to diminishing returns.328 Searchinger and Heimlich (2008) state that “The dominance of world agricultural production by mature agricultural economies that already use fertiliser and other inputs at high levels … suggests that technology-improvements, not price-induced increases in inputs, will drive 326 p. 3 327 p. 12 328 p. 16

Page 94: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

94

further yield increases”329 and that “It is widely accepted that in these countries [that have “made the transition to modern agriculture”], technology improvements will determine most yield growth… [T]here is relatively little room for most of the world’s agricultural production to increase output just by increasing inputs other than land”.330 They add that “Developing countries have more capacity to absorb inputs productively… But even in developing countries, subsistence producers may not respond to market signals331, and government policies in some of the countries that have the greatest opportunity to boost yields may in effect cushion price effects, which would reduce land expansion in those countries but direct it elsewhere.”332 According to Woods and Murphy (2009), “For the USA, Cassman (2008) has argued that as a result of the huge investment in maize production over the last 30 years and because of the very substantial existing infrastructure, biofuels are unlikely to result in an increase in the rate of long term yield gain. However, they may encourage the gap to be closed between the ‘best’ and ‘worst’ producers in terms of yield.”333 ECOFYS (2009) take the view that the barriers to the taking up of new agricultural technologies in Sub-Saharan Africa have included poor infrastructure, high transport costs, limited investment in irrigation and poor extension services.334 ECOFYS (2009) state that, in their literature review, “we did not find any literature that actually analyzed the actual behaviours of farmers in response to varying prices of inputs and outputs… While several sources find potential indications that prices of inputs and outputs influence yields, none of these sources provides proof for this causal relationship… Proving the relationship between short term price fluctuations and yield is challenging because other factors than price will have a significant impact on yields, such as weather conditions and technological change, and good data may not be available on these non-price variables (Rao, 1988)335. The ability of farmers to respond to price signals can be strongly limited due to [“non-price”] barriers. This is illustrated by the fact that in response to higher cereal prices, cereal production in developed countries increased by 11 per cent from 2007 to 2008, while production in developing countries rose by only 0.9 per cent. When Brazil, India and mainland China are excluded, production in developing countries actually fell by 1.6 per cent.336”337 ECOFYS (2009)338 quote Rao (1988)339, in a survey of the literature on agricultural supply response to price in developing countries, as follows: “Yield responses to prices are smaller and display much less stability than acreage elasticities” and “One reason for the wide range in crop-specific elasticities [found in research] is the fact that elasticities can systematically

329 pp. 1-2 330 p. 14 331 Referenced to Sain, G. and M.A. Lopez-Pereira, “Maize Production and Agricultural Policies in Central America and Mexico” (2002) 332 p. 13 333 p. 2 334 p. 4; sourced to IFPRI, “Green Revolution: Curse or Blessing?”, 2002 and [UNCTAD, 2009] 335 Referenced as J.M. Rao, “Agricultural Supply Response: A Survey”, Agricultural Economics 3, 1989 336 Referenced to IFPRI, “Sustaining and Accelerating Africa’s Agricultural Growth”, 2008 337 pp. 7-8 338 p. 21 339 Referenced as J.M. Rao, “Agricultural Supply Response: A Survey”, Agricultural Economics 3, 1989

Page 95: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

95

differ among crops and among countries. Evidence suggests that these determinants of supply elasticities include technological factors such as crop-specific yield risks, the feasibility of multiple-cropping and the availability of arable land; economic factors such as crop-specific price risks, the relative importance of the crop, farm incomes and farm size, and the incidence of tenancy; and sociological dimensions such as the level of farmer literacy.” Bouët et al. refer to “the high substitutability between fertilizers and land” but do not appear to provide evidence for this.

8.2.2. Increased frequency of cropping Kløverpris et al. (2008) use the term “optimisation of production” where the present paper uses the term “increase in inputs”. They identify cropping intensity (“the ratio between harvested area per year and the area of arable land”) as a way to do this. They state that all methods are subject to diminishing returns.340 ECOFYS (2009)341 quote Rao (1988)342, in a survey of the literature on agricultural supply response to price in developing countries, as follows: “One reason for the wide range in crop-specific elasticities [found in research] is the fact that elasticities can systematically differ among crops and among countries. Evidence suggests that these determinants of supply elasticities include technological factors such as... the feasibility of multiple-cropping.”

8.2.3. Faster technological development in response to demand Kløverpris et al. identify the following types of technological development: the improvement of mechanical aids; of crop strains; and of agricultural practices. They point out that unlike increases in inputs, “technological development ... will always lead to increased … yields …simply because technological improvements are not discarded in case of falling prices”. They go on to discuss the drivers of technological development: “The main driver for the development of better mechanical aids … is assumed to be … competition between suppliers of agricultural machinery. However, it cannot be ruled out that increased crop demand will also influence the speed of technological development within this field. The general drivers of better crop strains are assumed to be crop demand and … competition between companies developing and selling seeds. The demand for crops will influence the research priorities within these companies. More resources will be allocated to crops in high demand because these are being sold in large quantities. Furthermore, public funds may be allocated to this field of research in case of existing or perceived future societal food shortages. This can also be considered a type of demand… The general drivers for development of better agricultural practices will mainly be political decisions since this is mainly a public research field. This means that societal needs may also influence this development.” 343 According to the Millennium Ecosystem Assessment (2005), “it is the interplay between indigenous knowledge, access to new technologies, and risk aversion that are major determinants of decisions about cultivation practices and evaluation of farming systems”.344 340 p. 16 341 p. 21 342 Referenced as J.M. Rao, “Agricultural Supply Response: A Survey”, Agricultural Economics 3, 1989 343 p. 17 344 p. 774

Page 96: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

96

Biomass Research and Development Initiative (n.d.) identify the following types of technological development: increased genetic yield potential; greater resistance to pests, diseases or drought; and management innovations.345 Searchinger and Heimlich (2008) state that “Adoption of new technology was the primary driver of [recent] yield increases, occurred despite sharp declining real crop prices, and would have occurred at substantial levels regardless of new demand… The dominance of world agricultural production by mature agricultural economies that already use fertiliser and other inputs at high levels also suggests that technology-improvements, not price-induced increases in inputs, will drive further yield increases. The extent to which higher prices lead to improved technology is largely unknown, particularly because governments fund more than half of agricultural research and development.”346 In another statement on the same point they comment that “[H]igher demand and higher prices could also spur further investments in research that result ultimately in higher yields. Unfortunately, these effects, although plausible, are also speculative. Moreover, more than half of agricultural research worldwide is government-funded. In a policy context, this research should be viewed as independent of biofuels because such funding could, and should, be boosted with or without biofuels to held [sic] preserve forests and grasslands to avert global warming in the face of growing food demands by a rising population. In fact, anecdotal evidence suggests that at least in the short run, biofuel hopes have probably decreased investment in food crop yields as talent and funding has shifted to biomass crops”.347 Woods and Murphy (2009) draw attention to the scope to increase crop availability through “reduction in wastage/losses” and “efficiency gains from integrated supply chains”.348 They add that “Hazel and Wood (2007) argue that for most developing countries conventional crop yields are primarily a function of infrastructural investment”.349 ECOFYS (2009) express the issue as follows: “1. What new technologies in agricultural production (e.g. chemical fertiliser, crop breeding) made long term yield increases possible over the past decennia? 2. What were the key drivers behind the innovation and adoption of these technologies, and can these drivers be linked to increasing demand? 3. Have relative prices driven the direction of innovation and adoption of new technologies?”350 They take the view that “The adoption of new technologies by farmers has been incentivised by declining producer prices. Falling crop prices forced farmers to reduce input costs by adopting new technologies in order to maintain a sufficient margin. The resulting growing output or reduced input costs again reduced crop prices, forcing farmers to reduce input costs further. This ‘farm problem’ was first described by Cochrane 1958.”351

345 p. 34 346 pp. 1-2 347 p. 14 348 p. 1 349 p. 2 350 p. 3 351 p. 3

Page 97: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

97

They add that “Before the 1900’s labour was the scarce production factor and innovation was focused on reducing labour inputs through mechanization. From the 1900’s onward, land became an increasingly scarce production factor and advancements in agricultural production started to focus more on yield improvements352 … The key contributors to yield improvements can be divided into genetic improvements, resulting from crop breeding and genetic modification, and changes in crop management, including mechanization, application of chemical fertiliser and pesticides and irrigation.353”354 They go on, “Publicly funded R&D in agricultural technologies has been an important source of the technological innovations that made the dramatic yield increases of the past decennia possible. Publicly funded R&D played a major role in the early advancements in the developed world in the first half of the 20th century and also the transfer and adoption of these technologies to developing countries in the Green Revolution was largely made possible by not-for-profit institutions. This finding is confirmed by two findings of CIFAD (2002): • By the mid 1990’s, one third of the total spending on agricultural research was private. • At least in the 1990’s, the focus of privately funded research was very different from that

of publicly funded research. Publicly funded research focused on the farm (where yields are largely determined), whereas only 12% of privately funded research focused on the farm. Instead most private research focuses on food and other post-harvest elements…

The effectiveness of both public and private agricultural research are generally both high with IRRs ranging from zero to 300% (Shiva 1999)355. Furthermore, IFAD (2009) states that the decline in the growth rate of agricultural productivity from 3.5% in the 1980’s to 1.5% today, coincides with a reduction in public spending on agricultural research aimed at developing countries: • International support for agricultural research in developing countries declined from 18%

of total assistance in 1979 to 3% in 2006 • Domestically, government investment in agriculture in developing countries also fell, by

one third in Africa and by as much as two thirds in Asia and Latin America during this period.

This further strengthens the conception that yield developments are largely dependent on agricultural research. Similar findings that a decline in yield growth can be linked to a decline in agricultural research spending are reported by Tweeten and Thompson (2008)356, Abbott et al. (2008) and Fuglie (2008), although the latter does warn that the conclusions holds for land yields but does not necessarily hold for the total agricultural productivity. Vernon (2008)

352 Sourced to J.M. Reilly and K.O. Fuglie, “Future yield growth in field crops: what evidence exists”, Soil and Tillage Research 47, 1998 353 Sourced to IFPRI, “Green Revolution, Curse or Blessing”, 2002; P. Peltonen-Sainio et al., “Cereal yield trends in northern European conditions: Changes in yield potential and its realisation”, Field Crops Research 110, 2009; and [Bell et al., 1995] 354 p. 4 355 referenced as S.M. Shiva et al., “Returns to American Agricultural Research: Results from a Cointegration Model”, Journey of Policy Modeling 21(2), 1999 356 Referenced as L. Tweeten and S.R. Thompson, “Long-term Global Agricultural Output Supply-Demand Balance and Real Farm and Food Prices”, 2008

Page 98: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

98

states that much of the technical change that lead to output growth per hectare has been produced by public institutions (but gives no source for this claim).”357 Concerning the role of prices in agricultural innovation, ECOFYS (2009) conclude that “The more recent publications seems to agree that while some support for the theory of price-induced innovation can generally be found, other factors than price also play an important role… It should be noted that the price-induced innovation theory aims to explain which production factors innovation will focus on, depending on relative prices of different production factors. This is not the same as explaining how innovation will respond to increased prices of the output product. Neither does it predict whether the innovation will actually lead to yield increases – innovations may also be input saving without increasing yields per ha.”358 ECOFYS (2009) conclude that R&D has played an important role in long term yield improvements and that public R&D has played a larger role than private R&D. They derive from this the conclusion that “[F]ree-market driven mechanisms alone were not the major driving force behind past yield developments. While public spending on R&D may also be demand driven, e.g. in an effort to feed a growing world population, this can not be attributed to free-market mechanisms.” Bouët et al. (2009) state, “One recent analysis concluded that relative price changes have not encouraged innovation in US agriculture in the last 40 years. The paper concludes, ‘This finding cautions against the efficacy of policies based on the premise that price signals alone induce efficient technical change’”.359 Thirtle et al. (2003) present econometric analysis showing that investment in agricultural R&D in developing countries increases yields.360 They show this with a productivity equation in which agricultural R&D is lagged five years.361 This is an assumption rather than a result.362 They estimate that the rate of return on national public sector agricultural R&D averaged 22% in the African countries in the sample, 26% in the Asian countries and 10% in the countries in the Americas. The authors suggest that the reason for lower returns in the Americas is likely to be that these countries have higher GDP: “Yield-increasing innovations are no longer such a driving force in the development process as they are in the poorer continents”. If the true lag before the spending takes effect is longer than 5 years, the rates of return would be lower. The authors note, however, that results from detailed country-level studies are usually higher.363 Average national public sector spending on agricultural R&D was $(1995)0.66 in the African countries, compared to $3.13 in the Asian countries and $1.36 in the countries in the Americas.364

357 pp. 4-5 358 pp. 5-6; the sources cited are Y. Liu and C.R. Shumway, “Demand and Supply of Induced Innovation: An Application to U.S. Agriculture”, 2007 and C. Thirtle et al., “Testing the induced innovation hypothesis: an error correction model of South African agriculture”, Agricultural Economics 19, 1998 359 p. 17; referenced to Y. Liu and C.R. Shumway, “Demand and Supply of Induced Innovation: An Application to U.S. Agriculture”, 2007 360 p. 1960; see also section [8.2.6] 361 p. 1965 362 p. 1969 363 Referenced to J. Alston et al., “A meta-analysis of rates of return to agricultural R&D: Ex Pede Herculem?”, IFPRI, 2000 364 pp. 1968-1971

Page 99: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

99

Kløverpris et al. (2009) state that “no data is identified to form an empirical basis” for the relationship between increased demand and yield-improving technological development in agriculture.365

8.2.4. Lower yields on land converted from non-crop use Land supply curves are used to determine the yields of non-crop land converted to crop use. They do not appear to distinguish between yields under different crops. Kløverpris et al. (2008a) describe a workshop discussion which “characterised the land supply curves used in some economic models (Kløverpris, Banse) as a powerful concept although problems with the calibration exist. Furthermore, the suitability of the agricultural land expressed by the land supply curves is not the only decisive factor. Infrastructure and social aspects also determine which land is the next to be used.”366 Searchinger and Heimlich (2008) state, “As cropland acres expand, cropland also expands into more marginal land.”367 They do not provide evidence for this statement. In relation to Brazil they state, “Brazil’s government has projected that most new cropland will come from further conversion of the Cerrado, where remaining forests tend to be on hill tops, and even studies that show large potential expansion into the Amazon for soybeans show lower yields in much of it (Vera-Diaz 2007)368. Moreover, while Brazil experiences more expansion than any other country in our analysis, expansion will still primarily occur in other regions, including China and India, where newer croplands are probably less productive.”369 Woods and Murphy (2009) state that “[I]t is not yet clear that the best land is allocated to the most appropriate soils and climates in any country… Over time a rationalization will occur that matches soils, climates and crop/variety better through learning by doing.”370 Bouët et al. state, “Best lands (in the IIASA nomenclature, the very suitable and suitable land) are generally already in cultivation, Marginal land is therefore intrinsically of lower quality and marginal productively is therefore expected to decrease with land extension.”371 Lywood (2009) quotes CARB as saying that “little empirical evidence exists to guide modelers in selecting the most appropriate factor” for lower yields on converted land.372 ECOFYS (2009a) address the question, “Does land taken into production usually have a lower productivity than the average productivity of land already in production”. They state, “In our brief literature survey, we could find no study that systematically tries to answer this question… It is unclear whether the required data on land expansion and productivity exists.”373 365 p. 8 366 p. 181 367 p. 16 368 referenced as M. Vera-Diaz et al., “An Interdisciplinary Model of Soybean Yield in the Amazon Basin”, Ecological Economics (2007) 369 p. 16 370 p. 2 371 p. 72 372 p. 13 373 pp. 1-2

Page 100: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

100

They add, “Hong Yang et al.374 study … land cultivation in China over two decades. They observe a drastic reduction in the area of fertile land cultivated in the south-east of the country which is partially offset by land expansion in other areas. However, the net expansion in these other areas hides a dynamic balance of substantial abandonment of damaged land and reclamation of land to make up for these losses. In another study on the Chinese situation375 … Lin et al. find a large-scale conversion from unused and pasture land to cultivated and construction land between 1949 and 1996. As the other study, they observe the large losses of fertile land in the south-east to construction land which cannot be fully compensated for by the reclamation of low-grade farmland in the environmentally fragile frontier regions.”376 In a fuller summary of the Hong Yang et al. study ECOFYS state, “The study finds that the decline in cultivated land [during the previous two decades]… was mainly the result of a drastic reduction of fertile land in the southeast areas. The conversion of cultivated land to other types of agricultural uses and the encroachment of various constructions were the major causes of the loss there. Cultivated land increased in some northwest and frontier provinces, which partially offset the loss in the southeast. Reclamation was the primary source of the increase. This gain, however, has been made at the expense of environment, indicated by a substantial abandonment of damaged land in the major reclaiming provinces.”377 A study by Lubowski et al.378, also summarised in ECOFYS (2009a), provides some support for the hypothesis that the land in use at any one time for agriculture is that with the highest yields: “[US] lands moving between cultivated cropland and less intensive agricultural uses are, on average, less productive and more vulnerable to erosion than other cultivated lands, both nationally and locally”.379 Banse (2009) states, “Additional – currently not used – land is in most cases less productive. Any modeling of an expansion of land use should consider this.”380 Evidence for the assertion is not given.

8.2.5. Lower yields on land converted from one crop to another Searchinger and Heimlich (2008) state that “corn-based ethanol increases reliance on continuous corn, instead of corn/soybean rotations [in the US]”, and that this has a negative effect on yields.381 This is further explained as follows: “The most common U.S. rotation system for both corn and soybeans is a corn-soybean rotation… Studies have consistently found a sharp drop in yield of 10% to 15% in corn production for acres grown in second-year corn”.382

374 Referenced as Hong Yang et al., “Cultivated land and food supply in China”, Land Use Policy 17, 2000 375 Referenced as G.C.S. Lin et al., “China’s land resources and land-use change: insights from the 1996 land survey”, Land Use Policy 20, 2003 376 p. 1 377 ECOFYS (2009a) (pp. 4-5) 378 Referenced as R. N. Lubowksi et al., “Environmental effects of agricultural land-use change: the role of economics and policy”, USDA 2006 379 p. 7 380 pp. B3-B4 381 p. 2 382 p.15; referenced to B.J. Erickson and J.M. Lowenberg-Deboer’ “Weighting the Returns of Rotated vs. Continuous Crop” (2005), D.A. Hennessy, “On Monoculture and the Structure of Crop Rotations” (2006) and M. Duffy and D. Correll, “Where Will The Corn Come From” (2006).

Page 101: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

101

8.2.6. General Kløverpris et al. (2008) comment that “It is difficult to estimate the fraction of intensification derived solely from increased demand”.383 Keeney and Hertel (2008) state, “Despite the well known principles of producer theory, it is commonplace to attribute intensive increases of agricultural output to technological gains disembodied from the producer or market signals while ascribing year to year variations solely to weather. The predominance of research on supply response looks only to acreage movements, adopting the assumption that any yield change in response to commodity price movements is insignificant. Houck and Gallagher (1976)384 challenge the dominance of acreage response in agricultural supply studies, offering empirical estimates of yield response to commodity prices on par with those for acreage response to the same price changes… Induced innovation studies in the spirit of Hayami and Ruttan (1971)385 regularly find empirical support for the notion that the supply of technological advance is also responsive to changes in agricultural prices. Aggregate studies of agricultural technology find significant potential for non-land inputs to substitute for land in response to changes in prices as well (Abler, 2000386). Hertel, Stiegert and Vroomen (1996)387 show that substantial, econometrically estimated, aggregate substitution possibilities can be reconciled with limited response at the farm level through entry and exit (through turnover in the land market) by producers of heterogeneous managerial ability (and hence heterogeneous input intensities). Apart from prices, changes in agricultural policy incentives have been shown to generate dramatic changes in trend yields (Foster and Babcock, 1993)388… Keeney and Hertel (2005)389 identify the set of assumptions on factor supply and substitution in primary crop and livestock agriculture as well as food processing that lead to significantly different supply and demand responses to policy changes in general equilibrium context.”390 Keeney and Hertel (2008) report seven published estimates of corn yield response to price, giving elasticities ranging from 0.22 to 0.76. They conclude that “taken as a group, [these estimates] lead us to rejection the hypothesis of zero yield response in the long run.”391 (Searchinger and Heimlich (2008) comment on the same set of estimates, plus three others, and draw a different conclusion: “the few studies that have attempted to estimate the responsiveness of yields to price, once full use of nitrogen fertiliser had been widely adopted, have generally found yield highly… unresponsive”.392 This implies that Searchinger and Heimlich hold the view that where “full use of nitrogen fertiliser has been widely adopted”393, 383 p. 18 384 referenced as Houck, J.P. and P.W. Gallagher, “The Price Responsiveness of U.S. Corn Yields”, American Journal of Agricultural Economics 58, 2005 385 referenced as Hayami, Y. and V. W. Ruttan, “Induced Innovation and Agricultural Development”, 1971 386 referenced as Abler, D., “Elasticities of Substitution and Factor Supply in Canadian, Mexican, and United States Agriculture” in Market Effects of Crop Support Measures, OECD, 2000 387 referenced as Hertel, T.W., K. Stiegert and H. Vroomen, “Nitrogen-Land Substitution in Corn Production: A Reconciliation of Aggregate and Firm-Level Evidence”, American Journal of Agricultural Economics 78, 1996 388 referenced as Foster, W.E. and B.A. Babcock, “Commodity Policy, Price Incentives, and the Growth in Per-Acre-Yields”, Journal of Agricultural and Applied Economics 25, 1993 389 referenced as Keeney, R. and T.W. Hertel, “GTAP-AGR: A Framework for Analysis of Multilateral Changes in Agricultural Policies”, GTAP Technical Paper No. 25, 2005 390 pp. 6-7 391 pp. 9-10 392 p. 11 393 p. 11

Page 102: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

102

yield response to demand is less. They go on to state, “Obviously, for biological reasons, the responsiveness of yield to non-land inputs will decline with larger price increases. Estimated price increases for biofuels are dramatically beyond the level examined by these studies, and the responsiveness of yields to price should therefore be significantly lower.”394) Having decided to use the average of the seven results (0.4, to one d.p.395, with an assumed range of 0.0 to 0.8) in their work, Keeney and Hertel continue, “Other evidence on aggregate response due to input intensification in the long run (Tweeten and Quance, 1969; Peterson, 1979)396 marks this yield elasticity distribution as somewhat conservative.”397 Lywood (2009) states, “A large amount of work has been done on the reasons for increasing yields and trying to model yield changes. Most of the modelling has focused on short term relationships between price changes and yield changes to obtain a yield : price elasticity. In general a low and varied responsiveness of yield changes to price changes has been found (Keeney 2008).”398 ADAS UK Ltd (2008) say that “Further work [than that done in their report, which addresses yields in the baseline] would allow an “expected improvement” estimate to be made, requiring consideration of how yields may improve with likely changes in prices, investments, political and economic reforms etc.”399 They add that “The FAO have regarded declines in crop yield improvement to be a consequence of a decline in the growth of demand, as population increases begin to level off”.400 Defending their assumption that yields in the policy scenario would not be higher than in the baseline scenario, Searchinger and Heimlich (2008) state that “The mere fact that most recent increases in crop production are generated by yield increases does not mean that further demands will further increase yields beyond trends to supply most, or nearly all, the new demands. Adoption of new technology was the primary driver of yield increases, occurred despite sharp declining real crop prices, and would have occurred at substantial levels regardless of new demand… Evidence is compelling that higher prices trigger crop expansion, but studies of the impact of higher prices on crop yields generally conclude that the impact is small.”401 They go on to state that a “reasonable estimate of price-induced yield increases” would be “sufficient to supply 25% of the grain to replace corn diverted to ethanol”.402 No reason is given for the choice of this figure.

394 p. 13 395 the figure to two d.p. would be 0.44. The seven estimates include four from a single study (Houck and Gallagher (1976), referenced as Houck, J.P. and P.W. Gallagher, “The Price Responsiveness of U.S. Corn Yields”, American Journal of Agricultural Economics 58 (1976). If these four were first to be averaged, and the result were then to be averaged with the estimates of the other three studies, the figure would be 0.40. 396 referenced as Tweeten, I.G. and C.L. Quance, “Positivistic Measures of Aggregate Supply Elasticities: Some New Approaches”, American Journey of Agricultural Economics 59, 1969 and Peterson, W.L., “International Farm Prices and the Social Cost of Cheap Food Policies”, American Journal of Agricultural Economics 61, 1979 397 p. 11 398 p. 10; referenced as Keeney and Hertel, “Yield Response to Prices: Implications for Policy Modelling”, 2008 399 p. 4 400 p. 15; referenced to Bruinsma, J. (ed.), “World Agriculture: Towards 2015/2030: An FAO Perspective”, 2003 401 pp. 1-2 402 p. 2

Page 103: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

103

They add that “As our supporting materials [Searchinger et al. (2008)] point out, the extent of price-induced yield increase is uncertain because there are few studies that have tried to estimate them in the past and because this kind of analysis faces serious methodological obstacles.”403 They state that it is “obviously not true” that “in the absence of greater demand, there would have been no increases in yield [in recent decades]”: “Changes in agricultural technology are both yield-enhancing and cost-reducing, so farmers have a continuing incentive to adopt them even when demand is falling. Competitive pressures force the continuing adoption of new technology, leading to Professor Willard Cochrane’s famous “treadmill”404… Modern food-producing technologies already in existence… would have continued to diffuse, farmers would have continued to make investments in machinery, drainage and irrigation, and most importantly, research would have continued to generate better-producing crop varieties.”405 On the other hand, they accept that “it is also fair to assume that the growth in population and demand, through price effects, spurred greater improvements in yields”.406 Referring to the fact that past decades’ yield growth took place at a time of falling crop prices, they state that “Some economists have argued that this growth in yield negates the argument of price-induced yield increases. We do not agree. In the absence of increased food demand, prices would have declined yet further, and there is no way to know how much the relative maintenance of prices contributed to yield increases. Even so, the dramatic growth in yields despite large decreases in real prices hardly provides evidence that yield increases are closely tied to prices and aggregate demand.”407 Searchinger and Heimlich (2008) argue that the fact that there is a statistical correlation between higher prices and cropland expansion408 is the most important argument in favour of their assumption that “agricultural expansion will largely fulfill the growth in demand”. Having reviewed the question of whether yields grow as a result of increased inputs when demand grows, they conclude, “It is quite possible that yields are more responsive to price than that caused by increased input use alone because higher prices can spur technological improvements and perhaps encourage replacement of less successful farmers by those who are more successful. (In the developing world, that might mean replacement of small farmers with bigger operations.) Economists have great difficulty estimating long-run elasticities of supply, such as yield responses to price increases over many years, because it is difficult to isolate price effects over that period from other intervening factors. In addition, time-series data is not particularly reliable because high yields in a year will tend to produce abundant crops that lower prices, and low yields result in high prices, confusing the causal relationship. It is common sense that higher prices will spur at least some further investments in fertilizer, irrigation, seeds or drainage to boost yields, and over time, some additional research. However, our modeling approach focuses on using established relationships and data. Possible relationships that are quantitatively unknown are dealt with through sensitivity

403 p. 3 404 referenced to Cochrane, W.W., “Farm Prices: Myth and Reality” (1958) 405 p. 4 406 p. 4 407 p. 5 408 p. 9

Page 104: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

104

analysis. And the lack of data to support large demand-induced yield increases does not provide reason to assume them.”409 Lywood (2009a) plots yield growth against output growth. Each point represents a four year period; each year, except those at the start of the dataset, is taken into account in four different points. FAO PRODSTAT data for 1961-2007 are used. At a global level, regression analysis gives the result Yield growth [per annum] = -0.2% +0.85 (output growth).410 This would suggest that output growth is the main driver of yield growth. (Similar work is reported in Lywood (2009).411) Lywood (2009d) states, “The extent to which the output growth is met by yield or land increases will depend mainly on the crop. For annual crops… the tendency is to increase yields by growing more disease resistant or higher yielding varieties, increasing capital investment, or utilising more advantageous husbandry practices. For perennial crops the yield of any mature plantation from year to year is more constant, so additional output will tend to be obtained by planting more land.”412 Faaij (2009) states that Utrecht University is “looking at far more detail in historic developments in productivity increases in specific countries and understanding why these occurred. We hope that such case studies give us much better modelling data for scenario’s that assume such improvements in the future”. Lywood (2009) says, “[S]ince increased crop output is driven by higher prices, there must be relationships between land area growth, yield growth and prices. The reasons why these relationships have been difficult to determine is because increased output as a result of higher prices is in large part due to longer term investment in cultivating new land, land improvements, machinery, infrastructure and higher yielding crops for the years following high prices. The effect of prices takes several years to be fully reflected in yield and land area changes. The increase in output growth is therefore related more closely to relative price levels than to price changes.”413 Working with a sample of developing countries, Thirtle et al. (2003) obtain the following result:

Yield = 0.44 (national public sector R&D, $ per ha) + 0.01 (fertiliser, 100g per ha)2 + 0.34 (labour, agricultural workers per ha) + 0.65 (land quality index)414 – 0.21 (illiteracy) + 3.83

409 p. 13 410 Lywood (2009a) (pp. 3-4) 411 p. 11+ 412 p. 7 413 p. 10 414 This index is referenced to K. Wiebe, “Resource quality and agricultural productivity: a multi-country comparison”, 2000 (http://agecon.lib.umn.edu/aaea00/sp00wi01.pdf)

Page 105: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

105

This equation explains 82% of the variance in land productivity between the countries in the sample. It is interesting to note that there appears to be an increasing return to fertiliser use. Machinery per ha was not significant. The influence of illiteracy is ascribed to the fact that “Literate farmers are more able to assimilate information and make effective use of the new technologies that become available”. When the results are reported by continent, labour has a higher elasticity in Africa and the Americas, “but was insignificant for Asia, reflecting surplus labour, especially in South Asia”. Machinery is significant in Africa. Land quality has a high elasticity in Asia and the Americas while it is insignificant in Africa (reflecting surplus land?).415 Sanz Labrador (2009) comments, "The idea that an increase in agricultural production is not possible without using new farmland has been around for some time. Krugman and Wells (2008)416 argued that European politicians underestimated the reaction capacity of farmers by assuming that the high prices guaranteed by the CAP in the 70's and 80's would not lead to a significant increase in agricultural production. The mistake in this approach translated into an excess of agricultural production."417 They also state, "The … increase in production [in response to "the rise in food prices in 2008"] has occurred largely in developed countries. The reaction in developing countries has been weaker, due to rises in input prices (fertilizers, seeds and oil) and also to difficulties in receiving credit, particularly in a financial crisis environment. In these countries, investments in farm productivity improvement and distribution infrastructure are hampered by liquidity restrictions that also prevent farmers from benefiting from interest rate reductions."418

8.3. Yield response to demand in modelling exercises (Note: to help compare different studies’ approach, this section has been broken down between the different yield-affecting factors identified in section 8.1. This means that some studies’ results are divided up between sections. Where the presentation of a study’s methods or results did not permit this breakdown, they are covered in the final “general” section, which should ideally be read alongside each of the previous sections.)

8.3.1. Response of input use to demand According to Lywood (2009), CARB allows for “[y]ield growth with prices from additional fertiliser use”.419 In IFPRI there is “explicit use of fertiliser for modelling land productivity increase [endogenously]”420, also described as “[c]alibration on elasticities of yield to fertiliser prices (provided by IFPRI partial equilibrium models)”.421 In what is understood to be a separate way in which yields can respond to demand through input changes, there is “an endogenous factor distribution effect”.422

415 pp. 1963, 1965, 1966, 1967; values rounded 416 Referenced as P. Krugman and R. Wells, "Microeconomics", 2008 417 pp. 80-81 418 p. 90 419 p. 10 420 Valin (2009) (p. 6) 421 Valin (2009) (p. 21) 422 Valin (2009) (p. 21)

Page 106: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

106

Like IFPRI, Keeney and Hertel also use GTAP as the modelling platform. In Keeney and Hertel (2008) a distinction is made between two groups of non-land inputs to agriculture: (i) capital and labour; (ii) intermediate inputs. The study describes a model run in which intermediate inputs can vary in response to demand, while inputs of capital and labour cannot. It would appear from a diagram in Valin et al. (2009)423 that a similar distinction is operated in the modelling exercise of IFPRI. Fertiliser is modelled as a substitute for land, while labour and capital are modelled as a substitute for a “land composite” composed of both fertiliser and land. Concerning fertiliser/land substitution, it is stated in Valin et al. (2009) that “[t]he elasticity of substitution for this CES function vary between 0.1 and 2 according to the GTAP database, except for Northern countries… for which the default elasticity is fixed to 0.1”.424 In IFPRI, "Fertilizers are explicitly introduced in the global database and MIRAGE model to capture potential crop production, using more fertilizers, in response to increased demand for biofuel feedstock crops. The characterization of the crop production response to prices resulting from increased bioenergy demand is particularly important. Through improved modelling of fertilizers and its impact on crop yield, we introduce a better representation of yield response to economic incentives while taking into account biophysical constraints and saturation effects. The degree of crop intensification depends on the relative price between land and fertilizers… [C]rop yields in the model increase through three channels: - Exogenous technical progress […]; - Endogenous "factor" based intensification: land is combined with more labor and

capital; - Endogenous "fertilizers" (intermediate consumption) based intensification, the

mechanism described above".425 It is not clear how the different situation of developing countries (far from saturation) is addressed in the work. According to Kløverpris et al. (2008), in the GTAP model “the proportions between intermediate inputs are constant. In other words, crop production cannot be optimised by adjusting the application of fertilisers alone, but only by adjusting all inputs [other than land] equally.”426 According to Lywood (2009) , by contrast, “It is assumed in the GTAP model that the yield in any year changes, due to price changes, primarily by applying more or less fertiliser to the crop. An elasticity factor is used to relate the yield change to the price change.”427 In IIASA, “The rising agricultural prices in the biofuels scenarios provide incentives on the supply side, for intensifying production and for augmenting and reallocating land, capital and labor.”428 Having set maize’s yield price elasticity at about 0.3, Keeney and Hertel (2008) compare two assumptions: (a) it is not possible for the amount of labour and capital used in sectors of

423 p. 9 424 p. 9 425 Al-Riffai et al. (2010) (p. 33) 426 p. 19 427 p. 10 428 IIASA (2009) (p. 22)

Page 107: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

107

agriculture to change; (b) such change can take place, using appropriate elasticities. The results are shown in the table. Table – effect, in the modelling exercise of Keeney and Hertel, of taking into account, in the assessment of the impact of increased consumption of US maize for biofuels, of the possibility of yield increases in response to this increase in demand

Possibility for the amount of labour and capital used in sectors of agriculture to change

Such change is not possible

Such change is possible in the US429

% change compared to baseline scenario land use for coarse grains in the US +1.32% +1.71% yield of coarse grains in the US +0.11% +0.81% output of coarse grains in the US +1.43% +2.52% consumption of coarse grains in the US +2.97% +3.07% US exports of coarse grains -1.54% -0.55% land use in the rest of the world +0.46% +0.14%

Estimated consequent change in coarse grains area harvested (Mha) US +0.43 +0.56 rest of world +1.29 +0.39 total +1.72 +0.95 Source: Keeney and Hertel (2008)430, percentage changes in land use have been translated into estimated changes in area harvested by the Commission services431 In considering these results, a first question is why, in the second row, yield grows even under the assumption that increased inputs of labour and capital are not possible. It would appear that this is because even in this scenario, inputs of intermediate products that can be used to increase production are not capped.432 (It thus seems clear that this modelling treats yield increases as a function only of changes in inputs and not also of technological development.) It can be seen that modelling that permits yield increases through increased use of capital and labour gives substantially different results from modelling that does not. The estimated net global land impact in the coarse grains sector falls by 45%, from 1.72 to 0.95 Mha.

8.3.2. Increased frequency of cropping in response to demand In BLUM, for Brazil, “winter (wheat) and second crops (corn) [have been separated] from the land use analysis.433 Nassar (2009) states that this is not the case for AGLINK and FAPRI.434

429 The study also includes a scenario in which this is also permitted in the rest of the world. The comparison reported here is the one that most clearly shows the impact of permitting inputs to change. 430 pp. 13, 15, 30. The table on p. 30 does not explicitly refer to coarse grains for the various quantities, but this is strongly suggested by the discussion in the text. 431 Base data on harvested area taken from FAO Prodstat, data for 2006 (the base year used by Keeney and Hertel). 432 Keeney and Hertel (2008) (p. 17, footnote 2) 433 Nassar (2009) (p.2) 434 p. 2

Page 108: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

108

8.3.3. Response of technological development to demand In AGLINK there is “no representation of induced technical progress”.435 In IFPRI technology development, unlike input use, is entirely exogenous: "The model does not include endogenous technical progress based on private or public research and development expenditures in response to price changes." It is stated that "the increase of capital and labor [per] unit of land … plays a similar role", although it is hard to see how this could be: one such type of yield response to demand does not preclude the other.436 In any case, the question of technological development in response to demand is identified as one of the "directions for future research".437 According to Kløverpris et al. (2008), “In the standard GTAP model, the technological stage is assumed to be fixed. However, technological development can be incorporated either as an independent variable (determined outside the model) or as a function of another variable, e.g. crop prices. This decision depends on the relationship between crop demand and technological development.”438 Similarly, Piermartini and Teh (2005) state that technological change is “usually exogenous” in trade models439 (of which GTAP is one). They add that unlike the more commonly used comparative static analysis, dynamic analysis makes it “possible to examine whether changes in trade policy affect the rate of investment or accelerate the pace of technological innovation… Dynamic models tend to estimate larger gains from trade liberalization because they take into account the subsequent increases in the rate of investment and the diffusion of technological knowledge” (but have their own disadvantages).440 Smeets (2009) states that for renewable energy, the GREEN-X model includes “a dynamic cost-resource curve, which … incorporates such dynamic parameters as technological change (using the concept of experience curves or expert judgement)… The dynamic curve is endogenous to the model and is determined annually.”441 Kløverpris et al. (2009) carry out a study using GTAP. The effect of crop demand on technological development is ignored in the core scenarios but explored in a sensitivity test442 in which a 2% increase in the price of cultivable land automatically causes a 1% increase in the productivity of cultivable land. (This relationship is arbitrary.) “The relationship is asymmetric in the sense that a decrease in land price will not lead to a decrease in land productivity. This is to reflect the fact that technological development is not rolled back in case of decreasing demand.” The results are shown in the table in section 8.3.6. The amount of cropland required is forecast to fall by between 27% and 80%, depending on the geographical origin of the additional crops consumed. 443

435 von Lampe (2009) (p. 6) 436 Al-Riffai et al. (2010) (p. 33). 437 ibid (p. 72) 438 p. 19 439 p. 3 440 p. 4 441 p. 2 442 p. 4 443 p. 8

Page 109: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

109

Faaij (2009) states that REFUEL includes learning curves for various biofuel pathways.

8.3.4. Yields of converted non-cropland In the CAPRI modelling exercise, ex-fallow land (thus, a type of converted land) is assumed to have lower yields.444 According to Lywood (2009), “[R]eferences for the CARB GTAP model elasticities factors are not provided. The elasticity used to account for lower yield on new land used to grow biofuel crops is justified by “best judgement”.”445 He adds that CARB used “a range of the elasticity factor from 0.25 to 0.75” which “resulted in a 77% change in GHG emissions” (presumably, between the results given by the two factors)446 and that “The CARB work uses ‘best available professional judgement’ to choose a central case elasticity factor of 0.5, meaning that the average yield on new land will be half of that on existing cropland”.447 According to Kløverpris et al. (2008a), Kløverpris and LEITAP/Banse use land supply curves in their modelling exercises.448 LEITAP/ALTERRA uses land supply curves. Their role in the modelling is described as follows: “LEITAP presents total agricultural land supply in land supply function, specifying the relationship between land supply and a land rental rate in each region (van Meijl et al., 2006). Land supply to agriculture can be adjusted by idling agricultural land, converting non-agricultural land to agriculture, converting agricultural land to urban use, and agricultural land abandonment… When agricultural land use approaches potential land use…, farmers are forced to use less productive land with higher production costs (strongly increasing part of the supply curve). As a consequence, in land-abundant regions like South America and for members of NAFTA, an increase in demand … results in a large increase in land use … and a modest increase in rental rates …, while land scarce regions like Japan, Korea and Europe experience a small increase in land use and a large increase in the rental rate… These land price differences will affect competitiveness of biofuel production. The empirical implementation of this land supply curve for non-European regions is based on data from IMAGE, while CLUE with a more detailed spatial presentation provides data on land availability in LEITAP for the European regions. The modelling framework uses the IMAGE 2.4 framework version.”449 The authors add, “The land availability per world region is a very important driver of costs for biofuel production. A distinguishing feature of the LEITAP-IMAGE-method is the introduction of a land supply curve to represent the process of land conversion and land abandonment endogenously.”450 Woltjer and Prins (2009) add that the land supply curves are derived from IMAGE and CLUE data on the basis of "[l]and availability and suitability… National production and land use

444 answer given to a question during a presentation of Blanco-Fonseca and Peréz Domínguez (2009) 445 p. 7 446 p. 13 447 p. 14 448 p. 181 449 p. 7 450 p. 16; as well as van Meijl et al. (2006), the use of this land supply curve is referenced to Eickhout, B. et al., “The impact of environmental and climate constraints on global food supply”, in Hertel, T. et al. (eds.), “Economic Analysis of Land Use in Global Climate Change Policy”, 2009.

Page 110: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

110

data from LEITAP [are fed] into IMAGE and CLUE".451 In this implementation of the supply curves, land use is plotted against "average land rental price".452 The table shows the approximate results given by LEITAP/Banse et al.. The values shown are the combined result of the authors’ assumptions about the response (if any) of yields to (a) demand and (b) land conversion. Table – ratio of production increase to land increase in LEITAP/Banse et al. (change in policy scenario relative to change in reference scenario, both relative to base year) Region Ratio Africa 8.8 Asia 1.6 Central and South America 1.5 EU12 5.2 EU15 5.8 HighInc 2.5 Source: Banse et al. (2007)453; calculations of the Commission services. Note: the base year is 2001. The policy scenario is 2010, with an 11.5% biofuel share in the EU. It has not been possible to ascertain whether the reported results are expressed in financial units or physical units. These results should be treated as approximate since they were obtained by measuring off graphs. The term “HighInc” presumably refers to a group of high income countries, perhaps excluding Member States of the EU. It can be seen that in this study the required land increase far exceeds the required production increase. For example, in the EU every 1% increase in crop production for biofuels is associated with a 5% increase in land use. This presumably reflects the fact that Banse et al. use land supply curves (as attested by Kløverpris et al. (2008a))454 – a tool which rates converted land as less productive than land currently in use. It has not been possible to discover whether Banse et al. exclude the possibility of yields respecting to demand, or whether this phenomenon is simply weaker, in their work, than the tendency for converted land to have lower yields. It could also not be excluded that, if the results are expressed in financial units, the model predicts that land prices will rise faster than crop prices (i.e., that the ratio would not be so high if expressed in physical units). There is limited support for this hypothesis in the fact that the price of land in four EU Member States (Germany. France, Netherlands and UK) is expected to be an average of 46% higher in the policy scenario than in the reference scenario, while the price of four crops/crop products (cereals, oilseeds, vegetable oil and sugar) is expected to be an average of only 7% higher. On the other hand, the price of land in Brazil (the only other country for which data are given) is expected to be only 3% higher in the policy scenario than in the reference scenario, a fact which points in the opposite direction.455

451 p. 3 452 p. 8 453 pp. 11-12 454 p. 181 455 Banse et al. (2007), pp. 11 and 15; calculations of the Commission services, again relying on measurement off graphs

Page 111: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

111

It would seem that LEITAP modelling uses the land supply curve (which has a biophysical origin) both to determine how much of the extra crop demand caused by biofuels is modelled as coming from land that has to be converted, and also (in a way that is not clear, since the curve plots land rents against land) to determine the yield of that converted land. Tabeau et al. (n.d.) describe the creation of the land supply curves used in the LEITAP modelling: "In van Meijl et al. (2006) the land supply curve was conceptually implemented into the GTAP model. It was derived on theoretical considerations (see Abler, 2003)456 and calibrated using expert knowledge and FAO land use projections. In this paper, we show that detailed biophysical data concerning land use and associated land productivity provide an empirical foundation of the land supply curve in which both land availability and differentiated land quality are included… The assumption that the most productive, i.e. the less expensive to bring into cultivation, land is first taken into production leads to [an] agricultural land supply curve… The information about productivity is provided by biophysical data… Within the IMAGE model the crop productivity is calculated on a grid level of 0.5 by 0.5 degrees… The productivities for 7 food crops457 are calculated in the crop growth model of IMAGE 2.2… The crop growth model… is based on the FAO Agro-Ecological Zones Approach… To capture the overall productivity of each grid cell, the sum of the productivity of the seven crop types is simulated in each grid cell and the average crop productivity is calculated… [W]e transferred each crop productivity to a relative scale between 0 and 1 on the basis of a potential, maximum feasible crop productivity… By ordering all the grid cells in each region from high productivity to low productivity and cumulate the total area, land productivity curves are obtained… The land productivity curve can be translated into land supply curve under the assumption that the land rental rate is a function of the inverse of the land productivity."458 Lywood (2009) implies that EPA do not attribute a lower yield to converted land.459 In Tyner et al. (2009) it is assumed that “the productivity of marginal land is 2/3 of average land”.460

8.3.5. Yields of land converted from other crops Searchinger and Heimlich (2008) explain that “According to the modeling used in our study, as a result of the increase in corn ethanol studied, an additional 25% of the acres harvested for corn grain in 2016 would be corn after corn instead of corn after soybeans, which would have average yields of 154.8 bushels per acre instead of 172 bushels per acre (10% less)… [T]o replace this yield gap would require that demand-induced yields increase sufficiently to provide 12% of the replacement grain”.461 They go on, “We’ve plotted expected corn yields by soil type developed by NRCS, USDA from the best to the worst acres… At the level of corn acreage in the baseline modelling for 2016, the decline in yield between the average corn acre and the next acre of cropland that would be brought into crop production to meet the biofuel demand would be 14%. If we use this drop-off in productivity as a surrogate for the reduced productivity of all new cropland,

456 Referenced as D. Abler, "Adjustment at the Sectoral Level", 2003 457 Listed in a footnote as temperate cereals, rice, maize, tropical cereals, pulses, roots & tubers and oil crops 458 pp. 2, 3, 4, 5 459 p. 10, comparison with CARB 460 p. 19 461 p. 16

Page 112: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

112

demand-induced yield increases would have to provide at least 14% of the replacement grain for yield increases to just match current trends.”462

8.3.6. General In the AGLINK modelling exercise it seems that yield-price elasticities are used for some crops from some countries. It is not clear whether they have been used for all and if not, why not. (Lywood (2009b) states that the OECD model used in the AGLINK work does not take into account of yield growth.463 This seems unlikely.) In the CAPRI modelling exercise, yields respond to demand.464 The degree to which this can happen is, by assumption, capped.465 According to Kløverpris (2009), in CARB, yield increase is “accounted for outside the model”.466 Lywood (2009) states, “The CARB GTAP model … uses an exogenous yield growth rate, but has two additional factors for modelling yield growth: • Yield growth with prices from additional fertiliser use • Factor for lower yield on new land area The sensitivity check in the CARB modelling showed that the overall result was very sensitive to these factors.”467 He adds that the “elasticity factor” used is 0.25 (in a context implying that this is the “yield growth with prices”) and that the source for this value is not given.468 In IFPRI the question of yield response to demand is described as “[s]till [a] research topic”.469 According to Lywood (2009), EPA make an “assumption of constant yield growth rate”.470 He states that FAPRI and FASOM [the two models used by EPA] both do this.471 Keeney and Hertel (2008) similarly state that “Searchinger et al. … inherit the assumption of zero yield response to price from the … FAPRI … modelling framework”.472 However, ADAS UK Ltd (2008) state, “The stochastic models used by FAPRI and OECD do allow some modification of yields by price, though the mechanism for this is unclear”.473

462 p. 16 463 p. 3 464 Blanco-Fonseca and Peréz Domínguez (2009) (p. 10) 465 answer given to a question during a presentation of Blanco-Fonseca and Peréz Domínguez (2009) 466 p. 7 467 p. 10 468 p. 10 469 Valin (2009) (p. 21) 470 p. 7 471 p. 9 472 p. 5 473 p. 15

Page 113: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

113

In the GLOBIOM modelling exercise, “As we represent several crop management systems and allow for endogenous switches between rainfed and irrigated agriculture, the average yield will … be sensitive to the market signals”.474 Keeney and Hertel (2008) initially state that they will set maize’s yield price elasticity at 0.4, the average (to one d.p.) of the seven studies they identified.475 However, the scenario whose results they report appears to be a medium-term one (as opposed to a long term one) in which the yield response is just over 0.3.476 They say that in their work, “we find more than thirty percent of the medium run (five year) output response to a marginal ethanol demand shock is expected to be due to yield gains”.477 It is not clear how this result should be read alongside the results from the same paper relating to the effect of permitting the amount of labour and capital used in sectors of agriculture (see section 8.3.1). In the core policy scenarios478 of Kløverpris et al. (2009), input-driven yield increases are possible in response to demand, but technological development is not. The four scenarios model demand for an additional 0.5 Mt of wheat sourced respectively from Brazil, China, the US and Denmark. The table shows the estimated effects on yields. Table – Yield effects of additional demand for wheat in modelling exercise of Kløverpris et al.

Scenario: CORE TECHNOLOGICAL DEVELOPMENT

modelled demand is for an additional

0.5 Mt of wheat from…

impact on national yields of wheat

impact on national yields of non-wheat crops

expansion in agricultural land per t of wheat (ha)

reduction in expansion

Brazil +1.8% +0.04% 2.0 c. 27% China +0.12% +0.035% 0.26 c. 80% US +0.06% +0.01% 3.2 57% Denmark +1.7% +0.6% 1.7 c. 80% Source: Kløverpris et al. (2009)479 National wheat yields are estimated to grow by 1.8%, 0.12%, 0.06% and 1.7% respectively.

474 ECORYS (2009) (pp. 40-41) 475 p. 11 476 p. 14 477 p. 23 478 “Policy scenario” is a misleading term here, since this work investigates the effect of increased demand for wheat, not biofuels. 479 pp. 6-8, 15

Page 114: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

114

Searchinger and Heimlich (2008) state that the assumption of Searchinger et al. is that “crop yields would not grow faster [in the policy scenario] because positive and negative effects on yields of new demands and higher prices balanced each other out”.480 They add that “price-induced yield increases” would, if acting alone, have been “sufficient to supply 25% of the grain to replace corn diverted to ethanol”. However, this effect is estimated to have been fully offset by the facts that (i) “corn-based ethanol increases reliance on continuous corn, instead of corn/soybean rotations” and (ii) “corn-based ethanol… triggers expansion into more marginal land”.481 Sensitivity analysis explores how different assumptions about yields would affect the results. This is shown in the table. Table – Sensitivity analysis of Searchinger et al. on the impact of different assumptions regarding the effect of demand on yields Central

scenario Original sensitivity case482

Additional sensitivity case483

(A) Proportion of the grain needed to replace maize “diverted” to ethanol that would be compensated by price-induced yield increases

25% 45% 62.5%

(B) Offsetting reduction due to land conversion effects

-25% -25% -25%

(C) Net yield effect in policy scenario relative to baseline scenario

0% 20% 37.5%484

(D) Years during which maize ethanol is calculated to increase greenhouse gas emissions

167 133 84

480 p. 1 481 p. 2 482 In Searchinger et al. (2008) 483 Introduced in Searchinger and Heimlich (2008) 484 calculation of Commission services

Page 115: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

115

Source: Searchinger and Heimlich (2008)485 (For comparability it would be helpful to transform these into elasticities using price effects from Searchinger et al. (2008)) CARD rerun the modelling exercise of Searchinger et al. and test the effect of different assumptions than those made by the authors. One of the scenarios, described as “more along the lines of a thought experiment rather than a scenario run”, tests the effect of yield increases such that in 2018/19, yields of barley, maize, sorghum, soybeans and wheat would be 1% higher than in the central policy scenario.486 The yield response is assumed to be caused by higher prices. The base year and modelled year are probably 10 years apart487, implying a rate of yield increase in the sensitivity scenario that is less than 0.1% per year higher than that in the policy scenario. This small difference in yields has a big effect: greenhouse gas emissions from land use change fall from 113 g/MJ in the (Searchinger et al.-mimicking) policy scenario to 32 g/MJ, a fall of 72%.488 In Tyner et al. (2009), yield price elasticity is set at 0.25.489 It is not clear whether this reflects increased inputs only or also technological development; since the modelling platform is GTAP, the former seems more likely . The topic of yield growth is listed as one for future work.490

8.4. Commentary

8.4.1. Yield increases from increased inputs It is difficult to make sense of the statement on this topic in Searchinger and Heimlich (2008), used to justify the modelling choice of a low or zero yield response to demand. Even if their statement that “The dominance of world agricultural production by mature agricultural economies that already use fertiliser and other inputs at high levels … suggests that technology-improvements, not price-induced increases in inputs, will drive further yield increases”491 were to be correct in relation to mature agricultural economies, it ignores the potential for using inputs to improve yields in less “mature” economies, whose low contribution to world agricultural production can be explained precisely by the fact that yields are low. The authors note this potential in passing but go on to set it aside, without careful analysis, on the grounds that “subsistence producers may not respond to price signals … and government policies … may in effect cushion price effects”.492 That argument may be reasonable in its own right. But if it is to be used it should be substantiated.

485 p. 3 486 Dumortier et al. (2009) (p. 9) 487 According to Dumortier et al. (2009) (p. 2), the original Searchinger et al. work covered a 10 year period up to a modelled year of 2016. However, the text could also be read as meaning that the base year was 2007. The text on p. 3 suggests that the work reported in the sensitivity analyses of Dumortier et al. (2009) has a base year of 2008. This would fit in with 2018/2019 being used as the modelled year, if for example the intention is also to reflect southern hemisphere cultivation periods. 488 Amortisation over 20 years; calculations of the Commission services from 30-year-amortisation data in Dumortier et al. (2009) (p. 13). 489 p. 19 490 p. 33 491 pp. 1-2 492 p. 13

Page 116: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

116

CARB and IFPRI both use GTAP as a modelling platform. IFPRI incorporate a new method by which fertiliser inputs can respond to price; but it is understood that as part of the normal operation of the GTAP model, other non-land inputs can also respond in the same way. If this is understanding is correct, it seems unlikely that Kløverpris is correct to imply that in CARB’s use of GTAP, yield increases are accounted for solely outside the model. It would be good to obtain a better understanding of how IFPRI model fertiliser use, how they model other changes in non-land inputs, and how the two relate to each other. This would, inter alia, help understand how the results would differ if other inputs had been treated in the same way as fertiliser. A tentative conclusion from the detailed description of the modelling approach (“Modelling fertilizers is a delicate task since a simple CES assumption cannot be used to represent the impact of fertilizers on crop yield. Indeed, increasing fertilizer use could allow an increase in yields in the short run. However, some saturation can occur and some countries cannot get higher yield through fertilizers because of an already intensive use of them”493) is that the approach’s novelty consists in the introduction of a saturation effect, which would, presumably, reduce the extent to which the model depicts it being possible to increase fertiliser use as an alternative to increasing land use. It is not clear what empirical data have been used in determining the saturation effect (e.g. the maximum obtainable yield), nor how this effect (including this “maximum”) is modified, if at all, when other yield-improving inputs (say, irrigation) are increased. It would also appear that there is less scope for other non-land inputs to substitute for land than there is for fertiliser. (These inputs would appear to be modelled as a substitute for “land + fertiliser” rather than for land directly.) It is not clear why this is a correct approach.

8.4.2. Changes in frequency of cropping It seems that yield data and assumptions do not generally take into account the scope for changes in cropped area. Similarly, only one modelling exercise has been identified where this factor has explicitly been taken into account. This should be addressed.

8.4.3. Yield increases from technological development It is difficult to make sense of the statements on this topic (justifying the modelling choice of a low or zero yield response to demand) in Searchinger and Heimlich (2008): - It is not clear why the fact that “governments fund more than half of agricultural

research and development” is, as these authors state, a cause (moreover, the main cause) of the fact that “the extent to which higher prices lead to improved technology is largely unknown”.494

- It is not clear why government-funded agricultural research should be “viewed as

independent of biofuels because such funding could, and should, be boosted with or without biofuels”495. "Scenario modelling”, including that of Searchinger et al., aims at describing not what “could” or “should” happen but what will, most likely, happen in practice.

493 Al-Riffai et al. (2010) (p. 87) 494 p. 2 495 p. 14

Page 117: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

117

Cochrane’s “technology treadmill” argument, endorsed by ECOFYS (2009) among others, implies that once innovation has taken place and a small number of acts of adoption have also taken place, it ceases to make sense to see yield increase through technological development as a response to increases in demand or price. Instead, they should be seen as an inevitable response, by farmers who have not yet adopted the technology, to the price reductions that the technology’s initial introduction will trigger. It is not clear that this argument is correct. First, it is normally thought that the market price is determined by the costs of the marginal producer; this argument – if it hinges on the lower production costs permitted by the adoption of the new technology - suggests that it is, instead, determined by those of the cheapest producer. Second, if the argument hinges on the price reductions caused by the extra output that the new technology creates, it is not obvious that adoption by only a small number of farmers will create a large enough price effect to have a significant effect on the margins of the much more numerous remainder. Third, if some farmers nevertheless do face prices that are below their marginal cost of production, it is not correct to argue that they are “forced… to reduce input costs by adopting new technologies in order to maintain a sufficient margin”.496 If the units that they produce vary in respect of their production costs, they can also stop producing the more expensive units (e.g. those from lower-yield land). A more sophisticated theory is therefore needed to explain which of these choices they actually make. ECOFYS (2009) suggest that it is the reduction in public spending on agricultural research aimed at developing countries that has led to the decline in the growth rate of agricultural productivity since the 1980s.497 To evaluate this suggestion it would be helpful to know whether it is in developing countries that the reduced rate of productivity growth has been observed. ECOFYS (2009) present evidence that a decline in the growth rate of agricultural productivity has coincided with a decline in spending on agricultural research, and state that this “strengthens the conception that yield developments are largely dependent on agricultural research”.498 However, correlation (which is all that the evidence presented seems to show) is not causation. They cite studies which find the two phenomena to be “linked”499; without reading the studies it is, again, not possible to tell whether this link is correlative or causative. ECOFYS (2009) ask how far technological development responds to changes in prices.500 It would seem better to ask the broader question of how it responds to changes in demand. This would seem particularly relevant in this case, where it has been argued that most technological development in agriculture is publicly funded. ECOFYS make the point that while this public spending may be “demand driven” it is not “free-market driven”; but this is not relevant here. The question under consideration is whether increased demand for crops,

496 ECOFYS (2009) (p. 3) 497 p. 5 498 p. 5 499 p. 5 500 p. 5

Page 118: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

118

caused by biofuels, is likely to lead to faster yield growth. If increased demand for crops leads to increased public spending on agricultural research, and this leads to faster yield growth, the answer to the question is still “yes” even if market forces are not the mechanism through which this occurs. Commenting on low-carbon technologies in transport, van Dender (2009) states, “In order to create the conditions for such technologies to mature, a strong policy commitment to abatement and the creation of a stable environment for investments and research and development is… important”.501 It seems reasonable to believe that this comment is equally applicable to agricultural research. In that context it can be pointed out that government targets for the use of biofuels/renewable energy in transport, and other biofuel promotion measures, are likely to provide a more stable and predictable framework than other drivers of crop demand. Thus, to the extent that agricultural research is important (in yield growth) and is driven by demand, demand created by public biofuel promotion policies is likely to have a bigger effect (per unit of demand) than other types of demand.

8.4.4. Yields on land converted from non-cropland It seems clear from the description in Searchinger and Heimlich (2008) of Searchinger et al.’s method for estimating the lower yields of converted non-cropland that they assume that (a) it is optimal to use land in order from highest to lowest yield and (b) the current land use pattern is fully optimal. No assumptions could be made that would have deduced, from the dataset they use, a higher ratio between existing yields and “the next acre” of converted land. There are good reasons to consider that these assumptions are not correct (see below). It therefore seems clear that the value Searchinger et al. ascribe to this parameter (US corn yields being set at 14% lower) is higher than the true value. The modelling exercises that use LEITAP use land supply curves derived from van Meijl et al. (2006. Searchinger et al. use a different set of land supply curves. In both cases, the assumption is made that present land use is optimal (that is, that any other land use pattern would produce a lower yield. This assumption leads to a higher requirement for land conversion than would any other. There are good reasons for doubting the correctness of this assumption, for which no empirical evidence has been identified in this literature review. First, even if the market is fully competitive and farmers act as economic optimisers in deciding which land to use, economists would expect them to maximise profits, not yields. This would mean that other factors, such as accessibility and transport costs – and not yields alone - would play a part in determining which land is used.502 Second, there are plenty of reasons to doubt that the decisions about which land is used are in fact economically optimal. For example, - Individual farmers in the EU had to take 10% of their land out of use under the set-

aside policy. There is no reason to think that this led to the worst 10% of land overall being taken out of use;

- Even if the pattern was once optimal, climate change is changing land's potential. The

system may adapt – but not instantaneously; 501 p. 3862 502 The incorrect assumption that there is no difference between optimising for yields and economic optimisation is found, for example, in Tabeau et al. (n.d.). The authors speak of "The assumption that the most productive, i.e. the less expensive to bring into cultivation, land is first taken into production". (p. 3)

Page 119: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

119

- Cultural factors, as well as economic ones, affect farmers' decisions. It is worth noting that, while no data is available on this particular topic, there is plenty of data available on yields obtained from similar land in different countries. These data show great variation in yields, suggesting that decision-making in the agricultural system affecting yields does not, in general, produce results that are anything like yield-maximising. This supports the view that a yield-maximisation hypothesis in relation to land use decisions is untenable in the absence of firm empirical support. Finally, it needs to be asked whether lower-yield land has a tendency to have a lower soil carbon stock (so that its conversion would tend to release less carbon). The modelling assumes it does not: standard values are used for each type of land – see section 10.3. If this assumption is incorrect, then the assumption that the best (highest-yield) land is already in arable use would have a lower impact on the carbon stock losses attributed to biofuels. In this context it is relevant to point out that according to Tabeau et al. (n.d.), the factors used to calculate the productivity values used in the land supply curves include "three soil quality indicators: (1) nutrient retention and availability; (2) level of salinity, alkalinity and toxicity; and, (3) rooting conditions for plants".503 The land supply model of van Meijl et al. uses the concept of maximum potential agricultural land use. In the illustration of the model given in ALTERRA (2009)504, yields fall sharply when land use approaches this limit. It is not clear whether this sharp fall is present in all the regional datasets used (although this is implied by the description in ALTERRA (2009): “When agricultural land use approaches potential land use …, farmers are forced to use less productive land with higher production costs (strongly increasing part of the supply curve)”505) and, if so, whether it results from empirical data or is assumed. If the latter, it is important that the limit be correctly estimated. If, for example, nature protection areas are excluded in calculating potential land use, these could have high potential yields. If so, the assumption of sharply declining yields could be unjustified. To be classified as a land scarce region, in the modelling exercise of LEITAP/ALTERRA, means that yields of converted land will be considered to be much lower than those of existing agricultural land (because they are in a part of the land supply curve where marginal yields fall sharply). Europe is classified as such a region506; given the large amount of recently abandoned agricultural land in Europe, it is not clear why this is. In empirically implementing the concept of land supply curves, it seems that van Meijl et al. use data from IMAGE. It seems that LEITAP/ALTERRA use the same data except for Europe, where they use data from CLUE. It would be appropriate to compare the different land supply curves that result from these choices. The land supply curves in van Meijl et al. and LEITAP/ALTERRA plot land against land rents. It is not clear what the relationship is between these land rents and the biophysical data which underlie the curves. The land rent data are used to calculate how much of the extra demand has to come from new land. Presumably, the underlying biophysical data are then used to calculate how much new land will be needed; it is not clear how this is done. 503 p. 4 504 Figure 2, p. 8 505 p. 7 506 p. 7

Page 120: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

120

Lywood (2009) criticises the attribution, in modelling, of lower yields to converted land using the following arguments507: i) Historic yield data already take output growth into account. If the baseline rate of yield

growth between the base year and the modelled year is estimated from historic yield data, the effect of area changes is already taken into account.

(This argument does not seem valid. First, according to FAO data there was little net change in cropland from 1980 until relatively recently, while significant expansion is expected in most modelling exercises even without biofuel promotion. Second, at least in the case of the work that uses land supply curves, this argument is allowed for. It is acknowledged that existing cropland has different levels of productivity; the current average yield is attributed to the average-yielding piece of land that is estimated to currently be in use.)

ii) Analysis using Lywood’s technique for estimating the relationship between demand

and yield never show yield growth rates falling when prices or output growth rates rise; if they did, this would indicate lower productivity of converted land.

(It is true that evidence of a negative relationship between yields and output would strongly suggest lower productivity of converted land. But the absence of such a relationship does not prove that converted land has no lower productivity. This negative effect could simply be offset by other, positive yield responses to demand changes. In any case, this argument relies on the appropriateness of Lywood’s technique for estimating the relationship between demand and yield, which is questioned in section 8.4.6.)

iii) A country-level comparison of maize yields in 1995 and 2007 shows that increased

output has been achieved at all yield levels. (It is not clear why this is relevant.) As a Chinese example cited by ECOFYS (2009a)508 shows, degradation can lead the yields of existing agricultural land to fall. In such circumstances it cannot be excluded that converted land would have higher yields than existing land, not lower. Tabeau et al. (n.d.) state that in van Meijl et al. (2006) "the land supply curve was conceptually implemented into the GTAP model. It was… calibrated using expert knowledge and FAO projections". By contrast, the authors state that in Tabeau (n.d.) itself "we show that detailed biophysical data…. provide an empirical foundation of the land supply curve".509 This seems to imply that the same team510 have provided land supply curves that are calibrated using two different sets of data. If this is correct, it is not clear which set of data has been used by the modelling exercises that use land supply curves referenced to this work (i.e.,

507 p. 13; a fourth argument, on p. 14, appears to be a reprise of the first. 508 p. 1 509 p. 2 510 the authors of Tabeau et al. (n..d) are also authors of van Meijl et al. (2006)

Page 121: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

121

LEITAP). (On visual inspection, the curve for Canada used as an illustration in Tabeau et al. (n.d.)511 looks the same as that used in the IFPRI work512.) Tabeau et al (n.d.) state that "The land productivity curve can be translated into land supply curve under the assumption that the land rental rate is a function of the inverse of the land productivity."513 This appears to be the approach used to generate the land supply curves used in the LEITAP modelling. It seems, at least at first sight, to be a rather odd assumption: one would naturally expect rents to fall rather than rise as land quality declines. Other land use change modelling with GTAP - Tyner et al., CARB and IFPRI - uses arbitrary assumptions for the yield of new cropland (respectively 2/3, ½ 514 and ½ (but ¾ in Brazil515) of the yield of existing cropland). The first two of these exercises depict land use change as happening mainly in the US. Given that the US has a large stock of recently abandoned arable land, it seems appropriate to exclude the possibility of unmanaged land being converted (it is not clear whether these two modelling exercises do, in fact, exclude this). If that is so, it is worth noting that the land supply curves (derived from empirical data) appear to show that the minimum possible figure (based on the unrealistic assumption that the land currently in arable use is precisely that with the highest yields) is about 82% for the US (82% is also the median for all regions).516 If the data and method used in the calculation of the land supply curves can be relied on, it seems clear, therefore, that the figures used by Tyner et al. and GTAP are too high. IFPRI themselves point out in this context that "recent research seems to show that recent marginal land extension were taking place on land with at least average level yields".517 It should also be noted that more recent work by Tyner et al. replaces this assumption with one based on data on the actual yields of land used for crops, and derives yields on marginal land whose median is more than 90% of that of existing arable land.518

8.4.5. Yields on land converted from one crop to another Searchinger et al. take (negative) account of the fact that the encouragement of maize production for US biofuels makes continuous maize crops more likely than maize/soya rotations. If it is correct to do this, it would seem correct also to take (positive) account of the fact that the encouragement of rapeseed production for EU biofuels makes cereal/rape rotations more likely than continuous cereal crops.

8.4.6. General It is difficult to make sense of several of the statements on the topic of the response of yields to demand in Searchinger and Heimlich (2008):

511 p. 5 512 IFPRI (2009) 513 p. 5 514 With sensitivity tests for 0.25 and 0.75 515 Al-Riffai et al. (2010) (p. 98) 516 This conclusion is based on analysis of the land supply curves that, at one stage, IFPRI intended to use in determining the relative yield of converted and existing land – see Al-Riffai et al. (2010) (p. 98). 517 Al-Riffai et al. (2010) (p. 98) 518 Tyner et al. (2010) (p. 65)

Page 122: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

122

- The fact that there is a statistical correlation between higher crop prices and cropland expansion does not, as stated, support the hypothesis that extension rather than intensification can be expected to account for a large share of the response to demand increases.519 If higher crop prices reflect higher demand, the rate of use of all inputs to crop production can be expected to be correlated with crop prices, whether they play a large or a small share in meeting increases in demand.

- The fact that there are diminishing returns to scale in the use of non-land inputs does

not mean, as stated, that the share of land can be expected to be higher where demand/price increases are large520 – since the authors assume that there are also diminishing returns to land conversion. There is no obvious a priori reason why the rate of diminution should be expected to be higher for non-land inputs than for land.

Mathews and Tan (2009) refer to Searchinger et al.’s assumption that demand/price increases from biofuels will cause a (small) increase in yields. They then quote Searchinger et al. as stating that “Even if excess croplands in the US or Europe become available because of dramatic yield improvements beyond existing trends… biofuels would still not avoid emissions from land use change” and that this is because “truly excess croplands would revert either to forest or grassland and sequester carbon”. They conclude from these two citations that “[A]pparently it doesn’t matter what the yield improvements might be, according to Searchinger et al. If they are small, then more land will be converted to grain, thus releasing carbon. If the yield improvements are large, then fewer lands will be needed for agriculture and some can even revert to forest or grassland thus sequestering carbon (which might be characterised as negative land use change).”521 This would appear to be a misreading of Searchinger et al. The first-mentioned assumption refers to a yield effect that would differ between the policy scenario and the baseline scenario, while the passages that are then cited almost certainly522 refer to the Searchinger et al.’s evaluation of the impact of a higher rate of yield increase in the baseline (affecting both the baseline scenario and the policy scenario). There appears to be no reason to doubt that, to the extent that Searchinger et al. see biofuel promotion leading to higher yields, they evaluate this as reducing the amount of land use change attributed to the policy of biofuel promotion. Commending EPA’s modelling approach when compared to CGE modelling, Searchinger (2009a) comments on CARB as follows: “Production functions play a critical role in models such as GTAP, which allows them to adjust production levels in responses to changes in supply and price of various inputs… However, the empirical basis for these production functions is extremely weak, making them the subject of enormous criticism within the economics literature. The form of the production functions is also typically chosen, as in GTAP, for its ease of mathematical manipulation. The limitations are sufficiently strong that when Purdue University economists were adjusting the GTAP model to calculate indirect land-use change for the California Air Resources Board, they forced the production functions to reproduce a yield/price elasticity in theory derived from econometric studies. Even if that overall elasticity were valid (and its empirical basis was also weak), the overall elasticity would not tell you what variables to adjust to produce that elasticity. Because the relationship of the supply and price of these inputs to outputs is therefore based on limited empirical basis,

519 p. 9 520 p. 13 521 p. 8 522 [subject to verification]

Page 123: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

123

it is not particularly helpful to vary those input supplies and prices in responses to general equilibrium features.”523 Searchinger and Heimlich (2008) and Keeney and Hertel (2008) draw different conclusions from a similar set of studies of yield responses to price. An important reason for the difference seems to be that Keeney and Hertel think that data from the 1950s and 1960s, which dominate these data524, are relevant, while Searchinger and Heimlich think they are not, because this was before “full use of nitrogen fertiliser had been widely adopted”.525 A definite conclusion that can be drawn from this is that it would be good to obtain a wider collection of studies that use more recent data. A second, more tentative conclusion is that the approach adopted by Keeney and Hertel might still be relevant for economies with low use of fertiliser, even if it cannot yet be said whether it is relevant for those with higher fertiliser use. A final comment is that modelling exercises such as IFPRI are still being conducted on the assumption that there is scope to obtain yield gains through additional fertiliser application, implying that Searchinger and Heimlich’s expression “full use of nitrogen fertiliser” may be misleading. The method used in Lywood (2009a) to calculate correlations between yield growth and output growth appears methodologically weak. A time series approach would have been preferable. In addition, the fact that data for each year (except for those at the start of the dataset) are taken into account in four different points seems likely to create false correlations. Finally, it is not obvious that cause and effect (as between yield growth and output growth) have been correctly distinguished. According to Lywood (2009b), general equilibrium models “relate yield to price instead of yield growth to price”. (Another way of thinking about this might be to say that these models may allow for inputs to change in response to demand, but not for technologically to develop in response to demand. Input-related yield improvements will fall away when prices fall back; technological improvements will not.) He also comments that they cannot use historic data for yield growth. He adds that partial equilibrium models assume a fixed rate of yield growth based on technological development.526

8.5. Impact of public policy

8.5.1. Introduction The EU legislative framework includes measures to require biofuels used towards EU targets to achieve a minimum greenhouse gas saving and not to use raw material from certain types of land (including land with high carbon stocks). The modelling should assess and take into account the impact of these restrictions. Public authorities impose more general limits on the use of land, e.g. for nature protection purposes. These should also be taken into account.

523 p. C2 524 see Keeney and Hertel (2008) (p. 28) 525 p. 11 526 p. 3

Page 124: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

124

8.5.2. Taking into account in modelling of requirements for minimum greenhouse gas saving

Greenhouse gas savings vary between types of biofuel (both in relation to feedstocks and in relation to production processes). It has not been possible to identify how any of the modelling exercises determine which types of biofuel are considered to be consumed. In any case, none of the modelling exercises that assess the impact of the EU target for renewable energy in transport in 2020 appear to take into account in any way, in determining the types of biofuel consumed, the requirements for a minimum greenhouse gas saving. Examining cultivation practices in British farming, Woods et al. (2008) show that there is substantial variety in practices that can reasonably be expected to affect greenhouse gas emissions.527 WWF Germany (2007) do the same for palm oil.528 This reinforces the view that requirements for minimum greenhouse gas savings can be expected to affect the greenhouse gas savings from the biofuels consumed.

8.5.3. Limits on land use – empirical evidence There is a broad consensus in the literature that land use policies of one kind or another have an effect on land use. For example, - Tabeau et al. (n.d.) note that "The supply of agricultural land depends on… [inter alia]

institutional factors (agricultural and urban policy, policy towards nature)…".529 - Concerning the conversion of forest to cropland, Zanchi et al. (2007) state, "In some

[European] countries national legislation is already in place to restrict the re-conversion of afforested area back to other land uses. For example, in several countries the conversion to forest land is legally binding after a certain period… Where the law does not require an irreversible land-use change the area can be reconverted to the previous use and the risk of reconversion is generally higher on productive agricultural land."530

- Concerning the conversion of cropland to forest, Zanchi et al. (2007) state that

"Different proportions of forest area change are ascribable to afforestation activities… According to the figures reported or in the judgement of national expertises, Croatia, Czech Republic, Ireland, Finland, Hungary, Netherlands, Poland and the UK have an afforestation rate that accounts for all the forest expansion."531

- Commenting on a reduced rate of deforestation in Brazil in 2009, Greenpeace Brazil

commented that "whenever the government followed the law, deforestation fell".532

527 p. 25 528 p. 5 529 p. 3 530 p. 6 531 pp. 32-3 532 The Guardian, 23 November 2009

Page 125: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

125

- Lambin (n.d.) points to countries where a "Mix of good policies, economic reforms and cultural changes can restore forests and spare land… China, Vietnam, Bhutan, Costa Rica, El Salvador, Dominican Republic, Panama…"533

There is also some (anecdotal) evidence that farmers' choices about whether to fulfil sustainability criteria of one type or another depend on whether there is a premium payment for crops which can be shown to have done so. John Fagan of Cert-ID ("the leading US certifier of non-GM soya for import from Brazil to Europe") states that "Brazil will continue to plant non-GM as long as it gets paid to keep different supplies segregated".534 Palm oil certified by the Roundtable on Sustainable Palm Oil (RSPO) first came on the market in late 2008. It initially commanded a premium of $50/t. The premium fell to "practically nil" in the first half of 2009, before returning to $15-$20/t by August, with "certain big integrated players enjoying better premiums for traceability" (presumably through a mass balance system – as in the Renewable Energy Directive – or identity preservation system, as opposed to the book and claim system that can also be used under RSPO).535 In November 2009 the premium was quoted as $10-$40/t.536 According to UNEP (2009), "In the case of ecolabelling schemes, it has been shown that their introduction could – under some conditions – cause negative effects on the environment because of the rebound effect."537 The mechanism UNEP believes to be at work is not explained. Searchinger (2009) argues that as a matter of principle it is apparent that restrictions on land use change like those in the Renewable Energy Directive will have "little" effect: "The EU [Renewable Energy] Directive… [and] the [US] Energy Independence and Security Act… [both include] some restrictions on the kind of lands that can be directly plowed up to grow biofuels… Unfortunately these land use rules by themselves will have little significance because any producer could get around them simply by building two tanks. For example, a palm oil producer could use one tank to hold all food palm oil produced from already cleared forest and direct that oil to biodiesel and qualify for a European mandate. But a second tank could receive palm from newly cleared plantations and supply the world's great demand for palm oil for food. For the world's land use, the important factor is the total level of crop demand, not the precise origin of any particular crops."538

8.5.4. General limits on land use – treatment in modelling exercises It seems likely that the impact of limits that exist in the base year is taken into account. Where the "historical" approach is used to move from hectares to land types, this will certainly be the case. It is likely also to be the case, at least to some extent, where the "suitability" approach is used. (See chapter 9.) As an example of the latter, Tabeau et al. (n.d.) treat protected land as unavailable for conversion to agriculture in developing their land supply curves.539 533 p. 15 534 The Guardian, 27 October 2009, p. 23 [IY] 535 M.R. Chandran, Advisor to Roundtable on Sustainable Palm Oil, personal communication, 21 August 2009 [ES] 536 Daily Telegraph, 2 November 2009, p. B6 537 Referenced to J. v. Geibler, "Biomassezertifizierung unter Wachstumsdruck", Wuppertal Institute Paper 168, 2007 and to Bougherara et al. 2005. 538 p. 22 539 Hertel et al. (2008) (p. 16)

Page 126: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

126

However, the area of land covered by nature protection areas has grown consistently over time. By 2009 the protected area had reached 1.8 bn ha, 12.1% of global land area, having increased steadily from 1872540 and at a rate of 1.8% a year since 1990.541 It seems reasonable to assume that the area is likely to continue to grow in future up to the modelled year. Any reinforcement of the UNFCCC carbon accounting regime in relation to land use (LULUCF and REDD) will only tend to complement this trend. This should be taken into account in the modelling exercises but does not appear to have been addressed in any of them. Modellers and commentators have recognised inability to model restrictions relating to the use of land as a weakness: - Hertel et al. (2009) recognise that the modelling tools are still poor in relation to their

ability to take into account "non-primary demands for land": "In the long run the demand for land in parks for recreation and the preservation of ecological diversity is likely to be very important. But these sectors have not yet been well-developed in global CGE models. Improving their specification, as well as estimating how the demand for their services is likely to grow with higher incomes, will necessarily precede the incorporation of these land demands into CGE models."542

- Lambin et al. (2000) refer to the need for land use models to "provide plausible

representations of alternative futures where these are unknown, uncertain or may contain 'surprises'. Uncertainties and unknowns might include… the role of regulation, policy-making and political change."543

- Bouët et al. (2009) state, speaking of CGE models, "A number of issues which impact

on the land use dynamic, but are independent of commodity market effects, cannot be properly reproduced. This is the case, for example, for measures related to environmental protection…".544

- Kløverpris et al. (in press) comment, "[Our] modelling could be improved by

incorporating a mechanism simulating legal fertiliser and pesticide restrictions".545 - Babcock (2009) states, "More often than we want to admit, economists face situations

in which we do not have adequate data to make precise estimates of the response of a sector to a price change. The backup strategy is to rely on economic theory to determine the direction of the response, and then to make a reasonable assumption about the magnitude of the response… Thus, economists know that the Brazilian cattle herd will increase by some amount if U.S. meat supplies decrease. But an informed judgement about the magnitude of the change will rely [inter alia]… on [a] dedicated Brazilian agricultural economist with detailed knowledge of Brazilian environmental enforcement mechanisms… [making] an estimate of the extent to which pasture can expand in frontier forests… Most of the parameters used to capture supply and demand responses to price changes that populate the models economists use to

540 World Database on Protected Areas (2009) 541 Millennium development goals indicators, indicator 7.6, downloaded on 6 January 2010 [LA] 542 p. 20 543 p. 329 544 p. 73 545 p. 1

Page 127: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

127

estimate the impact of biofuels on land are based on less detailed knowledge."546 (An example of a similar approach, subsequently incorporated for Brazil in the FAPRI model for which Babcock is responsible, can be found in Nassar et al. (2009).547 This highlights the absence of such data in the modelling of other parts of the world and in other modelling exercises.)

- Wang (2009) comments, "While indirect effects can be caused by economic factors,

political and social factors can play an important role in the magnitude of the economic linkage between direct actions and indirect effects… Efforts have been made so far to examine land-use changes (LUCs) of biofuel production solely from economic factors. One could argue that the other two sets of factors, such as those through government intervention, can weaken (or strengthen) the economic linkage between direct actions and indirect effects, which has not been addressed in the current efforts of examining LUCs of biofuel production."548

- Houghton (2009) comments, "It's possible that future changes in the value of carbon

could encourage nations to manage wood products differently from the way they are managed currently. Thus, while forests cleared for soybeans today might be burned, with different incentives they might be harvested and turned into long-term products for the future. To the extent that such changes occur, it would not be justifiable to assume the same overall lifecycle impact."549

LEITAP/ALTERRA includes a modelling exercise that could be seen as estimating something that resembles an upper limit for the impact of one variant of such restrictions. They model the impact of global biofuel targets in 2020 (including 10% in the EU) with and without a "strict forest policy" comprising "protection of forest and woodland systems" and leading to "no forest conversion". This reduces the potential global agricultural area from about 8 bn ha to about 4 bn ha (of which a little less than 3 are already in agricultural use). Working with land supply curves, they estimate the land use change impact. If forest conversion were to be banned (reducing the attractiveness of extensification as opposed to intensification as a means of meeting the additional demand for crops), they estimate that the increase in the agricultural area between 2001 and 2030 would be 35% less, and greenhouse gas emissions from the resulting land use change would be 52% less.550 The GLOBIOM modelling exercise also includes sensitivity tests of the effects of preventing deforestation. If deforestation is not prevented, the modelling shows biofuel payback periods of about 15-114 years for 12 different scenarios (average: 46 years, median: 46 years). With deforestation prevented, the payback period falls, under every scenario, to 5 years or less.551

546 p. 5 547 p. 13 548 p. E1 549 p. C2 550 Calculations of the Commission services from data in Woltjer and Prins (2009) (pp. 4, 6, 7, 14, 15). Note: it is not clear whether the comparison is between the two variants of the policy scenario (with and without forest constraints) and (i) the baseline scenario or (ii) the base year. The former is the variable of interest and is what has been assumed here. If the comparison given in Woltjer and Prins is, instead, with the base year, it can be expected that the effect in relation to the baseline scenario would be even greater than the results cited. 551 Calculations of the Commission services from data in Havlík et al. (2009) (pp. 19 and 21). Note: these values should be treated as approximate since they were obtained by measuring off a graph.

Page 128: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

128

8.5.5. Restrictions on biofuels from certain types of land – treatment in modelling

New Fuels Alliance (2008) comment that "[T]he GTAP figures presented by [CARB] staff… were neither sensitive to U.S. federal biofuels policy, which contains land use provisions designed to discourage certain types of land conversion, nor the energy or land use policies in those countries where the land conversion actually takes place in the scenarios modeled".552

8.5.6. Commentary Searchinger's argument that restrictions on biofuels from certain types of land will have "little significance"553 ignores the fact that, if crops produced in a certain way command a premium price, it will be more attractive than it otherwise would have been to choose this way of expanding production rather than another way. It also ignores the potential cost to producers of the bad publicity that could come from evading sustainability requirements in the way that Searchinger describes. It is not immediately obvious how the land use restrictions laid down in the EU legislation, or more general restrictions on land use, should be modelled. One approach could be to take inspiration from one of the ways in which trade modellers have converted "various types of limitations on foreign service suppliers" into "tariff equivalents" in modelling barriers to trade in services. This has been done by estimating the volume of trade that would take place under free trade conditions and comparing it with the volume of trade that has in fact taken place.554 The analogy here would be to estimate the volume and type of land conversion that would take place if land suitability were the only deciding factor, and compare it with the land conversion that has, historically, actually taken place. Another approach could be to estimate the premium at which it is likely that biofuels proven as meeting the sustainability criteria will be able to sell, and model the effect of this premium on farmers' choices. In a text which has not yet revealed its full meaning, ECORYS (2009) allude to what appears to be one or two further possible approaches: "The calibration procedure for the G4M555 yields quantitative estimations of so called hurdle rates of land use change including net and gross deforestation. These hurdle rates contain information of governance and other policy factors impacting land use change, in particular, deforestation. "Visual" inspection of the numerical estimates across countries confirm that the values make sense and yield sensible interpretation. However, statistical analysis of the ensemble of hurdle rates in a global country cross section have not lead, so far, to statistically valid models of explaining hurdle rates with a regressor matrix of potentially interesting candidates such as degree of rule of law indicators, total equipment investment, GDP, remaining forest area. Thus, we suggest an alternative way to conduct scenarios with hurdle rates. We propose to conduct two scenarios: Driver governance:

552 p. 4 553 Searchinger (2008) (p. 22) 554 Piermartini and Teh (2005) (pp. 34-5) 555 It has not proved possible to identify what the G4M is.

Page 129: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

129

- Scenario option 1: Keep hurdle rates constant over time - Scenario option 2: Let hurdle rates converge to a pan-tropical target value (e.g.

current average +/- X) Currently there is modelling work under way in order to make hurdle rates consistent in both the GLOBIOM and G4M models. If successful we anticipate scenarios to be delivered which are consistent in both models with respect to "aggregate" governance and policy information."556

9. Model calculations: from hectares to land types

9.1. Introduction Where models calculate that land will need to be converted from non-cropland to fulfil demand for crops (see chapter 8), it is necessary – in order to estimate the carbon stock impact (see chapter 10) - to determine which types of land are considered to be converted. In the literature, two main methods of doing this have been identified: the historical approach and the suitability approach. Under the historical approach, data are obtained for the types of land that were converted during some previous period. It is assumed that in the future, land will be converted in the same proportions. In regions that experienced conversion from cropland to other land uses during the period in question, it is assumed that in the future, land will be converted from those uses to cropland in the same proportions as it was converted from cropland to those uses. Under the suitability approach, the land that is assumed to be converted is the land that is considered most suitable. This can be according to biophysical criteria or according to economic criteria.

9.2. The historical approach

9.2.1. Description of studies using the historical approach According to Kløverpris (2009), in the version of the GTAP model used by CARB, there is a “fixed amount of land for agriculture and forestry … à No expansion, only redistribution of land use”.557 Ensus use the historical approach. In the Ensus method, “The type of land used for increased crop demand is primarily determined by matching areas of increased crop land to areas where there are changes in natural vegetation, using historic data on the type of land use change for each crop. For those countries with the largest increases in land area for a crop, the increase in crop area has been compared with the changes in areas of forest and grassland and cropland in those countries. Where increases in land area of the crop correspond to decreases in forested area, it is assumed that the crop has been grown on deforested land.”558 The text goes on to 556 pp. 43-4 557 p. 7 558 Lywood (2009d) (p. 9)

Page 130: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

130

imply that a similar approach is taken for grassland and for idle land, so that in each country the growth in land area for each crop is assigned to one or more of these three land types. This approach is justified on the basis that, for any given crop, there is a good correlation between the change in land area in 1991-2001 and the change in land area in 2001-2007.559 Because the Ensus approach is crop-specific, the historic data can show that land for some crops grows at the expense of land for others. In a second-round calculation, this land is reattributed to one of the other categories.560 EPA use the historical approach. According to Valin (2009) they have built “a precise historical database relying on remote-sensing data”. This work has been done for EPA by MODIS.561 It is not clear whether these data show only net land use change, or whether they show what the previous land use was of land that has been converted specifically to cropland. According to ICF International (2009a), the work was done for EPA by Winrock International and consisted of “determining the extent of land use change using MODIS imagery from 2001 and 2004”. Winrock “conducted the satellite imagery change detection analysis and determined which land use types decreased or increased at the country level during this time period”.562 In the work of Kløverpris, “The nature types (biomes) affected by agricultural expansion are determined from land cover maps and FAOSTAT data.”563 It is not possible “in the model” (an application of GTAP) for land to shift between managed forest and agriculture, but “outside the model… modelled agricultural land expansion can take place at the expense of forest, managed or unmanaged”.564 Land supply curves are added to the standard GTAP model. The amount of land used for agriculture is a function of its price; its price is a function of (i) the proportion of land usable for agriculture in the region that is not used for agriculture and (ii) a region-specific coefficient.565 This modelling exercise assesses the impact of an increase in demand for wheat in a particular country. First, the net expansion in cultivated land is calculated. Then, “the likely geographical location of expansion within the region is …identified”.566 To do this, it seems that the following procedure is used: - Data from Ramankutty et al.567 are used to establish, for each region, the quantity of

land usable (under rain fed conditions) for (i) cultivation or grazing and (ii) grazing only (replacing the “single land type used in the standard GTAP model” with these two land types), as well as the proportions of these land types actually used for each of these purposes in a base year (possibly 2000)568;

559 Lywood (2009c) (p. 6). The correlations given are wheat 0.77, maize 0.93, rape seed 0.59, sugar cane 0.85, soya 0.98, oil palm 0.80 and sugar beet 0.74. 560 see Lywood et al. (2009d) (pp. 18-19) 561 p. 20 562 p. I-1 563 Kløverpris et al. (in press) (p. 2) 564 Kløverpris et al. (in press) (p. 2) 565 Kløverpris et al. (in press) (p. 3) 566 Kløverpris et al. (in press) (p. 5) 567 [Referenced as Ramankutty, N. et al., “The global distribution of cultivable lands”, Global Ecology and Biogeography 11 (2002) and Ramankutty, N. et al., “Farming the Planet: 1. Geographic distribution of global agricultural lands in the year 2000”, Global Biochemical Cycles, in press (2007).] 568 Kløverpris et al. (in press) (p. 4)

Page 131: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

131

- Data from Ramankutty and Foley for 1970 and 1990569, as well as FAOSTAT’s

annual statistics, are used to determine “the trend in utilisation of the two land types”570;

- “The trends in croplands and pastures are used as indicators of where in the region the

expansion predicted by the GTAP Model occurs”571; - “The potential natural vegetation (biome) of the areas affected by expansion is

determined by identifying these areas on a digital biome map with a global overview of 15 potential natural vegetation types.572

It is reasonably clear that this is a type of historical method; it is difficult to compare it with other examples of the historical method, because the way in which “trends … are used as indicators” (in the third step) is not explained. Although an annex to Kløverpris et al. (in press) gives several equations, it does not include those used at this step. Searchinger et al. use the historical approach. They state, “we used data compiled at the Woods Hole Research Center to estimate the net conversion of different types of ecosystems (for example, forests or grasslands) to cropland in each major world region over the period of 1990-99. We apportioned the additional cropland in each country to different ecosystem types according to the proportion of ecosystem converted in the 1990’s.”573 The ten regions used are the US; North Africa and the Middle East; Canada; Latin America; “Pacific Developed”; South and South East Asia; [Sub-Saharan] Africa; India, China and Pakistan; Europe; and the Former Soviet Union.574 An ad hoc modification to the method is made for the India/China/Pakistan region on the following grounds: “The data for land use change in Asia in the 1990s were dominated by conversion of changes [sic] in southeast Asia to forest. Because there is less forest in China, India and Pakistan and because our analysis predicts a large amount of conversion in those countries, we made a conservative, simplifying assumption for those countries that all conversion would come out of grassland.”575 This suggests that the historical land use change data used by Searchinger et al. are even more coarse grained than the 10-region breakdown would suggest (this text implies that the underlying data treat Asia as a single region). It is not clear why an ad hoc modification is appropriate here and not in other regions. Searchinger also states, “We … assumed that new cropland in the future would reflect the patterns of new cropland in the 1990s, roughly split between forest and grassland/savannahs”, adding, “Some observers misinterpreted the amount of forest conversion estimated in [Searchinger et al. (2009)] because the category of tropical open forest in Latin America, perhaps inappropriately labelled, essentially represented Brazilian Cerrado savannah”.576 Searchinger defends the use of the historical approach on the

569 [Referenced as Ramankutty, N. and J. Foley, “Estimating historical changes in global land cover: Croplands from 1700 to 1992”, Global Biogeochemical Cycles 13 (1999)] 570 Kløverpris et al. (in press) 571 Kløverpris et al. (in press) (p. 5) 572 Kløverpris et al. (in press) (p. 5); [map referenced as Ramankutty, N. and J. Foley, “Estimating historical changes in global land cover: Croplands from 1700 to 1992”, Global Biogeochemical Cycles 13 (1999)] 573 Searchinger et al. (2008) (p. 4) 574 Searchinger et al. (2008) (pp. 30-39) 575 Searchinger et al. (2008) (p. 37) 576 Searchinger (2009) (p. 15)

Page 132: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

132

basis that “That pattern [of the origin of new cropland in the 1990s] inherently reflected the various forces pushing land in one direction or another”.577 It can be seen that the method used by Ensus is crop-specific while that used by the other studies reported here is not.

9.2.2. Summary - time periods used As shown in the table, practitioners of the historical approach use different time periods to establish the assumed ratio of conversion from different land types. Table – Time periods used by practitioners of the historical approach Study Time period Searchinger et al. 1990-1999 Ensus not stated IFPRI (for exogenous land use change in baseline): 2000-2004

Kløverpris et al. 1970-2000 or later? EPA 2001-2004 Source: see above

9.2.3. Summary - regions used in the establishment of historical averages

The choice of regions of IFPRI is not clear. Valin (2009)578 implies that national data are used. Ensus only address the major producers, but do so at a more fine grained level, using country data. Kløverpris et al. use 8 world regions.579 Searchinger et al. use 10 world regions (although there are indications in the text that a cruder breakdown also plays a part ).

9.2.4. Summary - historical rates of land conversion used The table shows the rates of land conversion estimated (historically) and used (for the future) by some practitioners of the historical approach. (As a further step, it would be interesting to compare these results with data from FAO ResourceSTAT.)

577 Searchinger (2009) (p. 15) 578 p. 20 579 Kløverpris et al. (in press) (p. 9)

Page 133: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

133

Table – Origins of land conversion (to cropland) estimated by practitioners of the historical approach Searchinger et al. Ensus (crop-specific) EPA crop-specific?580

x ü x

US broad leaf forest 2% mixed forest 34% coniferous Pacific forest 2% grassland 62% total 0.5 Mha/year

grassland 100% forest 7% grassland 61% savannah 21% shrub 10% total 4.9 Mha/year

NA & ME tropical grassland 50% desert scrub 40% tropical woodland 10% total 6.0 Mha/year

Canada temperate evergreen forest 20% temperate grassland 80% total 2.2 Mha/year

idle land 100%

Latin America tropical evergreen forest 3%- tropical seasonal forest 22% tropical open forest 47% temperate evergreen forest 3% temperate seasonal forest 1% grassland 24% desert 1% total 22.1 Mha/year

Argentina (temperate crops): grassland 100% (soya): forest 19% grassland 67% idle land 15%581 Brazil (soya): forest 50% cropland 50% (sugar cane): grassland 100% Paraguay: forest 50% grassland 50%

Argentina: forest 8% grassland 40% savannah 45% shrub 8% Brazil: forest 4% grassland 18% savannah 74% shrub 4% total 2.7 Mha/year

“Pacific Developed”

tropical moist forest 15% tropical grassland 60% tropical woodland 25% total 3.3 Mha/year

Australia: grassland 100%582

South and South East Asia

tropical moist forest 76% tropical seasonal forest 19% open forest 5% total 24.3 Mha/year

Indonesia: forest 98% grassland 2% Malaysia: forest 97% other land 1% Thailand: forest 100%583

Indonesia: forest 34% grassland 5% or 4%584 savannah 58% shrub 4% total 1.0 Mha/year

580 In crop-specific work, a possibility that is taken into account is the conversion of land from another crop to the specific crop in question. In non-crop-specific (all-crop) work, this is not relevant. “Cropland” can thus appear as a land origin in crop-specific work, while it cannot in all-crop work. 581 figures obtained following a ‘3rd round’ reattribution 582 NB: figure for temperate crops

Page 134: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

134

Malaysia: forest 74% grassland 3% savannah 19% shrub 3% Philippines: forest 49% grassland 5% savannah 44% shrub 3%

Sub-Saharan Africa

tropical rain forest 11% tropical moist forest 38% tropical dry forest 22% shrub 14% montane forest 15% total 23.2 Mha/year

Nigeria: forest 86% grassland 14%585

Nigeria: forest 4% grassland 56% savannah 36% shrub 4% South Africa: forest 10% grassland 22% savannah 53% shrub 15%

India, China, Pakistan

grassland 100%586 India (soya): forest 40% grassland 60%587 (sugar cane): forest 31% grassland 69%588 China: grassland 40% idle land 60%

India: forest 7% grassland 7% savannah 33% shrub 53% China: forest 17% grassland 38% savannah 23% shrub 21%

Europe temperate evergreen forest 25% temperate deciduous forest 25% boreal forest 25% temperate grassland 25% total – 2.0 Mha/year (loss of cropland)

EU: idle land 100%

EU: forest 27% grassland 16% savannah 36% shrub 21%

FSU temperate evergreen forest 16% temperate deciduous forest 11% temperate grassland 74% total – 3.2 Mha/year (loss of cropland)

Ukraine: grassland 11% idle land 89%

period 1990-1999 not stated 2001-2004 Sources: Searchinger et al. (2008)589, Lywood (2009d)590, Valin (2009)591, Nassar (2009)592 583 NB: figure for oil palms 584 Valin (2009) gives 5%. Calculations from data in Nassar (2009) give 4%. 585 NB: figure for oil palms 586 ad hoc modification of the method (see above) 587 figures obtained following a ‘3rd round’ reattribution 588 figures obtained following a ‘3rd round’ reattribution 589 pp. 30-39 590 p. 10. The figures for Australia, Canada, EU and Ukraine are for “temperate crops”. The figures for Indonesia, Malaysia, Nigeria and Thailand are for oil palms. The figures for the US and Paraguay are for soya. The figure for China is for sugar cane.

Page 135: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

135

9.3. The suitability approach

9.3.1. Description of studies using the suitability approach It seems that BLUM model of land use change in Brazil uses the suitability approach with biophysical criteria. According to Nassar (2009), “Spatial information (availability and suitability) are used as input in the land allocation section of the model… Expansion is constrained by land availability for agriculture, which is calculated taking into account suitability (slopes, climate and soil characteristics)”. It is possible to add legal restrictions as an additional constraint. This availability assessment is done at the level of municipalities (of which there are about 5000 in Brazil).593 Land was classified as of low, medium, high or very high suitability on the basis of soil, climate and topographical characteristics.594 It seems that the classification describes suitability for agriculture in general rather than being crop specific.595 It is not clear exactly how the approach is applied, however. CARD apply what appears to be a suitability approach using the model GreenAgSiM. The model divides the globe into 518 administrative units. Using “vegetation maps and global ecological zones”, the average vegetation of each unit is calculated. It is assumed that when land for a particular crop expands in a particular country, it does so in administrative units that already have a high proportion of that crop.596 IIASA use a suitability approach. The steps are described in IIASA (2009) as follows: - “Land productivity and the expansion of cultivated land [are] determined in the world

food system model”597; - “The conversion of agricultural land is allocated to the spatial grid [of the resource

database598] in 10-year time steps by solving a series of multi-criteria optimisation problems for each of the countries/regions of the world food system model”599; the spatial grid of the resource database is 5 by 5 latitude/longitude grid cells, each with an estimated proportion of each type of land cover600;

- (in regions with a projected net decrease in cultivated land), “the main criteria and

drivers [used in the land conversion module] include demand for built-up land and abandonment of marginally productive cultivated land”601;

- (in regions with a projected net increase in cultivated land), “the land conversion

algorithm takes land demand from the world food system equilibrium and applies several constraints and criteria, including: (i) the total amount of land converted from

591 (for EPA) p. 20 592 (for EPA) p. 32 593 p. 3 594 p. 17 595 see for example p. 21 596 Dumortier et al. (2009) (p. 7). [Dumortier, J. and D.J. Hayes, Towards an Integrated Global Agricultural Greenhouse Gas Model, CARD Working Paper 09-WP 490, is given as a source for a full model description]1 597 p. 136 598 p. 161 599 p. 136 600 p. 161 601 p. 136

Page 136: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

136

and to agriculture in each region of the world food system model, (ii) the productivity, availability and current use of land resources in each country/region of the world food system model, (iii) suitability of land for conversion to crop production, (iv) legal land use limitation…, (v) spatial suitability/propensity of ecosystems to be converted to agricultural land, i.e. a priority ranking of ecosystems with regard to land conversion, and (vi) land accessibility, i.e. in particular a grid-cell’s distance from existing crop production activities.”602

At least in principle, the IIASA approach sounds more sophisticated than other approaches described here. One critical remark is that the assumption that the cultivated land that is abandoned is less productive need not necessarily be correct. In the vicinity of expanding urban areas, in both developed and developing countries, even highly productive agricultural land experiences urbanisation. For a more general critical discussion of the implied assumption of yield optimisation in land allocation, see chapter 8. JRC Ispra intend to use the suitability approach with biophysical criteria. To identify where cropland expansion could occur they intend to work with “maps for each agro-economical scenario”, “share of agricultural land at 10x10 km resolution”, and “indications on crop shares”. 603 The driving factor in their choice of method is the wish to work at 10x10 km resolution. It is stated that the historical approach cannot be used at this level because the necessary crop-specific time series data do not exist. An economic suitability criterion cannot be used because the necessary data (land cost, production cost, rates of return) also do not exist. An agronomical suitability criterion will therefore be used, in which new cropland is allocated to the best areas in terms of agronomical suitability.604 Two rules will be applied:

“When new areas of cropland are required, it is assumed they will appear in priority on most suitable areas in terms of climate, soil and landform

When suitability for a given crop is equal, the criterion of distance to current cropland is used”.605

It is not clear to what extent the first rule will be constrained by geographical allocation (e.g. to the level of countries) that is already present in the results of the models to which the JRC Ispra modelling will be applied. The description of the LEITAP modelling in Woltjer (2009) suggests that the land supply curve in the LEITAP model determines the land area required606, while the accompanying IMAGE model “allocates land on a grid level”. The suitability assessment in IMAGE is based on biophysical factors - population density, distance to water, distance to agricultural land and a random factor; it is added that “unsuitable land” is excluded.607

602 p. 136 603 Ramos et al. (2009) (p.4) 604 Ramos et al. (2009) (p. 6) 605 Ramos et al. (2009) (p. 7) 606 see p. 10 607 p. 13

Page 137: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

137

9.3.2. Summary – “suitability” factors taken into account The table shows the factors that different studies are stated to take into account in determining which land is converted.

Page 138: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

138

Table – “suitability” factors taken into account in determining land conversion BLUM CARD IIASA JRC Ispra LEITAP IFPRI Kløverpris

et al. soil suitability ü (for

crops in general)

ü (for crop in question)608

climate suitability ü (ditto) ü (ditto) landform/slope ü (ditto) ü (ditto) “suitability” (undefined)

ü ü (“unsuitable land” excluded)

distance to water ü yield ü pressure for urbanisation

ü (where cropland is projected to decrease)

population density

ü

land rent ü (from managed grassland & forest)609

? ü (from managed grassland)610

proximity to existing cultivation

ü (of crop in question)

ü (as proxy for accessibility)

ü (to decide between areas of equal suitability)611

ü (of “agricultural land”)

legal restrictions ü (as an option)

ü

geographical scale of assessment

Brazil, ~5000 units

world, 518 units

world, 5x5 latitude /longitude grid

world, 10x10 km squares

608 The basis for this interpretation is the statement “When suitability for a given crop is equal, the criterion of distance to current cropland is used.” (Ramos et al. (2009) (p. 7)) 609 For the rest, historical approach. 610 For the rest, historical approach 611 It is not clear if proximity is in relation to the cultivation of the crop in question or crops in general.

Page 139: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

139

cells Sources: see above

Page 140: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

140

It has not proved possible to compare how the factors in the table are taken into account in the studies. Unlike the historical method, which can in principle be applied without making an assessment of land availability for conversion in a particular region, it appears correct to say that ‘suitability’ studies necessarily make an (absolute or relative) estimate of land availability. The fact that certain study descriptions refer to this factor has not therefore been taken into account in the table above.

9.4. Other approaches

9.4.1. Studies using a hybrid approach (historical/suitability) Some studies use the suitability approach to estimate conversion to cropland from some types of land, and the historical approach to estimate conversion to cropland from other types of land. These studies are described here. Information about these studies is incorporated in summary tables for both approaches, as relevant. IFPRI use a hybrid approach. Land use in the base year is divided between “economic sectors” (cropland, pasture and managed forest) and “non economic land use categories” (unmanaged forest and other grassland).612 The model “computes” shifts to cropland from pasture and managed forest using an economic suitability approach.613 This is done using the same tool that is used to compute shifts between the quantities of land used for different crops. The model also computes “the quantity of land converted [from non economic] to economic use” 614. The model does not however compute “the origin of this land” 615. A historic approach is used to determine this. In regions that experienced cropland abandonment during the period in question, it is assumed that the proportions of conversion from cropland are maintained as rates of conversion to cropland. Once the quantity of land of each type to be converted into cropland has been calculated at the national level, it is divided between the country’s AEZs according to a formula that takes into account the quantity of cropland in the AEZ, the quantity of remaining land not already used as cropland and the share of the land for rain-fed crops in the macro-region that is located in the AEZ. Yield assumptions are then used to convert this result into ha. It is not clear what time period IFPRI use in determining historic averages for this analysis of endogenous change from non economic to economic land use. Valin et al. (2009) state, “We use the historical information on land use change”.616 The word “the” could be taken as referring to the FAO data for 2000-2004 referred to earlier in the same paper617 (these are used for determining exogenous land use change in the baseline). Later, a shift was made to the use of Winrock data. Thus, IFPRI use an economic suitability approach to depict the expansion of cropland into managed grasslands and forests, and a historical approach to depict its expansion into unmanaged grasslands and forests. 612 Valin et al. (2009) (pp. 6, 7 and 14) 613 Valin (2009) (p. 19) 614 Valin (2009) (p. 19) 615 Valin (2009) (p. 19) 616 p. 14 617 p. 12

Page 141: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

141

EPA use a hybrid approach. For the US, land use change is predicted using the FASOM model. For the rest of the world, “while FAPRI [the model used for outside the US] predicts land requirements, it does not predict how the requirements are to be met”.618 For this reason, a historical module is added to predict the types of land that are converted outside the US. It it is not clear how the work of Kløverpris et al. (described in section 9.2.1) addresses the expansion of cropland into managed grassland – it cannot be excluded that this is done in the same way as by IFPRI. However, the source states explicitly that the expansion of cropland into managed forest is not handled in that way.619

9.4.2. Econometric approach It would be possible to imagine addressing the problem using an economically estimated equation expressing the rate of conversion from each type of land as a function of a set of variables including demand for crops. No examples of such an approach have been identified.

9.4.3. Studies using other approaches In one AGLINK analysis it is assumed that all conversion is from grassland to cropland.620

9.5. Summary for both historical and suitability approaches

9.5.1. Summary - land types considered – general The tables show the types of land that can be converted and the types of land they can be converted into, in the different modelling exercises. It focuses on the types of land to which different carbon stocks can, in the models in question, be attributed.

618 Wang (2009) (p. E-4) 619 Kløverpris et al. (in press) (p. 2) 620 Von Lampe (2009) (p. 16)

Page 142: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

142

Table – Types of land considered as capable of having different carbon stocks in modelling exercises: (i) land types before conversion Searchinger et

al. Ensus (crop-specific) IFPRI EPA Kløverpris

et al. IIASA CARB Tyner et

al.

forest ü (many types)

ü ü (managed and unmanaged)621

ü ü ü ü (managed only)

grassland ü (many types, perhaps only natural)

ü ü (pasture and [other])622

ü (grasslands, savannah, shrub)623

ü ü (“pasture and other vegetation”)

ü (native grassland only)624

wetland/peatland ü (only as a type of SE Asian forest)

x ü x x

recently abandoned cropland

x ü (“idle land”)

?625 x x626 x x627 x

recently deforested forest, degraded forest land

x x x x x

cropland628 x (not needed) ü (reassigned in a 2nd- round calculation)

ü (“arable land and permanent crops”)

x ü (rain-fed and irrigated)

621 Managed forest consists of plantation and managed natural forest; unmanaged forest is also termed primary forest 622 Pasture includes permanent meadows. The “other” category is not explicitly defined as “other grassland” but this appears to be the case, since its contents are listed as “savannah, grassland, scrubland” (Valin et al. (2009) (p. 14)) and as “non pasture grasslands, savannah, scrubland, etc.” (Bouët et al. (2009) (p. 76). On the other hand, in Bouët et al. (2009) (p. 67) its contents are listed as “shrubland, urbanized land, desert, ice”. It would be helpful to look further into this. 623 These are the names of the three non-forest land types according to Valin (2009) and according to Lywood (2009) (p. 18). Nassar (2009) (p. 32) gives them as “pastagens” (Portuguese for “pasture”), “savana” and “other”; it is of course possible that “other”, if correct, is not a type of grassland. 624 According to Air Improvement Resource (2008) (p. 9) 625 Valin et al. (2009) (p. 22) gives a table of carbon stock values that includes values for “Land set aside”. The same table appears in other accounts of the IFPRI work. However, it is not clear whether or how this category is used in the modelling work. 626 “x” based on the statements that “All expansion in the USA occurs on grazable land because of the full utilisation of cultivable land” (p. 7) and “Denmark and China utilise all their cultivable land” (p. 6), set against the facts that the US is known to have significant quantities of recently abandoned cropland and Denmark is likely to have significant quantities of set-aside; the statement concerning the US is confusing since it appears from other parts of the text (e.g. p. 2) that while “cultivable land” can also be used for grazing, the reverse is not true. It may be, however, that the model depicts wheat expanding onto cultivable land presently used for grazing, driving grazing off that land and onto land that can only be used for that purpose. 627 According to Air Improvement Resource (2008) (p. 9)

Page 143: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

143

Table – Types of land considered as capable of having different carbon stocks in modelling exercises: (ii) land types after conversion Searchinger et

al. IFPRI

cropland only as residual, no explicit types

ü (different C stock attributed to rice)

Sources for both tables: Searchinger et al. (2008)629, Lywood (2009d)630, Valin et al. (2009)631, Valin (2009)632, Bouët et al. (2009)633, Kløverpris et al. (in press), Air Improvement Resource (2008)634

628 In applying a crop-specific approach, it may be found that land for the growth of a particular crop has historically come from land previous used for other crops. This does not arise with approaches that calculate only the overall historic change in the cropped area. 629 Searchinger et al. (2008) p. 11; for details of land types see section [10.3.4] 630 p. 9 631 pp. 6, 7, 14 632 (for EPA) p. 21 633 (for IFPRI) p. 66; (for Tyner et al.) p. 16 634 (for CARB) p. 9

Page 144: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

144

Because they treat the carbon stock of cropland as a residual, it is not possible within the methods of Searchinger et al. and Tyner et al. to explicitly attribute carbon stocks to cropland or consider variation within the category of cropland (though their method implicitly assumes such variation).

9.6. Commentary on land types considered

9.6.1. Abandoned agricultural land During the 1980s and 1990s, North America, Europe and the FSU all experienced cropland abandonment, both planned (in the form known in the European Union as “set-aside”) and unplanned.635 This land, if classified following abandonment as grassland, has a lower carbon stock than other grassland. Kline et al. (2009) state, “Land-use studies consistently acknowledge serious data limitations and uncertainties, noting that a majority of global crop lands are constantly shifting the location of cultivation, leaving at any time large areas fallow or idle that may not be captured in statistics. Estimates for idle grasslands, prone to confusion with pasture and grassland, range from 520 million acres to 4.9 billion acres [210 to 1980 Mha] globally.”636 “Rotational” set-aside land has a particularly low carbon stock compared to other currently uncropped land because it is ploughed regularly. It exists in the EU; Lywood (2009d)637 states that it also exists in the FSU. Because this category of land is not present in most studies (the land is considered to have been converted to ordinary grassland or to forest), it is not possible for this lower carbon stock to be taken into account. New Fuels Alliance (2008) criticise CARB on this point, stating “[T]he GTAP model does not include inputs for idle or CRP [Conservation Reserve Program] lands. This is a concern for two obvious reasons: (1) idle lands will be the first to be converted under any reasonable land conversion scenario; and (2), any model that does not include idle and CRP land will produce exaggerated forest effects because the major points of domestic agricultural land use expansion are disabled. Lands in developing countries without clear rents (economic values in a marketplace) cannot be analysed in GTAP. This includes much one-time cropland that is not accounted for or included in the GTAP estimates of effects.”638

9.6.2. Recently deforested land The approach adopted in Searchinger et al. and other studies reviewed here includes the following steps: (i) estimate by how much global crop production will grow as a result of additional

demand caused by biofuels/biofuel promotion policies;

635 FAO ProdSTAT; Koehler, 2009 [FG] p. 15) 636 p. 2 637 p. 9 638 p. 3

Page 145: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

145

(ii) estimate how much of this increased crop production will be located (in the modelled year, in the policy scenario) on land which (in the modelled year, in the baseline scenario639) would have been forest;

(iii) attribute the carbon stock loss (the difference between the carbon stock of the forest

and the crop) to the biofuels, considering them to have “caused” it. Commentators have made a number of criticisms of this procedure, notably including the following: (a) In reality, it often happens that land is deforested for some other reason than crop

production; left unused; would not revert to forest; and may subsequently be converted to crop production. In this case, the carbon stock loss attributable to the crop is the difference between that associated with the crop and that associated with the unused land. (Solution: include the category of ‘recently deforested land’ in the baseline, and use a procedure that permits new cropland to be attributed to land that would otherwise have fallen into this category.)

(b) Also in reality, it often happens that the deforestation decision has more than one

cause: for example, both to obtain the timber and to obtain the land for crop production. (Solution: a rule should be used to split the carbon stock cost (the difference between the stock associated with the forest and that associated with the crop) between the timber and the crop.)

Searchinger (2009) responds to criticisms relating to the multiple causes of deforestation. He makes the following points640: • “Some defenders have argued that biofuels are not the only cause of deforestation.

That statement is certainly true but irrelevant: The only question is whether additional demands on the world’s crops and croplands are an incremental cause.”

• “Others argue that the causes of deforestation are too varied to assume that agricultural

demand will translate into deforestation… To some extent, this argument confuses two questions: Where will deforestation occur, and what factors drive deforestation overall? Multiple factors always explain why deforestation occurs in some areas rather than others; for example, cropland will expand more where the infrastructure is available to support it. But studies641 have broadly concluded that increasing the economic return to agricultural use serves as a strong incentive to deforestation in general.”

• “Defenders argue that forestry is sometimes to blame for deforestation. That again is

the driving force in some locations – although forest will typically regenerate at least partially if land is not then converted to agriculture. In other regions, timber revenues combine with agricultural returns to provide the combined economic justification for land conversion. (In Brazil, for example, harvesting a few big trees may pay for some

639 It sometimes not entirely clear whether studies compare, instead, land use in the modelled year in the policy scenario with land use in the base year. To do so would be to introduce an additional source of error. 640 p. 15 641 [referenced to Kline , K. L., V. H. Dale and T. Searchinger, “Biofuels: Effects on Land and Fire: Exchange”, Science, 321, 2008]

Page 146: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

146

of the cost of ultimate agricultural conversion.) In fact, because timber markets may make land conversion cheaper, forestry may make land conversion more responsive to relatively small increases in agricultural prices.642 ”

• “Overall, many forces influence the amount and types of land that will be converted in

response to higher agricultural demand, and all the various models try in some ways to take these forces into account. That explains, in part, why all models so far estimate land conversion will come from a mix of forest and grassland.”

Searchinger’s responses do not address criticisms (a) and (b). The points made in the first two bullets are responses to the claim that no deforestation at all ought to be attributed to biofuels, a claim that is not made by criticisms (a) and (b). The second sentence of the third bullet correctly (but elliptically) describes criticism (a), without a response being made to this criticism. The third and fourth sentences of the third bullet correctly describe criticism (b). The fifth sentence appears to be intended as a response to this criticism, but does not in fact serve as one. Searchinger et al.’s modelling exercise depicts a world in which increases in agricultural prices lead, in a certain proportion, to deforestation. It must be supposed that Searchinger considers this proportion to be an unbiased estimate of the real rate. The fifth sentence says, in effect, “If there were no market for the timber, this proportion would be lower”. That may or may not be so. But criticism (b), inter alia as described in the third and fourth sentences, is not a statement about the size of this proportion. Rather, it is a statement about the way in which the carbon effects of this deforestation ought to be accounted for. Since the timber market is one of the causes of the deforestation of the land on which crops are then grown (and Searchinger’s third, fourth and fifth sentences all seem to recognise this), it ought also, according to criticism (b), to carry a share of the carbon cost – rather than (as in the methodology of Searchinger et al.) the whole carbon cost being attributed to the crops. Rather than responding to this argument, the fifth sentence speaks, if anything, to its correctness. The fourth bullet implies that all models try to take into account the multiple causes of deforestation, but does not explain how. It concludes by implying that the fact that all models “estimate land conversion will come from a mix of forest and grassland” is evidence that the multiple causes of deforestation have been taken into account; it is hard to see what relevance this has. It certainly has no relevance to criticism (a) (which claims that certain acts of cropland expansion which are treated as deforestation should instead be treated as replacement of unused, recently deforested land), nor to criticism (b) (which claims that responsibility for deforestation should, in certain circumstances, be attributed only partly to the crops that succeed the forest).

9.6.3. Degraded forest land ECORYS (2009) state that the FAO defines “deforestation” as the conversion of forest to another use or the long-term reduction of the tree canopy cover below 10%. If tree canopy cover is reduced but remains above 10%, the term used is “degradation”. They add, “Due to these definitions, activities such as logging often fall under the category of forest degradation

642 [A reference is made to a discussion of “these arguments” in Khosla, V., T. Searchinger and R.A. Houghton, “Biofuels Clarifying Assumptions: Exchange”, Science, 322 (2008).]

Page 147: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

147

and are therefore not included in the deforestation statistics provided by the FAO… It is of further interest to note that deforestation and forest degradation occur due to different driving forces and that deforestation does not necessarily follow degradation.”643 This suggests that even where logging has not led to forest clearance, it would still be appropriate to include a category of degraded forest land to which lower carbon stocks would be attributed.

9.6.4. Wetlands and peatlands Wetlands and peatlands have a higher carbon stock than other land uses (see section 10.3). When such land is converted and used for cultivation, the carbon stock is reduced. To the extent that this carbon stock loss is attributable to the crop, it should be taken into account. Swallow and van Noordwijk (2009) state that “Peatlands are found in many developing countries, with many other developing countries containing large areas of high-carbon wetlands facing similar threats from agricultural expansion”.644 Searchinger (2009a) states that “A few potentially large sources of greenhouse gas emissions are missing from [EPA’s] analysis… The most significant omission is wetlands… On a worldwide basis, wetlands have provided a significant percentage of cropland. In the U.S., agriculture is the estimated source of conversion for roughly 70% of wetland loss, or roughly 70 million acres [roughly 28 million ha]. That is roughly one fifth of the cropland actually planted and probably a significantly higher percentage of the total crop production. In the U.S., wetlands have provided the home for two of the three main sugarcane producing regions – south Florida and Louisiana, and outside of Brazil, wetlands could provide a main area of sugarcane expansion. Many of the best agricultural lands in Europe are also former wetlands. Wetlands are common in tropical forests. It is difficult to estimate what percentage of future land conversion around the world is likely to come from wetland conversion. However, the evidence is reasonable that one quarter of future palm oil expansion in Southeast Asia will go through peatlands, and in other regions, a scenario-based approach would also be reasonable. Wetlands store large quantities of carbon… [T]he EPA analysis should be modified to include their conversion.”645 Relatedly, according to another report in the EPA’s peer review, “Dr. Gibbs disagreed with [Winrock’s approach to the reclassification of the MODIS dataset] excluding [inter alia] the … “permanent wetland” categor[y] from the reclassification.”646 IFPRI assume that 10% of the land converted for oil palms in Malaysia and 27% of that converted in Indonesia will be on peatlands. They calculate how this could affect the total land use change impact of EU biofuel consumption using two sets of emissions coefficients, noting that "recent trends emphasise the underestimation of past values".647 With emissions coefficients derived from IPCC, land use change emissions would increase by 0.03 g/MJ.

643 p. 16 644 p. 10 645 p. C-7 646 ICF International (2009b) (pp. 6-7) 647 Al-Riffai et al. (2010) (p. 38)

Page 148: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

148

With emissions coefficients attributed to "Couwenberg (2009)648", they would increase by 0.23 g/MJ.649 Wetlands International suggest that a more recent report by the Couwenberg650 would imply emissions that were 25% higher than that estimate; another recent report would imply emissions that were more than twice as high; and there are reasons for considering that the true figure would be higher still.651 Searchinger et al. take wetlands into account in South East Asia652, and attribute the entire carbon stock difference between the wetland and the crop to the crop.

9.6.5. Pasture (including the postulated ‘cascade’ effect) Searchinger et al. make the following argument: “Much of additional cropland, in our analyses, is converted from grasslands. Because nearly all world grasslands that could be used for crops are grazed653, the conversion of cropland sacrifices forage from grazing land that now provides food for domestic animals. Replacing this forage should in turn lead to more land conversion either for grazing, which is a major source of deforestation in the Amazon or possibly to additional cropland. Our analysis ignores these effects because we wished to limit the analysis of land use change from agriculture to that which could be estimated with established models. As in the case of cropland, some forage could be replaced by intensification, in this case improvements of existing pasture, and some forage might never be replaced because of decreasing demand. In light of the role expansion of pasture is playing in deforestation, the additional land use change to replace forage is probably highly significant.”654 Searchinger and Heimlich (2008) make the same point: “Nearly all the world’s grasslands are grazed, so this conversion [to cropland] sacrifices ongoing forage and therefore some meat and dairy production, and there would be some additional conversion to replace this forage (either through more pasture or additional crops). There is no good model for estimating this additional conversion, however, so our analysis left it out… If a second round of conversion to pasture occurred for even one quarter of converted grasslands, the additional emissions from land use change could easily be 20% higher.” Searchinger (2009) makes the point more definitely: “Using grassland also sacrifices meat and dairy products because nearly all grassland is grazed. Replacing those livestock products would push grazing into yet other areas, such as a forest, which will release yet more carbon.”655

648 Cited as Couwenberg, J. (2009), "Emission Factors for Managed Peat Soils: An Analysis of IPCC Default Values", Wetlands International. 649 Calculations of the Commission services from data from Al-Riffai et al. (2010) (pp. 62 and 64). 650 Cited as Couwenberg, J., R. Donman and H. Joosten (2010), "Greenhouse gas fluxes from tropical peatlands in south-east Asia", Global Change Biology 16. The authors also put in doubt the calculations of a feedstock-specific emission factor for palm oil, suggesting that the peatland element of this could be understated by a factor of 100. It has not been possible to assess this. In any case there are other reasons to doubt the policy robustness of these feedstock-specific figures. (See section 11.2.) 651 Couwenberg and Silvius (2010) 652 Searchinger et al. (2008) (p. 11) 653 [referenced to J. Bruisnsma, Ed., “World Agriculture: Toward 2015/30, an FAO Perspective”, 2003] 654 p. 11 655 p. 9

Page 149: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

149

This postulated effect is described in this literature review as the ‘cascade’ effect.656 Searchinger and Heimlich (2008) make a second argument: “The CARD model used by our study assumes that all diverted feed grains are replaced by other crops. In reality, the rising price of meat due to higher feed grain prices will also spur at least some additional efforts to expand pastureland, which now occurs mostly in tropical forests. Using pasture, it takes more hectares to produce a unit of meat than using cropland. In this way, our study underestimated coverage.”657 The ‘cascade’ effect Lywood (2009c) states that Fritsche (2008), ECOFYS and “some partial equilibrium … models” incorporate the ‘cascade’ effect. He argues that this is not justified because historically, “the increased demand for meat production has been met entirely by increases in yield”, due to: “- the move from ruminants to monogastric animals, which are more efficient at meat

production - increased yields of cereal crops, used as animal feed - increased use of oil meals as feed supplements”. He further argues that data on meat production between 1961 and 2007 show that meat yields alter in response to demand for meat; and that a similar relationship can be expected in future, so that increases in demand for meat or reductions in land for meat production (which, through prices, will affect producers’ incentives in a similar way) will lead to a higher rate of yield increase rather than to conversion of land to grassland.658 Nassar (2009) states that BLUM has the advantage, relative to AGLINK and FAPRI (the partial equilibrium model used by EPA), of being able to “[p]roject pasture land endogenously” and “[c]apture pasture intensification due to competition with crops”.659 It is not clear whether “pasture intensification” refers to the production of more livestock for a given input (measured in ha) of pasture, or to the production of more livestock by supplementing the input of pasture with an input of feed. Going in the same direction as the argument of Lywood, Nassar (2009) gives data showing a correlation between Brazil’s total [cattle] herd and its stocking rate (animals/ha). These data suggest that rises and falls in the cattle herd do not lead to significant changes in the land area used. Searchinger (2009) criticises the work of Kim et al. because it “failed to recognise that … the sacrifice of grassland … will trigger at least some further land use change”.660

656 Note: this term is not used in the sources reviewed– it is used here as a convenient shorthand. 657 p. 17 658 pp. 8-10 659 p. 2 660 p. 12

Page 150: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

150

Searchinger (2009) states that “Former forests converted to grazing … receive enough rainfall to grow biomass well, but as a whole are underutilized even though they can supply meat and dairy products. Much of the case for biofuel production in Brazil rests on the broad availability of such low intensity cattle lands. The sugarcane industry believes that expanding into cattle lands avoids deforestation because the lost cattle production can be, and is, offset through greater use of fertilisation and better cattle breeds. On the other hand, conservation groups in Brazil argue that most of the expanded production of cattle in Brazil continues to occur through expansion into the Amazon. They contend that while Brazil could offset grazing lands converted to sugarcane through better management, that expansion is today also sending an additional signal for land conversion.”661 Searchinger and Heimlich’s second argument In a partial equilibrium modelling exercise, Saunders et al. (2009) conclude that increased demand for maize as a result of US promotion of biofuels will lead to increased feed prices and increased livestock prices. The New Zealand livestock sector, which predominantly uses pasture rather than feed as input, would see increased profits. The study does not go on to address whether this would lead to an increase in the area under pasture; more intensive use of the area already under pasture; an increase in feed inputs; or a continuation of high profits. General Searchinger et al.’s first argument is concerned with the reduced availability of pasture in the policy scenario (because some is converted to cropland). It appears to make the claim that in the policy scenario less nourishment for livestock is modelled as being made available than in the baseline scenario (because there is a reduction in nourishment from pasture and no compensating increase in feed consumption). While no definitive responses to this claim from other modellers have been identified, it seems odd that this is what the models would do. If they do indeed depict the world in this way, Searchinger et al.’s questioning certainly seems reasonable. Searchinger and Heimlich’s second argument is concerned with the reduced availability of feed in the policy scenario (because some cereals, otherwise used as feed, are diverted to biofuels). It appears to make the claims that (i) in the policy scenario this feed shortfall is always modelled as being compensated by the production of more feed from crops; (ii) in reality, a part of the response would come from the replacement of feed by pasture; (iii) this would be more land-intensive. Again, if (i) is a correct description of modelling practice, then (ii) certainly seems to a reasonable response. However, it is appropriate to make two further points. First, since biofuels (a fortiori EU biofuels) have co-products that are used as feed, it is not clear that the premise of a reduced availability of feed in the policy scenario is correct. In fact, the opposite seems more likely. If that is the case, the effect identified here would actually work in the opposite direction. Second, the passage from Searchinger and Heimlich implies that the “underestimated coverage” is evidently negative in carbon stock terms. This is not necessarily true. Searchinger and Heimlich’s argument is that what is depicted in the policy scenario as (i) X Mha of cropland (producing feed) should in fact be depicted as (ii) Y Mha of pasture

661 pp. 8-9

Page 151: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

151

(whose grass would nourish the livestock), where Y > X. Even if this is true, the carbon stock impact depends on what the land would have been in the baseline scenario. If the converted land would otherwise have been (ungrazed) grassland (whose carbon stock is similar to that of pasture) then the carbon stock effect of (ii) would be less than that of (i). At any rate, both the arguments point to a need for realistic modelling of the input that pasture provides to livestock production.

9.7. Commentary on other issues

9.7.1. General It is hard to avoid the conclusion that, while modellers may have succeeded in developing models that can to some extent predict the total amount of land that will be allocated in the future to different uses, they have not succeeded – for any purpose – in developing models that predict where, or on what type of land, these land use changes will take place. In the steps that precede the identification of the land that is converted, all the modelling exercises reported here apply the principle of economic optimisation. In all the modelling exercises, however, this principle appears not to be applied at this step. Economic optimisation is used to estimate how much production from new land is required in a particular country or region; the historical approach or the suitability approach is then used to determine which particular types of land will be used, and (in some cases) also its yields. The details of these choices could make the previous step (the decision about which region or country is affected) economically sub-optimal; but there is no feedback from this last step to those that precede it, and thus no opportunity to amend the choices made in the interests of optimality. The purpose of the “scenario modelling” reported here is to forecast what would happen under various conditions. This means that proponents of the historical approach need to argue why what happened in the past (at the level of geographical assessment they have chosen) is a good guide to what will happen in the future. Proponents of the suitability approach need to argue that the factors they have chosen to model are good predictors of the location of crop conversion. Little of this type of argument has been found in the literature.

9.7.2. Critiques of the historical approach Critics have pointed out that the results obtained under the historical approach seem to depend heavily on the years and data sources chosen. For example, Havlík et al. (2009) point out that where Searchinger et al. (using FAO data for the 1990s) estimate the proportion of forests in cropland expansion in Latin America at 75%, EPA (using MODIS data for 2001-2004) estimate it at 10%.662 Against this, Kløverpris et al., who appear to use long time periods (1970+), claim that “the trends in utilisation of cultivable and grazing land are fairly unambiguous for the eight regions

662 p. 16

Page 152: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

152

studied”.663 They go on to add, however, that “The subsequent identification of the areas affected within the regions (and thereby the biomes affected by expansion) is less certain.”664 The historical approach assumes that, in regions where cropland has historically been shrinking in favour of other land uses, including a proportion of forest, future increases in cropland (caused by biofuel-related demand) would lead, in the same proportion, to the loss either of forests which presently exist or of those that would otherwise come into existence ("future forests"). In relation to the first hypothesis, it is relevant to point out that "there are no significant examples of forest being displaced by temperate commercial crops".665 Dumortier et al. (2009) state that Searchinger et al. make an “assumption of deforestation in the United States”. This refers to the 38% of converted land in the US that Searchinger et al. consider to come from forest.666 Dumortier et al. make two criticisms of this assumption: (a) Empirical: “This seems to be unrealistic according to the Environmental Protection

Agency’s Greenhouse Gas Inventory Report (EPA, 2007). Between 1990 and 2005, forest area decreased by 4.1%, but it has remained stable over the last five years.”; “According to USDA and EPA GHG inventories, forest area in the US has remained constant between 1990 and 2005”.

(b) (Implicitly) methodological: “data from the Economic Research Service of the US

Department of Agriculture about major land uses in the US between 1945 and 2002 indicate that new cropland is taken from pasture and not forests (ERS/USDA)”; “In the last three years, most of the additional cropland in the US has come from grassland that was in pasture as part of the Conservation Reserve Programme”.667

The assumption of deforestation in the US seems to play an important role in the results of Searchinger et al.. Dumortier et al. estimate that the carbon payback time of US corn ethanol is less when US deforestation is excluded than when it is not by a factor variously stated as equivalent to about 1/3668 and to 23%.669 In the data used by Searchinger et al., the area of forest fell during the period used to determine the historical average (1990-1999), and so did the area of grassland. The method of Searchinger et al. is to assume that new cropland displaces each of these uses in proportion to the relative sizes of these two areas. Dumortier et al. seem to argue (a), methodologically, that this ignores the possibility that forest, when converted, might be converted to something other than cropland, so that the correct procedure to follow in jurisdictions where cropland is growing is to look at individual pieces of land converted to cropland and identify their immediately previous use; and (b), empirically, that in any case more recent historical data would have shown that the area of forest is no longer falling. 663 Kløverpris et al. (in press) (p. 9) 664 Kløverpris et al. (in press) (p. 9) [referenced to Kløverpris J., “Identification of biomes affected by marginal expansion of agricultural land use induced by increased crop consumption”, J Clean Prod (submitted)] 665 Lywood (2009c) (p. 11) 666 See table in section [9.1.5]; Dumortier et al. give the figure as 36%. (p. 8). 667 Dumortier et al. (2009) (pp. 1, 8, 9) 668 p. 2 669 p. 9

Page 153: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

153

Setting aside the empirical question, the methodological point, (a), certainly seems to have some merit. The table illustrates hypothetical three scenarios for land use change in a country. Table - Hypothetical scenarios for land use change in a country (A) (B) (C) T-1 (say,

1990) T0 (say, 1999), scenario with actual growth in crop production

T0’ (say, 1999), scenario with low growth in crop production

Crop production (quantity)

100 132 110

Crop yield (quantity/area)

1.0 1.1 1.1

Cropped area 100 120 100 Grassland area 100 90 110 Forest area 100 90 90 The method used by Searchinger et al. (and other practitioners of the historical method) in attributing land use change to historical growth in crop consumption (establishing a ratio which is then used to attribute land use change to future growth in crop consumption caused by biofuel use) is to compare column B with column A. Since 10 units of land ceased to be grassland between T-1 and T0 and 10 units of land ceased to be forest, it would follow from the method of Searchinger et al. that 50% of the land converted to cropland should be considered as coming from forest. In the hypothetical case illustrated in the table, the reality is different. Two separate trends are going on. There is a trend for conversion of forest to grassland, and a trend for conversion of grassland to cropland. The rate of forest to grassland conversion is independent of the rate of grassland to cropland conversion. One method to explore this might be the estimation of a counterfactual scenario (T0’ in the table). This would require the estimation of econometric equations for the relationship between forest, grassland and cropland areas – something found neither in the historical method, nor in the suitability method. Dumortier et al., who appear to have correctly identified this problem, appear to endorse another method: examining the immediately previous use of land that is converted to cropland. This method would appear to have the disadvantage of ignoring indirect effects (under which the conversion of grassland to cropland would drive the further conversion of forest to grassland, for example). It is not clear why Dumortier et al. do not explore the relevance of (a) to other jurisdictions where the cropped area is increasing, such as Brazil. As shown in section 9.2, practitioners of the historical approach use different time periods to establish the assumed ratio of conversion from different land types. None of the studies provides a justification for the length or start and end dates of the period used. As a general point, it could be considered that all these time periods are relatively short.

Page 154: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

154

As shown in section 9.2, practitioners of the historical approach use different sizes of regions to establish their historical averages. It would seem likely that the choice of regions concerned (including their size) also has an impact on the results. Again, practitioners have not generally provided justification for their choice of region. Nasser et al. have shown, in relation to Brazil, the difference that this choice makes. Valin (2009) comments that data on historical changes at the level of AEZs within countries would be needed “to be really effective” (presumably, in applying the historical approach).670 According to Searchinger et al., the US experienced substantial forest-cropland and grassland-cropland conversion during the 1990s. This is not consistent with FAO ProdSTAT data showing a reduction in US cropland during that period. Numerous other aspects of the data used by Searchinger et al. also appear strange. It seems doubtful that the clearance of natural ecosystems in North Africa and the Middle East is proceeding at 6 Mha/year. The source of the figure that 75% of land converted from cropland in the European Union in the 1990s went to forest (25% of this boreal forest) is unclear, and these figures do not look credible. It is claimed (for the most part, in their titles) that Searchinger et al.’s tables give data for conversion to cropland. But the depicted rate of conversion to cropland far exceeds that given in FAO data. If the past land conversion documented by users of the historical approach is indeed for a variety of purposes, not only for cropland, then, if this approach is to be used, it would seem appropriate to examine whether this is likely to introduce bias – in other words, whether there are systematic differences between the land typically converted to cropland and that typically converted to other uses, such as pasture. Searchinger, for example, fails to identify the need for such a step, stating that in applying the historical approach “We … assumed that new cropland in the future would reflect the patterns of new cropland in the 1990s”671 even though in fact that Woods Hole dataset used only shows the “net conversion”672 to cropland between 1990 and 1999, not the specific use in 1990 of each piece of land that was cropland in 1999. Valin (2009) may be alluding to this issue when he makes the comment on the historical approach “how marginal?”.673 Since pasture may have lower carbon stocks than natural grassland, it would also seem necessary to introduce the category of pasture to cropland conversion. Valin (2009) draws attention to the fact that “FAO has limited number of land use”.674 The round numbers in the land type shares given by Searchinger et al. and Ensus suggest that these represent estimates that are quite approximate. Valin (2009) asks “How accurate are the [FAO] data?” (used in the IFPRI work).675

670 p. 20 671 Searchinger (2009) (p. 15) 672 Searchinger et al. (2008) (p. 4) 673 p. 20 674 p. 20 675 p. 20

Page 155: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

155

Wang (2009) criticises EPA’s use of historical data to predict future land use patterns, and prefers the use of the GTAP model because it is “designed to predict land demand and supply for key individual countries, though the level of details of the GTAP model may need significant improvements”.676 Regardless of the merits of the historical work done for EPA, this comment on GTAP would seem to miss the point that GTAP is not designed to model conversion from non-commercial (non-managed) land uses. According to ICF International (2009a), the five peer reviewers (who did not include Wang, whose comment is reported above) who reviewed “the Winrock analysis of the historic satellite imagery”677 “generally agreed that the approach taken by EPA and Winrock was scientifically justifiable, especially given existing data and technology constraints. However, the reviewers highlighted several problematic areas of the analysis and recommended possible revisions. In general, these problematic areas were part of the satellite imagery analysis, rather than the emissions factor analysis [also done by Winrock and reviewed in the same report].” The main areas of concern, in descending order of frequency of comment by peer reviewers, were: “- The 3-year time period of the two MODIS data sets chosen and the error associated

with each of those data sets. - The coarse resolution of the satellite imagery. - The change detection analysis performed on the two MODIS data sets … - The reclassification analysis performed by Winrock on the satellite data, especially the

categories of excluded land and the role of the ‘mixed’ or ‘other’ category. - The methodology for projecting land use change patterns caused specifically by

biofuel production. - Evaluation of error and uncertainty associated with the satellite imagery analysis.”678 Further details of the five peer reviewers’ comments are given in ICF International (2009b). All agreed that the historical approach is scientifically justified, though one, Dr Gibbs, “recommended combining [it] with information on the drivers and causes of land use change”, another, Dr Tullis, recommended testing the accuracy of projections made in the past as a way of assessing the scientific justifiability of the approach and a third, Dr Houghton, said that “the justification becomes weaker for longer-term projects”.679 Four of the five reviewers agreed that any approach to this task would have to use remote sensing data as at least one of its sources of data. (It is not clear whether they explicitly considered and rejected the use of the FAO database, which would in principle offer an alternative – see chapter 3 – though one of the reviewers mentioned it as a possible source.)680

676 p. E-4 677 p. I-3 678 pp. I-4, I-5 679 p. 1 680 pp. 3, 4

Page 156: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

156

One reviewer, Dr Houghton, said that the 3 year gap between observations was too short. He recommended using 5-10 years.681 He commented that “identifying the (3-year) difference between two land-cover maps, each of which has been collapsed from 17 to 6 cover types, and each one of which has considerable error, is of questionable merit”.682 Three reviewers recommended using a dataset with a higher resolution than the MODIS dataset used (whose resolution was 1x1 km).683 Concerning the reclassification of the MODIS data (from 17 to 6 land categories), reviewers made a number of technical comments, including criticism of the exclusion of the “permanent wetland” category.684

9.7.3. Commentary on the suitability approach Kløverpris et al. (2008a) describe, as the view of some experts, the opinion that "characterise[s] … land supply curves … as a powerful concept although problems with the calibration exist. Furthermore, the suitability of the agricultural land expressed by the land supply curves is not the only decisive factor. Infrastructure and social aspects also determine which land is the next to be used."685. It is not clear that land supply curves (as used in the two examples cited - Kløverpris et al and LEITAP) actually include an element of “suitability”, at least as this term is used here. However, the reference to infrastructure and social aspects remains relevant. Social aspects are not taken into account in any of the “suitability” modelling reviewed here; infrastructure aspects are only taken into account if – as suggested by IIASA – proximity to existing cropland is a proxy for accessibility. IIASA do not present arguments in favour of this suggestion, and it is not obviously a powerful argument. (By contrast, it could be argued that – to the extent that infrastructure and social conditions in the future can be expected to be correlated with those in the past – these factors are taken into account in the “historical” modelling.) Lywood (2009c) states (wrongly, at least in the case of Searchinger et al.) that Searchinger et al. and OECD (presumably AGLINK) "assume that the land changes are based primarily on economics driven by land rents in different countries and different agro-economic zones".686 He criticises this approach on the basis that "changes in land use are determined by many factors other than economics. For example: security of food supply, employment and development policies."687 While explanations of the historical approach are often transparent, explanations of the suitability approach are less so, making it harder to assess whether they are an appropriate tool, appropriately handled. As one example, it is not clear what reference land use practitioners of the suitability approach assume, e.g. for land (if indeed there is such land in their modelling) that is cropland

681 pp. 7-8 682 p. 5 683 p. 5 684 pp. 5-6 685 p. 181 686 pp. 4-5 687 p. 5

Page 157: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

157

in the base year, forest or grassland in the modelled year in the baseline scenario and cropland in the modelled year in the policy scenario. (Note: this is the same as the question of ‘foregone sequestration’, addressed in chapter 10.)

9.7.4. Commentary on hybridity in approach Wang (2009) criticises EPA’s use of different models to predict land conversion within and outside the US, saying that this “poses a major methodological inconsistency”.688

9.8. Drivers of deforestation

9.8.1. Introduction In the present scientific debate concerning biofuels development, one of the major concerns is the link between biofuels and deforestation. The first question is to know if this link exists. If yes, is it a strong link? Must biofuels be blamed for deforestation by causing agricultural expansion? In order to bring elements to this debate, it seems necessary to take a step back in order to analyse the different drivers that can explain deforestation. By working that way, we can observe that deforestation is a complex process, which rarely can be explained by one single factor.

9.8.2. Deforestation: definition and main figures According to FAO, deforestation is "the conversion of forest to another land use or the long-term reduction of the tree canopy cover below the minimum of 10 percent". Deforestation must be differentiated from forest degradation which concerns the changes within the forest class which affect the forest stand, quality, or site negatively (reduction of the tree canopy above the original threshold of 10 percent is classified as forest degradation). In 2006, the FAO carried out a study on deforestation, which stated that the total forest area has decreased towards 2005 but that the rate of deforestation has slowed down due to increased plantation and landscape restoration. The overall net loss of forest per year from 2000 to 2005 is estimated at 7.3 million ha, compared to 8.9 million in the 1990-2000 period.

9.8.3. The multiple factors analysis: main principle Over the last years, many scientific studies working on land use change have reached the conclusion that deforestation can be explained most of the time by multiple factors. According to this rich academic literature, mainly based on Geist and Lambin work689 690 691, it is arbitrary and unreasonable to assume that all use change worldwide is driven primarily by agricultural expansion. Geist and Lambin took the form of a meta-analysis of 152 cases

688 p. E-4 689 Geist, H.J; Lambin, E.F Proximate Causes and Underlying Driving Forces of Tropical Deforestation.. BioScience. 2002, 52 (2), 143-150 690 Geist, H.J; Lambin, E.F ; Lepers E. Dynamics of Land-Use and Land-Cover Change in Tropical Regions. Annu. Rev. Environ. Resour. 2003, 28,0205-41 691 Geist, H.J; Lambin, E.F. 2001. What drives tropical deforestation? LUCC Project Report Series No.4. Belgium : Louvain la Neuve.

Page 158: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

158

studies to examine patterns and processes of deforestation in many locations around the world (78 cases from Latin America; 55 cases from Asia and 19 cases from Africa) This study focuses on four proximate causes (direct drivers), which can be defined as human activities in immediate action at the local level: Agricultural expansion; Timber harvest/ Wood extraction; Infrastructure development; other causes (e.g. predisposing environmental factors, biophysical factors, social disruption…) Ø In 25 % of cases, the three factors (expansion, wood extraction and infrastructure

expansion) are present Ø By getting into these categories into detail, the authors identify that permanent

cultivation, transport extension and commercial wood extraction are each present in 50% of the cases or more.

Ø There are some regional differences among these causes.

As regards the indirect driving forces (fundamental social processes that underpin the direct drivers and either operate at the local level or have an indirect impact from the national or global level), the study focuses on five: Economic factors; Institutional factors; Technological factors; Cultural Factors; Demographic Factors. All these multiple factors act synergistically rather than by single-factor causation, with more than one third of the cases being driven by the full interplay of economic, institutional, technological and demographic variables. From all the elements described ahead, it seems possible to confirm that deforestation must be seen as a complex and multiform process which cannot be represented by a mechanistic approach but only a system approach seems appropriate692. This general reasoning is strengthened by Kline et al work stating that there is little evidence that biofuels cause deforestation since key events leading to primary conversion of forests often proceeds from decades before it can be officially identified : "The regions of the world that are experiencing first-time land conversion and characterized by market isolation, lawlessness, insecurity, instability, and lack of land tenure (…) A cycle involving incremental degradation, repeated and extensive fires, and shifting small plots for subsistence tends to occur long before any consideration of crop choices influenced by global market prices"693.

9.8.4. Direct drivers of deforestation Agricultural expansion Agriculture expansion driver includes: § Permanent Cultivation (large scale versus smallholder ; subsistence versus

commercial) 692 de Sherbinin A. A Guide to Land-Use and Land-Cover Change (LUCC), Sept 2002, http://sedac.ciesin.columbia.edu/tg/ 693 Kline K, Dale V, Lee R, Leivy P, "In defense of biofuels, done right", Issues in Science and Technology, Spring 2009, vol. 25, issue 3, 75-84

Page 159: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

159

§ Shifting Cultivation (Slash and Burn versus tradition swidden) § Cattle ranching § Colonization (transmigration, resettlement projects…)

This driver is the leading factor since the expansion of cropped land and pastures is present (most of the time in combination with others) in 146 cases (96%) of the Geist and Lambin study. FAO estimates that, over the period [XXX], 16% of total deforestation was attributable to commercial agriculture which provides (inter alia) biofuel crops694. The Greenergy Econometrica study uses this assumption noting that the FAO assessment pre-dates the period under consideration and indicating that further research is required to better understand the contribution of different sectors to deforestation 695. Agriculture expansion affecting the rain forests varies depending on the regions. South East Asia: Rapid population growth in Southeast Asia represents one of the biggest challenges for sustainable resource use and mainly explains agricultural expansion (subsistence as well as large scale agriculture). As regards as export crops, the area of land occupied by palm oil trees in Indonesia kept increasing. This expansion is not fully due to biofuel demand (see timber harvest factor). Latin America: Several scientific studies focus on conversion to pasture for cattle ranching to be the most serious threat to the Amazon forest and the main driver for deforestation696 697. Brazil is the world's largest beef exporter. Some estimation concluded that 88% of the deforested area in this region would be occupied by ranches.698 According to a recent report, in 2007, for the first time, the legal Amazon passed the historical threshold of 10 million head of cattle slaughtered and the cattle herd in the region reached approximately 74 million head, or 3.3 head per inhabitant, triple to national average699. The cropland (mainly soy) production is also a high level production from commercial farming in this region. Morton et al. point out that cropland is expanding at an unprecedented rate in Mato Grosso700. Their finding refute the claim that new crop production in Amazonia is occurring only through intensified use of lands previously cleared for cattle ranching rather than adding a new pressure for forest loss. This finding implies that under favoured economic conditions, continued cropland expansion in the Amazon is possible. The predominance for commercial farming must not make forget that subsistence farming is also quite present in Amazonia.

694 FAO (1980), Global Ressources Assessment. 695 Econometrica, “A Practical Approach for Policies to Address GHG Emissions from Indirect Land Use Change Associated with Biofuels”, Technical Paper TP-080212-A, 2009 696Meizlish M, Spethmann D, Barbara M, "Financial analysis of Alternate Land use in the Amazon, Congo and Papua, Indonesia", New Forests, paper presented at UNFCCC COP 13 and Forest Day, Dec 2007, Bali Indonesia 697 Fearnside P, Deforestation in Amazonia, "Encyclopaedia of Earth, Eds Cultler J. Cleveland, Washington, DC Environmental Information Coalition, National Council for Science and the Environment. 698 Walker R and Moran E , (2000), "Deforestation and Cattle Ranching in the Brazilian Amazon ; External Capital an Household processes", World development Vol.28, No 4, p 683-699 699 Amigos de Terra, Amazonia Brazileira, "The Cattle Realm", January 2008 700 Morton DC, Defries RS, Shimabukuro YE, Anderson LO, Arai E, Espirito-Santo FdB, Freitas R, Morisette J, (2006)"Mapping land use of tropical regions from space", Proc Natl Acad Sci USA 103, 14637-14641

Page 160: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

160

Africa: Small scale agriculture is vital for livelihoods. This driver would be the most important to explain deforestation in this region. However, it is important to notice that in some Western Africa Countries, if large and small plantations were initially created at the expense of existing rain forests (mainly during the colonial period), more and more plantations are now established on already deforested areas which can be degraded forests (forest which have already been exploited for timber harvest or fire wood or for shifting cultivation), or in old abandoned cocoa or coffee plantations701. As regards as prices, some scientific studies came to the result that a rise in the price of crops (eg : soybean) has a significant effect on the rate of clearing of rain forest and grassland some countries (mainly Brazil702). However, these statements have afterwards been controverted explaining that this correlation broke down in 2005 because of the complexity and heterogeneity of causal factors for deforestation703. In a more general way, according to Keith Kline et al, "Current models attempt to predict future land use change based on changes in commodity prices. As conceived thus far, the computational general equilibrium models designed for economic trade do not adequately incorporate the processes of land use change. Although crop prices may influence short term land use decisions, they are not a dominant factor in global patterns of first time conversion)"704. Wood extraction This driver is a leading factor since this cause is mentioned in 67% of the cases studies by the Geist and Lambin meta-analysis. Wood extraction driver includes: § Commercial wood extraction (state managed or private) § Fuelwood (mainly domestic usage) § Polewood § Charcoal production

Timber extraction in tropical region countries is a very complex topic to deal with and to explain in details mainly because each country situation is very particular but also because the illegal logging activity can be predominant in some cases, while difficult to analyse because of its informal nature. Before entering into details in the regional differences and in the importance of each sub-driver, it is interesting to have a look at the official statistics provided by the International Tropical Timber Organization (ITTO) regarding ITTO producers705 production, exports and domestic consumption between 2004 and 2008. Logs production and exports volumes have slightly decreased over this period while domestic consumption has remained stable. In a general way, these figures point out there is still a clear need of wood at the domestic level in 701 Legoupil Jean-Claude, Ruf François, (2009), "Farmers's strategy and land use change in the perspective of biofuels development in West Africa, Natural resources forum, : 173 702 Morton DC et al, op. cit. 703 Woods Jeremy and Murphy Richard, Imperial College London, May 2009, comments on Greenergy Econometrica technical paper -TP-080212-A, January 2009 704 Kline K et al, op. cit. 705 Cameroon, Central African Republic, Congo, Côte d'Ivoire, Gabon, Ghana, Liberia, Nigeria, Togo, Cambodia, Fiji, India, Indonesia, Malaysia, Myanmar, Papua New Guinea, Philippines, Thailand, Vanuatu, Bolivia, Brazil, Colombia, Ecuador, Guatemala, Guyana, Honduras, Mexico, Panama, Peru, Surinam, Trinidad and Tobago, Venezuela

Page 161: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

161

these countries as well as a clear interest by foreign countries to import these woods. Therefore, commercial / domestic logging is an important activity by its own in these countries whatever the other reasons for taking wood out of the forest. Table 1-1-c. Production, Trade and Consumption of All Timber by ITTO Producers (1000 m 3)Country Product Species 2004 2005 2006 2007 2008* 2004 2005 2006 2007 2008* 2004 2005 2006 2007 2008*

Logs All 18005 17633 18805 18175 18063 2989 2904 3331 3545 3556 15016 14741 15475 14630 14507C 25 25 25 25 25 0 0 0 0 0 25 25 25 25 25NC 17980 17608 18780 18150 18038 2989 2904 3330 3545 3556 14991 14716 15450 14605 14481

Sawn All 4342 4688 4732 4586 4601 1832 1884 1723 1726 1679 2515 2880 3017 2864 2926C 12 12 12 12 12 1 0 1 1 1 15 80 14 13 13NC 4330 4676 4720 4574 4589 1832 1884 1722 1726 1678 2500 2800 3003 2851 2913

Ven All 692 757 711 827 913 440 389 352 321 309 261 375 360 506 605C 1 1 1 1 1 0 0 0 0 0 4 1 1 1NC 691 756 710 826 912 440 389 352 321 309 257 374 358 505 604

Ply All 402 440 449 436 407 161 135 194 263 242 263 346 262 185 176C 13 14 14 14 14 0 0 0 0 0 20 22 19 20 20NC 388 426 434 422 392 161 135 194 263 242 243 324 243 165 156

Logs All 85161 85260 89303 94359 94413 8926 10868 9348 9300 9319 79350 78528 84162 90021 90341C 5352 5356 5507 5498 5498 324 221 222 277 277 5506 5662 5868 6077 6074NC 79809 79904 83796 88861 88915 8602 10647 9126 9022 9042 73844 72865 78295 83944 84267

Sawn All 27388 29224 29399 29322 29346 8670 7889 8680 7860 7868 22297 25027 24476 25248 25224C 9454 10033 10057 10057 10057 98 106 84 84 81 9791 10328 10438 10469 10418NC 17934 19191 19342 19265 19289 8572 7783 8596 7776 7787 12506 14699 14038 14780 14806

Ven All 1622 1626 1564 1667 1688 626 559 500 546 557 1105 1188 1176 1226 1242C 100 95 91 97 102 37 23 30 40 48 86 94 94 94 89NC 1522 1531 1473 1570 1586 588 536 470 506 509 1019 1094 1082 1132 1153

Ply All 12766 12404 12830 12842 12834 8698 8218 9229 9058 8650 4377 4634 4268 4379 4781C 961 898 987 982 982 994 947 1132 1092 1094 92 183 260 267 290NC 11805 11505 11843 11860 11852 7705 7271 8097 7966 7556 4285 4451 4009 4112 4491

Logs All 123082 133439 134791 122236 122615 480 263 373 395 378 122830 133278 134547 122043 122350C 48839 59885 68134 50162 50697 95 24 2 7 7 48889 59934 68214 50302 50768NC 74243 73554 66657 72074 71918 386 239 371 388 371 73941 73345 66333 71741 71581

Sawn All 29592 29272 30024 30758 31941 4557 4389 3943 4075 4207 26444 26748 27917 28952 30177C 11167 12183 12605 13158 13259 2132 2050 1736 1667 1730 10033 11459 12135 12974 12970NC 18425 17089 17419 17600 18682 2425 2339 2206 2408 2477 16412 15289 15782 15978 17207

Ven All 1079 1079 1148 1176 1179 143 249 219 316 331 985 875 979 909 905C 637 652 729 765 765 27 38 31 146 154 620 623 709 630 630NC 442 428 419 410 413 117 211 188 169 177 365 252 271 279 275

Ply All 4847 4963 3897 3769 4282 3490 3922 3076 2693 2815 1949 1658 1464 1705 2031C 2849 3204 2695 2571 2990 2114 2938 2332 2098 2196 1016 564 717 854 1096NC 1998 1758 1202 1197 1291 1376 984 743 594 619 932 1095 747 851 935

Logs All 226248 236333 242899 234770 235091 12395 14035 13052 13239 13254 217196 226547 234184 226694 227197C 54216 65266 73667 55686 56220 418 246 225 284 284 54420 65621 74107 56404 56867NC 172032 171066 169233 179085 178871 11977 13789 12827 12955 12970 162776 160926 160078 170290 170330

Sawn All 61322 63184 64155 64666 65888 15059 14163 14346 13662 13754 51256 54655 55410 57065 58327C 20633 22228 22673 23227 23327 2231 2157 1822 1752 1811 19839 21866 22587 23456 23401NC 40689 40956 41482 41440 42560 12828 12006 12524 11910 11942 31418 32789 32823 33609 34926

Ven All 3393 3463 3423 3669 3780 1208 1197 1072 1183 1197 2352 2439 2515 2641 2752C 737 748 821 863 868 64 61 61 187 202 710 718 804 725 720NC 2656 2715 2602 2806 2911 1145 1136 1010 997 995 1641 1720 1711 1916 2032

Ply All 18015 17806 17176 17047 17522 12349 12274 12499 12013 11707 6589 6638 5994 6269 6987C 3823 4117 3696 3568 3987 3108 3885 3465 3190 3290 1129 769 996 1141 1406NC 14192 13689 13479 13479 13536 9241 8389 9035 8823 8417 5461 5869 4998 5128 5581

Asia-Pacific

Domestic Consumption

Africa

Production Exports

ProducersTotal

Latin America/Caribbean

Source: International Tropical Timber Organization (2008), Annual Review and Assessment of the World Timber situation 2006, http://www.itto.int/en/annual_review/ Venn: veneer Ply: plywood Wood extraction affecting the rain forests varies depending on the regions. South East Asia : Commercial logging is mostly mentioned in deforestation studies and fuel wood collection is mentioned to a lesser extent. Since the early 1970s, the Southeast Asia Pacific region has become the main source of tropical timber trade in the world. In Indonesia (Papua province), forest concessions allowing harvesting of natural forests have been granted over 6.5 million hectares of the province's 42 million hectares706 in order to extract wood for commercial using. These concessions are required to carry out impacts assessments and forest monitoring plans before extracting wood. Nevertheless, logging

706 Papua Ministry of Forestry (2006) List of Forest Concession Right (Hak Pengusahaan Hutan) and Timber Forest Product Utilisation Permit (Izin Usaha Pemanfaatan Hsil Hutan Kayu) holders, August

Page 162: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

162

practices do not often respect these requirements and result in over-exploitation and loss of the forest structure707. As regards as palm oil production, most of the global expansion of these plantations has occurred in Indonesia and Malaysia. This production is often blamed for being responsible for deforestation, since these plantations would allow to answer to the increasing biofuel international demand. Nevertheless, this statement should be mitigated. Indeed, while Indonesia has destroyed over 28 million hectares of forest since 1990, largely in the name of land conversion for plantations, yet, the area of oil palm or pulp wood plantations established in this period was only 9 million hectares708. Indeed, it seems that it is easier for companies to obtain land clearing permit that a logging permit so that oil palm is a mean to access to timber709. This clearly implies that most of the companies obtained permits to convert the forest only to gain access to the timber. Many palm oil companies are associated with logging companies. In 2003, although 12.5 million hectares of degraded land was available, most oil palm plantations were established in forested areas710, which means that it appears much more lucrative for companies to set up a "bogus" plantation, harvest the timber than abandon the area. Latin America: Wood is extracted from approximately 1.5 million hectares per year in the Amazonian region of Brazil (Azevedos Ramos, 2008). A substantial part of this production is exported to foreign market (about one third). As in Asia, harvesting for fuel wood is also a major cause of deforestation but is remains secondary compared to commercial logging. Brazil is the largest producer wood tropical timber products and as such, the forest industry is an important component of the economy and in particular the economy of the Legal Amazon711. The forestry sector is responsible for 3.5% of Brazil's gross domestic product (GDP), generating 2 million formal jobs and accounting for 8.4 % of Brazilian exports (Serviço Forestal Brasileiro, 2007) It is important to highlight that recent studies indicate that rather than contributing to deforestation by providing access to areas that are subsequently converted to agricultural use, selective logging appears to be a driver of deforestation on its own712. Africa : Most African use wood and charcoal for cooking, since there are no other affordable and available energy sources. Therefore, wood extraction for domestic fuel consumption remains at a high level in Africa. As regards as commercial logging, it has been increasing substantially since 1990. According to Lobet, the volume of timber exported annually from countries of the Congo Bassin has

707 Jarvis and Jacobsen (2006) "Working paper-Incentives to promote forest certification in Indonesia" Project: motivating sustainability, International Finance Corporation. 708 Greenpeace, "How the palm oil industry is cooling the climate", 8th November 2007 709 Sheil D, Casson A, Maijaard E, Van Noordwijk M, Gaskell J, Sunderland Groves J, Werts K, Kanninen M, "The impact and opportunities of oil palm in Southern Asia, CIFOR, June 2009 710Casson A, "Oil Palm, Soybeans and critical habitat loss", review prepared for WWF forest conversion initiative, August 2003 711 Banerjee O, Alavalapati JR, "Modeling forest sector in a Dynamic computable general equilibrium framework: The case of forest concessions in Brazil", paper prepared for the 12th annual conference on global economic analysis, Santiago, Chile, June 10-12, 2009 712 Foley et al (2007) citing Nepstad, D. Et al (1999) "Large scale impoverishment of Amazonia forests by logging and fire", Nature Issue 398, April 8, p 505-508 and Anser, G et al (2005) "Selective Logging in the Brazilian Amazon", Science Issue 310, p. 480-82

Page 163: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

163

increased tenfold713. Nearly 60% of the Democratic Republic of Congo's (DRC) total forest area is thought to be productive or commercially valuable714. Approximately 20 million hectares of the DRC's 145 million hectare forest estate are allocated as timber concessions to about 60 companies715. The tropical wood African production is principally driven by demand from to developed countries for furniture-grade wood (European demand plays an important role). Illegal logging: While analysing wood extraction in all these tropical regions, it must be referred to illegal logging. Illegal logging is a major concern in these countries (according to assessment by international institutions such as the World Bank or WWF, about 70 countries have substantial problems with illegal logging) Most of the scientific studies agree that this illegal commercial activity has a consistent impact on deforestation. Available figures and estimations must be dealt with cautiously. A joint UK-Indonesian study on timber industry in Indonesia in 1998 suggested that about 40% of throughout was illegal, with a value in excess of $365 million716. More recent estimates, comparing legal harvesting against known domestic consumption plus exports, suggest that 88% of logging in the country is illegal in some way717. According to WWF, In Brazil, 80% of logging in the Amazon violates government controls while in Africa, this rate vary from 50% to 80% depending of the countries718. Infrastructure This driver is a leading factor since this cause is mentioned in 72% of the cases study by the Geist and Lambin meta-analysis. Infrastructure extension includes: § Transports § Markets § Settlements (rural and urban) § Public service (water, electricity…) § Private company activities

In most of the studies, road extension is found to be one of the specific direct drivers of tropical deforestation. By contrasts, the development of private enterprise infrastructure (dams, mines, oil exploration) appears to be a minor direct driver of tropical deforestation globally, although it is important in some regions. In the tropical regions, commercial logging is closely connected to development of infrastructure. The international company that, most of the time, are responsible for harvesting and selling wood, create news roads in the area they are dealing with. Subsidies dedicated to road built can even more increase this effect. 713 Lobet I, "Living on Earth's", Interview 2nd of January 2009, Why tropical Forest fall 714 Congo Bassin Forest Partnership (2005) 715 Forests Monitor (2007), "The Timber Sector in the DRC : A Brief Oberview" 716 Indonesia-UK tropical forestry management programme (1999), Illegal Logging in Indonesia. ITFPM Report no. EC/99/03 717 Greenpeace (2003) Partners in Crime : "A greenpeace investigation of the links between UK ans Indoneisa's timber barons". 718 WWF international (2002), The timber footprint of the G8 and China

Page 164: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

164

Some scientific studies got to the conclusion that the infrastructure driver is the main one in the Amazonian region : "at the Amazon-wide scale (...) proximity to roads is the best predictr of deforestation. (…)When time is taken into account, roads emerge as the key companent of the deforestation process"719. In some cases, the construction of dams for the generation of hydroelectric power as well as oil and gas pipelines and new settlements can be seen as a cause of deforestation720.

9.8.5. Indirect drivers of deforestation Economic factors These factors are present in 81% of the cases studied by Geist and Lambin and are closely linked to economic development through a growing cash economy. Market growth is an underlying factor which is affecting forests in tropical countries. Timber products, agricultural products and minerals products account for a very significant part of this growth. This statement is strengthened in Latin America where the enforcement of Structural Adjustment policies (International Monetary Fund and World Bank) have brought to an economic development based on the export of raw material721. As a general comment, we can say that economic growth in tropical region puts forests under a substantial pressure in order to answer to the wood demand at the local level, but most of all, at the international level. Institutional factors These factors (78% of the cases) are, most of the time, not a positive driver to protect forests. Indeed, governance problems, weak and centralized regulatory system, endemic corruption, lack of clear forest management policies and the weaknesses public forestry agencies (Africa) tend to encourage the deforestation in these areas. Property right issues have been discussed in the deforestation literature. It is widely recognized that clear property rights are fundamental basis for instituting sustainable forest management. In general, governments in countries with large amounts of forest have traditionally opted to transfer access rights and management authority to large-scale private industry through logging concessions. In the last decades, some of them have also introduced reforms in forest ownership policies in favour of community access and ownership. There is no single, "correct" forest property rights regime for all cases. Each country must find its own balance among public, private and community rights but this equilibrium seems difficult to be found in these regions. Technological factors These factors are present in 70% studied cases. Over the last years, the tropical areas have experienced agro technological changes which allow them to answer to the substantial

719 Kathryn R. Kirby, William F. Laurance, Ana K. Albernae, Götz Schroth, Philip M. Fearnside, Scott Bergen, Eduardo M. Venticinque, Carlos da Costa, "The future of deforestation in the Brazilian Amazon", in Futures 38 (2006), 432-453 720 Manta Nolasco, M.I.(2007) : "Evaluacion de las causas naturales y socioéconomicas de los incendios forestales en America del Sur. Facultad de Ciencias Forestales, Universidad Nacional Agraria, Lima, Peru. 721 Grau, HR, Aide, M (2008) : Globalization and Land-Use Transitions in Latin America. Ecology Society 13(2):16

Page 165: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

165

increasing of their population. Thanks to these progresses, no one could wait for increased outcome of the existing agricultural land and a reduction of the pressure to new agricultural land. Nevertheless, these changes are still very slow and most of the farms remain quite small which cause more deforestation than big farms that tend to be more efficient. Furthermore, technological progress in Forestry can in some cases be damaging for forests since the new machinery used do not correspond with the forestry requirements in these regions (heavy inappropriate equipments). Cultural factors The cultural factor can lay into consumption patterns (e.g: increasing in the per capita consumption makes diet change and, therefore, makes land structure and uses change) but also into attitudes and perceptions such as unconcern for forests due to low morale and disregard for "nature". This factor can also be dominated by war and conflicts (Africa): forest can become the place to large movements of refugees, which can affect forest resources. Demographic factors It seems that population increase due to high fertility rates can sometimes be primary driver at a local scale. This can lead to logging, subsistence agriculture and a relevant rate of fuel wood. The demographic factor that also has a significant effect on deforestation is in migration of colonists. In some cases, this migration has been encouraged by the Governments to encourage people to establish new agricultural settlements very close to forests.

9.8.6. Does deforested land revert to forest? Searchinger (2009) states, “Forest will typically regenerate at least partially if land is not then converted to agriculture.”722

9.8.7. Conclusion The brief analyse of the drivers of deforestation which is here above is far from being exhaustive and comprehensive so much this topic is complex and the scientific literature is rich. The objective of this work was not to deeply analyse world deforestation but more to highlight and focus on the systemic approach (according to which deforestation can rarely be explained by only one cause) and to give an overview of main drivers that can be behind deforestation. It seems clear that the link between biofuel demand / production and deforestation is complex. It is not evidence what the "weight" of the biofuel driver is in the overall multi drivers' balance (in the drivers list above, biofuel would be linked to agricultural expansion driver). To date, it appears that the scientific studies carried out on land use change have not tried to

722 p. 15hodsopa

Page 166: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

166

achieve modelling all these drivers. That is maybe where the challenge lies in today in order to have a deep and more certain knowledge of the relation between bioenergy and deforestation. Further studies should be based on more detailed research at the case study level, in order to get a better understanding of likely patterns and dynamics between deforestation and biofuel development723. However, now and already, we can draw some conclusions that could be taken into account by the models. These findings are summarized in the outlines below.

723 CIFOR, "A global analysis of tropical deforestation due to bioenergy development", Bioenergy, sustainability and trade-offs : can we avoid deforestation while promoting bioenergy ?, project founded by the European Commission, Contribution agreement No EuropAid / DCI-ENV/2008/143936/TPS, 2009

Page 167: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

167

• The modelling should be capable of depicting substantial deforestation in the baseline

scenario, since it seems clear that substantial deforestation will take place without the biofuel policy.

• The modelling should be capable of depicting a situation in which the main driver(s)

of deforestation would lie more in logging and / or infrastructure than in agriculture expansion.

• In the policy scenario, it should be possible for commercial agriculture to expand onto

this deforested land. • In this case, depicted in Outline 1, the carbon stock impact should be attributed in

function of what could otherwise have been expected to happen this deforested land. In South East Asia at least, it seems clear that in many cases deforested land, when left idle, does not revert to forest. (Land might not revert for natural reasons, or because biofuel-induced conversion to crop simply accelerates a conversion that would otherwise have eventually taken place for other reasons, forestalling the building up of significant carbon stocks in trees.)

Outline 1

Today Baseline scenario Policy scenario

: Forest : Idle/waste land xxxx : Crops

xxxxx

Page 168: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

168

• When commercial agriculture expands directly onto forest in the policy scenario, as shown in Outline 2, 100% of the carbon emissions value due to deforestation should taken into account.

• In this case, it would be relevant to ask what happens to the timber when deforestation

took place. Where it can be expected that timber has been used for some purpose, it would seem appropriate for a share of the carbon stock loss to be allocated to the timber. (This point is acknowledged by Houghton (2009), but only in relation to changes that might occur in the future: "It's possible that future changes in the value of carbon could encourage nations to manage wood products differently from the way they are managed currently. Thus, while forests cleared for soybeans today might be burned, with different incentives they might be harvested and turned into long-term products for the future. To the extent that such changes occur, it would not be justifiable to assume the same overall lifecycle impact."724)

Outline 2

Today Baseline scenario Policy scenario

: Forest xxx : Crops x (for biofuel production)

724 p. C2

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Page 169: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

169

10. Model calculations: carbon stock changes

10.1. Data and assumptions: general Carbon stocks can be divided into ‘above ground’ and 'below ground' carbon. They can also be divided into ‘biomass’ or ‘vegetation’ carbon (which can be both above and below ground) and ‘soil’ carbon. The work reviewed here obtains the carbon stock values used as follows: BLUM use the emissions factors used by EPA for Brazil, except that a higher value for above-ground biomass is assigned to sugar cane.725 CAPRI use IPCC tier 2 methods for calculating greenhouse gas emissions.726 Air Improvement Resource (2008)727 criticises CARB for using Woods Hole data728 and imply that this has a single category of grassland. Air Improvement Research (2008)729 state that EPA uses data from the "validated CENTURY model" and imply that this distinguishes between native grassland, pasture and ex-cropland. Nassar et al. (2009) state that EPA use “emissions factors … proposed by Winrock International”.730 IFPRI state, “In order to determine greenhouse gases emission, we relied on IPCC guidelines for National Greenhouse Gas Inventories. We used Tier 1 method which does not require knowledge of the exact CO2 stock in each region but provides generic estimates for different climate zones that can be matched with the agro-ecological zones in the model.”731 For deforestation, the method used is to assume that all biomass carbon is lost as a result of conversion (see table below). It is stated that, “The formula for the computation of the CO2 stock in forest is: CO2 Stock … = Forest area … * CStock … * 0.47*44/12*(1+ Below ground ratio)”.732 It is somewhat difficult to understand the role of the multiplier of 0.47 in this equation. In explanation it is stated that “0.47 is the coefficient used to compute carbon mass by dry

725 Nassar et al. (2009) (p. 19) 726 Blanco-Fonseca and Pérez Domínguez (2009) (p. 2) 727 p. 9 728 This is confirmed by UNICA (2009) (p. 20) 729 p. 9 730 p. 19 731 Valin et al. (2009) (p. 14) 732 Bouët et al. (2009) (p. 79); this discussion relies on the relevant annex in that draft of the IFPRI study because there is no equivalent annex in the final study; it is however understood that this part of the work did not change between the draft and final study.

Page 170: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

170

matter”733; this explanation would appear, intuitively, to make more sense if the term “tons of dry matter” rather than “CStock” were to appear in the equation. The accompanying table734 is entitled “Carbon stock in forest for different climatic regions (tons dry mat per hectare). It includes columns headed “natural forest”, “forest plantation” and “below ground ratio”. It is not obvious whether the values given in the first two of these columns are to be understood as before or after multiplication by 0.47. The carbon stock values for “forest plantation” are attributed to managed forests. The values for “natural forest” are attributed to primary forest.735 The procedure for determining which is which (in the base year) is described as follows: “FAO data relative to forest areas are distributed between managed and unmanaged using data from Sohngen et al. (2007)736 on forest management practice. Tropical and forests with limited accessibility are considered to be unmanaged whereas temperate mixed forests with accessibility and forest plantations are considered to be managed forests.”737 For “the cultivation of new land” (which probably , but not necessarily, equates to grassland conversion), “We used Tier 1 methodology from IPCC and indicative release of carbon relative to different management practices to determine the additional emissions induced”. The different practices taken into account are non-cultivation, cultivation with full tillage, rice cultivation and set-aside. “The level of input was considered to be medium for each case (emission factor equal to unity).”738 Values are given (or can be calculated from other values) for the carbon stocks associated, in each agro-ecological zone, with each of these “management practices”.739 The most obvious conclusion to draw from the text cited in the last paragraph but one is that the “non-cultivation” value is attributed to grassland; the “rice cultivation” value is attributed to land used to grow rice; and the “cultivation with full tillage” value is attributed to all other cultivated land. It is not clear what role the value for “set-aside” plays in this work. It appears that grassland and cropland are assumed to contain only soil carbon and not biomass carbon stocks.740 Dumortier et al. obtain carbon stock data from the GreenAgSim model.741 Mortimer et al. quote ADAS as saying that there is “a high degree of uncertainty associated with this [carbon stock] data, either due to lack of relevant studies or variations caused by

733 Bouët et al. (2009) (p. 79) 734 Found in Bouët et al. (2009) (p. 79) as well as other accounts of the IFPRI work. 735 736 Cited as Sohngen, B. et al., “Global Forestry Data for the Economic Modelling of Land Use”, GTAP technical paper (2007) 737 Valin et al. (2009) (p. 7) 738 Valin et al. (2009) (p. 14) 739 A table of these values can be found in Bouët et al. (2009) (p. 80); the same table appears in other accounts of the IFPRI work. 740 This can be deduced from the fact that it appears that Table 5 on page 22 of Valin et al. (2009), and its equivalents in other IFPRI documents, is used for the calculation of carbon stock loss from grassland conversion. 741 Dumortier et al. (2009) (p. 1)

Page 171: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

171

contrasting management regimes on the same land-use type, particularly arable and grasslands”.742 Tyner et al. use “Woods Hole (regional data)” and “IPCC (aggregated for the whole globe)” for carbon stocks. They use Woods Hole data for forest sequestration and IPCC data for forest and grassland sequestration.743 They state that the IPCC data should be corrected to remove foregone emissions from grassland, and that the Woods Hole data should be corrected in respect of forest foregone emissions (these comments are not further explained). The former correction reduces the calculated impact of the land conversion due to the US ethanol programme by about 22%; the latter increases it by about 1.5%. Searchinger et al. state that the carbon stock data they use take into account the fact that forests that are "re-growing" have lower amounts of stored carbon in vegetation.744 They state that their data are more specific than, but generally comparable with, those cited by the IPCC.745 For grasslands, the estimated carbon stocks are for the natural ecosystem. The authors accept that some of the converted grassland will be pasture and that pasture may have lower carbon stocks. However they state that this is more than outweighed by the fact that their analysis fails to take into account the land use change needed to replace the annual forage no longer available from converted grasslands.746 (That issue is explored in section 9.6.5.) In relation to the carbon stocks themselves, it is stated that "These carbon stocks were initially determined from summaries of global vegetation… These analyses have been revised using biomass and soil carbon values from a variety of sources, usually specific for the region."747 The dataset of carbon stock values used by Searchinger et al. and maintained by co-author R. A Houghton748 is known as the Woods Hole dataset.749 O’Hare et al., Mortimer et al., Tyner et al., Searchinger et al. and IFPRI calculate carbon stock lost when forest or grassland is converted as a proportion of carbon stock previously present. The table shows the assumptions made in these studies. Among these authors, only Searchinger et al. give substantial data or arguments to justify their assumption. Even so, Mathews and Tan (2009) describe Searchinger et al.’s assumption concerning the proportion of soil carbon that is lost as “arbitrary”.750 IFPRI state that their assumption implies that “We therefore do not consider wood that could be used for building or furniture”751 – but do not address the more important issue of whether the land itself retains carbon stocks in its new use. 742 Cited as Kindred, D., “Effect of Land-Use Change: Total Greenhouse Gas Emissions from Land Use Change in the UK and EU”, ADAS 2008; from Mortimer et al. (2008) (p. 30) 743 Tyner et al. (2009) (p. 22) 744 p. 7 745 p. 12 746 Searchinger et al. (2008) (p. 11) 747 The global sources cited are G. Ajtay et al. in "The Global Carbon Cycle", ed. B. Bolin et al., 1979; J. Olson et al., "Carbon in live vegetation of major world ecosystems", US Department of Energy, 1983; R. Whittaker and G. Likens in "Carbon and the Biosphere", ed. G. Woodwell and E. Pecon, 1973; and R. Houghton et al., For. Ecol. Manage. 38 (1991). The specific sources cited are R. Houghton and J. Hackler, Global Change Biol. 5, pp. 481-492 (1999) and R. Houghton, Global Change Biol. 11, pp. 945-958 (2005) 748 p. 3 749 Searchinger et al. (2008) (p. 4) 750 p. 5 751 Bouët et al. (2009) (p. 79)

Page 172: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

172

Table – Proportion of carbon stocks lost when land is converted to cropland752 Proportion of carbon stock lost

O’Hare et al.

Mortimer et al.

Tyner et al. Searchinger et al.

IFPRI

Above ground carbon

100% 100%

Carbon in vegetation

forest: 75% grassland: 100%

100% forest: 100% grassland: no stock

Below ground carbon

25%753 forest: 100% grassland: no stock

Below ground biomass

10%

Carbon in soil 25% 25%754 calculated755 Sources: Mortimer et al. (2008) 756, Tyner et al. (2009)757, Searchinger et al. (2008)758, Valin et al. (2009)759, Valin (2009)760, Bouët et al. (2009)761, O’Hare et al. (2009)762

10.2. Data and assumptions: carbon stocks in cropland

10.2.1. Biomass/above-ground carbon Nasser et al. (2009) state that IPCC attribute 5 tCO2eq/ha to “crops” (taken here to mean above-ground carbon on cropland). They add that “since sugarcane is a semi-perennial crop, which is very rich in biomass, [this] value … does not accurately represent its carbon content. According to various specialists[763], sugarcane biomass above ground is on average about 17 MT CO2-e per ha.”764 In BLUM, a value of 17 tCO2eq/ha is therefore used for vegetation carbon for sugar cane, and a value of 5 tCO2eq/ha for other crops.765 UNICA (2009) state that CARB do not, and should, take into account “carbon uptake from crops”.766 752 The values shown are for the results obtained once the land has moved, over time, to its new equilibrium value. 753 25% of the carbon stored in the top metre of the soil. 754 25% of the carbon stored in the top metre of soil; the loss of soil carbon is attributed to “cultivation”. 755 Explicit values for soil carbon stocks in different land uses in different AEZs 756 Mortimer et al. (2008) (p. 35) This assumption is sourced to Smeets et al. (2007); the bibliography includes E. Smeets et al., “The sustainability of Brazilian ethanol – an assessment of the possibilities of certified production”, 2008 757 p. 22 758 p. 7 759 “When forest is converted to another use, we assume that the stock of carbon in this type of forest is released completely (both above ground and below ground stock).” (p. 14) 760 p. 24 761 p. 79 762 p. 4 763 Citation: Amaral et al. (2008) 764 p. 19 765 According to UNICA (2009), pp. 20-21, the value attributed by Brazilian experts to the vegetation carbon of sugar cane is 17.4 tC, not tCO2eq, /ha. 766 p. 21

Page 173: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

173

UNICA (2009) suggest that IPCC data cover soil carbon but not biomass carbon in cropland.767 By contrast, Nassar et al. (2009) state that IPCC attribute 5t CO2eq/ha to crops, and that this is the value used by EPA.768 Lywood (2009d) states: “It has been argued that the IPCC method is not fair for sugar cane because: - it does not take into account of the carbon stock in the root system - it does not take into account of the seasonal above ground carbon stocks, which are

harvested annually - mechanical harvesting (as opposed to burning) will increase carbon stocks”.769

10.2.2. Soil carbon Lal (2009)770 states that “most agricultural soils have lost 30 to 40 mt [presumably a misprint for t, and then equivalent to 110-147 tCO2eq] per hectare”. Kim et al. (2008) state that “Plow tillage … represents the “worst case” as far as environmental management of corn agriculture is concerned.” They criticise Searchinger et al. and Fargione et al. for assuming that all diverted and newly converted maize fields use plough tillage.771 Compared with Kim et al.’s preferred reference case in which current average tillage is used (40% conservation tillage, 60% conventional tillage), this assumption reduces cumulative greenhouse gas benefits over 100 years by 9-10%.772 Where new cropland comes from grassland, payback time for maize ethanol is 12 years in the reference case, 4 years with all-conservation tillage and 18 years will all-conventional tillage. Where new cropland comes from forest, the equivalent values are 31, 20 and 37 years respectively.773

10.3. Data and assumptions: conversion of wetland/peatlands Searchinger et al. (2008) state that the [soil] carbon losses associated with “drainage [of wetlands] for agricultural conversion can be much greater than 25% [the proportion assumed in their work – see section 10.1] and may continue indefinitely”.774 Swallow and van Noordwijk (2009) state, “A controversial 2006 report showed massive GHG emissions from conversion of peat forests and from poor management of peat soils converted earlier. Subsequent studies have made some adjustments to those results and emphasized the uncertainty surrounding these estimates.”775 It is not clear from the source what these "subsequent studies" are.

767 p. 21 768 769 p. 13 770 p. 11 771 pp. 10-11 772 p. 14 773 Kim et al. (2008a) (p. 19) 774 pp. 11-12 775 p. 10 [the “controversial 2006 report” appears to be A. Hooijer et al., “PEAT-CO2, Assessment of CO2 emissions from drained peatlands in SE Asia, Delft Hydraulics, 2006]

Page 174: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

174

CIFOR (2009) gives reason to doubt that the full carbon cost of a transition from forested peatland to oil palm plantation should be attributed to the oil palm plantation. It states, “Logging, draining or clearing peatlands allows drying, allows surface peat to become flammable or to decay, and releases large amounts of CO2”, and adds “Drainage incurs costs. Technically, there is no need to drain peat for oil palm production (the plant copes with waterlogged soils); however, it is necessary to create access.”776 It also adds that oil palms are increasingly being planted on already cleared land.777 (This raises the issue of why it would be necessary to drain to get access for planting, but not to get access for logging.) IIASA (2009) state, “Globally, one third of the carbon reservoir … is stored in peat lands covering 3 percent of the earth’s land surface. Through drainage, the dry peat is exposed to air and starts oxidizing, decomposing, and emitting large amounts of carbon dioxide. Additionally, drained peat land is prone to fire, further accelerating greenhouse gas releases. In Southeast Asia, large-scale drainage of former rainforest has occurred to enable logging of the peat swamp forests and the transportation of logs in the drainage canals. After deforestation, drainage continues to establish oil palm and pulp wood plantations.” In the peer review of Winrock’s work for EPA on soil carbon stocks and emissions, “Both Dr. Gibbs and Dr. Lal suggested improving the estimates of soil carbon stocks and emissions by improving estimates of peatlands. Dr. Gibbs commented that emissions from peat swamp clearing and other wetlands should be better accounted for. Dr. Lal noted that reliance on a single reference may not be sufficient for peatlands. He suggested comparing the analysis used with the subsidence and bulk density data published in the literature for specific peatlands.”778

10.4. Empirical evidence on foregone carbon sequestration Foregone sequestration can be discussed in relation to four categories of land use change: (i) land that is forest in the base year, remains forest in the baseline and is converted to

cropland in the policy scenario – "foregone sequestration by existing forests"; (ii) land that is grassland in the base year, remains grassland in the baseline and is

converted to cropland in the policy scenario – "foregone sequestration by existing grassland";

(iii) land that is cropland in the base year, is converted to forest in the baseline and remains

cropland in the policy scenario – "foregone sequestration by future forests"; (iv) land that is cropland in the base year, is converted to grassland in the baseline and

remains cropland in the policy scenario – "foregone sequestration by future grassland".

776 p. 27 777 pp. 23-24 778 pp. 13-14

Page 175: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

175

Mortimer et al. calculate sequestration rates for fallow land and non-rotational set-aside land. They point out that sequestration should not be attributed to rotational set-aside land because it is ploughed every few years. 779 Searchinger (2009) comments, “Most hectares productive enough to produce average U.S. corn yields would also regenerate into trees if left alone.” He adds, “Many corn acres are in land that was originally prairie … but without deliberate effort today to install fire on the landscape, most corn acres would regenerate as forest.”780

10.5. Assumptions on foregone carbon sequestration UNICA (2009) state that CARB do not, and should, take into account foregone sequestration by future forests.781 Dumortier et al. state that “the [GreenAgSim] model is able to capture carbon sequestration if cropland comes out of production and re-grows to natural vegetation... If cropland is taken out of production, we assume that the land sequesters carbon (biomass and soil) over 20 years.”782 O’Hare et al. account for foregone sequestration.783 Details of the values used are not given in their paper. Searchinger et al. account for foregone sequestration by both existing and future forests. They state, "We separated growing from re-growing forests"784 – this distinction appears to equate to the one made here between sequestration by future ("growing") and existing ("re-growing") forests. Concerning sequestration by existing forests, they state, "We … estimated the rate of carbon addition on those portions of forests that are re-growing. These rates are modest compared to many estimates of potential reforestation benefits because they are based on changes in carbon content of forests over time of all re-growing forests of a regional type, which implicitly accounts for the fact that some portion of forests are regularly disturbed and lose carbon (e.g. through fire)."785 (The logic of the second sentence is not clear.) The precise calculation method is not explained. Concerning sequestration by future forests, they state, "The GHG cost would be the loss of the carbon that would be sequestered on these lands over 30 years. This carbon gain is calculated as regaining 75% of the original 25% of carbon lost from the original conversion to agriculture786, i.e., 18.5% of carbon in undisturbed lands of the ecosystem type, plus a rate of growth of the vegetation equal to re-growing ecosystems of that type." Tyner et al. account for foregone sequestration. They refer to “the carbon that could have been stored by non-croplands if land conversion did not occur”.787 This would appear to refer to foregone sequestration by both existing forests and existing grassland. However, the data

779 Mortimer et al. (2008) (p. 38) 780 p. 8 781 p. 21 782 Dumortier et al. (2009) (p. 7) 783 O’Hare et al. (2009) (p. 4) 784 Searchinger et al. (2008) p. 7 785 p. 7 786 For this 25% figure see section [10.3.1]. 787 Tyner et al. (2009) (p. 22)

Page 176: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

176

presented in the source relate only to foregone sequestration by existing forests. The Woods Hole data that they use show (a) the total area of forest of a particular type in a particular country/region; (b) the “re-growing forest area” within this788; and (c) the gross uptake of carbon by that type of forest in that country/region (in tons per year).789 Foregone carbon sequestration per hectare of forest converted is calculated by dividing (c) by (a), thus assuming that the proportion of growing forest converted is the same as the proportion of growing forest in the total forest area. Tyner et al. do not refer to the question of sequestration by future forests. However it seems likely that they take it into account, since they present data for regions, such as the EU, where it would seem to be common ground that the cropped area is falling.

10.6. Results The tables shows values used in modelling exercises for the carbon stock associated with different land uses (first set of tables), for the carbon stock loss associated with land use change (penultimate table)790 and for foregone sequestration.

788 Presumably, the remaining forest area is considered to be mature and no longer sequestering carbon. 789 Based on example shown for Canada (p. 23) 790 It has been preferred to enter data in the first table where possible.

Page 177: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

177

Table – Carbon stock associated with land uses (soil and vegetation) – forest/plantation

tCO2eq/ha

Tyner et al.

Meizlish et al.

Fritsche WWF/Co- operative Bank

Searchinger et al. (Woods Hole)

IFPRI791 Fargione et al.

“German SBO draft”

Amaral et al.

tropical forest

549792 Brazil 345-539793 Indonesia 539

NA & ME 1161

natural: 359-1506 forest plantation: 154-753

Brazil 993

tropical moist/rain forest

Pacific developed 1161 S and SE Asia 1356 sub-Saharan Africa 642-1160

Malaysia 2572 Brazil 2700

971794

tropical deciduous forest

Latin America 872 S and SE Asia 843

tropical coniferous forest

Latin America 1092

tropical open forest

Latin America 454795 S and SE Asia 403

tropical dry forest

sub-Saharan Africa 303

montane forest

sub-Saharan Africa 659

tropical woodland

NA & ME 352 Pacific developed 352

wooded Cerrado

Brazil 605 Brazil 317

oil palm 403

Page 178: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

178

plantation temperate forest

EU 172 US 259

natural: 339-983 forest plantation: 145-626

temperate coniferous forest

Canada 1078

US 916-1319 NA & ME 1077 Canada 1077 Latin America 1107 Pacific developed 1077 Europe 1077 FSU 1077

temperate deciduous forest

Canada 986

US 1099-1209 Canada 986 Latin America 857 Pacific developed 986 Europe 931 FSU 986

temperate woodland

US 660 Europe 352 FSU 352

boreal forest

Canada 1085

Canada 601

Canada 1085 Europe 1085 FSU 1085

natural: 71-238 forest plantation: 71-191

Page 179: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

179

Table – Carbon stock associated with land uses (soil and vegetation) – wetland/peatland

tCO2eq/ha

Fargione et al.

Fehrenbach et al.

tropical peatland forest

Malaysia 12648

wetland 5130

Page 180: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

180

Table – Carbon stock associated with land uses (soil and vegetation) – grassland

tCO2eq/ha

Meizlish et al.

Searchinger et al. (Woods Hole)

CEPII/ IFPRI

Fargione et al.

“German SBO draft”

Amaral et al.

grassland Latin America 191 256 tropical grassland

NA & ME 220 Pacific developed 220

139-220 set-aside: 129-180

grassland Cerrado

Brazil 311 Brazil 262

savannah 491 Brazil 295796 shrub, desert scrub, desert

NA & ME 224 Latin America 234 sub-Saharan Africa 127

tropical cattle ranching

50797

managed pasture

Brazil 214

degraded pasture

Brazil 155

temperate grassland

US 330 Canada 718 India, China, Pakistan 718 Europe 718 FSU 729

139-348 set-aside: 129-324

US 605

tundra Canada 623 boreal grassland

249 set-aside: 204-232

Page 181: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

181

Table – Carbon stock associated with land uses (soil and vegetation) – cropland (for plantations see “forest”)

Sources for the preceding tables: calculations of the Commission services and data in Tyner et al. (2009)801, Meizlish et al. (2007)802, Fritsche (2007)803, WWF and Co-operative Bank (n.d.)804, Searchinger et al. (2008)805, Mortimer et al. (2008)806, Valin et al. (2009)807, Valin (2009)808, von Lampe (2009)809 (for “German SBO draft and Fehrenbach et al., (2007)), UNICA (2009)810 (for Amaral et al. (2008)811, Macedo and Seabra (2008)812 and IPCC data for cropland), Lywood (2009d)813

tCO2eq/ha

Tyner et al. (residual values)798

CEPII/ IFPRI

Mortimer et al. (residual values)

“German SBO draft”

IPCC799 Macedo & Seabra800

cropland Malaysia 732-9800 Brazil 247-847 US 409

202

tropical cropland

81-106

temperate cropland

Canada 492-515

111-240

boreal cropland

Canada 649 172-199

sugar cane

180 244

maize 134 148 soya 134 140

Page 182: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

182

Table – Carbon stock associated with change of land use (soil and vegetation) – estimated loss caused by land use change

tCO2eq/ha

Searchinger et al.

IPCC according to Searchinger et al.

IPCC according to Tyner et al.814

Tyner et al.815

IPCC according to Tipper et al.

tropical forest to cropland

604-824 553-824 312-690816

temperate forest to cropland

688-770 297-627

boreal forest to cropland

423-588817

tropical grassland/ savannah to cropland

75-305 189-214

temperate grassland to cropland

111-200 139-242

wetland to cropland 1146 (southeast Asia)

748 (global)

forest to cropland 1374 (global) forest to non-forest 322818 Sources: calculations of the Commission services and data in Searchinger et al. (2008)819, Tyner et al. (2009)820and Tipper et al. (2009)

Page 183: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

183

Table – annual carbon stock loss as a result of foregone sequestration when forest land is converted to cropland

Tyner et al. (approx.822) Searchinger et al.

tCO2eq/ha/year821

annual emissions

attributed to foregone

sequestration823

total emissions attributed to forest

conversion824

share from foregone

emissions825

annual emissions

attributed to foregone

sequestration

total emissions

attributed to forest

conversion826

share from foregone emissions

US 1.5 29 5% 0.3 39 1% Canada 0.3 23 1% 1.8 37 5% Sub-Saharan Africa 0.3 16 2% 0.2 16 1% Europe 4.6 28 18% 5.6 34 17% Russia 1.0 21 5% Central and South America 0.7 24 3% 0.5 21 2% Middle East and North Africa 2.8 18 16% 0 8 0% E Asia, Oceania and Japan 1.2 20 6% 'Pacific developed' 3.6 69 5% China, HK and India 3.1 35 10% Malaysia, Indonesia, rest of S Asia, rest of S E Asia 3.1 35 10%

3.8 61 6%

Sources: calculations of the Commission services from Tyner et al. (2009)827, Searchinger et al. (2008)828

Page 184: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

184

Concerning carbon stock values, it can be seen that there is wide variation in the values used, even within sub-global categories. It can also be seen that there are substantial differences between the values given in the two studies that cover foregone sequestration, both for absolute emissions and for the contribution of foregone sequestration. It can be seen that in both these modelling exercises, the inclusion of foregone sequestration makes a substantial difference in certain regions. In particular, the emission effect of forest conversion in the EU would be reported as 18% less if foregone sequestration were not taken into account – presumably because the EU is experiencing afforestation and thus has a smaller proportion of mature forests. Although Searchinger et al. use the figures in the fifth column of the table for the rate of foregone sequestration for both future and existing forests, Searchinger (2009) gives a higher figure for temperate regions for future forests: 7.5-12 tCO2eq/ha/year for temperate forests and 14-28 tCO2eq/year for tropical forests. A footnote suggests, nevertheless, that the temperate figures, at least, are derived from Searchinger et al. (2008).829 Mortimer et al. estimate foregone sequestration from the conversion of fallow land and non-rotational set-aside at 1.14 +/- 0.66 tCO2eq/ha/year. These figures are comparable with those for forests in the second and fifth columns of the table above.830

10.7. Commentary: foregone sequestration Foregone sequestration is taken into account in the modelling in two ways. Either it is assumed that sequestration continues at a constant rate indefinitely (exaggerating the impact, since sequestration stops when a forest reaches maturity) or it is assumed that the impact of foregone sequestration is the difference between the carbon stock of the new land use and that of a mature forest. The latter implies a baseline scenario under which any forest that exists is assumed to reach maturity (maximum carbon stock). This would appear to be an ‘environmentally optimal’ baseline scenario rather than a ‘what would otherwise happen’ baseline scenario, and therefore not compatible with the design principles of the baseline scenarios used in the modelling exercises in question (see chapter 4). After all, even without biofuels, it appears to be the case that forests are typically not mature. (If they were, there would be no future sequestration to forgo.) If it is logical to take into account foregone sequestration, it would seem logical to do so only in relation to a fixed number of years, after which the forest in question would anyway have reached maturity. This means that is probably incorrect for Tyner et al. to use the same annual rate of foregone sequestration in calculating the carbon stock consequences of an ethanol programme of 100 years duration as they do for one of 30 years duration.831 If it is logical to take into account foregone sequestration, it is also logical that the values used for carbon stock should be estimates of the average quantities per hectare that forests actually do contain rather than the quantities that they would contain if fully mature. It should be checked whether this is the case. Searchinger et al. state that in their case (using Woods Hole data), it is. The methods of Searchinger et al. and Tyner et al. appears to assume that when forest is converted, the mix between still-growing and not-growing forest is the same as the mix found in the country/region in question. If foregone sequestration is to be taken into account, it would be appropriate to reflect on whether this assumption is justified.

Page 185: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

185

If it is logical to take into account foregone sequestration in relation to forests, it should also be taken into account in relation to grassland. The sources do not do this, perhaps because the rate of carbon sequestration in grassland is low. It can be seen that the inclusion of foregone sequestration leads to higher estimates of land use change emissions per hectare of new cropland in regions where the required land is taken from recently abandoned cropland (eg Europe) than in those where it is taken from existing forest. It is unlikely that policy-makers or citizens share this view of the appropriate direction land use change should take. Moreover, this apparent effect is based on an illogicality. If it is correct to assess the former case on the basis of "afforestation opportunities foregone", then the same should be done in the latter case. In both cases it should be assumed that all the forest affected is new. (In other words, if this form of accounting makes sense, felling mature forests, using the trees for energy purposes and allowing them to be replaced with seedlings is precisely what we should be doing, at least in greenhouse gas terms.)

10.8. Commentary: carbon stock The means by which indirect land use change has a greenhouse gas impact is through changes in carbon stocks. This calculation is at the centre of the work reviewed here. In the light of this, it is notable that there are numerous significant differences in how the different studies calculate this, and surprising that (with the partial exception of Searchinger et al.) the literature contains little explicit discussion of the methodological and data issues involved. Each study tends to use its own assumptions without mentioning, let alone critically evaluating, others. A first point on which studies differ is on the proportion of carbon stocks lost when land is converted to cropland. For example, in the case of the conversion of a forest to cropland: - Searchinger et al. assume that all the carbon in vegetation (above and below ground)

and 25% of the soil carbon will be lost; - O'Hare et al. assume that all the above ground carbon and 25% of the below ground

carbon will be lost; - Mortimer et al. assume that all the above ground carbon and 10% of the below ground

"biomass" will be lost. - Tyner et al. assume that 75% of the carbon in vegetation and 25% of the carbon in soil

will be lost. Only Searchinger et al. provide data to justify (in part) their assumption. Among the studies mentioned, Searchinger et al.'s assumption gives the highest result for the quantity of carbon lost. O'Hare et al.'s assumption gives a lower result in the sense that only 25%, rather than 100% of below-ground vegetation is assumed to be lost. The assumptions of Mortimer et al. and Tyner et al. give results that are lower still. In Mortimer et al., 15% less below-ground carbon is lost832; in Tyner et al., 25% less above-ground carbon is lost.

Page 186: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

186

A second point on which studies differ is on the carbon stock attributed to vegetation on cropland (i.e., to crops). According to Nasser et al., for CARB, the figure is zero; for EPA it is 5 tCO2eq/ha; for BLUM it is 5 tCO2eq/ha for most crops and 17 for sugar cane. A third point on which studies differ is the treatment of foregone sequestration: the take-up of carbon from the atmosphere by forests that are present in the baseline scenario but not in the policy scenario. These may be (i) existing forests that are converted to cropland only in the policy scenario or (ii) existing cropland that remains cropland in the policy scenario, but is converted to forest in the baseline scenario. With some oversimplification, existing forests can be divided into those that are mature and those that are still growing. It can then be said that (i) should only be taken into account in relation to forests that are still growing. It seems that CARB do not take (ii) into account. It is not clear how they treat (i). Dumortier et al. take (ii) into account and attribute 20 years of sequestration to the forest. It is not clear how they treat (i). Searchinger et al. take both (i) and (ii) into account. The method to determine the share of growing forests assumed for (i) is not explained. For both (i) and (ii), 30 years of sequestration is attributed to the forest. (In Searchinger et al.'s paper, carbon stock changes are divided over 30 years.) Tyner et al. take (i) into account, assuming for each region that the proportions of growing and mature forest converted are the same as their proportions in that region's forest stock. They present analyses in which carbon stock changes are divided over either 30 or 100 years; this determines the amount of sequestration attributed to the forest. It seems likely that they also take (ii) into account.

Page 187: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

187

A fourth point on which studies differ is the carbon stock values attributed to different land uses. The table shows some examples for land types which appear in several different studies. Table – examples of differences in carbon stock values used

tCO2/ha

Tropical forest/rain forest, Brazil/SE Asia

Temperate forest, Europe

Boreal forest, Canada

Cerrado grassland, Brazil

Temperate grassland

Temperate cropland

Meizlish et al., 2007

549

Fritsche, 2007

539 172

Searchinger et al.

Brazil: 872-1092 SEA: 843

931-1077 1085 191 US: 330 Europe: 718

IFPRI 359-1506 339-983 71-238 139-220 139-348 111-240 Fargione et al., 2008

Brazil: 2700 SEA: 2572

311 US: 605

"German SBO draft"

971 256 202

Amaral et al., 2008

Brazil: 993 262

Tyner et al. 1085 492-515 WWF/ Cooperative Bank, n.d.

601

Mortimer et al., 2008

409

Some of the disparities may be due to differences in the precise pieces of land each value is meant to cover. Nevertheless, they appear rather significant, with the highest values of those found in the literature being between 2.2 and 15.3 times the lowest (median 5.7). Even when the source of the carbon stock values is supposed to be the same, these differences remain. Thus, Tipper et al. state that the carbon stock loss when forest is converted to non-forest use, calculated using IPCC data, is 322 tCO2/ha. However, Searchinger et al. state that IPCC data for forest to cropland conversion give losses of 299-627 tCO2/ha (temperate), and 553-824 tCO2/ha (tropical). Finally, Tyner et al. state that IPCC data give a loss of 1374 tCO2/ha for conversion of forest to cropland. Many studies state that the underlying source of carbon stock data is the IPCC. According to Koeble et al., the IPCC estimate of global soil carbon (0-100 cm), at 1500 Pg, is significantly higher than that obtainable from other sources (Harmonised World Soil Database: 1273 Pg; National Resources Conservation Service: 1376 Pg). This suggests that modelling exercises using IPCC carbon stock values may tend to overestimate soil carbon. Instead of IPCC, Searchinger et al. and Tyner et al. use the "Woods Hole" data-set. It can be seen that the carbon stock values used in these studies are generally at the high end of those in the literature. If studies using IPCC values do indeed tend to overestimate soil carbon, this

Page 188: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

188

suggests that studies using Woods Hole data may tend to do so to a greater degree (if indeed the difference is attributable to soil rather than vegetation values). A fifth point on which studies differ is the values attributed to foregone sequestration. Two studies go into detail on this: Tyner et al. and Searchinger et al. Both studies agree that the highest loss (around 5 tCO2/year per hectare of forest converted to cropland) occurs in the EU, and that the loss is also high (3-4 tCO2/ha/year) in South and South East Asia. However, for Tyner et al. there is also a significant loss (1.5 tCO2/ha/year) in the US, while the cost in Canada is close to zero. For Searchinger et al., the opposite is the case, with 2 tCO2/ha/year in Canada and close to zero in the US. None of the studies appears to take into account foregone sequestration from converted grassland. This could be because the amount of sequestration foregone is low. However, the value given in Mortimer et al., 2008 (about 1 tCO2/ha/year) does not appear insignificant.

11. Results for the land use change impact of biofuels

11.1. Estimated impact of biofuel promotion The table shows the results of some modelling exercises. Where studies assess several scenarios, preference has been given to: - scenarios with the highest gap, in biofuel consumption, between the policy scenario

and the baseline scenario; - scenarios with the assumptions that the authors appear to consider the most realistic.

Page 189: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

189

Table – land use change and biofuels – results of modelling exercises

emissions (gCO2eq/MJ biofuel) Date Exercise ha converted per toe biofuel

emissions per ha converted (tCO2eq)

land use change (over 20 years)

biofuel production

fossil fuel comparator

total

2008 Searchinger et al.833 maize ethanol 0.39 350 maize ethanol 156 biomass ethanol 167 soya biodiesel 165-270

64 19 46

-91 -91 -84

129 95

127-232

CARB834 maize ethanol 0.56 sugar cane ethanol 1.03 soya biodiesel 1.20

68 56 44

45 69 63

66 27

-96 -96 -95

15 -13

EPA835 maize ethanol 106 soya biodiesel 130

39 24

-104 -102

41 52

2009

Tyner et al.836 maize ethanol 0.14 216 36 67 -95 8 EPA837 maize ethanol 47

soya biodiesel 54 sugarcane ethanol 8 switchgrass ethanol 21

52 8 35 -24

-104 -102 -104 -104

-4 -40 -69

-128 AGLINK838 all biofuels 0.24839 Tyner et al.840 maize ethanol 0.12 225 32 64 -95841 1 Tyner et al.842 maize ethanol 0.08 289 27 64 -95 -4 Tyner et al.843 maize ethanol 0.06 287 21 64 -95 -9 Hertel et al.844 maize ethanol 0.15 222 40 65 -95 10

2010

IFPRI845 with peatland (IPCC) with peatland (Couwenberg)

all biofuels 0.11

130 130 132

17 17 (+0.16%) 17 (+1.30%)

31846 31 31

-92 -92 -92

-43 -43 (-0.22%) -43 (-1.80%)

Sources see individual references; calculations of the Commission847

Page 190: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

190

11.2. Variation of impacts among biofuels This section looks at land use change impacts of biofuels as reported by available studies by focusing on three dimensions: In how far does the extent of land use change and its related emissions as predicted by the studies lead to conclusions on whether to favour:

1. certain crops over others;

2. biofuels coming from certain geographical areas; 3. ethanol over biodiesel?

These dimensions are crucial in designing the appropriate policy response to land use change. In case of coherent results across models they could have concrete policy implications. However, most studies of ILUC focus on the overall magnitude of ILUC, and do not look systematically into these differences. The findings in this chapter are thus based on limited evidence and should be treated accordingly. 11.2.1. Results on feedstock Different feedstocks used for biofuel production might have varying land use change effects. Some studies present feedstock-specific land use change results; these could in principle be instructive in deciding whether the use of specific feedstocks should be limited. A first indicator to look at is the decomposition of production increases of different feedstocks as a result of biofuel policy and specifically at the relative roles played by land use change versus intensification for increasing production.

I. Decomposition of production increases into yield increases via fertiliser input, yield

increases by shifts in factor inputs and yield increases due to land use change. The table below compiles results from the IFPRI report (Al-Riffai et al., 2010). They indicate for which crops land use change is likely to be the primary source of production increase. Apart from land use change, crops output in the IFPRI model (MIRAGE) can be increased by intensification in terms of increased fertiliser input and in terms of increasing the ratio of other-factor inputs over land input. The figures do not provide a clear-cut picture in terms of biodiesel versus ethanol as feedstock. They point at a high reliance on land use change in Brazil for sugar cane and soybeans, in Latin America for soybeans, in Indonesia/Malaysia for palm oil and in CIS countries for sunflower expansion while wheat expansion is completely achieved via intensification (global land area for wheat cultivation actually falls, due to intensification, which results in the negative number below).

Page 191: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

191

Table: Decomposition of production increase, MEU_BAU compared to REF, % share of LUC out of total increase (Source: own calculation based on the IFPRI-report, p. 61)

PalmFruit World 76% Maize World 13% IndoMalay 70.1% USA 48.5% Rapeseed World 60% Sugar_cb World 69% EU27 43.1% Brazil 78.5% Soybeans World 55% EU27 0.8% Brazil 89.6% Wheat World -4% LAC 73.6% USA 26.9% Sunflower World 96% CIS 64.4% EU27 38.6%

The results suggest that EU27 is responding with yield increase rather that land expansion. But they do not tell us whether an incremental demand for a particular feedstock in the EU is likely to lead to a different result, in terms of land use change, than an incremental demand for the same feedstock elsewhere. This makes it difficult to know how far these figures, and those reported below, describe differences in feedstocks and how far they describe differences in place of production. The results follow the historical pattern of intensification in western countries and more reliance on land use change elsewhere, however, the marginal highest gains from intensification are probably located outside western countries, as these regions are less intensified from the outset.

II. Marginal land use change effects for individual biofuels The same authors provide feedstock-specific land use change results for ethanol and biodiesel. We here report figures from Tables 12 and 13 of the IFPRI report on the marginal land use change effect of adding an extra EU-demand for the respective biofuel of 1 million GJ in 2020 on top of a 5.6% share of biofuels from land. The results suggest that sugarcane/beet-based ethanol, both high energy yielding crops, have a considerably smaller land use change impact compared to especially maize-based ethanol and all biodiesel types, the highest emissions being found for soybean biodiesel due to its expansionary effects in Brazil. The authors compare the marginal land use change effects per feedstock to the average land use change effect they find of 17.73 gCO2eq/MJ (Table 11). Marginal effects are expected to exceed the average because of decreasing marginal productivity. Furthermore, the average land use change is much below the marginal biodiesel land use change effects because the additional production required to meet the 5.6% share is mainly met by ethanol and more specifically by sugar cane and sugar beet which have high energy yields per hectare. It is somewhat surprising that palmoil biodiesel performs better than rapeseed biodiesel in terms of estimated marginal land use change effects even when peatland effects are taken into account as they are in the figures reported below. One driving factor is the six-time higher energy yield per hectare for palm oil compared to rapeseed yields (p. 64). This effect dominates the fact that palmoil biodiesel leads to less by-products and the conversion of potentially higher-carbon-stock land. In fact, large amounts of by-products from rapeseed biodiesel production do not save net land according to this modelling exercise, as the

Page 192: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

192

dominating effect is increased meat consumption instead.848 A recent study of the US biodiesel feedstock market (Thompson et.al. 2010), indicates that fats and oils of various kinds are highly substitutable, which can partly explain why rapeseed performs not so well. However, this also depends on the Armington elasticises used for vegetable oils. (These measure the extent to which a good produced in one location is consumed as if it were a good substitute for the same good produced in another location.) In the IFPRI study, the researchers chose to replace the standard GTAP Armington elasticities with higher values, with the result that palm oil becomes a more attractive substitute for rape oil, Another critical assumption – while also holding for ethanol crops – are the decreasing crop yields in the EU over the period 2007 to 2020. Concerning peatland emissions, the authors take marginal coefficient of extension of palm tree plantations on peatlands of 10% for Malaysia and 27% for Indonesia849.

Table: Marginal land use change effect of different biofuels and feedstocks (Source: IFPRI report, pp.65-66)

Marginal Indirect Land Use emissions,

gCO2/MJ per annum, 20 years life cycle With peatland effect MEU_BAU MEU_FT Ethanol 17.74 19.18 SugarBeet 16.07 65.47 SugarCane 17.78 18.86 Maize 54.12 79.15 Wheat 37.27 16.12 Biodiesel 59.78 55.76 Palm oil 50.13 48.31 Rapeseed oil 53.68 51.24 Soybean oil 75.4 67.86 Sunflower oil 60.53 56.89

When the marginal land use change results from the BAU scenario compared with the "Free Trade" scenario, it is surprising to see how sugar beet goes from 16.08 g/MJ to 65.47 g/MJ, while the wheat ethanol figure is more than halved. The authors of the study have expressed uncertainty about the appropriateness of these marginal figures as a basis for policy-making given the high level of uncertainty attached to them850,. The final rule of EPA (EPA 2010) includes a comparison of FAPRI-CARD results with GTAP results of the same biofuel policy; this is found in chapter 2.4.11.1.3 of the final rule. Ethanol from maize uses globally 4.5 ha/billion BTU (FAPRI), while soybean biodiesel uses 12.4 ha/billion BTU (FAPRI), almost three times as much. Similar results are obtained with GTAP. While not explicitly stated it is therefore to be assumed that the larger part of production increase of maize for ethanol is met through intensification. Such a result would suggest that from a land use change perspective is maize preferred over soybean.

The final rule of EPA separates the analysis of domestic LUC occurring in the US, and international LUC outside the US. The following table sums up the main results

Page 193: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

193

Table: land use change according to EPA in g CO2/MJ and divided over 20 years

Fuel US LUC International LUC

Maize ethanol -3 51

Soybean biodiesel -14 68

Switchgrass Ethanol -3 24

Sugarcane 2 6

Judging from the table one can conclude that sugarcane ethanol and switchgrass ethanol have little total LUC emissions, while the opposite is true for maize ethanol and soybean biodiesel. For the feedstocks reported in both studies, these results thus point in the same direction as the feedstock-specific ILUC results from IFPRI reported above. The EPA final rule also contains the regional breakdown of where the international land use change takes place (table 2.4-47). The disaggregation is down to country-level, (except Brazil, with 6 regions), the results are summarized below.

The production of maize ethanol in the US leads to high land use change emissions especially in Brazil; 32 g CO2eq/MJ (Amazon, central-west cerrados and South) and to lesser extent Indonesia with 5.2 g CO2eq/MJ. However, for sugar cane, which is typically grown in Brazil, land use change is in total around 0 in Brazil. The low land use change result is mainly made up of 1.7 g CO2eq/MJ from the US, and 1.7 g CO2eq/MJ in Indonesia, the rest being small contribution spread all over the world. Soy biodiesel has yet another pattern, with –29.4 g CO2eq/MJ in Amazon, while a number of other regions in Brazil and the rest of the world in total leads to a total land use change impact of 67.3 g CO2eq./MJ. The land use change emissions from switchgrass is comparably lower at 23.9 g CO2eq./MJ, mainly made up of Brazil (around 22.2 g CO2eq./MJ). From this one can conclude that what happens in Brazil is very important for the overall results. Again, soybean is critical in terms of land use change, while sugarcane looks favourable.

CARB, in their “Low Carbon Fuel Standard (LCFS)” study include land use change results for both ethanol from maize and sugarcane and biodiesel from soybean. The results for maize ethanol are (average in brackets) 27.5 – 66.5 (45) g CO2eq/MJ. For soybean the range is 40.5 – 76.5 (63) g CO2eq/MJ, while for sugarcane it is 48.5 – 85 (69) g CO2eq/MJ (converted from 30 years to 20 years for all values). The CARB results suggest that sugarcane and soybean have higher land use change emissions than maize ethanol. In the LCFS those fuels are discouraged accordingly, being assigned with a higher land use change-factor. The opposing results for sugarcane derived from EPA and CARB question their credibility, as the context is similar, as most sugarcane in any case is coming from Brazil, and no domestic production takes place. It is, however, highly uncertain whether the results from both EPA and CARB are applicable to EU demand for biofuels. This question is related to how world feedstock markets work, and whether the withdrawal of one unit of production in one place (here the US) has different impact if it is withdrawn somewhere else (here the EU). In case of a perfect international market the land use change values found in the US should be similar to those found for the same crop in the EU. This again links to the question of whether certain regions are preferable over others, which is discussed in the next section.

Page 194: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

194

11.2.2. Results on feedstock-growing regions Results according to feedstock-growing regions can shed light on where land use change is likely to take place and which land cover types are prone to be converted. A high reliance on land use change in order to boost crop output in a region is likely to lead to an adverse GHG balance of biofuels grown in that region, especially with high carbon stock land such as forests being converted. A first indicator looks into cropland extension due to biofuel mandates in different regions and its split into different source land cover types.

I. Cropland extension and its split into different source land cover types The table below provides IFPRI results that indicate a high share of forest to cropland conversion of 69% for Indonesia/Malaysia. An even higher share is attributed to China, although the absolute magnitude of this conversion is less than in the case of Indonesia/Malaysia as can be seen in the second table. The aggregate world figures show a dominance of Savannah/Grassland conversion to cropland (34%) with forests still accounting for 22% on a global level. The biggest land use change both overall as well as for all individual land categories occurs in Brazil.

Table: Share of former land uses of land converted to cropland (Source: own calculation based on the IFPRI report, p. 60)

Brazil China EU27 Indo

Malay LAC World

Forest_total 16% 95% 34% 69% 28% 22%Pasture 14% 5% 0% 19% -13%851 12% SavnGrasslnd 58% 0% 0% 0% 0% 34% Other 12% 0% 66% 12% 85% 32% Total 100% 100% 100% 100% 100% 100%

Table: Absolute land use change in 1000ha (Source: own calculation based on the IFPRI report, p. 60)

Brazil China EU27 Indo

Malay LAC World

Cropland 4814 78 780 139 396 8196 Forest_total -791 -74 -266 -96 -112 -1765 Pasture -679 -4 -2 -27 52 -994 SavnGrasslnd -2783 0 0 0 0 -2783 Other -560 0 -512 -17 -336 -2655

Hertel et al. (2010) provide similar figures with the following important underlying differences: They distinguish only between three land cover types (cropland, forest, pasture), regions are partly defined differently and most importantly their study focuses on the US ethanol mandate by 2015, which is furthermore modelled as a mandate on production and not on consumption as is the case for the EU mandate. All this makes a comparison to the IFPRI study questionable. However, one could expect similar reaction patterns to a crop demand shock as imposed by a biofuel mandate. This expectation is not met comparing the tables

Page 195: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

195

below to the IFPRI ones above. Results are comparable for Brazil but not so for other regions. The largest shares of forest area cleared are found for the EU and Canada, while the largest absolute forest conversion takes place in the US. On a global level, the split between forest versus pasture conversion to cropland is roughly 20 versus 80%. The above results show which types of land cover are likely to be converted in different regions due to EU or US biofuel policy from which one can infer a likely land use change pattern in response to additional biofuel mandates. They do not uncover, however, the direct and indirect effects at work that lead to the reported land use changes and therefore policy conclusions are difficult to draw except for those that target forest protection legislation and its enforcement especially in the regions where forest conversion is likely.

Table: Share of former land uses of land converted to cropland (Source: own calculation based on Hertel et al., 2010, Table S2)

US RoW Brazil China EU27 SSAEn

Exp LAEn Exp

RofLat America Canada

forest 34% 10% 20% -225% 64% 2% 0% -133% 64% pasture 66% 90% 80% 325% 36% 98% 100% 233% 33%

Table: Absolute land use change in 1000ha (Source: own calculation based Hertel et al., 2010, Table S2)

US RoW Brazil China EU27 SSAEn

Exp LAEn Exp

RofLat America Canada

cropland 1590 2600 300 40 450 540 180 60 450 forest -540 -250 -60 90 -290 -10 0 80 -290 pasture -1050 -2350 -240 -130 -160 -530 -180 -140 -150

II. Land use effects of marginal crop demand

Alternative results on land use change in different regions are provided by Kløverpris et al. (2008b), though not in a study relating particularly to biofuels. The authors model one additional ton of household wheat consumption in respectively Brazil, China, Denmark and the USA (four separate scenarios). Using the GTAP model with land supply curves added, doubled Armington elasticities compared to the standard version and combined with FAOSTAT data and land cover maps for reporting results in physical units and for determining which nature types are affected by agricultural expansion they come to the following main result: Adding the extra ton of wheat consumption for, in turn, each of the regions mentioned leads to a net expansion of global agricultural area of 3200, 2000, 1700 and 260m2 for the USA, Brazil, Denmark, and China, respectively. Three quarters of the Brazilian cropland expansion go at the expense of tropical evergreen forest (for more detailed results for the remaining regions see Kløverpris et al., (2008b, p.98).

While these results are of limited use as they consider a narrow range of countries and only wheat, which is by far not the most important biofuel feedstock, they are reported here as an example of modelling results that would provide very valuable insights into the land use change debate: Knowing which regional extra biofuel feedstock demand would lead to largest LUC effects could provide ground for policies steering this demand by limiting biofuels from certain regions while favouring other regions’ biofuels.

Page 196: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

196

11.2.3. Results on fuel choice The question on fuel choice is very much related to the question on feedstock choice examined before. Considering the IFPRI results on the decomposition of production effects, we showed above that both ethanol (sugar cane) and biodiesel crops (especially soybeans and palm oil) cultivation rely to a large extent on land use change. The marginal land use change calculations in the same study show that ethanol fuels lead to lower marginal land use change effects with the exception of maize-based ethanol. EPA results on international land use change are in line with the IFPRI study by attributing the greatest international land use change effect to soybean biodiesel followed by corn ethanol, both clearly outperformed by switchgrass and sugarcane-based ethanol. The CARB results do not fit the picture that emerged from the IFPRI and EPA studies by predicting the highest land use change effects for sugarcane, which seems very questionable in light of the other studies’ results. The study by Hertel et al. 2010 is not of use neither for the purpose of distinguishing between crops nor between biofuels as they only report results related to corn ethanol. A previous study by Hertel et al. (2008) does provide some insight on this question, though. The authors look at the effects of simultaneously imposing the US and EU mandate up to the year 2015 (EU mandate modelled is 6.25% share by 2015). Looking at the figures in their table 1 (p.37), one can see that the EU mandate is predominantly met by biodiesel, which plays a minor role in the US. These figures provide some justification for using the split into US and EU policy effects as contributing to answering the question of ethanol versus biodiesel, respectively. When looking at crop cover changes,852 the authors find that the EU policy is the main driver behind crop cover changes from forest and pasture (p.23 and table 8 on p.44). The bulk of the cropland increase comes come pasture land, the largest decrease being found for Brazil where pasture decreases by 11% of which 8 % points are attributed to the EU policy. Their results thus add to the direction of increased international LUC from increased biodiesel demand.

The model comparison exercise coordinated by the JRC-IE (Edwards et al., 2010) compiled results from various models and marginal biofuel scenarios. Below we report results from different models on hectares of land use change per (marginal) toe. The second column shows relative results for each model by comparing all other scenarios to the ‘Wht eth EU’ scenario (chosen because every model ran this particular scenario with the exception of LEITAP that ran a French-specific scenario). No coherent picture emerges from the models compared. Biodiesel leads to somewhat higher LUC in FAPRI and EU/German biodiesel leads to much higher LUC than the remaining scenarios in the LEITAP model. However, in most cases this exercise suggests that bioethanol causes greater land use change than biodiesel. Further, these results do not convey information about emissions resulting from LUC. JRC-IE calculated emissions based on a uniform emission factor of 40 tC/ha, providing uncertainty ranges from 10-95 tC/ha. Looking at the total of results in their figure 22 shows that the highest emission values are found in biodiesel scenarios while the lowest are found in ethanol scenarios. However, comparing scenarios within models, again no clear-cut picture of biodiesel versus ethanol emerges for all models (Edwards et al., 2010, p.84). As the authors rightly acknowledge, using a uniform emission factor does not properly account for regional differences in soil carbon content due to varying climatic and soil conditions. Two models investigated, FAPRI-CARD and GTAP provide a more disaggregated analysis of LUC emissions. FAPRI-CARD compares a EU wheat ethanol scenario to a EU rapeseed biodiesel scenario and attributes roughly fourteen times higher emissions to the biodiesel scenario (221.6 versus 15.5 g CO2/MJ/year over 20 years). These results are driven by the extreme assumptions that the EU ethanol consumption increase comes entirely from domestic wheat with cropland expanding into either idle land or grassland within the EU. Most land use

Page 197: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

197

change resulting from the biodiesel consumption increase, on the other hand, takes place outside the EU (Edwards et al., 2010, p.86). Results based on LUC and regional emission factors from GTAP attribute the highest emissions to the EU wheat ethanol scenario (155 g CO2/MJ/year over 20 years) while the US coarse grain ethanol scenario, EU biodiesel mix and Mala-Indo biodiesel yields emissions of 62, 57, 47 g CO2/MJ/year, respectively (Edwards et al., 2010, p.88).

Table: Model comparison of LUC (Source: JRC-IE study, second column own calculations)

Model Scenario LUC (ha/toe) Wht Eth EU=1

AGLINK Wht Eth EU 0,57 1 AGLINK Maize Eth US 0,51 0,89 AGLINK Sugar Cane Eth Bra 0,13 0,23 AGLINK Biod EU 0,23 0,40 AGLINK Biod US 0,24 0,42 FAPRI Wht Eth EU 0,39 1 FAPRI Biod EU 0,40 1,03 GTAP Wht Eth EU 0,79 1 GTAP Maize Eth Us 0,16 0,20 GTAP Biod Mix Eu 0,38 0,48 GTAP Biod Ind/Mal 0,08 0,10 LEITAP Wht Eth EU-Fra 0,73 1 LEITAP Maize Eth Us 0,86 1,18 LEITAP Biod EU-Deu 1,93 2,64 LEITAP Biod INDO 0,43 0,59

Note: Results from the IMPACT model have been omitted because they only consider marginal ethanol scenarios.

11.2.4. Conclusions The above reported results show that various modelling exercises have not managed to present definite and detailed conclusions on whether or not to prefer certain feedstocks, feedstock-growing regions or fuel types. There can be large ranges of uncertainty within studies and partly contradicting results across studies. While no definite conclusions about the size of the effect nor about the preference for certain fuels/feedstocks/regions over others can be drawn, the results of some models nevertheless display a tendency to attribute higher land use change effects to biodiesel feedstocks – especially soya. However, even in these cases the results do not permit a reliable check of whether this result only applies to some vegetable oil feedstocks or to them all. There is also a tendency for the largest reliance on land use change as a means of increasing crop production and/or the highest absolute (forest) land conversion taking place in Indonesia/Malaysia, Brazil, other Latin American countries or, in other words, in countries with larger amounts of high-carbon stock land as opposed to e.g. Europe. The first point does provide justification for steering production and consumption towards more ethanol as opposed to biodiesel. The second point favours the use of higher energy yielding crops so that less land is used for biofuel feedstock production, thus reducing the replacement of food crops and weakening indirect land use change effects.

It is necessary to devote more reasesch into the question of whether feedstock, fuel or origin matters for the land use change effect. At the moment, the results are too uncertain to be a basis of firm conclusions.

Page 198: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

198

12. Model calculations: comparing land use effects with GHG savings from biofuels

12.1. GHG emissions from biofuel production

12.1.1. Objectives When working on biofuels indirect land use change potential effects, it seems relevant to look into GHG emissions from biofuels excluding land conversion since these figures are the basis of the final GHG emissions calculations. Therefore it is worth having an overview of the literature results regarding biofuels GHG emissions without land use and evaluating to what extent these results are coherent or not. At the European level, the Directive on the promotion of the use of energy from renewable sources (2009/28/CE) fixes the European method for calculating these emissions for regulatory purposes. The method is detailed in Annex V of the Directive and summarised in a formula. Co-products are taken into account using the energy allocation method. The Commission, through its Joint Research Centre (JRC) also calculates values on the basis of the same data, but taking co-products into account using the substitution method, which is thought inappropriate for regulatory purposes but more appropriate for policy analysis. The objective of this note is not to take any position on how biofuels GHG emissions savings (without land use effects) should be calculated but rather to list and compare what values have in fact been used in a range of scientific studies based on different assessments.

12.1.2. Introduction GHG are emitted at all stages of the biofuels production chain: fuel used for the production harvesting; collection and transportation of bioenergy feedstocks; energy required for producing fertilizers and pesticides; chemical processing of feedstocks; and distribution of biofuels to end users. At present, a number of different methods are being used for life-cycle analysis. When performing a life cycle assessment (LCA), one is usually interested only in one specific product. It is thus necessary to allocate / distribute the resource consumption and the emissions associated with the processes over the main product and its co-products853.

12.1.3. Main biofuels LCA analysis presentation We decided to focus on five LCA studies that, according to our readings, are relevant to point out the diversity of the results currently available on GHG savings from biofuels, excluding land conversion. RFA 2008 and RFA 2009 (UK) : for the current biofuels considered, default values assume the use of fossil fuel fired boilers and imported electricity in all cases apart from bioethanol production from sugar cane in Brazil where bagasse-fired CHP is adopted. Until recently, the RFA Technical Guidance adopted the use of substitution credits for all products other than the biofuels under consideration. However, it should be noted that, in the latest version of the

Page 199: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

199

RFA workbook, options are kept open by including facilities for allocation by energy content and price. ADEME (France): In September 2009, ADEME published on its internet website the synthesis of a study dedicated to LCA of first generation biofuels consumed in France. This study uses the substitution method when co-products are used for energy sources or as fertilizers and the allocation method by energy content when co-products are used to feed animal or for industrial uses. NB: this synthesis was removed by ADEME from its internet website a few weeks after the publication in order to be further improved thanks to the contribution of stakeholders. Therefore, the results presented are provisional and are listed as additional data on biofuels GHG emission savings. They will be updated when ADEME will publish the definitive results. EU: In the 2009/28 Directive on the promotion of renewable energy (Annex V), the method used for biofuels GHG emissions calculation is the allocation by energy content for co-products. However, recital 81 of the Directive states that "the substitution method is appropriate for the purposes of policy analysis, but not for the regulation of individual economic operators and individual consignments of transport fuels. In those cases the energy allocation method is the most appropriate method, as it is easy to apply, is predictable over time, minimises counter productive incentives and produces results that are generally comparable with those produced with the substitution method". As this paper's purpose is to carry out a policy analysis, the EU figures below are the substitution method figures. USA Environment Protection Agency (EPA): in order to calculate GHG emissions associated with the production of certain biofuels, the EPA uses the Greenhouse gases Regulated Emissions, and Energy use in Transportation (GREET) model, mainly based on the substitution method. Within this model, allocation by energy content is used for electricity but otherwise, displacement of product or use of substitution credits is adopted where possible. This model was used by Searchinger and al in several scientific papers. FAO and IEA854 : At this stage, no information or comments were found regarding the method used by FAO and IEA to calculate biofuels GHG emissions savings. However, since these figures are used in several scientific papers, we decided to include this study in this note.

12.1.4. Results

Page 200: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

200

RFA, carbon and sustainability reporting within the RTFO, technical guidance, carboner reporting - default values and fuel chains, 2008 and 2009Product GHG emissions (g CO2/MJ) GHG emissions savings %Gasoline 85diesel 86wheat to ethanol (high value : Ukraine) 103 -21%wheat to ethanol (low value : Germany) 59 30%Sugar beet to ethanol (UK) 50 41%corn to ethanol (low value : France) 49 42%corn to ethanol (high value : USA) 108 -27%Sugar cane to ethanol (low value : Brazil) 25 71%Sugar cane to ethanol (hight value : Pakistan) 115 -36%Oilseed rape to biodiesel (high value : USA) 93 -8%Oilseed rape to biodiesel (low value : Poland) 45 47%Soy to biodisel (high value : Brazil) 78 10%Soy to biodisel (low value : Spain) 46 46%Palm to biodiesel (hight value : Indonesia) 47 46%Palm to biodiesel (low value : Malaysia) 47 46%Sunflower to biodiesel (high value : Russia) 69 20%Sunflower to biodiesel (low value : USA) 21 76%Used cooking oil to biodiesel 13 85%Tallow to biodiesel (hight value : USA) 17 81%Tallow to biodiesel (low value : UK) 13 85%Oilseed rape to pure plant oil (hight value : USA) 84 2%Oilseed rape to pure plant oil (low value : Poland) 34 60%

Forestry residues to bioethanol 9 90%Miscanthus to bioethanol 7 92%Straw to bioethanol 23 73%Waste wood to bioethanol 5 94%Forestry residues to diesel (Fischer-Trop) 19 78%Miscanthus to diesel (Fischer-Trop) 17 80%Straw to diesel (Fischer-Trop) 8 91%Waste wood to diesel (Fischer-Trop) 22 74%

Page 201: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

201

ADEME, LCA of first generation biofuels consummed in France

Product GHG emissions (g CO2/MJ) GHG emissions savings %Gasoline (euro 4) 102diesel 96Sugar beet to ethanol 28 27%Wheat to ethanol 44 43%Corn to ethanol 37 36%Sugarcane to ethanol 23 22%Rapeseed to biodiesel 39 40%Sunflower to biodiesel 26 27%Soy to biodiesel 20 21%Palmoil to biodiesel 21 22%

Page 202: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

202

European Union

Petrol 85.8diesel 87.4sugar beet ethanol 38.1 56%wheat ethanol 79.3 8% (lignite-CHP)

61.2 29% (conv)48.5 43% (natural gas CHP)26.6 69% (straw CHP)

maize ethanol (EU produced) 47.0 45%sugar cane ethanol 13.1 85%rape seed biodiesel Average of 46.6 and 41.5 50%sunflower biodiesel Average of 31.0 and 25.9 67%soya biodiesel Average of 76.3 and 72.8 15%palm oil biodiesel Average of 51.9 and 48.4 43% (without methane capture)

23.6 73% (with methane capture)waste vegetable or animal oil biodiesel* 10 88%pure vegetal oil from rapeseed* 35 58%wheat straw ethanol* 11 87%waste wood ethanol* 17 80%farmed wood ethanol* 20 76%waste wood Fischer-Tropsch diesel* 4 95%farmed wood Fischer-Trop diesel* 6 93%

* : no difference between the substitution method results and the allocation method results set up by the RE Directive

GHG emission gCO2eq/MJ GHG saving

Page 203: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

203

FAO, The state of food and agriculture, 2008 : reduction in GHG emissions of selected biofuels relative to fossil fuels (exclude the effects of land use change) IEA, world energy outlook 2006Product GHG emissions (g CO2/MJ) GHG emissions savings %sugar beet ethanol (EU) 40 to 60%sugar beet ethanol (Brazil) 70 to 90%maize ethanol 15 to 35%maize ethanol (USA) 12 to 15%rape seed biodiesel (EU) 40 to 60%palm oil biodiesel 50 to 85%second generation biofuels 70 to 90%

Page 204: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

204

USA Environment Protection Agency (EPA) : At this stage, we only managed to get USA EPA figures on biofuels GHG emissions saving including land use change855 . It seems that to date, there is no official available information from EPA on biofuels GHG emissions saving calculation without land use effects.

12.1.5. Comment on main first-generation biofuels GHG emissions The table summarises the values given in the studies for the main biofuels. Table: greenhouse gas emissions attributed to biofuels (excluding land use change)

gCO2eq/MJ Highest result Lowest result Wheat ethanol 103 26.6 Sugar beet ethanol 50 28 Corn ethanol 108 37 Sugar cane ethanol 115 13 Rapeseed biodiesel 93 39 Sunflower biodiesel 69 21 Soya Biodiesel 78 20 Palm oil biodiesel 51.9 21 The chart presented above shows variations between GHG emissions savings calculations. Due to many factors, there is a wide range of results cited in the literature. Indeed, "Greenhouse gas balance published in the available literature differ widely among crops, locations, and conversion technologies, as well as the (…) method used in accounting for by-products and specific assumptions about energy sources used in the production of agricultural inputs and feedstock conversion to biofuels ".856 It is clear, therefore, that the choice of value for biofuel greenhouse gas emissions could have a significant impact on the results that studies report for the overall greenhouse gas impact of biofuel promotion. CARB, EPA and the other US studies use GREET, but the precise values used were not in the sources used for this literature review. For the EU studies, data on the values used have not been identified. Thus, it has not been possible to identify in what way the choices made may have affected the reported results. It is worth highlighting that the existing figures presented in the different scientific studies are likely to change in the following years, depending of the baseline improvement to the GHG emissions of the different chains. This scope for improvement was assessed by the Energy Research Centre (ECN) of the Netherlands857 :

Ø N2O emissions related to N fertiliser: since 2008, N2O emissions in the fertiliser

industry have become part of the EU-ETS. Recently several major parties in the fertiliser manufacturing industry have successfully implemented end-of-pipe technology to existing fertiliser plants. The ECN study assumes a 90% reduction of nitrous oxide emissions in N fertiliser to be realised on a relatively short term.

Ø CO2 emissions related to N fertiliser: these emissions will also be subject to

reduction ambitions, either through the EU ETS or through non ETS policy. The

Page 205: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

205

CEN study assumes a 25% reduction of CO2 emissions in N fertiliser production by 2020

Ø Emissions not related to N fertiliser production: major other sources of GHG

emissions in feedstock production are N2O emissions from fertiliser application and CO2 emissions related to energy inputs. A wide variety of emissions reduction options exists, but its mitigation potential is difficult to assess. The ECN study assumes a 5% reduction of GHG emissions in feedstock production by 2020, for emissions other than those related to N fertiliser.

Ø Biofuels processing: the biofuel processing industry will be subject in the

following years to CO2 reduction and energy efficiency improvement policies. The CEN study assumes a 10% reduction of CO2 emissions in the biofuel processing industry by 2020, with a linear development towards that level from 2008. Additionally, it assumes a 20% average reduction in methane emissions at palm oil mills by 2020.

12.2. The fossil fuel comparator

12.1.6. Introduction The fossil fuel comparator (FFC) is needed in order to determine the relative savings obtained through the use of biofuels. The literature on the subject is varied, and results vary, due to differing assumptions deciding on what kind of fossil fuels that biofuels will substitute i.e. what type or mix of fossil fuel is on the margin. In this kind of policy evaluation it is the marginal change that is of interest, as one is asking "what is the impact of the policy". Average values can be used as an approximation of the marginal, but this is unlikely to be the correct assumption. When one is discussing the marginal one has to distinguish between the short term marginal and the long term marginal, as both the premises of the analyses and the expected results vary considerably according to the timeframe. This is discussed in the first section.

12.1.7. Fixed versus variable capital A clear distinction to simplify the understanding of the matter is made: "Short term margin" refers to the operational decisions made in response to a shock in demand, while "long term margin" refers to the operational and investment decisions made in response to a shock in demand. The first assumes fixed capital while the latter includes the possibility to change the capital i.e. investments and closure of installations. The analysis of marginal fossil fuel is often limited to the short term margin, as economic modelling of the oil market is possible in the short term, when investment decisions are held outside of the analysis. The understanding of how operators act when investment is included is considerably more complicated as the competition is far from perfect, with a large cartel being a major player, geopolitics playing its part and general uncertainty of actor's decision-making further complicating the picture. Although companies are still expected to be profit maximizing actors; a wide range of uncertainty comes into play, making modelling difficult and bound to be uncertain. In this picture, when capital is flexible, also government policy plays its role by not only defining the economic framework of the industry, but also influencing the expectations of future attractiveness of different fossil fuel projects. Evaluating the impact of the Renewable Energy Directive and the Fuel Quality Directive one should assume capital to be variable, i.e. analyse the long term marginal, as the legislation is

Page 206: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

206

far-reaching in both time and volume, and can be expected to influence decision making concerning investments.

12.1.8. Marginal source of crude A general impression from literature is that conventional oil will peak within the next 15 years858, and that unconventional sources will fill the gap between demand and available conventional sources. This leads to unconventional sources setting the price for liquid fuels.859 Although general observation can be made, it is questionable whether one can assign a certain mix of fossil fuel emissions avoided to a specific policy. The range of possible outcomes ranges from synthetic fuels from coal on the one hand having the highest emissions to conventional fuels on the other hand with the lowest emissions. The section below summarizes some findings from the literature.

12.1.9. Marginal source of crude - Figures from literature At the F.O. Licht World Ethanol Summit 2009 BP biofuels presented the following cost curve for fuels in 2020, thus indicating the different sources that might be at the margin for different oil prices.

Figure 1: Cost curve for different quantities of oil equivalents The figure above gives an indication of which sources that are most sensitive to a long term change in oil price. To some degree higher production cost is linked with higher GHG emissions, but not systematically. E.g. "Deep water" and "Arctic" sources are more costly for other reasons than high energy consumption per barrel extracted crude. However; the general picture is that more expensive crudes are connected with higher emissions. This will be further discussed in relation to the paper by Brandt and Farrell (2007) further below. Persson et al. (2007) looked at what consequences a carbon constrained world would bring for the oil producers in the long run, up to the year 2100. A 450 ppm scenario is compared to a BAU scenario using the GET optimisation model. In the carbon constrained scenario mainly the supply of synthetic fuels (from coal and oil shale) is reduced, followed by reduction in heavy crude/tar sands from 2030. The difference in rate of extraction of conventional oil is less than 1% for the two scenarios, leaving synthetic fuels from coal and oil shale on the margin from 2010 and onwards. This implies that those sources are expected to be on the margin, while conventional extraction rates remains practically unchanged as a consequence

Page 207: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

207

of the climate policies put in place. This indicates that conventional oil is not the marginal long term source of crude. The Renewable Fuel Association (RFA 2009) calculated the percentage shares of different crudes which biofuels are supposed to displace. The starting point is figures for growth in conventional and un-conventional sources towards 2030 projected by Energy Information Administration (EIA), where a quantity of biofuel is among the un-conventional sources. The marginal source is assumed to be that mix of conventional and un-conventional sources. The question then asked is what fills that gap, if biofuels are taken out. In the worst case (highest emissions), RFA assumes that biofuels is replaced by only the non-conventional fuel growth although the marginal is still a mix of conventional and unconventional sources, but now with the small biofuel share of the margin substituted by other un-conventional sources. The best case (lowest emissions) is based on the assumption that the biofuels will be replaced by all sources predicted to be part of the projected growth mix of 58% conventional and 42% unconventional sources.860 This implies that growth in both conventional and un-conventional crude production is increased to fill the gap. The marginal mix for "worst" and "best" case is shown below in the table.

Source Worst Case Best Case Conventional 58.0% 69.6% Bitumen (Tar Sands) 23.3% 16.9% Coal-to-Liquids 10.2% 7.4% Extra Heavy Crude 4.5% 3.3% Gas-to-Liquids 2.4% 1.7% Oil Shale 1.4% 1.0% Other Unconventional 0.3% 0.2%

The resulting emissions are (using weighted average) 101 – 109 g CO2eq./MJ of fuel in the best case, and 104 - 115 g CO2eq./MJ in the worst case.861 Although these figures are from the US, the figures used are EIA’s global liquid fuel supply projections for 2030. In general one might argue that due to shuffling of crude sources, the only interesting and relevant projection would be the global supply and not a continental one like e.g. for the US or Europe.862 (In practice, however, markets may not behave exactly as this perfect-competition assumption implies.) Johnston (2009) states that more carbon-intensive petroleum fuels are the real competitor of biofuels, and when this is accounted for the net CO2 emissions of the fossil fuel the comparator increase by 17-30% for tar sands and heavy oils; 75% for coal-to-liquid syn-fuels and 30-250% for low-grade oil shale resources.863 OPEC assumes that OPEC crude is on the marginal in their assessment of biofuel policies, as described in the "World Oil Outlook" (WOO) 2008, but this is not a modelling result, rather an assumption used as input to the WOO-modelling864. Brandt and Farrell (2007) use the IPCC Special Report on Emissions Scenarios865, and assess among other things how the uncertainty of availability of petroleum sources affects the emissions in a business as usual scenario. The uncertainty with regard to availability, costs and emissions is summarized in the graph below, where the already consumed fuels are placed to the left of the y-axis. Shaded bright areas indicate increasingly uncertain resources. A conservative estimate would be to add up only the dark portions. The lower part of the graph indicates the upstream emissions together with the combustion emissions. Note that the unit of y-axis is grams C. To compare with grams of CO2 equivalents one has to multiply with

Page 208: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

208

3,664866. The numbers for the GHG emissions are converted and summarized in the table further below

Figure 2: Global supply of liquid hydrocarbons in dollars (top) and carbon emissions (bottom) The uncertainty of available resources and the variation in emissions is considerable. It is also clear the availability of fossil carbon in absolute term is not a medium term concern. The EPA (Environmental Protection Agency) in the impact assessment of their fuel regulation uses the average of petroleum-based fuels sold in 2005 as a fossil fuel comparator value.867 Those are 102.1 g CO2eq./MJ for diesel and 103.8 g CO2eq./MJ for gasoline. In the final rule by EPA, the values are changed slightly, due to a new set of assumptions on the 2005 baseline: Diesel 102.3 g CO2eq./MJ and Gasoline 103.6 g CO2eq./MJ. The overview below summarizes some of the results found in the literature on the topic. Pathway g CO2eq./MJ Source: RFA Technical Guidance value 86 Mortimer et al. (2008), p.24.

Page 209: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

209

– Diesel RFA Technical Guidance value – Petrol

85

GREET model, conventional low sulphur – Diesel

87

GREET model, conventional low sulphur – Petrol

97

GREET model, tar sands ~ 140 Kim et. al. Conventional Gasoline 93.3 Gas-to-Liquids 100 – 109 Bitumen (Tar sands) 107.7 – 131.5 Extra Heavy Crude 107.7 – 131.5 Oil Shale 120.9 – 256.9 Coal-to-Liquids 153.2 – 178.7

RFA (2009), p.4.

Conventional Gasoline / Diesel 94.2 / 93.5 Enhanced Oil Recovery 96.1 – 112.6 Gas-to-Liquids 100.1 – 108.9 Tar sands / Extra heavy oil 107.8 – 131.6 Oil shale 121 – 256.7 Coal-to-Liquid 153.3 –178.6

Brandt and Farrell (2007), p.249. (with own conversion calculations)

EPA proposed rule Diesel fuel 102.1 EPA proposed rule Gasoline 103.8 EPA (2009), p.313868.

EPA final rule Diesel fuel 102.3 EPA final rule Gasoline 103.6

EPA (2010), p.256 and 259869.

Renewable Energy Directive 83.8 Directive 2009/28/EC

12.1.10. Discussion To predict the long term marginal crude is not trivial. Will a policy like the Renewable Energy Directive influence investment in upstream petroleum equipment directly or only in-directly through a change in the oil price? In both cases, the question of which technology that will be substituted remains, although it is likely that it will be one of the more expensive ones. The more expensive crude sources are likely to be more polluting, as more energy is needed per barrel extracted. Although this is not always the case, it is a general picture that holds, since such sources, like arctic and deep water resources with high costs but not necessarily high emissions, are not expected to contribute considerably to the global crude supply. The different attempts of modelling the long term marginal have concluded in different ways, but none seem to assume that conventional crude is the marginal long term source. The fossil fuel comparator value chosen for the determination of the indirect land use change impact by IFPRI (91.7 gCO2eq/MJ)870 is thus low compared to unconventional fuels.871 The key question is, if biofuels were not to be used, what type of fossil fuel would be added to the mix to fill the gap? This means that the fossil fuel comparator should be the marginal source of fossil fuel in 2020, not the average. It should be the long term rather than the short term marginal, because the legislation that promotes the use of biofuels is far reaching in both time and volume, and can therefore be expected to influence decision making concerning investments.

Page 210: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

210

In practice, this is not the case in the studies reviewed here. EPA, CARB and IFPRI all use as their fossil fuel comparator an estimate based on average, present-day emissions of conventional crude (102-103, 95-96 and 91.7 gCO2/MJ, respectively). No study has been identified which takes a different approach. By contrast, while studies in the literature that attempt to identify the long term marginal fossil fuel source have concluded in different ways, none seems to assume that conventional crude is the marginal source. For Persson et al., 2007, synthetic fuels from coal and oil shale are on the margin. This implies emissions of at least 120 g/MJ. For Renewable Fuel Association, 2009, the margin will contain a mix of conventional and unconventional fuels, with emissions of 101-109 g/MJ in the best case and 104-115 g/MJ in the worst case. Johnston, 2009, also sees the more carbon-intensive fuels as being on the margin. Charts from BP, 2009 and from Brandt and Farrell, 2007 both suggest that at higher oil prices it is unconventional fuels that are on the margin. This difference can have a significant impact on studies' results. For example, if CARB and EPA used a fossil fuel comparator that was 15g higher than the one actually chosen (which would be consistent with RFA's worst case assumption and less than the values implied by Persson et al. and by Johnston), this would be equivalent to reducing the estimated land use change impact by about 30%.

12.3. Comparing flows with stock changes

12.3.1. Introduction Most policy decisions intrinsically involve a assessment of time, as the policy outcome often only occurs sometime in the future, while the cost to the society often has to be covered today, i.e. there is act of exchange not only between different actors, but also across time. This is certainly true in the field of climate change policies, where a range of value-laden judgement has to be made in order to strike the balance between benefits and costs across generations (as well as between different actors). One normally thinks of the valuing of time having most relevance for economic analysis, however it also has implications in the field of analysing e.g. damage occurring or determining the relative warming from the emission of greenhouse gases (GHG) in different circumstances. It is the latter issue that this section discusses and in the more specific field of land use change as a consequence of the use of biofuels, where a range of issues concerning accounting of time arises872. Two separate, although closely interlinked, groups of issues can be distinguished that simplifies the analysis: a) the valuation of warming potential from one pulse873 of different GHG over the lifetime of those gases and b) the valuation of emissions occurring from a project/policy at different times, i.e. the ploughing of grassland will have land use change emissions from the soil several years after the actual land use change took place. The first issue is discussed in the following section, while b) is discussed in the context of biofuels in the subsequent section.

12.3.2. Global warming potential Already the choice of how to index the radiative forcing of different GHG implies value judgment, as the timeframe is crucial for weighting the short-lived GHG against the longer lived ones. The IPCC global warming potentials are given for 20, 100 and 500 years. The values for 100 years have been adopted under the Kyoto protocol and also under the Renewable Energy Directive874, and are generally widely used. However, one should be aware that these figures applies no discounting, i.e. the radiative forcing that one unit of GHG

Page 211: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

211

causes in 50 years is valued as much as the radiative forcing that the same unit of GHG causes today, but the forcing occurring in one year after the timeframe chosen is assumed to be zero875. This is exemplified in the graph below, showing the reference gas (r) in green (normally CO2) and the gas to compare (c) in dark blue. The GWP is determined by the relation between the accumulated radiative forcing, i.e. the relation between the two areas. With a 100 year timeframe; gas c has a GWP of 0.7 (the reference gas has a GWP of 1). However if the timeframe is longer one can clearly see how the relative importance of the reference gas is increasing, as this gas has a longer lifetime and thus contributes to warming to a large extent also after 100 years. Thus having a timeframe of 500 years decreases the GWP of gas c to 0.3.

0 100 199 299 399 499Time

Rad

iativ

e fo

rcin

g

Reference GHG

Compared (c) GHG

Reference GHG >100 yr

Compared (c) GHG > 100 yr

Figure 3: Schematic overview of how the GWP are derived; relative accumulated warming This problematic is also evident not only for hypothetical gases as here, but also for e.g. methane, the second most important GHG in terms of global emissions, where the GWP is 72 in a 20 years timeframe, but only 7.6 in a 500 year perspective due to short lifetime876. It is evident that such choices as demonstrated here have considerable impact on policy making. The choice made does not only influence the relative weighting of different GHG, but also of the same GHG released at different times, as will be more thoroughly discussed in the next section. The figure below shows how 3 pulses of CO2 are accounted for under the framework of GWP, where one unit is released today, another is released in 25 years and then, in 75 years from now, one unit is sequestrated, shown as hatched area (for simplicity; the increased radiative forcing is here assumed to be linear to the quantity of carbon let out into the atmosphere).

Page 212: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

212

-1,5

-1

-0,5

0

0,5

1

1,5

2

2,5

0 100 199 299 399 499

Time

Rad

iativ

e fo

rcin

g

Figure 4: Carbon concentration over time according to the GWP with 100 years perspective Now, applying the Bern Carbon cycle model877 to these same pulses of emissions shows the actual accumulation of carbon in the atmosphere, which is rather different from the figure above, due to the decaying concentration of carbon in the atmosphere (the radiative forcing is calibrated assuming equal accumulated warming after 100 years – thus the area under the graphs after 100 years are equal).

-1

-0,5

0

0,5

1

1,5

2

2,5

3

3,5

0 100 199 299 399 499

Time

Rad

iativ

e fo

rcin

g

Figure 5: Carbon concentration over time according to the Bern Carbon Cycle model Still assuming that radiative forcing is linear to the concentration of GHG; one can see how the GWP-method puts different weight on emissions occurring at different times. The figure below shows the accumulated radiative forcing, both with and without the sequestration taking place after 75 years.

Page 213: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

213

0

50

100

150

200

250

300

350

400

0 100 199 299 399 499Time

Rad

iativ

e fo

rcin

g

GWP-100 with seq. Warming(ppm) with seq.GWP-100 Warming(ppm)

Figure 6: Accumulated radiative forcing using GWP and the Bern Carbon cycle model From figure XX4 one can see how the calculated warming follows the same pattern the first 100 years. For the time after 100 years; the benefit of sequestration grows with time applying the Bern Carbon cycle model compared with the GWP model, whereas for the latter the benefit remains stable after a transition period. The differences over time are not surprising, as one unit of GHG continues the warming also after 100 years applying the Bern Carbon cycle model, indicating that one unit released into the atmosphere causes warming long after the initial 100 years. Following the Bern Carbon cycle model it appears apparent that any reversion of emissions through sequestration taking place at a later stage will still cause warming through the period that this GHG was temporary in the atmosphere. This is linked to the fact that the net emissions to the atmosphere is not causing warming directly, but rather the contribution emissions have on the carbon stock in the atmosphere over time. As a conclusion one can summarize that using the GWP model with 100878 years timeframe implies a valuation of time, as emissions are only counted the first 100 years, although with a constant valuation over this time period. Compared to the more realistic representation of the carbon cycle and thus the warming effect, as described by the Bern Carbon cycle model, the change in warming calculated using GWP is lower, indicating that the use of GWP is equivalent in warming terms, to using a discounting factor (equal to 0.9% annual discount rate (Fearnside 2002)). For e.g. biofuel projects with associated land use change, this implies that the use of GWP is decreasing the warming compared to having no discounting and using the Bern Carbon cycle model, as shown in figure XX4, where the accumulated radiative forcing using the Bern Carbon cycle model is resulting in higher results for almost all time horizons. This topic in relation to biofuels is further elaborated in an article by O’Hare et.al. The main difference is that by using an indicator of cumulative radiative forcing, using GHG models instead of summing GWP, one captures the decaying concentration of GHG, and the relative increased importance of early emissions, as a fraction never ceases to increase warming (21,7% according to the Bern Carbon cycle model). A "break even" for a policy in GHG terms is thus only achieved when accumulated warming is equal to the baseline, and not just when the sum of GWP is zero (O'Hare). However: this discrepancy depends on the relationship between the size of the initial carbon emissions from land use change, and the yearly savings obtained by the biofuels.

Page 214: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

214

IPCC acknowledges the shortcomings of the GWP method and chooses deliberately to simplify the calculation of GWP by not applying any discounting879, as discussed above. However; any discounting rate would somehow be arbitrary and difficult to defend. Further; some commentators question whether one should discount physical numbers, as such value-laden calculations should not be part of the assessment of the physical properties.880 If a unit of GHG has decaying but eternal consequences the damage function is dependent of both a discount rate (that might be zero) and the timeframe chosen. This is also critical for land use change in the context of biofuels, as will be discussed in the next section. Different alternatives has been proposed to the GWP881, where some also tries to make the valuation of time a more explicit choice for policy-makers, where weight has to be given to different generations882 Using GWP with a distinct timeframe is not problematic when one is accounting emissions from projects where one GHG is dominant and emissions coincide in time with production and use. However; when different GHGs are emitted that have considerable different warming profiles (as seen above for e.g. methane), and/or emissions do not coincide with production and use; biases occur. Long GWP timeframe puts relatively more weight on long-lived GHG emissions versus short GWP timeframe that does the opposite. Consequences of emissions not coinciding in time with production and use is discussed below, but in short, the use of GWP with a fixed timeframe distorts the presentation of time-dynamics, as the summing of GWP values in fact is the summing of displaced and sometimes overlapping "accumulated warming over time" integrals. The ultimate solution would be to compare additional accumulated warming of climate policy options or projects, based on the modelled decaying concentration of different GHG, using e.g. the revised Bern Carbon cycle model for carbon, and other appropriate models for methane and N2O. Those values would then also necessary be a result of a choice of timeframe and discounting rate like for the GWP values, but with the advantage of sharing a common set of assumptions that are not displaced in time. Eventual non-zero discounting would then be a more attractive option, as the basis would be the physical concentration of GHG, and not discounting based on GWP that themselves are using 0% discounting during the timeframe and 100% discounting after the timeframe (O'Hare)883. It seems like there is an agreement in the literature884 that physical measures should not be discounted, only monetary units, or other units measuring human perception of damage.

Page 215: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

215

12.3.3. Relevance for biofuels The discussion in the section above demonstrate how valuation of time is already an issue when making the choice of a GWP timeframe. Such issues are relevant for all climate change policy making. Now the discussion turns to more specific policies where GHG emissions and avoided GHG emissions are dispersed in time, which is the case for e.g. a field of biofuel crops, where an initial land use change took place. Each harvest will displace fossil fuel, but at later stages than the emissions occurring initially from indirect land use change. Again the question arises; how to account for emissions occurring at different times? Should a discounting be applied and what should be the overall timeframe. The choices made have considerable impact on the results. The figure below shows the emissions per MJ of fuel, assuming that the production of biofuels leads to a certain amount of land use change, both indirectly and directly. The initial loss is around 20 ton of CO2 per ha, assuming a mixture of land use types being converted to arable land. The calculations follow the methodology laid down in the Renewable Energy Directive, Annex V, with rapeseed used as crop. Using energy allocation for co-products, the following values are obtained (for now the land use change emissions are assumed to be once-off, although in reality they follow a kind of exponential decay rate between the two land uses).

0

5

10

15

20

25

30

35

40

45

10 20 30 40 50

Time [years]

land

use

cha

nge

emis

sion

s [g

CO

2/M

J]

Land use change emissionsover different time horizons

Figure 7: Initial emissions from land use change divided on varying periods of cultivation Here, the issue shown in relation to time is the time horizon i.e. how many years are the initial emissions divided by. The emissions from land use change are assumed to happen at once (although the follow some kind of exponential decay. If one chooses a horizon of 20 years it is however still questionable whether 20 g/MJ (taken from the figure) is applicable to the consignment of fuel being produced in 20 years from now, as that fossil fuel displacement occurs 20 years later than the land use change emissions. If those where to be discounted, a range of results are obtainable, as demonstrated in the figure below. The timeframe is variable, and the crop in question is rapeseed. With a 2 % discount rate one assumes that the reduced damage from using biofuel today is 38 g CO2/MJ, but 25 g CO2/MJ in 20 years time.

Page 216: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

216

0

5

10

15

20

25

30

35

40

1 6 11 16

Time [years]

Dire

ct G

HG

sav

ings

[g C

O2/M

J]

0% 2% 4%

Figure 8: Impact of using biofuel diverting fossil fuel with varying discount rate Higher discount rates values savings occurring later less, as seen in figure XX6. As discussed earlier; the choice of discount rate relates to the choice of timeframe. A shorter timeframe and lower (zero) discount rate gives the same result as higher discount rate and longer timeframe. Using the rapeseed example again; a 100 years timeframe with 2 % discounting rate gives the same result in terms of carbon emissions as 43 years with 0 % discounting. Related to this is also the assumption made for what the land is used for after the cultivation period. If the land is left abandoned it might return to its former steady state and sequester the carbon lost during the initial land use change. As discussed above there is no direct discounting involved in the calculation of GWP, and thus discounting GWP values is not scientific correct way of attributing value to time. If one chooses to use GWP as metric it seems reasonable to avoid further complications of that metric, as it is itself a considerably simplified proxy. The Renewable Energy Directive uses no discounting (or a 0 % rate) and 20 divide the emissions on years, regardless of how many years the land is put to cultivation. Further is no credit offered for the carbon sequestration occurring on that land after those 20 initial years. The next section compares this with other relevant studies.

12.3.4. Choices made in other studies As discussed earlier the main issues related to accounting for time is:

1. A unit of GHG has varying impact over its lifetime; how to account for future warming occurring from a unit released to the atmosphere today? (standard approach used in most part of literature is the 100-years, -no discounting, -GWP values from IPCC)

2. Related to the first is the question of how to account for units of GHG emissions and sequestration occurring at different time, but being related to the same policy or project?

3. Extending the period of cultivation (or generally speaking the divisor) decreases the yearly land use change emissions attributed to each consignment of biofuel; what is a reasonable divisor? (number of years to divide land use change emissions by)

4. Assuming that the land returns to it's original state after the cultivation period may influence the overall results, but depends on choices made under 2 and 3; a 0 % discounting of emissions will lead no net land use change overall, while using a non-

Page 217: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

217

zero discounting depends on the discounting rate together with the length of the cultivation period. Longer cultivation periods (larger divisor) together with higher discounting rates diminishes the contribution of reversion of land use change.

The table below summarizes what other studies have assumed with regard to these crucial questions. Study/ legislation/ standard

Radiative forcing over lifetime of GHG

Discounting rate

Land use change is divided on how many years?

Reversion of land use change after cultivation

Renewable Energy Directive

IPCC GWP 100 yr

0 % 20 years No

CARB IPCC GWP 100 yr

0 % and 2 %

30 and 100 years No

EPA IPCC GWP 100 yr

0 % and 2 %

30 and 100 years No

Searchinger et.al. IPCC GWP 100 yr

0 % 30 years No

Fritsche et.al. (risk adder app.)

IPCC GWP 100 yr

0 % 20 years No

Delucchi M.A.885 LEM CO2-equivalanecy factors

Not 0 % 30 years Yes

British Standards886 (Defra)

IPCC GWP 100 yr

0 % 20 years No

O'Hare et.al. Fuel warming intensity

0 %, 3 %, 7 %

25 years, but variable analytic horizon

Yes and no

12.3.5. Commentary All the four categories of treatment of time has been discussed and analysed. The use of global warming potentials (GWP) for different GHG has become standard, where the GHG has equal warming over the chosen timeframe, normally 100 years. This choice is normally not discussed but has direct consequences for the planning, as well as the functioning of policies. The choice has at least two main implications: how to weight different GHGs against each other, and how to account for warming over time. One example of consequences of the first issue is the carbon market, where the price of carbon credits and thus which projects that are profitable depends on the timeframe chosen for GWP. With a shorter timeframe and /or a discounting rate more projects directed towards shorter-lived gases would have been executed, as e.g. more focus on methane emissions (which have a GWP of 72 in a 20-years perspective – instead of 23 under Kyoto). The possible discounting of those GWP has been suggested, and is done by some, e.g. CARB and EPA. Consequences of discounting are demonstrated in figure XX6. O’Hare et.al. however argues that such discounting is only further compounding the error of using the GWP, as GWPs apply no discounting within their defined timeframe and a 100% discounting after the end of the timeframe887. The GWP value is thus already a considerably simplified

Page 218: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

218

proxy. It thus seems arguable that discounting the GWP only sophisticates something that is already simplified. Scientific certainty can not be an expected result of such a process. The importance of the divisor (number of years that the land use change emissions are divided by) is shown in figure XX5, and the relationship is simple: using 30 years instead of 20 lowers the value by one third. Facing such a complex matter as accounting for time, where difficult decisions has to be made, it might be more suitable to choose a transparent, but simple methodology. Increased sophistication by discounting with non-zero rates of GWP values is not increasing the scientific quality, as the GWP themselves are proxies treating timing in its own way. The question should be addressed at the appropriate basic level, namely the determination of added warming, where the first assumptions for accounting of time have to be made. Finally it is a question of whether the added value of introducing more correct metrics instead of GWP outweighs the burden of introducing another metric that has to be explained to policy makers and is not part of the IPCC work so far. It seems that such a change in metric should be initiated through the appropriate institutions like the IPCC. With the known main parameters, as listed above, one can assess to what extent choices are biased. Assuming that one uses the standard 100 year timeframe GWP: The combination of long timeframe and no discounting, together with reversion of land use change after the cropping period is one extreme where timing barely matters; short timeframe and discounting (with a relatively high discount rate), together with no reversion of land use change is at the other end, where sequestration taking place in some years is valued at only a fraction of sequestration today. Comparing the choices made under the Renewable Energy Directive it seems that the treatment is among the more conservative. Comparing with CARB and EPA, and using the same assumptions888; the figure below is obtained, showing the emission savings (using 100 yr GWP) over a period of 100 years.

Page 219: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

219

0%

5%

10%

15%

20%

25%

30%

35%

40%

45%

0 10 20 30 40 50 60 70 80 90

Year

Savi

ngs

com

pare

d to

foss

il fu

el

Savings according to EPA (2% 100 yr)

Savings according to EPA (0% 30 yr)

Savings according to Renewable EnergyDirective (0% 20 yr)

`

To obtain one figure for the EPA discounting alternative, one can sum the accumulated savings over the whole timeframe of 100 years Savings Rapeseed with land use change-factor under Renewable Energy Directive methodology 21% Rapeseed with land use change-factor under EPA methodology (0% 30 yr) 29% Rapeseed with land use change-factor under EPA methodology (2 % 100 yr) 15%

12.3.6. Conclusion It is suggested to continue to use the GWP, although the imperfections are more considerable in the context of policies where emissions are substantially displaced from production and use, such as for biofuels. A change in metric should be done through e.g. IPCC to avoid confusion and parallel competing metrics in use. By using the GWP proxy, one should avoid further sophistication, as complexity based on a simplified proxy seldom provides added value neither in scientific terms nor for understanding purposes. Simple techniques as simply dividing the initial land use change emissions by a number of years is more helpful and intuitive to understand and compare than introducing complexity through e.g. discounting. If one wishes to improve the representation of time, a more helpful alternative would be to use GHG models to evaluate the accumulated added warming as suggested by O'Hare et.al.

Page 220: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

220

Acronyms and abbreviations AEZ agro-ecological zone bn billion BTL biomass to liquid [fuel made using the Fischer-Tropsch process] CARB California Air Resources Board IFPRI International Food Policy Institute CES constant elasticity of substitution CGE computable general equilibrium [model] CIS Confederation of Independent States889 DDGS Distillers Dried Grains and Solubles EEA European Environmental Agency EPA Environmental Protection Agency [of the United States] EU European Union EU15 European Union of 15 Member States (1995-2004) EU25 European Union of 25 Member States (2004-2007) FAO United Nations Food and Agriculture Organisation FAPRI Food and Agricultural Policy Research Institute FSU Former Soviet Union GAEZ global agro-ecological zone gal gallon, gallons GTAP Global Trade Analysis Project IEA International Energy Agency IFPRI International Food Policy Research Institute IIASA International Institute for Applied Systems Analysis IPCC Intergovernmental Panel on Climate Change IPTS Institute for Prospective Technological Studies [of the JRC] ISO International Organization for Standardization ha hectare JEC JEC consortium [members: JRC, EUCAR and CONCAWE] JRC Joint Research Centre [of the European Commission] Mha Million hectares Mt Million tons Mtoe Million tons of oil equivalent NA & ME North Africa and Middle East OECD Organisation for Economic Cooperation and Development OFID OPEC Fund for International Development OPEC Organisation of Petroleum Exporting Countries p.a. per annum PE partial equilibrium [model] PJ petajoule [1015 joules] RFA Renewable Fuels Agency [of the UK] t ton tC tons of carbon toe tons of oil equivalent; pedal digit UNEP United Nations Environment Programme UNICA Brazilian Sugarcane Industry Association WEO World Energy Outlook [of the IEA]

Page 221: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

221

References [ADAS UK Ltd (2008)] ADAS UK Ltd, “Anticipated and potential improvements in land productivity and increased agricultural inputs with intensification”, study commissioned by AEA Technology as part of the Gallagher Biofuels Review for Renewable Fuels Agency Department for Transport, 2008 [ADAS UK Ltd (2008a)] ADAS UK Ltd, “Critique of Searchinger (2008) and related papers assessing indirect effects of biofuels on land-use change”, study commissioned by AEA Technology as part of the Gallagher Biofuels Review for Renewable Fuels Agency Department for Transport, 2008 Air Improvement Resource, “Comments on ARB October 16 Land Use Emission Estimates for Corn Ethanol”, Air Improvement Resource for Renewable Fuels Association, California Air Resources Board Workshop 2 December 2008 Al-Riffai, P., B. Dimaranan and D. Laborde, “Biofuels II: Global Trade and Environmental Impact Study”, preliminary draft report, November 2009 Al-Riffai, P., B. Dimaranan and D. Laborde, “Global Trade and Environmental Impact Study of the EU Biofuels Mandate”, final report, March 2010. Amigos de Terra, Amazonia Brazileira, "The Cattle Realm", 2008 ALTERRA, Land use modelling – implementation: Preserving and enhancing the environmental benefits of “land-use services”, study commissioned by European Commission – Background document for the 26 June workshop, Version 21 June 2009 Babcock, B.A., "Measuring Unmeasurable Land-Use Changes from Biofuels", Iowa AG Review 15, 2009 Banerjee, O. and J. R. Alavalapati, "Modeling Forest Sector Illegality in a Dynamic Computable General Equilibrium Framework: The Case of Forest Concessions in Brazil", paper prepared for the 12th annual conference on global economic analysis, Santiago, Chile, June 10-12, 2009 Banse, M., “Impact of European Union Biofuel Policies”, powerpoint presentation to expert meeting on bioenergy policy, markets and trade and food security, FAO, Rome, 18-20 February 2008 Banse, M., “Dr. Banse Response to Charge Questions”, in “Lifecycle Greenhouse Gas Emissions due to Increased Biofuel Production – Model Linkage Peer Review Report”, prepared by ICF International for EPA, 2009 Banse, M., A. Tabeau, G. Woltjer, G. and H. van Meijl, “Impact of European Union Biofuel Policies on World Agricultural and Food Markets”, paper submitted for the GTAP Conference 2007

Page 222: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

222

Banse, M., H. van Meijl, A. Tabeau and G. Woltjer, “Will EU biofuel policies affect global agricultural markets?”, European Review of Agricultural Economics Vol 35 (2), 2008 Beckman, J. and T. Hertel, “Why Previous Estimates of the Cost of Climate Mitigation Might Be Too Low”, n.d. Berndes, G. and J. Hansson, “Bioenergy expansion in the European Union: Cost-effective climate change mitigation, employment creation and reduced dependency on imported fuels”, Energy Policy 35, 5965-5979, 2007 Bergerson, J., Keith, D. (2006) LCA of Oil Sands Technologies, Institute for Sustainable Energy, Environment and Economy (ISEEE) Biomass Research and Development Initiative, “Increasing Feedstock Production for Biofuels – Economic Drivers, Environmental Implications, and the Role of Research”, n.d. Blanco-Fonseca, M. and I. Peréz Domínguez, “Modelling Biofuels with CAPRI: Baseline and Marginal Scenarios”, powerpoint presentation to Inter Service Group on indirect land use change, Brussels, 23 July 2009890 Blanco-Fonseca, M., A. Burrell, H. Gay, M. Henseler, A. Kavallari, R. M'Barek, I. Pérez Domínguez, A. Tonini, "Impacts of the EU biofuel target on agricultural markets and land use: a comparative modelling assessment", 2010 Bouët, A., L. Curran, B. Dimaranan, M-P. Ramos and H. Valin, “Biofuels: Global Trade and Environmental Impact Study, Final Report – 29 April 2009”, report submitted to the European Commission, 2009 Brandt, A.E., Farrell, A.E. (2007) Scraping the bottom of the barrel: greenhouse gas emission consequences of a transition to low-quality and synthetic petroleum resources, Climatic Change 84, pp. 241-261 British Standards, "Specification for the assessment of the life cycle greenhouse gas emissions of goods and services (PAS 2050)", 2008 Britz, W., “Linking regional economic models to land use change modelling framework – benefits and challenges”, powerpoint presentation to workshop on “A land-use modelling framework for the European Commission”, 26 June 2009, Brussels CARB, "Staff report: initial statement of reasons proposed regulation to implement the low carbon fuel standard", 2009 Casson A., "Oil Palm, Soybeans and critical habitat loss", review prepared for WWF forest conversion initiative, 2003 [CE Delft (2008)] CE Delft, “Estimating indirect land use impacts from by-products utilization”, study commissioned by AEA Technology as part of the Gallagher Biofuels Review for Renewable Fuels Agency Department for Transport, 2008

Page 223: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

223

[CE Delft (2008a)] CE Delft, "Agricultural land availability and demand in 2020", part of the study commissioned from AEA Technology as part of the Gallagher Biofuels Review for Renewable Fuels Agency Department for Transport, 2008 IFPRI, land supply curves used in modelling exercise, personal communication, December 2009 CIFOR, "A global analysis of tropical deforestation due to bioenergy development", Bioenergy, sustainability and trade-offs : can we avoid deforestation while promoting bioenergy ?, project founded by the European Commission, Contribution agreement No EuropAid / DCI-ENV/2008/143936/TPS, 2009 [CIFOR Oil Palm study] Congo Basin Forest Partnership (2005) Couwenberg, J. (University of Greifswald) and M. Silvius (Wetland International), "Impact study EU biofuels Mandate", letter to IFPRI, 17 June 2010 de Sherbinin, A, “A Guide to Land-Use and Land-Cover Change (LUCC)” 2002, downloaded from http://sedac.ciesin.columbia.edu/tg/ Decreux, Y. and H. Valin, “MIRAGE, Updated Version of the Model for Trade Policy Analysis: Focus on Agriculture and Dynamics”, 2007 Delucchi, M. A.,"Lifecycle Analyses of Biofuels – Draft Report", 2006 Dumortier, J., D. Hayes, M. Carriquiry, F. Dong, X. Du, A. Elobeid, J. Fabiosa and S. Tokgaz, “Sensitivity of Carbon Emission Estimates from Indirect Land-Use Change”, Working Paper 09-WP 493, Center for Agricultural and Rural Development, Iowa State University, 2009 Econometrica, “A Practical Approach for Policies to Address GHG Emissions from Indirect Land Use Change Associated with Biofuels”, Technical Paper TP-080212-A, 2009 [ECOFYS (2009)] ECOFYS, “A review of demand-yield relationships”, draft dated 31 August 2009 [ECOFYS (2009a)] ECOFYS, “A review of yield in new agricultural areas”, draft dated 7 September 2009 ECORYS, “Study on the evolution of some deforestation drivers and their potential impacts on the costs of an avoiding deforestation scheme”, Draft interim report to the European Commission, 31 March 2009 Edwards, R., D. Mulligan and L. Marelli. Indirect Land Use Change from increased biofuels demand. Comparison of models and results for marginal biofuels production from different feedstocks. Final report of the contract n. 070307/2008/517067/C3. European Commission JRC-IE, Ispra, 2010.

Page 224: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

224

Ensus, “Indirect Effects of Biofuels – Study by the Renewable Fuels Agency – Evidence provided by Ensus Limited”, 2008 EPA, "Regulation of Fuels and Fuel Additives: Changes to Renewable Fuel Standard Program – Notice of proposed rulemaking", 2009. EPA, “Regulation of Fuels and Fuel Additives: Changes to Renewable Fuel Standard Program; Final Rule”, Federal Register / Vol. 75, No. 58 / Friday, March 26, 2010. See http://www.epa.gov/oms/renewablefuels/420f09024.htm and Proposed rules on Fuels and Fuel Additives: Changes to Renewable Fuel Standard Program, May 2009 (http://www.epa.gov/otaq/renewablefuels/rfs2_1-5.pdf) European Commission, “Terms of reference – The indirect land use change impact of biofuels”, 29 June 2009 European Commission, “Review of economic and environmental data for the biofuels progress report”, Commission Staff Working Document, SEC (2006) 1721, 10 January 2007 Faaij, A.P.C., “RE: Iluc”, personal communication, 8 June 2009 FAO, "Global Resources Assessment", 1980 FAO, "The State of Food and Agriculture", 2008 Fearnside M. P., "Time preferences in global warming calculations: a proposal for a unified index", Ecological Economics 41, 2002 Fearnside P, Deforestation in Amazonia, "Encyclopaedia of Earth, Eds Cultler J. Cleveland, Washington, DC Environmental Information Coalition, National Council for Science and the Environment. Foley et al (2007) citing Nepstad, D. Et al (1999) "Large scale impoverishment of Amazonia forests by logging and fire", Nature Issue 398, April 8, p 505-508 and Anser, G et al (2005) "Selective Logging in the Brazilian Amazon", Science Issue 310, p. 480-82 Forests Monitor, "The Timber Sector in the DRC : A Brief Overview", 2007 Forster, P., V. Ramaswamy, P. Artaxo, T. Berntsen, R. Betts, D.W. Fahey, J. Haywood, J. Lean, D.C. Lowe, G. Myhre, J. Nganga, R. Prinn, G. Raga, M. Schulz and R. Van Dorland, "Changes in Atmospheric Constituents and in Radiative Forcing", in Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M.Tignor and H.L. Miller (eds.), "Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change", 2007 Fritsche, U., "GHG Accounting for Biofuels: Considering CO2 from Leakage", 2007 Fritsche et.al. Draft document, Personal communication

Page 225: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

225

Gay, S. and A. Kavallari, “Impact of European Union renewable energies Directive on agricultural markets – simulations with AGLINK”, powerpoint presentation to Inter Service Group on indirect land use change, Brussels, 23 July 2009891 Geist, H.J and E.F. Lambin, "Proximate Causes and Underlying Driving Forces of Tropical Deforestation", BioScience 52, 2002 Geist, H.J, E. F. Lambin and E. Lepers , "Dynamics of Land-Use and Land-Cover Change in Tropical Regions", Annu. Rev. Environ. Resour. 28, 2003 Geist, H.J and E.F. Lambin, "What drives tropical deforestation?", 2001 Grau, H.R. and M. Aide, "Globalization and Land-Use Transitions in Latin America", Ecology Society 13, 2008 Greenpeace, "How the palm oil industry is cooling the climate", 8th November 2007 Greenpeace, "Partners in Crime : A Greenpeace investigation of the links between UK and Indonesia's timber barons", 2003 GLOBIOM, powerpoint presentation to indirect land use change meeting, European Commission, Brussels 30 June 2009 Groves, M.C.E. and Maurer, R., "N2O Abatement in an EU Nitric Acid Plant – A case study, The International fertiliser Society - Proceedings 539, 2004 Havlík, P., M. Obersteiner, E. Schmid and U. Schneider, “Indirect land use change from biofuels: modelling with XXXX Hertel, T.W., “Strengths and limitations of the GTAP Modeling Framework”, powerpoint presentation, n.d. Hertel, T.W., A. Golub, A.D. Jones, M. O’Hare, R.J. Plevin and D.M. Kammen, “Effects of US Maize Ethanol on Global Land Use and Greenhouse Gas Emissions: Estimating Market-mediated Responses”, BioScience, Vol. 60 No. 3, March 2010, 223-231 and supporting online material. Hertel, T.W., S. Rose and R. S. J. Tol, "Land Use in Computable General Equilibrium Models: An Overview", GTAP Working Paper no. 39, 2008 Hertel, T.W., W.E. Tyner and D.K. Birur, “Biofuels for all? Understanding the Global Impacts of Multinational Mandates”, Center for Global Trade Analysis Department of Agricultural Economics, Purdue University, USA, 2008 Houghton, R.A., “Dr. Houghton Response to Charge Questions”, in “Emissions from Land Use Change due to Increased Biofuel Production – Satellite Imagery and Emissions Factor Analysis Peer Review Report”, prepared by ICF International for EPA, 2009 Hyungtae K, Seungdo, Dale B, Biofuels, Land Use and Greenhouse Gas Emissions: Some Unexplored Variable, Environmental Science and Technology, January 2009, 961-967

Page 226: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

226

[ICF International (2009a)]: ICF International, “Introduction”, in “Emission from Land Use Change due to Increased Biofuel Production – Satellite Imagery and Emissions Factor Analysis Peer Review Report”, prepared by ICF International for EPA, 2009 [ICF International (2009b)]: ICF International, “Peer Reviewer Responses to Charge Questions”, in “Emission from Land Use Change due to Increased Biofuel Production – Satellite Imagery and Emissions Factor Analysis Peer Review Report”, prepared by ICF International for EPA, 2009 IIASA, “Biofuels and Food Security”, OFID study prepared by IIASA, 2009 Indonesia-UK tropical forestry management programme (1999), Illegal Logging in Indonesia. ITFPM Report no. EC/99/03 International Institute for Applied Systems Analysis [IIASA], “Biofuels and food security – Implications of an accelerated biofuels production”, Summary of the OFID [OPEC Fund for International Development] study prepared by IIASA, 2009 IPCC, 2000 - Nebojsa Nakicenovic and Rob Swart (Eds.) Emission Scenarios Cambridge University Press; http://www.ipcc.ch/ipccreports/sres/emission/index.php?idp=0 IPTS, "Biofuel Modelling (AGLINK, ESIM, CAPRI), draft dated 30 October 2009.892 IPTS, “Impacts of the EU Biofuel Target on Agricultural Markets and Land Use – A Comparative Modelling Assessment”, final draft report, May 2010.893 ISO, "ISO 14044 - Environmental Management-Life Cycle Assessment – Requirements and Guidelines", 2006 Jarvis and Jacobsen (2006) "Working paper-Incentives to promote forest certification in Indonesia" Project: motivating sustainability, International Finance Corporation. Johnston, M. (2009) Agricultural Biofuel Yields: Spatial Variation and Productivity Gaps, Power Point Presentation held on July 22, 2009 Kampman, B., C. Leguijt, D. Bennink, L. Wielders, X. Rijkee, A. de Buck and W. Braat, "Green Power for Electric Cars – Development of policy recommendations to harvest the potential of electric vehicles", report by CE Delft for Friends of the Earth Europe, Transport and Environment and Greenpeace, 2010 Kathryn R. Kirby, William F. Laurance, Ana K. Albernae, Götz Schroth, Philip M. Fearnside, Scott Bergen, Eduardo M. Venticinque, Carlos da Costa, "The future of deforestation in the Brazilian Amazon", in Futures 38 (2006), 432-453 Keeney, R. and T. W Hertel, “The Indirect Land Use Impacts of U.S. Biofuel Policies: The Importance of Acreage, Yield, and Bilateral Trade Responses”, GTAP working paper No. 52, 2008

Page 227: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

227

[Kim et al. (2008)]: Kim, H., S. Kim and B. Dale, “Biofuels, Land Use Change and Greenhouse Gas Emissions: Some Unexplored Variables”, accepted for publication in Environmental Science and Technology, 2008 [Kim et al. (2008a)]: Kim, H., S. Kim and B. Dale, supporting information for “Biofuels, Land Use Change and Greenhouse Gas Emissions: Some Unexplored Variables”, 2008 Kline K., V. H. Dale, R. Lee and P. Leiby, "In Defense of Biofuels, Done Right", Issues in Science and Technology, Volume 25, issue 3, 2009 [Kløverpris et al. (2008a)]: Kløverpris, J., H. Wenzel, M. Banse, Ll. Milà í Canals and A. Reenberg, "Conference and Workshop on Modelling Global Land Use Implications in the Environmental Assessment of Biofuels", International Journey of Life Cycle Analysis 13(3), 2008 [Kløverpris et al. (2008)]: Kløverpris, J., H. Wenzel and P. Nielsen, “Life Cycle Inventory Modelling of Land Use Induced by Crop Consumption – Part 1: Conceptual Analysis and Methodological Proposal”, International Journal of Life Cycle Analysis 13(1), 2008 [Kløverpris et al. (2008b)] Kløverpris, J., K. Baltzer and P. Nielsen, “Life Cycle Inventory Modelling of Land Use Induced by Crop Consumption – Part 2: Example of wheat consumption in Brazil, China, Denmark and the USA”, International Journey of Life Cycle Analysis, in press Kløverpris J., “Biofuels and Indirect Land Use Change (land use change)”, powerpoint presentation on behalf of Novozymes, 2009 Koeble, R., A. Leip, V. Blujdea, R. Hiederer and F. Carré, “Spatial Estimations of Land Carbon Stock Changes and N2O Emission from Cultivation of Crops for Biofuel Production”, powerpoint presentation to workshop on marginal yields and land allocation in land use change emissions estimates, Brussels 22 July 2009 Koehler, 2009 […] Lal. R, “The Potential for Soil Carbon Sequestration”, in IFPRI, “Agriculture and Climate Change: An Agenda for Negotiation in Copenhagen”, 2009 Lambin, E., "Planetary boundaries & sustainability transition", powerpoint presentation, n.d. Lambin, E. F., M.D.A. Rounsevell and H. J. Geist, "Are agricultural land-use models able to predict changes in land-use intensity?", Agriculture, Ecosystems and Environment 82, 2000 Lambin, E.F., B.L. Turner, H.J. Geist, S.B. Agbola, A. Angelsen, J.W. Bruce, O.T. Coomes, R. Dirzo, G. Fischer, C. Folke, P.S. George, K. Homewood, J. Imbernon, R. Leemans, X. Li, E.F. Moran, M. Mortimore, P.S. Ramakrishnan, J.F. Richards, H. Skånes, W. Steffen, G.D. Stone, U. Svedin, T.A. Veldkamp, C. Vogel and J. Xu, "The causes of land-use and land-cover change: moving beyond the myths", Global Environmental Change 11, 2001 Legoupil Jean-Claude, Ruf François, (2009), "Farmers's strategy and land use change in the perspective of biofuels development in West Africa, Natural resources forum, : 173

Page 228: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

228

Lobet I, "Living on Earth's", Interview 2nd of January 2009, Why tropical Forest fall Londo, M. “Impacts of sustainability criteria on biomass availability and costs”, study report by ECN Policy Studies for European Commission, draft of 3 August 2009. [Lywood (2009)]: Lywood, W., “Issues of concern with models for calculating GHG emissions from indirect land use change”, draft, August 2009 [Lywood, (2009a)]: Lywood, W., “Global Agricultural Yield Changes”, 2009 [Lywood (2009b)]: Lywood, W., “Inventory and Comparison of Assessment Methodologies for determining GHG Emissions from Indirect Land Use Change”, draft, 6 May 2009. [Lywood (2009c)]: Lywood, W., “Natural vegetation changes resulting from changes in land requirements for increased biofuels production”, draft, 8 June 2009. [Lywood (2009d)]: Lywood, W., “Modelling of GHG emissions from Indirect Land Use Change from increased EU demand for Biofuel”, draft 2, June 2009. [Lywood (2009e)]: Lywood, W., “RE: Ensus approach for land use change Modelling”, personal communication, 29 June 2009 Lywood, W., J. Pinkney and S. Cockerill, “Impact of protein concentrate co-products on net land requirement for European biofuel production”, 2009 Manta Nolasco, M.I.(2007) : "Evaluacion de las causas naturales y socioéconomicas de los incendios forestales en America del Sur. Facultad de Ciencias Forestales, Universidad Nacional Agraria, Lima, Peru. Mathews, J. and H. Tan, “Biofuels and indirect land use change effects: the debate continues”, 2009 Meizlish, M., D. Spethmann and M. Barbara, "Carbon Finance for Reduced Emissions from Deforestation and Degradation at the Forest Frontier – Financial Analysis of Alternate Land Uses in the Amazon, Congo and Papua New Guinea", New Forests, 2007 Millenium Economic Assessment, "Ecosystems and Human well-being", http://www.millenniumassessment.org/en/index.aspx, 2005 Mohr, E. (2010) Long term prediction of unconventional oil production, Energy policy 38, pp. 265-276 Mortimer, N., A. Ashley, A. Evans, A. Hunter and V. Shaw, “Support for the [Gallagher] Review of the Indirect Effects of Biofuels”, 2008 Morton D.C., R.S. Defries, Y.E. Shimabukuro, L.O. Anderson, E. Arai, F.d.B. Espirito-Santo, R. Freitas, J. Morisette, "Mapping land use of tropical regions from space", Proc Natl Acad Sci USA 103, 2006

Page 229: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

229

Nassar, A., L. Harfuch, M. Moreira, L. Bachion and L. Antoniazzi, “Impacts on Land Use and GHG Emissions from a Shock on Brazilian Sugar cane Ethanol Exports to the United States using the Brazilian Land Use Model (BLUM)”, 2009 Nassar, A. M., “Land Use and Land Allocation in Brazilian Agriculture: BLUM (Brazilian Land Use Model)”, powerpoint presentation to workshop on marginal yields and land allocation in land use change emissions estimates, Brussels, 22 July 2009 New Fuels Alliance, letter to Ms M. Nichols, Chairman, California Air Resources Board, 23 October 2008 O’Hare, M, R. Plevin, J. Martin, A. Jones, A. Kendall and E. Hopson, “Proper accounting for time increases crop-based biofuels’ greenhouse gas deficit versus petroleum”, Environmental Research Letters, 2009 Özdemir, E., M. Härdtlein and L. Eltrop, “Land substitution effects of biofuel side products and implications on the land area requirement for EU 2020 biofuel targets”, Energy Policy 37, 2009 Papua Ministry of Forestry, “List of Forest Concession Right (Hak Pengusahaan Hutan) and Timber Forest Product Utilisation Permit (Izin Usaha Pemanfaatan Hsil Hutan Kayu) holders”, 2006 Persson et.al. (2007) Major oil exporters may profit from rather than loose, in a carbon-constrained world, Energy Policy 35, pp. 6346-6353. Piermartini, R. and R. The, “Demystifying Modelling Methods for Trade Policy”, WTO Discussion Paper no. 10, 2005 Ramos, F., O. Gomez and J.-M. Terres, “Spatial allocation of extra areas resulting from land use change modeling", powerpoint presentation to workshop on marginal yields and land allocation in land use change emissions estimates, Brussels, 22 July 2009 REFUEL, “Potentials and costs of biofuel feedstocks – Key outcomes of the REFUEL feedstock assessment”, n.d. [RFA (2008)] Renewable Fuels Agency, “The Gallagher Review of the indirect effects of biofuels production”, July 2008 RFA (2009) What do Biofuels displace and why does it matter?. http://www.ethanolrfa.org/objects/documents/2489/what_do_biofuels_replace.pdf Sands, R., M. Brady and M.-K. Kim, “Survey of Land Representation in Economic and Biophysical Models”, n.d. Santakos G., T. Corbière and C. Opal, "Substitution versus allocation-definitions", Roundtable on sustainable biofuels, 2008 Sanz Labrador, I., "Biofuels: A Decisive Instrument for Sustainable Development", transl. E. Schmid Cartes, 2009

Page 230: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

230

Saunders, C., W. Kaye-Blake, L. Marshall and S. Greenhalgh, “Impacts of a United States’ biofuel policy on New Zealand’s agricultural sector”, Energy Policy 37, 2009 Searchinger, T., "GREET and UK Default Values CO2 Emissions for various fuels, grams CO2 Eq. per Megajoule of energy expanded", 2008 [Searchinger (2009)]: Searchinger, T., “Evaluating biofuels – the consequences of using land to make fuel”, Brussels Forum paper series, 2009 [Searchinger (2009a)]: Searchinger, T., “Mr. Searchinger Response to Charge Questions”, in “Lifecycle Greenhouse Gas Emissions due to Increased Biofuel Production – Model Linkage Peer Review Report”, prepared by ICF International for EPA, 2009 Searchinger, T., R. Heimlich, R. Houghton, F. Dong, A. Elobeid, J. Fabiosa, S. Tokgoz, D. Hayes and T.-H. Yu, “Supporting Online Material for Use of U.S. Croplands for Biofuels Increases Greenhouse Gases Through Emissions from Land Use Change”, Science Express, 2008 Searchinger, T. D. and R. Heimlich, “How much can demand-increased increases in yields replace crops or cropland diverted to ethanol” (Discussion Draft, 13th March 2008) Sheil D, Casson A, Maijaard E, Van Noordwijk M, Gaskell J, Sunderland Groves J, Werts K, Kanninen M, "The impact and opportunities of oil palm in Southern Asia, CIFOR, June 2009 Smeets, E., “Review of GREEN-X assumptions on biomass availability and costs”, study report by ECN Policy Studies for European Commission, draft of 3 August 2009 Smukler, S. and C. Palm, “Monitoring, Reporting and Verification Methodologies for Agriculture, Forestry and Other Land Use”, in “Agriculture and Climate Change: An Agenda for Negotiation In Copenhagen”, International Food Policy Research Institute, Focus 16, May 2009 Stern 2007 STERN REVIEW: The Economics of Climate Change, Chapter 2: Economics, ethics and climate change, available at: http://www.hm-treasury.gov.uk/stern_review_report.htm Swallow, B. M. and M. van Noordwijk, “Direct and Indirect Mitigation Through Tree and Soil Management”, in “Agriculture and Climate Change: An Agenda for Negotiation In Copenhagen”, International Food Policy Research Institute, Focus 16, May 2009 Tabeau, A., B. Eickhout and H. van Meijl, "Endogenous agricultural land supply: estimation and implementation in the GTAP model", n.d. Taheripour, F., T. Hertel, W. Tyner, J. Beckman and D. Birur, “Biofuels and their By-products: Global Economic and Environmental Implications”, Selected paper prepared for presentation at the American Agricultural Economics Association Annual Meeting, Orlando, FL, 27-29 July 2008

Page 231: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

231

Thirtle, C., L. Lin and J. Piesse, “The Impact of Research-Led Agricultural Productivity Growth on Poverty Reduction in Africa, Asia and Latin America”, World Development 31, 2003 Thompson et.al (2010) The US biodiesel use mandate and biodiesel feedstock markets Biomass and Bioenergy 34, pp 883-889. Tipper, R., C. Hutchison and M. Brander, “A Practical Approach for Policies to Address GHG Emissions from Indirect Land Use Change Associated with Biofuels”, 2009 Tonini, A. and M. Henseler, “Biofuel scenario analysis with the European SIMulation (ESIM) model, powerpoint presentation to Inter Service Group on indirect land use change, Brussels, 23 July 2009894 Tyner, W., F. Taheripour and U. Baldos, “Land Use Carbon Emissions Due to the US Ethanol Program”, powerpoint presentation, 26 January 2009 Tyner, W. E., F. Taheripour and U. Baldos, "Land Use Change Carbon Emissions due to US Ethanol Production", 2009 Tyner, W.E., F. Taheripour, Q. Zhuang, D. Birur and U. Baldos, "Land Use Changes and Consequent CO2 Emissions due to US Corn Ethanol Production: A Comprehensive Analysis", 2010 UNEP, "Towards sustainable production and use of resources: Assessing Biofuels", 2009 UNICA (Brazilian Sugar cane Industry Association), “Proposed Low Carbon Fuel Standard”, letter to Mary Nichols, Chair, Air Resources Board, 16 April 2009 United States Environmental Protection Agency (EPA) (2009) Draft Regulatory Impact Analysis: Changes to Renewable Fuel Standard Program. http://www.epa.gov/otaq/renewablefuels/420d09001.pdf United States Environmental Protection Agency (EPA) (2010) Final Rule Preamble: http://www.epa.gov/otaq/renewablefuels/rfs2-preamble.pdf Valin, H., “Modelling land use change in a CGE model – the experience from IFPRI with MIRAGE”, powerpoint presentation to workshop on marginal yields and land allocation in land use change emissions estimates, Brussels 22 July 2009 Valin, H., B. Dimaranan and A. Bouët, “Biofuels in the world markets: CGE assessment of environmental costs related to land use changes”, XIIth GTAP conference paper, 15 April 2009 van Dender, K., “Energy policy in transport and transport policy”, Energy Policy 37, 2009 von Lampe, M., “Land Allocation Effects of Biofuel Support Policies – Methodology and Experiences at the OECD”, powerpoint presentation to workshop on marginal yields and land allocation in land use change emissions estimates, Brussels, 22 July 2009

Page 232: THE IMPACT OF LAND USE CHANGE ON GREENHOUSE GAS … · 5 Executive summary The modelling of the land use change impact of biofuels is new. The first study (by Banse et al.) appeared

232

Walker R and Moran E , (2000), "Deforestation and Cattle Ranching in the Brazilian Amazon ; External Capital an Household processes", World development Vol.28, No 4, p 683-699 Wang, M., “Dr. Wang Response to Charge Questions”, in “Lifecycle Greenhouse Gas Emissions due to Increased Biofuel Production – Model Linkage Peer Review Report”, prepared by ICF International for EPA, 2009 Woltjer, G., “Tackling the tension between localized land in land allocation models and CGE modelling”, powerpoint presentation to workshop on marginal yields and land allocation in land use change emissions estimates, Brussels, 22 July 2009 Woltjer, G. and A. G. Prins, "European Biofuel policy alternative: effect of forest exclusion", powerpoint presentation to workshop on "A land-use modelling framework for the European Commission", Brussels, 12 November 2009 World Database on Protected Areas, "Growth in Nationally Designated Protected Areas (1872-2008)", 2009 Woods, J., G. Brown, A. Gathorne-Hardy, R. Sylvester-Bradley, D. Kindred and N. Mortimer, "Facilitating carbon (GHG) accreditation schemes for biofuels: feedstock production", HGCA Project MD-0607-0033, 2008 Woods, J. and R. Murphy, response to comments of I. Hodgson on Greenergy-Econometrica technical paper TP-080212 A, 2009 WWF and Co-operative Bank (n.d.), "Unconventional Oil – Scraping the bottom of the barrel?", n.d. WWF Germany, "Rain Forest for Biodiesel? – Ecological effects of using palm oil as a source of energy", 2007 WWF International, "The timber footprint of the G8 and China", 2002 Zanchi, G., D. Thiel, T. Green and M. Lindner, "Afforestation in Europe", Specific Targeted Research Project no. SSPE-CT-2004-503604, 2007