Ic

HAa

b

c

d

e

f

g

a

AA

KPICGLMDL

1

UiTthpt2e

foci

0d

Ecological Modelling 221 (2010) 2153–2166

Contents lists available at ScienceDirect

Ecological Modelling

journa l homepage: www.e lsev ier .com/ locate /eco lmodel

ntegrated assessment of agricultural policies with dynamic land usehange modelling�

edwig van Deldena,∗, Tomasz Stuczynskib, Pavel Ciaianc,d, Maria Luisa Paracchinie, Jelle Hurkensa,rtur Lopatkab, Yu-e Shia, Oscar Gomez Prietoe,f, Silvia Calvoe,g, Jasper van Vlieta, Roel Vanhouta

Research Institute for Knowledge Systems (RIKS), P.O. Box 463, 6200 AL Maastricht, The NetherlandsInstitute of Soil Science and Plant Cultivation (IUNG), Czartoryskich 8, 24-100 Pulawy, PolandKatholieke Universiteit Leuven, P.O. Box 5005, 3000 Leuven, BelgiumEuropean Commission, Joint Research Centre (JRC), Institute for Prospective Technological Studies (IPTS), C/ Inca Garcilaso 3, 41092 Sevilla, SpainEuropean Commission, Joint Research Centre, Institute for Environment and Sustainability, Via E. Fermi 1, I-21020 Ispra (VA), ItalyEuropean Environment Agency, Kongens Nytorv 6, 1050 Copenhagen, DenmarkDepartment of Marine Geosciences and Land Use Planning of the University of Vigo, Campus Lagoas-Marcosende, 36310 Vigo, Spain

r t i c l e i n f o

rticle history:vailable online 4 May 2010

eywords:olicy supportntegrated assessmentommon Agricultural Policyeonamica

a b s t r a c t

With about half of its territory being farmed, agriculture is the main land use in the European Union(EU). As over 10% of the total EU manufacturing output comes from the agri-food sector, it also is aneconomic factor of great importance. Moreover, EU policy in this sector has far-reaching consequencesranging from the EU’s status as a global trade partner to landscape preservation and development. TheLUMOCAP Policy Support System is targeted towards policy makers in the European Commission (EC)and its Member States (MS) and aims to provide support in the field of sustainable agricultural and ruraldevelopment. To this end it incorporates an integrated model with socio-economic and bio-physical

and use changeodel integrationecision Support SystemUMOCAP

processes, operating at different spatial scales. For supporting integrated assessment, a large number ofpolicy levers is included as inputs for these models and outputs are transformed into policy-relevantsocial, economic and environmental indicators. The whole system is framed in a flexible, modular andeasy to use software package that is useable for process experts and policy-analysts alike.

This paper describes the integrated model, the individual models and a first calibration of the system.m’s bem an

It demonstrates the systecurrent status of the syst

. Introduction

As agriculture covers about half of the territory of the Europeannion (EC, 2007a), the Common Agricultural Policy (CAP) (EC, 2005)

s an important driver for land use structure and landscape quality.he emphasis of the early CAP was to promote agricultural produc-ivity and supporting farmers’ incomes. In recent years its focusas shifted more and more to rural development, including the

reservation of landscapes and sustainability of agri-ecosystemshrough facilitating a proper management of natural resources (EC,006a). The LUMOCAP project – Dynamic Land Use change MOd-lling for CAP impact assessment on the rural landscape – directly

� This publication has been funded under the EU 6th Framework Programmeor Research, Technological Development and Demonstration, Priority 8.1. Policy-riented research (European Commission, DG Research, contract 006556). Itsontent does not represent the official position of the European Commission ands entirely under the responsibility of the authors.∗ Corresponding author.

E-mail address: hvdelden@riks.nl (H. van Delden).

304-3800/$ – see front matter © 2010 Elsevier B.V. All rights reserved.oi:10.1016/j.ecolmodel.2010.03.023

ehaviour for typical scenario runs and concludes with a reflection on thed some recommendations for further development.

© 2010 Elsevier B.V. All rights reserved.

contributes to an assessment of the impact of policies targetingthese objectives.

The LUMOCAP Policy Support System (PSS) aims to assess howdifferent policy scenarios will impact the land use and landscapein the 27 Member States of the European Union (EU-27). Becauseof the inherent complexity of land use change processes, agricul-tural policies at the European level have their effect not only ondevelopments in the agricultural sector, but also on for instanceregional ecological coherence and socio-economic dynamics ofrural areas. This means that a model for policy impact assess-ment should reach beyond EU agricultural policies and includepolicies and processes at other levels and sectors such as local zon-ing regulations, infrastructure planning and interaction betweensectors as well as external factors like climate change and socio-economic drivers. The LUMOCAP PSS allows investigating the

relation between EU policies, agricultural economics, land suit-ability and land use dynamics through dynamic simulation. Itincorporates an integrated model, tools to set-up scenarios for (acombination of) policy measures and external factors and tools tovisualise and analyse indicators.

2 al Mod

tcgc(SSottob

p(fttdpba

aideLatme

teiphaet–BLte

pmitbof

2

ebpooS

154 H. van Delden et al. / Ecologic

Over the past years different systems have been developed withhe aim to provide support to policy makers in the field of agri-ulture and rural development. These can be divided in two mainroups: those focusing on a comprehensive modelling of the agri-ultural sector like CAPRI (Britz and Witzke, 2008) and SEAMLESS-IFVan Ittersum et al., 2008), and the Integrated Decision Supportystems (IDSS) like EURURALIS (WUR and MNP, 2007), the SEN-OR SIAT tool (Sieber et al., 2008) and the LUMOCAP PSS focusingn broader issues of land use and land use change. Compared tohe agricultural systems, LUMOCAP includes a wider variety of sec-ors at the cost of detail in the agricultural sector; it has less focusn (agricultural) economics and more on the reciprocal relationetween economics and land use.

As a result, different mechanisms for integrating model com-onents have been used. CAPRI and SEAMLESS-IF calculate anequilibrium) end-condition and use iterations to obtain the resultsor a selected end-year. Similar to SEAMLESS-IF and CAPRI, the SIATool calculates an end-condition, instead of a development overime. LUMOCAP includes a simulation model that calculates futureevelopments in yearly time steps based on the set of drivers incor-orated. Results for each year build on the results from the yearefore. This allows for mutual interactions between the economicnd the land use components during each simulation step.

Only two systems, EURURALIS and LUMOCAP, model EU-27 athigh level of spatial detail (1 km grid cells or less). However, the

nteraction between models operating at different spatial scales isifferent. In EURURALIS land use demands are calculated by mod-ls at national level and subsequently allocated to 1 km grid cells.UMOCAP incorporates four spatial scales (EU, national, regionalnd local) and there is top-down as well as bottom-up interac-ion. An example of the bottom-up interaction is that results from

odels at local level (land use and land suitability) are used in theconomic model at European level.

Another distinguishing characteristic between the systems ishe inclusion of models in them. EURURALIS pre-calculates differ-nt scenarios and provides the results in the EURURALIS tool, whilen LUMOCAP, SEAMLESS-IF and CAPRI, the actual models are incor-orated. Incorporation of the models allows users of LUMOCAP toave access to their underlying data and parameters and to creatend run scenarios based on any combination of policy options andxternal factors included in the model, without having to resort tohe developers team. The SIAT tool incorporates simplified relationsnamed response curves – that are derived from existing models.y incorporating the actual models instead of simplified relationsUMOCAP incorporates detailed process knowledge into the sys-em which enables to focus on the interaction between models atach time step.

This paper gives an overview of the integrated model incor-orated in the LUMOCAP PSS. First an overview of the integratedodel is provided, followed by a detailed description of each of the

ndividual models. Subsequently we describe the procedure used toest and calibrate the model and provide the main results of the cali-ration and typical scenario runs. The paper closes with a reflectionn the current status of the system and some recommendations forurther development.

. The integrated model

The core of the LUMOCAP PSS consists of a selection of mod-ls, all linked into a single integrated model simulating the linked

io-physical and socio-economic developments in the entire Euro-ean Union (EU-27) up to 30 years forward. To capture processesccurring at different spatial scales, the system includes modelsperating at four different levels: EU-27, country (the Membertates of EU-27), region (so-called NUTS 2 regions, which approxi-

elling 221 (2010) 2153–2166

mately match provinces within the Member States; NUTS standingfor Nomenclature of Territorial Units for Statistics in the EuropeanUnion) and local. At local level the system operates at a 1 km grid forEU-27 and at a 200 m grid for selected case regions. The temporalresolution of the system is a year, its temporal horizon 2030. Thedifferent models and their linkages are schematised in the systemdiagram in Fig. 1. This section provides an overview of the mod-els and their interactions. Details about the individual models aredescribed in the next section.

2.1. European-wide agricultural production and socio-economicfactors

At the highest spatial level of the model, EU-27, the LUMOCAPagricultural economic model (Section 3.1) calculates acreages percrop type, average yields and production. Expectations regardinggrowth or decline of population, Gross Domestic Product (GDP) andjobs are seen as external driving forces and can as such be enteredand/or adapted in the socio-economic scenario component. Climatechange scenarios can also be selected at this spatial level.

Some of the socio-economic drivers (GDP, inflation, marketprices and population) are used as input in the agricultural eco-nomic model. Population, jobs and the acreages per crop type aredisaggregated from EU level to the national level. Climate change(rainfall and temperature) directly impacts the land suitabilitymodel at local level.

2.2. Competition between countries and regions for allocation ofactivities

A spatial interaction and distribution model (Section 3.2) dis-aggregates population, jobs and hectares for different crop typesto the second spatial level, the individual Member States. Withineach country a second spatial interaction model distributes thenational figures to the NUTS 2 regions. The relative attractivenessof the countries and regions plays a crucial role in the migrationand distribution of activities and the allocation of crop areas. Thespatial interaction models simulate the competition between coun-tries and regions respectively and use a cross-sectoral approach,in which agricultural activities become integrated with the othersocio-economic activities (jobs in other sectors and population). Atregional level, activities are converted to land use demands (areasper land use category) that are then fed to the next spatial level.

2.3. Local dynamics simulating changes in land use and landsuitability

Within the NUTS 2 regions, a constrained cellular automatamodel (CCA) allocates the area demands for the different landuse categories – as calculated by the regional spatial interactionmodel – to cells of 1 km × 1 km (Section 3.3). This model is usedfor broad land use categories (residential areas, industry & com-merce, recreation, agriculture, forest and natural vegetation) andsimulates the competition for space between land use categories,in order to obtain the most preferential locations. Since this modelis not very well suited to simulate crop choice decisions, a separatemodel is used to determine what crops will grow on the agricul-tural land, and where. The third model at local level calculates theland suitability for different crop types for each location, based onlocal characteristics and impacts of climate change. Results of theland suitability model are used for the allocation of the broad land

use categories and crop types in the land use and the crop choicemodel, respectively. Aggregate suitability information is used asone of the factors determining the attractiveness of countries andregions in the spatial interaction models at regional and nationallevel and in the agricultural economic model at EU level. In these

H. van Delden et al. / Ecological Modelling 221 (2010) 2153–2166 2155

CAP s

mycsa

2

mommRL1tttaitm2a

Fig. 1. The LUMO

odels it is – together with technology factors – used as a proxy forield. Besides aggregate suitability values, also other relevant localharacteristics such as existing practices, accessibility and availablepace are used as factors influencing the attractiveness of regionsnd countries.

.4. A system for policy support

To ensure the policy-relevance of the LUMOCAP system, policyeasures at different spatial levels are included that can be entered

r adapted by the user (Fig. 2). Examples of these policies are CAPeasures from pillar 1 at EU level (e.g. intervention prices, ani-al premiums, crop area compensations) (EC, 2007b; OECD, 2003),

ural Development Policies (RDP) (EC, 2006a) at national level andess Favoured Areas (LFA) (EC, 1999; CEC, 2009), Natura2000 (EEC,979, 1992) and construction of infrastructure at local level. Sincehe impact of policy interventions depends heavily on other drivershat cannot be defined by policy makers, LUMOCAP also enableshe user to select different external factors such as climate changend socio-economic scenarios to analyse the robustness of the pol-

cy options under different external conditions. On the output sidehe model results are converted into social, economic and environ-

ental indicators (EEA, 2005; Petit et al., 2008; Farrington et al.,008), such as agricultural profit, production and yield, nationalnd regional jobs and population, crop diversity, degree of open-

ystem diagram.

ness, hemeroby, abandoned land, afforested land and land use inhigh nature value (HNV) farmland.

LUMOCAP is developed using the Geonamica software environ-ment that is specifically designed for the development of spatialdecision support systems that integrate a number of non-spatialand spatially explicit models (Hurkens et al., 2008). To facilitate theuse of the system for different users with different needs, the graph-ical user interface (GUI) is split into a policy interface for integratedassessment and a modeller interface for updating data and calibra-tion parameters and for fine-tuning the models incorporated in thesystem. The policy interface only gives access to a limited numberof drivers and indicators; the modeller interface gives access to alldata and parameters and to most of the equations.

3. Individual model components

3.1. Models and data at European level

At the highest spatial level of the model, scenarios for the fol-lowing drivers are included:

• Population (EC, 2007c),• GDP (FAPRI, 2007a),• Inflation rates (FAPRI, 2007a),

2156 H. van Delden et al. / Ecological Modelling 221 (2010) 2153–2166

F res cai nd onw

•

•

covsamct(

foOEmtcmodstFdi

among different Member States of the EU, three separate agricul-tural economic models were constructed, which are linked throughcommodity prices and agricultural policies:

ig. 2. Screenshot of the LUMOCAP PSS. The main window shows how CAP measumpact on, instruments are entered as numerical values or maps. Results can be fou

indow.

Jobs in main economic sectors (industry & commerce, tourism &recreation and agriculture). Initial data comes from Eurostat (EC,2007c), projections are derived from historic data (EC, 2007c) andGDP projections (FAPRI, 2007a),World market prices per crop type. The main source of worldprices used in the LUMOCAP PSS is the OECD’s AGLINK model(Conforti and Londero, 2001; OECD, 2004).

Bio-physical changes are assumed to be induced by climatehange scenarios. Four scenarios from the Intergovernmental Paneln Climate Change (IPCC, 2007) have been incorporated that pro-ide monthly information on rainfall and temperature for the entireimulation period. The only model operating at EU level is angricultural economic model (Fig. 3). This model is a dynamiculti-product supply model of EU agriculture and covers all agri-

ultural areas (Table 1). The overall design of the model focuses onhe potential influence of agricultural policy on land use changesCiaian, 2007a,b).

The structure of the agricultural economic model containseatures of the AGLINK model of the Organisation for Economic Co-peration and Development (OECD) (Conforti and Londero, 2001;ECD, 2004) and of the AGMEMOD model (Chantreuil et al., 2005;rjavec and Donnellan, 2005; AGMEMOD, 2006). It is an econo-etric model for which the coefficients used were derived through

hree main approaches: econometric estimation, calibration andoefficients taken from economic literature (Ciaian, 2007a,b). Theajority of the data used in the agricultural economic model was

btained from the EC (EC, 2007a) and Eurostat (EC, 2007c). Whenata from these two sources were missing, other sources were used,

uch as the Food and Agricultural Organisation (FAO), the OECD andhe United Nations (UN). The Eurostat data supplemented by theAO data (FAO, 2007) was used as source for commodity marketata (e.g. production, yields, prices, etc.). Policy information (e.g.

ntervention prices, RDP, direct payments, quotas, etc.) came from

n be entered for specific years and regions. Depending on the spatial level policiesthe land use map shown in the back and through the indicator section in the main

EC databases. From the OECD world prices were extracted as well aspre-accession policies for New Member States (NMS) (OECD, 2003,2007).

In order to take into account the differences in policies applied

Fig. 3. System diagram of the agricultural economic model. The arrows show theflows of information; black arrows represent current values and dashed arrowslagged values (values from the previous time step).

H. van Delden et al. / Ecological Modelling 221 (2010) 2153–2166 2157

Table 1Structure of the agricultural economic model.

Stage 1 Stage 2 Stage 3 Stage 4

Agricultural land loss to urban areasAbandoned land

Utilised Agricultural Area (UAA) Arable land Cereal–oilseed area CerealsOilseedsSet-aside

RicePotatoesSugar beetsTobaccoVegetablesFodder from arable landOther arable land

Permanent grasslandLand under permanent crops Vineyards

Olives

••

•

aoblt

eapwaa

3

rTwsws

cpTt

3

crai2ha

EU-15: comprising of the 15 “old” (pre-2004) Member States,NMS-10: comprising of the 10 New Member States which joinedthe EU in 2004 (Czech Republic, Cyprus, Estonia, Hungary, Latvia,Lithuania, Malta, Poland, Slovakia, and Slovenia),NMS-2: comprising of the 2 New Member States which joinedthe EU in 2007 (Bulgaria and Romania).

The agricultural economic model covers the entire agriculturalrea and all major crop and animal sectors of the agricultural econ-my: beef, dairy, sheep, pigs, cereals, oilseeds, rice, potatoes, sugareets, tobacco, vegetables, fodder from arable land, other arable

and, vineyards, olives, fruit crops and other permanent crops. Addi-ionally, set-aside area is modelled.

The model incorporates all major CAP instruments, which arexogenously introduced into the model. The modelling approachssumes that intervention prices affect EU prices. Coupled directayments are assumed to affect per hectare (or per animal) returns,hile the decoupled payments and rural development payments

re assumed to affect only the total area for agricultural and non-gricultural use.

.1.1. Price formationFor each regional sub model (EU-15, NMS-10, and NMS-2) one

epresentative price is used for each crop and for each animal sector.he representative price is taken from the most important marketithin the EU (for example French wheat price is used as repre-

entative price for wheat in the EU-15 sub model, while Hungarianheat price is used as representative price for wheat in the NMS-10

ub model).The EU market prices are endogenously determined in the agri-

ultural economic model. They are assumed to depend on worldrices and market intervention policies (e.g. intervention prices).his modelling of prices implies that they are not determined fromhe market-clearing conditions.

.1.2. Land demandsThe agricultural economic model calculates land demands for

rops at EU-15, NMS-10 and NMS-2 level. To derive consistentesponses of land use to external drivers and policies, the model

ssumes that farmers behave in a rational manner and aim at max-mising their profits (Guyomard et al., 1996; Erjavec and Donnellan,005; Britz and Witzke, 2008). Model drivers that are assumed toave an impact on profits are agricultural policies, market pricesnd technologies.

Fruit crops (excluding wine and olives)Other permanent crops

The basic decision tree for the division of land among cropsis provided in Table 1. The decision tree contains four stages. Atstage 1, a decision is taken on the amount of land for agriculturalactivities, the amount of abandoned land and the amount of landtransferred to non-agricultural uses. The division among these landcategories is driven by agricultural returns, animal stock, agricul-tural subsidies, GDP growth and population growth. The GDP andpopulation growth reflect the pressures from non-agricultural sec-tors on the land. At stage 2, agricultural land is subdivided amongthree main sectors: arable land, permanent grassland and perma-nent crops. Each of these three sectors is further split in morespecific sub-sectors at stage 3. At stage 4 there is a further divisionof the cereal–oilseed area.

The total demand for land and the division into specific land usecategories is based on a set of linear equations at each decision stageof the LUMOCAP approach (as provided in Table 1). Dependent vari-ables are represented in hectares (at stage 1) or as area shares of thehigher level total area (at stages 2–4). Main explanatory variableswhich determine the land division between crops are own returns,returns of competing crops, policy variables, animal stocks andmacro-variables such as GDP and population. Crop returns affectland use directly as they affect profitability of a particular crop.The animal sector affects land use indirectly through demand forfodder. Macro-variables represent the pressure of non-agriculturalsectors on land use.

3.1.3. Yields, slaughter weights and total productionThe yield (Y) equation is a function of market price (p), techno-

logical development, physical characteristics and climate change.The last three factors are calculated in the land suitability modeland provide together the value for the yield elements calculated atlocal level (YLS):

Y = f (p, YLS)

Slaughter weights are a function of market price and techno-logical development. A positive impact of prices on yields andslaughter weights is expected. Prices change marginal profits ofoutputs. Yields are used to calculate the per hectare returns inthe agricultural economic model (price times yield plus coupleddirect payments) while slaughter weights are used to calculate

animal returns (price times slaughter weight plus coupled directpayments). The total production for a specific crop is obtained bymultiplying the crop area with the yield. The slaughter weightsdetermine the productivity of the animal sector. All equations andparameter estimations are provided in Ciaian (2007a).

2 al Mod

3

slhrststaddlStma2sv2

3a

siaempg

Fv

158 H. van Delden et al. / Ecologic

.2. Models and data at national and regional level

An important challenge in the development of the LUMOCAPystem was to bridge the gap between EU level information andocal level land use. On the one hand crop areas, population and jobsad to be downscaled to land use at local level; on the other hand,elevant information from the local level that plays a role at higherpatial levels (e.g. local yield calculations that can be aggregatedo NUTS 2, MS or EU level) had to be embedded in the processesimulated at those levels. An approach was selected that treatshe Member States and the NUTS 2 regions separately, since it wasssumed that processes and policy options play different roles atifferent levels. This means that information from EU level is firstownscaled to the national level and subsequently to the regional

evel. Local level information is aggregated to NUTS 2 or Membertate level depending on its relevance to contribute to the attrac-iveness of regions and countries. At both spatial levels the same

odelling concept is applied, but variables and parameter settingsre different. Data used in these models come from Eurostat (EC,007c) and FAO (FAO, 2007). The approach builds on an existingpatial interaction and distribution model that has been used in pre-ious projects and products (Environment Explorer, Engelen et al.,003; Xplorah, Van Delden et al., 2008; Metronamica, RIKS, 2009).

.2.1. Modelling competition for population and economicctivity between countries and regions

This model caters for the fact that the overall growth will notpread evenly over the modelled area, but rather that regionalnequalities will influence the location and relocation of residents

nd economic activity and thus drive regional development. Fig. 4xplains the relations between the different components of theodel at national level and their relation to the other models incor-

orated in LUMOCAP. It shows that the model allocates overallrowth (a) provided by the socio-economic scenarios (b) as well

ig. 4. System diagram of the national (and regional) spatial interaction and distributionalues and dashed arrows lagged values (values from the previous time step).

elling 221 (2010) 2153–2166

as the interregional migration (c) of activities and residents basedon the relative attractiveness of each region (White, 1977, 1978).In modelling the allocation of socio-economic growth and migra-tion, distance plays a crucial role. The underlying assumption forthis is that countries or regions benefit from other attractive coun-tries or regions as long as they are sufficiently close to each other.Furthermore, people and jobs get higher reluctance to migrate asthe distance of that migration gets bigger. The attractiveness forthe socio-economic sectors (population, jobs in agriculture, jobsin industry and commerce, jobs in tourism and recreation) (d) isbased on the existing socio-economic activity (e) as well as regionaland local characteristics (f). Local characteristics that are taken intoaccount are the suitability for different land use categories, theavailable space and the local accessibility. For the national spa-tial interaction and distribution model these factors are providedthrough the model at regional level.

3.2.2. Converting activities into area demands for land useFig. 4 shows that calculated activity levels create demand for

land (g) which is used as an input in the local level land use model.Cell-productivity (h) expresses the amount of activity in a sectorthat is located in one cell. The average cell-productivity in eachregion is modelled differently for economic and population sectorsand for natural sectors. In the latter case, the activity is defined interms of surface area, so the productivity is equal to the surface areaof one cell. In the former case, the activity is expressed in terms ofthe number of jobs or people.

3.2.3. Adaptations made to the original model to incorporate

agriculture

For the agricultural classes we use a simplified version of theabove-mentioned spatial interaction and distribution model. Sincethe agricultural economic model (i) already provides informationon the surface of each crop type, the model requires no conversion

model. The arrows show the flows of information; black arrows represent current

al Modelling 221 (2010) 2153–2166 2159

fdbcafrete

••

••

•

fit

mt

3

afafit2

3

iMErso

boti

•

•

•

H. van Delden et al. / Ecologic

rom activity to area. Furthermore it is assumed that agricultureoes not migrate between regions in the same way that people andusinesses do and does not benefit from good practices in regions atloser distance. Since the distance component in the spatial inter-ction model mainly impacts on those processes, it was omittedrom the equation for the distribution of agricultural areas. Whatemains is a disaggregation based on the relative attractiveness ofach Member State and NUTS 2 region (j). Factors that contribute tohe attractiveness are a combination of physical, social and financiallements:

Land suitability on the basis of which the yield is calculated (k),The number of large farms, to take into account economies ofscale for those crops for which this is relevant (e.g. cereals),Jobs in agriculture,Relative advantages related to the way policies (subsidies) areimplemented,Tradition or existing practices (l).

Since relative attractiveness is modelled, only those elementsor which national or regional differences could be found werencluded. Information from the local level was provided throughhe regional spatial interaction and distribution model.

This section described the spatial interaction and distributionodel at national level. The model at regional level works according

o the same concepts.

.3. Models and data at local level

At local level, there are three models that play a role: a land usellocation model, a crop choice model and a land suitability modelor agricultural crop types (Fig. 1). Main data sources for the modelst local level are the CORINE land cover maps (EEA, 2007a,b), mapsor height, slope and soil characteristics (Lambert et al., 2003), pol-cy maps such as Less Favoured Areas and Natura2000, the GISCOransport network (EC, 2006b), and yield data from Eurostat (EC,007c).

.3.1. Land useIn LUMOCAP the allocation of land use demands from the spatial

nteraction model at NUTS 2 level, is modelled by means of theetronamica cellular automata based land use model (White and

ngelen, 1993; Barredo et al., 2003; RIKS, 2009) in which EU-27 isepresented as a mosaic of 1 km grid cells, each occupied with apecific land use. All cells together constitute the land use patternf the European Union.

In principle, it is the relative attractiveness of a cell as viewedy a particular spatial agent, as well as the local constraints andpportunities that cause cells to change from one type of land useo another. Changes in land use at the local level are driven by fourmportant factors (Fig. 5):

Physical suitability, represented by one map per broad land usecategory modelled. The term suitability is used here to describethe degree to which a cell is fit to support a particular land useand the associated economic or residential activity. The agricul-tural suitability maps are calculated in the land suitability modeldescribed below.Zoning or institutional suitability, represented by one map perland use category modelled. Zoning maps are used for enforcingspatial restrictions on the allocation of land uses. For each land

use there is a time series of zoning maps, specifying which cellscan and cannot be taken in by the particular land use allowingchanging zoning regulations over periods of time.Accessibility, represented by one map per land use category mod-elled. Accessibility is an expression of the ease with which an

Fig. 5. Drivers of the land use model at local level. For agriculture, suitability iscalculated on a yearly basis in the physical suitability model based on climate changescenarios and local characteristics.

activity can fulfil its needs for transportation and other infras-tructure in a particular cell, based on the infrastructure network.

• Dynamic interaction of land uses in the area immediately sur-rounding a location is represented by the neighbourhoodpotential. For each land use function, a set of spatial interactionrules determines the degree to which it is attracted to, or repelledby, the other functions present in its surroundings; a 196 cellneighbourhood with a radius of 8 cells (8 km). If the attractive-ness is high enough, the land use will try to occupy the location,if not, it will look for more attractive places. New activities andland uses invading a neighbourhood over time will thus change itsattractiveness for activities already present and others searchingfor locations. This process constitutes the non-linear character ofthis model.

On the basis of these four elements, the model calculates forevery simulation step the transition potential for each cell and landuse. Over time and until regional demands are satisfied, cells willchange to the land use for which they have the highest transitionpotential. Consequently, the transition potentials reflect the pres-sures exerted on the land and in this way constitute importantinformation for those responsible for the design of sound spatialplanning policies.

The local land use model in LUMOCAP includes 19 land useclasses of which 6 are dynamically modelled: agriculture, residen-tial, industry & commerce, tourism & recreation, forest and naturalvegetation. The remaining classes (e.g. water bodies, dump sites,airports) are assumed to remain stable during the simulation periodor entered manually as policy decisions (e.g. construction of infras-tructure).

3.3.2. Crop choiceAfter the land use model has allocated the broad land use cate-

gories, the crop choice model specifies what crop types will occupythe agricultural areas. All cells that have been designated as ‘agri-culture’ by the land use model are allocated a surface area for eachcrop type, summing to the surface area of a cell (100 ha). The totalsper NUTS 2 region sum to the regional demands for each crop type.

To allocate crops to agricultural areas, the crop choice modelcalculates the potential for each cell and each crop type on thebasis of land suitability, policies and existing practices. Then foreach NUTS 2 region an iterative approach is taken to allocate theregional demands for crops to all agricultural cells. It is assumed

2 al Mod

tafocprtdst

3

tfyfsf

C

mitGuiptttat

afstT1iambnttFwopM

t

160 H. van Delden et al. / Ecologic

hat the regional demands sum to the surface area taken up by allgricultural cells in each region. To meet the requirement that theull surface area of a cell (100 ha) is allocated to crops, the ratiof the potential of a crop type and the sum of the potential for allrop types determines the surface area of a crop type in a cell. Asotential values are non-negative, the allocated surface areas willange between 0 and 100 ha and sum to 100 ha. The requirementhat the surface area of each crop in a region sums to the regionalemand for that crop is handled by rescaling the potentials with acalar value for each crop type. The scalars are iteratively adaptedo approximate the regional demands.

.3.3. Land suitabilityLand suitability plays an important role in the allocation of crop

ypes. It is used as a proxy for yield and reflects locational dif-erences. The potential yield is defined as a product of maximalield and land suitability which mainly depends on environmentalactors such as terrain conditions, temperature, soil moisture andpecific crop features. The land suitability is normalized and rangesrom 0 to 1.

The above assumptions can be formalized by the followingobb–Douglas type production function:

(1)

In which YLS is the yield as calculated in the land suitabilityodel; YLS,potential is the potential yield based on physical character-

stics; LS is the land suitability; GDP is the Gross Domestic Product ofhe Member State in which the cell is located; GDPmean is the meanDP in EU-27; Ymax is the maximum yield that can be obtainednder optimal conditions; slope is the slope of the location; Tmean

s the average temperature of the location; Topt is the optimal tem-erature for crop growth; ETa is the actual evapotranspiration onhe location; ETp is the potential evapotranspiration, ˛ is the impor-ance of terrain constraints in calculating LS; ˇ is importance ofemperature in calculating LS; � is the importance of water avail-bility in calculating LS; ı is the impact of GDP in relating potentialo actual land suitability.

The first term in the LS equation is a function which producesvalue of 1 if the slope is 0◦, as there are no terrain constraints

or crop growth. For a slope of 90◦, it equals 0, and for intermediatelopes the LS value depends on the ˛ coefficient. The second term inhe LS equation characterizes the impact of temperature on yields.his function is of a skewed bell shape and has a maximum value ofat a temperature optimal for plant growth, whereas the value of 0

s reached at 0 ◦C, which is a temperature where growth processesre stopped. The third term represents the ratio between the dryatter biomass produced in a vegetation season and the dry matter

iomass that could be produced in optimal conditions – if there iso water stress this term would be equal to 1. It is assumed thathe dry matter produced is proportional to the amount of evapo-ranspiration (De Wit, 1958) which is calculated according to theAO methodology (Allen et al., 2000). Within this term there is aater balance calculation included, relating the stage of crop devel-

pment to the water available to plants in the soil, the amount ofrecipitation and the actual evapotranspiration (Thornthwaite andather, 1955).The use of this type of approach allows assessing the current sta-

us of land suitability for specific crops as well as projecting changes

elling 221 (2010) 2153–2166

in land suitability resulting from climate change trends as reflectedby different scenarios.

It is worth noting that the actual yield is not only controlledby natural conditions but also influenced by management andinput levels. It has been established that the utilisation of potentialland productivity correlates strongly with socio-economic condi-tions, characterized by indicators such as GDP. It was observedthat in regions with a high GDP the quality of agricultural man-agement, production inputs, means and technology are far moreadvanced relative to areas with low GDP. A combination of theseGDP related factors apparently provides a better compensation ofenvironmental constraints and therefore the actual yield based onland suitability (YLS) approaches its potential level (YLS,potential) inmore developed regions (Lopatka and Stuczynski, 2007). The firstterm in Eq. (1) is a coefficient that relates potential to actual landproductivity.

4. Calibration and behaviour testing: procedure and results

Calibration of integrated (land use) models is not easy, becausethey contain a multitude of parameters that influence the eventualresults. The approach followed in the calibration of the LUMOCAPsystem was to first test and calibrate the individual models in iso-lation, next in pairs and small groups and finally all together in thecomplete integrated model. Because of the differences in modelconcepts, different approaches were used for the calibration of theindividual models. In general, a historic analysis over the period1990–2000 was used to understand the drivers and processes thathave shaped the present. This knowledge was used to fine-tunethe parameters in such a way that the model was able to simu-late realistic developments. To determine the success of the modelcalibration, different assessment methods were used that best fitthe individual models. For the agricultural economic model andthe land suitability model, simulation results for 2000 were com-pared to data for 2000 and the model’s accuracy was assessed bycalculating the standard error; for the spatial interaction and distri-bution models and the local land use model, simulation results from1990 to 2000 were compared against results provided by a neutralmodel (Hagen-Zanker and Lajoie, 2008). If the model outperformedthe neutral model and parameter values could be related to real-world processes it was assumed that the model performed well. Theassessment method used for the spatial interaction and distributionmodels is the Root Mean Square Error (RMSE). The goodness of fit ofthe local model is measured based on a combination of assessmentmethods that include visual interpretation, cell by cell comparisoncalculated by the Kappa statistics (Monserud and Leemans, 1992)and measures for the land use structure, such as the clumpinessindex (McGarigal et al., 2002) and the cluster size–frequency distri-bution (White, 2006). Although visual interpretation is a techniqueoften used in calibrating land use change models, calibrating a sys-tem with the spatial extent and detail that LUMOCAP encompassescan hardly be done purely by visual comparison. Moreover, spa-tial metrics provide an objective measure which complements thesubjective visual interpretation.

It should be noted that although calibration is important forscenario studies, this does not mean that the future will be anextrapolation of the past. Understanding the drivers and processesthat have shaped the present as well as the (changes in) drivingforces that might occur in the future, are crucial factors for any

scenario study. For this reason also the long-term behaviour of themodel was tested. The model was run until 2030 under variousassumptions for the main drivers to test if hypothesis would bemet. The results of the historic calibration and the scenario runsare described below.

H. van Delden et al. / Ecological Modelling 221 (2010) 2153–2166 2161

Table 2Standard errors for land demand equations (1996–2004) expressed as the percent-age deviation of the simulated value from the observed value.

EU-15 NMS-10 NMS-2

Utilised Agricultural Area (UAA) 1.0 6.1 0.9Permanent grassland 2.3 4.6 2.2Arable land 1.3 8.4 2.3Land under permanent crops 4.5 28.3 7.5

Arable crop area equationsCereal–oilseed area 1.7 2.3 2.7Rice area 6.2 21.2 38.3Potato area 3.7 26.9 5.8Sugar beet area 2.8 12.3 40.2Vegetable area 2.4 9.0 10.5Fodder from arable land area 1.9 13.7 24.0Tobacco area 5.6 18.2 15.8

Permanent crop area equationsVineyard area 2.3 21.7 4.9

4

rtdaaAot

rldc

cilmrdkts

4

dvdnc

TR

Table 4RMSE results for the demand for land use (in cells).

RMSE simulation RMSE constantshare

Residential land use 260 274

Olive area 10.4 15.0Fruit crop area 2.5 14.4 15.4

Average 3.5 14.4 13.1

.1. Calibration of models at EU level

Econometric estimation techniques were used to obtain modelesponse parameters (e.g. price response of yield, land use responseo returns). When estimation techniques could not be applied (e.g.ue to short time series available), response parameters from liter-ture were used (e.g. Choi and Helmberger, 1993; Guyomard etl., 1996; Lekakis and Pantzios, 1999; Moro and Sckokai, 1999;GMEMOD, 2006; FAPRI, 2007b). Model calibration was used tobtain constants of the model equations particularly for those equa-ions for which parameters from literature were used.

The model’s performance was measured by comparing itsesults to historic trends. Table 2 reports standard errors for theand division equations for the period 1996–2004. The EU-15 landivision equations have smaller standard errors than the land allo-ation equations of NMS-10 and NMS-2.

The land division equations do not incorporate institutionalhanges implemented in the NMS. Wide ranging reforms weremplemented from the early 1990s on (e.g. land reform, priceiberalisation), with the aim to replace a planned economy with

arket institutions and later on with the aim to join the EU. Theseeforms are an important factor that significantly affected farmecision making in the NMS, besides agricultural policies and mar-et returns. Similar changes are not expected in the future andherefore setting parameter values for NMS for scenario studiesolely based on historic data would not be appropriate.

.2. Calibration of models at national and regional level

The main parameters to be calibrated at this level are those thatetermine the attractiveness and the productivity of a region. RMSE

alues were computed for all activities (Table 3) and for all pro-uctivities (Table 4), where the latter ones are measured from theumber of cells rather than the productivity directly. Results wereompared with the constant share solution for the same period. The

able 3MSE results for the activities.

RMSE simulation RMSE constantshare

Population 496,241 555,498Jobs in agriculture 252,609 317,448Jobs in industry & commerce 1,143,804 1,305,408Jobs in tourism 81,352 117,737

Agricultural land use – –Commercial and industrial land use 105 117Recreational land use 48 45

results for the productivity of agricultural land use are not given inTable 4 since they are provided by the agricultural economic modeland therefore exogenous to the spatial interaction and distributionmodels.

The results presented in Table 3 show that the calibration of thespatial interaction and distribution model was successful. For allactivities, model results outperform the constant share model by amargin of 10–30%. However, it should be noted that Romania wasexcluded in the assessment of the distribution of jobs in agriculture.It was also possible to outperform the constant share model whileincluding Romania, but that would require nonsensical parametersettings. Therefore, and because it is suspected that the increasein farmers in Romania does not correspond to reality, is against alltrends, and seems to be generated by the data itself (collection orrepresentation), it was excluded.

The results for the productivity (Table 4) show that the simula-tion outperforms the constant share model for almost all land uses.However, the difference is much smaller than the margin observedfor the activities. This might be partly caused by the typical differ-ences between Member States. Some of the new Member States arevery much in a transition phase which heavily influences their pro-ductivity. For example, Eastern European countries are typicallychanging from an industrialised economy towards a service ori-ented one. The latter requires much less space which translates ina higher productivity. Since these transition processes are typicalfor only some countries, there was a higher variation among them.

Details of the calibration results are available in the final reportof the LUMOCAP project (LUMOCAP, 2009).

4.3. Calibration of models at local level

Main parameters that need calibration in the land use model arethe interaction rules, the extent of the random effect, the accessi-bility parameters, the weighing factors for the suitability and thedifferent elements in the total potential.

Simulation results and reference results were both compared tothe actual land use map. This comparison indicates that the sim-ulation results are more accurate, as the Kappa values are 0.972and 0.957, respectively. Analysis of land use patterns over the cal-ibration period indicates that the model simulates plausible landuse developments, since it corresponds better than the referenceresults with actual land use patterns. The clumpiness index for res-idential land use for the actual land use, the simulated land use andthe reference result are 0.33, 0.34, and 0.29, respectively. Similarresults are obtained for agricultural land use: 0.67, 0.63, and 0.57,respectively.

Parameters in the crop choice model that need calibration arethe impact of the suitability, the impact of LFA policies and the iner-tia of the crop allocation. Since no crop type maps were availablefor the crop types modelled in LUMOCAP, these maps were createdbased on NUTS 2 data from Eurostat, Corine land cover maps andancillary information such as the LUCAS data (EC, 2000). The latter

provides information on land use and land cover for approximately10,000 locations in 18 European countries.

Because a similar procedure was used for the creation of mapsand for modelling crop allocation, it would not be appropriate to

2162 H. van Delden et al. / Ecological Modelling 221 (2010) 2153–2166

Table 5Calibration results for the land suitability model.

Crop R2 na

Wheat 0.81 215Barley 0.78 218Rye 0.71 191Potato 0.67 218Rape 0.70 184Soya 0.69 79Pulse 0.58 149Sunflower 0.63 146Maize grain 0.52 197Flax 0.52 85Cotton 0.13 12Sugar beet 0.47 180Tobacco 0.32 105Rice 0.30 41Vine 0.23 84Maize fodder 0.13 196Olive 0.03 51

cae

amNpbr

GyciEtealdmoap

4

pt

1

2

Fig. 6. Percentage of land used for agriculture per agricultural suitability value in

over Europe in the baseline scenario. Fig. 6 confirms this decline

Fruit 0.36 101Permanent crops 0.41 76

a n: number of NUTS 2 units in which given crop is grown.

alibrate based on these maps. The model was therefore testednd calibrated based on its long-term behaviour under differentxternal conditions.

In the land suitability model, parameters that need fine-tuningre ˛, ˇ, � , ı, Ymax and Topt. The calibration of the model throughultiple linear regression was based on the Eurostat yield data forUTS 2 regions. The performance of the model is assessed by com-aring model results to average yield at NUTS 2 level as reportedy Eurostat. The assessment method used to compare the modelesults against real data is the least square method.

Results of the model calibration (after correcting the impact ofDP on yields) show that the model can explain 80% of the observedield variability for wheat across EU-27, whereas for other mainrops R2 usually approaches 0.6 (Table 5). The model performances significantly poorer for crops which are not widely grown in allU regions such as cotton, vines and olives – for which very little ofhe variability reported in Eurostat can be explained by the param-ters in the land suitability model. One possible reason is that themount and quality of the data is insufficient. However, it is moreikely that there are specific management and environmental con-itions influencing crop productivity which are not reflected in theodel. Aside from this and considering that cereal crops cover 60%

f arable land in EU-27, results indicate that the generated land suit-bility maps are able to explain the variability of land productivityotential in Europe.

.4. Testing the model’s long-term behaviour

For EU-27 no independent data set is available for validationurposes. Therefore the long-term behaviour of the LUMOCAP sys-em was tested through the following exercises:

. Assessing the temporal evolution of land use change from 2000to 2030. The assumption was that the baseline scenario shows acontinuous decline of the total Utilised Agricultural Area (UAA)following the past observed trend. Locations with low suitabilityare expected to be the first ones to be abandoned.

. Testing the impact of changes in CAP policy measures on thetotal agricultural area and yield. The assumption is that fewer

subsidies would result in a decrease of the agricultural area; thiswould especially cause a decline in crop types which are nowheavily subsidised. No major changes in the average yield percrop type are expected.

the baseline scenario. The graph shows that in the period 2000–2030 there has beena decline in agricultural area and that land abandonment has mainly taken place atlocations with low suitability. In 2030 locations with suitability values below 50 areno longer occupied by agriculture.

3. Testing the impact of GDP on the total agricultural area andyield. The assumption is that higher GDP would result in anincreased competition for land. As a result an increase in urbanarea and a decrease in agricultural area are expected. Because ofthe increased possibility to invest in agriculture average yieldsare expected to increase.

4. Testing the impact of climate change on the total agriculturalarea and yield. The assumption is that some crops benefit fromclimate change, while others experience negative consequences.

5. Testing the development of urban clusters. The clustersize–frequency distribution is assumed to show the behaviourobserved in the calibration period. Furthermore, to be able tomodel urban dynamics realistically, the model should be able toshow the emergent behaviour of (urban) developments.

6. Testing the spatial developments over time. At the local level twoprocesses are expected to cause a decline in agricultural area:urbanisation and land abandonment. The first process shouldoccur mainly around existing urban areas, the second process inmarginal regions.

Five different scenarios were run with the LUMOCAP PSS. Abaseline scenario representing business as usual conditions, a liber-alisation scenario assuming an elimination of agricultural subsidiesafter 2013 with other conditions kept similar to the baseline, agrowth scenario assuming a growth in GDP which is 50% higherthan the growth in the baseline, and a climate change scenarioin which the HadCM2GSd1 scenario is used compared to theHadCM2GGa1 scenario in the baseline simulation. The IPCC sce-narios differ in precipitation and temperature. In Southern Europetemperatures will be approximately 2 ◦C lower relative to thebaseline, whereas for Northern Europe no significant difference isexpected.

Testing the long-term behaviour gives the following results(Tables 6–8):

1. Table 6 shows that there is a decline of UAA from 2000 to 2030 all

and shows that land abandonment has mainly taken place atlocations with low suitability.

2. Comparing the liberalisation scenario against the baseline sce-nario indicates that no significant changes have occurred in the

H. van Delden et al. / Ecological Modelling 221 (2010) 2153–2166 2163

Table 6Change in agricultural area (%) during the period 2000–2030 for the baseline scenario and the climate change scenario and the difference between both scenarios in 2030.

Change in agricultural area inthe baseline scenario from2000 to 2030

Change in agricultural area inthe climate change scenariofrom 2000 to 2030

Difference between the climatechange scenario and thebaseline scenario for 2030

Crop type EU-15 NMS-10 NMS-2 EU-15 NMS-10 NMS-2 EU-15 NMS-10 NMS-2Cereals 20.36 −1.51 −7.79 20.62 −1.51 −7.35 0.22 0.00 0.47Other arable land 2.17 −26.19 −36.65 0.12 −27.63 −36.44 −2.01 −1.96 0.33Rice −12.88 −15.08 −12.63 −13.85 −15.08 −12.73 −1.11 −0.01 0.11Potatoes −27.36 −41.61 −24.40 −26.64 −41.30 −24.26 0.99 0.53 0.19Sugar beets −55.21 −26.57 74.51 −55.17 −26.69 69.56 0.08 −0.16 −2.84Oilseeds 25.12 −4.22 −6.65 25.36 −4.24 −7.68 0.19 −0.02 −1.11Tobacco −49.86 −15.08 −12.63 −49.31 −15.08 −12.54 1.09 −0.01 0.11Vegetables 6.28 1.32 2.63 6.42 1.85 2.66 0.13 0.52 0.03Fruit crops −22.07 44.80 −5.91 −21.74 43.57 −5.73 0.43 −0.85 0.19Vineyards −20.04 −10.36 −20.79 −18.20 −10.67 −20.79 2.30 −0.35 0.00Olives −45.75 63.21 – −45.81 62.20 – −0.12 −0.62 0.00Fodder from arable land −12.24 −52.48 −3289 −11.71 −50.73 −32.93 0.60 3.69 −0.07Permanent grassland −16.46 −31.63 −20.82 −16.45 −31.63 −20.93 0.01 0.01 −0.14Other permanent crops 168.79 9.11 −21.62 159.31 8.82 −21.67 −3.53 −0.26 −0.06Utilised Agricultural Area (UAA) −3.31 −17.50 −15.49 −3.31 −17.53 −15.47 −0.00 −0.03 0.03

Table 7Impact of different scenarios on the average yield (in %). Comparison for the year 2030.

Change in the liberalisation scenariocompared to the baseline scenario

Change in the growth scenariocompared to the baseline scenario

Change in the climate change scenariocompared to the baseline scenario

Crop type EU-15 NMS-10 NMS-2 EU-15 NMS-10 NMS-2 EU-15 NMS-10 NMS-2Cereals −0.01% 0.01% 0.02% 2.85% 11.89% 16.64% 0.17% −0.22% 0.98%Rice 0.00% – −0.09% −6.84%Potatoes −0.04% 0.02% 0.01% 3.03% 9.08% 14.21% 0.20% −0.49% 0.44%Sugar beets 0.63% −1.97% 0.15% 4.31% 9.02% 13.61% −2.22% −2.22% −0.36%Oilseeds 0.13% 0.03% −0.05% 3.20% 7.66% 6.07% 0.21% −0.29% 0.61%

3

TI

Tobacco −0.49% 0.03% 0.11% 3.19%Vegetables −0.06% 0.02% 0.00% 1.70%Vines −0.08% −0.13% 0.00% 7.53%Olives 0.10% −0.37% −4.14%

average yield for the different crop types (Table 7). The total UAAhas declined to some extent, but changes become apparent in theallocation of land among the different crop types (Table 8). Theremoval of support influences the relative returns of crops: cropsthat currently rely heavily on subsidies become less attractivewhen subsidies disappear.

. Of all scenarios, the growth scenario shows the largest decline inUAA (Table 8). New Member States experience a larger decline

than Member States in EU-15. This could be due to the fact thatthe New Member States are expected to experience a higher eco-nomic growth which will lead to a stronger competition betweenother sectors. There is not a large impact on the share of differ-ent agricultural sectors as this scenario leaves relative returns

able 8mpact of different scenarios on the total area per crop type (in %). Comparison for the ye

Change in the liberalisation scenario comto the baseline scenario

Crop type EU-15 NMS-10Cereals −0.42 −1.45Other arable land −0.57 −35.16Rice 7.78 −3.04Potatoes −8.16 3.90Sugar beets −22.66 −6.32Oilseeds 14.26 −0.12Tobacco 19.49 −3.04Vegetables 2.16 3.04Fruit crops −3.05 8.73Vineyards −0.86 18.75Olives −27.18 −15.01Fodder from arable land 0.37 42.89Permanent grassland −3.03 −2.90Other permanent crops 35.62 16.96Utilised Agricultural Area (UAA) −0.56 −2.37

3.80% 5.86% −5.75% −2.21% −1.15%9.23% 6.79% 0.26% −0.25% 0.08%5.63% 2.87% 4.74% −0.71% −0.37%6.78% −15.16% −2.67%

largely unchanged. The high economic growth has an impact ontechnology and management practices which results in higheraverage yields in this scenario compared to the other scenarios(Table 7).

4. Use of a different climate scenario does not have a large impacton the total UAA (Table 8). Yields of different crop types are how-ever substantially impacted by changing assumptions on rainfalland temperature (Table 7). Analysing, e.g. sugar beets, tobacco

and olives indicates that higher yields are expected in the base-line scenario than in the climate change scenario. The reason forthis is that sugar beets, tobacco and olives are crops that prefer awarm climate. Declining temperature in relation to the baselinescenario results in a decrease of suitability and hence yields.

ar 2030.

pared Change in the growth scenario compared tothe baseline scenario

NMS-2 EU-15 NMS-10 NMS-2−14.38 −2.45 −6.96 −3.68−7.38 0.15 6.28 7.36−2.89 −2.73 −7.03 −8.21−0.24 −2.46 −14.21 −9.1322.49 −1.42 −6.97 −31.7034.01 −2.90 −6.58 −28.79−2.89 −6.99 −7.03 −8.21−1.96 −2.46 −7.63 −9.53−5.46 −2.93 −12.32 −12.54

0.88 −3.79 −14.49 −13.820.00 −3.34 −13.54 0.00

10.64 −3.03 −32.19 −10.802.56 −4.82 −19.85 −17.874.49 −2.57 −15.39 −14.94

−1.11 −3.09 −9.56 −11.43

2164 H. van Delden et al. / Ecological Mod

Fig. 7. Cluster size–frequency distribution of residential clusters based on 1990 data(green triangles), 2000 data (blue diamonds) and 2030 model results (pink stripes).The y-axis shows the cluster size of residential clusters, the x-axis their frequency.(r

5

6

5

t

For this reason process-based models were selected when possi-ble and calibration against (historic) data was used to fine-tune the

For interpretation of the references to colour in this figure legend, the reader iseferred to the web version of the article.)

Not all regions show the same behaviour: average yields ofcereal in EU-15 and NMS-2 are higher in the climate change sce-nario, while average yields in NMS-10 are higher in the baselinescenario. The reason for this is that in some countries changes intemperature result in improved conditions (actual temperaturescome closer to the optimal temperatures as included in Eq. (1)),while in other countries they have the reverse effect.

. Fig. 7 shows the cluster size–frequency distribution of res-idential clusters for 1990 (data), 2000 (data) and 2030(simulation results). Although the cluster size–frequency distri-bution remains rather constant during the period 1990–2000,larger clusters seem to increase which is consistent withobserved urbanisation trends. In the results for 2030 the clustersize–frequency distribution is similar to that of 1990 and 2000,but larger clusters seem to grow slightly more than average.

Details of the land use maps show a similar behaviour(Fig. 8a–c). Urban development is attracted by existing urbanareas, causing (most) existing clusters to grow. On the other handalso new clusters develop. Both at city level and at higher spatiallevel complex structures seem to emerge.

. Analysing the spatial consequences of the scenarios shows thaturban expansion often takes place on agricultural areas. In thebaseline scenario 75% of the new urban area in 2030 is allocatedon land previously occupied by agriculture. Fig. 8a–c shows acut-out of the land use map (the region between London andBirmingham) that confirms the growth of urban areas at theexpense of agriculture. The decline of agricultural area is alreadyconfirmed in the first test that also showed that land abandon-ment mainly takes place on locations with limited suitability.

. Discussion

In developing models for integrated assessment of EU policieshe main challenge is to develop a model that:

elling 221 (2010) 2153–2166

1. is able to provide the user with information at the level that isdemanded for integrated assessment of EU policies,

2. simulates processes at their appropriate scale and resolution toensure a (scientifically) correct representation of processes, and

3. makes use of existing databases (and process knowledge) at EUlevel.

Conflicts can easily arise between the first two points, especiallygiven the current availability of data and knowledge. Rather thanfavouring policy needs over scientific rigour or vice versa, in thedevelopment of LUMOCAP we aimed to find a balance betweenboth.

To be able to do this we decided to include a number of modelsoperating at the spatial level they can best represent their subse-quent processes. For this reason agricultural economics is modelledat a high level of aggregation, while local dynamics are modelled ata 1 km grid resolution and competition between regions and coun-tries is modelled at NUTS 2 and Member State level, respectively.Although this approach enabled the inclusion of a wide range ofmodels, it also resulted in a discussion on the appropriate level tomodel certain processes. For some processes, like local dynamicsand interaction between regions or countries, it is fairly easy toselect the appropriate scale. For other processes like crop choice itis much more difficult. On the one hand farmers decide upon thecrops they plant, which requires a model operating at the (local)farm level. On the other hand, macro-processes (CAP, GDP, worldmarkets) also play an important role, which requires a model at ahigher spatial level. In LUMOCAP, the decision to work top-downwas to a large extent driven by data availability, in particular bythe lack of detailed spatial data required for a bottom-up approach,since it was decided to work at the scale of the entire EU. Althoughthe disaggregation mechanism used seems to work rather well, itwould be worthwhile to investigate if applying the agricultural eco-nomic model at Member State or NUTS 2 level would give improvedresults. Most of the data is available at Member State level, so thiswould in any case be feasible.

Calibration and testing of the LUMOCAP system was carried outbased on a combination of assessment methods at different spatialscales. Its accuracy was assessed using RMSE, squared error andKappa statistics, its temporal behaviour through visual interpre-tation and the spatial patterns through the clumpiness index, thecluster size–frequency distribution and visual interpretation. Sinceno assessment method is able to capture all important aspects, weargue that a combination of methods is the appropriate way to eval-uate this kind of integrated model. Using several methods of whichsome are qualitative and others quantitative raises the questionhow to combine these methods and which method to prefer whenresults are conflicting. We have tried to find a balance betweenaccuracy and spatial patterns, but – in case of conflict – we havechosen to focus more on simulating correct patterns, since we feltthat this would contribute most to the overall aim of the system:analysing the impact of the CAP on the European landscape.

One of the difficulties experienced in the calibration of theLUMOCAP system is the lack of long time series of consistent datasets. Since land use changes occur over relatively long periods,deriving relations of land use change based on a 10-year period isa real challenge. However, the better the land use model is able toactually capture the process of land use change, the more the datacan be used for fine-tuning the parameters (rather than for under-standing the process of land use change). In case data is missing orunreliable, parameters can be fine-tuned with process knowledge.

parameters. The only model that depends heavily on historic datais the agricultural economic model, since it is using econometricestimation techniques to obtain the coefficients in the equations.

H. van Delden et al. / Ecological Modelling 221 (2010) 2153–2166 2165

F nd 20l catesi to th

Nuitp

6

iuessfc

ppdits

s

ig. 8. Cut-out of the land use map for the London–Birmingham area for 2000 (a) aocations where residential development was found in 2000 and 2030; blue indinterpretation of the references to colour in this figure legend, the reader is referred

onetheless, also for this model knowledge from literature wassed when data was missing, unreliable or inappropriate for cal-

brating a forecasting model (because of the structural changeshat occurred in NMS-10 and NMS-2 during the historic calibrationeriod).

. Conclusion and recommendations

The LUMOCAP PSS is a modular, transparent, PC-based, analyt-cal system, enabling a user to interactively select policy optionsnder a specific set of natural and socio-economic conditions asxternal driving forces. The user can formulate potential land usecenarios and assess their impacts on the quality of rural land-capes through selected indicators. Consequently, the modellingramework developed enables identifying areas of adverse land usehanges caused by non-sustainable agricultural systems.

To capture the interrelation between different disciplines androcesses operating on various spatial scales, emphasis has beenut on the interaction between the different model componentsuring each time step. LUMOCAP is thus able to show that a pol-

cy measure at one scale and aimed at one particular sector, will –hrough positive and negative feedback – have its effect at othercales as well as other sectors.

The LUMOCAP system is able to outperform neutral models inimulating an historic period (1990–2000) and shows a plausi-

30 (b). (c) The differences in residential area between both maps. Green indicatesthe areas that have become residential during the course of the simulation. (Fore web version of the article.)

ble behaviour for future scenarios (2000–2030). Nonetheless, it isimportant to stress that at this stage only an initial calibration hasbeen carried out and results could be improved when more timewould be dedicated to this task. Other factors that would contributeto better representing real-world processes are a larger EU-widedatabase with good quality (spatial) data and model improvementssuch as mentioned in the discussion.

The approach followed in the calibration to start with the indi-vidual models and gradually build up to the integrated model hasproven very successful. Breaking the calibration up in small piecesmakes it easier to understand the observed behaviour and to under-stand what parameters have to be changed.

For simplification reasons, most of the models incorporated usethe same parameter settings for the whole European Union. Fora future study it would therefore be interesting to investigate towhat extent calibration parts of the model for countries or groupsof countries would improve the model behaviour and calibration.

Acknowledgements

The authors wish to thank the European Commission forthe financial support of the LUMOCAP project (contract num-ber SSPE-CT-2005-006556). The authors are solely responsiblefor the content of the paper. The views expressed are purelythose of the authors and may not in any circumstances

2 al Mod

bC

R

A

A

B

B

C

C

C

C

C

C

D

E

E

E

E

EE

E

E

E

EEE

E

E

E

FF

F

F

White, R., 1978. The simulation of central place dynamics: two sector systems andthe rank-size distribution. Geographical Analysis 10, 201–208.

166 H. van Delden et al. / Ecologic

e regarded as stating an official position of the Europeanommission.

eferences

GMEMOD, 2006. Impact Analysis of the CAP Reform on Main Agricultural Com-modities. Report 4, Baseline and Scenario Results of Country and EU 25/27Aggregate Models, AGMEMOD Partnership.

llen, R.G., Pereira, L.S., Raes, D., Smith, M., 2000. Crop evapotranspiration. FAOIrrigation and Drainage. Paper No. 56. FAO, Rome, Italy, pp. 163–165.

arredo, J., Kasanko, M., McCormick, N., Lavalle, C., 2003. Modelling dynamic spa-tial processes: simulation of urban future scenarios through cellular automata.Landscape and Urban Planning 64, 145–160.

ritz, W., Witzke, P. (Eds.), 2008. CAPRI Model Documentation 2008: Version 2.Available on http://www.capri-model.org/docs/capri documentation.pdf.

EC, 2009. Towards a better targeting of the aid to farmers in areas with naturalhandicaps. COM, 161 final.

iaian, P., 2007a. Parametric Rules for CAP Induced Land Use Dynamics. LUMOCAPdeliverable report, Leuven.

iaian, P., 2007b. Land use changes in the EU: policy and macro impact analysis.Agricultural Economics 53 (12), 565–579.

hantreuil, F., Levert, F., Hanrahan, K., 2005. The Luxembourg Agreement Reform ofthe CAP: An Analysis using the AGMEMOD Composite Model. Paper presentedat the 89th EAAE Seminar. Parma, Italy.

hoi, J.S., Helmberger, P.G., 1993. How sensitive are crop yields to price changes andfarm programs? Journal of Agricultural and Applied Economics 25, 237–244.

onforti, P., Londero, P., 2001. AGLINK: The OECD Partial Equilibrium Model. Work-ing Paper No. 8. National Institute of Agricultural Economics, Italy.

e Wit, C.T., 1958. Transpiration and Crop Yields. Agric. Res. Report, 64.6, PUDOC,Wageningen.

C, 1999. Council Regulation 1259/1999 on Establishing Common Rules for DirectSupport Schemes under the Common Agricultural Policy, 17.05.1999.

C, 2000. Decision 1445/2000/EC of the European Parliament and of the Council onthe Application of Area-frame Survey and Remote-sensing Techniques to theAgricultural Statistics for 1999 to 2003, 22.05.2000.

C, 2005. Council Regulation 1290/2005 of 21 June 2005 on the Financing of theCommon Agricultural Policy, OJ L 209, 11.8.2005, pp. 1–25.

C, 2006a. Council Decision 2006/144/EC of 20 February 2006 on Community Strate-gic Guidelines for Rural Development (Programming Period 2007 to 2013), OJ L55, 25.2.2006, pp. 20–29.

C, 2006b. Website: GISCO. Acquired 15.6.2006: http://epp.eurostat.ec.europa.eu.C, 2007a. Website: Agriculture and Rural Development. Acquired on 30.4.2007:

http://ec.europa.eu/agriculture/.C, 2007b. Implementation of Direct Payments in the Member States. European

Commission Directorate-General for Agriculture and Rural Development.C, 2007c. Website: Eurostat. Acquired on 30.4.2007: http://epp.eurostat.

ec.europa.eu.EA, 2005. Agriculture and Environment in the EU-15—The IRENA Indicator Report.

Office for Official Publications of the European Communities, Luxembourg.EA, 2007a. CORINE Land Cover 1990, 100m raster. EEA, Copenhagen.EA, 2007b. CORINE Land Cover 2000, 100m raster. EEA, Copenhagen.EC, 1979. Council Directive 79/409/EEC of 2 April 1979 on the Conservation of Wild

Birds.EC, 1992. Council Directive 92/43/EEC of 21 May 1992 on the Conservation of

Natural Habitats and of Wild Fauna and Flora.ngelen, G., White, R., De Nijs, T., 2003. Environment explorer: spatial support sys-

tem for the integrated assessment of socio-economic and environmental policiesin the Netherlands. Integrated Assessment 4 (2), 97–105.

rjavec, E., Donnellan, T., 2005. Development of the AG-MEMOD Country Level Agri-cultural Policy Analysis Tool in the New Members States of EU. Paper presentedat the 89th EAAE Seminar. Parma, Italy.

AO, 2007. Website: Faostat. Acquired 30.4.2007: http://faostat.fao.org.APRI, 2007a. Website: Food and Agricultural Policy Research Institute

(FAPRI). U.S. and World Agricultural Outlook. Acquired 30.4.2007:http://www.fapri.iastate.edu.

APRI, 2007b. Website: Food and Agricultural Policy Research Institute

(FAPRI). The FAPRI Searchable Elasticity Database. Acquired 30.4.2007:http://www.fapri.iastate.edu.

arrington, J.H., Kuhlman, T., Rothman, D.S., Imrichova, Z., Reid, L., Konkoly Gyuró,E., 2008. Reflections on social and economic indicators for land use change. In:Helming, K., Perez-Soba, M., Tabbush, P. (Eds.), Sustainability Impact Assessmentof Land Use Changes. Springer.

elling 221 (2010) 2153–2166

Guyomard, H., Baudry, M., Carpentier, A., 1996. Estimating crop supply response inthe presence of farm programmes: application to the CAP. European Review ofAgricultural Economics 23, 401–420.

Hagen-Zanker, A., Lajoie, G., 2008. Neutral models of landscape change as bench-marks in the assessment of model performance. Landscape and Urban Planning86, 284–296.

Hurkens, J., Hahn, B.M., Van Delden, H., 2008. Using the Geonamica® softwareenvironment for integrated dynamic spatial modelling. In: Sànchez-Marrè, M.,Béjar, J., Comas, J., Rizzoli, A., Guariso, G. (Eds.), Proceedings of the iEMSsFourth Biennial Meeting: “Integrating Sciences and Information Technology forEnvironmental Assessment and Decision Making”. International EnvironmentalModelling and Software Society, Barcelona, Spain.

IPCC, 2007. Website: IPCC Data Distribution Centre. Acquired 20.2.2007:http://www.ipcc-data.org.

Lambert, J.J., Daroussin, J., Eimberck, M., Le Bas, C., Jamagne, M., King, D., Mon-tanarella, L., 2003. Soil Geographical Database for Eurasia & The Mediterranean:Instructions Guide for Elaboration at scale 1:1.000.000. Version 4.0. EUR 20422EN. Office for Official Publications of the European Communities, Luxembourg.

Lekakis, J.N., Pantzios, C., 1999. Agricultural liberalization and the environmentin Southern Europe: the role of the supply side. Applied Economics Letters 6,453–458.

LUMOCAP, 2009. Final Report of the LUMOCAP Project—Dynamic Land Use ChangeModelling for CAP Impact Assessment on the Rural Landscape. Available onhttp://www.riks.nl/projects/LUMOCAP.

Lopatka, A., Stuczynski, T., 2007. A simple approach to land suitability assessmentfor delineation on LFA in the EU. In: Proceedings of Workshop on Common Bio-physical Criteria for Defining Areas Which are Less Favourable for Agriculturein Europe. Joint Research Centre Ispra, April 19–20.

McGarigal, K., Cushman, S.A., Neel, M.C., Ene, E., 2002. FRAGSTATS: Spatial PatternAnalysis Program for Categorical Maps. Computer Software Program Producedby the Authors at the University of Massachusetts, Amherst, Available onwww.umass.edu/landeco/research/fragstats/fragstats.html.

Monserud, R.A., Leemans, R., 1992. Comparing global vegetation maps with theKappa statistic. Ecological Modelling 62, 275–293.

Moro, D., Sckokai, P., 1999. Modelling the CAP arable crop regime in Italy: degree ofdecoupling and impact of agenda 2000. Cahiers D’économie et Sociologie Rurales53, 49–73.

OECD, 2003. Agricultural Policies in Transition Economies. OECD, Paris.OECD, 2004. Analysis of the 2003 CAP Reform. OECD, Paris.OECD, 2007. Website: OECD. Acquired 30.4.2007: http://webnet.oecd.org.Petit, S., Finn, P.V., Verkerk, P.J., Firbank, L.G., Halberg, N., Dalgaard, T., Kjeldsen, C.,

Lindner, M., Zudin, S., 2008. Indicators for assessing the environmental impactsof land use change across Europe. In: Helming, K., Perez-Soba, M., Tabbush, P.(Eds.), Sustainability Impact Assessment of Land Use Changes. Springer.

RIKS, 2009. Metronamica - Model descriptions. RIKS, Maastricht, The Netherlands.Sieber, S., Müller, M., Verweij, P., Haraldsson, H., Fricke, K., Pacini, C., Tscherning, K.,

Helming, K., Jansson, T., 2008. Transfer into decision support: The Sustainabil-ity Impact Assessment Tool (SIAT). In: Helming, K., Perez-Soba, M., Tabbush, P.(Eds.), Sustainability Impact Assessment of Land Use Changes. Springer.

Thornthwaite, C.W., Mather, J.R., 1955. The water balance. Climatology VIII (1).Van Delden, H., Gutiérrez, E.R., Van Vliet, J., Hurkens, J., 2008. Xplorah, a multi-scale

integrated land use model. In: Sànchez-Marrè, M., Béjar, J., Comas, J., Rizzoli, A.,Guariso, G. (Eds.), Proceedings of the iEMSs Fourth Biennial Meeting: “Integrat-ing Sciences and Information Technology for Environmental Assessment andDecision Making”. International Environmental Modelling and Software Society,Barcelona, Spain.

Van Ittersum, M.K., Ewert, F., Heckelei, T., Wery, J., Alkan Olsson, J., Andersen, E.,Bezlepkina, I., Brouwer, F., Donatelli, M., Flichman, G., Olsson, L., Rizzoli, A.,Van der Wal, T., Wien, J.E., Wolf, J., 2008. Integrated assessment of agriculturalsystems—a component-based framework for the European Union (SEAMLESS).Agricultural Systems 96, 150–165.

WUR, MNP, 2007. Eururalis 2.0 CDrom. Alterra Wageningen UR, Wageningen, TheNetherlands.

White, R., 1977. Dynamic central place theory: results of a simulation approach.Geographical Analysis 9, 227–243.

White, R., 2006. Pattern based map comparisons. Geographical Systems 8, 145–164.White, R., Engelen, G., 1993. Cellular automata and fractal urban form: a cellular

modelling approach to the evolution of urban land use patterns. Environmentand Planning A 25 (8), 1175–1199.