are agricultural land-use models able to predict changes in land-use intensity?

Agriculture, Ecosystems and Environment 82 (2000) 321–331

Are agricultural land-use models able to predictchanges in land-use intensity?

E.F. Lambin∗, M.D.A. Rounsevell, H.J. GeistDepartment of Geography, University of Louvain, Place Louis Pasteur 3, B-1348 Louvain-la-Neuve, Belgium

Abstract

Land-use and land-cover change research needs to pay more attention to processes of land-cover modification, and es-pecially to agricultural land intensification. The objective of this paper is to review the different modelling approaches thathave been used in land-use/land-cover change research from the perspective of their utility for the study and prediction ofchanges in land-use intensification. After clarifying the main concepts used, the different modelling approaches that havebeen used to study land-use change are examined, case study evidence on processes and drivers of land-use intensificationare discussed, and a conclusion is provided on the present ability to predict changes in land-use intensity. The analysis sug-gests there are differences in the capability of different modelling approaches to assess changing levels of intensification:dynamic, process-based simulation models appear to be better suited to predict changes in land-use intensity than empirical,stochastic or static optimisation models. However, some stochastic and optimisation methods may be useful in describing thedecision-making processes that drive land management. Case study evidence highlight the uncertainties and surprises inherentin the processes of land-use intensification. This can both inform model development and reveal a wider range of possiblefutures than is evident from modelling alone. Case studies also highlight the importance of decision-making by land managerswhen facing a range of response options. Thus, the ability to model decision-making processes is probably more important inland-use intensification studies then the broad category of model used. For this reason, landscape change models operatingat an aggregated level have not been used to predict intensification. In the future, an integrated approach to modelling — thatis multidisciplinary and cross-sectoral combining elements of different modelling techniques — will probably best serve theobjective of improving understanding of land-use change processes including intensification. This is because intensification isa function of the management of physical resources, within the context of the prevailing social and economic drivers. Some ofthe factors that should be considered when developing future land-use change models are: the geographic and socio-economiccontext of a particular study, the spatial scale and its influence on the modelling approach, temporal issues such as dynamicversus equilibrium models, thresholds and surprises associated with rapid changes, and system feedbacks. In industrialisedregions, predicting land-use intensification requires a better handling of the links between the agriculture and forestry sectorsto the energy sector, of technological innovation, and of the impact of agri-environment policies. For developing countries,better representation of urbanisation and its various impacts on land-use changes at rural-urban interfaces, of transport infras-tructure and market change will be required. Given the impossibility of specific predictions of these driving forces, most ofthe modelling work will be aimed at scenario analysis. © 2000 Elsevier Science B.V. All rights reserved.

Keywords:Land use; Land-use change; Agricultural intensification; Land cover

∗ Corresponding author. Tel.:+32-1047-4477; fax:+32-1047-2877.E-mail address:[email protected] (E.F. Lambin).

0167-8809/00/$ – see front matter © 2000 Elsevier Science B.V. All rights reserved.PII: S0167-8809(00)00235-8

322 E.F. Lambin et al. / Agriculture, Ecosystems and Environment 82 (2000) 321–331

1. Introduction

Land-use and land-cover change, as one of the maindriving forces of global environmental change, iscentral to the sustainable development debate. Land-use and land-cover changes have impacts on a widerange of environmental and landscape attributes in-cluding the quality of water, land and air resources,ecosystem processes and function, and the climatesystem itself through greenhouse gas fluxes and sur-face albedo effects. Whilst, a few years ago, mostland-use and land-cover change research was fo-cused on land-cover conversions (e.g., deforestation,urbanisation), researchers have increasingly realisedthat more subtle processes leading to a modificationof land cover deserves greater attention. Land-covermodification is frequently caused by changes in themanagement of agricultural land use including, e.g.,changes in levels of inputs and the effect of this onprofitability or the periodicity of complex land-usetrajectories, e.g., fallow cycles, rotation systems orsecondary forest regrowth.

Agricultural land intensification has been one ofthe most significant forms of land-cover modifica-tion, with dramatic increases in yields being the mainfeature during the previous 30 years. Yields of foodcrops (per area of land) have outpaced global humanpopulation growth (Matson et al., 1997), but if currenttrends are extrapolated linearly into the future, inten-sification of agriculture will have major detrimentalimpacts on non-agricultural terrestrial and aquaticecosystems (Tilman, 1999; Socolow, 1999). Intensifi-cation levels can also be an indicator of the ability ofland-use systems to adapt to changing circumstances,e.g., because of policy or climate change. For exam-ple, many extensive land-use systems are marginal inproductivity terms (e.g., uplands, semi-arid regions,high latitude areas, etc.) and these types of land usesoften have little capacity to adapt. This does not fol-low, however, where extensive land use is a resultof deliberate policy constraints on land that is notmarginal in productivity terms. In this paper, it is con-tended, therefore, that land-use change research wouldbenefit from a better understanding of the complexrelationships between people and their managementof land resources, and that land-use intensification isa vital consideration in these processes. This impliesthat, to fully understand and predict human impacts

on terrestrial ecosystems, there is a need for morecomprehensive theories of land-use change (Lambin,1997, p. 389; Lambin et al., 1999, pp. 37–46, 89).

Much land-use/land-cover change research has beenbased on the use of models. Modelling, especially ifdone in a spatially-explicit, integrated and multi-scalemanner, is an important technique for the explorationof alternative pathways into the future, for conduct-ing experiments that test our understanding of keyprocesses, and for describing the latter in quantitativeterms. Many different modelling approaches have beenadopted in the study of land-use/land-cover change,although most have been concerned with issues ofland use conversion (Lambin, 1997; Kaimowitz andAngelsen, 1998). Few modelling studies have explic-itly sought to evaluate potential changes in land-useintensification resulting from changes in management.Note, however, that economists have a long traditionin studying agricultural intensification in relation tomanagement practices and conditions (e.g., prices ofinputs, production functions). However, most studieshave not revealed the driving factors, apart from eco-nomic incentives, that cause management to change.In this paper, the different modelling approaches thathave been used in land-use/land-cover change researchwere examined from the perspective of their utility forthe study and prediction of changes in land-use inten-sification. After clarifying the main concepts used, thedifferent modelling approaches that have been used tostudy land-use changes are examined, case study evi-dence on processes and drivers of land-use intensifica-tion are discussed, and a conclusion is provided on thepresent ability to predict changes in land-use intensity.

2. Definitions

It is important to clarify terminology and definitionsused in land-use/land-cover change research. Such ter-minology can be esoteric and, thus, affect the under-standing of land-use/land-cover change research by abroad readership. The term land cover refers to theattributes of a part of the Earth’s land surface and im-mediate subsurface, including biota, soil, topography,surface and groundwater, and human structures. Landuse refers to the purposes for which humans exploitthe land cover. Forest, e.g., is a type of land cover thatis dominated by woody species and may be exploited

E.F. Lambin et al. / Agriculture, Ecosystems and Environment 82 (2000) 321–331 323

for land uses as varied as recreation, timber productionor wildlife conservation. Furthermore, one may dis-tinguish between land-cover conversion, i.e., the com-plete replacement of one cover type by another, andland-cover modification, i.e., more subtle changes thataffect the character of the land cover without changingits overall classification (Turner et al., 1993).

The concept of intensification in land use often,although not exclusively, refers to agriculture. Agri-cultural intensification has been defined by Brookfield(1972) as the substitution of inputs of capital, labourand skills for land, so as to gain more production froma given area, use it more frequently, and hence makepossible a greater concentration of production. Inten-sity is usually measured in terms of output per unitof land or, as a surrogate, input variables against con-stant land (Turner and Doolittle, 1978). Thus, one candistinguish between input intensification, which mea-sures the increases in input variables, e.g., chemicalfertiliser, pesticides, etc., and output intensification,which measures the increases in production againstconstant units of land area and time, e.g., food-tonnesor number of calories/hectare/number of years (Turnerand Doolittle, 1978, p. 298). Because of data prob-lems, surrogate measures are often employed (Kateset al., 1993, p. 12). Regardless of the measures andvariables applied, any finding will vary from whatis typically thought of as intensification since meth-ods used in data gathering and analysis will stronglyinfluence the relationship between variables. Conven-tional methods of measurement are, e.g., frequency ofcultivation or number of harvests per plot over a stan-dard time frame (after Boserup, 1965) as comparedto, e.g., farm produce-generated income per hectareas a reflection of yields per hectare (Dorsey, 1999,p. 187).

Beginning at least with von Thünen in 1842, agri-cultural intensity (viewed in terms of production oryield per unit area and time) has long been regardedas a key concept in numerous explanations of agricul-tural growth and change (Turner et al., 1977; Turnerand Doolittle, 1978). It had been pointed out by Kateset al. (1993, p. 21) that long-term population growthand economic development usually do not take placewithout intensification and agricultural growth, al-though intensification and agricultural growth donot inevitably follow population growth and are notnecessarily beneficial or sustainable (see Mortimore

(1993) for a review and a discussion of the theoriesof Boserup and neo-Malthusians).

3. Categories of land-use models

The modelling of land-cover change processesshould aim to address at least one of the followingquestions:1. Which environmental and cultural variables con-

tribute most to an explanation of land-coverchanges — why?

2. Which locations are affected by land-cover changes— where?

3. At what rate do land-cover changes progress —when?

Table 1 shows the four broad categories of land-usechange models that may be used to address differentquestions, but which all require a different set of pre-liminary information: empirical–statistical, stochastic,optimisation and dynamic (process-based) simulation.

3.1. Empirical–statistical models

Empirical, statistical models attempt to identifyexplicitly the causes of land-cover changes using mul-tivariate analyses of possible exogenous contributionsto empirically-derived rates of changes. Multiple lin-ear regression techniques are generally used for thispurpose. The finding of a statistically significant asso-ciation does not establish a causal relationship. More-over, a regression model that fits well in the regionof the variable space corresponding to the originaldata can perform poorly outside that region. Thus,regression models cannot be used for wide rangingextrapolations. Such models are only able to predictpatterns of land-use changes which are represented inthe calibration data set. Thus, these models are onlysuited to predict changes in land-use intensity wheresuch changes have been measured over the recentpast: in most studies this assumption is not valid.Note however that most empirical–statistical modelsare based on cross-sectional analysis of a series offarms, districts or counties. Because spatial variabilityin land-use systems is sometimes large, there will be,in some cases, empirical evidence of intensification.So, the derived regression model could be used to“predict” intensification of the observations lagging


Table 1Categories of land-use change models

What is alreadyknown on LUCC

What one needs toknow on LUCC

Modelcategory

Modellingapproach

Where and whenin the past

When in the future (short-term) Stochastic Transition probability models

Why in the past (proximate causes) Empirical, statistical Multivariate statistical modellingWhere in the future (short-term) Spatial statistical (GIS-based) models

Where, when andwhy in the past

When in the future (long-term) Process-based, mechanistic Behavioural models anddynamic simulation models

When and where in the future(long-term)

Dynamic spatial simulation models

Why in the future (underlying causes) Analytical, agent-based,economic

Generalised von Thünen models

Why in the future (underlyingcauses; scenarios)

Deterministic and stochasticoptimisation models

behind in the intensification process. Problems ofcausality do of course remain.

3.2. Stochastic models

Stochastic models which, for land-use change, con-sist mainly of transition probability models, describestochastically, processes that move in a sequence ofsteps through a set of states. For land-use change, thestates of the system are defined as the amount of landcovered by various land uses. The transition probabil-ities can be statistically estimated from a sample oftransitions occurring during some time interval. Proba-bilities of transitions are defined for changes from oneland-cover category to another. Transition probabilityapproaches are limited in their application to the ques-tion of land-use intensification because they only usetransitions which have been observed in the recent past— which is similar to empirical–statistical models.However, some other forms of stochastic models, suchas spatial diffusion models, do appear to be useful inresearch on intensification. Hägerstrand (1968) devel-oped stochastic approaches based on Monte Carlo sim-ulation to explain the diffusion of innovation throughSwedish farming communities. The way in which in-formation on management options moves through alandscape must be an important process in understand-ing intensification. Thornton and Jones (1998) recentlypresented a, so far, purely conceptual model of agri-cultural land-use dynamics, based on Markov chainsgoverned by a few simple decision rules. They statethat the construction of this top–down model will be

used in further development to interpret possible eco-logical consequences of changes in input conditionson the landscape. This could lead to the derivationof some simple indices or measures of potential eco-logical impact of technological and economic changeon agricultural land-use, which could be of value ina range of impact assessment studies (Thornton andJones, 1998, p. 519).

3.3. Optimisation models

In economics, many models of land-use changeapply optimisation techniques based either onwhole-farm analyses using linear programming, at themicroeconomic level, or general equilibrium models,at the macroeconomic scale (Kaimowitz and An-gelsen, 1998). Many of these approaches originatefrom the land rent theory of von Thünen and that ofRicardo. Any parcel of land, given its attributes andits location, is modelled as being used in the way thatearns the highest rent. Such models allow investiga-tion of the influence of various policy measures onland allocation choices. However, models of urban andperiurban land allocation appear to be much more de-veloped than their rural counterparts (Riebsame et al.,1994). The agricultural land rent theory of von Thü-nen (1966) does not primarily concern the dynamicprocess of intensification. It rather explains optimalcrop production allocation following degrees of agri-cultural intensity. In this theory, agricultural systemsare found to be centred around a single, “isolated”market place in the form of land-use intensity rings.


However, the central-peripheral concept of decreasesin land-use intensity due to land rent differenceshas been and still is pervasive in modelling. A lim-ited number of variables used in the basic, mainlystatic and deterministic model of agricultural systemsexplains intensity as depending on the achievableeconomic rent. This rent is determined by marketdemands in the consumer centre, transportation costs,production costs and degrees of perishability of goodsproduced for the central market. By adopting themodel to the case of rotational agriculture, e.g., rota-tional time (fallow length) was shown to be inverselylinked to system parameters as given above. Price ex-pectations and interest rates have also been shown tobe important factors. Thus, the results are consistentwith the Boserupian contention about a link betweenpopulation density and degree of agricultural inten-sity while extending the argument to the effects ofthe market economy on agricultural intensity (Jonesand O’Neill, 1993, 1994). However, the von Thünenmodel is only partly suitable for prediction in a re-vised form, e.g., including more variables such asthe actual transportation network (Alonso, 1964) andunder certain conditions. Such conditions are that de-mand for products exceeds supply, that transport costsare crucial part of overall costs, and that the model isapplied in urban hinterland and isolated commercialmarket/settlement areas of today’s developing worldor in pre-industrial societies of today’s developedworld (Peet, 1969). These restrictions are mainly dueto market changes and transportation infrastructurewhich, due to their highly political economy nature,are difficult to predict in the long-term and thus limitpredictive power (Guyer and Lambin, 1993).

Optimisation models suffer from other limitations,such as the somewhat arbitrary definition of objectivefunctions and non-optimal behaviour of people, e.g.,due to differences in values, attitudes and cultures.While, at an aggregated level, these limitations arelikely to be non-significant, they are more importantas one looks at fine scale land-use change processesand is interested in the diversity between actors.

3.4. Dynamic (process-based) simulation models

Patterns of land-cover changes in time and spaceare produced by the interaction of biophysical andsocio-economic processes. Dynamic (process-based)

simulation models have been developed to imitatethe run of these processes and follow their evolution.Simulation models emphasise the interactions amongall components forming a system. They condense andaggregate complex ecosystems into a small number ofdifferential equations in a stylised manner. Simulationmodels are therefore based on an a priori understand-ing of the forces driving changes in a system. In thecase of agricultural intensification, this understand-ing is rooted in the models of Boserup (1965, 1975,1981) and Chayanov (1966). These models werepowerful stimuli to researchers dealing with agricul-tural change beyond the confines of economics. Thelargely population-driven approaches of a Boserupianor Chayanovian type of economy relate agriculturalintensification to household needs and wants (am-plifying into a consumption-based “needs theory”)according to which households balance consump-tion and leisure (least-effort means) and/or follow asubsistence target or strategy (“full belly” approach).

Boserup’s intensification model touches thelong-run processes of intensification of cultivationas driven by population growth and farmers’ as-sumed preferences for leisure. It measures increasesin frequency of cultivation against constant land andtime period and defines a “continuum of agriculturalintensity”. The model is mainly applicable in subsis-tence economies and is valid for the broader sweep ofagrarian change history rather than for individual, lo-cal cases. The model is based on a rather mechanisticview and is, therefore, hardly suitable for prediction(Lambin, 1994, p. 66, 71). It constitutes a largelyverbal and not spatially explicit model that was notdesigned for numerical prediction — despite attemptsto formalise it mathematically at a macro-scale level(Turner et al., 1977). In Chayanov’s “theory of peasanteconomy”, it is argued that (Russian) farm householdsdid not seek to produce as much as was possible (inprofit maximisation), but sought a more restrained andless elastic goal to provision the household. In otherwords, models of the Chayanovian type refer to situ-ations where agricultural households maximise utilitywith a trade-off between consumption and leisure,and where farming units are seen not to be integratedinto “perfect” markets, i.e., no off-farm labour isassumed.

Different from population approaches that assumesubsistence behaviour and limited market integration,


more realistic theorising on longer-term agriculturalintensification integrates demands from the market.In a numerical simulation model, Angelsen (1999,pp. 197–201, 204–206) considers land-use intensityas one of 12 variables besides exogenously givenprices and a newly introduced labour market — inthat any amount of labour can be sold or hired withmigrational flows in and out of the local economy.Within market-based theorising, the dominant viewsused to be based on a purely market or commodityapproach. This postulates that farmers accept com-modity production and respond to market demandswithin constraints placed upon them by maximisingproduction to the level of maximum reward. The dom-inant view has now shifted to the theory of “inducedinnovation”. Here, risk aversion behaviour, in theform of maintenance of some minimum productionfor subsistence needs, has gained momentum towardshardly achievable full rational economic efficiency inresource allocation. Kates et al. (1993, pp. 9 and 10)summarised that “induced innovation” implies thattechnological and institutional changes required fordevelopments in agriculture are endogenously derivedas a result of changes in resource endowments anddemand. Numerous studies (Binswanger and Ruttan,1978; Hayami and Ruttan, 1985; Pender, 1998) havefocused since then on agricultural innovation andtechnological change. These factors were consideredcollectively, e.g., by Goldman (1993), Tiffen and Mor-timore (1994), and Tiffen et al. (1994). From thesestudies, agricultural change emerges as an extremelycomplex process in the course of which opportuni-ties are as significant as constraints, and the sourcesof innovation by farmers are multiple (Goldman,1993, p. 68). While the above studies did not resultin dynamic simulation models per se, they allow fora more realistic representation of processes of agri-cultural intensification in broader simulation modelsof land-use change (e.g., Stephenne and Lambin,2001).

The scale issue is difficult to deal with in dynamicsimulation models. Process-based models can beparametrised based on local observations of decision-making. These relations can, however, not be used ina straightforward way to model aggregate behaviour.At the landscape level, behaviour is more complexgiven the numerous interactions between actors andwith the biophysical environment (Ahn et al., 1998).

3.5. Integrated modelling approaches

Whilst the previous discussion provides a tidy clas-sification of the various types of models that have beenused in land-use change research, newer approachesare increasingly based on combining elements ofthese different modelling techniques. In principal, thebest elements are combined in ways that are mostappropriate in answering specific questions. Thesetypes of models are increasingly referred to as inte-grated models, although in many cases they are betterdescribed as hybrid models because the level of in-tegration is not always high. Wassenaar et al. (1999),e.g., demonstrated how a dynamic, process-based cropmodel could be applied at the regional scale throughthe derivation of statistical relationships between themodelled crop productivity outputs and easily mappedsoil parameters. New statistical relationships are de-rived each time the dynamic model is run, so avoidingthe problem of the limited applicability (or transfer-ability) of statistical functions. Such approaches areespecially useful where the types of spatially-variabledata required by a dynamic model are lacking at theregional scale, but which may be available at welldocumented sites. White et al. (1997) demonstratedthe use of a land-use change model that combined astochastic, cellular automata approach with dynamicsystems models of regional economics. The approachallows spatially-explicit geographic processes to beconstrained by less spatially-precise economic pro-cesses within the framework of a Geographic Infor-mation System (GIS). The approach has been usedas a decision support system, by allowing regionalland-use planners to investigate the consequences ofalternative management strategies.

The combination of dynamic, process-based mod-els with optimisation techniques underpinned thedevelopment of the Integrated Model to PredictEuropean Land Use (IMPEL; Rounsevell et al., 1998).The requirement for IMPEL was to be able to assessmodifications to the spatial distribution of agriculturalland-use in response to climate change. This requireda decision-maker (optimisation) orientated approachthat was also able to deal explicitly with the impactof climate change on the biophysical components offarming systems (e.g., crop productivity) through theuse of dynamic simulation. Whilst, the use of suchintegrated modelling approaches seem to provide


useful insights into complex land-use systems, oneneeds to be aware that such models are no longerwithin the domain of individual researchers, but areincreasingly developed within the framework of large,multidisciplinary research teams. This trend awayfrom the simple approaches proposed in earlier mod-els towards increasing size and complexity is perhapsbest exemplified by a group of models commonlytermed Integrated Assessment Models, e.g., IMAGE(Alcamo et al., 1998). Usually operating at globalscales, although still relatively finely resolved in geo-graphic space, models such as IMAGE seek to inte-grate a wide range of sectors and process descriptions.Integrated Assessment Models are, thus, often wellplaced to enhance understanding of the consequencesof intensification, although seldom able to model themanagement and decision-making processes of indi-viduals. This stems from their requirement to operateat global scales which often results in simplistic treat-ment of intensification processes. It is important tonote, however, that such models were not designedto assess intensification, but are concerned more withbroad scale changes in land cover.

In the context of developing countries, increasingefforts, aimed at integrating various model elements(e.g., guided by the major population and market the-ories) are obvious from even predominantly economicmodelling. An increasing awareness of the signifi-cance of spatial explicitness (or regional situations,at least) is also emerging. This is obvious, e.g., fromsome recent modelling of subsistence agriculture atthe forest frontier in the developing world. Angelsen(1994) discusses how variables of the populationmodels such as fallow period and labour input canbe integrated in the open economy models. As anoutcome of comparing the modelling approaches, An-gelsen (1994) suggested consideration of input pricedevelopments (e.g., price of fertilisers), risk factorsand crop choices, but also the recognition of differ-ences between settled (intensive margin) and frontier(extensive margin) agriculture when evaluating theimpact of technological change (Walker, 1999).

4. Case study evidence

The Land-Use/Land-Cover Change (LUCC) prog-ramme (Turner et al., 1993), identified the importance

of using case study evidence to supplement modellingactivities. This leads us to ask, what can be derivedfrom case study evidence to improve our ability topredict land-use intensity? For example, what vari-ables should be included in more “realistic” mod-elling? The outcomes of two empirical case studiesaimed at testing theories and reflecting recent effortsto merge subsistence and market demands (“inducedintensification”) are presented below.

In a time series analysis between 1950 and 1986 ofthe induced intensification theory for 265 householdsand five villages representing a range of agroecolog-ical and socio-economic conditions in Bangladesh,Turner and Ali (1996) stated that the model pro-vided relatively high levels of explained variance incropping intensity. However, the model also indi-cated the relative impacts of other important variablessuch as household class, cropping strategies, envi-ronmental constraints (location and water control)and other impediments such as impoverishment, in-adequate transportation infrastructure for marketingand unfavourable state policies (e.g., no governmentassistance in regulating access to water pumps).Though, during the study period, all of the small-holder farmsteads actually produced a small surplusalong the lines of induced intensification, the resultswere achieved under conditions of increasing socialpolarisation in the course of time, with larger hold-ers accounting for surplus production and increasinglandless households suffering from production short-falls and malnutrition. Thus, adding to a broader socialor political economy understanding of land-use inten-sity changes, the authors also point to the existenceof Malthusian-like “thresholds” of intensification thatconstitute critical junctures in the process and createa series of steps or staircase-like intensification pro-cess. Thresholds may serve as major impediments tointensification and either lead to conditions of invo-lution — in the sense of production increases undersignificant declines in the marginal utility of inputs(Geertz, 1963) — or stagnation, i.e., no or decliningproduction. Thus, notwithstanding environmental andsocio-economic constraints to intensification, agricul-tural growth in Bangladesh was achieved by means ofincreasing land productivity in the form of intensifica-tion through increased irrigation (facilitating doubleand triple cropping) and the use of green revolutioninputs with some thresholds suggesting responses of


involution and stagnation. The latter were advertedfirst in the 1960s by the adoption of high yielding ricevarieties, in the 1980s by a shift to garden crops withhigh market value, and later with the removal of eco-nomic and policy barriers to irrigation technologies.Noteworthy are both the socio-economic impedimentsor “distortions” (Schultz, 1978) in the form of stateregulations or institutional structures and similarlyoperating environmental constraints (Meertens et al.,1996) to the “ideal” trajectory of induced intensifi-cation. Though the diversity of constraints will bedifficult to generalise, the point is made that suchstructures can be so significant as to mask the under-lying processes (Turner and Ali, 1996, p. 14 986). Itis concluded that it is not the process as a whole thatis less developed in conceptual terms, but more thespecifics that divert intensification into the involutionand stagnation paths.

In an evaluation of survey results from 67 smallfarms in the Kiriyaga District of Kenya between1981 and 1995, Dorsey (1999) provides findings thatexpand upon induced innovation and intensifica-tion theory supporting the thesis that — despiteclaimed negative food production trends in Africa— farmers are practising highly productive agri-cultural strategies resulting from both subsistence-and commodity-based production. This rests on themore general observation found in Nigeria by Netting(1993, p. 32) that the main strategy for combininghigh production per unit area with risk reductionand sustainability is diversification. The Kenyan casestudy takes the scale of agricultural production (i.e.,number of hectares under cultivation per farm) to ar-rive at a better understanding, and perhaps predictionof intensification and innovation under the constraintsof declining land availability. Indicators of agricul-tural intensification have been identified as a decreasein grazing land (coupled with the effects of a govern-ment “zero grazing” programme), rising percentageof available land under production, competition forand fragmentation of land (as measured by a decreasein farm size against rising population density), andwidespread capital availability constraints on furtherproduction increases. Especially, the latter is consid-ered to be a crucial indicator. The author states thatif the scale at which inputs and credit availabilitybecome constraints can be specified, a more preciseconception of induced intensification may be obtained

(Dorsey, 1999, p. 181). Strongly correlated are com-mercial specialisation and diversification. They bothhave a strong effect on net income per unit area(reflecting intensification), while the latter is largelyseen as a function of how much food productiongoes toward consumption versus sales (Dorsey, 1999,p. 193). Variations in the interconnectedness of thelatter may be explained by constraints and behaviour.Some farmers are risk averse whilst others engage inmore risk-taking behaviour. Two other conclusionsseem noteworthy. First, no significant relationship hasbeen found between farm size and intensification and,secondly, it is stated that the market demand path mayhave a greater influence on increased production thanpopulation pressure, given that decreases in holdingsize do not generally lead to declining income per unitarea (Dorsey, 1999, p. 192). For other compilationsand reviews of studies supporting induced intensifi-cation, see Brush and Turner, 1987, and Kates et al.,1993.

5. What is required to be able to predict land-useintensity?

Any model prediction can only be based on whatis currently known about processes of change. Casestudy evidence, however, can highlight the uncer-tainties and surprises inherent in the processes ofland-use intensification. This can both inform modeldevelopment and reveal a wider range of possiblefutures than is evident from modelling alone. More-over, an inductive approach to the understanding ofchange processes uses a much larger number of vari-ables than are typically represented in a model. Theunderstanding of this complexity helps in turn in de-ciding what reductions can be performed in modeldevelopment whilst maintaining the validity of themodel. Finally, case studies highlight the importanceof decision-making by land managers when facing arange of response options. External driving forces ofland-use change open new and/or close old options,but final land management decisions are made by ac-tors who are influenced by socio-cultural and politicalfactors as well as by economic calculations.

Can the question raised in the title be answered?Inevitably, perhaps, the answer is “partly”, althoughthere are clear differences in the capability of


different modelling approaches to assess changinglevels of intensification. The answer to the questionalso depends strongly on the geographic location orsocio-economic context of a particular study. It seemsthat different modelling approaches are more or lessappropriate in tackling the question of intensificationat different locations. One could anticipate, e.g., us-ing different modelling tools (or a different mix ofthese tools) when working in developing rather thandeveloped countries. In general, however, dynamicsimulation models are better suited to predict changesin land-use intensity than empirical, stochastic orstatic optimisation models, although some stochasticand optimisation methods may be useful in describ-ing the decision-making processes that drive landmanagement. Moreover, it appears that an integratedapproach to modelling (that is multidisciplinary andpossibly cross-sectoral) will probably best serve theobjective of improving understanding of land-usechange processes including intensification. This is be-cause intensification is a function of the managementof physical resources, within the context of the pre-vailing social and economic drivers. Because intensifi-cation is driven by management options, the ability topredict changes in land-use intensity requires modelsof the process of decision-making by land managers.Thus, the ability to model decision-making processesis probably more important in land-use intensificationstudies then the broad category of model used. For thisreason, most landscape change models operating atan aggregated level have not been used to predict in-tensification. However, there may be opportunities todevelop landscape scale models further that explicitlytake account of decision-making processes. Actually,issues of intensification are also very interesting —and often more policy relevant — at aggregate scales.

A trend toward integrated models is already evident,although the logistical difficulties in developing suchlarge and complex modelling approaches needs to berecognised. In part, therefore, the future developmentof such models may depend as much on the willing-ness of the organisations that fund research as theresearchers themselves because large modelling exer-cises are expensive. Whilst many integrated modelsdraw heavily on existing approaches, there may alsobe the need to develop new modelling techniquesthat address intensification issues, specifically. Thedevelopment of new techniques, however, should be

undertaken firmly within the context of the purpose ofland-use change models, which is to better understandland-use change processes. Finally, the developmentof models of land-use intensification has importantdata requirements as one needs spatially-explicitvariables on land management, input use, croppingsystem, frequency of cultivation, rotations, etc.

Within this context of current model development,there are some simple initiatives that would enhancethe ability of the land-use change modelling commu-nity to improve its understanding and representationof intensification processes. This includes the devel-opment of models that explicitly incorporate man-agement and intensification processes. This will needto consider the spatial scale and its influence on themodelling approach, temporal issues such as dynamicversus equilibrium models, thresholds and surprisesassociated with rapid changes, and system feedbacks(e.g., the interaction between changes in land manage-ment and the quality of soil and biological resources).These models should lead to the formulation of sce-narios which would provide plausible representationsof alternative futures where these are unknown, un-certain or may contain ‘surprises’. Uncertainties andunknowns might include (bio)technological change,social change, or the role of regulation, policy-makingand political change. Scenarios would provide an op-portunity to analyse the sensitivities of land-use andmanagement systems as well as allowing us to exploredifferent (sustainable) development strategies and op-tions. Within this context there is a clear need to de-velop “integrated” scenarios that recognise, in an inter-nally consistent way, that the future will be shaped bysocio-economic as well as biophysical change drivers.

6. Conclusion

The type of model needed to predict changes inland-use intensity depends on the specific researchquestion. So far, the emphasis of spatially-explicitlandscape models has been on thelocation issue —where will land-use change take place? With thesemodels, land-use intensification could be treatedequally as land-cover conversion. Therefore, for thisquestion, land-use intensification does not require afundamentally different modelling approach. How-ever, we believe that the most relevant questions


concerning intensification, e.g. for scientific under-standing, to predict environmental impacts and to sup-port policies, are related to thequantitiesof land-usechange rather than to the location issue. By quantity,we mean the amount of change that is taking place,i.e. in the case of land-cover conversion, the amountof area changed or, in the case of land-use intensifica-tion, the amount of inputs used and/or production perunit area gained or lost, as a function of managementlevel. Model requirements to generate scenarios onthe quantity of land-use intensification are more com-plex compared to models of land-cover conversion. Inindustrialised regions, predicting land-use intensifica-tion requires a better handling of the links betweenthe agriculture and forestry sectors to the energy sec-tor, of technological innovation, and of the impact ofagri-environment policies. For developing countries,better representation of urbanisation and its variousimpacts on land-use changes at rural-urban interfaces,of transport infrastructure and market change willbe required. Given the difficulty of specifying futurepredictions of these driving forces, most of the mod-elling work will be aimed at scenario analysis. Thesescenarios should contribute to a better understandingof thresholds involving changes in land-use intensity.

Acknowledgements

Part of this work was supported by the Services ofthe Prime Minister, Office for Scientific, Technical andCultural affairs of Belgium (contract GC/XX/201). Ananonymous reviewer is thanked for his contribution inimproving the manuscript.

References

Ahn, T., Ostrom, E., Gibson, C., 1998. Scaling issues in the socialsciences. IHDP Working Paper No.1, Bonn, 85 pp.

Alcamo, J., Leemans, R., Kreileman, E., 1998. Global ChangeScenarios of the 21st Century. Pergamon, Oxford, 296 pp.

Alonso, W., 1964. Location and Land-use: Towards a GeneralTheory of Land Rent. Harvard University Press, Cambridge,p. 204.

Angelsen, A., 1994. Shifting cultivation expansion and intensity ofproduction: the open economy case. Working Paper 3. ChristianMichelsen Institute, Fantoft, Bergen.

Angelsen, A., 1999. Agricultural expansion and deforestation:modelling the impact of population, marked forces and propertyrights. J. Dev. Econ. 58 (1), 185–218.

Binswanger, H.P., Ruttan, V., 1978. Induced Innovation: Techno-logy, Institutions and Development. Johns Hopkins UniversityPress, Baltimore, MD.

Boserup, E., 1965. The Condition of Agricultural Growth. Allen& Unwin, London.

Boserup, E., 1975. The Impact of population growth on agriculturaloutput. Quart. J. Econ. 89, 257–270.

Boserup, E., 1981. Population and Technological Change.University of Chicago Press, Chicago.

Brookfield, H.C., 1972. Intensification and disintensification inPacific agriculture. Pacific Viewpoint 13, 30–48.

Brush, S.B., Turner, B.L., 1987. The nature of farming systems andviews of their change. In: Turner, B.L., Brush, S.B. (Eds.), Com-parative Farming Systems. Guilford Press, New York,pp. 11–48.

Chayanov, A.V., 1966. Peasant farm organization. In: Thorner, D.,Kerblay, B., Smith, R.E.F. (Eds.), A.V. Chayanov on the Theoryof Peasant Economy. Irwin, Homewood, pp. 21–57.

Dorsey, B., 1999. Agricultural intensification, diversification, andcommercial production among smallholder coffee growers inCentral Kenya. Econ. Geogr. 75 (2), 178–195.

Geertz, C., 1963. Agricultural Involution: the Process of EcologicalChange in Indonesia. University of California Press, Berkeley,CA.

Goldman, A., 1993. Agricultural innovation in three areas ofKenya: neo-Boserupian theories and regional characterization.Econ. Geogr. 69, 44–71.

Guyer, J.I., Lambin, E.F., 1993. Land-use in an urban hinterland:ethnography and remote sensing in the study of Africanintensification. Am. Anthropologist 94 (4), 839–859.

Hägerstrand, T., 1968. Innovation Diffusion as a Spatial Process.University of Chicago Press, Chicago.

Hayami, Y., Ruttan, V., 1985. Agricultural Development: anInternational Perspective. Johns Hopkins University Press,Baltimore, MD.

Jones, D.W., O’Neill, R.V., 1993. Human-environmental influencesand interactions in agriculture. In: Lakshmanan, T.R., Nijkamp,P.N. (Eds.), Structure and Change in the Space Economy.Springer, Berlin, pp. 297–309.

Jones, D.W., O’Neill, R.V., 1994. Development policies, rural landuse, and tropical deforestation. Reg. Sci. Urban Econ. 24 (6),753–771.

Kaimowitz, D., Angelsen, A., 1998. Economic Models of TropicalDeforestation: a Review. Centre for International ForestryResearch, Jakarta, 139 pp.

Kates, R.W., Hyden, G., Turner, B.L., 1993. Theory, evidence,study design. In: Turner, B.L., Hyden, G., Kates, R.W.(Eds.), Population Growth and Agricultural Change in Africa.University Press of Florida, Gainesville, pp. 1–40.

Lambin, E.F., 1994. Modelling Deforestation Processes: a Review.TREES Series B. Research Report 1. Office of OfficialPublications of the European Community, Luxembourg, 113 pp.

Lambin, E.F., 1997. Modelling and monitoring land-cover changeprocesses in tropical regions. Prog. Phys. Geogr. 21 (3), 375–393.

Lambin, E.F., Baulies, X., Bockstael, N., Fischer, G., Krug, T.,Leemans, R., Moran, E.F., Rindfuss, R.R., Sato, Y., Skole,


D., Turner, B.L., Vogel, C., 1999. Land-use and land-coverchange (LUCC): implementation strategy. A core projectof the International Geosphere–Biosphere Programme andthe International Human Dimensions Programme on GlobalEnvironmental Change. IGBP Report 48, IHDP Report 10.IGBP, Stockholm, p. 125.

Matson, P.A., Parton, W.J., Power, A.G., Swift, M.J., 1997.Agricultural intensification and ecosystem properties. Science277 (5325), 504–509.

Mortimore, M., 1993. Population growth and land degradation.GeoJournal 31 (1), 15–21.

Meertens, H.C.C., Fresco, L.O., Stoop, W.A., 1996. Farmingsystems dynamics: impact of increasing population density andthe availability of land resources on changes in agriculturalsystems. The case of Sukumaland, Tanzania, Agriculture.Ecosyst. Environ. 56 (3), 203–215.

Netting, R.McC., 1993. Smallholders, householders: farm familiesand the ecology of intensive, sustainable agriculture. StanfordUniversity Press, Stanford.

Peet, R., 1969. The spatial expansion of commercial agriculturein the nineteenth century: a von Thunen interpretation. Econ.Geography 45 (4), 283–301.

Pender, J.L., 1998. Population growth, agricultural intensification,induced innovation and natural resource sustainability: anapplication of neoclassical growth theory. Agric. Econ. 19 (1/2),99–112.

Riebsame, W.E., Meyer, W.B., Turner, B.L., 1994. Modelingland-use and cover as part of global environmental change.Clim. Change 28 (1/2), 45–64.

Rounsevell, M.D.A., Evans, S.P., Mayr, T.R., Audsley, E., 1998.Integrating biophysical and socio-economic models for landuse studies. In: Proceedings of the ITC-ISSS Conference onGeo-information for Sustainable Land Management, Enschede,17–21 August 1997, 368 pp.

Schultz, T.W. (Ed.), 1978. Distortions of Agricultural Incentives.Indiana University Press, Bloomington.

Socolow, R.H., 1999. Nitrogen management and the future offood: lessons from the management of energy and carbon. In:Proceedings of the National Academy of Sciences of the UnitedStates of America, Vol. 96 (11), pp. 6001–6008.

Stephenne, N., Lambin, E.F., 2001. A process-based, dynamicsimulation model of land-use changes in the African Sahel.Agric. Ecosyst. Environ., in press.

Thornton, P.K., Jones, P.G., 1998. A conceptual approach todynamic agricultural land-use modelling. Agric. Syst. 57 (4),505–521.

Thünen, J.H. von, 1966. Der isolierte Staat in Beziehung aufLandwirtschaft und Nationalökonomie. Neudruck nach derAusgabe letzter Hand (1842/1850). Fischer, Stuttgart, 687 pp.

Tiffen, M., Mortimore, M., 1994. Malthus controverted: the roleof capital and technology in growth and environment recoveryin Kenya. World Dev. 22, 997–1010.

Tiffen, M., Mortimore, M., Gichuki, F., 1994. More people, lesserosion: environmental recovery in Kenya. Wiley, Chichester,311 pp.

Tilman, D., 1999. Global environmental impacts of agriculturalexpansion: the need for sustainable and efficient practices. In:Proceedings of the National Academy of Sciences of the UnitedStates of America, Vol. 96 (11), pp. 5995–6000.

Turner, B.L., Ali, A.M.S., 1996. Induced intensification:agricultural change in Bangladesh with implications for Malthusand Boserup. In: Proceedings of the National Academy ofSciences of the United States of America, Vol. 93 (25),pp. 14984–14991.

Turner, B.L., Doolittle, W.E., 1978. The concept and measure ofagricultural intensity. Professional Geographer 30 (3), 297–301.

Turner, B.L., Moss, R.H., Skole, D.L., 1993. Relating land useand global land-cover change: a proposal for IGBP-HDP coreproject. IGBP Report 24, HDP Report 5. IGBP, Stockholm.

Turner, B.L., Hanham, R.Q., Portararo, A.V., 1977. Populationpressure and agricultural intensity. Ann. Assoc. Am.Geographers 67 (3), 384–396.

Walker, R., 1999. The structure of uncultivated wilderness: landuse beyond the extensive margin. J. Reg. Sci. 39 (2), 387–409.

Wassenaar, T., Lagacherie, P., Legros, J.-P., Rounsevell, M.D.A.,1999. Modelling wheat yield responses to soil and climatevariability at the regional scale. Clim. Res. 11, 209–220.

White, R., Engelen, D., Uljee, I., 1997. The use of constrainedcellular automata for high-resolution modelling of urbanland-use dynamics. Environ. Plann. B 24 (3), 323–343.

are agricultural land-use models able to predict changes in land-use intensity?

Documents