4 Discussion: Modelling of the land system

In this paper, eight integrated model approaches were reviewed. We found that a large variety of techniques with different levels of complexity were used to simulate the processes of the human-environment system and feedback between the sub-systems. Table 2 summarizes the main characteristics of the models. In allusion to the review criteria, this section addresses commonalities and differences between the models in respect of their representation of the key elements of land systems (i.e. human sub-system, the environment sub-systems and the linkages between them). In addition, questions of model application and adaptation to new geographic regions are discussed.

4.1 Representation of the human sub-system

Most of the reviewed models concentrate on the simulation of local land-use decisions and the resulting changes of the spatial land-use pattern (proximate level) and do not explicitly model processes on the underlying level. Instead regional or global demographic and economic developments are exogenous variables for the decision making process, for example in form of demands for housing area (PLM), policies and planning regulations by payments of subsidies (ITE2M) or land-use restrictions on natural conservation areas (CLUE, LANDSHIFT, SITE). Among the three examples that explicitly model processes on the underlying level (IMAGE, SYPRIA and GEONAMICA) only IMAGE includes feedback mechanisms from the local level to the regional level. This feedback is realized by land supply curves that are employed to indicate scarcity of suitable agricultural land which affects the calculation of regional agricultural demands by the GTAP model.

A major difference between the models is the level of land-use decision making. Verburg et al. (2006a*) distinguish between individual entities (people) and grid cells (pixel) as reference units for land-use decisions. The reviewed models cover a wide range of different approaches which fall between these extremes. We only found one agent based model (SYPRIA) that puts the “people” in the centre of interest by modelling the land-use related decisions of households and institutions. This may be explained by the comparatively high data demands of models that very explicitly represent human decision making (e.g. SYPRIA requires detailed data from household surveys). Contrastingly, CLUE, IMAGE, GEONAMICA and ITE2M strictly focus on grid cells (pixels) as decision units. LANDSHIFT and SITE implement an intermediate level of complexity. They distinguish between land-use activities as aggregated decision makers with different behaviour, which are competing for available land-resources, in ways that may differ between simulated sub-regions or countries.

Regarding the procedure of modelling land-use change, we found that with the exception of ITE2M in which decisions are made individually for each grid cell (or land parcel) all the reviewed models (also the agent based SYPRIA) can be categorized as top-down approaches (Verburg et al., 2006a). The first simulation step is an analysis of the suitability of each raster cell for the modelled land-use types. This is done either by methods from economic theory such as cost-benefit analysis (ITE2M) and logit models (PLM), by means of multi-criteria analysis (GEONAMICA, IMAGE, LANDSHIFT, SITE, SYPRIA) or by statistical regression (CLUE). In the second step, rule based and heuristic approaches (IMAGE, LANDSHIFT, SITE) as well as cellular automata (GEONAMICA) and iterative procedures (CLUE) are used to translate the demand either for area of different land-use types (CLUE, GEONAMICA, SYPRIA) or the amount of different commodities (IMAGE, LANDSHIFT, SITE) to a spatial land-use pattern. While the majority of models do at least in a simple way account for scale effects, in that they include more than one spatial level (e.g. region and grid), spatio-temporal hierarchies or lag-effects such as the question “how fast and to which degree does land use change, when one or more drivers have changed” are either ignored, or represented in a static manner (e.g. in CLUE-S, SITE).

The number of actively modelled land-use types depends on the scientific goals of the case study that the respective models are applied to. The potential of handling an arbitrary number of land-use types is given by the GEONAMICA and the SITE frameworks (which are rule-based models) and by the CLUE approach (mainly based on regression analysis). In other models, land-use types are associated to specific a decision making process (e.g. residential area in PLM or cropland in SYPRIA). These models cover a much smaller number of “active” land-use types but at the same time support a more differentiated representation of decision making processes. On the other hand, for each additional “active” land-use type, a dedicated decision making process has to be programmed and included into the respective model.

4.2 Representation of environmental processes

In the models reviewed, we identified different levels of complexity of modelling the environment sub-system. The rule-based cellular automata approach in SYPRIA and empirical models for crop yields in IMAGE and ITE2M are contrasted by process-oriented models. For instance, LANDSHIFT, PLM and SITE use very detailed ecosystem models (DayCent or GEM), whereas GEONAMICA achieves a similar degree of complexity by linking separate components for soil, hydrology and vegetation. Under the perspective of a clear model formulation, problems arise when environmental processes are represented not uniquely within a model or when processes are only poorly connected. In these cases, environmental variables are calculated by different methods which may lead to inherent inconsistencies in the model results. We found an example for the redundant formulation of an ecosystem process in ITE2M. Here, the ProLand model component uses an empirical approach (Liebig function) to calculate crop growth while the same process is simulated by the EPIC model within the hydrology component. An example for poorly connected processes can be found in the current IMAGE model, where the global nitrogen cycle is modelled independently from vegetation and carbon dynamics.

Another important finding was that most of the reviewed integrated models limit their representation of the environment on “point processes”. For this task they employ ecosystem models that operate on grid level and do not account for horizontal matter and energy fluxes between grid cells (SITE, GEONAMICA, IMAGE). Horizontal water fluxes (e.g. river discharge) are only simulated by the PLM model. ITE2M and LANDSHIFT also use complex hydrological models but in a loosely coupled fashion. These results indicate that the potential of using the reviewed models in context of water resources related research (e.g. questions of water availability and water quality) is only marginally explored, yet.

Table 2: Main characteristics of the reviewed models. (Note that several models can be applied at different spatial resolutions, which may not all be listed in this table.)


Grid resolution (typical)

Representation of the human sub-system and land-use change

Representation of environmental processes

Human effects on the environment

Effects of environmental change


CLUE: 7×7 km – 32×32 km

CLUE-S: 1×1 km

Change of spatial land-use pattern (various classes)

Methane emissions

Nutrient balances in agriculture

Erosion and sedimentation

Land-use pattern

Soil depth by erosion


100×100 m –

500×500 m

Water management

Change of spatial land use pattern (various classes)

Farmer’s choice of field crops on arable land

Climate & weather

Hydrology and soil

Natural vegetation

Crop yields

Land-use pattern

Agricultural management (sowing, ploughing, harvest, irrigation etc.)

Crop yields


30 arcmin

(50×50 km at the Equator)

Energy supply and demand

Consumption and trade of agricultural commodities

Change of spatial land use pattern (agriculture, forestry)


Potential crop yields

Carbon cycle

Nutrient cycles


Land-cover pattern

Crop yields

Land supply


25×25 m

Land-use change (agriculture, forestry)


Potential yield


Fate of heavy metals

Land-use pattern,

Agricultural management (fertilizer, ploughing etc.)

Crop yields


5 arc-min

(9×9 km at the Equator)

Change of spatial land-use pattern (agriculture, grazing, settlement)

Crop yields, NPP

Soil water, carbon and nutrients


Land-use pattern

Crop yields


Water availability


200×200 m –

1×1 km

Change of spatial land-use pattern (residential area)

Agricultural management


Nutrient dynamics (soil, water)


Land-use pattern

Sewage disposal,

Soil sealing,

Agricultural management (fertilizer, planting etc.)


250×250 m –

500×500 m

Change of spatial land-use pattern (settlement, agriculture, forestry)

Biodiversity, crop yields, NPP

Soil water, carbon and nutrients

Land-use pattern

Agricultural management (fertilizer, ploughing etc.)

Crop yields


28×28 m

Institutional boundary conditions

Choice and location of agricultural systems by households

Soil fertility

Vegetation succession

Land-use pattern

Agricultural management (fertilizer, fallow cycle)

Soil fertility

4.3 Model integration

The link between the human and the environment sub-system is modelled via the effects of changing land-use patterns and the associated management actions on environmental factors and processes. Examples for the effects of changing land use patterns cover impacts on biodiversity (ITE2M, SITE) and on hydrological parameters (LANDSHIFT, ITE2M). Modelling approaches to quantify the influence of the associated land management on the environment concentrate on the agricultural sector (soil fertility, crop growth, nutrient input). Effects of urbanization on water quality and surface run-off are only addressed by PLM. The potential effects of other types of land management for example in the forestry sector (in terms of management practices beyond deforestation) or in the field of nature conservation are addressed by none of the models. The same holds true for the modelling of effects of water use on available water resources (see last paragraph).

An important process for the agricultural sector, which is modelled very differently, is agricultural intensification, which is considered a key issue in land-use modelling by Lambin et al. (2000). In CLUE, the effects of intensification are modelled indirectly by changing the exogenous area demands for cropland, while IMAGE and LANDSHIFT use correction factors to account for yield changes due to technological change over time. On the other hand, approaches that explicitly model intensification processes are limited to the simulation of expansion of irrigated area (LANDSHIFT) and to the simulation of future trends of fertilizer application rates affecting crop productivity (PLM, SITE). A major limitation for the latter example is the lack of solid spatially explicit data on fertilizer use (Costanza et al., 2002). Other means of intensification such as genetic manipulated crops or pesticide application are not taken into account by any of the models.

A central element of the conceptual model of land systems are feedback effects of environmental change (caused by human activities) on the human system. Which feedbacks are implemented strongly depends on the model purpose. Most of the reviewed models again concentrate on the links between agricultural management and their effects on changes of land productivity. These effects are modelled as soil degradation (erosion), changes of soil fertility or changes of crop productivity, and feed back both on the suitability assessments and on the land allocation procedures. The only examples for linking data on changes in biodiversity to human decision making can be found in ITE2M and SITE. From these examples it becomes obvious that at the moment only a rather limited number of the possible environmental change impacts on human systems are covered by the individual models. Even more important is the fact that there is a significant lack of studies that systematically explore the influence of the implemented feedback effects on the model behaviour. The only attempt that we could identify is the publication by Verburg (2006).

4.4 Model application and adaptation to other geographic regions

The application of the reviewed models is predominantly in the context of scenario analysis with emphasis on the quantity and location of land-use change and the effects on environmental services such as crop yields, water quality, biodiversity and soil quality. In contrast, only GEONAMICA is explicitly addressed as Decision Support System. Types of scenario analysis include the comparison of different future pathways defined by the trends of exogenous driving variables such as population or economic development as well as the analysis of specific policy (tax) and land management options such as agricultural intensification in terms of fertilizer input and/or irrigation (ITE2M, LANDSHIFT, PLM, SITE, SYPRIA). A major difference in the underlying modelling philosophy of land-use changes either based on inductive (e.g. statistical regression models like CLUE) or deductive approaches (e.g. rule based like SITE) imply different possibilities for the use in context of a scenario analysis. Concerning the time horizon of a scenario analysis, inductive approaches make long-term projections difficult (Verburg and Denier van der Goon, 2001), since the empirical relationships cannot necessarily be assumed to be constant over long time periods. This shortcoming is potentially overcome by deductive approaches that explicitly model the interaction between underlying drivers of land-use change and decision making processes. They allow the modelling of changes over time in these relationships as part of the scenario assumptions and therefore increase their suitability to model the changing dynamics of land systems over a longer time period. Nevertheless time dependency of decision making is targeted by none of the reviewed models in a sufficient way, yet. As most of the models do support this feature, we see great potential for the models in the development of more differentiated scenario experiments.

Regarding the possibility to adapt the integrated models to other geographic regions, we have to distinguish between the large scale models IMAGE and LANDSHIFT which can be applied globally for any study region, like (large) countries or catchments, and the models predominantly used on the regional and watershed-level. The analysis of the latter ones shows that SYPRIA and ITE2M are developed specifically for single case studies. Here, particularly the modules representing the human sub-system require very specific data for defining household decision strategies (based on data from household surveys) or for economic data to describe cost structures of agricultural production systems. This extensive data demand makes it difficult to transfer these models to other study regions or scale levels.

In contrast, the CLUE, GEONAMICA and SITE are designed as frameworks that support the construction of region specific model applications, though with very different means. The CLUE team at Wageningen University provides information of how to adapt their model framework. Nevertheless the framework definition is limited to the land-use component. The GEONAMICA framework, as a rule-based system, among other factors, relies strongly on the positive or negative spatial interaction of all simulated land-use types and provides several user interfaces to adjust amplitudes and distances of spatial interactions, and other model parameters. Furthermore, it provides a consistant technical framework for the integration of both human and environment models in the form of so called “Model Building Blocks”. Finally, SITE provides a generic land-use modelling framework, for which case study specific rule-sets can be developed in the widely used scripting language Python, e.g. based on existing applications. It also offers technical solutions to couple different model components.

  Go to previous page Scroll to top Go to next page