3 Integrated models of the land system

In this paper, we review recent integrated models that simulate the functioning of the land system. The selection of models is based on criteria that we regard as characteristic for this type of model. (1) The first criterion is the argument that the simulation of the dynamics of the coupled human-environment system requires the inclusion of processes of both the human and environment sub-system into the model. (2) In order to stress the integrated approach, only models which focus on the interplay and competition between different land-use activities (e.g. between agriculture and urban development) and their environmental consequences are taken into account. (3) As third criterion, we concentrate on geographically explicit models that are designed for regional and watershed scale level (> 1000 km²) up to the global scale level. Based on a review of the literature, we identified eight models that meet our criteria (Table 1).

The review framework is based on the conceptual model for “land systems” described in the last section. On an abstract level these systems are composed of three basic elements: the human sub-system, the environment sub-system and the links and feedbacks between them. In our framework, we use these elements as categories for identifying the different aspects of land systems that are considered by the selected models and for analyzing the implemented modelling concepts. Additionally, a category describing model purpose and application is introduced.

Model purpose and application

This category covers the scientific question that the model is built for and general aspects of the underlying model philosophy and the spatial scale levels, the model is used on. Moreover, we give examples of studies conducted with the model in order to illustrate its field of application.

Model concepts for the human sub-system

Here, we discuss the modelling concepts representing the different aspects of the human sub-system. We identify, which parts of the human sub-system are endogenously modelled and which are assumed as exogenous variables. For the endogenous parts, we review the modelling approaches both for the level of underlying causes (e.g. economic development and demography) and for land-use decision making as well as the spatial scale level these approaches are designed for.

Model concepts for the environment sub-system

Under this headline we examine, which parts of the environment sub-system are included in the integrated model, their scale level and implemented modelling approaches.

Model integration

Model integration describes the linkages and feedbacks between the models for the human and environmental sub-systems. We examine how the coupled model represents the influence of human activities on environmental factors and in turn, how changing environmental properties affect human land-use decisions and other aspects of the human sub-system.

3.1 CLUE modelling framework

Model purpose and application

The CLUE (Conversion of Land Use and its Effects) modelling framework is described as a tool for quantitative multi-scale analysis of actual land use and the modelling of land-use change scenarios (Verburg et al., 1999a). Based on this framework, a family of different models has been developed. The original application of CLUE focuses on the national and continental scale with grid cell sizes between 7×7 km and 32×32 km. To each cell a relative cover of land-use types is assigned. Examples of model applications are studies concerned with land-use change in China (Verburg et al., 1999c), the influence of high population pressure on land-use change on Java, Indonesia (Verburg et al., 1999b) and deforestation caused by the expansion of grazing activities in the Neotropics (Wassenaar et al., 2007). In contrast, the CLUE-S model version (Verburg et al., 2002*) uses a finer grid resolution of 1×1 km. Here, each grid cell has a dominant land-use type. Originally designed for the regional or watershed level, it has recently been applied on the continental scale for a European-wide scenario analysis, together with the IMAGE model (Verburg et al., 2006b). While the CLUE framework focuses on the human aspects of the land system, in several studies the derived models have been coupled to environmental models for the estimation of greenhouse gas emissions (Verburg and Denier van der Goon, 2001*), soil erosion (Verburg, 2006*) and carbon and nutrient fluxes (Priess et al., 2001*).

Model concepts for the human sub-system

The model concepts for the representation of the human sub-system differ between the various models derived from the CLUE framework. The original CLUE model combines empirical-statistical elements with dynamic simulation to model the spatial dynamics of land-use change. The empirical-statistical model part comprises methods to analyse and model statistical relations between the observed land-use pattern and socio-economic (e.g. income, population density), geographical (e.g. distance to cities, markets or roads) and biophysical (e.g. slope, precipitation) explanatory factors by means of multiple regression analysis (Veldkamp et al., 2001). Based on this analysis, each cell’s suitability for the modelled land use types is determined. In the second step, land-use change is simulated by dynamic modelling of competition between the land-use types. Competitive advantage is based on the suitability values of a cell. Moreover land-use change is driven by changes of national level demand for each land-use type, for example as a result of population growth or an increasing crop production. This information is provided as exogenous variables.

In CLUE-S, the empirical statistical analysis is supplemented by decision rules that determine which land-use type is allowed to change in each simulation time step. With this element it is possible to restrict conversion from urban area to agriculture or to exclude nature conservation areas from land-use changes. Overmars et al. (2007) replace this empirical procedure for determining the cell suitability values by a deductive approach (Action in Context method) that models human decision making in terms of “options” and “motivations”. The method is based on household surveys and applied in a case study for a municipality in the Philippines. Based on the suitability maps and the decision rules in combination with the actual land-use map, then CLUE-S calculates land-use change. This is done by an iterative procedure which allocates area demands for different land-use types to the best suited grid cells (Verburg et al., 2002). Similar to CLUE, changes of area demands are specified as exogenous driving variables which reflect underlying causes like population growth or changes of economic boundary conditions.

Model concepts for the environment sub-system

The CLUE framework does not explicitly represent processes of the environment sub-system. Nevertheless, different studies illustrate how the CLUE models can be combined with environmental models of different levels of complexity. For China, based on output of the large scale model version of CLUE, the effects of changes in rice cultivation area and livestock management on methane emissions were analysed (Verburg and Denier van der Goon, 2001*) by using empirically derived emission factors (IPCC, 1997). Another example is presented by Priess et al. (2001). They couple a CLUE model with a spatial resolution of 9×9 km to the NUTMON tool (Smaling and Fresco, 1993) for the assessment of soil nutrient balances in tropical agricultural systems. NUTMON is an input-output model that relates nutrient inputs such as fertilizer application and biological fixation to output fluxes in (among others) harvested products, crop residues and by gaseous losses. Moreover, Verburg (2006*) couples the CLUE-S model to the process based LAPSUS model (Schoorl et al., 2004) in order to calculate the impact of land-use change on erosion erosion/sedimentation processes for a region in Southern Spain.

Model integration

The linkage of CLUE to the methane emission models is realized as a loose coupling approach where results from CLUE serve as input for further processing. In contrast, the coupling to the LAPSUS model is done by dynamically linking the models. In this example, erosion processes are influenced by agricultural management (e.g. tillage, changes in water infiltration and changes in vegetation cover). Erosion and sedimentation processes change the soil depth. This information on environmental change affects land-use decisions in the human sub-system. For example, when soil depth becomes too shallow to be productive or in case of the occurrence of gully-erosion a grid cell is evaluated as unsuitable for agriculture and consequently is abandoned. The case study shows significant differences between the simulated land-use patterns, compared to the results from simulations that neglect the environmental feedback.

3.2 GEONAMICA framework

Model purpose and application

GEONAMICA is a commercial application framework for constructing integrated models of the land system. It further elaborates the concepts on cellular automata based land-use modelling presented by White and Engelen (1997*) and (Engelen et al., 1997*). The framework supports spatial scale-hierarchies consisting of macro-levels (regional level) and spatial grids, with a resolution typically ranging from 100×100 m to 500×500 m. Based on this framework, numerous regional integrated models have been developed for use as Decision Support Systems (Oxley et al., 2004*; de Kock et al., 2001*; MODULUS, 2000*) and Policy Support Systems (van Delden et al., 2007*). Examples are MODULUS (MODULUS, 2000; Oxley et al., 2004), which addresses physical, economic and social aspects of land degradation in the Mediterranean region, RamCo for coastal zone management in South Sulawesi in Indonesia (de Kock et al., 2001*) and MedAction for the support of planning and policy making in the fields of land degradation, desertification, water management and sustainable farming in Mediterranean watersheds (van Delden et al., 2007*). Sharing most features with Modulus, in the following the more recent MedAction system (van Delden et al., 2007*) is chosen to illustrate the concepts of GEONAMICA-based models.

Model concepts for the human sub-system

Processes of the human sub-system are simulated by three model components: (1) water management, (2) land-use, and (3) crop choice and profit. The first component calculates water use of different land-use activities. The land-use model (Engelen et al., 1997) simulates land-use change on two different scale levels. On the regional level (macro-level), it calculates the land demands for different economic sectors such as agriculture, forestry and industry (de Kock et al., 2001) by representing a mixture of underlying causes such as demography and market mechanisms, according to Geist and Lambin (2002). These land demands are then spatially allocated to the grid level (micro-level). Here, land-use decision making is simulated by the constraint cellular automata approach introduced by White and Engelen (1997). As important factors for location choices, the model considers the cell’s suitability for different types of land-use, its spatial neighbourhood, zoning restrictions and accessibility of transport infrastructure. In the next step, the crop choice and profit model is applied to determine the best suited crop type for each cell that has been classified as agricultural land before. This is done by portraying the decision making process on level of individual farmers, which among other factors regards farming history, yield and expected profits. While crop prices are provided as exogenous scenario variables, each cell’s suitability for a particular crop is determined dynamically by analysing important soil and landscape parameters.

Model concepts for the environment sub-system

The environment sub-system consists of (1) a climate and weather model, (2) a hydrology and soil model and (3) a vegetation model. Simulations are performed on grid level with different temporal resolution. The climate and weather model regionalises data from meteorological stations and general circulation models to the grid level. In the hydrology model the processes interception, runoff, evapotranspiration, infiltration, soil moisture, aquifer recharge, river flow and transmission loss are implemented. Based on this information, together with soil characteristics such as texture and thickness, the model simulates processes of soil erosion and sedimentation as well as soil and aquifer salinity. The vegetation model consists of two elements. The plant growth element determines structural components of plants and crop yields. The natural vegetation element applies a rule-based approach for calculating succession processes and the response of natural vegetation to disturbances like fire. It determines the vegetation type for all grid cells that are not used by human activities like settlement and agriculture.

Model integration

The level of integration within and between the sub-systems is high. Within the environment sub-system, numerous feedback mechanisms are realized. For example, soil moisture as calculated by the hydrology model is used by the plant growth model for the simulation of biomass growth. The calculated biomass, in turn influences soil moisture due to the dependence of soil evaporation and transpiration on structural variables such as leaf area index and root biomass (van Delden et al., 2007). Feedback between the human and environment systems is realized for agricultural management. On the one hand management events such as sowing, ploughing, harvest and irrigation have a direct effect on crop growth. In turn, the computed crop yields influence the farmer’s choice for a specific crop in the profit and crop choice model. If an agricultural cell is taken out of production, it changes back to natural vegetation over a defined time period.

3.3 IMAGE model

Model purpose and application

The Integrated Model to Assess the Global Environment (IMAGE 2.4) is not a specialized model for the global land system but a dynamic integrated assessment model that aims at the simulation of the major societal and environmental processes of the Earth System and their feedbacks (MNP, 2006*; Alcamo et al., 1998). The model is designed for global level analysis with a time horizon of 100 years and operates on two different scale levels. A macro-level is defined by 24 world regions while the Earth surface is represented as a 30 arcminutes grid equalling a cell size of approximately 50×50 km at the equator. IMAGE has been applied to calculate land-use emissions (Strengers et al., 2004) as well as impacts of land-use change on ecosystems and the environment on global (GEO, 2002; Leemans and Eickhout, 2004; Carpenter et al., 2005; Eickhout et al., 2006) and (in context of the EURURALIS project) on the continental level (Eickhout et al., 2007). IMAGE output in form of land-cover maps serves as input for different types of studies, ranging from bio-energy potentials (Hoogwijk et al., 2005; de Vries et al., 2007) to the assessment of the impacts of land-use change on the carbon balance of the terrestrial biosphere by coupling to the global vegetation model LPJ (Sitch et al., 2003; Müller et al., 2007).

Model concepts for the human sub-system

The human sub-system of the global land system is represented by three major model components which are driven by exogenous data on demographic and economic trends. (1) Energy supply and demand is simulated by the TIMER model for each world region. While the energy demand is estimated based on aggregated economic indicators such as GDP and household consumption, the supply side regards different technology paths of energy production. The model output includes information on energy consumption by different sectors such as residential or industry, the mix of energy carriers (coal, oil, gas, bio-fuels etc.) as well as on the resulting emissions of greenhouse gases. (2) Consumption and trade of agricultural products are calculated by the GTAP model. GTAP is an economic Computable General Equilibrium model (CGE) that takes into also account the impact of non-agricultural sectors on agriculture (van Meijl et al., 2007). (3) Finally, land-use change is calculated by a spatial allocation model on the grid level. The land-use decision making is formulated in a rule based manner and includes two steps. First, the suitability of each raster cell for agriculture and timber extraction is calculated, taking into account spatial neighbourhood to other land-use types, population density as well as potential crop yields and forest productivity. Then, the demand for agricultural and forest commodities (computed by GTAP and TIMER) as well as for bio-energy crops (computed by TIMER) is distributed to the best suited grid cells within each region, according to a set of heuristic “land allocation rules”.

Model concepts for the environment sub-system

The environment sub-system consists of model components for terrestrial vegetation, terrestrial carbon and nutrient cycles, and biodiversity. All these components operate on grid level. The terrestrial vegetation part is further separated into a model to compute the potential distribution of natural vegetation and a model for potential crop yields, both taking into account local soil and climate conditions. Potential crop yields are computed with the global agro-ecological zones approach (Fischer et al., 2002) that uses both rule-based and process-based modelling techniques. The terrestrial carbon model simulates carbon fluxes between atmosphere and biosphere taking into account the impact of land-use change as well as climate change, represented by the processes of soil respiration and net primary production of natural vegetation. The nutrient cycle model computes the surface balances of phosphorus and nitrogen, including nutrient emissions to the atmosphere and to groundwater and surface water. It accounts for point sources like wastewater treatment as well as soil nutrient dynamics of agricultural soils (non-point sources). The impact of stress factors like climate and land-use pattern on biodiversity is assessed by the GLOBIO model (Nellemann et al., 2001). Being formulated as a statistical regression model, it is used to estimate the mean species abundance in relation to the species abundance in primary undisturbed vegetation.

Model integration

The model components of IMAGE are tightly coupled with feedback mechanisms between the different components and across scales. Within the human sub-system, for instance regional demands for agricultural commodities and bio-energy crops are provided to the land- use model by GTAP and TIMER. In turn, yields and yield changes due to climate change as well as information on the expansion of agriculture to less productive area are fed back to GTAP in order to calculate land supply curves (MNP, 2006). Regarding the environment sub-system, land use and the type of natural vegetation serve as input to the carbon cycle model while the land-use pattern and nutrient loadings have an impact on biodiversity. Aiming at the representation of the whole Earth System, IMAGE also models feedbacks between the land-system and the atmosphere-ocean system. For example, natural vegetation patterns and crop yields are influenced by climate parameters. At the same time, activities in the different land use (via the carbon model) and energy sectors produce greenhouse gas emissions (e.g. by deforestation or combustion processes) that contribute to climate change.

3.4 ITE²M model network

Model purpose and general design

The ITE²M (Integrated Tool for Ecological and Economic Modelling) model network has been developed to simulate the functioning of complex human-environment systems and to evaluate the influence of changing land use on environmental services. ITE²M links interdisciplinary models, covering aspects of agro-economy, agricultural policy and environmental services (Fohrer et al., 2002; Frede et al., 2002*; Reiher et al., 2006*). Scale levels of ITE²M are specific to the model components, ranging from a 25×25 m grid over fields to sub-catchment level. Its first application was for the “Lahn-Dill Bergland” region (1200 km²) in Germany. Simulation studies focus on the effects of changes in land-use policy and agricultural management on environmental services like economic output, water balance and biodiversity (Frede et al., 2002*; Fohrer et al., 2001; Gottschalk et al., 2007*).

Model concepts for the human sub-system

The human sub-system is modelled by the models ProLand (Möller et al., 1999; Kuhlmann et al., 2002*) and CHOICE (Borresch and Weinmann, 2006). ProLand is a static bio-economic model to simulate the spatially explicit allocation of land-use systems for a specific point of time in the future, dependent on legal and economic boundary conditions, and environmental factors. Land-use systems include various types of cropland, grassland, forest and fallow systems whereas settlement area and infrastructure are assumed as static. For each decision unit (field level or 25×25 m grid cell) the land-use system with the highest land rent is selected, assuming a profit maximizing behaviour of the land-owners. Land rent is calculated as the difference between profit and costs of production. While profit is defined as a function of potential yield, market prices and transfer payments, the cost side considers the land-use system specific labour and machinery costs, which are influenced also by local environmental factors like slope and soil clay content. The CHOICE model consists of two parts. First, the model evaluates landscape functions with an economic cost-benefit approach. Model input is delivered by ProLand and the environmental assessment models described below. Then, a sub-model for agricultural trade (AGRISIM) calculates price effects of agricultural policies that serve as input for the ProLand model.

Model concepts for the environment sub-system

Processes of the environment sub-system are represented by the yield estimation module of ProLand and by different models used for the assessment of environmental services. ProLand uses yield information for cropland, grassland and forest to determine the income side of the land rent calculations. Potential yield is calculated for each grid cell with a Liebig production function, based on site conditions like soil, temperature, water and genetic potential (Kuhlmann et al., 2002). Models for environmental services cover aspects of hydrology, biodiversity and soil pollution. The eco-hydrological model SWAT (Soil Water Assessment Tool, Arnold et al., 1993) is used to calculate the water balance and water quality on level of sub-catchments. Hydrological processes include potential evapotranspitration, surface run-off and soil percolation. Moreover, crop growth is computed by a stripped-down version of the EPIC-model (Williams et al., 1990). The biodiversity side is represented by the cellular automata model ANIMO for habitat specific species numbers and botanical species richness (Frede et al., 2002) and the GEPARD model for species richness of birds and carabids (Gottschalk et al., 2007). Reiher et al. (2006) also describes the application of the ATOMIS model to estimate the fate of heavy metals in top soils. These three models operate on the 25 m grid level.

Model integration

The linkages between the models within ITE²M are straight forward and do not account for feedback loops. ProLand computes land-use scenarios and serves them as input to the environmental assessment models. Based on the collected model results, CHOICE then evaluates the environmental services. Human-environment interactions are modelled by the yield module of ProLand that provides information to the decision making process. Furthermore the calculated land-use pattern influences biodiversity and the management activities associated to each land-use system (e.g. fertilizer input) have an effect on the water balance and quality as well as on crop growth as calculated by SWAT.

3.5 LANDSHIFT model

Model purpose and application

The aim of the LANDSHIFT model is to simulate land-use change dynamics on global and continental scale (Alcamo and Schaldach, 2006; Schaldach et al., 2006*). It integrates functional components that represent human and environmental aspects of the land system. LANDSHIFT operates on a spatial scale-hierarchy that consists of three different levels: a macro-level defined by countries, an intermediate level (global 30 arcminutes grid) and a micro-level that is defined by a global 5 arcminutes grid. Currently there are two applications documented. First is a continental scale analysis on impacts of land use changes on the spatial extent of natural vegetation in Africa (Schaldach et al., 2006) and on future trends of the extent of irrigated areas (Heistermann, 2006*). Second application is an analysis of the impact of bio-fuel development on country-wide land-use change in India (Schaldach et al., 2008).

Model concepts for the human sub-system

The human sub-system of the land system is represented by the land-use-change module (LUC-module). Societal and economic data like population growth, production of agricultural commodities (crops and livestock) and yield increases are provided as exogenous variables to the module. In current studies, the Partial Equilibrium Model IMPACT (Rosegrant et al., 2002) is used to calculate agricultural production and crop yield changes for countries. The LUC-module determines the locations where land-use change takes place by regionalizing the country level demands for area intensive commodities and services to the micro-level, i.e. to the grid cells of the particular country. Land-use decisions are made by different sub-modules, each representing one land-use activity. Currently sub-modules for settlement (METRO), crop cultivation (AGRO) and grazing (GRASS) are implemented. Underlying model concept is a demand and supply approach. Each sub-module is responsible to allocate specific demands to the most suitable cells. The supply side is defined by the available land (number of grid cells) and the amount of a commodity that can be produced on the cell (e.g. crop yields, net primary productivity (NPP) of grassland). This information is calculated by “productivity modules” that are part of the representation of the environment sub-system (see below). Consequently, changes of land-use pattern are either induced by changes of the demand-side or by changes of local productivity. Regionalization is done in a two step procedure. Firstly, the suitability of each grid cell for a particular land-use type is determined by a multi-criteria analysis conducted by each sub-module (Eastman et al., 1995*). The analysis considers sector-relevant landscape properties (e.g. biomass productivity and slope) as well as spatial neighborhood relations and socio-economic factors (e.g. population density and road infrastructure) in form of a linear utility function. Furthermore, land-use constraints like protection of nature reserves are taken into account. The second step is to allocate land within a particular land-use sector. For each sub-module, a specific allocation strategy is implemented. For instance, the allocation of different crop types on arable land (AGRO) is computed by a modified version of the Multi Objective Land Allocation Algorithm (Eastman et al., 1995*) while METRO uses a rule-based approach to distribute new population to the raster cells. Competition between the land-use sectors is modelled by defining a hierarchy between the responsible sub-modules that reflects their economic importance. This results in the sequential execution, starting with METRO, followed by AGRO and GRASS.

Model concepts for the environment sub-system

Within LANDSHIFT, the environment sub-system is defined by productivity modules. They generate information on biomass productivity in terms of crop yields and grassland net primary production (NPP), on each 5 arcminutes raster cell under local climate, soil and management conditions. These data are used for suitability evaluation by the land-use sub-modules and also determine the potential local production (supply) for each agricultural commodity. Currently, these calculations are carried out by a grid version of the agro-ecosystem model DayCent (Parton et al., 1998*; Stehfest et al., 2007*), which has a detailed process-oriented representation of plant growth, soil water fluxes, soil carbon dynamics and nutrient pool dynamics. The model calculates global maps on a 30 arcminutes grid for yields of major crop types and for grassland which are mapped to the 5 arcminutes cells within the boundary of each 30 arcminutes cell. Calculations are conducted for 10-year time slices of climate change scenario data under country-specific crop management assumptions. Hence, data on crop management and climate variables are both exogenous driving variables to the model.

Model integration

The model components of LANDSHIFT are coupled by a soft-link. Both productivity and economic data are calculated separately and then are provided as exogenous input variables to the LUC-module. The link between the human and environment sub-system is established by the variable “biomass production”, as it influences location choices for cropland and grazing. The crop yields generated by DayCent are adjusted each time step by the IMPACT model outcome on yield increases, which reflect improvements of agricultural management due to technological and societal change. Moreover, simulation results from the global hydrology model WaterGAP2 (Alcamo et al., 2003) on water availability are used as exogenous variable to determine suitable grid cells for the expansion of irrigated area (Heistermann, 2006). While between the components for the human and environment subsystem, feedback mechanisms are not implemented, yet, the internal structure of each component shows numerous complex feedback mechanisms. For example, the DayCent model represents feedbacks between the different parts (e.g. vegetation and soil) of the modelled ecosystem, while the LUC-module simulates competition between the different land-use sectors in dependence on the actual spatial land-use pattern.

3.6 Patuxent Landscape Model (PLM)

Model purpose and application

The Patuxent Landscape Model (PLM) is an integrated modelling framework to simulate ecosystem processes on the watershed level and their linkages to human factors (Voinov et al., 1999a*; Costanza et al., 2002*). The model operates on a grid with a cell size ranging from 200×200 m up to 1×1 km. It has been developed for the Patuxent River watershed in the State of Maryland in the U.S., in order to address problems of water quality management. Principal aims of the model are to develop a better understanding of the complex interactions between human and natural systems and to serve as a predictive policy tool for supporting environmental management efforts to improve water quality by reducing nutrient loading resulting from land use. Results from a scenario analysis which capture the impact of land-use change as well as of changes in agricultural management practice are presented by Costanza et al. (2002*).

Model concepts for the human sub-system

The human sub-system is modelled by an economic model which computes the probability of conversion from agriculture or forest to residential land use within the seven counties of the Patuxent River watershed. It models the decision making process with a Markov Chain approach, interpreting the Markov transition probabilities as discrete choice probabilities (Bockstael, 1996). As decision criteria the model applies the value of a grid cell for different types of land use, based on factors like distance to public infrastructure and proximity to other land-use types as well as on historical land conversion decisions and land conversion costs. Model output is the likelihood of land-use conversion for each grid cell. The result is interpreted as the spatial pattern of development pressure (Voinov et al., 1999a). The actual quantity of land-use change is then computed by combining these probability maps with assumptions of regional growth pressure for settlement development, defined as part of the scenarios, for example as number of new housing units in the future (Costanza et al., 2002*). In PLM the aspect of land management practices of agricultural and residential land use and its impacts on the environment, which, in context of this review, we understand as a part of the human sub-system, is modelled within the “ecological component” by the “human dominated systems” sub-model (see below).

Model concepts for the environment sub-system

The environment sub-system is described by the “ecological component” which includes a grid level model and a spatial model. The grid level model (unit model) simulates ecological processes for the local “habitat type” (Costanza et al., 2002*) like forest, cropland, grassland, urban and open water within each grid cell. It builds on the General Ecosystem Model (GEM) by Fitz et al. (1996) and beside the sub-model for “human-dominated systems” includes 3 other linked sub-models (Costanza et al., 2002*) to describe the different functional parts of an ecosystem. (1) The hydrology sub-model computes vertical water fluxes within each cell including infiltration, transpiration and evaporation. (2) The nutrient sub-model simulates the dynamics of phosphorus and nitrogen. It models plant uptake and soil processes like organic matter decomposition. (3) The macrophyte sub-model simulates processes of plant growth for different habitat types (both in aquatic and terrestrial environments) in response to temperature, nutrient levels and water availability. The unit models are connected by the spatial model which is responsible for computing horizontal fluxes of water and the associated nutrient fluxes between the grid cells (Voinov et al., 1999b).

Model integration

PLM is characterized by a high level of integration. It models complex inter-linkages between hydrology, plant growth and nutrient cycling as well as between the human sub-system and the environmental sub-system. The model considers land-use conversion as well as land-use modification in the agricultural sector by changing management practices, realized by the “human-dominated system” model component. This component acts as an interface between the human and the environment sub-system. While agricultural land use affects plant growth and nutrient dynamics by fertilizer input, planting, harvesting and crop rotations (Binder et al., 2003), residential land use has an effect both on surface run-off (hydrology) and nutrient loading by road run-off or point sources like discharges from septic tank of households, located within the watershed.

3.7 SITE model

Model purpose and application

The SITE modelling framework (Simulation of Terrestrial Environments) is a spatial explicit integrated land-use model, developed for studies at the regional scale. The model is driven by demographic, economic, and climatic dynamics and by demands for spatially relevant commodities, such as agricultural and forest products and space for housing and commerce. SITE provides generic land-use modelling functionalities to implement different case studies. The Sulawesi case study, for example is covering 7,200 km² and employs a spatial resolution of 250 m (sub-region) or 500 m (Priess et al., 2007a; Mimler and Priess, 2008*). SITE allows for a multi-scale approach, e.g. implementing different spatial scales to simulate the study region in Indonesia (sub-districts, communities, grid cells). Additional important spatial units such as catchments, protected areas, or preference zones are implemented and may crosscut other spatial units such as administrative boundaries. According to Priess et al. (2007b*), the major purpose of SITE is to serve as a scientific tool in scenario analysis, to study socio-economic and environmental impacts of land-use dynamics and some of their important feedback mechanisms. Secondly, SITE is used to simulate stakeholder perceptions of perceived future opportunities and challenges, and to inform stakeholders and decision makers of the impacts of potential pathways of change (Priess et al., 2007b). In order to match the scientific purposes, the land-use module is dynamically coupled to biophysical models, which are simulating crops, natural vegetation (Parton et al., 1998*) and changes in pollinator diversity (Klein et al., 2003*). The biophysical models inform the land-use module of changes in for example crop productivity. This information is used (i) to calculate economic and environmental indicators and (ii) to adjust suitability and allocation ‘decisions’ in subsequent time steps. Model parameters without sufficient empirical basis can be calibrated using genetic algorithms and map comparison algorithms, which are implemented in additional components of the modelling framework (Mimler and Priess, 2008).

Model concepts for the human sub-system

The modelling framework can represent an arbitrary number of behavioural rules, including different types of decision makers. The complexity of the rules and the degree of detail represented, depend on (i) the purpose of the study, (ii) the availability of empirical descriptive information (regional preferences and believes; management of crops, etc), and (iii) structural information on decision makers and the decision making processes. In the Indonesia case study, the human component is represented as an ‘average decision maker’, implemented as a set of decision rules and constraints, e.g. with respect to the types of land-use conversions that are possible. Based on a multi-criteria analysis, which is carried out separately for every land-use type and every grid cell, a ranking of suitable cells is calculated. After the suitability analysis, the SITE model allocates demands for all commodities in a hierarchical fashion. This task is carried out by the allocation modules, which check (i) whether a cell is eligible for land-use change, (ii) which is the optimal land use for the present cell, and (iii) whether the transition from one land-use type to another is permitted. For example, cells recently converted to crops carrying high establishment costs like coconuts or cocoa, can only be converted after 15 years, and thus would not be eligible for change for example after 4 years. Additionally, the suitability of the new land-use type needs to surpass the suitability of the current type (i) to a predefined amount and (ii) for a number of years specific to each land-use type, mimicking the conservative behaviour of farmers. The allocation hierarchy is based on regional allocation priorities. Competition of different land-use types (for example crops) for favourable locations is simulated explicitly by the allocation algorithm, which identifies the local optima for all land-use types based on their overall suitability.

Model concepts for the environment sub-system

The biophysical environment is represented by either static maps (elevation, rivers), or dynamic maps (soil parameters, land cover), which are updated every simulated annual time step. The dynamics of natural vegetation and crops are represented by different, partly modified modules of the DAYCENT model (Parton et al., 1998; Stehfest et al., 2007), which simulates plant growth, the production of crop yields and resulting changes in soil parameters in daily time steps. Like other models, DAYCENT employs several component models to simulate different processes such as evapotranspiration and runoff, nutrient uptake and turnover, but also agricultural management in terms of irrigation, fertilisation and ploughing. Output of the DAYCENT model is usually aggregated to monthly or annual values and transferred to the land-use model. Weather and crop management data, which are required to drive the DAYCENT model, are read from a scenario database. An additional component simulates the biodiversity of certain groups of organisms, currently based on an empirical model (Klein et al., 2003).

Model integration

One of the novel features of the SITE modelling framework is the close coupling of the simulated environmental or biophysical dynamics and the decision making. As a consequence, the simulated decision makers can take environmental changes into account, irrespective whether they are caused by the simulated use and management of the land (e.g. intensification of agriculture or over-exploitation) or by global change processes. Similarly, the simulated decision makers are informed about land-use related changes in biodiversity, e.g. of pollinators. The current version of SITE uses the (negative) impact on crop yields as a feedback, because it is known that changes in biodiversity as such are not meaningful to Indonesian farmers, whose reasoning and decision making are simulated. SITE calculates a number of environmental indicators (land-cover change, productivity of natural vegetation, soil fertility, trace gas emissions, etc.) and socio-economic indicators (crop yields, gross margins, land-use change, encroachment of protected areas, etc.), providing an integrative insight into land-use dynamics and their potentially beneficial or detrimental consequences for society and environment.

3.8 SYPRIA model

Model purpose and application

SYPRIA (Southern Yucatán Peninsular Region Integrated Assessments) is a spatially explicit model to simulate scenarios of land-use change in the Southern Yucatán Peninsular region (∼ 22.000 km²) of Mexico. SYPRIA falls into the class of agent-based models (Parker et al., 2003). It is a probabilistic model that produces likelihoods of change. Consequently a large number (100–160) of Monte Carlo runs is needed to produce stable results (Manson, 2006*). Conceptually, the model consists of three components: actors, institutions and environment. Spatial scale levels are administrative units (ejidos) and an underlying 28×28 m grid, which reflects the unit of observation of the applied satellite imagery from the Thematic Mapper sensor (Turner II et al., 2001). The goals of the model are to examine links between land manager decision-making strategies and environmental and institutional factors, and furthermore to explore scenarios that relate to land use, population growth, agricultural systems and institutional influences on household decision making (Manson, 2005*). Results from a scenario analysis are presented by Manson (2006*). The described scenarios concentrate on the influence of population growth on the quantity of land-use change and on the influence of cultivation strategies (shifting cultivation vs. market oriented) and land tenure policies on the location of land-use change.

Model concepts for the human sub-system

The human sub-system is represented by actors (smallholder households) and institutions as proxy for social systems (Manson, 2005). Both are implemented as software-agents. Land-use decisions are made by the actor-agents on grid level and are influenced by the boundary conditions and constraints computed by institution agents and by the local environmental conditions. Each actor-agent makes location decisions for different agricultural activities (e.g. the cultivation of subsistence or commercial crops). The land-demand of each household for these activities is defined probabilistically within a certain range by an institutional agent (see below). Land-use change is computed in two steps. First, each actor-agent determines the suitability of each raster cell using multi criteria evaluation by a linear utility function in addition with Boolean constraints (Eastman et al., 1995). Utility factors include environmental factors on grid level like soil quality and distance to roads and markets. Constraints consider land-use limitations that are provided by institutions (see below). In the second step, each household chooses a number of cells to fulfil its land demand by maximizing the aggregate suitability. Here, each household is characterized by an individual location choice strategy (specific weights for the utility factors). These decision strategies are computed by genetic programs based on location data obtained for a calibration period. Actor decision making is influenced by institution-agents that define boundary conditions and constraints. A population-agent is responsible for the number of active actor-agents in each ejido, dependent on the observed population density and the future population growth. Agents on the ejido-level assign land demands to households (see above) and define spatial constraints for actor land-use decisions by preventing forest and natural reserves from agriculture and by deciding whether actor-agents are allowed to locate their land demand outside their home ejido. Furthermore, a market-agent makes available to all actors travel-cost surfaces for markets, population centres and roads.

Model concepts for the environment sub-system

A generalized cellular automata model is used to simulate the environment sub-system on grid level. Implemented processes include changes in soil fertility and vegetation succession. Both are defined by transition rules. According to Manson (2006), the change of soil fertility depends on land cover and soil fertility of the past time step, the soil type with respect to its value for agriculture, the duration of the cells’ current land-use type and the amount of fertilizer applied by actor-agents. Moreover, vegetation succession is calculated as the transition between different succession stages, taking into account the time since the last transition, the total number of neighbouring cells acting as seed sources and the fallow cycle dynamic that affects soil fertility. Other parts of the environment, which are relevant for actor decision making, like elevation, terrain slope and precipitation are treated as static variables.

Model integration

As institutions primarily provide constraints and boundary conditions for the actors, there is no feedback from actor decision making implemented. In contrast, the actor and environment components are tightly coupled. The effects of land use on environmental properties like changes of soil fertility by agricultural management and on natural vegetation (fallow cycle) directly feed back to the actor decision making process. For instance, decreasing soil fertility due to long term crop cultivation without additional fertilizer use makes a grid cell less attractive for further agricultural use.