Agent-based approach to environmental assessment

In the assessment, we use an empirical agent-based model (ABM) that is capable of simulating the long-term consequences (i.e. to 2013) of CAP reform on land use and farming practices in a real agricultural region. This was done by extending an existing ABM of regional structural change in agriculture, the Agricultural Policy Simulator or AgriPoliS (Balmann, 1997; Happe et al., 2006), for the purpose. The agent-based approach allows us to represent important aspects of the heterogeneity of farms, and their behaviour in space and time. We provide here only a short description of AgriPoliS and our modelling assumptions. For more details, see the chapter by Brady et al. (2010) in these proceedings, or for full documentation, see Kellermann et al. (2008).

Overview of the AgriPoliS model

The observed population of farms in a region is modelled in AgriPoliS as a multiagent system in which individual farm-agent behaviours and their interactions — principally competition for land — are defined through an optimization framework, with land use resulting as an emergent property of the system. The “optimizing” behaviour of farm agents is modelled using mixed integer programming, which is well suited to the task of combining economic, ecological and biophysical aspects of landscape evolution. Anonymous survey data on individual farms (i.e. FADN) and regional economic statistics were used to calibrate the model to a real agricultural landscape (Sahrbacher and Happe 2008).

Spatial representation in AgriPoliS is by a two-dimensional grid of equally sized cells or plots (Happe, 2004). Five different landscape layers are used to represent the structure of agriculture and the landscape in each region (Kellermann et al., 2008).

  • • The ownership layer denotes the ownership or rental of a specific plot.
  • • The soil layer reflects the distribution of any number of different land or soil quality types, which determines what types of (endogenous) agricultural land use are feasible on a particular plot.
  • • The block layer replicates the distribution of contiguous areas of a particular land type that are separated from land of the same type, by either another land type or physical borders that are protected through say legislation (e.g. hedgerows, ditches, roads), and hence, for all intensive purposes, can be assumed to be permanent boundaries that are not affected by agricultural policy.
  • • The allocation layer represents the allocation of plots to farms and reflects farmers’ land rental decisions.
  • • The fifth layer reflects a farm’s cropping decisions, i.e. a field comprising a number of contiguous plots used for a particular activity (e.g. wheat).

Consequently, the modelling framework can simulate from policy to individual farms and changes in cropping patterns at the plot level based on farm-agent behaviour. In this idealized representation, all land uses other than agricultural ones, such as forestry, lakes, and urban development, are subsumed into a single plot type: non-agricultural land. This abstraction is based on the assumption that only agricultural land use is affected by changes in agricultural policy, and hence all other features of the landscape remain unchanged. A further simplification is that AgriPoliS models the landscape synthetically, rather than as the actual location of farms and land as seen on a map. Using a landscape calibration algorithm, AgriPoliS generates a statistically similar landscape based on the size distribution of agricultural blocks and non-agricultural land in the region. This approach captures some important characteristics of the actual landscape (field size distribution and fragmentation) while other characteristics are ignored (field shape).

< Prev   CONTENTS   Source   Next >