Overview

Spatial explicit Agent-Based Models (ABM) within the specific agricultural context was pioneered by Balmann (1997) with the Agricultural Policy Simulator (AgriPoliS) model.

ABM have the benefit of catching the fundamental behaviour at the micro-level of the individuals farms, without the need of aggregating them in ``representative'' agents. Maybe even more important, ABM can catch the iterations of the heterogeneous farms when they deal with competition over common finite resources, e.g. land.

Boero (2006) and Parker (2003) have review several other ABM involving land use changes in various scientific areas, including agricultural economics, natural resource management, archaeology and urban simulation, but this section will briefly describes AgriPoliS as RegMAS borrows many concepts from it, in primis the utilisation of a profit-maximisation algorithm to derive farmers behaviours.

AgriPoliS allows to model heterogeneous farms behaviours under various external situations (typically, under different policy scenarios) and observe regional results by aggregating these micro-level behaviours.

In AgriPoliS agents are mainly farmers⁴. They have their own goals; in AgriPoliS, the farmer's objective is the maximisation of household income. To achieve this objective, farmers solve a Mixed Integer Programming (MIP) problem that, in some aspects, is specific to each farmer. Outside the linear programming problem, they can also decide to rent other agricultural plots or to release rented land.

Using a mixed integer linear programming approach to simulate each agent behaviour on one hand is very flexible, as it can cover the whole range of farm activities, from growing specific crops to investing in new machinery or hiring new labour units. Furthermore, it is simple to add new regional-specific activities.

On the other hand, however, linear programming techniques require a long calibration phase to assure a balanced choice of farm activities, avoiding unrealistic outcomes ⁵.

Any farmer in the model is a real farmer whose data are taken from farm-level datasets (in Europe, FADN) and explicitly associated to a spatial location. Due to privacy-protection regulations, however, we don't have access to the real farm localisation. Therefore, we have to distribute farms randomly in the virtual region. Space (i.e. location) is important in the model because it influences transport costs and indirectly makes the farmers interact each other, e.g. by competing for the same land plots.

AgriPolis, as it takes into account many aspects of a real farm, is a very complex model, with lot of code dedicated to cover specific aspects (e.g. quotas, generational changes, multi-years investments). A detailed description of AgriPolis can be found in Happe et al. (2004) or in Kellermann et al. (2007). While Happe et al. (2004) focus is on the methodological advantage of using ABM in agriculture as compared with other instruments as partial and general equilibrium models on one side and individual farm-level models on the other, Kellermann et al. (2007) details the latest implementation of AgriPoliS (2.0). In addiction to this two papers, Sahrbacher et al. (2005) describes AgriPoliS implementation over several case-study regions and Lobianco (2007) presents an adaptation of AgriPoliS for the Mediterranean regions, further adding some general background on agent-based modelling and to its motivations.

As AgriPoliS, RegMAS is spatially explicit, a characteristic that can not be neglected when modelling the agricultural sector. For example the spatial heterogeneity allows the model to associate on each plot a different rental price and investigate possible land abandonment phenomenas even when the land is on average profitable.

Differently from AgriPoliS, the spatial dimension is initialized from real land-use data, using satellite information, and plots are explicitly modeled in the decision matrix as individual resources.

As a further distinction, RegMAS has been designed from the ground-up to explicitly consider farmers as one type of several possible type of agents. In RegMAS farmers have sensitivity of the overall environment, including extra-agricultural variables. On a technical point, ``farmer'' agents in RegMAS derive from a more general type of ''spatial`` agents that in turn derive from a ''base`` type. Each agent type has its own ''manager`` agent that dialogue with a ''Super Agent Manager``. The formers are a sort of interface ''agent side`` while the latter implements the same interface on the program core side. In this way the model core doesn't need to know anything about agents internal logic. While this approach allows for rapid development of different agent types (only specific characteristics need to be modelled) at current RegMAS development stage only farmer agents are fully implemented.