Coupling Environmental, Social and Economic Models to Understand Land-Use Change Dynamics in the Mekong Delta

Abstract

The Vietnamese Mekong Delta has undergone in recent years a considerable transformation in agricultural land-use, fueled by a boom of the exportation, an increase of population, a focus on intensive crops, but also environmental factors like sea level rise or the progression of soil salinity. These transformations have been, however, largely misestimated by the 10-year agricultural plans designed at the provincial levels, on the predictions of which, though, most of the large-scale investments (irrigation infrastructures, protection against flooding or salinity intrusion, and so on) are normally planned. This situation raises the question of how to explain the divergence between the predictions used as a basis for these plans and the actual situation. Answering it could, as a matter of fact, offer some insights on the dynamics at play and hopefully allow designing them more accurately. The dynamics of land-use change at a scale of a region results from the interactions between heterogeneous actors and factors at different scales, among them institutional policies, individual farming choices, land-cover and environmental changes, economic conditions, social dynamics, just to name a few. Understanding its evolution, for example, in this case, to better support agricultural planning, therefore requires the use of models that can represent the individual contributions of each actor or factor, and of course their interactions. We address this question through the design of an integrated hybrid model of land-use change in a specific and carefully chosen case study, which relies on the central hypothesis that the main force driving land-use change is actually the individual choices made by farmers at their local level. Farmers are the actors who decide (or not) to switch from one culture to another and the shifts observed at more global levels (village, district, province, region) are considered, in this model, as a consequence of the aggregation of these individual decisions. The central component of our hybrid model is then an agent-based model of farmers, provided with a sophisticated mechanism of decision-making that is influenced, at different degrees, by their perception of the contexts in which they act or interact with other actors. The economic context, accessible by them through the market prices of crops, plays a role, as well as the changes observed or forecasted in their physical context (land-cover changes, salinity rise) or the decisions made by others in their social context (neighbors, family members, opinion leaders). The model of farmers is coupled, through this decision-making mechanism, with other independent sub-models, each of them carrying out a realistic description of one of these contexts. Since the dynamics depicted in these sub-models obey to different logics, operate at different scales and rely on different data, they are represented using appropriate modeling techniques: the spatial model is based on GIS information on parcels, soils, and rivers; a cellular automaton is used to account for the evolution of land-cover changes and the diffusion of salinity; an aggregated mathematical model represents the fluctuation of prices on the regional and national markets; and a graph-based social network model is used to represent familial networks of influence. Beyond the descriptions of these models, the paper is organized around a discussion about the two main outcomes of this research work. The first one is applicative: the way we have calibrated, coupled together, and experimented in different scenarios these five models is presented and we show that some findings obtained with the resulting hybrid model could not have been obtained with more traditional techniques. The second one is methodological: the underlying co-modeling architecture we used for declaring and running this assembly of heterogeneous models, implemented in the GAMA modeling and simulation platform, is presented and we show how it can be generalized to arbitrary hybridizations of models.