Integrated modeling of agricultural land use change in Romania: From retrospective causal analysis to future developments

Agricultural land use changed rapidly in Central and Eastern Europe following the collapse of socialism. Cropland is arguably the most dynamic land use class, with cropland abandonment typically exceeding cropland expansion. Yet to date there has been little empirical evidence on the rates, patterns, and processes of cropland change for Central and Eastern Europe. To remedy this, we integrated socioeconomic and environmental data and employed exploratory statistics with predictive simulations to study the causes of past changes, as well as to forecast future cropland development in Arges County, Romania. In a first step, we used spatially explicit logistic regressions to estimate the direction and strength of the influences underlying factors that led to a change in the extent of cropland. The regressions focused on the exogenous, underlying variables that foster land change, allowed us to rank the importance of factors, and was used to test causal hypotheses of land use change processes. In a second step, we calibrated artificial neural network models with the statistically significant variables to predict the likely spatial arrangement of future cropland change. Such pattern recognition techniques are computationally efficient tools for forecasting locations that are most likely to undergo future cropland changes, given a user-defined quantity of change. Both of the employed modeling approaches are commonly used in land change science and we show an empirical example of how they complement each other. The combination of exploratory and predictive findings are of particular importance for understanding and dealing with complex processes in regions such as postsocialist countries, where empirical evidence on the local driving factors and possible future developments is scarce,. The proposed multi-method modeling approach based on the spatial analysis of integrated human-environment data allows the generation of causal inferences that inform the development of land use change forecasts. We believe that combining both approaches generates insights that are greater than the sum of their parts.