Measuring Allocation Errors in Land Change Models in Amazonia

Top-down land change models use computer simulation to allocate demands for change over the spatial region under study. This paper presents a metric to estimate the goodness of fit of land change models. It focuses only on changed areas, preventing inflated goodness of fit values due to large fractions of the landscape remaining unchanged. We use the proposed metric to evaluate land change models for Amazonia. Despite large quality differences between them, all models have problems to predict new frontiers and expansion areas. The best model considered in this paper only performs slightly better than a simple model that predicts a cell’s deforestation based on the deforestation in neighboring cells.