Assessing the Robustness of Predictions in Spatially Explicit Models of Land Use

We propose an information-theoretic approach to assess the performance of a discrete choice model used to analyze land use and land use change. We show that our disaggregated measure can be used to compare robustness of predictions across land use categories and across models. Furthermore, a proper reformulation of the problem shows that a disaggregated (observation by observation) log-likelihood lends itself to an information theoretic interpretation, which allows comparisons performance across models.