Spatial Disaggregation of Agricultural Production Data by Maximum Entropy

In this paper we develop a dynamic data-consistent way for estimating agricultural land use choices at a disaggregate level (district-level), using more aggregate data (regional-level). The disaggregation procedure requires two steps. The first step consists in specifying and estimating a dynamic model of land use at the regionallevel. In the second step, we disaggregate outcomes of the aggregate model using maximum entropy (ME). The ME disaggregation procedure is applied to a sample of California data. The sample includes 6 districts located in Central Valley and 8 possible crops, namely: Alfalfa, Cotton, Field, Grain, Melons, Tomatoes, Vegetables and Subtropical. The disaggregation procedure enables the recovery of land use at the district-level with an out-sample prediction error of 16%. This result shows that the micro behavior, inferred from aggregate data with our disaggregation approach, seems to be consistent with observed behavior.