Fine-Scale Land Use Allocation Using Maximum Likelihood

Despite the growing need for fine-scale resolution land allocation data on a large scale, land allocation data is typically available only at an aggregate level. We take a unique perspective by building a probability model designed to predict land use patterns at a fine-scale resolution, based on biophysical land characteristics observable on a fine-scale as well as maintaining shares of land allocation within a region that are observable only at an aggregate level. A sampling strategy is developed to reduce the computational burden brought on by the dimensionality of the combinatorics involved in evaluating our likelihood model. Empirical results from an application of our model to predicting maize throughout North, Central and South America show that common biophysical factors are sufficient for predicting maize production at a fine-scale. This method is generally applicable to other cases in which only aggregated level data is available but fine-scale data is desired.