Machine learning based estimation of land productivity in the contiguous US using biophysical predictors

Abstract

Estimation of land productivity and availability is necessary to predict land production potential, especially for the emerging bioenergy crop production, which may compete land with food crop production. This study provides land productivity estimates in the contiguous United States (CONUS) through a machine learning approach. Land productivity is defined as the potential in producing agricultural outputs given biophysical properties including climate, soil, and land slope. The land productivity is approximated by the potential yields of six major crops in the CONUS, i.e. corn, soybean, winter wheat, spring wheat, cotton, and alfalfa. This quantitative relationship is then applied to estimating the availability of marginal land for bioenergy crop production in the CONUS. Furthermore, the levels of uncertainties associated with land productivity and marginal land estimates are quantified and discussed. Based on the modeling results, the total marginal land of the CONUS ranges 55.0–172.8 mha, but the 95% inter-percentile distance of the estimated productivity index reaches up to 60% of its expected value in data-scarce regions. Finally, in a cross-check analysis, marginal lands estimated based on biophysical criteria are found to be comparable to those based on an economic criterion.