Using Machine Learning to Assess Yield Impacts of Crop Rotation

To overcome the constraints for policy and practice posed by limited availability of data on crop rotation, this paper applies machine learning to freely available satellite imagery to identify the rotational practices of more than 7,000 villages in Ukraine. Rotation effects estimated based on combining these data with survey-based yield information point toward statistically significant and economically meaningful effects that differ from what has been reported in the literature, highlighting the value of this approach. Independently derived indices of vegetative development and soil water content produce similar results, not only supporting the robustness of the results, but also suggesting that the opportunities for spatial and temporal disaggregation inherent in such data offer tremendous unexploited opportunities for policy-relevant analysis.