Simulating regional grain yield distributions to support agricultural drought risk assessment

Abstract Most food insecure countries do not have long-term records of either agricultural drought or the impacts of agricultural drought on food security. This lack of data impedes famine early warning and crop insurance programs. One recent paper addresses this issue by using resampled rainfall data, a basic crop yield model, and linear regression to simulate distributions of grain yield. We expand on this process by incorporating flexible regression models and defining a set of criteria to test model performance. We also examine how well a model fit on national data can emulate yield distributions at regions within a country. We find that models with spatially varying coefficients are better able to simulate distributions than basic linear regression models. Generalized additive models also perform well but do not offer substantial improvement over varying coefficient models. We also find that simulated yield distributions are most accurate in higher producing regions that have lower within region diversity of yields.