Modeling drought effects on rainfed crop yields using probabilistic and machine learning approaches

Drought is a major hazard with significant impacts on agriculture, water resource availability, and terrestrial ecosystems. Under climate change drought events are expected to increase in frequency, severity, duration, and propagation with consequent impacts on crop yields. Given these circumstances, a thorough understanding of drought is needed to increase societal preparedness to drought effects on food production particularly in regions where agriculture is dominantly rainfed. Unfortunately, drought events remain very unpredictable suggesting the need to enhance the understanding of drought effects on rainfed crops. Hence, this study aims to examine the relationships between drought characteristics and rainfed crop yields. Particularly, the study uses probabilistic and machine learning (i.e., random forest) approaches to investigate the influence of standardized precipitation and evapotranspiration index (SPEI) severity and duration on the yield of corn, cotton, peanuts, and soybeans in the southeast region of the United State (US). County wise analyses were conducted for three contiguous southeastern States including North Carolina, South Carolina, and Georgia. Preliminary results outlined different performances depending on the approach, the counties, and the crops. Highly performing approaches could be considered for modeling drought effect on crops at county, State, or regional levels.