Projecting impacts of extreme weather events on crop yields using LASSO regression

Abstract

Extreme weather events are recognized as major drivers of crop yield losses, which threaten food security and farmers’ incomes. Given the increasing frequency and intensity of extreme weather under climate change, it is crucial to quantify the related future yield damages of important crops to inform prospective climate change adaptation planning. In this study, we present a statistical modeling approach to project the changes in crop yields under climate change for eight majorly cultivated field crops in Germany, estimating the impacts of nine types of extreme weather events. To select the most relevant predictors, we apply the least absolute shrinkage and selection operator (LASSO) regression to district-level yield data.

The LASSO models select, on average, 62% of the features, which align with well-known biophysical impacts on crops, suggesting that different extremes at various growth stages are relevant for yield prediction. We project on average 2.5-times more severe impacts on summer crops than on winter crops. Under RCP8.5, crop yields experience a mean change from −2.53% to −8.63% in the far future (2069–98) for summer crops and from −0.80% to −2.88% for winter crops, without accounting for CO₂ fertilization effects. Heat impacts are identified as the primary driver of yield losses across all crops for 2069–98, while shifting precipitation patterns exacerbate winter and spring waterlogging, and summer and fall drought.

Our findings underscore the utility of LASSO regression in identifying relevant drivers for projecting changes in crop yields across multiple crops, crucial for guiding agricultural adaptation. While the present analysis can identify empirical relationships, replicating these findings in biophysical models could provide new insights into the underlying processes.