Identifying compound meteorological drivers of extreme wheat yield loss using Lasso regression

Compound weather events may lead to extreme impacts that can affect many aspects of society including agriculture. The identification of the underlying mechanisms that cause extreme impacts, such as crop failure, is of crucial importance to improve their understanding and forecasting. Here we investigate whether key meteorological drivers of extreme yield loss can be identified using Least Absolute Shrinkage and Selection Operator (Lasso) in a model environment. 
We use yearly wheat yields as simulated by the APSIM crop model driven by 1600 years of daily weather data from a global climate model (EC-Earth v2.3) under present-day conditions for the Northern Hemisphere. We define extreme yield loss as years with yield below the 5th percentile. We apply logistic Lasso regression to predict whether weather conditions during the growing season lead to crop failure. Lasso selects the most relevant variables from a large set of predictors that best explain the target variable via regularization. Our input variables include monthly averaged values of maximum temperature, vapour pressure deficit and precipitation as well as established extreme event indicators such as maximum and minimum temperature during the growing season, diurnal temperature range, total number of frost days, and maximum five-day precipitation sum.
We obtain good model performance in Central Europe and the American Corn Belt, while yield losses in Asian and African regions are less accurately predicted. Model performance and mean wheat yield strongly correlate, i.e. model performance is highest in regions with relatively large mean yield. Based on the selected predictors, we identify regions where crop loss is predominantly influenced by a single variable and regions where it is driven by the interplay of several variables, i.e. compound events. Especially in the Midwest and Eastern regions of the USA, several variables are required to correctly predict yield losses. This illustrates the importance of accounting for the interplay of various weather conditions over the course of the growing season to be able to determine crop yield losses more precisely.
We conclude that the Lasso regression is a useful tool to detect the compound drivers of extreme impacts, which can be applied for other impact variables such as fires or floods. As the detected relationships are of purely correlative nature, more detailed analyses are required to establish the causal structure between drivers and impacts. Furthermore, using the same model environment, the robustness of the identified relationships will be tested in a climate change context.