Applying machine learning for drought prediction in a perfect model framework using data from a large ensemble of climate simulations

Abstract

Abstract. There is a strong scientific and social interest in understanding the factors leading to extreme events in order to improve the management of risks associated with hazards like droughts. In this study, artificial neural networks are applied to predict the occurrence of a drought in two contrasting European domains, Munich and Lisbon, with a lead time of 1 month. The approach takes into account a list of 28 atmospheric and soil variables as input parameters from a single-model initial-condition large ensemble (CRCM5-LE). The data were produced in the context of the ClimEx project by Ouranos, with the Canadian Regional Climate Model (CRCM5) driven by 50 members of the Canadian Earth System Model (CanESM2). Drought occurrence is defined using the standardized precipitation index. The best-performing machine learning algorithms manage to obtain a correct classification of drought or no drought for a lead time of 1 month for around 55 %–57 % of the events of each class for both domains. Explainable AI methods like SHapley Additive exPlanations (SHAP) are applied to understand the trained algorithms better. Variables like the North Atlantic Oscillation index and air pressure 1 month before the event prove essential for the prediction. The study shows that seasonality strongly influences the performance of drought prediction, especially for the Lisbon domain.