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

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