Concurrent climate extremes have severe consequences on societies, economies, and natural systems. Multi-hazard risk-oriented early warning systems are essential to reduce impacts, enhance preparedness, and boost adaptation. Yet, the growing volume and variety of spatio-temporal data combined with the increasing frequency of concurrent extremes pose challenges to the rapid detection and tracking of harmful events.
Artificial intelligence offers an opportunity to deal with these challenges, especially when interpretability and explainability are ensured. Here, we show how expert-driven and explainable artificial intelligence models can probabilistically detect multiple agriculture-related hazards. The models are trained using the work of agro-climatic experts who, over decades, operationally identified multiple climate hazards affecting agriculture in Europe. The models identify the main drivers leading to the detection of affected areas while effectively dealing with large datasets to provide probabilistic results and uncertainty estimation.
Results highlight the added value of expert-driven and explainable artificial intelligence models in supporting risk management as well as effective and sustainable adaptation, particularly when integrated into early warning systems and sectoral climate services. Grounded on expert-driven information, the models contribute to a better understanding of the complex dynamics behind the onset and spatio-temporal evolution of climate extremes and to enhanced trust in defining and communicating affected areas.
DRM Phases
Sectors
Risk drivers