Climate extremes, including heatwaves, droughts, and storms, have severe societal impacts, making reliable predictions crucial in a warming world. The demand for accurate forecasts, particularly on timescales ranging from weeks to decades, is growing among policymakers and stakeholders. However, traditional climate models face limitations in predicting such events due to the inherent complexity of weather systems.
In recent years, AI has emerged as a game-changer for predicting extreme weather events. AI-based systems, especially those leveraging machine learning (ML) and deep learning (DL), can identify patterns in vast climate datasets.
A paper, recently published on WIREs Climate Change journal, explores AI’s potential to improve the prediction of extremes at the seasonal to decadal (S2D) timescale, and to reveal their links to large-scale and local drivers. By reviewing recent literature on AI applications for climate predictions of extreme events and the prospects brought by the combination of empirical and dynamical methods, it discusses the challenges of the data-driven approach and future perspectives, providing climate scientists with a state-of-the-art framework available for rigorous future applications.
The review work includes contributions from the AI4Drought project.