Speaker: Antonello Squintu, CMCC

Moderator: Annalisa Bracco, CMCC

The use of AI techniques in climate sciences allows the processing of large amounts of data and to handle non-linear relationships, opening new frontiers in the extreme-event analysis at different time horizons, from weather forecasts to climate projections. A fundamental step in these applications is the thorough identification of the precursor of the targeted events, namely the drivers. These may represent, on the one hand, known connections or, on the other hand, undocumented relationships that link variables in different locations and at different lags. Providing AI models with a slim and meaningful set of drivers increases the significance of the final results.

In this seminar, the selection of drivers via the Probabilistic Coral Reef Optimization (PCRO) for two case studies in Northern Italy is described, with a focus on its contribution to enhancing the scientific understanding of the phenomena. In the first case, we’ll show how the selected drivers contribute to the development of an AI module that can produce hybrid subseasonal forecasts of high sea level events in Venice. While in the second case we’ll show how the PCRO output can be employed to develop a driver-based validation of climate simulations and to construct storylines of heatwaves in a warmer climate.