SCEWERO

25-26 June 2026 – Training for AI tools for extreme events

Training for AI tools for extreme events will be coordinated by Fondazione CMCC and delivered together with UANTWERPEN. It will consist of the AI-driven identification of large-scale patterns associated with extreme events for better seasonal forecasting. Building on the knowledge gained by Fondazione CMCC in previous projects (CLINT, CYCLOPS, etc.) relative to the improvement of the seasonal forecast of heatwaves and/or extreme precipitation over certain domains (e.g., European area, Mediterranean subbasins, Como Lake, etc), we aim to apply a similar approach over Romania sub-domains. The AI-driven association of large-scale conditions to observed local extreme events (lagged in space and time) in ERA5 data will be the base of the improved seasonal forecast. Then, instead of directly forecasting the probability of extreme events on the seasonal time horizon, we evaluate the probability of multiple large-scale conditions that lead to such extreme events. This is consistent with the model’s superior ability to represent large-scale conditions rather than single extreme events when compared to the observations.

Training for AI tools for extreme events will be coordinated by Fondazione CMCC and delivered together with UANTWERPEN. It will consist of the AI-driven identification of large-scale patterns associated with extreme events for better seasonal forecasting. Building on the knowledge gained by Fondazione CMCC in previous projects (CLINT, CYCLOPS, etc.) relative to the improvement of the seasonal forecast of heatwaves and/or extreme precipitation over certain domains (e.g., European area, Mediterranean subbasins, Como Lake, etc), we aim to apply a similar approach over Romania sub-domains. The AI-driven association of large-scale conditions to observed local extreme events (lagged in space and time) in ERA5 data will be the base of the improved seasonal forecast. Then, instead of directly forecasting the probability of extreme events on the seasonal time horizon, we evaluate the probability of multiple large-scale conditions that lead to such extreme events. This is consistent with the model’s superior ability to represent large-scale conditions rather than single extreme events when compared to the observations.  

This two-day training focuses on the application of machine learning (ML) techniques for the detection, analysis, and forecasting of climate extremes. The programme combines theoretical sessions with hands-on exercises, providing participants with practical skills in working with climate data and ML models.

The first day is dedicated to heatwaves and intense precipitation, covering the identification of drivers, forecasting approaches, and practical implementation using ML methods, including dimensionality reduction and model-based prediction.

The second day addresses drought detection using parametric and non-parametric indices, followed by applied exercises. The programme concludes with statistical downscaling techniques, focusing on the use of machine learning methods such as Empirical Quantile Mapping to extend seasonal forecasts to local scales.

The training is structured to encourage interaction, applied learning, and discussion, offering a comprehensive introduction to AI-based approaches in climate extreme analysis.

 

Program

Day 1 ~ Thursday, 25 June 2026

9:00 – 9:30  – Registration and welcome coffee

9:30 – 10:30 – ML for Heatwaves Detecting and forecasting heatwaves, and detecting their drivers

10:30 – 11:00 – Break

11:00 -13:00 – ML for Heatwaves: Hands-on Dimensionality reduction of predictors, introduction to ML models

13:00 – 14:30 – Lunch Break

14:30 – 15:00 – ML for Intense Precipitation – Extreme Precipitation intro Forecasting intense precipitation through identification of its drivers

15:00 – 16:30 – ML for Intense Precipitation: Hands-on

Precipitation ½ h hands on + 1h discussion

Selection of drivers and forecasting through ML models

Day 2 ~ Friday 26 June 2026

9:30 – 10:30 – Introduction to the detection of drought events. Parametric and non-parametric indices for drought detection.

10:30 11:00 – Break

11:00 -13:00 – Hands-on drought events. Hands-on estimating indices for drought events.

13:00 – 14:30 – Lunch Break

14:30 – 15:00 – Statistical downscaling using machine learning techniques. Using the Empirical Quantile Mapping to extend seasonal forecast applications to the local scale

15:00 – 16:00 – Statistical downscaling using machine learning techniques.
Hands-on Downscaling & Discussion on ML methods.
Using k-Nearest Neighbours to extend seasonal forecast applications to the local scale

16:00 – Closing activities