The eiLab contributed to a study developing an explainable machine learning model to estimate intense Mediterranean cyclone activity from large-scale climate variables.

In the study, published in Journal of Geophysical Research: Machine Learning and Computation, the authors developed a convolutional neural network to estimate monthly Mediterranean cyclone activity, measured through Accumulated Cyclone Energy, which combines cyclone intensity, frequency, and duration.

The model outperforms both a simple climatology and a statistical benchmark. It identifies absolute vorticity as the most informative variable, followed by relative humidity and mean sea-level pressure, while also focusing on regions historically associated with Mediterranean cyclogenesis. The proposed approach offers a first data-driven tool for estimating Mediterranean cyclone activity and may support future analyses of seasonal forecasts and climate model projections.

The paper is available here: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2025JH000807

The code and data are available here: https://zenodo.org/records/15031785