Toward a data-driven approach for large-scale storm surge prediction from tropical cyclones – Waves Workshop 2025

This record contains the slides of the oral presentation (L1, Session L: Tropical Cyclones II) given at the 4th International Workshop on Waves, Storm Surges and Coastal Hazards, incorporating the 18th International Workshop on Wave Hindcasting and Forecasting, hosted by the Universidad de Cantabria in Santander, Spain, on 22–26 September 2025. Storm surges are among the deadliest hazards associated with tropical cyclones (TCs), but computational limitations restrict large-scale and long-term assessments, especially in countries with limited technical infrastructure. This presentation describes the data exploration that underpins a data-driven storm surge model for TCs (now called CycloNet). Tropical cyclone tracks well represented in ERA5 are combined with storm surge simulations from GTSM. The workflow includes (1) KM-DBSCAN, a tracking system that combines dynamic K-means and DBSCAN clustering to filter storm surge and its drivers; (2) event augmentation, which splits each TC track into multiple events and expands 199 TCs into about 70,000 events; (3) K-means clustering of storm surge hydrographs and TC characteristics into storm surge types; and (4) Random Forest feature importance, which identifies the TC variables that control surge peak and duration for each type. This work is part of the TRANSCLIMA project (Detection and Global Risk Assessment of Coastal Transitional Climate and Neo-Climate Regions in a Warming World), funded by the European Union's Horizon Europe research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 101111262 (HORIZON-MSCA-2022-PF-01).