Projections of changes in extreme storm surges for European coasts using statistical downscaling
Abstract. Understanding future changes in extreme storm surges (ESSs) is critical for coastal risk assessment and adaptation. However, existing projections in Europe are often based on computationally expensive dynamical models, limiting ensemble sizes and thus confidence in projected changes. In this study, we develop a cost-effective statistical downscaling model (SDM) trained to replicate dynamically downscaled storm surges, enabling the generation of a pan-European ensemble of ESS projections based on 17 global climate models (GCMs) – substantially expanding previous efforts.
The SDM is trained on a storm surge hindcast and demonstrates stable skill across historical and future climates, effectively capturing projected ESSs changes given by dynamical simulations. Ensemble projections reveal robust multi-model mean (MMM) changes in the 10-year return level (RL10) of ESSs by 2100. Negative MMM changes are identified in the Mediterranean Sea (−7 %), Moroccan Atlantic coast (−10 %), and Danish Straits (−6 %), while positive changes of around +6 % are projected for the Celtic and Irish Seas, western Denmark, and the Gulf of Finland. Despite these robust signals, inter-model spread is substantial, with likely ranges (17th–83rd percentiles) extending from −25 % to +17 % across Europe, and changes of up to ±35 % in individual models. The southern North Sea and northern Baltic Sea emerge as low-confidence regions, marked by particularly strong inter-model spread. Higher return levels (e.g., 100-year) show larger changes but increased uncertainty.
Our results underscore the importance of extended ensembles in projecting ESSs in Europe and demonstrate the value of statistical models for applications that demand extensive simulations – such as climate projections based on large ensembles, multi-scenario climate analyses, and detection and attribution studies – which can complement computationally expensive traditional dynamical downscaling methods.