Abstract Estimating low‐probability storm surges is essential for flood risk management, yet conventional asymptotic extreme value methods use only a small subset of observations and therefore produce large uncertainty at poorly gauged and ungauged sites. We present MEVD‐S, a new framework that combines the Metastatistical Extreme Value Distribution with an original regionalization scheme to overcome data scarcity in rare storm surge analysis. MEVD‐S uses all observed events to infer annual maxima and incorporates nearby observations through weights inversely related to site heterogeneity, avoiding rigid homogeneous regions and fixed spatial boundaries. This allows information to be integrated seamlessly across space and time while better representing the continuity of storm‐surge‐generating processes. Evaluation with tide‐gauge records from North America, Europe, and Japan shows that MEVD‐S outperforms at‐site Generalized Extreme Value (GEV) estimates and yields predictions at ungauged locations, with bias and uncertainty comparable to those obtainable from long local records.

Read original article