Improved Estimation of Extreme Sea Levels via Non-asymptotic Statistical Methods for Coastal Hazard Assessment

Abstract. Understanding the likelihood of extreme sea level events intensified by climate change is vital for effective coastal management. This study focuses on the Mediterranean and part of the European Atlantic coastline. Because sea level records are often short, non-asymptotic extreme value modeling is more accurate, as it does not depend on large-sample limiting assumptions. However, selecting independent events (IEs) is critical and depends on the chosen threshold sea level (TSL) and time window width (TWW) used. Existing literature often lacks a definitive methodology for IE selection. This study aimed to fill this gap by proposing two methodologies for selecting IEs (overlapping and sorting approaches) and a methodology to find the optimal combination of TWW and TSL. The identified IEs were employed to model extreme sea level events utilizing the Peak Over Threshold-Generalized Pareto Distribution (POT-GPD) and the metastatistical extreme value distribution (MEVD) approaches. A cross-validation approach, along with statistical metrics, was used to rigorously assess performance. Findings indicate that MEVD outperforms POT-GPD and is effective in estimating extreme events with longer return periods (RPs). Overall, MEVD with the overlapping method proved more effective at predicting long RP events, although MEVD with the sorting approach had comparatively lower uncertainty.