Comparative analysis of storm surge reconstructions in the Western North Pacific implications for coastal flood risk assessment

This study evaluated four storm surge models: the Global Tide and Surge Model (GTSMv3.0), ERA20C neural network (ERA20C_nn), ERA20C multiple linear regression (ERA20C_ml), and 20th Century Reanalysis multiple linear regression (20CR_ml), using data from 160 tidal stations. Storm surges in the Western North Pacific cause significant economic damage and loss of life, highlighting the need for accurate storm surge predictions.

The results show that the ERA20C_nn model outperformed others, with the highest correlation to tide-gauge observations. The GTSMv3.0 model follows closely, although slightly less accurate. The ERA20C_ml and 20CR_ml models were less effective, especially in predicting extreme surges. The ERA20C_nn model also provided more reliable estimates for 100-year return surge levels, outperforming other models. These findings suggest that neural network-based models, particularly ERA20C_nn, are better suited for assessing coastal flood risks in the region.