Development and Evaluation of an Ensemble Regional Storm Surge Forecasting Model for the Coasts of Korea

Abstract

This study developed an Ensemble Regional Tide⋅Storm surge Model (ETSM) with 26 atmospheric ensemble members to generate diverse storm surge prediction scenarios along the Korean coast. The predictive accuracy and probabilistic performance were evaluated through seasonal experiments in 2024 and case studies of typhoons from 2022 to 2024, comparing results with observed sea levels and the Regional Tide⋅Storm surge Model (RTSM). The seasonal analysis showed a weak positive bias along the east and south coasts and a weak negative bias along the west coast. While both models exhibited comparable RMSE at short lead times, ETSM demonstrated improved performance as lead time increased, reducing RMSE by approximately 12% compared to RTSM at 120-hour forecasts averaged over all coastal regions. In probabilistic verification, the Brier Score ranged from 0.24 to 0.27, with the lowest values along the South Coast. The ROC AUC ranged from 0.63 to 0.70, indicating that the ensemble prediction provides a reliable level of probabilistic skill. In the typhoon case analysis, the ETSM reduced RMSE by 25% and 4% compared to the RTSM for Typhoons HINNAMNOR (2022) and JONGDARI (2024), respectively. However, for Typhoon KHANUN (2023), the ETSM exhibited a larger RMSE, which was approximately 4 cm higher at the initial forecast lead time (0hr). This difference is attributed to the ETSM overestimated sea levels during the early stages of the forecast compared to the RTSM. Overall, the ensemble-based approach effectively captured the potential for storm surges better than the deterministic model, suggesting that an ensemble-based forecast can significantly enhance the reliability of storm surge forecasting systems along the Korean coast.