MEGA-ENSEMBLE STORM SURGE PREDICTIONS BASED ON JMA MESO ENSEMBLE PREDICTION SYSTEM GPV

Abstract

 In this study, a mega-ensemble storm surge prediction system was developed using the JMA Meso Ensemble Prediction System (MEPS) and validated for Typhoon Lan (2023), which hit the main island of Japan. Since the 21 members of MEPS are not sufficient to represent the uncertainty of the storm surge, a total of 1,000 members were extended for the typhoon attribute parameters (e.g., typhoon center, central pressure, and maximum wind speed radius) using SMOTE, which is an oversampling technique used in the field of machine learning, and mega-ensemble storm surge predictions in Ise Bay were conducted using an empirical typhoon model and a one-layer storm surge model. The track forecast errors were classified into five patterns, and the frequency of sea level anomalies was visualized as a stacked histogram for each classification. The information would help to provide a flexible storm surge mitigation strategy considering the ongoing errors in typhoon track forecasts.