Method Considering Uncertainties of Tropical Cyclone and Storm Surge Forecasts Due to Physics Scheme Selections in the Weather Research and Forecasting Model

This study explored the use of a decision-tree-based method to optimize the selection of physics schemes in the Weather Research and Forecasting (WRF) model for tropical cyclone (TC) prediction to improve TC and subsequent storm surge prediction accuracy. First, sensitivity analyses were conducted for TCs in Japan using various combinations of physics schemes as an extension of Shirai et al. (Shirai et al. in Coast Eng J 64:506–532 5). Next, a comprehensive database was created that included those results and several variables considered important in the analysis for TC prediction performance of WRF, such as observed TCs and initial values of sea surface temperatures and other parameters. We then applied a random forest classifier (RFC) to the database to identify the optimal WRF physics scheme combination for predicting TC parameters. Specifically, the minimum central pressures (representing TC intensity) and the center positions of TCs (indicating TC track) were considered, as they significantly influence storm surge prediction. Using a leave-one-out evaluation method, we found that our approach enhances TC and subsequent storm surge prediction accuracy.