A generalized parameter calibration framework for typhoon wave models using pre-optimization dimensionality reduction in the South China Sea

Accurate simulation of wave fields under extreme typhoon sea states is of great significance to ocean engineering and disaster prevention and mitigation. However, third-generation wave models still exhibit considerable uncertainty under strong wind conditions, and their empirical parameters lack systematic calibration. Targeting the South China Sea (SCS) as the study area, this paper proposes an automated parameter calibration framework based on “Morris global sensitivity analysis and Tree-structured Parzen Estimator (TPE) Bayesian optimization” utilizing the ST6 source term package of the WAVEWATCH III (WW3) model. First, a sensitivity analysis is conducted on nine key physical parameters, identifying CDFAC, SINWS, SINA0, SWLB1, and GAMMA as the dominant parameters. Subsequently, aimed at multi-typhoon joint calibration (Kompasu, Maon), an objective function is constructed based on the mean root mean square error (RMSE) from 10 typhoon buoy samples, and 200 optimization iterations are performed. The results indicate that the optimal parameter combination reduces the mean RMSE of the calibration set from 0.5167 m to 0.3800 m, an improvement of 26.5%, yielding a parameter combination significantly superior to the default configuration for the SCS. Validated by four independent typhoons (Nesat, Mangkhut, Molave, Saudel) involving 17 typhoon buoy samples, the independent validation set showed an improvement of 29.69% in the mean RMSE and an approximately 61.86% reduction in the mean absolute Bias, which verifies its robustness and generalization capability. This study provides an efficient and transferable parameter calibration method, and the obtained optimal parameters demonstrate excellent applicability for typhoon wave simulations in the SCS.