A stochastic discrepancy data-physics fusion framework for tropical cyclone model and applications to typhoon wind hazard assessment

This study proposes a stochastic discrepancy framework to enhance physics-based tropical cyclone (TC) models and risk assessments by systematically integrating observational data.Current deterministic TC models exhibit structural biases from oversimplified physics and fail to quantify epistemic uncertainties inherent in their formulations.We reframe existing models as prior knowledge and derive parameter-efficient stochastic governing equations to characterize their discrepancies against TC observations.Validated through hindcasts of historical TCs in the Western North Pacific, the framework improves track and intensity simulations while quantifying uncertainty.For wind hazard assessment, our estimated wind speed generally agrees with the code recommendations with well calibrated confidence intervals to include results from alternative models.Overall, the proposed framework provides a principled approach to enhance the physics-based TC models, paving the future way for more informed TC simulation under changing climates.