Hybrid Physics-Data Driven Ocean-Atmosphere Coupling Model for Tropical Cyclone Track Prediction

Abstract:

Tropical cyclones (TCs) are severe marine meteorological disasters whose tracks are closely regulated by ocean-atmosphere coupling processes, such as sea surface temperature (SST) anomalies, latent heat flux exchange, and upper-ocean thermal structure. Traditional TC track prediction models, which rely on parameterized ocean-atmosphere coupling schemes or pure statistical learning methods, often fail to capture the complex nonlinear interactions between oceanic and atmospheric variables, leading to significant prediction errors, especially for long-lead-time forecasts (72 h and above). This study proposes a Hybrid Physics-Data Driven Ocean-Atmosphere Coupling (HPD-OAC) model for TC track prediction. The model integrates a physics-based ocean-atmosphere coupling core (derived from the Regional Ocean Modeling System (ROMS) and Weather Research and Forecasting (WRF) model) with a data-driven dynamic correction module (based on a Temporal-Spatial Attention Long Short-Term Memory (TSA-LSTM) network). Multi-source data, including satellite remote sensing SST, altimeter-derived sea level anomaly (SLA), in-situ buoy observations, and reanalysis data, are assimilated to optimize the initial conditions and coupling parameters of the model. Validation results based on 150 TC events in the Northwest Pacific and North Atlantic basins from 2000 to 2023 show that the HPD-OAC model reduces the average track prediction error by 35% (72 h lead time) and 42% (120 h lead time) compared to the traditional WRF-ROMS coupled model. For intense TCs (category 4-5), the prediction error reduction reaches 48% at 120 h lead time. Further analysis indicates that the model accurately captures the regulating effect of upper-ocean cold wakes on TC intensity and the steering flow adjustment caused by ocean-atmosphere heat exchange, which are key factors improving track prediction accuracy. Under future climate change scenarios (CMIP6 SSP2-4.5 and SSP5-8.5), the HPD-OAC model projects a westward shift of TC tracks in the Northwest Pacific by 0.8-1.2° longitude and an increase in TC landfall frequency along the East Asian coast by 15-22% by the end of the 21st century. This study provides a new hybrid framework for improving TC track prediction accuracy, enriching the research methods of computational Earth system dynamics, and offering important scientific support for marine disaster prevention and mitigation.