A physics-guided integrated global to regional method for high-resolution weather forecasting

High-resolution regional forecasts support disaster mitigation and sectoral decisions, yet numerical regional models remain costly. Here we introduce AERO-ODE, a physics-guided framework that combines global dynamical information, parameterization increments, and regional terrain constraints. From global initial conditions, AERO-ODE generates 72 h regional forecasts at hourly intervals and 3 km resolution in approximately 21 s, covering pressure-level Z , T , S , U and V and near-surface MSLP, U 10 , V 10 and T 2 m , without separate global surface lateral-boundary inputs. At 48 h, the median reductions in RMSE across the 29 outputs are 29–32% and 28–33% compared with WRF-ARW and YingLong-WRF, respectively; all reductions relative to WRF-ARW are significant in both regions ( p < 0.05). Median anomaly correlation is 0.04–0.05 higher than WRF-ARW, while extreme-event CSI and ETS improve for most variables. AERO-ODE generally shows lower errors than interpolated NeuralGCM, IFS and PanGu-Weather fields. These results demonstrate rapid, high-resolution regional forecasting from global dynamical information and regional constraints.