PINN-Based Atmospheric Thermal Forecasting for Digital Twins

This physics-informed neural network (PINN) forecasts potential temperature by constraining predictions with a pressure-coordinate thermodynamic advection-source equation. Evaluated against ERA5 reanalysis, it demonstrates RMSE improvements over standard neural network baselines in sparse-observation scenarios.