Tides of tomorrow: Physics-informed neural operator (PINO) for 24-hour tidal forecasting in Manila Bay, Philippines

Abstract

This study presents a physics-informed neural operator (PINO) model for 24-hour tidal level forecasting in Manila Bay, Philippines. The model incorporates a linearized form of the shallow-water equation into its training loss, enabling it to learn from data while maintaining physical consistency and enhancing both stability and model generalization. Tidal gauge measurements, bathymetry data, and TPXO-generated true tidal fields were preprocessed and projected onto a structured 64 x 64 grid. The model was trained to forecast the next 24 hours of PINO-predicted tidal fields using the preceding 24-hour sequence and static bathymetry. The evaluation showed excellent agreement with both TPXO-generated true tidal levels and NAMRIA-observed tidal levels, achieving root mean square error (RMSE) of 1.25 cm and 11.06 cm, respectively. Percentage-based errors remained below 4%. Spatial plots confirmed consistently low errors across the Manila Bay domain. These results demonstrate that the PINO model can provide accurate, physically coherent short-term tidal forecasts and serve as a promising tool for operational coastal prediction and risk reduction applications.