Physics-Guided Spatiotemporal Learning for Coastal Wave Peak Period Estimation from Video

A new Physics-Guided Deep Spatiotemporal Learning Framework has been proposed for estimating nearshore wave peak periods from passive video streams, addressing the limitations of traditional monitoring systems. The framework integrates automated region-of-interest detection, multi-stage Sim-to-Real transfer learning, and physics-informed regularization, with evaluations showing that transformer-based architectures excel in prediction accuracy while lightweight recurrent-convolutional models provide greater temporal stability. This approach enhances the feasibility and cost-effectiveness of coastal wave monitoring, making it a valuable tool for practitioners in oceanography and coastal engineering.