This study develops a real-time tsunami inundation prediction method using machine learning and data from Japan's extensive offshore tsunami observing system (S-net). The model, trained on 3093 hypothetical tsunami scenarios and tested on 480 unseen scenarios and three historical events, achieves accuracy comparable to physics-based models with a 99% reduction in computational cost. Direct use of offshore observations increases forecast lead time and eliminates uncertainties from tsunami source estimations.
