CL-PINN: A Physics-Informed Neural Networks Framework for Equatorial Tsunami Wave Propagation in Geophysical Ocean

Wave dynamics are governed by linear and nonlinear partial differential equations, where prior physical knowledge in the form of differential equations plays a crucial role in simulating dynamical models. Traditional analytical models often drift over time due to noise in real-world data, while data-driven methods struggle to uncover underlying patterns when labeled data are scarce and frequently fail to yield meaningful scientific insights. Nonetheless, physics-informed neural networks bridge this gap by embedding governing physical laws directly into the learning process. This paper presents the CL-PINN model, a novel curriculum learning-based physics-informed neural networks to study travelling wave solutions of a geophysical Boussinesq-type equation modeling equatorial tsunami wave propagation. The analyzed wave structure exhibits explicit travelling wave solutions with linear propagation speed and small-amplitude near field variations. Our approach achieves at least a 74% improvement over the baseline models, reduces computational cost by 40%, and attains an error as low as 10-4 compared to the ground truth. This demonstrates the potential of physics-informed neural networks for modeling oceanic wave phenomena.