WorksPhysics and Astronomy
Reconstruction and Forecast of Extreme Waves from a Single Observation
Abstract
The state space of a dynamical system can, under suitable conditions, be reconstructed from a single observational time-series through delay-coordinate embedding. However, reconstructing the spatiotemporal evolution of nonlinear water waves from highly sparse measurements remains a considerable challenge. Here, by combining deep learning, compressed sensing, and the self-similarity properties of unidirectional nonlinear water wave evolution equation, we show that the evolving surface elevation of a highly nonlinear wave field can be reconstructed from observations collected at a single measurement location. We subsequently use these reconstructions to forecast rogue waves associated with self-focusing soliton dynamics from point observations obtained in controlled water wave tank experiments, demonstrating excellent agreement with measurements of the fully evolved wave field. Our results show that extreme wave events can be forecast from remarkably sparse observations and provide a framework that may be extended to other nonlinear dispersive systems governed by the interplay of dispersion and nonlinearity.