AI and the World's Hard Problems · Earth & climate science · arXiv · 2026-10-21
Continuous and Differentiable Neural Network Representations of the Surfzone Sea-Surface from Lidar Point Clouds
Multi-beam scanning lidar measures the sea surface as time-sequenced point clouds, which normally must be gridded, interpolated, or mapped before wave analysis. The authors instead train a SIREN neural network on 60 seconds of lidar data from a drone hovering over large (4 m) breaking waves in Monterey, California, yielding a continuous, differentiable, grid-free representation of the sea surface that can be evaluated, together with its derivatives, at any location and time. SIREN errors are comparable to instrument errors and insensitive to local data density, though the representation filters the highest-wavenumber variability such as splash-up and sea spray, and computing derivatives and estimating wave velocity from a space-time slice demonstrate its potential for directional wave analysis.