UAV Video-Based Estimates of Nearshore Bathymetry

Abstract

Nearshore bathymetry estimated from video acquired by a hovering UAV is compared with ground truth. Individual wave crests (distinguished from the breaking wave toe that can move down the wave front face) in video timestacks are determined with a deep-learning neural network and surfzone depth estimates are computed from the wave celerity. Time-2D spatial transforms (cBathy) are used to estimate wave phase speed and depth between the surfzone and 10m depth. Composite profiles (cBathyCT), formed by joining cBathy and crest-tracking solutions near the surfzone seaward edge, based on a newly determined 𝛾(𝑥) parameter, avoid the large cBathy errors associated with the onset of breaking. Incident wave heights were relatively constant on each day, but varied over days between 0.55 − 2.15m. Averaged over all 17-min hovers and cross-shore transects (130 total), surfzone depths errors were relatively small (average
root-mean-square error ⟨RMSE⟩ = 0.24m, ⟨Bias⟩ = −0.02m) after including a nonlinear correction to the linear phase speed. Between the seaward surfzone edge and 10m depth, errors are similar to previous cBathy studies: ⟨RMSE⟩ = 0.96m, ⟨Bias⟩ = 0.61m with the largest errors in deepest water. Beach profiles were generally similar for all 8 test days, concave up with a slight terrace (no sandbar) and small alongshore depth variations. Accuracy was lower on one transect with a shallow reef

Keywords: Bathymetry, Remote Sensing, Machine Learning, UAV