Predicting beach profiles with machine learning from offshore wave reflection spectra

Tracking and forecasting changes in coastal morphology is vital for development, risk reduction, and overall coastal management. One challenge of current coastal research and engineering is to find a method able to accurately assess the bathymetry profile along the coast and key parameters such as slope and sandbars. Traditional bathymetry measurements are obtained through echo-sounding, which is time-consuming, hazardous and costly. Using a variety of simulated cases, we test the potential of machine learning and in particular Neural Networks to reconstruct the coastal bathymetry profile from offshore sensed waves, based on shore-based wave reflection. Features such as foreshore slope, curvature, sandbars amplitude and positions can be captured. • Machine learning, specifically ANNs, used to estimate nearshore bathymetry from wave spectra. • Direct bathymetry estimation achieves high accuracy (R 2 = 0.9), but struggles with detailed sandbar features. • Direct feature estimation predicts sandbar height (R 2 = 0.89), position (R 2 = 0.74) and width (R 2 = 0.71). • First steps in the development of a novel technique for beach profile monitoring from offshore waves.