Predicting coastal variations in non-storm conditions with machine learning

Abstract Coastal erosion, driven by natural factors and human activities, is a major threat to vulnerable regions like Narrabeen, Australia. This study investigates shoreline changes, berm crest elevation variations, and horizontal berm crest positions under non-storm conditions. Using a decision tree algorithm, key features influencing these phenomena were identified. For shoreline changes, berm width changes (∆BW), berm slope, sea level rise (SLR), and wave breaking index (ζ) were critical. Berm crest elevation was linked to BC height, ∆xShoreline, ∆xBC, and wave power (P), while horizontal berm crest positions were influenced by BW, berm slope, ∆yBC, BC height, wave energy (E), SLR, and ζ. The feedforward neural network (FNN) algorithm was then applied to predict these objectives. Shoreline changes were predicted with a root mean squared error (RMSE) of 3.3 m and R 2 of 92% (DS4 scenario). Berm crest elevation predictions achieved an RMSE of 0.35 m and R 2 of 75% (DY4 scenario), while horizontal berm crest positions reached an RMSE of 9.28 m and R 2 of 85.8% (DX7 scenario). These results demonstrate that parameter classification via decision trees enhances neural network predictions. The FNN proved to be a reliable tool for forecasting coastal dynamics, supporting effective monitoring and management strategies.