Prediction of shoreline wave setup on beaches using artificial neural networks

This study explores the application of artificial neural networks (ANNs) for predicting shoreline wave setup, a critical parameter in coastal flood risk assessment and infrastructure design. A dataset comprising 772 field observations, the largest dataset available for shoreline wave setup prediction, was used to train, validate and test two distinct models. ANN model 1 incorporates the deep-water wave height, deep-water wavelength and beach foreshore slope as input variables. ANN model 2 includes these variables as well as the median grain size. Both models were developed by using feedforward architectures, with training performed through backpropagation and hyperparameters optimized via the Optuna framework. The results demonstrate that the ANN models significantly outperform traditional empirical formulas in terms of predictive accuracy and generalizability. Further analysis reveals that the models perform robustly across different beach states, with particularly notable advantages for intermediate and reflective beaches. The incorporation of the median grain size as an input variable enhances the model performance across all sediment types, with substantial gains observed on medium to coarse sand beaches. Overall, this study offers reliable tools for shoreline wave setup prediction, thus contributing to coastal management and engineering applications.