Shoreline dynamics prediction using machine learning models: from process learning to probabilistic forecasting

Coastal zones are experiencing notable changes attributed to natural and anthropogenic effects. This study investigates the potential of machine learning (ML) in predicting shoreline changes, a developing field still in its early exploration phase. Traditional methods, while insightful, have faced challenges in terms of adaptability, accuracy, and computational demands. ML, as a data-driven approach, potentially offers flexibility, computational efficiency, and can avoid the constraints associated with physics-based models. This study aims to evaluate various machine learning models’ efficacy in predicting shoreline changes using synthetic data. Through comprehensive testing across one complex shoreline evolution scenario, this research identifies the ConvLSTM model—trained on 2D gridded data— as the optimal machine learning approach suited for addressing specific shoreline complexities and evolution patterns. This approach can learn shoreline evolution, predict it, and serve as a foundational component of a proposed method for probabilistic shoreline position prediction. Additionally, the study shows that the choice of ML model depends on the complexity of shoreline evolution and the desired level of accuracy.