RESUMEN
Sea level change, a major consequence of climate change, presents significant threats to coastal regions and demands precise, timely forecasting for effective management and adaptation. This review assesses methodologies and approaches essential for developing robust machine learning (ML) models for predicting and forecasting sea level change (SLC). Key findings reveal that artificial neural networks (ANNs), especially deep learning models and their hybrid variants, outperform traditional regression and simpler ML techniques in short-term sea level anomaly prediction. Supervised learning approaches dominate the field, while semi-supervised methods excel in short-term projections. Simpler models, such as regressions and support vector machines perform better with sufficient training data, however, often exhibit lower accuracy in handling complex, non-linear scenarios. The selection of relevant input variables, such as atmospheric, oceanic, and geological factors, significantly influences model accuracy, and the balance between training and testing data is crucial for avoiding overfitting and underfitting. This review also clarifies the distinction between ML prediction and forecasting as used in the literature. The study recommends that future research should focus on integrating physics-based general circulation models (GCMs) with ML techniques and exploring innovative methodologies to improve regional long-term forecasting, which is critical for effective coastal management and resilience.