Comparative Evaluation of Machine Learning Models for Sea-Level Prediction Using Multi-Mission Satellite Altimetry: A Case Study of Vietnam's Coastal Waters

Accurate characterization of coastal sea-level variability is essential for climate-risk assessment and adaptation planning. This study evaluates machine-learning approaches for estimating sea-level anomaly (SLA) in Vietnam's coastal waters using harmonized observations from six satellite altimetry missions—CryoSat-2, SARAL, Jason-3, Sentinel-3A, Sentinel-3B, and Sentinel-6 Michael Freilich—spanning 2016–2026. Following multi-mission harmonization and quality control, SLA (m) was defined as the prediction target. Five approaches—Deep ST-FNN, a Bayesian-optimized ensemble, boosted trees, random forest, and a conventional neural network—were evaluated using a common spatiotemporal-block validation framework to reduce spatial and temporal dependence between training and validation samples. Performance was assessed using R², RMSE, and MAE. Random forest achieved the best overall validation performance (R² = 0.1479, RMSE = 0.0407 m, MAE = 0.0275 m), closely followed by Deep ST-FNN (R² = 0.1440, RMSE = 0.0408 m, MAE = 0.0275 m) and the Bayesian-optimized ensemble (R² = 0.1425, RMSE = 0.0408 m, MAE = 0.0276 m). The satellite-based analysis revealed pronounced regional heterogeneity in sea-level behaviour, consistent in direction with independent tide-gauge evidence indicating a decreasing trend of approximately −2.02 mm yr⁻¹ in the Gulf of Tonkin and an increasing trend of approximately +3.73 mm yr⁻¹ in the Mekong Delta. This regional agreement provides independent support for the physical consistency of the modelling framework, although the modest R² values indicate that substantial local and short-term SLA variability remains unresolved. The results demonstrate the potential of integrating multi-mission satellite altimetry, machine learning, spatiotemporal validation, and independent tide-gauge observations for regional sea-level assessment and long-term coastal monitoring in Vietnam.