Machine Learning Applications in Sea Level Prediction

Machine learning has emerged as a powerful toolkit for modelling and forecasting sea level changes by harnessing vast observational and modelled datasets. Traditional approaches relied on linear statistical methods or physics-based simulations, which can struggle to capture complex, nonlinear interactions between oceanographic, atmospheric and climatic drivers. By contrast, machine learning techniques—including neural networks, ensemble algorithms and data-driven clustering—can learn patterns directly from time series of tide gauges, satellite altimetry and environmental covariates. These methods have been applied to regional forecasts, global extreme‐value estimation and decomposition of seasonal, trend and residual components. Hybrid frameworks that combine autoregressive models with deep-learning elements offer enhanced performance by isolating deterministic variability before modelling nonlinearity. Advanced workflows also integrate topological data analysis and dimensionality-reduction schemes to identify region-specific modes of variability, improving local accuracy. Collectively, these innovations provide near-term projections (monthly to multi-year horizons), probabilistic uncertainty estimates and interpretable insights that inform coastal adaptation, hazard assessment and infrastructure planning worldwide.

Research from Nature Portfolio

Recent studies have demonstrated the capability of machine learning to predict regional coastal sea level variability by leveraging ocean temperature proxies. A supervised learning framework was developed to model thermosteric contributions to sea level at coastal tide gauges across multiple basins. The approach uses key temperature estimates as inputs to train ensemble neural networks, producing skillful forecasts at timescales from months to several years. Model evaluations reveal strong agreement with observed records in areas dominated by internal climate variability, enabling tailored projections of near‐future sea level tendencies. Moreover, this work illustrates how uncertainty quantification can be integrated into data-driven predictions, supporting robust decision-making for coastal protection strategies.

Research from all publishers

A multi-layer clustering framework has been applied to global subsurface ocean temperature data to extract regional modes of variability associated with sea level fluctuations. By combining k-means variants with empirical orthogonal function analysis, researchers identified depth-layer contributions that explain regional sea level anomalies, facilitating automated detection of subtle climate patterns and improving regional forecast skill. Another study employed ensembles of site-specific artificial neural networks to estimate total sea level, including tides and storm surges, at over 600 tide gauges worldwide. These networks capture nonlinear interactions—including tide-surge and tide-tide dynamics—yielding consistent probabilistic forecasts that outperform traditional linear regressions. In the South Pacific, a hybrid deep-learning pipeline decomposed tide-gauge records via adaptive noise filtering before using convolutional neural networks and gated recurrent units for trend and residual prediction. This CEEMDAN-CNN-GRU model achieved over 96% correlation accuracy, demonstrating the value of combined decomposition and recurrent architectures for small-island sea level projections.

Machine Learning Applications in Sea Level Prediction publication trend

The graph below shows the total number of articles in machine learning applications in sea level prediction across all publications each year (not limited to Nature Index journals).

Technical terms

Artificial Neural Network (ANN): A computational model inspired by biological neurons, consisting of interconnected layers that learn complex nonlinear relationships through weighted connections and activation functions.

Long Short-Term Memory (LSTM): A recurrent neural network variant designed to capture long-range temporal dependencies in sequence data through gated mechanisms that regulate information flow.

Convolutional Neural Network (CNN): A deep-learning architecture that applies convolutional filters to extract hierarchical spatial or temporal features, often used for time-series and image data.

Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN): A signal processing technique that decomposes non-stationary time series into intrinsic mode functions and residuals, enhancing the subsequent prediction of trend and irregular components.

Empirical Orthogonal Function (EOF) Analysis: A statistical method for decomposing spatial–temporal datasets into orthogonal basis patterns (modes) ranked by explained variance, often used to identify dominant variability structures.

References

1. Unveiling Regional Climate Patterns Through Global Subsurface Ocean Temperature Data: An AI Multi-Layer Analysis Framework. Earth Systems and Environment (2024).
2. Estimation of global coastal sea level extremes using neural networks. Environmental Research Letters (2020).
3. Predicting regional coastal sea level changes with machine learning. Scientific Reports (2021).
4. Sea Level Prediction Using Machine Learning. Water (2021).