Enhancing Regional Sea Level Predictions: A Unified Structural Equation Modeling Approach

Accurate forecasting of regional sea-level rise is becoming increasingly vital as coastal communities face the intensifying impacts of climate change.Traditional physics-based models, although grounded in well-established scientific principles, require strict boundary conditions, and substantial computational resources.Conversely, many current machine learning models, although innovative, often suffer from limited interpretability and offer only short-term predictions.This study introduces a Unified Structural Equation Modeling (uSEM) framework to enhance regional sea level forecasting.Unlike black-box approaches, uSEM provides a transparent and interpretable structure that explicitly models the relationship between the Global Mean Sea 1 Downloaded from https://spj.science.orgon October 27, 2026 Level (GMSL), regional climatic drivers, and Relative Sea Level (RSL).We employ equation-byequation LASSO variable selection to automatically learn the optimal physical structure for each location, followed by Maximum Likelihood Estimation to ensure robust parameter identification.By utilizing a recursive forecasting scheme with full covariance propagation, the model generates long-term projections through 2100, with quantified uncertainty.Compared to the projections from the Intergovernmental Panel on Climate Change (IPCC), our data-driven approach provides greater sensitivity to regional variability by capturing local climatic and geodetic factors in sea level trends across global coastlines.Furthermore, a spatial clustering analysis revealed distinct geographical patterns of sea level responses, underscoring the pronounced regional heterogeneity that global models can overlook.Our framework provides a scalable, interpretable, and computationally efficient solution for long-term predictions.