Sea Surface Height Dataset for Western Pacific Marginal Seas (2010-2024): Physics-Inspired Geometrically Guided Deep Learning for SSH Prediction

Sea Surface Height Dataset for Western Pacific Marginal Seas (2010-2024): Physics-Inspired Geometrically Guided Deep Learning for SSH Prediction Description This repository hosts a curated regional sea surface height (SSH) dataset designed for spatiotemporal predictive learning, physics-informed neural networks (PINNs), and oceanographic forecasting. The dataset comprises long-term, high-quality gridded SSH fields over two dynamically distinct marginal seas in the Western Pacific: the East China Sea/Yellow Sea (ECS/YS) and the South China Sea (SCS), spanning from January 1, 2010, to December 31, 2024. 1. Data Source and Domains The dataset is extracted from the gridded Level-4 sea surface height product provided by the Copernicus Marine Environment Monitoring Service (CMEMS). This product integrates multi-satellite altimeter observations using optimal interpolation, offering high spatial continuity and suppressed high-frequency noise. The dataset is partitioned into two sub-regions, both resolved at a daily temporal resolution and a 0.125×0.125 spatial resolution (corresponding to 64×64 grid points per region): East China Sea / Yellow Sea (ECS/YS): Longitude: 122.0625°E to 129.9375°E Latitude: 26.0625°N to 33.9375°N South China Sea (SCS): Longitude: 111.0625°E to 118.9375°E Latitude: 14.0625°N to 21.9375°N 2. Preprocessing for Deep Learning To facilitate direct utilization in deep learning and spatiotemporal sequence forecasting frameworks, the dataset has been fully preprocessed: Land Masking: Land areas and islands are strictly masked based on geographical boundaries, with valid ocean cells preserved to avoid numerical contamination near coastlines. Normalization: The daily SSH values are normalized to the range of [0,1] based on the training set statistics to facilitate stable model convergence. Chronological Partitioning: The data is partitioned chronologically to avoid temporal data leakage: Training Set: January 1, 2010 – December 31, 2021 (12 years) Validation Set: January 1, 2022 – December 31, 2022 (1 year) Test Set: January 1, 2023 – December 31, 2024 (2 years) 3. Technical Specifications Data Format: NetCDF-4 (.nc) Spatial Dimensions: 64×64 grid points for each sea basin Temporal Coverage: 5,479 daily frames (including leap years) Variables: ssh (normalized sea surface height above the geoid), mask (binary land-sea mask) Research Context This dataset supports the experimental validation and physical diagnostics presented in the manuscript: "PIGG-SDNet: Physics-Inspired Geometrically Guided Network for Sea Surface Height Prediction" Usage and Citation Please cite the aforementioned paper when using this dataset for research or operational forecasting tasks. For questions regarding the PIGG-SDNet architecture or replication of the 10-day rolling forecasts, please refer to the associated source code or contact the authors.