Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural ocean forecasting from sparse satellite-derived observations: a case-study for SSH dynamics and altimetry data

Daria Botvynko, Pierre Haslée, Lucile Gaultier, Bertrand Chapron, Clement de Boyer Montégut, Anass El Aouni, Julien Le Sommer, Ronan Fablet

TL;DR

This study tackles the challenge of short-term global ocean forecasting from sparse satellite altimetry by introducing an end-to-end neural framework that forecasts sea surface height anomalies (SLA) and derived surface currents over a 7-day horizon. The approach fuses two advanced architectures, 4DVarNet (a neural data-assimilation interpolation model) and UNet (a convolutional image-to-image network), and trains them in an OSSE-based setting using synthetic nadir-like sampling on the GLORYS12 reanalysis. The models are evaluated against independent SLA and drifter-derived velocity data, and outperform the operational Mercator GLO12 baseline across lead times, with notable gains in high-variability and small-scale regimes. All experiments are conducted within the OceanBench benchmarking framework to promote reproducibility, and the results demonstrate the feasibility of end-to-end neural forecasting for operational, data-sparse oceanography.

Abstract

We present an end-to-end deep learning framework for short-term forecasting of global sea surface dynamics based on sparse satellite altimetry data. Building on two state-of-the-art architectures: U-Net and 4DVarNet, originally developed for image segmentation and spatiotemporal interpolation respectively, we adapt the models to forecast the sea level anomaly and sea surface currents over a 7-day horizon using sequences of sparse nadir altimeters observations. The model is trained on data from the GLORYS12 operational ocean reanalysis, with synthetic nadir sampling patterns applied to simulate realistic observational coverage. The forecasting task is formulated as a sequence-to-sequence mapping, with the input comprising partial sea level anomaly (SLA) snapshots and the target being the corresponding future full-field SLA maps. We evaluate model performance using (i) normalized root mean squared error (nRMSE), (ii) averaged effective resolution, (iii) percentage of correctly predicted velocities magnitudes and angles, and benchmark results against the operational Mercator Ocean forecast product. Results show that end-to-end neural forecasts outperform the baseline across all lead times, with particularly notable improvements in high variability regions. Our framework is developed within the OceanBench benchmarking initiative, promoting reproducibility and standardized evaluation in ocean machine learning. These results demonstrate the feasibility and potential of end-to-end neural forecasting models for operational oceanography, even in data-sparse conditions.

Neural ocean forecasting from sparse satellite-derived observations: a case-study for SSH dynamics and altimetry data

TL;DR

This study tackles the challenge of short-term global ocean forecasting from sparse satellite altimetry by introducing an end-to-end neural framework that forecasts sea surface height anomalies (SLA) and derived surface currents over a 7-day horizon. The approach fuses two advanced architectures, 4DVarNet (a neural data-assimilation interpolation model) and UNet (a convolutional image-to-image network), and trains them in an OSSE-based setting using synthetic nadir-like sampling on the GLORYS12 reanalysis. The models are evaluated against independent SLA and drifter-derived velocity data, and outperform the operational Mercator GLO12 baseline across lead times, with notable gains in high-variability and small-scale regimes. All experiments are conducted within the OceanBench benchmarking framework to promote reproducibility, and the results demonstrate the feasibility of end-to-end neural forecasting for operational, data-sparse oceanography.

Abstract

We present an end-to-end deep learning framework for short-term forecasting of global sea surface dynamics based on sparse satellite altimetry data. Building on two state-of-the-art architectures: U-Net and 4DVarNet, originally developed for image segmentation and spatiotemporal interpolation respectively, we adapt the models to forecast the sea level anomaly and sea surface currents over a 7-day horizon using sequences of sparse nadir altimeters observations. The model is trained on data from the GLORYS12 operational ocean reanalysis, with synthetic nadir sampling patterns applied to simulate realistic observational coverage. The forecasting task is formulated as a sequence-to-sequence mapping, with the input comprising partial sea level anomaly (SLA) snapshots and the target being the corresponding future full-field SLA maps. We evaluate model performance using (i) normalized root mean squared error (nRMSE), (ii) averaged effective resolution, (iii) percentage of correctly predicted velocities magnitudes and angles, and benchmark results against the operational Mercator Ocean forecast product. Results show that end-to-end neural forecasts outperform the baseline across all lead times, with particularly notable improvements in high variability regions. Our framework is developed within the OceanBench benchmarking initiative, promoting reproducibility and standardized evaluation in ocean machine learning. These results demonstrate the feasibility and potential of end-to-end neural forecasting models for operational oceanography, even in data-sparse conditions.
Paper Structure (13 sections, 13 equations, 7 figures, 4 tables)

This paper contains 13 sections, 13 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Problem statement and Related work
  3. Proposed approach
  4. End-to-end neural schemes for short-term forecasting
  5. OSSE-based training scheme
  6. Data Preparation
  7. Model Training
  8. Evaluation Metrics
  9. Results
  10. Benchmarked models
  11. Global-scale performance
  12. Regime-dependent performance
  13. Conclusion and Discussion

Figures (7)

  • Figure 1: Proposed end-to-end neural scheme for short-term forecasting: input satellite observations are fed to the end-to-end neural architecture providing a full SSH forecasted field.
  • Figure 2: Input and target of the proposed end-to-end neural scheme for short-term forecasting: (a) the input to the end-to-end neural model is a set of 14 past days of satellite observations and the corresponding target is (b) a set of 7 future days of complete SSH fields.
  • Figure 3: SLA predicted by proposed end-to-end neural forecasts 4DVarNet (left column) and UNet (right column) at leadtime 0 on 2023-01-18.
  • Figure 4: Benchmark of SSH forecasts (and derived SSC), provided by GLO12, GloNet, XiHe and proposed end-to-end neural forecasts 4DVarNet and UNet. The benchmark presents the following metrics: nRMSE score and average effective resolution of the SSH with respect to SARAL/AltiKa, % correctly predicted SSC angles and magnitudes with respect to drifters. Colors indicate relative gain in % compared to the baseline GLO12, with blue indicating better performances.
  • Figure 5: SLA reconstructed by OI-based DUACS product vs baseline forecast GLO12 and proposed end-to-end neural forecasts 4DVarNet and UNet at leadtimes 0 and 5 in the Gulf Stream on 2023-01-18 and 2023-01-23.
  • ...and 2 more figures