Year: 2025
Type: article
Abstract: The European Space Agency’s Sentinel-1 (S-1) satellite mission has captured more than 10 million images of the ocean surface using C-band synthetic aperture radar (SAR WV mode). While machine learning is a promising approach for detecting and quantifying various geophysical signatures in these images, scientists are limited by the cost of manual data annotation for any particular task. We propose to use contrastive self-supervised learning on the full archive of unannotated WV-mode images to train a semantic embedding model named WV-Net. In experiments, we show that WV-Net embeddings outperform those from models that were pretrained with natural images (ImageNet) on four downstream tasks: multilabel classification [0.96 average area under the receiver operating characteristic (AUROC) vs 0.95], wave height regression [0.50 root-mean-square error (RMSE) vs 0.60], near-surface air temperature regression (0.90 RMSE vs 0.97), and unsupervised image retrieval [0.41 class-averaged mean average precision (mAP) vs 0.37]. WV-Net embeddings also scale better in data-sparse settings, and fine-tuned WV-Net models are more robust to hyperparameter choices. The WV-Net foundation model is publicly available and can be adapted to a variety of data analysis and exploration tasks in geophysical research. Significance Statement Satellite imagery of the ocean from the Sentinel-1 (S-1) mission contains global and temporal information about ocean waves, atmospheric patterns, biological signatures, sea ice, and ships. However, the dataset is currently not fully utilized due to the size and complexity of the data. Our WV-Net foundation model makes this large dataset more accessible to researchers by providing a pretrained model that converts images into semantic vector embeddings, making the dataset amenable to analysis using advanced data-driven methods.
Institutions: University of Hawaiʻi at Mānoa, University of Washington, University of Washington Applied Physics Laboratory, University of New Hampshire, Centre National de la Recherche Scientifique, Ifremer, Université de Bretagne Occidentale, Institut de Recherche pour le Développement, Université de Bretagne Occidentale
Language: English
Subfield: Oceanography
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Open Access status: green
Funders: National Science Foundation, National Aeronautics and Space Administration, Institut Français de Recherche pour l'Exploitation de la Mer, European Space Agency, Division of Ocean Sciences, National Science Foundation Graduate Research Fellowship Program, NASA Earth Science Technology Office
Awards: Critical Early DKIST Science: Spectropolarimetric Inversion in Four Dimensions with Deep Learning, G2331673178, G356238214, G4354503405, Graduate Research Fellowship Program (GRFP), CC* Data Storage: KoaStore: A High Performance and Flexible Research Storage Resource, RII Track-4:FAST: An investigation of tropical cyclone intensity using synthetic aperture radars and complementary satellite ocean observations, G5737237736, G6801603785, G8037292183, G8040576089, THERE IS AN EVOLVING EARTH SYSTEM SCIENCE INTEREST TO CAPTURE UNRESOLVED MIXING PROCESSES IN BOTH ATMOSPHERIC AND OCEANIC BOUNDARY LAYER MODELING AND IN THE DIAGNOSIS OF THESE MODELS USING FIELD DATA SETS. ONE IMPORTANT EXAMPLE IS HOW TO INCLUDE THE IMPACT OF THE FREQUENTLY OCCURRING ORGANIZED SECONDARY CIRCULATION IN THE MARINE ATMOSPHERIC BOUNDARY LAYER AN EDDY FIELD HAVING SCALES OF 1-10 KM VARIABLE TWO- AND THREE-DIMENSIONAL (3D) FEATURES AND NONLOCAL MIXING CONTRIBUTIONS THAT IMPACT AIR-SEA SURFACE FLUXES CLOUD FORMATION AND ATMOSPHERIC BOUNDARY LAYER EVOLUTION. ANOTHER IS THE COUPLED AIR-SEA INTERACTION THAT TAKES PLACE ACROSS OCEAN CURRENT AND TEMPERATURE FRONTS AT THE OCEAN SUB-MESOSCALE (SPATIAL SCALES OF .05-5 KM) WHERE RECENT FOCUS IS BEING GIVEN TO RESOLVING SPATIAL WIND STRESS VARIATION AND ITS IMPACT ON 3D UPPER OCEAN MIXING. THESE DIFFERING PROCESSES SHARE THE FACT THAT BRIGHTNESS CONTRASTS IN VISIBLE OR RADAR SATELLITE OCEAN IMAGERY HAVE BEEN SHOWN TO DELINEATE FLOW FIELD CHARACTERISTICS VIA VARIABLE SURFACE WAVE ROUGHNESS THAT REFLECTS THE AIR-SEA INTERACTION. WHAT HAS BEEN LACKING FROM A REMOTE SENSING PERSPECTIVE IS A MEANS TO COLLECT HIGH-QUALITY MEASUREMENTS OVER THE GLOBAL OCEAN AND WITH SUFFICIENT COVERAGE IN SPACE AND TIME TO SUPPORT THE NEEDED MODEL ADVANCEMENTS. NASA EARTH SYSTEM MISSIONS GOING FORWARD INCLUDE PROJECTS DESIGNED TO RESOLVE OCEAN CURRENTS WINDS AND SEA SURFACE HEIGHT AT KM SCALE AND THIS PROPOSAL SEEKS TO APPLY A RECENT SATELLITE SYNTHETIC APERTURE RADAR DATASET TO CATALOGUE BACKSCATTER VARIABILITY WITH NEARLY GLOBAL COVERAGE AND WITH SPECIFIC APPLICATION TO DIAGNOSIS OF OCEAN-ATMOSPHERE COUPLING AT SCALES INSIDE OF 5 KM. THE TARGET DATASET IS THE WAVE MODE 20X20 KM SYNTHETIC APERTURE RADAR IMAGETTES OF THE SENTINEL 1A AND 1B SATELLITES. THESE DATA ARE DESIGNED TO SUPPORT OCEAN WAVE FORECASTING APPLICATIONS AS WAS DONE FOR THE ENVISAT MISSION BUT THE LARGER IMAGE SIZE HIGHER RADIOMETRIC QUALITY 5 M RESOLUTION AND VAST QUANTITY AND COVERAGE OF THE RECORDED IMAGES INDICATES CLEAR POTENTIAL FOR EXPANDED OCEANOGRAPHIC APPLICATIONS THAT CAN SUPPORT BOTH SCIENTIFIC AND FUTURE NASA MISSION GOALS. THIS PROJECT WILL DEVELOP NEW REMOTE SENSING ANALYSIS TECHNIQUES THAT WILL PRODUCE MULTI-YEAR DATASETS AND ANALYSES OF SENTINEL-1 SAR IMAGETTES COLLOCATED WITH NASA SATELLITE BUOY AND WEATHER PREDICTION MODEL DATA TO INVESTIGATE ATMOSPHERIC ADJUSTMENT TO VARYING OCEAN AND ATMOSPHERIC CONDITIONS WITH SPECIFIC FOCUS ON AIR-SEA FLUX MODIFICATIONS ASSOCIATED WITH ORGANIZED LARGE EDDIES AND WITH SUB-MESOSCALE SEA SURFACE TEMPERATURE FRONTS., CC* Compute: Koa - A High Performance and Flexible Research Computing Resource
OpenAlex ID: W4414596133