Ocean-Land Mask Update in our Synthetic Aperture Radar Data Pipeline Solved
Update: This post was updated May 29 and Aug. 21 following our announcement about transitioning to an updated ocean-land mask for our synthetic aperture radar (SAR) data processing.
Overview of the change
Beginning March 28, 2026, we transitioned to an updated “ocean-land mask” for processing SAR satellite imagery. This mask acts as a filter, allowing our algorithms to “crop” the ocean and ignore landmasses to ensure we only detect objects in navigable waters.
A technical update in our cloud assets required this transition. To ensure platform continuity, our team implemented a new version of our masking tool.
This issue impacted the following datasets in our map, APIs and data download portal:
- SAR vessel detections
- Sentinel-1 and Sentinel-2 fixed infrastructure
After a thorough quality assurance review, we successfully resolved the issue. However, some minor impacts persist in our fixed infrastructure data. To maintain full transparency, the impact assessment below outlines how these earlier updates affect our global vessel and infrastructure datasets.
Ocean-land mask overall coverage
Assessment period: March 13–26, 2026
Methodology: Detection percentages and coverage changes are calculated against previous dataset totals to identify net gains and losses.
The mask update produces small coverage gains in some areas and small losses in others.
- Coverage gain: The new mask adds approximately 10,000 to 16,000 square kilometers per day to our processing pipeline (roughly 0.15 percent to 0.25 percent of total global coverage).
- Net benefit: The area added in previously “masked-out” zones outweighs the minimal area lost during the transition.
Vessel detections: Net gains and specific losses
The updated mask allows us to detect closer to the shore in some areas while also removing some legacy detections in specific nearshore environments.
- Gains (2.1 percent to 2.9 percent ): New detections have emerged in nearshore areas previously excluded by the old mask. Coverage gains slightly exceed losses across all continents.
- Losses (1.7 percent to 2.1 percent): Some existing detections were dropped, primarily in nearshore areas no longer covered by the new ocean mask.
- Top affected EEZs: Japan, Chile, Turkey, China, Indonesia and Norway saw the highest volume of change. Notably, Australia and South America experienced higher loss rates (2 percent to 5 percent) compared to the global average.
- Regional focus: Europe and Asia account for the majority of changes, while Africa remains the least affected continent (less than 2 percent change across all categories).
Impact on fixed infrastructure
Assessment period: December 1–31, 2025
Switching to the new ocean-land mask affects few real structures. Of the 57,823 oil and wind detections previously identified, 116 were lost (0.20 percent) and 212 were gained (0.37 percent), representing a net gain of 96 infrastructure detections. The overall distribution of detections remains consistent. Including detections labeled as noise, the net change across all detections is an increase of 0.06 percent.
Summary of gains and losses
- Noise in existing areas: Most raw gain and loss differences are noise in already-identified areas, especially along the Arctic coastline around Russia and Svalbard. These differences are likely driven by less accurate shoreline data and Arctic ice rather than real offshore structures appearing or disappearing.
- Valid gains: These are driven mainly by offshore wind detections in Europe, especially Germany and the United Kingdom. Germany showed the largest increase from wind farms 50 to100 kilometers offshore. Africa, Australia and South America are minimally affected.
- Valid losses: These are highly limited. Few non-noise detections were removed, and all were labeled “low confidence.” China lost 17 offshore detections but gained 29, while Venezuela had 15 losses in Lake Maracaibo and no corresponding gains.
- Distance from shore: The differences are driven by mismatches between the shoreline definitions in the distance-from-shore raster and the new ocean-land mask. In the new mask, some features, such as artificial islands, are correctly classified as land, so detections on or around them are no longer covered.
We have detailed the exact locations and IDs of the oil platforms and wind turbines that are no longer detected. Imagery and thumbnails for selected detections are available, up to 48 per continent.
Addressing data losses: What’s next for the fixed infrastructure dataset
Although we have restarted the SAR fixed infrastructure pipeline with the updated mask, our long-term focus is Infrastructure Version 2. This will be a complete rebuild of Global Fishing Watch’s offshore infrastructure dataset, designed to replace Version 1 while preserving existing infrastructure IDs. The update will feature a new detection approach, broader sensor coverage, improved clustering and machine-learning classification. These updates should resolve current boundary discrepancies and improve classification accuracy for fixed infrastructure, such as oil platforms and wind turbines.
Questions?
We are committed to providing the most accurate open-access data available. For technical questions about specific datasets or how this may affect your work, please contact [email protected].
