2025 March – Global static dataset of AIS-based apparent fishing effort v3: difference with the map

Users of the Global Fishing Watch (GFW) Map and APIs can access near real-time vessel activity data, subject to an approximate 3-day processing delay via an automated pipeline, supporting in-depth analysis. This static data release differs in that it provides a fixed, quality-assured dataset spanning to December 2024, achieved through more extensive cleaning and manual review.

• As in the previous release of this dataset, we remove apparent fishing hours on transits. Our algorithm can make false positive fishing detections when a vessel is transiting (in other words, incorrectly identifying the vessel as engaged in fishing behavior when it is underway from point A to point B). The GFW Map and APIs do not attempt to remove these false positives. However, in the static fishing effort dataset, we do attempt to remove these false positives by removing any fishing detections when the vessel is moving fast (that is, above fishing speeds) in a straight line.
• In the static dataset, we more rigorously filter vessel lists to include only vessels that we have high certainty are fishing vessels. For example, for vessels that do not self-report as fishing vessels, to assign the vessel a fishing class in this dataset we require a high level of confidence from our classification neural net and/or registry information listing the vessel as a fishing vessel. However, the level of confidence required is lower for the vessel to be shown on the GFW Map and APIs as a likely fishing vessel.
• For vessels with invalid MIDs and no registry information, the GFW Map and APIs give no flag information, while as previously mentioned this dataset assigns a flag of UNKNOWN-[ISO] for vessels that spend a majority of their fishing hours in a particular state’s EEZ.
• To handle and display a high volume of data interactively, the GFW Map and APIs thins vessel tracks to one position per hour when producing rasters. Since in this static dataset we are not constrained by the same engineering concerns, we do not thin vessel positions. Therefore, the fishing effort rasters will differ slightly between the two, especially at the 1/100th degree level, since fishing effort data for a given vessel is more granular in the static raster dataset.
• The Map and APIs rasters have cells centered around every 1/10th or 1/100th degree interval, whereas the static rasters in this dataset have cells with every 1/10th or 1/100th degree interval in the corner of cells. This difference occurs because of differences in the code frameworks underlying the two different data products; in the future, we hope to eliminate this discrepancy so that data from the all products can be used in tandem.