Pipeline version 5 Release (Oct 2026)

API Datasets created in version 5

AIS

• public-global-fishing-effort:v5.0
  • Note on terminology: in this Pipeline version 5 release we are moving away from "fishing effort" to "fishing activity", see more details here. The API dataset name keeps "effort" until our next pipeline version 6.
• public-global-presence:v5.0
• public-global-fishing-events:v5.0
• public-global-loitering-events:v5.0
• public-global-encounters-events:v5.0
• public-global-gaps-events:v5.0
• public-global-port-visits-events:v5.0
• public-global-vessel-insights:v5.0

Identity

• public-global-vessel-identity:v5.0

Imagery

• public-global-sar-presence:v5.0
• public-global-sentinel2-presence:v5.0

See all API datasets here.

Data caveats and full data release notes

Read the full version 5 release notes to learn more about the data changes and limitations here.

Changes in Vessel API

Starting Oct. 21, 2026, the Vessel API returns vessel identity data from version 5 through the public-global-vessel-identity:v5.0 dataset. The latest alias also points to v5.0. Version 5 adds vessel type and characteristics from our new vessel classification model, and changes how the source of each vessel type is reported.

Vessel IDs don't change. A vessel keeps the same vesselId in version 4 and version 5, so stored IDs keep working.

Keep the following in mind regarding vessel identity data:

• The discrepancy vessel type no longer exists. The new model produces a class even where registry and neural network inputs conflict, so there is no longer a category for that disagreement.

• insufficient_data is returned for vessels the model cannot classify. A vessel needs more than 500 positions in a year, after thinning to 5 minutes, for the model to make a prediction. Below that threshold the class is reported as insufficient_data rather than guessed.

• Predictions fall back to a general class when confidence is low. The model assigns a specific class only when its score passes a single threshold that applies uniformly across all gear types and regions. A low-scoring POLE_AND_LINE prediction, for example, is reported as the generic FISHING class instead. This is why the generic FISHING gear type grows substantially while specific gear types shrink.

Updated fields in combinedSourcesInfo

The secondary classifier model updates these fields:

• shiptypes
• geartypes

See the updated diagram on how the ship type and gear type are calculated here. We strongly recommend you check the details and limitations of our new vessel classification model here.

New fields in combinedSourcesInfo

These new fields come from the secondary classifier, the second model in our new vessel classification process. It combines vessel movement patterns with identity information broadcast over the automatic identification system (AIS). See more details of our new vessel classification model here.

Each field reports source: "SECONDARY_CLASSIFIER_MODEL".

Changed: the source value for shiptypes and geartypes

• Carrier, support and bunker vessels now report source: "GFW_VESSEL_LIST".
• All other vessels now report source: "SECONDARY_CLASSIFIER_MODEL".
• Version 4 reported COMBINATION_OF_REGISTRY_AND_AIS_INFERRED_NN_INFO, which is no longer used in version 5.

No longer populated: prodGeartypeSource

The field stays in the response so existing integrations don't break, but its source is now null. Don't rely on it in new code.

What you need to do

• If your code checks for COMBINATION_OF_REGISTRY_AND_AIS_INFERRED_NN_INFO, update it to handle SECONDARY_CLASSIFIER_MODEL and GFW_VESSEL_LIST.
• If you need to reproduce earlier results, pin public-global-vessel-identity:v4.0. Version 4 is available as a static dataset, with no daily updates, until April 2027.
• The gfwr (R) and Python packages return version 5 data starting Oct. 21, 2026.

Changes in 4Wings API - AIS Apparent fishing activity dataset

The new general fishing model uses a transformer architecture instead of the previous convolutional neural network, and was retrained on a substantially expanded label set. It also runs differently: predictions are made at the AIS segment level rather than at the vessel level, positions are thinned to one every 10 minutes instead of one every 5 minutes, and intermediate positions only inherit a prediction when two thinned positions less than 12 hours apart agree.

The result is a more precise and more conservative model. Keep the following in mind when comparing version 5 to earlier versions:

• Lower fishing activity overall. Large volumes of false-positive fishing in ports and during transits are removed.

• Activity shifts between gear types. Most of the change comes from the new vessel classification model rather than from the fishing model itself: many vessels move from a specific gear type to the generic FISHING class. See the vessel classification caveats.

• Known false negatives. The shift toward precision means the model now misses some real fishing, particularly for seines, fixed gear, pots and traps, set gillnets, set longlines and trollers, and for activity close to shore in shallow water. Where fishing hours fall for these gear types, it is difficult to separate genuine improvement from missed activity.

• New activity in poor-reception regions. The previous model required a minimum number of positions per vessel and left many positions permanently unclassified. Version 5 removes that requirement, so positions that were previously null are now classified. This can appear as a large increase in predicted fishing in coastal areas with poor AIS reception.

• Squid jiggers are unaffected. They are classified by night loitering rather than by the fishing model. However, vessels reclassified into or out of the squid jigger class will still show a change.

• New spatial measures. Distance from port, distance from shore and bathymetry are now computed on a single aligned grid, with an updated anchorage list and an updated bathymetry source. This affects distance-based filters such as distance_from_port_km.

Learn more about the new general fishing model here

Changes in 4Wings API - AIS Presence dataset

speed was previously available only as a parameter. It is now also returned in the response schema.

• When temporal-resolution = HOURLY and group by vessel or mssi, we return each speed.
• In other combination: we return avg speed.

Changes in Events API

AIS Encounters events dataset

Encounters are filtered to remove those close to port, using a distance-from-port threshold of 10 km. Version 5 rebuilds that distance against an updated anchorage list, so borderline encounters can move to either side of the threshold. The most visible single change is the addition of a port at 26.5;50 (al-Qatif, Saudi Arabia) that was built in 2023 but missing from the previous anchorage list: it produced more than one million pseudo-encounters, mostly fishing to fishing, which are removed in version 5.

Encounters also depend on which vessels are classified as fishing, carrier, support and bunker, so they are affected by the new vessel classification model.

Check full data changes related to Pipeline version 5 here

AIS Loitering events dataset

Loitering is essentially unchanged in version 5 (around −0.25%). The small difference comes from applying the 20 nautical mile distance-from-shore filter against the rebuilt distance-from-shore raster, which can add or drop events that sit right at the threshold.

Check full data changes related to Pipeline version 5 here

AIS Port visits dataset

Version 5 uses an updated anchorage list, shared for the first time between our AIS and VMS pipelines. In total 5,366 anchorages were added, the largest share of them in Brazil, and a small number of erroneous anchorages that could produce false visits were removed. Named anchorages feed directly into the distance-from-port raster, so the change affects both port visit detection and distance-based filters.

Where a new anchorage sits closer to a vessel than the one previously matched, an existing visit can now register at the new anchorage instead, which also shifts when its entry and exit events fire. Visits at removed anchorages disappear.

Check full data changes related to Pipeline version 5 here

AIS apparent fishing events dataset

Apparent fishing events inherit every change described for apparent fishing activity, since they are built from the same underlying predictions. In addition:

• Fewer events, of shorter duration and distance. Events located in ports are largely removed, consistent with the reduction in port fishing.

Check full data changes related to Pipeline version 5 here