Global Fishing Watch AIS Datasets and APIs

Most AIS products tell you where a ship says it is. This one asks what it was doing there - and what it was hiding. Global Fishing Watch publishes research-grade maritime activity data derived from broadcast AIS, government VMS partnerships and deep-learning classification of satellite imagery, organized into four product groups: gridded apparent fishing effort and vessel presence, vessel identity resolution, a multi-class event stream, and per-vessel insights.

The scale is the story. Apparent fishing effort runs as daily global grids at 0.01-degree resolution back to 2012 - twelve-plus years of where fishing pressure concentrated, split by flag state and gear type - alongside monthly 0.1-degree grids and per-MMSI daily hours. The identity layer resolves MMSI, IMO number or name against GFW's own AIS identity records plus more than 40 public regional and national registries. The events layer classifies behaviour: apparent fishing events, encounters between vessel types (fishing-carrier, tanker-fishing, carrier-bunker and more), loitering, port visits, and AIS-off gaps - the dark-vessel signature. A separate layer carries Sentinel-1/2 detections of fixed offshore infrastructure from 2017 onward.

Inside Datadory's marine transportation shelf this record holds a unique seat: it is the only one that reads behaviour rather than movement or cargo. MarineCadastre counts positions; Eurostat counts tonnes; this counts what ships did when nobody was meant to be watching. Get a sample of this dataset cut to your region, gear types and years.

One observation per grid cell per day in the effort layers; one observation per detected event in the events layers. The shape below follows the verified field dictionary:

# Effort layer: one row per grid cell, per day
dataset          : public-global-fishing-effort:latest   # layer selector
mmsi             : <vessel MMSI>                         # identity joins on mmsi / imo / shipname
geartype         : trawlers                              # trawlers, longliners, purse seines, ...
flag             : <flag state>                          # effort splits by flag state and gear
hours            : <apparent fishing hours>              # aggregated per cell or region report
# Events layer adds:
event_type       : encounter                             # fishing | encounter | loitering | port_visit | gap
start_time       : <event start>                         # with end_time bounding the window

An honesty note first: the record ships no sample file, so the block above shows structure rather than pasted extracts - concrete values arrive with your sample request instead of being invented here. Three things worth noticing anyway. hours is an apparent measure - model-inferred fishing behaviour from AIS patterns, not observed catches - so it reads activity pressure, not landings. The identity fields are the join key that lets you attach behaviour to a specific hull across years even when its MMSI changes. And the event taxonomy is typed by vessel-pair class, which means "carrier meeting bunker near the EEZ edge" is a query, not a manual review project.

Eleven documented attributes carry the model, split between an identity spine (mmsi, imo, shipname, flag, geartype) and an activity payload (hours, event_type, temporal bounds). Definitions were inferred during research from the publisher's documentation rather than shape-exact against live responses - the exact response naming for your pipeline locks when your sample is prepared. The dictionary below is the full set.

• Geography: global ocean, built from AIS receivers, satellite AIS providers and partner VMS programs - no regional gating, the same cells cover the North Atlantic and the Pacific.
• Temporal: AIS-derived effort, presence and identity run from 2012 to within about 96 hours of the present; SAR vessel detections start 2017 and run to roughly five days ago; fixed-infrastructure detections run from 2017 onward; the static fishing-effort release maps 2012 through 2024.
• Granularity: three resolutions stacked - 0.01-degree daily grids, 0.1-degree monthly grids by flag state and gear, and 0.1-degree daily hours per MMSI, plus per-event and per-vessel identity records.

On Datadory's 0-10 quality rubric the record scores 9, against a 7.81 average across the 1,744 cataloged datasets - held up by twelve-plus years of daily global grids, a 40-plus-registry identity backbone and a typed event taxonomy, and held back mainly by inferred (rather than shape-verified) field documentation.

API, files, or your warehouse. Daily, weekly, or hourly.

• IUU-fishing investigation and enforcement support - pull apparent effort by gear type and flag state, then overlay AIS-off gaps to flag vessels that went quiet exactly where the pressure was.
• Transshipment detection - query encounters typed by vessel pair (fishing-carrier, tanker-fishing, carrier-bunker) to screen for at-sea transfer corridors instead of reading track plots by hand.
• Supply-chain and ESG due diligence - check whether carriers and reefers on your route actually called at ports with loitering-heavy approaches before signing the charter.
• Port-call and market context - read port visits and loitering around a hub as an activity index no customs table publishes.
• Insurance and risk screening - use gaps, encounters and loitering histories as behavioural risk features alongside flag and age.
• Academic and cited journalism - anchor papers and investigations to a named, versioned product whose provenance runs to satellite-AIS providers and VMS partnerships.

Data scientists and ML engineers rank first: typed event tables with a stable identity spine train anomaly models without a labeling sprint (their marine-transportation workflows). Journalists, academics and students get citable, versioned layers behind every dark-fleet headline (their use cases). Competitive-intel teams watch rival fleets' port-call and loitering patterns quarter over quarter (their use cases). Market researchers and consultants read effort and port visits as regional activity indices (their use cases). Developers building data products get map-ready grids and typed events that render as monitoring dashboards (builders' use cases), and e-commerce operators benchmark ocean-leg exposure for seafood and perishable supply chains (their workflows).

• MarineCadastre AIS Vessel Traffic Data (USCG Nationwide AIS) - the raw-position counterpart for US waters: billions of timestamped broadcasts where GFW contributes the classified interpretation (MarineCadastre).
• EMODnet Vessel Density Maps (EU Waters) - the intensity-grid counterpart for EU basins, counting presence per square kilometre where GFW labels what the presence was doing (EMODnet).
• vs UNCTADstat Maritime Transport Data Centre - behavioural event records against official fleet-capacity and trade statistics; field-by-field breakdown in the dedicated comparison.
• Glossary notes - why a dark vessel is defined by silence, how a transshipment event shows up as an encounter record, and what an AIS position report contributes before any classification runs.

Two honesty notes before you request a sample. "Apparent" is doing real work throughout: effort and events are model-inferred from broadcast behaviour, so a vessel that never carried AIS stays invisible to every layer here regardless of what it did. And the field dictionary was inferred from documentation rather than shape-verified against live responses, so confirm exact response naming against your prepared sample before hard-coding parsers. Full slice ranking sits on the marine transportation data hub, with the ranked view at best marine-ports-services datasets.