ocean · agriculture · family: the model was never trained on this
fishing boats that go dark
IUU fishing dark vessel detection and identification
Problem statement
Illegal, unreported, and unregulated (IUU) fishing has been measured worldwide only in part: illegal and unreported catch (leaving out discards and unregulated artisanal catch) is estimated at 11–26 million tonnes annually — $10–23.5 billion in value — an overall loss of 13–31% (mean 18%) in the fisheries studied (Agnew et al. 2009). The main public tracking tool is the Automatic Identification System (AIS), which broadcasts vessel coordinates picked up by satellites and shore receivers. But vessels engaged in illicit activity often turn off their AIS transponders or manipulate the locations they broadcast ("going dark"), and satellite radar mapping shows that 72–76% of the world's industrial fishing vessels are not publicly tracked, with much of that untracked fishing taking place around South Asia, Southeast Asia and Africa (Paolo et al. 2024). Detecting and identifying these "dark vessels" using satellite imagery or other remote sensing requires distinguishing fishing vessels from thousands of non-fishing contacts (cargo ships, pleasure craft, natural features) across vast ocean areas.
Why this matters
IUU fishing threatens food security for the 3.2 billion people who obtain at least 20 percent of their animal protein from aquatic foods (FAO 2024), is associated with depleted stocks (in the Eastern Central Atlantic, where illegal fishing is highest, 60% of assessed demersal stocks were overexploited in the early 2000s), has been linked to organized crime where governance is poor, and captures rents that would otherwise go to legitimate fishers and management authorities (Agnew et al. 2009). Developing countries are the most exposed: Agnew et al. found illegal fishing rates highest in the Eastern Central Atlantic, with total estimated catches off West Africa running 40% higher than reported. Enforcement agencies cannot inspect what they cannot see — and patrol-based monitoring covers only a small fraction of the ocean area where IUU fishing occurs.
What’s been tried and why it hasn’t worked
Global Fishing Watch uses AIS data combined with machine learning to classify vessel behavior (fishing vs. transiting), but this is useless for vessels with AIS disabled. Synthetic aperture radar (SAR) satellites can detect vessels in daylight or darkness and in most weather, but the Sentinel-1 SAR imagery used by Paolo et al. has a resolution of about 20 m (so most vessels under 15 m are missed), and the two Sentinel-1 satellites together mapped the world's coastal waters only about every 6 days — during which a vessel can travel a long way. Optical satellite imagery (Sentinel-2, commercial providers) provides better classification potential but is blocked by cloud cover (which obscures much of the ocean at any given time) and darkness. VIIRS nighttime lights data can detect vessels using fishing lights but misses vessels fishing during the day. Patrol vessel and aircraft surveillance covers tiny fractions of EEZ areas. The fundamental challenge is that no single sensor modality provides the temporal coverage, spatial resolution, and all-weather capability needed for reliable dark vessel detection.
What would unlock progress
Multi-sensor fusion systems that combine SAR detection (all-weather, day/night) with optical classification and AIS correlation to build persistent maritime domain awareness. Machine learning models trained on labeled SAR signatures of known vessel types could enable automated classification from radar data alone. Cubesat constellations providing higher revisit rates (hours rather than days) at lower cost would reduce the gap between detections. On the policy side, mandatory, publicly accessible position tracking for all commercial fishing vessels — AIS requirements currently vary by country, vessel size and activity, and vessel monitoring system (VMS) data are proprietary — would shrink the dark fleet.
Entry points for student teams
A team could build a multi-sensor vessel detection pipeline using freely available Sentinel-1 SAR and Sentinel-2 optical data (via Copernicus Open Access Hub), correlating detected vessels with AIS data (available from MarineTraffic or Global Fishing Watch) to identify "dark" contacts. The classification problem — distinguishing fishing vessels from other contacts in SAR imagery — is a tractable computer vision challenge. A policy-focused team could analyze the coverage gaps in current VMS mandates across jurisdictions and model the impact of extending requirements.
Genes
Structural cousins: stuck the same way, in other fields
Sources
Agnew DJ, Pearce J, Pramod G, Peatman T, Watson R, Beddington JR, Pitcher TJ, "Estimating the Worldwide Extent of Illegal Fishing," PLOS ONE 4(2): e4570, 2009, Paolo FS, Kroodsma D, Raynor J, et al., "Satellite mapping reveals extensive industrial activity at sea," Nature 625: 85–91, 2024, Kroodsma et al., "Tracking the global footprint of fisheries," Science, 2018, via Global Fishing Watch publications, FAO, "The State of World Fisheries and Aquaculture 2024 – Blue Transformation in action," All accessed 2026-08-21. go to source 1 ↗ go to source 2 ↗ go to source 3 ↗ go to source 4 ↗
Working notes
Notes made while writing and checking this brief, shown as written. They use working shorthand.
The `stakeholders:multi-institution` tag passes: flag states (vessel registration), coastal states (EEZ enforcement), regional fisheries management organizations (quota management), and satellite data providers each control non-substitutable pieces — no single institution can solve IUU detection alone. The `temporal:worsening` tag passes: satellite mapping reveals high densities of vessel activity in large areas of ocean that showed little or no activity in public tracking (Paolo et al. 2024; this shows how much goes unseen, not a trend), Agnew et al. estimate that illegal and unreported catch has increased in 5 areas and declined in 11 since the 1990s, with a steady increase in the Eastern Central Atlantic (mixed trajectory evidence), and Agnew et al. note strong economic drivers for illegal fishing, name growing world demand for fish and significant overcapacity of the world's fishing fleet among the drivers of its mid-1990s peak, and say illegal and unreported fishing may be contributing to the overexploitation of stocks (drivers; the paper does not describe a feedback loop). Related to `ocean-fisheries-subsidies-overcapacity-crisis` (which addresses the economic drivers of overfishing) and `ocean-efficient-fishing-gear-bycatch-paradox` (which addresses gear-level issues). This brief focuses on the remote sensing and data fusion challenge.
Reconciliation 2026-08-21: The Source line was largely reconstructed-from-memory and has been replaced with verified citations. Global Fishing Watch publishes no report titled "Illuminating the Global Footprint of Fishing" (checked against https://globalfishingwatch.org/publications/) — the real anchors are Kroodsma et al., "Tracking the global footprint of fisheries" (Science, 2018) and Paolo FS, Kroodsma D, Raynor J, et al., "Satellite mapping reveals extensive industrial activity at sea" (Nature 625: 85–91, 2024), both now cited. The "INTERPOL, International Law Enforcement Cooperation in the Fisheries Sector, 2023" citation could not be verified after multiple attempts (INTERPOL's fisheries-crime page names only Projects ALIOS and GAIA, no such 2023 report) and was dropped; no body claim depended on it. Headline numbers re-anchored to Agnew DJ et al., "Estimating the Worldwide Extent of Illegal Fishing" (PLOS ONE 4(2): e4570, 2009): 11–26 million tonnes and $10–23.5 billion confirmed, but the brief's "up to $23.5 billion" understated the range's floor attribution and "approximately 20% of global marine catch" overstated the paper's 18% mean across studied fisheries — now "on the order of one-fifth" with the full $10–23.5B range. The unsourced "30–50% of the global fishing fleet never carry AIS" was replaced with Paolo et al.'s verified finding that 72–76% of industrial fishing vessels are not publicly tracked. "Coastal developing nations lose an estimated $10 billion annually in their EEZs" appears to be a garble of the global $10–23.5B range and could not be sourced; replaced with Agnew et al.'s verified regional finding (Eastern Central Atlantic highest illegal-fishing rates; West Africa catches 40% higher than reported). "3 billion people who depend on seafood as a primary protein source" corrected to FAO SOFIA 2024's actual statistic: 3.2 billion people obtaining at least 20 percent of per capita animal protein from aquatic foods ("In 2021, they contributed at least 20 percent of the per capita protein supply from all animal sources to 3.2 billion people," FAO newsroom release for SOFIA 2024, https://www.fao.org/newsroom/detail/fao-report-global-fisheries-and-aquaculture-production-reaches-a-new-record-high/en). Two unsourced precise figures softened: "monitoring covers less than 5% of the ocean area" → "only a small fraction," and "clouds cover ~60% of the ocean" → unquantified; "VMS required only for vessels over 15m in some jurisdictions" softened to jurisdiction-varying length thresholds (thresholds differ, e.g., 12m vs 15m, across regimes). FAO SOFIA 2024 full title ("Blue Transformation in action") and Agnew/Paolo citation strings copied from the fetched publisher pages 2026-08-21.
Reconciliation 2026-09-30: All four sources on the Source line and the FAO newsroom release named in the earlier note were read on 2026-09-30, and a second reader went over the repairs against the same sources on the same day. Agnew et al. 2009 was read in full text at the PLOS address; Paolo et al. 2024 in full text at the Nature address; the Global Fishing Watch publications page as the full list; the FAO SOFIA landing page as the full page (it is a short catalogue page, not the report); the FAO newsroom release in full. The Kroodsma et al. 2018 Science paper itself is closed, so only its citation line on the Global Fishing Watch page and its abstract (Europe PMC record for DOI 10.1126/science.aao5646) were read. (1) The text said IUU fishing "accounts for" 11–26 million tonnes and "on the order of one-fifth of the global marine catch." Agnew et al. confine their estimate to illegal and unreported catch, excluding discards and unregulated artisanal catch, and give "the overall loss from our studied fisheries" as "13–31% (lower and upper estimates) with a mean of 18%", not as a share of the global catch. The sentence now says that. (2) "IUU vessels routinely disable AIS transponders" is given in Paolo et al. as "Vessels engaged in illicit activities often turn off their AIS transponders or manipulate the locations they broadcast"; the text now says "often" and includes the manipulated positions. "Not publicly tracked by AIS at all" lost "by AIS at all": the paper's term is "not publicly tracked", meaning not visible on publicly accessible AIS data, and it notes blind spots where satellite reception is poor. "Concentrated in South and Southeast Asia and Africa" is the paper's "much of that fishing taking place around South Asia, Southeast Asia and Africa", now worded that way. "Broadcasts vessel position via satellite" now says coordinates picked up by satellites and shore receivers, which matches the paper's mention of satellite reception and terrestrial receivers. (3) "Drives species toward collapse (particularly in West African and Southeast Asian waters), funds organized crime networks, and undermines the economic viability of legal fishing operations" is not what Agnew et al. say. They report a correspondence between illegal and unreported fishing and the number of depleted stocks, with 60% of assessed demersal stocks in the Eastern Central Atlantic overexploited in the early 2000s; that illegal fishing in poorly governed regions "has often been linked to organised crime"; and that rents are "captured by illegal fishermen but lost to legitimate fishermen and management authorities." Southeast Asian waters are not singled out. The sentence now reports those three findings. (4) SAR "in all weather and lighting conditions" with "revisit times of 1–6 days" and vessels travelling "thousands of kilometers": Paolo et al. say SAR is "unaffected by light levels and most weather conditions", that the Sentinel-1 imagery they used has a resolution of about 20 m so most vessels under 15 m are missed, and that the two Sentinel-1 satellites together map coastal waters "approximately every 6 days". No source gives a 1-day revisit or a travel distance; the sentence now carries the paper's figures, tied to Sentinel-1 as that paper used it, and an unquantified travel distance. (5) VMS "currently required only above vessel-length thresholds that vary by jurisdiction" is not in any source. Paolo et al. say AIS is not required of all vessels "as regulations vary by country, vessel size and activity" and that VMS is proprietary; the sentence now says that. (6) In the tag rationale, "industrial fleets move into previously unfished waters" and "FAO estimates of IUU catch have not decreased over 15 years despite enforcement efforts" are not in any source, and the second is contradicted by Agnew et al., who estimate that the level of illegal and unreported catch "has declined in 11 areas and increased in 5" since the 1990s, with "a steady increase" in the Eastern Central Atlantic. "Overcapacity in legal fishing drives more operators to IUU as stocks decline (feedback loop)" is supported only in part: Agnew et al. name "significant overcapacity of the world's fishing fleet" among the drivers of the mid-1990s peak in illegal fishing and say illegal and unreported fishing may be contributing to overexploitation, but they do not describe a loop. The rationale now cites Paolo et al.'s mapping of vessel activity in areas with little or no publicly tracked activity, Agnew et al.'s mixed regional trends, and the drivers Agnew et al. name. The `temporal:worsening` tag itself is untouched, but the sources do not show a worsening trend (Agnew et al. place the worst period in the mid-1990s) and the tag deserves review. Confirmed in the sources and left as written: 11–26 million tonnes and $10–23.5 billion annually; illegal fishing highest in the Eastern Central Atlantic and West African catches 40% higher than reported; "Developing countries are most at risk"; 72–76% of industrial fishing vessels not publicly tracked; the 3.2 billion people getting at least 20 percent of their animal protein from aquatic foods in 2021 (FAO newsroom release); all four citation strings on the Source line, including the Nature volume and pages and the Science volume, issue and pages listed on the Global Fishing Watch page. Not in any source and left as written, as general background: that Global Fishing Watch classifies fishing versus transiting with machine learning (the Kroodsma abstract says only that 22 billion AIS messages were processed to track more than 70,000 vessels); the cloud-cover and VIIRS day/night statements; patrol and aircraft coverage as "a small fraction" or "tiny fractions"; cubesat revisit rates; and the entry-point data services (Copernicus Open Access Hub, MarineTraffic). Titles unchanged. Genes untouched.