# MarineAware — full content > MarineAware is an AI-native maritime analytics firm. We build agentic intelligence on AIS, satellite (SAR/EO), and metocean data for carriers, insurers, commodity traders, ports, and governments. MarineAware is an AI-native maritime analytics firm (founded 2024, remote-first · London · Singapore). We build agentic intelligence on AIS, satellite radar/optical (SAR/EO) and metocean data: dark-vessel and sanctions-evasion detection, fleet emissions (CII/EEXI/EU ETS/FuelEU Maritime), voyage and port optimisation, and market/flow intelligence — for carriers, insurers, commodity traders, ports and governments. A live demonstrator runs at https://live.marineaware.com. Site: https://www.marineaware.com/ Short index: https://www.marineaware.com/llms.txt Contact: contact@marineaware.com · https://www.marineaware.com/contact Updated: Sun, 26 Jul 2026 01:04:37 GMT --- ## Solution lines ### Maritime Domain Awareness See the vessels that do not want to be seen. For: Navies, coast guards, fisheries & sanctions enforcement Problem: Up to ~75% of industrial vessel activity is absent from public AIS. Dark ships disable transponders, spoof positions, and hide behind shell ownership. Legacy tooling and analyst shortages leave a blind ocean. Approach: Fuse cooperative AIS with non-cooperative SAR (Sentinel-1) and optical (Sentinel-2, Planet, ICEYE) detections; CFAR / CNN ship detectors with detection-to-AIS association and gap reasoning; Pattern-of-life modelling, rendezvous & loitering detection, RF cross-cueing; Analyst-in-the-loop copilots that explain every alert with cited evidence. Outcomes: A common operating picture that includes the dark fleet; Prioritised, explainable leads instead of raw detections; Faster tip-and-cue from wide-area SAR to high-res tasking. Data used: Satellite & terrestrial AIS, Sentinel-1 SAR, Sentinel-2 / Planet / ICEYE, Global Fishing Watch. ### Dark-Fleet & Sanctions Screening Predictive risk scoring for the 600–800-ship shadow fleet. For: Marine insurers, P&I clubs, reinsurers, compliance teams Problem: The shadow fleet is now ~10–15% of the tanker fleet. Accidental exposure to a sanctioned vessel is an existential compliance and reputational risk, and P&I withdrawal has become the sharpest enforcement lever. Approach: Behavioural risk scoring from AIS gaps, spoofing signatures, STS transfers and flag-hopping; Identity resolution across MMSI ↔ IMO ↔ beneficial owner using Equasis, S&P and Clarksons; Ownership-network graph analysis to expose shell structures; Explainable, auditable models with full data lineage for regulators. Outcomes: Know-your-vessel screening before a policy is bound; Portfolio-level exposure aggregation to designated & high-risk tonnage; Defensible methodology that underwriters and regulators accept. Data used: AIS gap & spoofing analysis, OFAC / OFSI lists, Equasis & registries, SAR corroboration. ### Fleet Decarbonization & Compliance One coherent strategy across CII, EU ETS and FuelEU Maritime. For: Shipowners, operators, charterers & sustainability teams Problem: Overlapping regimes — CII ratings, 100% EU ETS from 2026, FuelEU Maritime penalties, the IMO Net-Zero Framework — each carry different metrics and penalties. Owners need to cut cost and carbon without a rip-and-replace. Approach: Bottom-up emissions modelling from AIS activity × vessel-specific fuel curves, calibrated against verified EU MRV data; Cost modelling across CII, EU ETS, FuelEU and UK ETS with pooling & banking optimisation; Voyage and weather routing that trades time vs fuel vs CII penalty; Compliance strategy financiers and charterers will accept. Outcomes: Lowest-cost compliance pathway quantified per vessel and per voyage; CII rating improvement without blanket slow-steaming; MRV-grade emissions numbers for lenders and Poseidon Principles. Data used: EU MRV / THETIS-MRV, AIS activity model, CMEMS / ECMWF metocean, Registry fuel curves. ### Market & Flow Intelligence Turn raw AIS and cargo data into a proprietary trading signal. For: Commodity traders, energy majors & research desks Problem: Desks buy Kpler/Vortexa-grade feeds but need to turn them into defensible alpha — floating storage, refinery runs, port queues — before the market prices it in, without conflicts across competing desks. Approach: Feature engineering from AIS: floating storage, STS transfers, port congestion, draft-derived utilisation; Trade-flow context from UN Comtrade fused with port-call time series; Backtested signals with explicit handling of alpha decay and data latency; Integration into existing trading and research models — we sit between vendors and your book. Outcomes: Earlier read on supply shifts than the printed market; A proprietary layer on top of purchased data; Quant-rigorous, backtested, conflict-free. Data used: Satellite AIS, UN Comtrade, Port-call analytics, Floating-storage models. ### Voyage & Port Optimization Arrival predictions more accurate than carrier ETAs. For: Ports, terminals, carriers & cargo owners Problem: Unreliable ETAs and berth under-utilisation ripple through the supply chain as demurrage, congestion and buffer inventory. Carrier-reported ETAs are not good enough to plan against. Approach: ML ETA regressors on historical voyages plus physics-aware weather routing; Metocean forcing from CMEMS currents, WaveWatch III and ECMWF, constrained by GEBCO/EMODnet bathymetry; 72-hour congestion forecasting from port-call time series; Just-in-time arrival to cut at-anchor waiting and emissions. Outcomes: Predictive ETA that beats carrier estimates; Berth and yard planning against a reliable horizon; Lower demurrage, congestion and at-berth emissions. Data used: AIS voyage histories, CMEMS / WW3 / GFS, GEBCO bathymetry, Port-call data. ### Diligence & Advisory Domain-literate diligence on a category growing ~192% YoY. For: VCs, PE, corporate development & strategics Problem: Maritime and dual-use AI is heating up fast. Investors need to know whether the technology is real, whether the data moat is defensible, and whether the TAM rests on durable regulatory tailwinds. Approach: Technical due diligence: is the model real, is the data moat defensible; Market sizing anchored to dated regulatory drivers (EU ETS, FuelEU, IMO Net-Zero, sanctions); Competitive teardowns of Kpler, Windward, Spire, Veson and the challenger set; A point of view, an operator network for reference calls, and investor-grade rigor. Outcomes: Fast, expert diligence in a competitive process; A defensible investment thesis with named sources; Confidence in — or red flags on — the technology and moat. Data used: Market & funding data, Competitive intelligence, Technical audits, Operator network. --- ## Data sources we work with - Satellite & terrestrial AIS (Vessel tracking, Commercial): Kpler AIS, Spire, S&P/ORBCOMM; T-AIS near-instant, enhanced S-AIS ~6 min — used for Live positions, port calls, dark-activity. - NOAA MarineCadastre AIS (Terrestrial AIS, Open): US coastal historical archives — used for Free historical traffic analysis. - Copernicus Sentinel-1 (SAR) (C-band radar, Open): ~5–20 m, all-weather day/night; 1C/1D carry onboard AIS — used for Non-cooperative / dark-ship detection. - Copernicus Sentinel-2 (optical) (Multispectral, Open): 10 m, 5-day revisit, cloud-limited — used for Optical vessel & coastal detection. - ICEYE / Umbra / Planet (Commercial imagery, Commercial): SAR to ~25 cm; PlanetScope ~3 m daily — used for Confirm & classify detections. - Copernicus Marine (CMEMS) (Ocean physics, Open): Currents, SST, SSH, waves; ~8 km global to 500 m regional — used for Routing & fuel optimisation. - NOAA GFS / WaveWatch III (Metocean forecast, Open): Wind ~13 km; waves ~50 km, to 5–16 days — used for Weather routing. - ECMWF ERA5 / Open Data (Reanalysis / NWP, Freemium): ERA5 ~31 km hourly 1940→; open subset 0.25° — used for ML training & backtesting. - GEBCO / EMODnet bathymetry (Bathymetry, Open): GEBCO 15 arc-sec; EMODnet ~115 m (EU) — used for Under-keel clearance, depth routing. - Global Fishing Watch (Fishing / events, Open): AIS effort 2012–2024, events & vessels APIs, SAR layers — used for IUU, encounters, transshipment. - UN Comtrade (Trade flows, Freemium): Bilateral HS trade, 1962→ — used for Demand-side flow context. - EU MRV (THETIS-MRV) (Emissions, Open): Per-ship verified annual CO₂ & fuel, >5000 GT, 2018→ — used for Emissions ground truth & CII context. - Equasis / Clarksons / S&P Sea-web (Registries, Freemium): Identity, ownership, PSC, ship particulars — used for Identity resolution & due diligence. --- ## Methods & modelling approaches ### Trajectory modelling & anomaly detection Learn normal behaviour per ship-type and route from AIS kinematics, then flag deviations — course/speed anomalies, gaps, loitering, impossible kinematics. Sequence models (LSTM/GRU, Transformers), reconstruction and WGAN-GP scoring, over unsupervised route baselines. ### Dark-vessel & sanctions-evasion detection The flagship capability: correlate AIS with SAR/optical detections — a radar hit with no matching AIS is a candidate dark ship — fused with gap, loitering and rendezvous reasoning. Anchored by the 2024 Nature result that ~75% of fishing vessels are untracked. ### ETA & voyage / weather routing ML ETA regressors on historical voyages plus physics-aware routing over gridded metocean (CMEMS, WW3, ECMWF) with bathymetry constraints and vessel fuel curves. Objective trades time vs fuel vs CII penalty. ### Emissions estimation (CII / EEXI, MRV-grade) Bottom-up activity model — AIS speed/time × vessel-specific power/fuel curves → fuel → CO₂ — calibrated against verified EU MRV totals so AIS-based estimates are credible to lenders and verifiers. ### Demand & flow forecasting Fuse port-call time series (AIS geofencing) with trade signals (UN Comtrade) and seasonality; classical and ML time-series, increasingly time-series foundation models for zero/few-shot forecasting across many ports. ### Foundation models for geospatial & time-series Fine-tune geospatial FMs (Prithvi-EO, TerraMind) for SAR/optical vessel detection, and apply time-series FMs (TimesFM, Chronos) to ETA and demand. Maritime-specific fine-tunes are R&D, not turnkey — we treat them accordingly. ### LLM / agentic maritime-intelligence pipelines Retrieval over vessel, registry and regulatory corpora; agents orchestrating AIS, SAR and weather APIs; natural-language alerting with cited evidence — explainable copilots in the spirit of research systems like AIS-LLM. ### Identity resolution & network analysis Resolve MMSI ↔ IMO ↔ beneficial owner across Equasis, S&P and Clarksons; graph analytics over ownership, management and co-occurrence to expose the shell structures behind sanctioned or IUU fleets. --- ## Representative case studies Illustrative of how we work; not claims of named client results. ### Predictive dark-fleet screening for a marine mutual Sector: Marine insurance · Client type: A top-10 P&I club (anonymized) Summary: We rebuilt know-your-vessel screening around behavioural risk scoring, catching shadow-fleet exposure before cover was bound rather than after a sanctions listing. Challenge: The club faced growing exposure to the sanctioned shadow fleet — vessels that disable AIS, spoof positions and hide behind shell ownership. Screening was reactive: a vessel was flagged only after it appeared on a designation list, by which point cover was already in force. Approach: We built a behavioural risk score from AIS gap analysis, spoofing signatures, ship-to-ship transfer detection and flag-hopping, then resolved identity across MMSI, IMO number and beneficial owner using Equasis, S&P and Clarksons registries. Ownership-network graph analysis surfaced shell structures, and every score shipped with its data lineage so underwriters and regulators could defend it. Outcome: Underwriters now receive an explainable risk score at the point of quote, with portfolio-level aggregation of exposure to designated and high-risk tonnage. Screening shifted from reactive to predictive without adding headcount. Metrics: 3–6 wks earlier risk detection vs list-based screening; ~80% of flagged vessels had prior AIS-gap signatures; 100% audit-ready lineage on every score. ### One compliance strategy across CII, EU ETS and FuelEU Sector: Shipping & decarbonization · Client type: A mid-size dry-bulk operator (anonymized) Summary: We modelled the true cost of overlapping 2026 emissions regimes per vessel and per voyage, then found the lowest-cost compliance pathway without blanket slow-steaming. Challenge: With 100% EU ETS coverage from January 2026, the first FuelEU Maritime penalties landing by mid-2026, and several vessels drifting toward a CII "D" rating, the operator faced three overlapping regimes with different metrics and no coherent view of total exposure. Approach: We built a bottom-up emissions model from AIS activity multiplied by vessel-specific fuel curves, calibrated against the fleet's verified EU MRV totals so the numbers were MRV-grade. On top we modelled cost across CII, EU ETS, FuelEU and UK ETS, optimised FuelEU pooling and banking, and ran weather routing that trades voyage time against fuel and CII penalty. Outcome: The operator now has a per-vessel, per-voyage compliance cost model, a defensible plan to lift at-risk vessels off a "D" rating, and emissions figures its lenders and charterers accept under the Poseidon Principles. Metrics: 12–18% modelled reduction in 2026 compliance cost; 4 of 6 vessels moved off a projected CII "D"; MRV-grade emissions numbers reconciled to EU MRV. ### A floating-storage signal for a crude trading desk Sector: Commodity trading · Client type: A commodity trading house (anonymized) Summary: We turned purchased AIS and cargo feeds into a backtested floating-storage indicator the desk could trade against, sitting between the data vendors and the trading model. Challenge: The desk already bought high-grade AIS and cargo data but could not turn it into a proprietary edge. Floating storage, ship-to-ship transfers and port queues were visible in the raw feed but arrived as noise, not signal — and always later than the desk wanted. Approach: We engineered features directly from AIS: floating-storage tonnage from laden tankers idling beyond normal dwell, STS transfer detection, port-congestion queues and draft-derived utilisation. We fused these with UN Comtrade trade-flow context, then backtested the resulting indicator with explicit handling of alpha decay and data latency before wiring it into the desk's existing research model. Outcome: The desk gained a proprietary floating-storage indicator with a documented lead time over the printed market, delivered as a clean signal into models it already trusted — with no conflict of interest across competing desks. Metrics: 5–9 days median lead over published inventory prints; 6 yrs backtested signal history; 0 conflicts with competing desks. ### SAR-and-AIS fusion for a fisheries enforcement agency Sector: Government & defense · Client type: A national maritime agency (anonymized) Summary: We stood up a dark-vessel detection pipeline that fuses Sentinel-1 radar with AIS, turning wide-area imagery into a short, ranked list of leads for a stretched analyst team. Challenge: The agency had a vast EEZ, a handful of analysts, and a mandate to counter illegal, unreported and unregulated fishing — where up to three-quarters of vessels never appear in public AIS. Raw satellite tasking produced more detections than the team could triage. Approach: We fused cooperative AIS with non-cooperative Sentinel-1 SAR: CFAR and CNN ship detectors on wide-area radar, detection-to-AIS association, and gap reasoning so any radar hit without a matching transponder became a candidate dark vessel. Loitering, rendezvous and pattern-of-life scoring ranked leads, and an analyst-in-the-loop copilot explained each one with the evidence behind it. Outcome: The team moved from drowning in detections to working a short, ranked, explainable lead list — with a clear tip-and-cue path from free wide-area SAR to high-resolution commercial tasking only where it mattered. Metrics: ~70% reduction in detections needing manual triage; every alert dark-vessel leads ranked & explained; ↓ sharply cost per confirmed lead. ### Predictive ETA that beats carrier estimates for a terminal Sector: Ports & terminals · Client type: A container terminal operator (anonymized) Summary: We replaced unreliable carrier-reported ETAs with weather-aware arrival predictions and a 72-hour congestion forecast the berth-planning team could actually plan against. Challenge: Carrier-reported ETAs were too noisy to plan berths and yard against, driving demurrage, anchorage queues and wasted at-berth emissions. The terminal needed arrival predictions it could trust across a 72-hour planning horizon. Approach: We trained ML ETA regressors on historical AIS voyages and layered physics-aware weather routing on top — metocean forcing from CMEMS currents, WaveWatch III waves and ECMWF wind, constrained by GEBCO and EMODnet bathymetry. A congestion model on port-call time series projected queue build-up 72 hours out, enabling just-in-time arrival coordination with inbound carriers. Outcome: Berth planners now work from a reliable arrival horizon rather than carrier guesswork, cutting anchorage waiting and at-berth emissions and smoothing yard utilisation. Metrics: ~35% improvement in 72h ETA accuracy vs carrier ETA; double-digit % reduction in avg anchorage wait; 72 hrs planning horizon. --- ## Guides & insights ### Open datasets for maritime domain awareness: a practical catalog (Guide) URL: https://www.marineaware.com/guides/open-datasets-for-maritime-domain-awareness Date: 2026-07-11 A practitioner's catalog of the free, open datasets you can use to build a real maritime domain awareness picture — open AIS, Copernicus Sentinel-1 SAR and Sentinel-2 optical, metocean, thermal, fishing, trade and registry data — with what each is good for, how to access it, and where it runs out. You can build a real maritime domain awareness (MDA) picture without buying a single feed. We did — [live.marineaware.com](https://live.marineaware.com) runs on nothing but the datasets below. This guide is the practical catalog: what each open dataset is, how to reach it, what it is genuinely good for, and where it runs out. For the terms used here, see the [MDA glossary](/glossary); for how to turn these feeds into findings, see the [maritime OSINT guide](/guides/open-source-intelligence-for-maritime-domain-awareness). ## How to read this catalog MDA is a fusion problem: no single feed answers a real question. Group the open datasets by the role they play, and the picture assembles itself: - **Cooperative tracking** — where vessels say they are (AIS). - **Non-cooperative imagery** — where they actually are (SAR, optical). - **Environment** — why they behave as they do (metocean, bathymetry). - **Activity & context** — fishing, thermal, trade, identity, emissions. ## Cooperative tracking: open AIS AIS is the backbone — VHF broadcasts of a vessel's identity and position — but the open versions are sampled, not exhaustive. - **AISStream.io** — a free, live WebSocket of AIS position reports, filterable by bounding box. Best for: a live snapshot of who is in a zone right now. Limits: the free tier throttles hard across many areas at once. - **Global Fishing Watch (GFW)** — open AIS-*derived* events (gaps, loitering, encounters) and satellite vessel detections, with APIs and bulk download under a CC BY-SA licence. Best for: pre-computed behavioural signals without building your own pipeline. - **NOAA MarineCadastre.gov** — free bulk historical US coastal AIS. Best for: training and backtesting on clean archives. **The catch, quantified:** a 2024 *Nature* study found roughly **75% of industrial fishing vessel activity** absent from public AIS. Open AIS tells you about cooperative vessels; the interesting ones are often silent — which is why you need imagery. ## Non-cooperative imagery: Copernicus Sentinel The European Copernicus programme publishes two open datasets that are the heart of open dark-vessel work, both free via the **Copernicus Data Space Ecosystem** (STAC catalog, OData, S3). - **Sentinel-1 (C-band SAR)** — radar that sees hulls in any weather, day or night, at ~5–20 m. This is the open dataset for **dark-ship detection**: cross-reference radar detections against AIS, and a hull with no transponder is a candidate. Limits: 6-day-class revisit and metre-scale-to-tens resolution — it samples a chokepoint, it does not stare. - **Sentinel-2 (optical, multispectral)** — 10 m visible/near-infrared, 5-day revisit. Best for: daylight confirmation and coastal/port context. Limits: cloud cover, and no imaging at night. For a zero-setup daily image, **NASA GIBS / Worldview** serves pre-rendered global true-color snapshots (MODIS/VIIRS) with no key — coarse (250 m class) but instantly usable as context. ## Environment: metocean and bathymetry Behaviour without weather is ambiguous — a loitering vessel may simply be holding station in heavy seas. Open metocean resolves it. - **Copernicus Marine (CMEMS)** — currents, sea-surface temperature, salinity, sea level, waves; global to regional, hindcast to forecast. Open via Toolbox API and NetCDF. - **NOAA GFS / WaveWatch III / RTOFS** — global wind, wave and current forecasts, open via NOMADS/GRIB. - **Open-Meteo** — a free, no-key REST API for wind, gusts, wave and swell height at a point; the fastest way to add a transit-condition read (this is what the live dashboard uses). - **GEBCO_2024 / EMODnet Bathymetry** — open depth grids for under-keel clearance and routing bounds. ## Activity & context - **NASA FIRMS** — near-real-time thermal anomalies (fires, gas flares, hot spots); a useful activity proxy near ports and industrial coasts. Free key. - **Global Fishing Watch fishing effort** — open, gridded AIS-derived fishing effort, 2012 onward; the backbone of open IUU-fishing analysis. - **UN Comtrade** — open bilateral commodity trade flows for demand and floating-storage context (free key, generous limits). - **Equasis** — a free-registration vessel registry: identity, ownership, Port State Control and ISM records for due diligence and identity resolution. - **EU MRV / THETIS-MRV** — open, ship-level verified CO₂ and fuel data for ships calling the EU/EEA; the calibration anchor for open emissions estimation and the context behind [CII, EU ETS and FuelEU](/insights/cii-eu-ets-fueleu-2026). ## A minimal open MDA stack If you want the shortest path to a working picture, this is the stack behind our demonstrator: 1. **AISStream.io** for live positions in your areas of interest. 2. **Global Fishing Watch** for derived gap/loitering/encounter events and SAR detections. 3. **Copernicus Sentinel-1** for dark-ship cross-checking; **Sentinel-2** / **GIBS** for optical context. 4. **Open-Meteo** for metocean. 5. **NASA FIRMS** for thermal activity. 6. **Equasis** / **UN Comtrade** for identity and trade context. Fuse them, and you have MDA. The full source catalog — including the licensed options — is on our [data & methods](/data-and-methods) page. ## Where the open stack ends Open datasets are a genuine baseline, not a ceiling. Free AIS is rate-limited and cooperative; open SAR revisit is multi-day and resolution coarse; licences vary (Sentinel and NASA are permissive; GFW is CC BY-SA; always check before redistributing). The moment you need **persistent** monitoring — a vessel count that is a count, an AIS gap that is a real gap, a dark ship you can re-observe within hours — you are into licensed satellite AIS (Kpler, Spire) and high-revisit commercial SAR (ICEYE, Umbra, Capella). How that transition works, and what fine-tuned models add on top, is covered in [from open source to operational](/insights/osint-to-operational-proprietary-data-finetuned-models). --- **See these datasets fused and scored, live, at [live.marineaware.com](https://live.marineaware.com).** To build the operational version on your own areas of interest, [talk to us](/contact). --- ### Open-source intelligence (OSINT) for maritime domain awareness: a practical guide (Guide) URL: https://www.marineaware.com/guides/open-source-intelligence-for-maritime-domain-awareness Date: 2026-07-11 How maritime open-source intelligence actually works — the collection-to-assessment workflow, the signals that matter (AIS gaps, loitering, ship-to-ship transfers, dark ships via SAR-AIS cross-matching, thermal, news corroboration), how to avoid false positives, and where open methods scale into licensed data and fine-tuned models. Open-source intelligence at sea is not a data-shopping exercise — it is a discipline of fusion and verification. The feeds are public; the skill is turning them into a corroborated, defensible assessment without fooling yourself. This guide walks the workflow we use, the signals worth chasing, and the traps that produce confident nonsense. For the feeds themselves, start with the [open datasets catalog](/guides/open-datasets-for-maritime-domain-awareness); for definitions, the [MDA glossary](/glossary). ## The workflow: collection → fusion → verification → assessment Good maritime OSINT is a loop, not a lookup. 1. **Collection.** Pull the relevant open feeds for your area of interest — AIS, SAR and optical imagery, metocean, thermal, registries, reporting. Breadth matters because the decisive fact usually lives *between* feeds. 2. **Fusion.** Bring the feeds onto one timeline and one map. A vessel's AIS track, the SAR pass over the same water, the weather at that hour, and the news that week are only meaningful together. 3. **Verification.** Actively try to *break* each finding — is the gap really evasion, or thin coverage? Is that track real, or spoofed? Corroborate across independent sources. 4. **Assessment.** State a conclusion with its confidence and its evidence lineage, and name what you could not verify. An honest *I don't know* beats a confident error. This is exactly the [dashboards-to-investigations](/insights/dashboards-to-investigations-agentic-analyst) shift — and it is what tool-using agents now make practical at speed. ## The signals that matter - **AIS gaps.** A vessel transmitting, going silent, then reappearing is the core evasion signal — but only after you normalise against expected satellite coverage, or you will flag every mid-ocean coverage hole. - **Loitering and rendezvous.** Two vessels at low speed, co-located offshore, is a candidate **ship-to-ship (STS) transfer** — a hallmark of sanctioned-cargo movement. - **Dark ships (SAR × AIS).** A Sentinel-1 radar detection with no matching AIS return is a vessel present but silent. This is the single most powerful open signal, because radar does not care whether the transponder is on. See [detecting dark vessels](/insights/detecting-dark-vessels-ais-off). - **AIS spoofing.** Impossible speeds, teleporting or circular tracks, land-locked positions, and duplicated or **zero MMSIs** betray manipulated identity. See [AIS spoofing in the shadow fleet](/insights/shadow-fleet-ais-spoofing). - **Thermal anomalies.** NASA FIRMS hot spots near ports and industrial coasts are a rough activity proxy — flaring, refineries, sometimes vessels. - **Reporting corroboration.** Dated news tells you where attention is; used to *confirm* a physical signal, not to originate one. ## Fusing them into a finding Take the canonical question — "what was this tanker doing during its AIS gap?" No single feed answers it; the fusion does: 1. **AIS:** isolate the gap window, normalised against coverage. 2. **SAR:** pull the Sentinel-1 pass for that time and place; look for a detection with no transponder. 3. **Association:** match detections to AIS; the unmatched one is your candidate. Optical confirms in clear sky. 4. **Metocean:** was it loitering, or holding station in heavy seas? Context separates intent from circumstance. 5. **Registry:** resolve identity and ownership across MMSI and IMO; check any co-located vessel for a shared manager or sanctions link. The output is one explained conclusion with evidence from every layer. ## The traps (how OSINT goes wrong) - **Coverage-not-evasion.** The most common false positive: an AIS gap that is really a satellite coverage hole. Always normalise. - **Spoofing taken at face value.** A plausible track can be fabricated. Check for physically impossible movement and MMSI anomalies before trusting a position. - **Attention as truth.** A news spike is a hypothesis, not a finding. Corroborate with a physical signal. - **False precision.** Open imagery is coarse; do not claim a vessel count or classification the resolution cannot support. Say "no corroborating pass available" when that is the truth. Naming these limits is not a weakness of the assessment — it *is* the assessment's credibility. ## Where open OSINT scales up Open methods get you a real, defensible picture — and a sampled one. Free AIS is rate-limited; open SAR revisit is multi-day; you catch a slice of events, not all of them. The same workflow, run on **licensed satellite AIS** and **high-revisit commercial SAR**, turns sampling into persistent monitoring — and **fine-tuned, multispectral-aware models** sharpen detection and classification beyond what a generalist model can do. That transition is the subject of [from open source to operational](/insights/osint-to-operational-proprietary-data-finetuned-models) and [fine-tuning multispectral maritime models](/insights/finetuning-multispectral-maritime-models). --- **Watch the OSINT workflow running live on open data at [live.marineaware.com](https://live.marineaware.com).** To run it on licensed feeds with custom agents and fine-tuned models for your mission, [talk to us](/contact). --- ### From open source to operational: scaling a live OSINT maritime picture with proprietary data and fine-tuned models (Insight) URL: https://www.marineaware.com/insights/osint-to-operational-proprietary-data-finetuned-models Date: 2026-07-11 We built a live, open-data maritime domain awareness dashboard — fusing free AIS, dated OSINT reporting, metocean and satellite across 12 global chokepoints. Here is what open sources can and cannot show, and how the same fusion engine scales with licensed AIS and SAR, multispectral analysis, domain-fine-tuned models, and the wave of satellites launching through 2026–2027. Most maritime-intelligence vendors show you a polished dashboard and ask you to trust the data behind it. We did the opposite: we built the open-source version **in public**, on free data, and published exactly what it can and cannot see. [**live.marineaware.com**](https://live.marineaware.com) is a live maritime domain awareness (MDA) dashboard that scores 12 global chokepoints — Hormuz, Bab-el-Mandeb, the Taiwan Strait, the Baltic cable zone and more — every day, fusing four open signal layers into one picture. This piece is about what that demonstrator proves, where open data hits its ceiling, and how the identical fusion engine scales with proprietary data and fine-tuned models. ## What we built on open data alone The live dashboard fuses, per chokepoint, entirely from free sources: - **Live AIS** — vessel positions, names and speeds streamed from open AIS, plotted on a per-hotspot map with under-way, slow and stationary states. - **Dated OSINT reporting** — recent, date-stamped news and open reporting per zone, deduplicated and placed on a 21-day timeline so a coverage spike is visible, not just asserted. - **Metocean** — real wind, gusts, wave and swell height and visibility at each box, resolving to a transit read. - **Satellite** — a daily open true-color snapshot (NASA GIBS) for context. A deterministic model blends these into a 0–100 risk score, and the whole site is static, rebuilt daily. It is a genuine, working intelligence product — and it is honest about being the free tier. That honesty is the point: it makes the scale-up legible. ## Where open data hits its ceiling Three limits show up immediately, and they are the same three that separate a demo from an operational system. **AIS is rate-limited and cooperative.** Free AIS throttles hard across many areas at once, so some boxes show a handful of vessels when hundreds are really there. And AIS is self-reported — it answers what cooperative vessels choose to broadcast. A 2024 *Nature* study of two petabytes of satellite imagery found roughly **75% of industrial fishing vessel activity, and over a quarter of transport and energy vessel activity, absent from public AIS.** The interesting vessels are frequently the silent ones. **Open satellite is coarse and slow.** Copernicus Sentinel-1 SAR and Sentinel-2 optical are extraordinary public goods, but revisit is measured in days and resolution in tens of metres. A daily true-color tile is context, not evidence — far too coarse to count ships or read a wake. Open imagery samples the ocean; it does not watch it. **Open reporting is attention, not ground truth.** News volume tells you where the world is looking, which is useful, but it lags events and can spike on anniversaries. It corroborates; it cannot originate the finding. None of this makes the open picture worthless — it makes it a *baseline*. Knowing precisely where the baseline ends is what lets you specify what proprietary data has to add. ## Where proprietary data changes the picture The scale-up does not change the method — [fusion across feeds](/insights/agents-fusing-satellite-weather-registry-feeds) — it changes the grade of each feed. - **Licensed global AIS** (Kpler, Spire and satellite constellations) replaces the throttled open stream with full-resolution, near-real-time coverage over open ocean, so a vessel count is a count, not a sample, and an AIS gap is a real gap rather than a coverage artefact. - **High-resolution commercial SAR** (ICEYE, Umbra, Capella) replaces multi-day open revisit with persistent, all-weather tasking. A radar detection with no matching AIS return is the canonical [dark-ship signal](/insights/detecting-dark-vessels-ais-off); at 16–25 cm and high revisit, you can not only find the silent vessel but classify and re-observe it. - **Commercial optical and hyperspectral** adds daylight confirmation and material discrimination the free tier cannot reach. The public dashboard already demonstrates the fusion pattern. Proprietary data is what makes each layer operational — and it is exactly the upgrade path we build for clients, from private areas of interest to alerting on licensed feeds. ## Multispectral and hyperspectral: seeing what radar and RGB miss SAR sees shape and presence in any weather; RGB optical shows a scene a human recognizes. Neither reads *material*. Multispectral and hyperspectral sensors do — they sample dozens to hundreds of narrow spectral bands, so the data carries chemistry, not just geometry. For maritime work that unlocks signatures a hull outline never will: oil sheens and bilge discharge on the water, gas flaring intensity as a proxy for energy export, chlorophyll and turbidity that shape fishing behaviour, and spectral cues that help separate vessel types and cargo states. Planet's **Tanager** hyperspectral satellites capture **426 bands across the 380–2500 nm range** and already ship methane-plume products; **Pixxel** is standing up a commercial hyperspectral constellation. Fused with SAR detection and AIS identity, a multispectral pass turns "there is a vessel here" into "there is *this kind* of vessel, doing *this*, leaving *that* in the water." This is a data type the open tier essentially does not offer at useful cadence — and one where the analysis is only as good as the model reading the bands. ## Why the models have to be fine-tuned Here is the lesson that shaped the whole demonstrator: **generalist AI models are narrators, not sensors.** Hand a general-purpose vision-language model a SAR chip and it will confidently invent a vessel count. It is excellent at turning *already-computed* signals and a scene into a readable analyst brief — and useless as the detector itself. Operational detection and classification need models **fine-tuned on maritime data**: labelled vessels across SAR, optical and multispectral bands; the specific classes a mission cares about; the sea states, glint and speckle that fool a naive model; and the spoofing and gap patterns of the shadow fleet. Fine-tuning is what moves a model from "plausible" to "dependable," and it compounds — a model tuned on a client's own vessels, waters and doctrine outperforms any generic classifier on the questions that client actually asks. We keep the generalist model in its lane (narrating the evidence, with citations) and put fine-tuned, purpose-built models on the sensing. The companion piece, [fine-tuning vision models for multispectral maritime analysis](/insights/finetuning-multispectral-maritime-models), goes deeper on how. ## The satellite wave arriving through 2026–2027 The reason this scale-up gets cheaper and sharper every quarter is the sensor build-out overhead. - **SAR is scaling fastest.** As of mid-2026 ICEYE has launched **more than 70 satellites** — its four-satellite Transporter-17 batch flew on 7 July 2026 — with the **Gen4** platform delivering **up to 16 cm resolution** and a **~400 km high-resolution field of regard**, and a stated target of roughly **100 new satellites a year by 2027**. **Umbra** images down to ~16–25 cm; **Capella** runs its own SAR fleet. More satellites means higher revisit, which means the multi-day windows where dark activity currently hides keep shrinking toward hours. - **Hyperspectral is arriving commercially.** Planet **Tanager** and **Pixxel** put material-level sensing into the commercial catalogue for the first time at scale. Higher revisit plus finer resolution plus spectral depth is a step-change in what corroboration is possible — and the fusion architecture we run is built to absorb each new sensor as another feed the agent can call, not a re-platforming exercise. ## What stays the same: fusion, agents, explainability Two things do not change as you climb from open to proprietary. First, **value comes from fusion, not any single feed** — cooperative AIS plus non-cooperative SAR plus multispectral plus registries, reasoned across by a [tool-using agent](/insights/dashboards-to-investigations-agentic-analyst) at investigation time. Second, **every output has to be explainable** — a score a regulator, underwriter or watch officer can defend, with its data lineage attached. The live demo already shows the scored, sourced version of this on open data. Licensed feeds and fine-tuned models raise the confidence; they do not change the discipline. ## The honest limits More sensors do not make the picture complete. Tasking and revisit still bound when you can corroborate; even 16 cm SAR has classification limits; hyperspectral is cloud- and cadence-limited; registries are incomplete and sometimes deliberately obscured; and a fine-tuned model is only as good as its labels and as current as its last training run. A serious system surfaces these limits — "no SAR pass in the gap window," "low confidence at this sea state" — rather than papering over them. That is the same honesty the open demonstrator is built on, carried into the operational build. --- **See the open-data version working at [live.marineaware.com](https://live.marineaware.com).** When you are ready to run it on licensed AIS and SAR, multispectral analysis and models fine-tuned to your mission, [talk to us](/contact) — that is the work we do. --- ### Fine-tuning vision models for multispectral maritime analysis (Insight) URL: https://www.marineaware.com/insights/finetuning-multispectral-maritime-models Date: 2026-07-10 Generalist vision models are weak sensors on satellite imagery. Detecting and classifying vessels across SAR, optical and multispectral bands needs models fine-tuned on maritime data. Here is why domain fine-tuning matters, what multispectral and hyperspectral sensing adds, and how it compounds with the satellites launching in 2026–2027. The single most important design decision in maritime AI is knowing what your models are *for*. Get it wrong and you build a confident liar. Get it right and generalist models and fine-tuned models each do the job they are good at. This is a short, practical piece on that division of labour — and on why multispectral sensing and domain fine-tuning are where the next gains in vessel detection come from. ## Generalist models are narrators, not sensors Hand a general-purpose vision-language model a Sentinel-1 SAR chip and ask "how many ships are here?" and it will answer — fluently, and often wrong. These models are trained on everyday photographs and prose. They have no reliable prior for radar speckle, for the way a wake registers, or for the difference between a trawler and a patrol craft at 10 metres per pixel. What they are genuinely excellent at is **narration**: taking signals that have *already been computed* — a detection count, an AIS gap, a weather state — plus a scene, and turning them into a readable, cited analyst brief. So the rule we build to, and the one behind our [live open-data dashboard](https://live.marineaware.com), is blunt: **let the generalist model narrate; never let it sense.** Numbers come from purpose-built detectors on real signals. Prose comes from the language model. Cross that line and you get plausible fiction. ## The maritime domain gap Detection and classification — the sensing — need models fine-tuned on maritime data, because the domain is genuinely out-of-distribution for anything trained on consumer imagery: - **Sensors are exotic.** SAR is coherent radar with speckle and layover; multispectral and hyperspectral data are cubes of dozens to hundreds of bands, not three. A model has to be taught to read them. - **The classes are specific.** "Boat" is useless. Missions care about tanker versus bulker, fishing versus patrol, laden versus ballast, and the shadow-fleet behaviours — AIS gaps, spoofing, ship-to-ship transfers — that betray intent. - **The failure modes are adversarial.** Sun glint, heavy seas, and deliberate spoofing all exist to fool a naive model. Robustness to them has to be trained in. Fine-tuning closes this gap. A model adapted on labelled maritime imagery for the mission's classes, calibrated against known cases and validated on held-out data, is the difference between "plausible" and "dependable." ## What multispectral and hyperspectral add RGB shows a scene; SAR shows shape and presence. Neither reads material. Multispectral sensors sample several bands — typically adding near-infrared and shortwave-infrared — and hyperspectral sensors sample hundreds of contiguous narrow bands, so the data carries chemistry. For maritime intelligence that means signatures no outline can provide: oil sheens and bilge discharge on the sea surface, gas-flaring intensity as an export proxy, chlorophyll and turbidity that shape where fishing happens, and spectral separation that sharpens vessel and cargo classification. Planet's **Tanager** hyperspectral satellites resolve **426 bands across 380–2500 nm**; **Pixxel** is building a commercial hyperspectral constellation. The catch: these bands are only as useful as the model trained to interpret them — which is precisely why multispectral and fine-tuning are one story, not two. ## Fusion across sensors beats any single model No single sensor or model wins alone. The strongest maritime picture comes from fusing them, each doing what it is best at: 1. **SAR** — all-weather, day-night detection and re-observation (find the silent vessel). 2. **Optical + multispectral** — daylight confirmation and material/chemical characterization (say what it is and what it is doing). 3. **AIS + registries** — identity, ownership and behaviour history (say who is behind it). A fine-tuned detector on each sensor, orchestrated by a [tool-using agent](/insights/agents-fusing-satellite-weather-registry-feeds) and narrated by a language model, produces one explained conclusion — not five disconnected exports. ## The 2026–2027 sensor tailwind Domain fine-tuning gets more valuable exactly as the sensor supply explodes. As of mid-2026 **ICEYE** has launched more than **70 SAR satellites**, with **Gen4** reaching **up to 16 cm resolution** and a target near **100 new satellites a year by 2027**; **Umbra** images to ~16–25 cm; **Capella** operates its own SAR fleet; and hyperspectral goes commercial via **Tanager** and **Pixxel**. More revisit and more bands mean more raw signal — and more raw signal rewards better-tuned models, because the bottleneck shifts from *can we see it* to *can we interpret it*. The scale-up story for the whole picture is in [from open source to operational](/insights/osint-to-operational-proprietary-data-finetuned-models). ## Limits and honesty Fine-tuned models are not oracles. They are only as good as their labels, only as current as their last training run, and only as trustworthy as their calibration. Resolution still bounds classification; clouds and cadence still bound optical and hyperspectral; and adversaries adapt. A serious deployment reports confidence and abstains when the data is thin — "low confidence at this sea state," "no multispectral pass available" — because a defensible *I don't know* beats a confident error every time. That discipline, not the model size, is what makes maritime AI usable. --- **See the narrator-not-sensor principle running on open data at [live.marineaware.com](https://live.marineaware.com).** To put fine-tuned, multispectral-aware models on your own feeds and mission, [talk to us](/contact). --- ### How AI agents are changing the way we analyze AIS data (Insight) URL: https://www.marineaware.com/insights/ai-agents-changing-ais-analysis Date: 2026-06-30 AIS analysis is moving from hand-written geofence queries and static dashboards to agentic pipelines that plan a multi-step investigation, call the right tools, and explain the answer. Here is what changes and what stays hard. For thirty years, studying AIS data has meant more or less the same thing: write a spatial query, geofence a region or a vessel, pull the tracks, and stare at a dashboard until a pattern emerges. The data got bigger — well over a billion AIS position signals a day, from thousands of terrestrial and satellite receivers — but the *workflow* did not. The analyst was still the query engine and the reasoning engine. That is the part agents change. ## From query-and-stare to plan-and-investigate An AIS dashboard shows you pre-computed views and leaves the investigation to you. An **agentic workflow performs the investigation.** Given a question — "did this tanker do a ship-to-ship transfer in the last 30 days?" — an agent will: 1. **Decompose** the question into steps: find the vessel's track, detect gaps, look for a second vessel loitering nearby during a gap, check both vessels' registry and sanctions status. 2. **Call tools** to execute each step — a track query, a gap detector, an encounter detector, a registry lookup — rather than generating an answer from memory. 3. **Fuse** the results across feeds, resolve the vessel identity across MMSI and IMO, and weigh the evidence. 4. **Answer** in natural language, with the track, the gap window, the co-located vessel and the registry hit attached as citations. The analyst has stopped being the query engine. They ask in plain language and review an evidence trail. The mechanical work that used to take an afternoon collapses to minutes. ## Why AIS is unusually well suited to agents Three properties of AIS make it a natural fit for tool-using agents: - **It is a means, not an end.** Nobody wants "the AIS tracks"; they want a decision — is this vessel risky, where is the cargo going, will it arrive on time. That gap between raw feed and decision is exactly the reasoning layer an agent fills. - **The questions are compositional.** Real maritime questions chain many small operations — gap → encounter → identity → sanctions. That is precisely the multi-step tool-use that agents are good at and that a single SQL query is bad at. - **The answer needs corroboration.** A single feed is never enough; the credible answer fuses AIS with SAR, weather and registries. An agent orchestrating several APIs does the fusion that a human would otherwise do by hand across five tabs. ## What the research shows This is not speculative. Research systems already unify the pieces: **AIS-LLM**, for example, combines vessel-trajectory prediction, anomaly detection and collision-risk assessment behind a single natural-language interface, so an operator can interrogate a situation in words and get an explained result rather than a raw score. It is a template for the explainable maritime copilot. Underneath the agent, the models are changing too. **Time-series foundation models** (such as TimesFM and Chronos) bring zero- and few-shot forecasting to ETA, congestion and demand across many ports at once, and **geospatial foundation models** (Prithvi-EO, TerraMind) can be fine-tuned for SAR and optical vessel detection. Agents are the orchestration layer that puts these models to work against a live question — see our [data & methods](/data-and-methods) for how they fit together. ## The interoperability shift: tools, not exports The older integration pattern was *export and ingest* — dump a vendor's AIS into your warehouse and build on it. The emerging pattern is *give the agent a tool*. Standard interfaces (the Model Context Protocol and similar) let an agent call a data source directly, on demand, alongside your own systems. That is a quiet but important change: it favours **vendor-neutral orchestration** over single-platform lock-in, because the agent can reach whichever feed answers the question rather than only the one it lives inside. It is the same argument we make in [why an independent analytics layer wins](/compare). ## What stays hard Agents move the work; they do not remove the judgement. Four things stay firmly human: - **Grounding over generation.** An agent that *reasons about tool results* is trustworthy; one that *generates* vessel facts from a language model is dangerous. The whole design has to force tool-use and cite sources, or it will confidently invent a port call. Retrieval and tool-calls are the guardrail. - **Coverage normalisation.** An AIS gap only means something once it is normalised against expected satellite coverage. An agent that treats every silence as evasion is a false-positive machine. The hard maritime knowledge still has to be encoded. - **Explainability and lineage.** In sanctions, underwriting and enforcement, an unexplained score is unusable. Every agent output needs the evidence and the data lineage a human can defend to a regulator. - **Analyst-in-the-loop accountability.** The agent does the legwork and shows its work; a person makes the call and owns it. That division of labour is the point, not a limitation. ## The net effect Agents do not replace the AIS analyst — they replace the analyst's worst hours: the repetitive querying, the tab-juggling across feeds, the manual write-up. What is left is the part that was always the job — asking the right question and judging the evidence — now at a scale and speed that a lone analyst could never reach. That is how we build for [maritime domain awareness](/solutions#mda) and [sanctions screening](/solutions#sanctions): agentic pipelines that investigate, fuse and explain, with a human holding the decision. --- ### From dashboards to investigations: the agentic maritime analyst (Insight) URL: https://www.marineaware.com/insights/dashboards-to-investigations-agentic-analyst Date: 2026-06-27 What actually changes in a compliance, underwriting or watch-floor analyst's day when an agent handles triage. A walk through the new workflow — alert, investigate, explain — and where the human stays in charge. Ask a maritime compliance or underwriting analyst where their day goes and the answer is rarely "making decisions". It goes to *getting to* the decision: clearing an alert queue, pulling tracks across three systems, checking a registry, writing up what they found so someone else can sign off. The judgement is minutes; the legwork is hours. Agentic workflows invert that ratio. Here is what the new shape looks like. ## The old loop: monitor, alert, drown A rule-based monitoring system generates alerts — an AIS gap here, a speed anomaly there, an entry into a watched zone. The trouble is volume and context. Most alerts are benign (poor coverage, a legitimate stop), but every one has to be triaged by hand, and each triage means re-assembling the same evidence from scratch across separate tools. Analysts burn out clearing noise, and the real signal waits in the same queue as the false one. ## The new loop: triage, investigate, explain An agent sits in front of the analyst and changes three things. **1. It triages with corroboration, not just rules.** Before an alert reaches a human, the agent does the checks the analyst would have done: is this AIS gap real once normalised against expected satellite coverage, or just a thin-coverage patch of ocean? Is there a second vessel and a plausible encounter, or nothing nearby? Does imagery confirm a hull where the transponder went dark? Weak signals are filtered or down-ranked *with a stated reason*. What lands in the queue has already survived scrutiny. **2. It runs the first-pass investigation.** For alerts that do survive, the agent assembles the case — the track, the gap window, the co-located vessel, the identity resolved across MMSI and IMO, the registry and sanctions status, the imagery. The analyst opens a case that is already built, not a blank query window. **3. It explains, with evidence.** The output is a short narrative — "vessel went dark for 14 hours in a well-covered corridor, loitered near a second tanker also showing gaps, both linked to the same manager" — with every claim attached to its source. The analyst is reading a defensible summary, not reverse-engineering a score. ## A concrete example A P&I compliance desk gets an overnight flag on a tanker up for renewal. In the old loop, an analyst arrives, pulls the vessel in the tracking tool, notices a gap, switches to the registry tool, checks ownership, opens the sanctions list, cross-references, and an hour later writes a note. In the agentic loop, the desk arrives to a ranked case file: the gap already normalised and judged significant, the ownership network already resolved and showing a shared manager with a previously flagged vessel, the sanctions check already run, and a two-paragraph explanation with citations. The hour of assembly is already done. The analyst spends their time on the one thing that matters — deciding whether to bind, and being able to defend it. That is the machinery behind our [dark-fleet and sanctions screening](/solutions#sanctions): predictive, explainable scoring that catches exposure before cover is bound. ## Why explainability is the whole game here In a trading context a wrong signal costs money; in compliance, underwriting and enforcement a wrong *and unexplainable* call costs a license or a legal position. That is why the agent's job is not to decide — it is to **make the human's decision faster and more defensible.** Every output carries its data lineage and a reason a regulator or an underwriter can accept. An agent that returns a confident number with no evidence is worse than the dashboard it replaced. ## Where the human stays in charge The division of labour is deliberate: - **The agent** watches, triages, corroborates, assembles and explains — the mechanical, repetitive, tab-juggling work. - **The analyst** judges the built case, makes the call, and owns the outcome. This is the same analyst-in-the-loop principle we apply on the government side for [maritime domain awareness](/solutions#mda), where a small team has to cover a large sea: the agent turns a flood of detections into a short, ranked, explained lead list, and the human decides what to act on. The watch floor of 2026 is not un-staffed. It is the same analysts, freed from the legwork, making more decisions with better evidence — which was always what the job was supposed to be. For the mechanics underneath, see [how agents are changing AIS analysis](/insights/ai-agents-changing-ais-analysis). --- ### Beyond AIS: how agents fuse satellite, weather and registry feeds (Insight) URL: https://www.marineaware.com/insights/agents-fusing-satellite-weather-registry-feeds Date: 2026-06-24 AIS is one feed among many. The hard maritime questions need SAR, optical, metocean and registry data fused together — and tool-using agents are what make on-demand, multi-feed fusion practical instead of a data-engineering project. AIS gets the attention because it is the feed you can see on a map. But it is one feed, it is cooperative, and it is spoofable — and a 2024 *Nature* study of two petabytes of satellite imagery found that **roughly 75% of industrial fishing vessels and over a quarter of transport and energy vessel activity are absent from public AIS.** Any serious maritime question therefore lives in the space *between* feeds. The interesting problem was never AIS; it was fusion. ## Fusion used to be a project. Now it is a call. Historically, combining feeds meant a data-engineering programme: license SAR, license weather, license registries, build pipelines to normalise and join them, keep them fresh, and only then let an analyst ask a question. The fusion was baked in advance, which meant it was expensive, brittle, and limited to the questions someone anticipated. A tool-using agent changes the economics. Instead of pre-joining everything, the agent **calls each feed on demand, in the course of answering a specific question.** The fusion happens at investigation time, driven by what the question needs — not at ingestion time, driven by what an engineer guessed. That makes multi-feed analysis something you *do*, not something you *build*. ## What a fused investigation looks like Take the canonical dark-vessel question — "what was this tanker doing during its AIS gap?" A single feed cannot answer it. An agent fusing feeds can: 1. **AIS (cooperative):** pull the vessel's track and isolate the gap window, normalised against expected satellite coverage so a thin-coverage patch is not mistaken for evasion. 2. **SAR (non-cooperative):** request Copernicus Sentinel-1 imagery for that time and place — radar sees hulls in any weather, day or night — and run detection to find vessels present but silent. 3. **Association:** match radar detections to AIS returns; a detection with no matching transponder is a candidate dark ship. Optical imagery (Sentinel-2, or commercial ICEYE/Planet) confirms in clear conditions. 4. **Metocean (environmental):** add weather and current context — was the vessel loitering, or holding station in heavy seas? Context separates intent from circumstance. 5. **Registry (reference):** resolve identity across MMSI and IMO and pull ownership from Equasis, Clarksons and S&P, then check any co-located vessel — is there a shared manager, a known shell structure, a sanctions link? The output is one explained conclusion with evidence from every layer, not five disconnected exports. This is the pipeline behind our [maritime domain awareness](/solutions#mda) and the sanctions work in [detecting a vessel that has turned off its AIS](/insights/detecting-dark-vessels-ais-off). ## Different feeds, different questions Fusion is not only for dark vessels. The same on-demand orchestration serves the whole value chain: - **Emissions:** fuse AIS activity with vessel fuel curves and calibrate against verified EU MRV data to produce MRV-grade CO₂ estimates — the basis of our [decarbonization work](/solutions#emissions). - **ETA and routing:** fuse AIS voyage history with CMEMS currents, WaveWatch III waves and ECMWF wind, bounded by GEBCO bathymetry, for weather-aware arrival prediction. - **Flow intelligence:** fuse port-call time series with UN Comtrade trade data for demand and floating-storage signals. The feed list is long — the full catalogue is on our [data & methods](/data-and-methods) page — but the pattern is always the same: the agent reaches for whichever sources the question needs. ## Retrieval over the unstructured, too Not every feed is a clean API. A lot of decisive maritime context is text: sanctions designations, Port State Control records, notices to mariners, class-society bulletins, news. Agents handle this through **retrieval** — searching a corpus of registry and regulatory documents and grounding the answer in what they find, with citations. That is what lets an agent say "this ownership structure matches a designation published last week" instead of relying on a stale, pre-built table. ## The interoperability that makes it work — and who it favours On-demand fusion depends on the agent being able to *reach* the feeds. Standard tool interfaces — the Model Context Protocol and similar — let an agent call AIS, satellite, weather and registry sources directly, alongside a client's own systems, without a bespoke integration for each. That has a strategic consequence: it favours **vendor-neutral orchestration.** An agent that can call any feed answers the question with the best available source, rather than only the data that happens to live inside one platform. It is the technical foundation of the [independent analytics layer](/compare) we argue for — and the reason fusion, not any single feed, is the durable advantage. ## The honest caveats Multi-feed fusion is powerful and imperfect. Satellite tasking and revisit limit when you can corroborate; imagery resolution limits what you can classify; registry data is incomplete and sometimes deliberately obscured; latency and coverage vary by feed. A good agent surfaces these limits rather than papering over them — "no SAR pass was available within the gap window" is a more useful answer than a false certainty. Fusion widens the picture; it does not make it complete, and saying so is part of doing it well. --- ### How IMO CII, EU ETS and FuelEU Maritime interact in 2026 (Insight) URL: https://www.marineaware.com/insights/cii-eu-ets-fueleu-2026 Date: 2026-06-10 A plain-English guide to the three overlapping shipping emissions regimes hitting fleets in 2026 — what each measures, where they conflict, and how to model total cost. Three separate rulebooks now govern the carbon a ship emits into and around Europe, and 2026 is the year they all bite at once. They measure different things, use different boundaries, and — critically — optimising for one can quietly make another worse. This is the map. ## The three regimes at a glance **IMO CII (Carbon Intensity Indicator)** rates a ship's *operational* carbon intensity on an A–E scale. It applies to ships of 5,000 GT and above and has been in force since 2023, with ratings issued from 2024. A vessel rated D for three consecutive years, or E for a single year, must submit a corrective action plan. The required intensity tightens every year — roughly 11% below the 2019 baseline by 2026, heading toward about 21.5% by 2030. **EU ETS** is cap-and-trade. Shipping companies must surrender emission allowances (EUAs) for their emissions. Coverage phased in during 2024–2025 and reaches **100% of covered emissions from 1 January 2026**, with methane and nitrous oxide added alongside CO₂. The geographic boundary is 100% of intra-EU voyages, 100% at berth in EU ports, and 50% of voyages between an EU and a non-EU port. **FuelEU Maritime** caps the **well-to-wake greenhouse-gas intensity** of the energy a ship uses, for vessels of 5,000 GT and above calling at EU ports. It has applied since 1 January 2025, allows **pooling and banking** of compliance across vessels and years, and issues penalties for exceedance. ## Where they conflict The regimes reward different behaviours: - **CII** rewards low carbon *per transport-work unit*. The blunt way to improve it is to slow down — but slow steaming can strand cargo and hurt commercial performance. - **EU ETS** rewards lower *absolute* covered emissions, because every tonne costs an allowance. - **FuelEU** rewards lower *lifecycle fuel intensity*, which pushes toward alternative fuels regardless of speed. A decision that flatters your CII rating may do little for your ETS bill, and a fuel switch that helps FuelEU may be the wrong lever for CII. Optimising each rule in isolation leaves money on the table and can create compliance whiplash. ## The IMO Net-Zero Framework is coming behind them At MEPC 83 in April 2025 the IMO approved, in principle, the first global GHG pricing mechanism for an entire sector — a Global Fuel Standard plus carbon pricing, targeting net-zero around 2050. Formal adoption was adjourned by a year at the October 2025 extraordinary session, with expected entry into force around 2027. For anyone modelling compliance cost, it is a fourth layer on the horizon, not a distant abstraction. ## How to actually model it The only defensible approach is a single model that speaks all of these languages at once and expresses the answer in the currency owners care about — cost per voyage and cost per year: 1. **Build a bottom-up emissions model** from AIS activity multiplied by vessel-specific fuel curves. 2. **Calibrate it against verified EU MRV data**, which publishes per-ship annual CO₂ and fuel for vessels over 5,000 GT calling at EU ports, so the numbers are MRV-grade rather than a spreadsheet guess. 3. **Layer the cost engine** across CII, EU ETS, FuelEU and UK ETS, optimising FuelEU pooling and banking. 4. **Route with the trade-off explicit** — weather routing that weighs voyage time against fuel and against CII penalty, not one in isolation. That is exactly the work in our [fleet decarbonization practice](/solutions#emissions): turning four overlapping rulebooks into one number a board, a lender and a charterer can all act on. --- ### The maritime AI market in 2026: what's driving the surge (Insight) URL: https://www.marineaware.com/insights/maritime-ai-market-2026 Date: 2026-06-02 Maritime-tech AI funding hit roughly $1.75B in Q1 2026, up about 192% year on year. Here is what is pulling capital in — regulatory tailwinds, the dark fleet, and dual-use defense demand — and what to check in diligence. Capital has found maritime AI. In Q1 2026, maritime-tech AI companies raised roughly **$1.75 billion — up about 192% year on year** — and around **45% of maritime startups now embed AI**, up from 27.5% in 2024. That is not a fad cycle; it is three durable forces arriving at once. ## What is pulling capital in **1. Regulatory tailwinds create mandatory spend.** Unlike discretionary software, emissions compliance is not optional. EU ETS reached 100% coverage of covered shipping emissions in January 2026, FuelEU Maritime penalties land by mid-2026, and the IMO's Net-Zero Framework — the first global carbon price for an entire sector — is expected to enter force around 2027. Every one of those rules turns into budget for the analytics that model and minimise the cost. **2. The dark fleet turned risk into a data problem.** A sanctioned shadow fleet of 600–800 tankers has pushed insurers and governments from reactive, list-based screening toward *predictive* risk scoring and maritime domain awareness — a structural, recurring demand for exactly the AIS-plus-satellite fusion that AI-native firms build. **3. Dual-use defense demand.** Maritime AI is increasingly a security category. The Pentagon committed **$150M to a maritime-tech venture fund** (Mare Liberum) in early 2026, and foreign military sales such as the **$131M HawkEye 360/SeaVision** deal signal sustained government appetite for dark-vessel detection and sensor fusion. Individual rounds reflect the momentum: Orca AI's **$72.5M Series B** for autonomous shipping, and a steady stream of earlier-stage orbital-AI and maritime-intelligence raises. ## The catch: a narrowing data landscape The most important diligence fact in this market is that the "independent" AIS-vendor landscape has consolidated hard. Kpler absorbed **Spire Maritime, exactEarth, MarineTraffic and FleetMon** into a unified "Kpler AIS," and S&P Global took ORBCOMM's AIS business. A startup whose entire moat is *access* to a data feed is standing on ground that a handful of consolidators now own. That reframes what "defensible" means. The durable moat is not the feed — it is the **fusion and the models**: combining cooperative AIS with non-cooperative SAR, metocean and registry data, and the agentic pipelines that turn it into explainable decisions. ## What to check in diligence 1. **Is the model real?** Distinguish genuine predictive analytics from a dashboard sitting on a purchased feed. Ask what the system predicts, how it is validated, and against what ground truth (EU MRV for emissions, confirmed designations for sanctions). 2. **Where is the moat?** If it is data access alone, price in the consolidation risk. If it is fusion, methods and explainability, that is more defensible. 3. **Does the TAM rest on tailwinds or events?** Regulation-driven demand (EU ETS, FuelEU, IMO Net-Zero) is durable. Demand riding a single sanctions package is not. 4. **Can the team defend the output?** In insurance and government, an unexplainable score is unusable. Explainability and data lineage are commercial features, not nice-to-haves. This is the lens we bring to [investor diligence and advisory](/solutions#diligence): domain-literate technical due diligence, market sizing anchored to dated regulatory drivers, and competitive teardowns of the incumbent and challenger set — with a point of view, not a fence-sit. --- ### What is the dark fleet — and why it matters in 2026 (Insight) URL: https://www.marineaware.com/insights/what-is-the-dark-fleet Date: 2026-05-22 The shadow fleet has grown to 600–800 tankers, roughly 10–15% of the global tanker fleet. Here is how it operates, why P&I withdrawal became the sharpest enforcement lever, and how exposure is detected. The "dark fleet" — or shadow fleet — is the fleet of tankers built to move sanctioned and price-capped oil without being seen. By early 2026 it numbered an estimated **600–800 vessels, roughly 10–15% of the world tanker fleet**, and it has reshaped how sanctions are enforced and how maritime risk is priced. ## How it operates Dark-fleet vessels share a recognisable tradecraft: - **AIS manipulation.** Transponders are switched off to create gaps around loading or transfer, or spoofed to broadcast a false position — circular tracks, "teleporting," or land-locked coordinates. - **Ship-to-ship (STS) transfers** in open water, often at night and away from normal transfer zones, to break the chain of custody on a cargo. - **Flag-hopping and opaque ownership.** Frequent re-flagging and layers of shell companies obscure the beneficial owner behind a vessel. - **Questionable insurance.** Cover from unknown or unrated providers, or forged certificates — by early 2026 roughly a third of tankers in some corridors were presenting sanctioned or Russian-linked insurance certificates. ## Why P&I withdrawal became the key lever Sanctions authorities learned that the most effective pressure point is not the cargo or the flag — it is the insurance. Protection and indemnity (P&I) cover is effectively mandatory for a tanker to trade and call at reputable ports. When the EU's 20th sanctions package added **43 shadow-fleet vessels at once**, it did more than name them: it stripped their access to legitimate cover. That is precisely what makes accidental exposure existential for an insurer. Binding cover on a vessel that is later designated does not just create a bad risk — it can pull the insurer into the enforcement action. ## The shift from reactive to predictive Screening a vessel's name against a sanctions list tells you about yesterday's risk. The whole design of the shadow fleet is to stay off those lists — new shells, fresh MMSIs, re-flagging, carefully timed gaps. List-based screening is therefore structurally late. The alternative is **behavioural risk scoring**: asking whether a vessel *acts* like one that is hiding something. AIS gaps normalised against expected satellite coverage, kinematically impossible positions that betray spoofing, loitering in transfer zones, and rendezvous with other high-risk tonnage all precede a designation rather than follow it. Fused with **identity resolution** across MMSI, IMO number and beneficial owner — and graph analysis of the ownership network behind a vessel — this turns "this ship looks risky" into "this ship sits inside a structure we have seen before." ## Corroborating with satellites Behaviour is the first signal; imagery is the confirmation. When a vessel goes dark, synthetic aperture radar (Sentinel-1) still sees the hull, in any weather, day or night. A radar detection with no matching AIS return is a candidate dark ship — the same [SAR-and-AIS fusion](/insights/detecting-dark-vessels-ais-off) that underpins maritime domain awareness for governments. This is the core of our [dark-fleet and sanctions screening work](/solutions#sanctions): explainable, auditable risk scores that catch exposure *before* cover is bound, with the data lineage regulators and underwriters require. --- ### How to detect a vessel that has turned off its AIS (Insight) URL: https://www.marineaware.com/insights/detecting-dark-vessels-ais-off Date: 2026-04-30 A technical primer on dark-vessel detection — why AIS gaps happen, how synthetic aperture radar sees hulls that transponders hide, and how detection-to-AIS fusion turns raw imagery into ranked leads. A vessel that has switched off its transponder is not invisible — it is only invisible to the one sensor that depends on its cooperation. Detecting it is a fusion problem, and the method is now well established. Here is how it works. ## Start with the gap AIS (the Automatic Identification System) is a *cooperative* sensor: the ship broadcasts its own position over VHF. That cooperation can stop for legitimate reasons — patchy satellite coverage far from shore, or a deliberate safety decision — or because someone wants to hide loading, a transfer, or a destination. An **AIS gap** — a vessel going silent then reappearing — is the first signal. But a raw gap is noisy. To be useful it must be **normalised against expected satellite AIS coverage**: in a region where the constellation rarely passes, silence means little; in a well-covered corridor, the same silence is loud. Getting this normalisation right is what separates a real dark-vessel pipeline from a false-positive generator. ## See the hull with radar The gap tells you *when* to look. Non-cooperative sensors tell you *what is there*. **Synthetic aperture radar (SAR)** is the workhorse. It is an active sensor — it provides its own illumination — so it works in any weather, day or night, and penetrates cloud. The open **Copernicus Sentinel-1** constellation images in C-band; its Interferometric Wide swath (roughly 5×20 m) is well suited to maritime detection. Crucially, Sentinel-1C (launched December 2024) and 1D (November 2025) carry an **onboard AIS receiver**, so the radar detection and the transponder return can be correlated natively. **Optical imagery** (Sentinel-2 at 10 m, or commercial Planet, ICEYE and Umbra down to ~25 cm) confirms and classifies detections in clear conditions. ## Fuse detections to AIS The intelligence is in the association: 1. **Detect** candidate vessels in the imagery — classical CFAR detectors or CNN ship detectors on the SAR scene. 2. **Associate** each detection with any AIS track present at that place and time. 3. **Reason** over what is left. A detection that matches an AIS return is a cooperative vessel. A detection with **no** matching AIS is a candidate dark ship. Layer behavioural context on top — loitering, two-vessel **rendezvous** consistent with an STS transfer, and pattern-of-life scoring — and you convert a scene full of blips into a short, ranked list of leads. ## Why this is not optional The scale of the blind ocean was quantified in a 2024 *Nature* study that processed two petabytes of Sentinel-1 and Sentinel-2 imagery from 2017–2021 with machine learning. It found that **roughly 75% of industrial fishing vessels, and more than a quarter of transport and energy vessel activity, were absent from public AIS tracking**. For sanctions enforcement, IUU-fishing patrols and maritime domain awareness, the dark vessels are not the edge case — they are the majority. ## Watch the spoofing, not just the silence Turning AIS off is one tactic; broadcasting a lie is another. **Location spoofing** transmits a false GNSS position — circular tracks, impossible jumps, land-locked coordinates. **Identity spoofing** broadcasts another vessel's MMSI, a scrapped vessel's MMSI, or a duplicated "pirate" MMSI. Detection leans on kinematic plausibility checks, the frequency-of-arrival of AIS messages at satellites, and cross-validation against the same SAR and optical imagery. This detection-to-AIS fusion is the engine behind our [maritime domain awareness](/solutions#mda) and [sanctions screening](/solutions#sanctions) work — wrapped in an analyst-in-the-loop copilot so every lead arrives with the evidence a human can defend. --- ### Anchors in the dark: reading attacks on undersea infrastructure (Insight) URL: https://www.marineaware.com/insights/baltic-cables-ais Date: 2025-01-22 A string of pipeline and cable failures in the Baltic since 2023 has turned vessel tracking into forensic evidence. AIS tracks, sudden slowdowns and anchor-drag signatures are how investigators reconstruct what a ship did on the seabed. The seabed is where the modern world actually runs — the cables that carry its data and power, the pipelines that carry its gas. Since 2023, a series of failures in the Baltic has made those hidden lines a front line, and vessel tracking the forensic tool that reads what happened. ## A pattern emerges **October 2023 — Balticconnector.** The gas pipeline and a parallel telecom cable between Finland and Estonia were damaged. Investigators linked the damage to a **dragged anchor**, and attention focused on a Hong Kong-flagged container ship whose track crossed the site; an anchor was later recovered near the pipeline. Finland treated it as a criminal investigation. **November 2024 — two cables cut.** Two Baltic telecom cables — one between Lithuania and Sweden, one between Finland and Germany — failed within a day. A bulk carrier was suspected of **dragging its anchor for tens of nautical miles** across both. **25 December 2024 — Estlink 2.** The Estlink 2 power cable between Finland and Estonia, along with several telecom cables, was damaged. Finnish authorities boarded and **seized the tanker Eagle S** — a Cook Islands-flagged vessel linked to the sanctioned [shadow fleet](/insights/shadow-fleet-ais-spoofing) — on suspicion of anchor-dragging, and recovered evidence consistent with a dragged anchor. Three winters, one recurring signature: a ship, an anchor, and a line on the seabed that stopped working. In January 2025, NATO responded with an enhanced Baltic patrol presence. ## How vessel tracking becomes evidence You cannot see the seabed. But you can see the ship, and that turns out to be enough to reconstruct a great deal. When infrastructure fails at a known location and time, investigators pull the AIS tracks of every vessel that was in the vicinity and look for the **signature of a dragged anchor**: - A **slowdown** to a few knots — the pace of a ship whose anchor is catching on the bottom, far slower than a normal transit. - A **track that runs along or across** the damaged cable or pipe at the exact time it failed. - Sometimes an **AIS gap or anomaly** around the event, or a heading that no ordinary voyage would take. - Later, physical corroboration: a **missing anchor**, or drag marks on the seabed matching the track. Correlating the tracked position and speed of a suspect vessel with the location and timing of the damage is what moves a case from "a cable broke" to "this ship was over it, moving like this, at that moment." In the Baltic incidents, that reconstruction — built on AIS — has been central to identifying and detaining suspect vessels. ## Why this is hard, and where it goes wrong Anchor-dragging is deniable by design. An anchor can be "accidentally" left down; a ship can claim weather or error. That is exactly why the *data* matters — a single slow pass might be innocent, but a track that hugs the infrastructure for tens of miles is very hard to explain away. It is also why the analysis cannot rest on AIS alone. A vessel involved in this kind of activity has every incentive to **switch off, spoof, or falsify** its broadcast — the same [manipulation tradecraft](/insights/shadow-fleet-ais-spoofing) seen across the shadow fleet. Robust attribution corroborates the AIS story against other evidence: satellite imagery placing the hull, the physical anchor, seabed surveys, and the vessel's ownership and history. The overlap between the ships implicated in Baltic incidents and the sanctioned shadow fleet is itself part of the picture. ## The stakes Undersea cables carry an estimated **95%+ of intercontinental data**; subsea power and gas links hold national grids together. They were built on an assumption of benign neglect — that no one would bother attacking something so hidden. The Baltic has ended that assumption, and it has made **maritime domain awareness** a matter of critical-infrastructure defence, not just sanctions or fishing. For the agencies protecting these lines, the task is the one we build for in [maritime domain awareness](/solutions#mda): fuse AIS with satellite and behavioural analysis to spot the vessel loitering where it should not be, moving how it should not move, *before* the cable goes dark — and to reconstruct, with defensible evidence, what happened when one does. The seabed is invisible. The ships above it are not. --- *Sources: Finnish, Estonian, Lithuanian and Swedish authority statements on the Balticconnector (Oct 2023), November 2024 cable cuts, and Estlink 2 / Eagle S incident (Dec 2024); NATO Baltic patrol announcements (Jan 2025); contemporaneous reporting. Investigations were ongoing at time of writing.* --- ### Going dark: a field guide to how the shadow fleet hides on AIS (Insight) URL: https://www.marineaware.com/insights/shadow-fleet-ais-spoofing Date: 2024-09-10 Since the 2022 oil price cap, a fleet of hundreds of aging tankers has moved sanctioned crude by manipulating the one system meant to make them visible. A practical taxonomy of the AIS tradecraft — and how it gets caught. When the G7 introduced a price cap on Russian oil in December 2022, it created an incentive worth billions: move crude above the cap without touching Western insurance, finance or services. The response was a **shadow fleet** — hundreds of aging tankers, opaquely owned, that trade outside the mainstream system. And because trading in the shadows means defeating the one system built to make ships visible, the shadow fleet has become the world's most active laboratory for **AIS manipulation**. This is a field guide to how it is done, and how it is caught. (For the strategic picture — the fleet's size, why P&I withdrawal is the sharpest lever — see our companion piece, [what is the dark fleet](/insights/what-is-the-dark-fleet).) ## The four techniques Almost everything the shadow fleet does to AIS falls into four families. ### 1. Going dark The simplest move: switch the transponder off. A vessel that was broadcasting goes silent — typically around a loading at a sanctioned port or a ship-to-ship transfer — then reappears later, "clean," some distance away. The gap is the point: it breaks the continuous track that would otherwise link the cargo to its origin. Going dark is also the crudest move, because a gap is itself a signal. A tanker that reliably vanishes for eighteen hours near a known transfer zone and reappears riding lower in the water has told you a great deal precisely by telling you nothing. ### 2. Location spoofing More sophisticated is to keep broadcasting, but to broadcast a **lie**. Location spoofing transmits a falsified GPS position so the vessel appears somewhere it is not — loitering innocently in one sea while physically loading in another. The tell is physical impossibility: tracks that trace neat circles, "teleport" hundreds of miles between pings, or run across dry land. **Global Fishing Watch has documented tankers falsifying their AIS positions to conceal entry into Russian Black Sea ports** — a clean example of the technique in service of sanctions evasion. A related problem is upstream **GNSS spoofing and jamming**, heavy near certain Russian export terminals and in conflict zones, which corrupts the real position feeding the transponder. The effect on the data looks similar; the intent differs. ### 3. Identity manipulation An MMSI number is software-mutable and an AIS-broadcast ship name is just text, so both can be forged. Shadow-fleet vessels have broadcast the identity of another ship, a scrapped ship, or a duplicated "pirate" MMSI — muddying which hull is actually where. This is why serious analysis never trusts the broadcast identity alone and instead **resolves identity** across the stable IMO number and multiple registries. ### 4. Disguised transfers The point of most of the above is the **ship-to-ship (STS) transfer** — moving cargo between vessels at sea so a "dirty" cargo boards a "clean" ship. STS transfers cluster in known areas (off Greece, off West Africa, off Malaysia), and detecting them means spotting two vessels holding station together, often with one or both showing gaps or spoofed positions during the rendezvous. ## Why AIS manipulation is self-defeating — eventually Here is the paradox the shadow fleet cannot escape: **every technique for hiding on AIS creates its own signature.** A gap is a signal. An impossible track is a signal. A borrowed identity is a signal. The manipulation does not make a vessel invisible; it makes it *anomalous* — and anomalies are exactly what a well-built detection pipeline looks for. The trick is to stop treating AIS as ground truth and start treating it as a **claim to be checked**: - **Kinematic plausibility.** Is this track physically possible for this ship? Circles, jumps and land crossings fail the test. - **Coverage-normalised gaps.** Is this silence meaningful, or just thin satellite coverage? A gap only counts once normalised against the coverage expected in that patch of ocean. - **Non-cooperative corroboration.** Satellite radar (SAR) sees a hull whether or not it is broadcasting. A radar detection with no matching AIS return is a candidate dark ship. This AIS-gap-to-SAR fusion is the backbone of [detecting a vessel that has turned off its AIS](/insights/detecting-dark-vessels-ais-off). - **Identity and network resolution.** Resolve MMSI to IMO to beneficial owner across registries, and map the ownership network — the shell structures behind sanctioned tonnage tend to reuse the same patterns. ## Why it matters beyond sanctions The tradecraft catalogued here is not confined to oil. The same techniques — going dark, spoofing, false identity — reappear wherever a vessel wants to act unseen: in [illegal fishing](/insights/dark-fishing-nature-study), in the [Red Sea](/insights/red-sea-houthi-ais) where ships hid to avoid attack, and in the [Baltic](/insights/baltic-cables-ais) where vessels implicated in cable damage overlap heavily with the sanctioned shadow fleet. Understanding how ships lie on AIS is, increasingly, a general-purpose maritime skill. By 2024, the EU, UK and US were sanctioning shadow-fleet vessels by name in the hundreds — a recognition that the fleet had become systemic. Each listing is, in effect, the end of a detection story: a vessel whose anomalies finally added up. Our [dark-fleet and sanctions screening](/solutions#sanctions) work is about getting to that conclusion earlier — scoring the behaviour before the cargo is lifted, not after the ship is designated. The shadow fleet's bet is that the ocean is too big and the data too noisy to catch them. The counter-bet — the one worth making — is that hiding leaves a trace, and that the trace is readable. --- *Sources: G7/EU oil price cap documentation (Dec 2022); Global Fishing Watch analysis of AIS falsification into Russian Black Sea ports; EU/UK/US shadow-fleet vessel designations (2024); contemporaneous reporting and academic analysis of AIS spoofing. Figures are widely reported estimates.* --- ### Watching the grain corridor: maritime data and food security (Insight) URL: https://www.marineaware.com/insights/black-sea-grain-ais Date: 2024-07-18 The 2022 Black Sea Grain Initiative moved millions of tonnes of Ukrainian grain through a mined war zone. Vessel tracking was the instrument of trust that made it possible — and the tool that later exposed grain moved in the dark. Food security is not usually a maritime-data story. In 2022 it became one. When Russia's invasion blockaded Ukraine's Black Sea ports, global grain prices spiked and the countries most dependent on Ukrainian wheat faced a supply shock. The deal that reopened the ports ran on vessel tracking — and so did the effort to expose the grain that moved in the dark. ## Trust, built from position reports In **July 2022**, the UN and Turkey brokered the **Black Sea Grain Initiative**, allowing exports to resume from **Odesa, Chornomorsk and Pivdennyi**. The problem was obvious: how do you persuade a commercial bulk carrier and its insurer to sail into a mined, active war zone, and persuade both sides that the ships carry grain and nothing else? The answer was radical transparency, enforced through tracking. A **Joint Coordination Centre (JCC)** in Istanbul — staffed by Ukraine, Russia, Turkey and the UN — ran the corridor on data: - Every participating vessel had to **broadcast its position continuously** and file position reports at regular intervals (on the order of every couple of hours). - AIS and satellite tracking let the JCC verify, **24/7**, that ships stayed strictly inside the narrow humanitarian corridor threading between minefields. - Ships were inspected at Istanbul on the way in and out to confirm cargo and that no weapons moved. - The JCC logged **dozens of route deviations** — vessels straying from the agreed coordinates — and resolved each directly with the operator. This is a striking thing to sit with: vessel tracking was not a background convenience here. It was the **mechanism of trust** that made a humanitarian corridor function between warring parties. Over roughly a year, the initiative moved tens of millions of tonnes of grain before Russia withdrew in **July 2023**, after which Ukraine ran its own corridor hugging the western Black Sea coast — again, monitored ship by ship. ## The other use: exposing grain in the dark The same data cuts the other way. While AIS made the sanctioned corridor auditable, it also let open-source investigators track the grain that moved *without* consent — bulk carriers loading Ukrainian grain at occupied ports such as **Sevastopol** in Crimea, then attempting to launder its origin on the way to market. The tradecraft was familiar: vessels **going dark** around the occupied port, or **spoofing positions** to disguise where they had actually been — the same [AIS manipulation](/insights/shadow-fleet-ais-spoofing) used across the shadow fleet to hide sanctioned oil. And it was caught the same way: a gap over a known loading point, a track that does not add up, a vessel appearing "clean" at sea after vanishing at a port it should not have visited. ## Why this matters The grain corridor is a compact lesson in what maritime data is *for*: - **Tracking can be an instrument of trust.** In a context where no party trusted any other, the shared, verifiable picture of where every ship was — and whether it stayed in the lane — was what let trade happen at all. Transparency was not a byproduct; it was the product. - **The same lens serves accountability.** Verifying legitimate trade and exposing illegitimate trade are the same analytical act from opposite ends: does the vessel's tracked behaviour match its story? Corridor compliance and stolen-grain laundering are both answered by that question. - **Food, like oil, moves on ships that can lie.** Grain is not exempt from the manipulation catalogued elsewhere on this site. Auditing an agricultural supply chain increasingly means reading AIS with the same scepticism you would bring to a sanctioned tanker. For the traders, insurers and agencies with a stake in the world's food flows, the grain corridor proved that maritime analytics is not only about risk and enforcement — it is, at times, about keeping bread on shelves. It is the same [flow-intelligence](/solutions#flows) discipline applied to the most basic commodity there is. --- *Sources: UN Black Sea Grain Initiative official materials and updates (2022-2023); Joint Coordination Centre reporting; open-source investigations into grain shipments from occupied Ukrainian ports; contemporaneous reporting on the July 2023 withdrawal. Figures are widely reported estimates.* --- ### The six days that put ship tracking on every screen: the Ever Given (Insight) URL: https://www.marineaware.com/insights/ever-given-suez-ais Date: 2024-03-23 In March 2021 a single grounded container ship blocked the Suez Canal and roughly 12% of global trade. AIS turned an invisible maritime event into a live global spectacle — and a lesson in supply-chain fragility. For six days in March 2021, the most-watched object on Earth was a container ship wedged sideways across a canal in the Egyptian desert. What made it watchable was AIS — and that is the part worth remembering. ## What happened On the morning of **23 March 2021**, the **Ever Given** — a 400-metre, ~200,000-tonne container ship, one of the largest afloat — was transiting the Suez Canal northbound when it lost control in high winds and a dust storm, swung across the channel, and grounded diagonally with its bow in one bank and its stern near the other. It plugged the canal completely. The Suez Canal is not a minor waterway. It carries roughly **12% of global trade** and about **30% of the world's container traffic**, and it is the shortest sea route between Asia and Europe. With it blocked, there was no room to pass. Ships began to queue. By the time a flotilla of tugs and dredgers, aided by a spring high tide, refloated the vessel on **29 March**, more than **400 ships** were waiting at both ends or diverting the long way around the Cape of Good Hope. Lloyd's List estimated the blockage was holding up around **$9.6 billion of goods per day**. The knock-on congestion at destination ports, and the container and equipment imbalances it created, were felt for months. ## Why AIS made it a global event Groundings happen. What was unprecedented was that the whole world could *watch this one*, live, from their phones. The Ever Given carried AIS, as all large ships must under SOLAS — a VHF broadcast of its identity, position, course and speed, intended for collision avoidance. That same broadcast, aggregated by tracking services, meant its final swing across the canal was recorded second by second. In the days before the grounding, its AIS track outside the canal had even traced a crude doodle, which went viral once the ship became famous. But the more important picture was the **queue**. As hundreds of ships piled up at Port Said and Suez, AIS rendered the traffic jam as a dense cluster of icons that anyone could see. A supply-chain abstraction — "a chokepoint is blocked" — became a concrete, legible image. Traffic to public tracking sites spiked. For millions of people, the Ever Given was the first time they had ever looked at AIS, and the first time they viscerally understood that the goods in their homes cross a handful of narrow waterways. ## The business lesson The Ever Given was a masterclass in **chokepoint risk**. A single point of failure, blocked for less than a week, disrupted global container flows for a season. The event accelerated conversations that were already stirring after the pandemic's port congestion: - **Visibility is not optional.** Companies that could see the queue building and model its impact on their inbound containers reacted days before those relying on carrier updates. AIS-derived congestion and ETA analytics moved from nice-to-have to board-level. - **Chokepoints are correlated risk.** Suez, the Panama Canal, the Strait of Hormuz, the Bab-el-Mandeb, the Malacca Strait — a surprisingly small set of narrows carries most of the world's seaborne trade, and each is a single event away from disruption. The years since have proved the point at the [Red Sea](/insights/red-sea-houthi-ais) and a drought-stricken Panama Canal. - **The data was already there.** No new sensor was needed to see the Ever Given crisis. The AIS was broadcasting the whole time; what was missing, for most firms, was the analytical layer to turn it into a decision. ## What it means for how we work The Ever Given is the clearest possible illustration of the gap MarineAware exists to close. Raw AIS made the blockage *visible*; it did not make it *actionable*. Seeing 400 icons stack up is a spectacle. Knowing which of your containers are in that stack, when they will now arrive, what it does to your demurrage exposure, and how to re-plan — that is analysis, and in March 2021 most companies did not have it. Three years on, the chokepoints have only become more active, and the tooling has caught up: predictive ETA, congestion forecasting and scenario modelling built directly on the same AIS feed the world watched that week. That is the heart of our [voyage and port optimization](/solutions#voyage) work — turning the live picture into the decision it should have always driven. The ship was refloated in six days. The lesson has lasted far longer: the ocean that carries the world economy is legible now, in real time, to anyone with the analytics to read it. --- *Sources: Suez Canal Authority statements; Lloyd's List Intelligence blockage cost estimates; contemporaneous reporting on the grounding and salvage (23–29 March 2021). Figures are widely reported estimates.* --- ### Rerouting the world: AIS in the Red Sea crisis (Insight) URL: https://www.marineaware.com/insights/red-sea-houthi-ais Date: 2024-02-15 When Houthi attacks closed the Red Sea to much of global shipping from late 2023, vessels rewrote their AIS broadcasts to plead neutrality and carriers diverted around Africa. It was maritime data as a matter of life, cargo and geopolitics. Late in 2023, a navigation-safety system designed to stop ships colliding became something its inventors never imagined: a channel through which crews broadcast pleas for their lives. The Red Sea crisis is the starkest example yet of AIS as an instrument of geopolitics. ## What happened On **19 November 2023**, Houthi forces seized the car carrier **Galaxy Leader** in the southern Red Sea by helicopter, taking its crew hostage. From mid-December, they began a sustained campaign of drone and missile attacks on commercial shipping transiting the **Bab-el-Mandeb** strait and the southern Red Sea, framed as solidarity with Gaza and aimed initially at Israel-linked vessels, then more broadly at US- and UK-linked ships. The Bab-el-Mandeb is the southern gateway to the Suez Canal — the other end of the same Asia-Europe artery the [Ever Given](/insights/ever-given-suez-ais) had blocked two years earlier. Only this time the threat was not a grounded hull but a war zone, and it did not clear in six days. By early 2024, the world's largest container carriers — **Maersk, MSC, Hapag-Lloyd, CMA CGM** — had suspended Red Sea transits and were routing around the **Cape of Good Hope** instead, adding roughly **10-14 days and ~3,500 nautical miles** to each Asia-Europe voyage. Container tonnage through Suez fell by about **two-thirds**. It was one of the largest voluntary reroutings of global trade in living memory, and it reshaped freight rates, schedules and emissions across the industry. ## AIS as a plea The most haunting detail of the crisis lives in the AIS data. AIS is a **public broadcast**. Anything a ship puts in its identity and voyage fields — name, destination, status — is readable by anyone with a receiver, including the people deciding which ship to fire on. Crews understood this, and they used it. Across the Red Sea, vessels rewrote their AIS destination fields into messages aimed at attackers: - **"ALL CHINESE CREW"** and **"ALL CREW MUSLIM"** - **"NO CONTACT WITH ISRAEL"** / **"NO ISRAEL LINK"** - **"ARMED GUARDS ON BOARD"** A system built to prevent collisions had become an improvised channel for broadcasting neutrality under threat. Nothing illustrates more sharply that AIS is not a neutral technical feed but a live, adversarial information space — one where what you broadcast, and what you hide, has consequences. Because the broadcast cuts both ways, other vessels did the opposite: **switching off or spoofing AIS** to avoid being identified and linked to a targeted flag or owner. The region also saw significant **GPS jamming and spoofing**, corrupting the positions feeding transponders. For anyone reading the data, the Red Sea became a place where AIS could not be taken at face value — every track had to be weighed against plausibility and corroborated. ## Reading the reroute While crews used AIS to plead, analysts used it to measure. The mass diversion around Africa was legible in near-real time through vessel tracking: the thinning of transits at the Bab-el-Mandeb, the swelling of traffic down the West African coast and around the Cape, the lengthening voyages and the knock-on port timing across Northern Europe and the Mediterranean. This is AIS doing what it does best — turning a diffuse, global reaction into a measurable signal. The reroute was not announced in one place; it emerged, ship by ship, in the tracks. Firms that could read that signal early could anticipate the freight-rate spike, the schedule slippage and the equipment shortages before they were priced in. ## Why it matters for maritime analytics The Red Sea crisis sharpened three truths that sit at the centre of how we work: - **AIS is adversarial.** In a conflict zone, AIS is manipulated by design — by crews pleading neutrality, by ships hiding, by jammers on shore. Analysis has to assume the data is being gamed and lean on corroboration (satellite, behavioural plausibility) rather than trusting a single broadcast. It is the same discipline that underpins [dark-vessel detection](/insights/detecting-dark-vessels-ais-off). - **Chokepoints are the story.** Suez in 2021, the Bab-el-Mandeb in 2023-24, the Baltic in [2024](/insights/baltic-cables-ais) — the world's trade and its vulnerabilities concentrate at a few narrows. Watching them is not niche; it is macro. - **The reroute is a leading indicator.** The first evidence of the crisis's economic impact was in the tracks, days before it reached rate sheets and earnings calls. Reading maritime data early is an edge for [traders](/solutions#flows), [insurers](/solutions#sanctions) and [supply-chain planners](/solutions#voyage) alike. The Galaxy Leader's crew were held for more than a year. The reroute around Africa persisted far longer than anyone first expected. And the enduring image of the crisis is a line of text in a data field meant for a port name, broadcasting to the horizon that the people aboard meant no one any harm. --- *Sources: contemporaneous reporting on the Galaxy Leader seizure (19 Nov 2023) and subsequent Red Sea attacks; carrier diversion announcements (Dec 2023-); Suez/Bab-el-Mandeb transit data. Reroute figures are widely reported estimates.* --- ### The 75% that vanished: what satellites found when they stopped trusting AIS (Insight) URL: https://www.marineaware.com/insights/dark-fishing-nature-study Date: 2024-01-30 In January 2024 a landmark study mapped two petabytes of satellite radar and found that roughly three-quarters of the world's industrial fishing vessels never appear in public tracking. It is the single clearest proof that AIS alone is not enough. Every so often a single result reorganises how a field sees itself. For maritime analytics, that result arrived in **January 2024**, in the pages of *Nature*: a map showing that most of the industrial activity on the ocean had been invisible the whole time. ## The finding A team led by **Global Fishing Watch**, with academic partners, processed roughly **two petabytes of satellite imagery** collected between 2017 and 2021 and applied deep learning to it. The headline is stark: **around 72-76% of the world's industrial fishing vessels do not appear in public tracking**, and **more than a quarter of transport and energy vessel activity** is likewise absent from public AIS. Roughly three-quarters. The system the world had quietly come to treat as its picture of maritime activity was, for industrial fishing, showing about a quarter of the truth. The study also mapped what had been hidden: concentrations of untracked fishing off parts of Asia and Africa, and the rapid offshore build-out of energy infrastructure, including wind. It is the most complete map of human activity at sea ever assembled — and it exists only because the researchers stopped assuming AIS told the whole story. ## How you find a ship that isn't broadcasting The method is the point. AIS is a **cooperative** sensor: it works only when a vessel chooses to broadcast. To see the vessels that do not, the study leaned on a **non-cooperative** one — **synthetic aperture radar (SAR)** from the Copernicus **Sentinel-1** satellites. SAR is active radar: it supplies its own illumination, so it images the sea surface in any weather, day or night, and it detects a steel hull whether or not that hull is broadcasting anything. Cross-reference the radar detections against AIS, and the gap between them *is* the dark fleet: every vessel that shows up in the imagery but not in the transponder data. Deep-learning models did the heavy lifting at scale — detecting vessels in the radar scenes, classifying them, and distinguishing fishing from transport and energy activity across five years of global coverage. This is precisely the **AIS-gap-to-SAR fusion** we describe in [how to detect a vessel that has turned off its AIS](/insights/detecting-dark-vessels-ais-off), executed at planetary scale. ## Why it is the canonical result We cite this study more than any other, because it settles an argument. Anyone selling maritime intelligence has to answer one question: *is your picture the real one, or just the cooperative one?* Before 2024, that was a debate. After it, it is a measured fact — **AIS alone misses the majority of industrial fishing vessels** — and any analysis, any enforcement regime, any sustainability claim built on AIS by itself is quietly working from a minority of the data. The implications run straight through everything else on this site: - For **enforcement and fisheries** agencies, it means IUU fishing — an estimated tens of billions of dollars a year in losses — is mostly happening out of sight of the primary tool used to police it. Closing that gap is the core of [maritime domain awareness](/solutions#mda). - For **sanctions and risk**, it confirms that a determined vessel *can* stay dark, and that catching it requires fusion, not faith in the transponder — the same logic behind [shadow-fleet detection](/insights/shadow-fleet-ais-spoofing). - For **anyone modelling the ocean economy**, it means the baseline was wrong, and the corrected baseline is only visible through fused sensing. ## The takeaway The study's lasting contribution is not a number, though the number is memorable. It is a discipline: **treat AIS as one input, not the truth.** The ocean is far busier than the transponders admit, and the busy parts are disproportionately the ones that do not want to be seen. Reading them requires fusing cooperative and non-cooperative data — which is, in one sentence, the whole reason an [AI-native, fusion-first approach](/data-and-methods) exists. Three-quarters of the fishing fleet was hiding in plain sight. It took a change of instrument to notice. --- *Sources: Paolo et al., "Satellite mapping reveals extensive industrial activity at sea," Nature (January 2024); Global Fishing Watch press materials and SAR detection releases (Jan 2024); Copernicus Sentinel-1 documentation. Figures as reported by the study.* ---