Dark vessel detection by fusing AIS gaps with RF and SAR
When a vessel goes dark on AIS, its radar cross-section and incidental RF emissions remain. Fusing spaceborne AIS gaps with SAR imagery and HawkEye 360 RF fingerprints produces detections that no single sensor can achieve alone.
Sensors
- Spire Global spaceborne AIS: A constellation of over 100 LEO nanosatellites receiving AIS VHF messages (161.975 MHz and 162.025 MHz). Global revisit of any ocean point is typically every 20–40 minutes. Gaps or absence of expected MMSI messages flag potential transponder disablement. Message latency to ground is typically under 30 minutes.
- HawkEye 360 RF geolocation: Clusters of three formation-flying satellites detect and geolocate RF emissions across VHF, UHF and X-band ranges using time-difference-of-arrival and frequency-difference-of-arrival. Positional accuracy is typically 1–3 km CEP. Can detect radar altimeters, fishing echo sounders and communication uplinks that a vessel cannot easily suppress, even with AIS off.
- Sentinel-1 SAR (ESA): C-band (5.405 GHz) synthetic aperture radar with 5×20 m resolution in Interferometric Wide Swath mode and a 250 km swath. Revisit at mid-latitudes is roughly every 6 days per satellite; the two-satellite constellation halves that to around 3 days. Detects vessel radar cross-sections down to approximately 50–100 m² in moderate sea states, corresponding broadly to vessels over 30–40 m length, though sea-state clutter degrades this floor significantly.
- ICEYE SAR: X-band (9.65 GHz) commercial SAR offering Stripmap imagery at approximately 3 m resolution and Spot mode at around 1 m. Tasking latency can be under 24 hours and revisit is flexible through commercial scheduling. X-band is more sensitive to smaller metallic targets than C-band, improving detection of smaller fishing vessels, but is also more affected by rain attenuation.
Why AIS alone is an unreliable witness
AIS was designed for collision avoidance, not surveillance. The protocol is unauthenticated: any vessel can broadcast a false MMSI, a spoofed position, or simply nothing at all. Switching off the transponder takes seconds and leaves no trace in the AIS record beyond an absence. For maritime domain awareness, that absence is the signal, but only if you have a prior expectation of presence.
Spire Global's spaceborne AIS constellation collects messages globally, giving analysts a near-continuous picture of declared vessel positions. When a known vessel's MMSI goes silent in open ocean, or when a vessel appears in a SAR image with no corresponding AIS echo within a 20–40 minute window, that discrepancy is the starting point for a dark-vessel investigation. The gap alone is not proof of wrongdoing: transponder failures and shaded coverage zones are real. The gap is a hypothesis that the other sensors must test.
What a floating hull gives away even in silence
A steel vessel displacing thousands of tonnes has a radar cross-section that Sentinel-1 or ICEYE will register regardless of what the crew does with the AIS switch. In SAR imagery, ships appear as bright point targets against a darker sea background, and their wakes can persist for kilometres. The detection challenge is not whether the ship is visible, but whether it can be unambiguously separated from sea clutter, sidelobes and other artefacts.
Sea state matters considerably. In Beaufort 5–6 conditions, wave backscatter in C-band can mask vessels below roughly 500–1000 m² radar cross-section, which excludes a meaningful fraction of the fishing fleet. ICEYE's X-band is somewhat better at resolving smaller targets in moderate clutter, though it covers a narrower swath per pass and must be tasked deliberately rather than collected systematically.
HawkEye 360 adds a dimension that SAR cannot provide: the RF environment around the vessel. A fishing boat running its echo sounder, a tanker communicating via satellite phone, or a vessel using a navigation radar all emit signals that HawkEye's formation clusters can geolocate to 1–3 km accuracy. That is not precise enough to identify a specific ship, but it is precise enough to cue a SAR collect or to confirm that something active exists in a region where AIS shows nothing.
The fusion logic: turning three imperfect sensors into one defensible detection
No single layer is sufficient. AIS provides identity but can be falsified. SAR provides position and approximate size but not identity. RF geolocation provides activity evidence but not precise position. The analytical value comes from correlating all three in time and space.
A practical workflow runs as follows. AIS gap analysis identifies ocean regions where expected vessels have gone silent, or where vessel density appears anomalously low given historical patterns. HawkEye RF passes over those regions flag active emitters. SAR imagery, either from Sentinel-1 systematic collection or a tasked ICEYE pass, then resolves individual targets within the RF uncertainty ellipse. Where a SAR target exists, an RF emitter is present, and no AIS message has been received, the confidence in a genuine dark vessel is substantially higher than any single indicator would support.
Temporal alignment is critical. Sentinel-1 and HawkEye passes are asynchronous, and a vessel moving at 12 knots travels roughly 22 km in an hour. Analysts must propagate positions forward or backward in time using assumed or estimated speed and heading before matching across layers. Errors in that propagation are a real source of false associations.
Ambiguities the honest analyst must acknowledge
SAR incidence angle affects how a vessel's hull is illuminated. A vessel heading directly toward or away from the satellite's look direction presents a different cross-section than one broadside to it. Long, narrow vessels can appear as point targets at coarser resolutions, making length estimation unreliable. At Sentinel-1's 5×20 m resolution, distinguishing a 60 m fishing vessel from an 80 m coastal freighter is genuinely difficult.
Vessel heading also affects wake visibility. A vessel moving parallel to the SAR azimuth direction may show little or no Kelvin wake, removing one of the key secondary cues. And in high sea states, the wake signal is buried entirely.
RF geolocation has its own ambiguities. Not every vessel runs every emitter at all times. A vessel that has turned off its AIS may also have powered down non-essential electronics. Conversely, a legitimate vessel with a malfunctioning AIS transponder will appear identical in the RF and SAR record to a deliberately dark one. Corroborating evidence, such as prior voyage history, port state records, or flag-state data, is needed before any detection becomes an actionable finding rather than a lead.
From detection to decision: what the analytic products look like
The output of a dark-vessel fusion workflow is typically a ranked list of candidate detections, each carrying a confidence score derived from how many independent layers agree, the quality of the temporal alignment, and the sea-state conditions at the time of the SAR collect. A detection supported by an AIS gap, an RF emitter, and a SAR target within a 5 km radius and a 2-hour window is categorically more credible than one resting on SAR alone.
Satellize structures these outputs as GIS-compatible alert feeds, allowing clients to ingest detections directly into maritime operations centres or compliance workflows. The Overhead column has covered the underlying physics and policy context of AIS manipulation in open-ocean settings for readers who want the analytical background before commissioning a monitoring programme.
The practical use cases span sanctions enforcement, illegal fishing detection, environmental compliance monitoring and insurance underwriting. Each has a different tolerance for false positives. A sanctions analyst may accept a 20% false-positive rate in exchange for high recall; an insurer writing hull cover needs near-certainty before acting. Configuring the confidence thresholds and the sensor tasking cadence to match that tolerance is the actual design problem, and it has no universal answer.
Typical figures
| AIS revisit (Spire Global) | 20–40 minutes globally; message latency to analyst typically under 30 minutes |
| RF geolocation accuracy (HawkEye 360) | 1–3 km CEP; frequency range covers VHF through X-band |
| SAR spatial resolution (Sentinel-1 IW) | 5 m range × 20 m azimuth; 250 km swath |
| SAR spatial resolution (ICEYE Spot) | Approximately 1 m; Stripmap approximately 3 m |
| Sentinel-1 revisit at mid-latitudes | Approximately 3 days (two-satellite constellation); 6 days per satellite |
| Minimum detectable vessel size (Sentinel-1, calm sea) | Approximately 30–40 m length; degrades significantly in Beaufort 5+ conditions |
| SAR archive depth (Sentinel-1) | From April 2014 (Sentinel-1A launch); systematic global ocean coverage |
| Temporal alignment tolerance for fusion | Typically ±1–2 hours; position propagation required for moving targets |
| Delivery formats | GeoJSON alert feed, GIS layer (GeoPackage or Shapefile), PDF investigation report |
Analytics Satellize can run
| AIS gap map | Temporal dropout analysis on Spire AIS message archive; comparison against historical vessel density baselines | Daily or weekly GeoJSON layer of ocean regions with anomalous AIS silence, ranked by deviation from baseline |
| SAR vessel detection layer | Constant false alarm rate (CFAR) ship detection on Sentinel-1 IW or ICEYE imagery; clutter filtering by polarisation ratio | GIS point layer of detected targets with estimated length, heading and confidence score per scene |
| RF emitter correlation report | Spatial and temporal intersection of HawkEye 360 geolocation ellipses with AIS gap regions | Tabular report of RF detections with no matching AIS MMSI within the geolocation uncertainty radius |
| Fused dark-vessel candidate list | Multi-layer evidence scoring: AIS gap + RF presence + SAR target co-location within defined space-time tolerance | Ranked candidate list with per-detection confidence tier (high / medium / low), delivered as GeoJSON or PDF |
| Voyage reconstruction for flagged vessels | Back-interpolation of SAR positions combined with last known AIS fix and port-state entry records | Vessel track report covering the dark period, with positional uncertainty bounds stated explicitly |
| Persistent monitoring alert feed | Automated pipeline combining systematic Sentinel-1 ingestion, Spire AIS stream and scheduled HawkEye tasking over a defined area of interest | Near-real-time alert feed (latency typically 2–6 hours post-SAR acquisition) delivered via API or email digest |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.