AIS correlation and dark-ship detection
Cross-referencing SAR-detected vessel positions against spaceborne AIS records exposes ships that have gone deliberately dark. This page explains the physics, the method, and the honest limits of each data layer.
Sensors
- Sentinel-1 SAR (C-band, ESA): 5 m × 20 m resolution in Interferometric Wide Swath mode, 250 km swath, ~6-day revisit at mid-latitudes (improving to ~3 days with both satellites active). Detects vessels by their radar cross-section regardless of transponder state. Minimum detectable vessel length is roughly 15–20 m in calm sea states; clutter from high sea states raises that floor.
- ICEYE SAR (X-band, commercial): Sub-1 m spotlight resolution available on tasking, with revisit on a specific location achievable within hours using the full constellation. X-band penetrates rain better than optical but is more sensitive to sea clutter than C-band at equivalent incidence angles. Useful for rapid re-imaging of a dark-ship candidate identified in a Sentinel-1 pass.
- Spire Global spaceborne AIS: Over 100 LEMUR nanosatellites carrying VHF AIS receivers. Global AIS message collection with latency typically under 20 minutes for high-traffic areas. Coverage degrades in very dense shipping lanes where message collision rates reduce decode probability; Spire publishes this as a known limitation.
- exactEarth spaceborne AIS: Dedicated S-AIS constellation with dynamic message slots designed to reduce collision rates in congested waters. Provides historical AIS archive going back to 2009, useful for building vessel behavioural baselines and identifying anomalous gap patterns over time.
Why a ship's radar cross-section is not optional
The Automatic Identification System was designed for collision avoidance, not accountability. A Class A AIS transponder broadcasts a vessel's MMSI, position, course and speed on VHF channels 87B and 88B at intervals as short as two seconds when underway. Switching it off takes one button press. What cannot be switched off is the physical fact that a steel hull, superstructure and cargo geometry scatter microwave energy back toward an imaging radar in a pattern that is highly distinctive and largely independent of what the crew chooses to broadcast.
Sentinel-1 operates at C-band (5.405 GHz). A large tanker or bulk carrier presents a radar cross-section that can exceed 40 dBsm, producing a bright, elongated return that stands out sharply against the ocean background. The ship's heading can be inferred from the orientation of that return and from the wake signature it leaves in the backscatter field. None of this requires the vessel's cooperation.
The correlation method: matching what the radar sees to what AIS reports
The core analytical step is straightforward in concept and fiddly in execution. A SAR image is acquired over an ocean tile. An automated constant false alarm rate (CFAR) detector identifies candidate vessel returns above the clutter background. Each detection is assigned a centroid position, an estimated length derived from the bright-return extent, and a heading estimate. Separately, spaceborne AIS records are pulled for the same tile and a time window bracketing the SAR acquisition, typically plus or minus 30 minutes to account for interpolation uncertainty in vessel tracks.
Detections that can be spatially matched to an AIS-reporting vessel within a position tolerance (usually 500 m to 1 km, depending on AIS latency and vessel speed) are marked as correlated. What remains unmatched is the dark fraction. A vessel 200 m long moving at 12 knots travels roughly 370 m in the 30-second gap between AIS broadcasts, so the matching tolerance must account for that drift. In practice, the unmatched fraction on a busy ocean tile during normal commercial operations is low, typically a few per cent. A tile where 15–20 per cent of SAR detections have no AIS counterpart warrants attention.
Spoofing signatures and the difference between dark and lying
Going dark means broadcasting nothing. Spoofing means broadcasting a false position. These are different problems with different detection signatures, and conflating them is a common analytical error. A spoofing vessel continues to appear in AIS records but its reported position does not match the SAR detection. The mismatch distance is the tell: a genuine GPS fix error is typically under 50 m; a spoofed position displaced by tens of kilometres is not navigational noise.
Spoofing detection via SAR correlation is therefore a byproduct of the same correlation pipeline. Any AIS record that claims a position more than a few kilometres from the SAR-detected return, while the vessel geometry and length are consistent, is a candidate spoofing event. The caveat is that SAR position accuracy itself has limits. Sentinel-1 IW mode geolocation accuracy is documented at roughly 5–10 m for well-characterised scenes, but local ionospheric and tropospheric delays can introduce additional error. Claiming sub-100 m spoof detection is not credible with Sentinel-1 alone; ICEYE spotlight imagery tightens that considerably.
What the archive reveals that a single pass cannot
A single SAR acquisition catches a ship dark at one moment. An archive of S-AIS records going back years, combined with periodic SAR coverage, reveals behavioural patterns: vessels that consistently go dark in the same geographic box, vessels whose AIS gaps cluster around known ship-to-ship transfer corridors, or vessels that reappear with a different reported cargo after a multi-day silence. exactEarth's archive extends to 2009, which is long enough to build a meaningful behavioural baseline for most commercial vessels.
Gap analysis on AIS tracks is a distinct but complementary product. A vessel that disappears from AIS for 48 hours in mid-ocean and reappears on a different heading has almost certainly done something the owner prefers not to document. The gap itself is not proof of wrongdoing, but its location, duration and the vessel's subsequent port call create a testable hypothesis. Pairing gap timestamps with SAR coverage schedules shows whether the vessel was in a region imaged during the gap and, if so, whether it appears in the imagery.
Honest limits: what this method cannot do
SAR coverage is not continuous. Sentinel-1's ~6-day revisit at mid-latitudes means a vessel can transit an entire ocean basin between passes. Commercial tasking from ICEYE or similar providers can fill gaps, but only if an analyst knows where to look. Detecting a dark ship requires either routine area surveillance or a tip that focuses tasking resources on a specific tile.
Small vessels present a genuine detection floor. A wooden-hulled fishing boat of 10 m length has a radar cross-section that may fall below the CFAR threshold in anything above a moderate sea state. This method is most reliable for vessels above roughly 50 m in length. Additionally, the spaceborne AIS decode rate in very dense shipping lanes, the Strait of Malacca being the canonical example, is reduced by message collision. A vessel that is technically transmitting may not appear in S-AIS records simply because its messages were lost in the noise. Absence from AIS is not always absence of intent. Satellize's correlation pipeline flags unmatched detections with a sea-state confidence score and a lane-density qualifier precisely because those two factors drive most false positives.
Turning detections into actionable intelligence
The raw output of a correlation run is a list of unmatched SAR detections with position, estimated length, heading, sea-state confidence and, where available, a nearest-AIS candidate and its displacement distance. That list becomes useful when it is ranked by anomaly score, cross-referenced against vessel registries and sanctions lists, and delivered with enough context for a non-specialist analyst to decide whether to task a follow-up image.
Satellize builds this pipeline on open Sentinel-1 acquisitions and licensed S-AIS feeds, adding commercial SAR tasking where a client's area of interest justifies the cost. The analytic output is a ranked alert feed, not a raw detection dump. It is worth being direct about what that means: the system will surface genuine anomalies alongside vessels that went dark for mundane reasons such as transponder failure or crew error. Human review of the top-ranked alerts remains necessary. The physics is reliable; the intent behind any individual dark event is always an inference.
Typical figures
| SAR spatial resolution (Sentinel-1 IW mode) | 5 m × 20 m (range × azimuth) |
| SAR spatial resolution (ICEYE spotlight) | < 1 m |
| Sentinel-1 revisit (mid-latitudes, single satellite) | ~6 days; ~3 days with dual-satellite constellation |
| ICEYE tasking revisit | Hours to sub-daily for a specific location |
| Spaceborne AIS latency | Typically < 20 minutes for high-traffic areas (Spire published figure) |
| Minimum detectable vessel length (SAR, calm sea state) | ~15–20 m (Sentinel-1); smaller targets possible with ICEYE spotlight |
| AIS archive depth | exactEarth archive from 2009; Spire from ~2015 |
| Position matching tolerance (SAR vs AIS) | 500 m–1 km typical, depending on AIS interpolation window |
| Sentinel-1 geolocation accuracy | ~5–10 m (documented; ionospheric conditions add uncertainty) |
| Delivery formats | GeoJSON alert feed, GeoTIFF annotated imagery, CSV ranked detection report |
Analytics Satellize can run
| Dark-vessel fraction map | CFAR SAR detection correlated against S-AIS records; unmatched detections aggregated by ocean tile | GIS layer (GeoJSON polygon grid) with per-tile dark fraction score and vessel count |
| Ranked dark-ship alert feed | Anomaly scoring combining vessel length estimate, sea-state confidence, lane-density qualifier and distance to nearest AIS record | Daily or per-acquisition JSON alert feed, top-N candidates with supporting metadata |
| Spoofing candidate report | SAR-to-AIS position displacement analysis; detections matched to AIS records with displacement exceeding 2 km flagged as spoof candidates | PDF or structured report listing candidate vessels, displacement distance, SAR acquisition timestamp and nearest port |
| AIS gap analysis with SAR coverage overlay | Historical S-AIS track segmentation to identify gaps; Sentinel-1 acquisition schedule cross-referenced to determine whether gap period was imaged | Per-vessel timeline showing gap events, whether SAR coverage existed, and detection or absence result |
| Behavioural baseline and deviation scoring | Multi-year S-AIS archive analysis to establish typical routes, port calls and AIS-on patterns per MMSI; current behaviour scored against baseline | Vessel-level risk score updated on configurable cadence, delivered as CSV or API feed |
| Annotated SAR dark-ship imagery | CFAR detections overlaid on Sentinel-1 or ICEYE GeoTIFF with matched and unmatched detections colour-coded | GeoTIFF with vector annotation layer; suitable for direct import into QGIS or ArcGIS |
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.