Ship-to-Ship Transfer Location Pattern Analysis
Recurring ship-to-ship transfer zones betray themselves through spatial clustering, even when individual events are missed. Density mapping of SAR detections and AIS gap logs turns scattered incidents into actionable patrol polygons.
Sensors
- Sentinel-1 SAR (C-band, ESA): Interferometric Wide Swath mode delivers 10-metre ground resolution across a 250 km swath. Revisit at the equator is roughly 6 days per satellite; the two-satellite constellation halves that to about 3 days at mid-latitudes, shorter at higher latitudes. Detects vessel pairs regardless of cloud or darkness, though vessels below roughly 30 metres in length approach the noise floor in IW mode.
- Spire spaceborne AIS: A constellation of over 100 LEO cubesats receiving Class-A AIS messages globally. Gap logs, where a vessel's MMSI disappears and reappears with a positional discontinuity, are the primary AIS-derived indicator of deliberate transponder suppression during a transfer. Latency to gap detection is typically under 2 hours in operational feeds.
- ICEYE SAR (X-band): Spotlight mode achieves sub-1-metre resolution, sufficient to resolve fender rigging and hose connections that confirm physical contact. Tasked on demand with revisit to a specific point possible within hours using the full ICEYE constellation. Higher cost per acquisition than Sentinel-1 makes it appropriate for confirmation rather than area survey.
- Capella Space SAR (X-band): Spotlight and sliding-spotlight modes reach 0.5-metre resolution. Like ICEYE, useful for evidentiary-quality imagery once a hotspot polygon is already defined. Capella's 25-minute revisit target (across the full constellation) makes it viable for near-real-time monitoring of a known zone.
Why location patterns matter more than individual events
A single ship-to-ship transfer is an event. A cluster of transfers returning to the same 20-nautical-mile patch of sea, month after month, is a policy. The distinction matters because individual events are often missed: cloud obscures optical sensors, Sentinel-1 revisit leaves gaps of days, and AIS suppression means no transponder record exists to trigger an alert. Pattern analysis is resilient to those gaps in a way that event-by-event monitoring is not.
The underlying logic is geographic preference. Vessels conducting transfers for sanctions evasion or undocumented cargo movement choose locations deliberately. Shallow radar cross-section environments such as sea clutter near island chains can complicate shore-based radar. Jurisdictional seams between overlapping EEZs or high-seas corridors reduce the probability of interception. Certain anchorage depths suit specific vessel draught combinations. These constraints are stable across time, which means the locations recur even as the individual vessels rotate.
Building the detection layer: SAR meets AIS gaps
The input dataset is a fusion of two independent streams. Sentinel-1 IW-mode acquisitions, archived continuously since 2014 for most ocean regions, are processed through a CFAR (constant false-alarm rate) detector to extract vessel centroids and, where resolution permits, estimated length. Pairs of detections within roughly 50 metres of each other and aligned bow-to-stern or beam-to-beam are flagged as candidate STS events. At 10-metre resolution, physical contact is not always unambiguous; the confirmation threshold typically requires the pair to persist across two sequential passes.
The AIS gap layer adds a behavioural dimension. A gap is defined as an MMSI that broadcasts in open ocean, goes silent for more than a defined threshold (commonly 4 to 12 hours depending on the analyst's tolerance for false positives), and reappears with a position inconsistent with the vessel's last known speed. When a SAR-detected vessel pair spatially coincides with one or two AIS gaps in the same time window, confidence in a deliberate transfer rises substantially. Neither stream alone is sufficient. SAR without AIS cannot distinguish a breakdown from a transfer. AIS gaps without SAR cannot confirm physical proximity.
From point detections to hotspot polygons
Once a multi-year catalogue of candidate STS events is assembled, kernel density estimation converts the point cloud into a continuous probability surface. The choice of bandwidth is consequential: too narrow and the map fragments into noise; too wide and operationally distinct zones merge. A bandwidth of 15 to 25 nautical miles is common in published maritime surveillance literature, though analysts should test sensitivity. The output is a set of contour polygons ranked by event density, clipped to international waters or the specific jurisdictional zones of interest.
These polygons are the operationally useful product. A coast guard or naval authority can define patrol boxes around the top-density contours rather than attempting to cover open ocean. The polygons also serve as tasking cues for commercial SAR: instead of ordering broad-area surveillance, an operator tasks ICEYE or Capella spotlight acquisitions over the known hotspot during the time windows when transfers historically concentrate, which published studies on Gulf of Guinea and Strait of Malacca activity suggest are often correlated with low-wind, low-swell conditions that ease hose handling.
Honest limits deserve stating plainly. A hotspot polygon shows where transfers have occurred, not where they will occur next. If enforcement pressure displaces activity, the historical polygon becomes stale within weeks. The map requires continuous updating, not one-time production.
Jurisdictional geometry as an analytic variable
One underused dimension of this analysis is the deliberate mapping of hotspot polygons against maritime boundary geometry. Transfers that cluster just outside an EEZ boundary, or in a corridor where two coastal states have overlapping or disputed claims, are almost certainly not coincidental. The spatial relationship between the density peak and the nearest jurisdictional line is itself an intelligence product: it tells an analyst which authority would need to act, and reveals whether the operating model depends on a specific legal seam.
This is straightforward to compute. UNCLOS-derived EEZ boundaries are publicly available through VLIZ (the Flanders Marine Institute) and can be overlaid directly on the density surface. The distance from each hotspot centroid to the nearest boundary, combined with the historical spread of the event cluster, gives a measure of how precisely the operators are navigating the legal geography. A cluster whose 90th-percentile contour sits entirely beyond 12 nautical miles from any coast, but within 5 nautical miles of an EEZ boundary, is a different problem from one centred in open high seas.
Temporal structure: when the hotspots are active
Density maps collapse time, which is useful for patrol planning but hides operational rhythm. Layering a time-of-year breakdown onto the spatial analysis often reveals seasonal concentration driven by weather windows, commodity shipping cycles or enforcement calendars. In the Gulf of Guinea, for instance, publicly reported STS activity linked to crude oil has shown concentration in specific quarters aligned with tanker scheduling patterns documented in shipping press.
A second temporal dimension is the gap between AIS silence onset and SAR confirmation. Where the archive is dense enough, this gap distribution tells an analyst how long transfers typically take in a given zone, which constrains the revisit requirement for effective monitoring. If the median transfer duration is 8 hours and Sentinel-1 revisit is 3 days, the probability of catching any single event is low. That calculation makes the case for supplementing open-data SAR with tasked commercial acquisitions over confirmed hotspots, and it is the kind of quantified argument that Satellize builds into programme design for clients who need to justify sensor expenditure.
What the method cannot do
SAR-based STS detection has a firm size floor. Vessels shorter than roughly 30 metres in Sentinel-1 IW mode are unreliable detections; small product tankers or fishing vessels used as intermediaries may be missed entirely. X-band commercial SAR pushes that floor lower, but not to zero: sea state, incidence angle and vessel orientation all affect detectability.
Cloud is not a problem for SAR, but latency is. Sentinel-1 data is typically available within 1 to 3 hours of acquisition through the Copernicus Data Space, but the processing pipeline from raw acquisition to a calibrated, geocoded, vessel-detected output adds time. Operational maritime intelligence rarely tolerates latency beyond 2 to 4 hours for a live event; the hotspot polygon approach sidesteps this by treating the pattern as the product rather than the individual detection. The polygon is updated weekly or monthly, not in real time, and patrol tasking is planned accordingly. Real-time interception remains a different problem requiring different architecture.
Typical figures
| Primary SAR resolution (Sentinel-1 IW) | 10 m ground range, 250 km swath |
| Confirmation SAR resolution (ICEYE / Capella spotlight) | Sub-1 m to 0.5 m |
| Sentinel-1 revisit (two-satellite) | ~3 days at mid-latitudes; ~1 day above 60° N/S |
| Commercial SAR tasked revisit | Within hours for a named point (ICEYE / Capella full constellation) |
| AIS gap detection latency (Spire) | Typically under 2 hours in operational feed |
| Minimum detectable vessel length (Sentinel-1 IW) | Approximately 30 m; smaller vessels approach noise floor |
| Sentinel-1 archive depth | From 2014 (Sentinel-1A launch) for most ocean regions |
| Hotspot polygon update cadence (typical) | Weekly to monthly; not a real-time product |
| Delivery formats | GeoJSON / Shapefile polygons, GeoTIFF density rasters, PDF patrol-tasking briefs |
| Sensor frequency (Sentinel-1) | C-band, 5.405 GHz |
Analytics Satellize can run
| Multi-year STS candidate event catalogue | CFAR vessel detection on Sentinel-1 IW SLC/GRD, vessel-pair proximity filter, AIS gap temporal co-registration | Timestamped point GeoJSON with vessel-pair centroids, estimated lengths and AIS gap flags |
| Hotspot density polygons | Kernel density estimation (bandwidth 15-25 NM) over event catalogue; contour extraction at analyst-defined percentile thresholds | GeoJSON / Shapefile polygon set ranked by event density, updated monthly |
| Jurisdictional proximity analysis | Spatial join of hotspot centroids against UNCLOS EEZ boundaries; distance and overlap statistics | Table and map layer showing hotspot-to-boundary distances and responsible coastal-state attribution |
| Seasonal and temporal activity profile | Time-series decomposition of event catalogue by month, quarter and time-of-day; correlation with publicly available weather and swell records | Heatmap chart and summary report identifying peak-activity windows for patrol scheduling |
| Commercial SAR tasking cue list | Ranking of hotspot polygons by recent activity rate and estimated detection probability given current Sentinel-1 revisit geometry | Prioritised polygon list with recommended ICEYE or Capella tasking windows, formatted for operator upload |
| Confirmed STS evidentiary package | Sub-metre spotlight SAR acquisition over flagged event, co-registered with AIS gap record and vessel identity cross-reference | PDF report with annotated imagery, timeline and vessel MMSI / IMO data for regulatory or legal use |
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.