Vessel wake pattern analysis for illegal trawl corridor identification in SAR
Trawling vessels leave Kelvin and turbulent wake signatures in SAR imagery that persist for tens of minutes, enabling retrospective corridor mapping even without AIS. Overlaying detected corridors against MPA and closed-area boundaries flags probable violations with documented probability limits.
Sensors
- Sentinel-1 (C-band SAR, IW and EW modes): IW mode delivers 10 m ground range resolution across a 250 km swath; EW mode covers 400 km at 25 m resolution. Revisit at mid-latitudes is 6 days per satellite, 3 days with the A/B pair. C-band (5.405 GHz) is sensitive to Bragg-scale capillary waves and wake-induced surface roughness changes, but wind speeds above roughly Beaufort 4 raise background clutter and suppress wake contrast.
- ICEYE X-band SAR constellation: X-band (9.65 GHz) offers finer capillary-wave sensitivity than C-band and spotlight modes down to approximately 0.5 m resolution, useful for resolving narrow turbulent wake cores. ICEYE's growing constellation supports sub-daily tasking over specific areas of interest, reducing the temporal gap between passage and detection.
- Capella Space X-band SAR: Spotlight mode at approximately 0.35 m resolution can resolve individual wake arms and, in calm sea states, biogenic slick boundaries associated with trawl gear. Tasking latency of hours rather than days makes it practical for near-real-time cuing when a dark vessel is already suspected.
- Umbra X-band SAR: Umbra's spotlight capability reaches approximately 0.16 m resolution in its highest mode, sufficient to distinguish vessel hull geometry alongside wake morphology. Useful for confirming vessel class when wake geometry alone is ambiguous.
What a wake actually records
A vessel moving through water generates two physically distinct signatures. The Kelvin wake is a V-shaped pattern of surface gravity waves with a half-angle of approximately 19.5 degrees, a value that is independent of vessel speed in deep water and was derived analytically by Lord Kelvin in 1887. In SAR imagery the Kelvin arms appear as bright lines because they tilt the water surface toward the radar look angle and increase local backscatter. The turbulent wake is the narrow, chaotic zone of foam and bubble plumes directly behind the hull; it appears as a dark stripe in SAR because bubble-covered water suppresses Bragg scattering.
Trawlers add a third signature. The trawl gear disturbs bottom sediment and releases biogenic material, producing a surfactant slick that damps capillary waves and appears as a dark, elongated patch behind the vessel. In calm conditions this slick can persist for 30 to 60 minutes after passage. Together, the three signatures allow analysts to reconstruct not just that a vessel was present, but the direction of travel, approximate speed from Kelvin arm geometry, and whether gear was deployed.
Detection probability and its honest limits
Wake detection in SAR is not reliable under all conditions. Published studies using Sentinel-1 data consistently report that Kelvin wake detection probability falls sharply when wind speed exceeds roughly 7 to 8 m/s (Beaufort 4 to 5). Above that threshold, background ocean clutter from wind-driven capillary waves overwhelms the subtle backscatter contrast of the wake arms. Rain cells cause additional artefacts: rain roughens the surface and can mimic or mask wake signatures. Any compliance report that omits these conditions is overstating confidence.
Turbulent wake length depends on vessel speed and sea state. At typical trawling speeds of 3 to 5 knots, the turbulent wake may extend 1 to 3 km behind the vessel in calm conditions, giving a temporal window of 20 to 40 minutes for detection. At Beaufort 5 or above, that window collapses. Biogenic slick persistence is similarly state-dependent and is essentially undetectable in choppy water. Shallow-water trawling in turbid coastal areas introduces further ambiguity because suspended sediment plumes can resemble biogenic slicks. These limits must be flagged explicitly in any output delivered to a flag state or regional fisheries management organisation.
From single wake to trawl corridor
A single SAR pass records a snapshot. Corridor reconstruction requires temporal stacking across multiple passes and, where available, fusion with AIS data to anchor the timeline. The method is straightforward in principle: each detected wake segment is assigned a heading vector and a rough timestamp derived from the satellite overpass time minus the estimated wake age. Segments are then chained spatially to produce a probable vessel track. Where AIS is present but shows a vessel stationary or absent, the wake geometry provides an independent ground truth.
The practical challenge is ambiguity. Two vessels on converging headings can produce overlapping wake patterns that are difficult to disentangle at 10 m resolution. ICEYE or Capella tasking at sub-metre resolution can resolve individual wake arms and reduce this ambiguity, but commercial tasking adds cost and requires advance notice of the area of interest. Sentinel-1's open-access archive, which extends back to 2014, is the practical starting point for historical corridor analysis before any commercial tasking is commissioned.
Spatial overlay with closed areas and MPAs
The analytical value of corridor reconstruction depends entirely on the quality of the boundary data it is tested against. Marine protected area polygons vary considerably in their legal status, coordinate precision and update frequency. Some national MPA boundaries held in regional databases differ by hundreds of metres from the legally gazetted coordinates. A wake corridor that appears to cross a boundary by 200 m may or may not represent a genuine violation depending on which polygon version is used.
Best practice is to use the most recent officially gazetted boundary, to apply a conservative buffer zone when boundary precision is uncertain, and to report the corridor position with its own spatial uncertainty derived from the SAR resolution and wake-age estimate. The output should be a probable violation flag, not a confirmed one. Confirmation requires corroborating evidence such as vessel monitoring system data, observer reports or a follow-up high-resolution SAR task. Satellize structures its corridor-overlay outputs accordingly, distinguishing between 'probable corridor inside closed area' and 'confirmed vessel inside closed area' as separate confidence tiers.
Automated detection versus analyst review
Several published methods exist for automated wake detection in SAR. The Radon transform is the most widely cited: it detects linear features at arbitrary angles and has been applied to Sentinel-1 data to extract Kelvin arm orientations at scale. Matched-filter approaches tuned to the expected Kelvin angle also perform well in low-clutter conditions. Machine learning classifiers trained on labelled SAR chips have shown precision and recall figures in the range of 70 to 90 percent in published evaluations, with performance degrading predictably as sea state rises.
Automated outputs still require analyst review before being used in a compliance context. False positives from ship wakes of non-fishing vessels, current boundaries and internal wave fronts are common enough that an unsupervised pipeline would generate unacceptable noise in an enforcement report. The practical workflow is automated detection to flag candidates, followed by analyst triage to remove obvious false positives, followed by structured reporting with confidence levels attached. This is slower than a fully automated feed but it is what a flag state or RFMO can actually act on.
What the archive enables that real-time monitoring cannot
Sentinel-1's archive back to 2014 makes it possible to characterise historical trawling pressure in a corridor over years, not just days. Repeated wake detections along the same heading in the same area, across dozens of passes, build a statistical picture of habitual fishing behaviour that is far more compelling to a regulator than a single-pass flag. This retrospective analysis is particularly useful for identifying corridors that are systematically avoided during known monitoring periods but reappear when coverage lapses, a pattern consistent with deliberate evasion.
The archive also allows baseline characterisation before a new MPA is gazetted, which is necessary for measuring whether protection has actually changed vessel behaviour. Without that baseline, post-designation monitoring has no reference point. Satellize has applied this kind of multi-year archive analysis in its Tonga crop-estimation programme for agricultural baselines and the same temporal stacking logic transfers directly to marine corridor work.
Typical figures
| Sentinel-1 IW mode resolution | 5 m range × 20 m azimuth (10 m ground range after multi-look) |
| Sentinel-1 EW mode resolution | 25 m ground range, 400 km swath |
| ICEYE spotlight resolution | Approximately 0.5 m (single-look spotlight) |
| Sentinel-1 revisit (mid-latitudes, A+B) | 3 days; 6 days per satellite |
| Commercial SAR tasking latency | Hours to sub-daily (ICEYE, Capella, Umbra) |
| Wake persistence window (calm, Beaufort ≤3) | Kelvin arms: up to 10–20 min; turbulent wake: 20–40 min; biogenic slick: 30–60 min |
| Detection probability floor (sea state) | Degrades sharply above Beaufort 4 (~7–8 m/s wind); near-zero above Beaufort 6 |
| Sentinel-1 archive depth | 2014 to present (open access) |
| Radar frequency | C-band: 5.405 GHz (Sentinel-1); X-band: ~9.6 GHz (ICEYE, Capella, Umbra) |
| Delivery formats | GeoTIFF wake-segment layers, GeoJSON corridor polygons, PDF compliance report with confidence tiers |
Analytics Satellize can run
| Wake candidate detection layer | Radon transform or matched-filter applied to calibrated SAR sigma-nought imagery to extract linear features at Kelvin angles (~19.5°) | GeoTIFF or GeoJSON of detected wake centrelines with heading, length and confidence score per segment |
| Vessel track reconstruction | Temporal chaining of wake segments across multi-pass Sentinel-1 stack, fused with AIS where available, to produce probable vessel trajectories | GeoJSON track layer with timestamp range, estimated speed, and AIS-reconciliation flag |
| Biogenic slick mapping | Normalised difference of SAR backscatter in wake zone versus ambient ocean; thresholded against wind-speed ancillary data to suppress false positives | Polygon layer of probable trawl-associated slicks with sea-state validity flag |
| MPA corridor overlap report | Spatial intersection of reconstructed corridors with official MPA and closed-area boundary polygons; uncertainty buffer applied based on SAR resolution and wake-age estimate | PDF compliance report distinguishing 'probable corridor inside closed area' from 'confirmed vessel inside closed area', with confidence tier and caveats |
| Historical trawling pressure index | Multi-year Sentinel-1 archive stacking; wake-detection frequency per grid cell normalised by valid-observation count (excluding high-sea-state passes) | Gridded GeoTIFF of trawling pressure intensity, suitable for MPA baseline characterisation or post-designation change detection |
| Evasion pattern flag | Comparison of trawling pressure index across known monitoring periods versus non-monitored periods to identify statistically significant corridor avoidance correlated with patrol schedules | Analytical summary report with time-series charts and flagged corridor segments |
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.