Sensors
- Sentinel-1 SAR (ESA): C-band (5.405 GHz) SAR; IW mode delivers 10 m ground range resolution across a 250 km swath. Repeat pass over most ocean areas every 6–12 days with a single satellite; the two-satellite constellation historically achieved 6-day revisit before Sentinel-1B failure in 2021 reduced coverage. Free and open archive from 2014.
- ICEYE X-band SAR constellation: X-band (9.65 GHz) SAR; spotlight mode achieves approximately 0.5 m resolution, strip mode 3 m at a 30 km swath. Commercial tasking enables same-day or next-day revisit over a specific area of interest. Particularly sensitive to small metal targets due to shorter wavelength.
- Capella Space X-band SAR: X-band SAR constellation; spotlight mode resolution down to approximately 0.5 m. Offers rapid tasking response, sometimes under two hours for priority requests. Well-suited to vessel length estimation and hull-type discrimination at fine resolution.
- Spire satellite AIS: Space-based AIS receiver constellation of over 100 satellites; global message latency typically under 30 minutes for most ocean areas. Provides the broadcast AIS track archive against which SAR detections are correlated to identify non-reporting vessels.
What a metal hull gives away
Synthetic aperture radar illuminates the ocean surface with microwave pulses and records what bounces back. A steel or aluminium fishing vessel is a near-perfect retroreflector. Corner-reflector geometry formed by the hull, superstructure and deck returns a signal orders of magnitude stronger than the surrounding sea surface. At C-band (Sentinel-1's 5.405 GHz) a vessel of 20 metres or more is reliably detectable in moderate sea states. X-band systems from ICEYE and Capella Space, operating around 9.65 GHz, are more sensitive to smaller targets because the shorter wavelength interacts more strongly with compact metal structures, though X-band also suffers higher attenuation in heavy rain.
L-band SAR (around 1.2 GHz, used by ALOS-2 and the forthcoming NISAR mission) penetrates cloud and rain more effectively and produces strong returns from large vessels, but its coarser typical resolution makes it less useful for discriminating small fishing boats from background clutter. The practical consequence: for a coastal patrol zone where vessels of interest may be 15–40 metres in length, X-band commercial SAR is the sharper instrument, while Sentinel-1 provides the broad-area, low-cost sweep that flags anomalies worth tasking.
The correlation step that makes detection meaningful
Detecting a bright point target in a SAR image is straightforward. Knowing whether that target is a legitimate vessel operating normally is not. The analytic value lies in the correlation layer. A satellite AIS feed, such as those aggregated by Spire or exactEarth, provides a time-stamped record of every vessel broadcasting its identity and position. When a SAR scene is acquired, detected radar targets are matched against the AIS positions interpolated to the scene acquisition time. Targets with a corresponding AIS broadcast are flagged as reporting vessels. Targets without one are dark vessels, candidates for further investigation.
The matching is not trivial. AIS position interpolation introduces error when message cadence is low, particularly for vessels at anchor or moving slowly. A positional tolerance of roughly 500 metres to 1 kilometre is typical, depending on AIS latency and SAR geolocation accuracy. Vessels that have recently switched off their transponders may have drifted beyond their last known position. The Global Fishing Watch programme, which publishes its methodology openly, has demonstrated that this correlation approach can be applied at ocean-basin scale using Sentinel-1 imagery, identifying thousands of candidate dark vessels per month across monitored regions.
False positives are the operational problem nobody advertises
Any honest account of this method must address the false-positive rate. Ocean clutter produces bright returns, particularly from breaking waves in high sea states. Offshore platforms, buoys, fish aggregating devices and aquaculture infrastructure all appear as persistent bright targets. Icebergs and sea ice are strong C-band reflectors. Rain cells create distributed high-backscatter regions that can mask or mimic vessel returns.
Constant false alarm rate (CFAR) detection algorithms adapt the detection threshold to local clutter statistics, which reduces but does not eliminate spurious detections. A well-tuned CFAR detector on Sentinel-1 IW mode imagery achieves a vessel detection probability exceeding 90 percent for targets longer than 50 metres in sea states below Beaufort 5, but detection probability drops sharply for smaller vessels and in rough conditions. X-band imagery at sub-metre resolution allows analysts to examine the shape and shadow of a return, which helps distinguish a vessel from a buoy or platform. Cross-checking against charts of known fixed infrastructure removes another class of false positives. What remains after those filters is a list of candidates, not a list of confirmed illegal vessels. Probable cause for interdiction requires additional corroboration.
Revisit rates and the gap between detection and interdiction
The most significant operational constraint is time. Sentinel-1 revisits a given ocean area every 6–12 days under current single-satellite operations. A vessel detected as dark in a Sentinel-1 scene acquired on Monday morning could be anywhere within a radius of several hundred nautical miles by the time a patrol asset responds. Commercial X-band constellations reduce this gap: ICEYE and Capella can task a specific area within hours of a request, enabling follow-up imaging that confirms whether a dark vessel is still present and refines its position.
Even with same-day commercial tasking, satellite SAR is not a real-time surveillance system. It is a detection and evidence system. The realistic operational model is: Sentinel-1 broad-area sweep identifies a cluster of dark vessels in a zone of interest; commercial SAR provides a higher-resolution confirmatory image; the combined evidence package is passed to maritime patrol authorities for asset deployment. Satellite AIS feeds provide continuous track data for vessels that are broadcasting, which can indicate whether a known vessel switched off at a suspicious moment and in a suspicious location. The combination is considerably more powerful than either data stream alone.
What the analytics actually produce
A dark-vessel detection workflow produces several distinct outputs. The primary deliverable is a georeferenced alert layer: point features representing unmatched SAR detections, attributed with estimated vessel length (derived from the radar return footprint), the time of detection, sea state at acquisition, and a confidence score based on CFAR threshold margin and clutter conditions. Secondary deliverables include a corroborated-vessel layer showing all SAR detections that matched an AIS broadcast, which provides context and validates the detection pipeline.
Over time, repeated passes build a pattern-of-life picture. A vessel that appears dark in the same fishing ground across multiple Sentinel-1 acquisitions, while neighbouring vessels broadcast normally, is a qualitatively different finding from a single anomalous detection. Historical SAR archives extend to 2014 for Sentinel-1, enabling retrospective analysis of fishing pressure in zones where regulatory interest has recently increased. Satellize runs this correlation and pattern analysis on open Sentinel-1 data and can add commercial SAR tasking for priority areas under client licence, in the same configuration used to support analytics programmes in maritime-dependent economies.
What this method cannot do
SAR dark-vessel detection does not identify who owns a vessel, what species it is targeting, or whether it holds a valid licence in the jurisdiction where it operates. It produces a list of physically present, non-broadcasting vessels. Determining whether non-broadcast constitutes an IUU violation requires knowledge of the applicable legal framework, the vessel's flag state obligations, and the specific maritime zone. Some vessels legitimately operate without AIS, particularly small-scale artisanal craft below the mandatory carriage threshold (generally vessels under 15 gross tonnes or 15 metres in many jurisdictions, though thresholds vary).
Cloud cover does not affect SAR, which is a genuine advantage over optical methods. Rain at X-band is a partial exception. Shallow coastal waters with complex bathymetric backscatter patterns can complicate detection near shorelines. And the method says nothing about what is happening below the surface: a vessel with its gear deployed but its engine off may present a smaller radar cross-section than expected, potentially falling below the detection threshold in moderate clutter conditions. Honest deployment of this capability means presenting it as a screening and prioritisation tool, not a definitive enforcement record.
Typical figures
| Sentinel-1 SAR spatial resolution (IW mode) | 10 m ground range × 10 m azimuth (multi-looked product); swath 250 km |
| ICEYE / Capella SAR spatial resolution (spotlight) | Approximately 0.5 m; strip mode approximately 3 m |
| Sentinel-1 revisit (current single-satellite) | 6–12 days depending on latitude and acquisition plan |
| Commercial X-band SAR tasking latency | Same-day to next-day; priority tasking under 2 hours for some operators |
| SAR frequency bands used | C-band 5.405 GHz (Sentinel-1); X-band ~9.65 GHz (ICEYE, Capella) |
| Minimum detectable vessel length (C-band, Beaufort ≤5) | Approximately 20 m reliably; smaller targets detectable at higher false-positive rate |
| Satellite AIS global message latency | Typically under 30 minutes (Spire constellation) |
| Sentinel-1 archive depth | From April 2014 (Sentinel-1A launch) |
| Positional matching tolerance (SAR to AIS) | 500 m to 1 km, depending on AIS cadence and SAR geolocation accuracy |
| Delivery formats | GeoJSON alert layer, GeoTIFF SAR chip, CSV detection log, PDF evidence summary |
Analytics Satellize can run
| Dark-vessel alert layer | CFAR vessel detection on SAR imagery correlated against interpolated satellite AIS positions; unmatched detections flagged as non-reporting | GeoJSON point layer with vessel length estimate, confidence score, sea state flag, and acquisition timestamp; updated per SAR pass |
| Corroborated-vessel reference layer | SAR detections matched to AIS broadcasts within positional tolerance; validates pipeline and provides operational context | GeoJSON layer of confirmed reporting vessels in the same scene, for comparison with dark detections |
| Pattern-of-life dark-vessel report | Multi-temporal stacking of detection layers across Sentinel-1 archive passes; persistent non-reporting targets identified by spatial clustering | PDF report with mapped recurrence hotspots, vessel count time series, and annotated SAR chips for selected high-confidence targets |
| High-resolution confirmatory SAR chip | Commercial X-band tasking (ICEYE or Capella) triggered by Sentinel-1 alert; sub-metre imagery allows hull-shape discrimination and false-positive rejection | GeoTIFF spotlight chip with analyst annotation distinguishing vessel from fixed infrastructure or clutter |
| AIS behaviour anomaly flag | Analysis of Spire AIS track history to identify vessels with a pattern of transponder gaps in specific zones or at specific times; correlated with SAR detection events | CSV anomaly log per vessel MMSI, with gap duration, last known position before gap, and SAR detection record if available |
| Zone-level IUU pressure index | Normalised dark-vessel detection density per unit area per month, derived from Sentinel-1 archive; benchmarked against total SAR-detected vessel count to estimate non-reporting fraction | Monthly GIS raster layer and summary table for defined exclusive economic zone or managed area; suitable for regulatory reporting |
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.