SAR Doppler anomaly measurement for vessel velocity
Synthetic aperture radar encodes a moving vessel's radial velocity as a measurable image displacement. This page explains the physics, the geometry, the numbers, and where the method breaks down.
Sensors
- Sentinel-1 (ESA): C-band (5.405 GHz), IW mode ground range resolution approximately 5 x 20 m, 12-day exact repeat per satellite (6-day with both units operational). Doppler centroid anomaly routinely extractable from Level-1 SLC products. Free and open archive from 2014.
- RADARSAT Constellation Mission (CSA): C-band, three satellites, revisit as short as 4 days at mid-latitudes and near-daily at high latitudes. Multiple beam modes from 3 m spotlight to 100 m ScanSAR. Higher revisit improves the chance of catching a vessel mid-transit rather than at anchor.
- ICEYE SAR (commercial): X-band (9.65 GHz), spotlight mode resolution down to approximately 0.5 m, strip mode around 3 m. Tasking on demand with sub-24-hour revisit to a specific area of interest. X-band's shorter wavelength increases sensitivity to small metallic targets but also to sea clutter in high sea states.
- TerraSAR-X / TanDEM-X (DLR): X-band, spotlight resolution to approximately 1 m, staring spotlight to sub-0.25 m. Used extensively in published Doppler velocity research because of its geometric precision. Commercial tasking; archive from 2007.
What a moving vessel does to a radar image
A synthetic aperture radar works by transmitting pulses and recording the phase of the returned signal across the satellite's flight path. For a stationary target, the geometry is predictable and the focusing algorithm places it correctly. A vessel moving with a radial velocity component (towards or away from the satellite) shifts the Doppler frequency of its return relative to the surrounding sea surface. The processor, not knowing the target is moving, misinterprets that frequency shift as a change in along-track position. The vessel appears displaced from its true location in the final image.
The displacement is proportional to the vessel's radial velocity and inversely proportional to the satellite's velocity divided by its slant range. For Sentinel-1 in IW mode, a vessel moving at 10 knots with a radial component of 5 knots will be displaced by roughly 200 to 400 metres in the along-track direction, depending on incidence angle and look geometry. That displacement is measurable. The satellite's orbital parameters and imaging geometry are precisely known, so the radial velocity can be recovered from the offset.
Extracting velocity from the Doppler centroid anomaly
The Doppler centroid anomaly (DCA) is the difference between the Doppler centroid estimated from the data and the theoretical centroid expected from the satellite's motion alone. For the open ocean, the DCA is dominated by sea surface currents and wind-driven surface motion, both of which are small and slowly varying. A vessel return sits on top of that background as a localised, high-amplitude anomaly. By fitting the DCA of the vessel's pixel cluster and comparing it to the local sea background, analysts can isolate the vessel's contribution and convert it to a radial velocity estimate.
Published work using Sentinel-1 SLC data reports radial velocity estimation accuracy in the range of 0.5 to 1.5 knots (roughly 0.25 to 0.75 m/s) under good conditions. The method requires access to Single Look Complex (Level-1 SLC) products, not the detected (GRD) products that most casual users download. SLC data preserves the phase information that carries the Doppler signal.
The geometry problem: what you cannot see
The critical limitation is that SAR only measures the radial component of velocity, the component along the line of sight from satellite to target. A vessel moving exactly parallel to the satellite's flight path has zero radial component. Its Doppler return is indistinguishable from a stationary target. It will appear in the correct position with no anomaly, and the method yields nothing useful about its speed.
In practice, Sentinel-1 ascending and descending passes have different look directions, typically separated by around 150 degrees in azimuth over mid-latitudes. Combining passes from both geometries constrains the ambiguity considerably, though it requires two acquisitions that may be hours or days apart. For a single pass, the method is most reliable for vessels whose heading is within roughly 30 to 60 degrees of the satellite's look direction. Below about 3 knots of actual vessel speed, the radial component often falls below the noise floor of the DCA estimation, particularly in moderate sea states where wave orbital velocities add scatter.
High sea states are a compounding problem. Wave orbital velocities at the surface can reach 1 to 2 m/s in sea state 4 or above, which is the same order of magnitude as the vessel signal. The DCA background subtraction becomes less reliable, and velocity estimates degrade accordingly.
Combining displacement with wake geometry
The along-track displacement gives radial velocity. The vessel's Kelvin wake, where visible, gives heading. Combining the two allows a vector velocity estimate rather than just a scalar radial component. This is not always possible: wake visibility depends on wind speed, sea state, and the vessel's size and speed. At wind speeds above roughly 8 to 10 m/s, the wake is often suppressed by surface roughness. At very low speeds, the wake may be too faint to detect at Sentinel-1's resolution.
When both signals are available, the combined approach is substantially more informative than either alone. The displacement method is covered here; wake-based heading and speed reconstruction is treated separately in the vessel wake analysis page in this library.
Practical intelligence value and honest caveats
The primary use case is vessels that have switched off their AIS transponder. A detected SAR target with no AIS correlation is already suspicious. Adding a velocity estimate, even an approximate one, narrows the possibilities considerably. A vessel drifting at 0.5 knots behaves differently from one making 12 knots on a consistent heading. Speed-over-ground, even with its angular ambiguity, is operationally meaningful for tracking, interception planning, and pattern-of-life analysis.
Revisit rate is the binding constraint for most operational users. Sentinel-1's 6-day revisit (with both satellites) means a vessel in a given area may be imaged once or twice per week at best. RADARSAT Constellation Mission improves this at higher latitudes. Commercial tasking through ICEYE or similar operators can reduce the gap to hours for a defined area of interest, at cost. Latency from acquisition to processed SLC product is typically 1 to 3 hours for Sentinel-1 NRT products via the Copernicus Data Space, though full SLC processing and DCA analysis adds further time.
Satellize applies DCA-based velocity extraction as part of its dark-vessel analytics pipeline, combining it with AIS gap detection and optical cross-cues where cloud cover permits. The method is one layer in a multi-source picture, not a standalone answer.
Archive depth and historical pattern analysis
Sentinel-1 SLC data is available from late 2014 for Sentinel-1A and from 2016 for Sentinel-1B (now retired). The archive is free and openly accessible through the Copernicus Data Space Ecosystem. This depth allows retrospective analysis: if a vessel of interest was transiting a region two years ago, the SAR archive may contain the pass that imaged it, and DCA analysis can be applied post hoc.
TerraSAR-X offers a commercial archive from 2007 with higher geometric precision, useful for research-grade validation. The practical limit on historical analysis is not the archive but the computational cost of processing large volumes of SLC data to extract DCA fields across wide areas. Targeted processing on a known vessel track or a defined chokepoint is tractable; global retrospective surveys are expensive.
Typical figures
| Primary frequency | C-band 5.405 GHz (Sentinel-1, RADARSAT); X-band 9.65 GHz (ICEYE, TerraSAR-X) |
| Spatial resolution (SLC) | Sentinel-1 IW: ~5 x 20 m (range x azimuth); ICEYE spotlight: ~0.5 m; TerraSAR-X spotlight: ~1 m |
| Revisit (open constellation) | Sentinel-1: 6 days (single satellite), 12-day exact repeat; RADARSAT Constellation: 4 days mid-latitude, near-daily high latitude |
| Radial velocity accuracy | 0.5 to 1.5 knots (0.25 to 0.75 m/s) under sea state 1-3; degrades above sea state 4 |
| Minimum detectable vessel speed | Approximately 3 knots radial component; lower speeds fall into sea-surface noise floor |
| Data product required | Level-1 SLC (Single Look Complex); GRD products are insufficient for DCA extraction |
| NRT latency (Sentinel-1) | 1 to 3 hours from acquisition to SLC product via Copernicus Data Space; DCA analysis adds 30 to 90 minutes |
| Archive depth | Sentinel-1A from late 2014; TerraSAR-X commercial archive from 2007 |
| Coverage per pass | Sentinel-1 IW swath: 250 km wide; ICEYE spotlight: ~5 x 5 km tasked area |
| Geometric blind spot | Vessels moving parallel to satellite flight path yield zero radial component; method fails for those headings |
Analytics Satellize can run
| Radial velocity estimate per dark vessel detection | Doppler centroid anomaly extraction from Sentinel-1 SLC IW data, background sea-surface DCA subtraction | GeoJSON point layer with vessel centroid, displacement vector, estimated radial speed, and confidence flag per acquisition |
| True position correction for SAR-detected vessels | Along-track displacement inversion using published satellite geometry parameters | Corrected vessel position shapefile with displacement magnitude and direction attributes |
| Velocity-filtered alert for vessels above threshold speed | DCA-derived radial speed threshold applied to all non-AIS SAR detections in an area of interest | Near-real-time alert feed (JSON or email) triggered when a dark vessel exceeds a client-defined speed threshold |
| Multi-pass velocity time series for a named vessel | Retrospective SLC archive processing across all available passes intersecting a vessel's known operational area | CSV time series of radial velocity estimates with pass metadata, delivered as a structured report |
| Ascending/descending geometry fusion for vector velocity | Dual-geometry DCA combination to resolve along-track and cross-track velocity components, following published two-look inversion methods | Vector velocity estimate (speed and bearing range) per vessel, GIS layer with uncertainty ellipse |
| Sea-state quality flag per acquisition | ECMWF ERA5 or NOAA GFS wind/wave fields co-registered to SAR scene to flag acquisitions where DCA background noise exceeds velocity signal | Per-scene quality metadata appended to all vessel velocity products |
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.