Oil spill detection and extent mapping with SAR
Synthetic aperture radar detects surface oil by the backscatter suppression it causes on centimetre-scale ocean waves. Sentinel-1 is the operational standard, but false positives from biogenic films and wind shadows demand careful contextual analysis.
Sensors
- Sentinel-1 (C-band SAR): Primary operational sensor. Interferometric Wide Swath mode delivers 10 m resolution across a 250 km swath in VV and VH polarisation. Revisit is 6 days at the equator for a single satellite, 3 days with both Sentinel-1A and 1B active. Free and open archive from 2014.
- RADARSAT-2 (C-band SAR): Canadian commercial system offering flexible polarimetric modes including full quad-pol, which aids slick characterisation. Fine-beam modes reach 3 m resolution; standard wide mode is 25 m at 500 km swath. Tasked on demand, so latency after an incident depends on orbital geometry and licence.
- COSMO-SkyMed (X-band SAR): Italian constellation of four satellites offering 1 m spotlight resolution and sub-daily revisit under tasking. X-band is more sensitive to fine surface texture than C-band, but the shorter wavelength also increases false positives from rain cells and foam. Best used for confirming extent and edge detail after a C-band detection.
- TerraSAR-X / TanDEM-X (X-band SAR): German commercial system with spotlight modes down to 1 m and a 2.5-day mean revisit. Like COSMO-SkyMed, X-band sensitivity is high but so is rain-cell confusion. Useful for high-resolution follow-up mapping of confirmed spills.
What a floating film does to radar
Ocean radar backscatter is dominated by Bragg resonance: centimetre-scale surface ripples scatter C-band energy (roughly 5.4 GHz, 5.6 cm wavelength for Sentinel-1) back toward the sensor. Oil at the surface, even in films only micrometres thick, suppresses these capillary and short gravity waves through increased surface viscosity. The result is a dark patch in the SAR image, typically 2 to 10 dB below the surrounding ocean backscatter in VV polarisation.
The physics is well understood and the detection is genuinely reliable under the right conditions. Wind speed is the critical variable. Between roughly 3 m/s and 14 m/s, the surrounding ocean is rough enough to produce strong backscatter, making the dampened slick stand out clearly. Below 3 m/s the whole sea surface calms and darkens; above 14 m/s wave breaking re-roughens slick areas. Outside that window, detection degrades significantly and extent estimates become unreliable.
The false-positive problem is not a footnote
Any process that suppresses surface roughness produces a dark patch that looks, to a naive algorithm, like oil. Biogenic slicks from phytoplankton, fish schools and natural seeps are the most common confusion class; they are widespread, especially in productive coastal waters, and can cover hundreds of square kilometres. Wind shadows behind islands, ships or offshore structures create localised calm patches. Rain cells produce circular or elliptical dark areas that can persist for tens of minutes after the rain stops. Low-wind areas can produce large dark regions with no pollution cause at all.
Dual-polarisation data (VV and VH simultaneously, as Sentinel-1 IW mode provides) helps. The ratio of VV to VH backscatter differs between mineral oil and biogenic films because their damping mechanisms differ. Polarimetric decomposition methods, including the use of the degree of polarisation and the normalised difference polarisation index, have been published in peer-reviewed literature and improve discrimination. They do not eliminate ambiguity, particularly in low-wind conditions. Honest operational practice treats any SAR-only detection below 3 m/s wind speed as unconfirmed until contextual evidence is added.
Contextual discrimination in practice
Reducing false positives requires layering evidence. Wind field data from ECMWF numerical weather prediction, or from the SAR image itself via inversion of backscatter in clean ocean areas, identifies whether wind speed is within the detectable range. AIS vessel tracks, cross-referenced against the slick position and drift trajectory, can link a candidate spill to a specific ship. Ocean current and wind-drift modelling allows backward trajectory analysis to estimate the spill origin time and location, which is critical for enforcement.
Shape and texture also carry information. Biogenic slicks tend to have diffuse, fractal edges and low internal contrast. Mineral oil slicks, particularly from continuous discharges such as tank washing, often show elongated shapes aligned with vessel heading. Rain cells are typically circular with sharp edges. None of these rules is absolute, but together they shift a probabilistic classification toward a defensible conclusion. Documented false-positive rates in published validation studies for Sentinel-1-based systems typically range from 20 to 40 percent before contextual filtering, dropping to under 10 percent after.
Minimum detectable size and archive depth
At 10 m pixel spacing in Sentinel-1 IW mode, the practical minimum detectable slick area is roughly 0.01 km², though this depends on the signal-to-noise ratio of the surrounding ocean. Very thin films near the detection threshold of the sensor can be missed entirely. The Sentinel-1 archive runs from April 2014, giving over a decade of consistent C-band coverage globally, which is sufficient for baseline studies of chronic discharge locations such as shipping lanes and offshore production areas.
Commercial X-band systems can resolve slick edges and internal structure at 1 to 3 m, which is useful for estimating volume in combination with thickness models, but that thickness estimation is a separate problem addressed by optical sensors and is outside the scope of this page. SAR gives extent and shape; it does not give thickness.
From detection to enforcement evidence
A detection polygon is not an enforcement case on its own. The chain of evidence typically requires: a confirmed dark anomaly with backscatter statistics distinguishing it from background, a wind-speed verification showing conditions were within the detectable range, a drift-corrected trajectory linking the slick to a source, and ideally AIS or vessel monitoring system data placing a specific vessel at the origin point. Each of these steps introduces uncertainty, and any competent legal challenge will probe them.
Sentinel-1's 3-to-6-day revisit means that a spill reported on day one may not have a follow-up SAR pass for several days, by which time the slick has dispersed, weathered or sunk below detection. For enforcement applications requiring rapid confirmation, commercial tasking on RADARSAT-2, COSMO-SkyMed or TerraSAR-X is necessary. Satellize structures analytics pipelines that combine open Sentinel-1 monitoring for chronic discharge detection with commercial tasking triggers for incident response, similar in architecture to the multi-source approach it applies in its Tonga crop-estimation programme.
Latency from satellite pass to delivered alert depends on processing pipeline design. Near-real-time services based on Sentinel-1 can achieve 1 to 3 hours from acquisition to detection output. That is fast enough to direct aerial or vessel verification before the slick disperses significantly, provided the overpass occurs close to the incident.
What SAR cannot tell you
Extent, yes. Thickness, no. Volume, only approximately and with large uncertainty. Whether the oil is fresh crude, weathered emulsion, or diesel requires optical or in-situ data. Whether a dark patch is definitely mineral oil rather than a biogenic film requires contextual analysis, and sometimes that analysis is genuinely inconclusive. SAR is a detection and mapping tool, not a complete characterisation tool.
Cloud cover is irrelevant to SAR, which is one of its principal advantages over optical sensors for maritime monitoring. But heavy precipitation degrades image quality, and the rain-cell false-positive problem is worst precisely when weather is poor. No satellite system provides continuous monitoring of a given ocean point; gaps between passes are real and operationally significant. Understanding these limits before designing a monitoring programme saves considerable time and credibility later.
Typical figures
| Primary sensor | Sentinel-1 IW mode, C-band (5.405 GHz) |
| Spatial resolution | 10 m (Sentinel-1 IW); 1–3 m (COSMO-SkyMed / TerraSAR-X spotlight) |
| Swath width | 250 km (Sentinel-1 IW); 15–100 km (commercial X-band, mode-dependent) |
| Revisit (open data) | 3–6 days (Sentinel-1A+B combined, latitude-dependent) |
| Revisit (commercial tasking) | Sub-daily possible with COSMO-SkyMed or RADARSAT-2 on demand |
| Effective wind-speed window | Approximately 3–14 m/s; detection unreliable outside this range |
| Minimum detectable slick area | Approximately 0.01 km² under favourable wind conditions at 10 m resolution |
| Polarisation | VV + VH dual-pol (Sentinel-1 IW standard); quad-pol available on RADARSAT-2 |
| Archive depth | Sentinel-1 from April 2014; RADARSAT-2 from 2008 (commercial access) |
| Alert latency (Sentinel-1 pipeline) | 1–3 hours from satellite acquisition to detection output, typical NRT pipelines |
Analytics Satellize can run
| Slick detection and extent polygon | Adaptive thresholding on VV backscatter with morphological filtering to remove speckle artefacts; published EMSA CleanSeaNet methodology class | GeoJSON or shapefile polygon with area, perimeter, centroid, acquisition time and backscatter statistics |
| False-positive confidence score | Dual-pol discrimination using VV/VH ratio and normalised difference polarisation index, combined with ECMWF wind-speed verification | Per-detection confidence rating (high / medium / low / unconfirmed) appended to polygon attribute table |
| Drift-corrected source trajectory | Backward Lagrangian particle tracking using Copernicus Marine Service ocean current and ECMWF surface wind fields | PDF report with trajectory map, estimated origin time window and uncertainty radius |
| AIS vessel cross-reference | Spatial and temporal intersection of slick trajectory with AIS position history from licensed provider; MMSI and flag-state lookup | Tabular report listing candidate vessels with proximity scores, suitable for submission to maritime authority |
| Chronic discharge monitoring report | Multi-temporal stacking of Sentinel-1 detections over a defined area; hotspot density mapping using kernel density estimation | Monthly GIS layer showing recurring slick locations, with time-series chart of detection frequency by grid cell |
| Commercial tasking trigger alert | Sentinel-1 NRT detection used as trigger condition; automated tasking request issued to commercial SAR operator for high-resolution follow-up pass | High-resolution (1–3 m) confirmation image with refined extent polygon, delivered within the next available overpass window |
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.