Disaster-triggered oil spill and chemical release detection from SAR
Earthquakes, floods, and industrial accidents release hydrocarbons into waterways before optical sensors can see through cloud cover. SAR detects surface oil films through Bragg-scattering suppression, producing dark signatures in C- and X-band imagery within hours of an event.
Sensors
- Sentinel-1 A/B (C-band SAR): 20 m resolution in Interferometric Wide Swath mode, 250 km swath, repeat pass every 6–12 days at mid-latitudes (reduced to 12 days with single satellite after Sentinel-1B failure in 2021). Free, open archive from 2014. The workhorse for initial spill extent mapping.
- ICEYE SAR constellation (X-band): Spotlight mode delivers approximately 0.5 m resolution; Strip mode approximately 3 m at 30 km swath. Constellation revisit can reach sub-daily for priority targets. X-band is more sensitive to thin films but also more prone to wind-state ambiguity.
- Capella Space (X-band SAR): Spotlight scenes to approximately 0.5 m resolution, tasked on demand with typical delivery latency of a few hours after collection. Useful for resolving spill boundaries at harbour or riverbank scale that Sentinel-1 cannot separate.
- COSMO-SkyMed (X-band SAR, ASI): First and Second Generation constellations. ScanSAR mode covers 200 km at 30 m; Spotlight to approximately 1 m. Italy's civil protection authority has operational spill-detection protocols built on this system, providing a documented precedent for disaster-response tasking.
What a surface film does to radar backscatter
Open water returns radar energy through Bragg scattering: centimetre-scale capillary waves resonate with the incident C-band (5.6 cm) or X-band (3.1 cm) wavelength and scatter energy back to the sensor. Oil and many other surface-active contaminants damp those capillary waves. The result is a patch of anomalously low backscatter, appearing dark against the surrounding water in a normalised radar cross-section image.
The physics is well established. The degree of damping depends on the film's viscoelastic properties, which vary with oil type, weathering state, and temperature. Fresh crude oil suppresses backscatter by 4–8 dB relative to clean water under moderate wind conditions. Emulsified or weathered oil produces weaker damping and can become nearly invisible to SAR at low wind speeds, where clean water is also calm and the contrast collapses. This is not a defect to hide; it is a calibration constraint every analyst must carry into interpretation.
The look-alike problem is not trivial
Several phenomena produce dark patches in SAR imagery that are not oil. Wind shadows behind islands, headlands, or large vessels suppress capillary waves and mimic slick signatures. Biogenic films from algal blooms, fish schools, and decaying organic matter create genuine surface damping that is physically identical to the oil signature at C-band. Rain cells, internal waves, and areas of naturally low wind speed all generate dark patches.
Discrimination relies on shape, context, and auxiliary data rather than backscatter alone. Oil slicks from a point source tend to elongate downwind with relatively sharp leading edges and diffuse trailing margins. Biogenic films are typically more irregular and patchy. Wind shadows are geometrically constrained by the obstacle that casts them. Overlaying a near-simultaneous wind field from ECMWF or ASCAT scatterometry is standard practice: if wind speed is below roughly 2–3 m/s, the scene is largely uninterpretable for spill detection, and any dark feature should be flagged as ambiguous rather than confirmed. Above approximately 10–12 m/s, wave breaking roughens the surface sufficiently to suppress the damping contrast, and thin films become undetectable.
In disaster contexts, the prior probability of a real release is elevated. A magnitude-7 earthquake near a refinery cluster, or a major flood event in a petrochemical corridor, shifts the statistical weight toward genuine contamination. That context should be documented explicitly in any assessment, not silently folded into a confident-sounding classification.
Combining wide-area and high-resolution SAR for boundary delineation
Sentinel-1 IW mode is the natural first pass. Its 250 km swath means a single acquisition can cover an entire river delta or coastal bay, and the 6–12 day repeat (or faster under emergency tasking via the International Charter) provides a baseline against which change is measured. Polygon extraction from a single post-event scene requires a clean pre-event reference; without one, biogenic films present before the disaster will be misattributed.
Once a candidate slick is identified in Sentinel-1, a same-day or next-day ICEYE or Capella tasking resolves the boundary at sub-metre scale. At 0.5 m Spotlight resolution, the edge of a slick against a seawall or river bank is mappable to within a few metres, which matters for containment boom placement and shoreline impact assessment. The two datasets are complementary, not redundant: wide-area detection followed by targeted high-resolution confirmation is the operationally sensible sequence.
Cosmo-SkyMed's documented use in the Italian civil protection system, including the 2012 Costa Concordia incident, provides a real-world reference for what this workflow produces under operational pressure. The physics and the processing chain are the same regardless of which X-band asset is tasked.
Quantitative limits the buyer should know before commissioning
Minimum detectable film thickness for SAR is not a fixed number. Published studies suggest that films thicker than roughly 0.1–0.5 micrometres produce detectable damping under favourable wind conditions (3–10 m/s), but this range reflects experimental variation across oil types, sea states, and incidence angles. Thin sheens from minor leaks may fall below detection entirely.
Spatial resolution sets a floor on feature size. A 20 m Sentinel-1 pixel cannot resolve a narrow river channel or a slick confined to a harbour basin. Minimum mappable spill area in IW mode is realistically several hectares. High-resolution commercial SAR closes much of that gap but cannot be pre-positioned everywhere; tasking latency of a few hours is excellent by historical standards but still means the first post-event acquisition may be 12–36 hours after rupture, depending on orbital geometry.
Chemical releases other than hydrocarbons are harder. Some solvents and industrial chemicals do not produce a persistent surface film and leave no SAR signature. Dissolved contaminants are invisible to SAR entirely. Optical or hyperspectral sensors can detect certain dissolved plumes through water-leaving radiance changes, but that is a different sensor class and a different page.
From detection to response: what the data products look like
The primary deliverable is a georeferenced polygon layer of candidate slick extent, classified by confidence level (confirmed, probable, possible) based on shape, context, and wind-state auxiliary data. A second layer marks the probable source point or upstream origin, inferred from slick geometry and known infrastructure locations. Both layers are delivered as GeoJSON or shapefile for direct ingestion into emergency management GIS platforms.
Time-series stacking across multiple acquisitions allows drift trajectory estimation. If Sentinel-1 acquisitions are available at 6-day intervals, the displacement of the slick centroid between scenes, combined with surface current data from Copernicus Marine Service, supports backward advection to confirm or narrow the source location. This is relevant when a disaster has ruptured multiple facilities and attribution matters for liability or containment priority.
Satellize runs this workflow on open Sentinel-1 data and supplements with commercial tasking where clients hold the necessary licences. The same analytic infrastructure that supports the Tonga crop-estimation programme handles multi-source SAR ingestion; the sensor changes, the processing logic for change detection and polygon extraction does not. Analysts reviewing outputs should always receive the wind-state metadata alongside the slick classification, not as an appendix but as a condition of the confidence rating.
Typical figures
| Spatial resolution (Sentinel-1 IW) | 20 m ground range, 250 km swath |
| Spatial resolution (ICEYE / Capella Spotlight) | Approximately 0.5 m; Strip mode approximately 3 m |
| Revisit (Sentinel-1, single satellite) | 6–12 days at mid-latitudes; faster under Charter emergency tasking |
| Revisit (commercial X-band, tasked) | Sub-daily for priority targets with sufficient constellation assets |
| Radar frequency | C-band 5.405 GHz (Sentinel-1); X-band ~9.6 GHz (ICEYE, Capella, COSMO-SkyMed) |
| Operational wind window | Approximately 2–3 m/s minimum to ~10–12 m/s maximum for reliable slick contrast |
| Minimum detectable slick area (Sentinel-1 IW) | Realistically several hectares; sub-hectare features require high-resolution X-band |
| Cloud and daylight sensitivity | None: SAR is all-weather, day/night |
| Archive depth (Sentinel-1) | From 2014 (Sentinel-1A launch); accessible via Copernicus Data Space |
| Delivery formats | GeoJSON, Shapefile, GeoTIFF (sigma-nought); wind-state metadata included |
Analytics Satellize can run
| Candidate slick extent polygon | Adaptive thresholding on normalised radar cross-section (sigma-nought) with morphological filtering to remove speckle artefacts | GeoJSON polygon layer with confidence classification (confirmed / probable / possible) |
| Source-point attribution | Backward geometry tracing from slick centroid combined with known infrastructure inventory and surface current vectors | Point layer with ranked candidate sources and supporting rationale note |
| Look-alike discrimination report | Contextual analysis using ECMWF or ASCAT wind fields, slick shape metrics, and pre-event biogenic baseline | PDF assessment with wind-state metadata and false-positive risk rating |
| Spill drift trajectory estimate | Multi-temporal centroid displacement across Sentinel-1 acquisitions fused with Copernicus Marine Service surface currents | Vector trajectory layer with timestamps; suitable for containment boom positioning decisions |
| High-resolution boundary delineation | Segment-anything or edge-detection applied to commercial X-band Spotlight scenes after Sentinel-1 initial detection | Fine-scale GeoJSON boundary accurate to within a few metres, for shoreline impact and boom placement |
| Multi-date spill evolution stack | Change detection across registered SAR time series to track areal growth, contraction, or fragmentation | Animated GeoTIFF stack with per-acquisition area statistics table |
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.