Oil spill detection from satellite imagery using AI
Find and outline slicks on the sea surface across wide areas, day or night, in most weather.
Darkocean's oil spill detection applies machine learning to satellite imagery, chiefly synthetic aperture radar, to find candidate slicks over large stretches of sea. It is built for monitoring offshore assets, shipping lanes and coastlines, and for giving responders a location to investigate.
Sees through cloud and in darkness
Depends on satellites and latitude
Hundreds of kilometres per image
With confidence scores for review
How does satellite oil spill detection work?
Synthetic aperture radar (SAR) satellites send microwave pulses to the sea and record the echo. A wind-roughened sea scatters a good share of that energy back and looks bright. Oil damps the short capillary waves that cause the scattering, so a slick appears as a dark patch against a brighter background. Radar needs no daylight and passes through cloud, which makes it the standard sensor for marine oil monitoring.
The difficulty is that many other things also make the sea look dark. Low-wind areas, rain cells, algal blooms, natural films, internal waves and the shadows of land or ships can all resemble oil. A model that simply thresholds dark pixels produces many false alarms. Darkocean's approach uses machine learning trained to separate likely oil from look-alikes, using the shape and context of each dark feature, the local wind conditions, and its position relative to ships, platforms and pipelines.
Optical imagery has a supporting role. In daylight and clear sky, optical and thermal data can help confirm a candidate slick, but it cannot replace radar for routine monitoring. Detection also has limits. Very low winds, below roughly 2 to 3 m/s, leave no background for a slick to contrast with, and very high winds, above roughly 10 to 12 m/s, can disperse or hide it.
The output is a set of candidate slicks with outlines, area and a confidence level. It is a screening tool that tells a human analyst where to look and where to send an aircraft or vessel. It does not identify the type of oil or the source on its own.
How it works
Acquire
Task or collect SAR imagery over the area of interest, with optical data added where it helps.
Prepare
Calibrate, mask land and correct the imagery so that sea-surface features can be compared.
Detect
Run the model to find dark features and score how likely each is to be oil rather than a look-alike.
Review
Check candidates against wind data, vessel positions and known infrastructure.
Report
Deliver outlines, areas and confidence levels in formats your response team can use.
What you receive
Detections
a. Candidate slick polygons
b. Area and confidence score for each
c. GIS layers in common formats
Context
a. Wind and sea-state conditions at acquisition
b. Nearby vessel and platform positions where available
c. Annotated image extracts
Monitoring
a. Repeat-coverage change summaries
b. Alerts for new detections in a defined area
Where it is used
Offshore platformsWatch the sea around installations and export pipelines for unexplained slicks.
Shipping lanesScreen busy routes for illegal discharges and bilge dumping.
Port and terminal approachesMonitor anchorages and loading areas for chronic or accidental releases.
Natural seepsMap repeated slicks from the seabed as an indicator of hydrocarbon seepage.
Coastal protectionGive environmental agencies early warning that a slick may be heading for shore.
Incident responseLocate and track a known spill to guide aircraft and vessel deployment.
Frequently asked questions
Can satellites detect all oil spills?
No. Detection depends on wind, the type and thickness of oil, and satellite timing. Thin sheens and very low or high wind conditions can be missed, and a satellite can only see the sea when it passes overhead.
How do you avoid false alarms?
Radar shows many dark features that are not oil. The model scores each candidate using its shape, the local wind and nearby ships or infrastructure, and an analyst can review the low-confidence cases. Some false positives will remain, which is why we report confidence rather than a yes or no.
Is this real-time monitoring?
It is near-regular rather than continuous. Revisit depends on which satellites are used and how far north or south the area lies, and processing takes time after each pass. For continuous surveillance, satellite detection is usually combined with aircraft, vessels or fixed sensors.
Can you tell where the oil came from?
Not from the image alone. Position, shape and nearby vessel or platform data can suggest candidate sources, but attribution needs additional evidence such as vessel tracking history and samples.
Need to watch the sea around your assets?
Tell us the area and how often you need coverage, and we will outline a monitoring approach.
