MarSlick
SIH 26143
Smart India Hackathon 2026 · PS 26143

Findtheshipthatspilledit.

A tanker washes its tanks at night. By the time a satellite passes, the slick has drifted tens of nautical miles and a thousand ships have crossed the area. Coastal authorities take two to three days to work backward by hand, and the vessel is in foreign waters by then.

MarSlick does that reconstruction in seconds — and is built to say “insufficient evidence” when the data does not support naming anyone.

48–72 hManual attribution today
~10 sEnd-to-end, this pipeline
500Drift hypotheses per case
SHA-256Sealed, reproducible record
01

Open a case

Upload a SAR scene. Detection runs on the image; drift and attribution require a georeference.

Drop a Sentinel-1 SAR scene
GeoTIFF preferred · PNG and JPEG accepted with manual coordinates
Quick SAR Demo Scenes:
Georeference

Optional for GeoTIFF. Without coordinates the pipeline still detects the slick but withholds drift and attribution rather than invent a position.

Layer 1 · SAR gateLayer 2 · look-alikeLayer 3 · georeference
02

How the attribution is made

Detect, then interrogate

A U-Net segments dark regions in Sentinel-1 backscatter. Because a dark patch is not automatically oil, every candidate is then tested against the physics that separate a slick from a calm zone, a biogenic film or a rain cell.

Rewind the ocean

Five hundred particles are advected backward through an hourly met-ocean field, each drawing its own wind, current, windage and drift duration. The result is a probability distribution over the release point, not a single dot.

Ask who was there

Each candidate vessel's AIS track is interpolated to every ensemble member's release time. The share of hypotheses that place it at the scene becomes its likelihood — joint in space and time, so being near at the wrong moment counts for nothing.

03

Recent cases