About this project
About Seahound
A personal project. Work in progress. Not a commercial product.
What this is
Seahound is a personal side project that processes publicly available Sentinel-1 synthetic aperture radar (SAR) imagery from the European Space Agency's Copernicus programme to estimate vessel congestion at a small number of global maritime chokepoints.
The supply-chain graph — connecting commodity sources to ports to end markets — is hand-curated and represents a rough sketch of trade flows, not a verified commercial dataset. The intent is to explore whether open satellite data can produce a useful early-warning signal for supply-chain risk.
Current status
Work in progress
This application is under active development. Data updates are semi-automated — the processing pipeline runs when triggered manually, not on a continuous schedule. There will be gaps between satellite passes, processing runs that fail silently, stale risk scores, and missing observations. This is expected.
Updates are driven by Sentinel-1 revisit cadence — typically every 6–12 days per location — and only when the processing pipeline has been run against the latest available scene. Not every overpass produces a valid observation.
What the data does and doesn't mean
“Likely vessel detections” are not confirmed vessels
The detector identifies bright SAR targets — high-backscatter connected components — consistent with large metal objects on water. These include ships, but also offshore platforms, buoys, and processing artefacts. The algorithm is intentionally conservative and should be read as an MVP signal, not confirmed vessel identification.
Risk scores are simple anomaly estimates
A risk score is derived from the ratio of detected targets to a static per-zone baseline count. It is not a model trained on historical congestion events. Scores can be thrown off by cloud cover confusion, SAR noise, detection failures, or an outdated baseline.
SAR has fundamental detection limits
Oil and vessel detection from SAR fails at wind speeds above ~12 m/s, under rain cells, and for targets smaller than the sensor resolution. A “low risk” reading may simply reflect conditions where the sensor cannot see the scene clearly — not an absence of activity.
What this is not
- Not financial, trading, or investment advice of any kind.
- Not a commercial maritime intelligence product. It has no SLA, no uptime guarantee, and no support.
- Not a real-time system. Data can be hours or days behind the latest satellite pass.
- Not a replacement for AIS data, commercial vessel tracking, or professional maritime analysis.
Data sources
Satellite imagery: Copernicus Sentinel-1 GRD scenes, sourced via the Element84 Earth Search STAC API on AWS Open Data. Contains modified Copernicus Sentinel data processed by Seahound.
Supply-chain graph: hand-curated, drawing on publicly available trade statistics, port authority data, and commodity flow reports. Not verified against any proprietary source.