You draw a boundary on a map around somewhere you care about — a concession, a construction site, a stretch of forest, a supplier’s land. TerrAlert then fetches new Sentinel-2 satellite imagery of that area as it becomes available, compares each new clear image against the previous one, and tells you where the ground has changed.
It is built for the case where nobody can visit often enough: land you do not own, sites you audit rather than run, or areas too remote or too numerous to inspect.
Realistically, several days — and it is worth understanding why, because no satellite service can honestly promise same-day.
Sentinel-2 passes over a given point roughly every two to three days, not daily. Imagery becomes downloadable around half a day to a day after that pass. Cloud then removes a large share of what is left: across our own monitored areas, only about 42% of scenes are clear enough to use. And a change can only be found by comparing two clear images, so what matters is the gap between consecutive usable passes — about five days on average.
So for most areas, expect a detection within roughly a week of a change, sometimes faster after a run of clear weather, and longer in a persistently cloudy region. Anyone quoting you minutes for optical satellite monitoring is describing the alert delivery, not the observation.
Sentinel-2 Level-2A optical imagery at 10 metres per pixel, from the European Space Agency’s Copernicus programme, accessed through the Copernicus Data Space. It is open data, so there is no per-image licensing cost and no minimum commitment.
Detections can additionally be cross-checked against independent public signals — NASA FIRMS active fire, GLAD and RADD forest-loss alerts, and flood extent — so that a finding is not resting on a single model output.
We do not publish an accuracy figure, because we do not yet have one we would stand behind. The model has been evaluated on public benchmark imagery, but not on a labelled set built from our own Sentinel-2 pipeline, and quoting the benchmark number as though it described the live system would be misleading. Building that evaluation set is current work.
What we do instead is make every detection checkable. Each one carries a confidence score you can filter on, a before-and-after view of the actual imagery, and where available a corroborating independent signal. The intended workflow is that you look at the pair of images and decide — not that you trust a number.
At 10 metres per pixel, an area needs to be at least about 100 metres on a side — roughly 0.01 km² — before change inside it can be resolved at all. Smaller than that and the whole area is a handful of pixels, so nothing can be detected no matter how long you wait. Areas below that size are refused when you draw them, with an explanation.
At the other end, the maximum area per boundary depends on your plan, from 200 km² on the free tier up to 3,000 km² on Pro.
Nothing useful, which is the honest answer. Optical satellites cannot see through cloud. A scene where cloud covers the area is recorded but not compared, because comparing a cloudy image against a clear one produces confident nonsense.
In practice this means monitoring is faster in dry seasons and slower in wet ones, and some areas go weeks without a usable image. The comparison view shows the cloud share of each scene so you can see exactly why a period is quiet.
By email, or by webhook to a URL you supply. Webhooks are signed with HMAC-SHA256 so you can verify the payload came from us, and a Slack or Microsoft Teams incoming-webhook URL is detected automatically and posted as a readable chat message rather than raw JSON.
You set a minimum confidence per area, so you decide where the line sits between hearing about everything and hearing only about the clear-cut cases.
Yes, on Pro and Enterprise. It is a REST API authenticated with a bearer token, covering your monitored areas, detections and alert history, with a default limit of 200 requests a minute. Detections can also be exported as CSV or GeoJSON from the interface on any paid plan.
Free for one area up to 200 km² with weekly checks, with no card required. Paid plans run from $149.99 a month for three areas up to $599.99 for twenty, with larger areas, more frequent checks, webhook alerts, longer history and API access. Enterprise is priced per deployment.
Pricing is flat monthly. There is no per-image or per-scene charge, because the underlying imagery is open data.
Your boundaries and detections are yours. Every detection can be exported as CSV or GeoJSON, and you can request a full export or complete erasure of your account data at any time.
The underlying satellite imagery is Copernicus open data, free for anyone to use under its own terms.
See pricing or the TerrAlert overview.