TerrAlert Team · 2026-09-09
A change detection system compares two pictures of the same place, taken at different times, and works out what is different between them. That is the whole idea. Everything else (satellites, AI models, alerts) exists to make that comparison fast, frequent and reliable enough to be useful.
Imagine you own a small patch of forest two hours' drive away. You cannot visit every week. So you take a photo from the same hill every time you do visit, and you compare each new photo with the last one. If a section of trees is missing that was there last time, you have detected a change, by eye, slowly, and only as often as you can make the trip.
A change detection system automates exactly that comparison, using pictures a satellite takes instead of ones you take yourself, and a computer doing the comparing instead of your eyes. The trade you make is speed and reach: you can watch a hundred places you could never physically visit, in exchange for a kind of attention that is narrower than a human's. It looks for that something changed, not why, and it needs a human to interpret an unusual result. See How to Spot Important Changes Before They Become Bigger Problems for why that automated comparison is worth setting up in the first place.
A security camera watches continuously and shows you a live picture. A change detection system does not do that. It only has something to say when a new comparison is made, and a new comparison can only happen when a new, usable image exists. For satellite imagery, "usable" is the operative word: cloud cover routinely removes a large share of what a satellite captures, so the true rhythm of detection is set by how often a clear image becomes available, not by how often the satellite passes overhead.
This distinction matters because it is where most disappointment with monitoring tools comes from. A tool that says "daily monitoring" is describing how often it checks, not how often it finds something new to compare against. If no clear image arrived since the last check, there is nothing new to compare, and nothing to report, correctly.
A well-built system does not try to name every kind of change in the world. It is built to notice specific, defined categories (a building went up, tree cover disappeared, a lake got bigger, a burn scar appeared) because narrowing the target is what makes the comparison reliable. A system that claims to detect "anything different" tends to also flag things nobody cares about: a shadow moved, a cloud's edge sat differently, the sensor's own calibration drifted a little between passes.
Is this the same as a security camera? No. A camera streams continuously; a detection system only produces a result when two comparable images exist, which for satellite-based systems can be days apart.
*Can it tell me why something changed? Not directly. It tells you that* an area changed and roughly what kind of change it resembles. A human still has to look at the imagery and decide what happened.
See how this applies in practice on the TerrAlert overview or read How TerrAlert Works.
All posts