Tree monitoring
Finding individual trees is the first step. At satellite resolution it is a hard problem, and it is the one our pipelines are built for. On a new site we can deliver a first count within weeks. Following the same tree for years, across sensors and seasons, turns those detections into a record.
What we are working toward now is early warning, which means seeing in multispectral imagery that a tree's health is declining while there is still time to act. Our research so far is showing fascinating results.
Early warning
A dying tree changes over years, and a sequence of captures can show what a single image can't.
Where the imagery can't support a claim about one tree, we say so and report at stand level.
How it adapts to your site
A model trained on one forest usually struggles on the next, and on different imagery or a different season. We take a detector from open data down to your domain: your sensor, your forest, your season. Three stages.
Trained in-house on crown annotations from around the world, across resolutions, so one model handles drone, aerial and satellite imagery.
Starting point for everything below.We correct the model's output on a sample of your own imagery and train on those corrections. Correcting is much faster than labeling from scratch, so this takes days rather than months.
We measure the review effort needed on your imagery.We freeze and version the model so results remain reproducible. We check each new capture on a reviewed sample before relying on its predictions. When an update helps, we test and release a new version.
Each capture is checked on a reviewed sample.Every step is scored on imagery held back from training, and we show you the numbers. It works with the imagery you already have, or we help you get it: drone flights, public aerial archives with years of history, or 30 cm commercial satellite imagery when you need to cover a lot of ground.
Try it on a place you care about
If there is a forest, a park or a city you care about and some imagery of it, we'll run a session and show you what comes out, and what it would take to run inside your own systems. If you are building tools of your own, or want to join the mission, we'd like to hear from you.