Labels
Classify
Name a class, select the label icon, and select or paint the representative pixels that belong to that class to label them.
Labelling tips. Spread labels across the whole map: seven neighbouring cells describe one neighbourhood, not the class. Keep each class visually coherent (gravel, bare dirt and asphalt make a smeared average). Avoid boundary cells such as shorelines and park edges: each cell was embedded with its surroundings in view, so edge cells are genuinely mixed.
Adjust
Global settings
Explore
Find similar
Pick the Similar tool (magnifier in the bottom bar), then click any cell to see which other places look like it to the model.
Similarity ranks places against your example. It is not a probability, and one click is one example: a single cell can be a poor stand-in for its category.
Read first
About this map
- Each square is 40 m × 40 m (about a third of a city block). No individual trees or yards.
- One date only: 22 Aug 2025. Nothing about seasons or change over time.
- Similarity is not probability. A high score means "looks like the example", not "is 90% likely to be".
- Model: …, run once on Sentinel-2 imagery. Nothing runs live in your browser.
- Data: …. Each cell is described by a list of numbers from the model, and the map compares those lists.
- Licences and credits: the numbers come from Ai2's OlmoEarth model, under the OlmoEarth Artifact License. Its use restrictions apply to this data: no military, defense, surveillance or policing uses, and no extractive uses such as mining or deforestation. Imagery: contains modified Copernicus Sentinel data (2025). Font: Rubik (SIL Open Font License).
Cell information
Click the map to inspect a cell.
Classification
!
Is it settling?
One bar per recent stroke, newest on the right. Small, steady bars mean more labels of the same kind won't help much: try a different place or class. A rough rule of thumb.