The colors don't lie, but they're not obvious
Open any land-cover map and you'll see a patchwork of colors. Green, brown, gray, blue. It looks like a map in the usual sense, but the information is different. You're not seeing what's visible from space — you're seeing what the satellite's classifier decided is the dominant cover in each pixel.
This is a short guide to reading these maps without over-interpreting them.
The big four: tree cover, cropland, built-up, water
- Tree cover (dark green). Anywhere canopy exceeds 10%. A dense Amazon rainforest pixel is 95%+ canopy; a sparse savanna with scattered trees is 15% but still classified as tree cover. Don't assume "tree cover" means "dense forest."
- Cropland (orange / yellow). Plowed or planted land, whether actively in use or fallow. Edge cases: a reverted field growing weeds for a season may still show up as cropland from radar memory; recently cleared forest before planting may briefly show up as bare.
- Built-up (red / magenta). Any impervious surface, most often buildings and roads, but also parking lots, airport runways, quarries. A dense forest with a paved road running through it will have built-up pixels on the road.
- Water (blue). Permanent open water — lakes, large rivers, reservoirs. Seasonal ponds and intermittent streams end up as "wetland" or whatever the surrounding dry-season cover is.
The tricky ones: wetland, grassland, shrubland, mangrove
- Herbaceous wetland (teal). Marsh, bog, seasonally-flooded grassland. Edge cases are common because the water level changes with season.
- Grassland (pale green / khaki). Native grasses, pasture, prairie. On the continuum with cropland (intensively managed) and shrubland (woody invasion).
- Shrubland (light brown). Woody vegetation under ~5m tall. Desert margins, chaparral, recently burned forest still regrowing. Often mistaken for grassland by casual observers.
- Mangroves (dark olive). Coastal salt-tolerant forests. Classified separately because they're ecologically distinct and economically valuable (storm buffer, fish nursery).
The last three: bare, snow, moss/lichen
- Bare / sparse vegetation. Deserts, rock outcrops, active quarries, and small urban edges. A pixel with a driveway and a patchy lawn can land here.
- Snow and ice. Permanent glaciers and year-round snowpack. Seasonal snow gets reclassified as whatever's under it.
- Moss and lichen. Tundra, alpine meadows above treeline. Often next to shrubland in mountain transitions.
What to look for at the edges
Pixel boundaries between classes are where classifiers struggle most:
- Forest → grassland edges: often a zone of shrubland that gets classified either way.
- Cropland → built-up: suburban development on former fields can take a year or two to reclassify.
- Water → wetland: reservoirs with seasonal drawdown oscillate annually.
- Bare → urban: parking lots and small commercial sites sometimes flip.
If something looks "wrong," check at 10-meter pixel resolution — the classifier may be right but the physical scale may surprise you.
Reading the breakdown on a dashboard
When a tool like GeoLab gives you "40% cropland, 30% tree cover, 20% built-up, 10% water," those percentages are the fractions of pixels within the radius you queried (in our case, 5 km). That's the most useful way to think about a small area: not what one pixel is, but the mix.
Want to see it for your own spot? GeoLab has the full breakdown for any coordinate, with color-coded bars.