GeospatialGeoLab

What Is ESA WorldCover? The Free 10-Meter Map of Earth

How the European Space Agency's global land-cover map works, what the 11 classes mean, and why it's the gold standard for open earth-observation data in 2026.

Ryan Bethencourt
April 19, 2026
7 min read

The open map that changed earth observation

Before 2021, if you wanted a global land-cover map at better than 30-meter resolution, you needed a commercial license and a non-trivial budget. ESA WorldCover changed that.

Released in 2022 (with a 2021 annual snapshot) and updated annually, WorldCover is a free, global, 10-meter-resolution land-cover classification derived from Sentinel-1 (radar) and Sentinel-2 (optical) satellite imagery. It's the foundation that most open geospatial tools — including ours — use to answer "what's really at this pin?"

Tip
Short version: 10-meter global pixels, 11 land-cover classes, CC-BY 4.0, free to download, commercial-use permitted. The default answer when someone asks "what is this place classified as?" in 2026.

The 11 classes

  • Tree cover (code 10) — forests and woodlands >10% canopy cover.
  • Shrubland (20) — woody vegetation <5 m tall.
  • Grassland (30) — natural or semi-natural grasses.
  • Cropland (40) — cultivated or plowed land.
  • Built-up (50) — urban, roads, buildings.
  • Bare / sparse vegetation (60) — deserts, rock, <15% vegetation.
  • Snow and ice (70) — permanent ice, glaciers.
  • Permanent water bodies (80) — lakes, reservoirs, large rivers.
  • Herbaceous wetland (90) — marsh, seasonally flooded grasslands.
  • Mangroves (95) — coastal salt-tolerant forest.
  • Moss and lichen (100) — tundra, high-altitude ground cover.

Why 10 meters matters

At 10-meter resolution, one pixel is about the size of a backyard — ~100 square meters. That lets you see:

  • Individual buildings in urban areas.
  • Narrow creeks and waterways.
  • Field boundaries in agricultural areas.
  • Thin strips of riparian forest along rivers.

At 30 m (the old standard from Landsat), these get blurred together. At 10 m, a backyard garden shows up; a roadside forest buffer shows up; a reservoir's outline is crisp.

How it's made

The WorldCover team combined Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 multispectral imagery — both free Copernicus satellites. They ran a gradient-boosted tree classifier trained on ~2 million human-labeled pixels worldwide, then applied temporal consistency filters to stabilize the output.

The radar component is critical: Sentinel-1 sees through clouds, so tropical regions (normally cloud-covered year-round in optical imagery) get classified from radar signals sensitive to vegetation height and roughness.

Who uses it

  • Environmental monitoring: WWF, Global Forest Watch, Carbon Plan.
  • Insurance underwriters: flood exposure and wildfire risk assessment.
  • Agtech companies: first-pass cropland extent before running per-field models.
  • Urban planners: measuring city growth over annual time steps.
  • Journalists: cross-referencing claims in environmental reporting.

Try it on any spot

GeoLab pulls WorldCover data for any latitude/longitude and gives you the breakdown in plain English — no GIS software needed. Free.

Frequently Asked Questions

Is WorldCover free to use commercially?

Yes. The ESA WorldCover products are released under CC-BY 4.0, which means you can use them in any application — commercial or research — as long as you credit ESA / Copernicus. That's why it's the default land-cover source for most open geospatial tools.

How does it compare to the USGS National Land Cover Database (NLCD)?

NLCD is higher resolution for the US (30 m) and more specific to US ecoregions, with finer-grained classes. WorldCover is 10 m resolution, global, and uses 11 broadly consistent classes. For global or international work, use WorldCover. For US-only federal reporting work, use NLCD.

How often is it updated?

WorldCover has released 2020 and 2021 annual maps. A 2023 version is expected in 2025. For weekly change tracking you need Sentinel-2 time-series, which WorldCover is derived from.

What's the accuracy?

Globally, ESA reports ~74-77% overall accuracy against independent ground-truth. Class-level accuracy varies: water is >95%, built-up is 85%+, crop/grass boundaries are harder (65-75%). Urban edges and wetland-to-forest transitions are the known weak spots.

Can I download the full map?

Yes — ESA hosts the full global map as Cloud-Optimized GeoTIFFs on AWS Open Data. About 80 GB for the full 10 m global product. For most applications you just need a small area around your coordinates, which tools like ours pull on demand.

Run this yourself — no GPU, no install

Free for researchers. Pick a tool, paste your input, see results in seconds.