I processed the ESA WorldCover 2021 dataset — a global 10 m land cover map derived from Sentinel-1 and Sentinel-2 — into PMTiles archives using geotiff2pmtiles. The tilesets can be served from any static file host, no tile server required. The dataset is useful for painting a blank OpenStreetMap layer. And it’s pretty to look at.
Land Cover Classification#
Open the land cover map in full screen
The tiles are grayscale PNGs (512 px) where each pixel value maps to a land cover class, colored on the fly by the viewer:
| Value | Color | Class |
|---|---|---|
| 10 | Tree cover | |
| 20 | Shrubland | |
| 30 | Grassland | |
| 40 | Cropland | |
| 50 | Built-up | |
| 60 | Bare / sparse vegetation | |
| 70 | Snow and ice | |
| 80 | Permanent water bodies | |
| 90 | Herbaceous wetland | |
| 95 | Mangroves | |
| 100 | Moss and lichen |
This map uses mode resampling to avoid artefacts, prevent color bleed, and avoid falsifying any data. Nearest-neighbor would have worked as well, but mode makes the higher zoom levels more believable. Ocean tiles were filled in using pmtransform, which replaces transparent tiles with ocean, since the source data only extends about 10 km from the coast. Antarctica is not covered.
Satellite RGB Composite#
Open the satellite RGB composite in full screen
WebP tiles (512 px, zoom 0–13) covering -60° to 83° latitude. Quality suffers in mountainous and cloudy regions where clouds were stripped from the source data, leaving black pixels.
Both tilesets were generated with geotiff2pmtiles. The land cover archive is 274 GiB; the RGB composite is 258 GiB.