A global land cover GeoTIFF was recently released by Impact Observatory (IO) and Esri. To create this geospatial layer, hundreds of thousands of satellite photos were classified into ten unique land use/land cover (LULC) classes using a deep learning model in partnership with Microsoft AI for Earth.
Sentinel-2 imagery was used to divide the world into ten categories of land use cover:
Water (areas that are predominately water such as rivers ponds, lakes, and ocean)
Trees (clusters that are at least 10 meters high)
Grasslands such as open savannas, parks, and golf courses
Flooded vegetation such as wetlands, rice paddies, and
Crops
Scrubland
Built areas such as urban/suburban, highways, railways, and paved areas.
Bare ground in areas with little or no vegetation such as exposed rock/soil and sparsely vegetated deserts.
Permanent snow and ice areas
Cloud cover areas where the persistent cloud cover prevents an analysis of the underlying land cover.
The end product is a 10-meter resolution GeoTiff that the developers have released under a Creative Commons 4.0 license. The machine learning model was run on multiple dates throughout the year with the results folded into one consolidated layer to represent land use cover for the year 2020.
The 2020 Esri Land Cover dataset can be browsed using Esri’s online Map Viewer. Users can also access the full global GeoTIFF zip file or use Esri’s tool for accessing the land use data by tile.