Description
This dataset classifies the land cover class present on the land surface at 10 m spatial resolution, with global coverage annually from 2017 onwards. Pixels are assigned one of 9 land cover classes.
Usage notes
Data are delivered as annual composites of predictions made on multiple Sentinel-2 observations.
Licence
Open
Version
1.2
Release date
Made to order
CRS
EPSG:3857
Update schedule
January for previous year
Provider cost
$0.00/ha
Description
Class value provides the index of the predicted land cover class for a pixel. Predictions are made using a deep learning segmentation model trained to predict > 5 billion manually annotated land cover classifications from Sentinel-2 imagery.
Usage notes
Grass and Shrub classes are merged into a single Rangeland class (value 11). Overall accuracy is 76% compared to a majority consensus. Lowest accuracy is for Flooded Vegetation (57%) and Rangeland (58%). Highest accuracy is for Crops (90%) and Water (87%).
Variable category
Land cover class
Variable type
Integer
Metadata & resources