Radiant Earth Foundation releases a global land cover classification training dataset

The open geospatial library,  LandCoverNet, is available for download at  Radiant MLHub. It will enable accurate and regular land cover mapping allowing for timely insights into natural and anthropogenic impacts on the Earth. 

Radiant Earth Foundation releases a global land cover classification training dataset
Radiant Earth Foundation releases a global land cover classification training dataset

Radiant Earth Foundation releases a global land cover classification training dataset

Radiant Earth Foundation announced the release of a human-labeled global land cover training dataset, “LandCoverNet”. The open geospatial library,  LandCoverNet, is available for download at  Radiant MLHub. It will enable accurate and regular land cover mapping allowing for timely insights into natural and anthropogenic impacts on the Earth. The release also contains data across Africa, accounting for 1/5 of the global dataset.

Labels for the multi-spectral high-quality satellite imagery from Sentinel-2 satellites, covering Africa, Asia, Australia, Europe, North America, and South America will be covered in this annual land cover classification training dataset, LandCoverNet. The annual land cover classes are labeled based on 24 scenes of Sentinel-2 for each tile throughout 2018.

Radiant Earth’s technology team selected 300 geographically diverse tiles of Sentinel-2 imagery spanning all continents to capture the diversity of land covers globally to generate the training data. Then, 30 image chips of 256 x 256 pixels at 10-meter spatial resolution were generated in each tile resulting in 9000 global chips of Sentinel-2 L2A observations. The team then built machine learning algorithms for each Sentinel-2 tile to generate a “guess land cover label”. It was then independently validated by three different individuals using Sinergise’s Classification App.

The first version of LandCoverNet, which contains image chips across Africa, provides high-quality training data for pixel-wise land cover classification and a consensus score to indicate the uncertainty in human interpretation of each class. LandCoverNet can be used by data scientists and practitioners to develop new land cover classification models or validate their own models’ accuracy. Land cover maps created with LandCoverNet can also identify underrepresented areas where more data is required.

Hamed Alemohammad, chief data scientist of Radiant Earth Foundation who leads the technology team, called LandCoverNet a benchmark training dataset, which is necessary for developing and validating accurate and scalable classification algorithms across diverse geographies. He said that their focus on Africa adds to the geodiversity of global land cover models, a feat that only leads to balanced results.

Anne Hale Miglarese, founder and CEO of Radiant Earth Foundation, said that they are incredibly grateful to Schmidt Futures for investing in this project. They believe that their investment solidified the need to diversify training data geographically, leading Radiant Earth to focus its efforts on advancing the curation and sharing of geospatial training datasets to address complex challenges like food security and climate change through Radiant MLHub.

Thomas Kalil, Chief Innovation Officer for Schmidt Futures, commented that want to congratulate the Radiant Earth Foundation for the progress that they have made in using satellite imagery and machine learning to tackle global development challenges. He added that they hope that other foundations and philanthropists will support projects like LandCoverNet that harness machine learning to achieve the Sustainable Development Goals.