Potential distribution of land cover classes (Potential Natural Vegetation) at 250 m spatial resolution

Potential distribution of land cover classes (Potential Natural Vegetation) at 250 m spatial resolution based on a compilation of data sets (Biome6000k, Geo-Wiki, LandPKS, mangroves soil database, and from various literature sources; total of about 65,000 training points). We used a comparable thematic legend used to produce the Dynamic Land Cover 100m: Version 2. Copernicus Global Land Operations product (Buchhorn et al. 2019), which is based on the UN FAO Land Cover Classification System (LCCS), so that users can compare actual (https://lcviewer.vito.be/) vs potential (this data set) land cover. Two classes not available in the LCCS were added: "subtropical/tropical mangrove vegetation" and "sub-polar or polar barren-lichen-moss, grassland". The map was created using relief and climate variables representing conditions the climate for the last 20+ years and predicted at 250 m globally using an Ensemble Machine Learning approach as implemented in the mlr package for R. Processing steps are described in detail <strong>here</strong>. Maps with "_sd_" contain estimated model errors per class. Antarctica is not included. Produced for the needs of the <strong>NatureMap</strong> which is project run by the <strong>International Institute for Applied Systems Analysis</strong> (IIASA), the <strong>International Institute for Sustainability</strong> (IIS), the <strong>UN Environment Programme World Conservation Monitoring Centre</strong> (UNEP-WCMC), and the <strong>UN Sustainable Development Solutions Network</strong> (SDSN). NatureMap is funded by Norway’s International Climate Initiative (NICFI). Maps will also be made available via: OpenLandMap.org. These are initial predictions for testing purposes only. A publication explaining all processing steps is pending. If you discover a bug, artifact or inconsistency in the predictions, or if you have a question please use some of the following channels: Technical issues and questions about the code: https://github.com/Envirometrix/PNVmaps/issues All files internally compressed using "COMPRESS=DEFLATE" creation option in GDAL. File naming convention: pnv = theme: potential natural vegetation, potential.landcover = variable: potential land cover type (e.g. "open forest, evergreen needleleaf"), probav.lc100 = classification model: ProbaV-based land cover mapping legend (LCCS), c = factor, 250m = spatial resolution / block support: 250 m, b0..0cm = vertical reference: land surface, 2000..2017 = time reference: period 2000-2017, v0.1 = version number: 0.1,