A global reference dataset for natural and improved grasslands for 2020

This is a globally consistent, open reference dataset for grassland classification derived through expert visual interpretation using the Geo-Wiki platform. The dataset contains approximately 50,045 globally distributed sample locations interpreted using very high-resolution satellite imagery, Sentinel time series, as well as other ancillary information such as ground photographs. The interpretation protocol distinguishes between natural/semi-natural grasslands, cultivated grasslands, and other land-cover classes while also capturing information related to management intensity and livestock presence. The dataset was specifically designed to support statistically rigorous validation and area estimation for global grassland mapping products (e.g., https://zenodo.org/records/15647681). This record includes three files: Aggregated annotations at ~90m x 90m resolution; Initial annotations at ~30m x 30m resolution; Revisions at ~30m x 30m resolution A document containing annotation examples. Aggregated data set at 90m x90m contains such fields: rowid: unique row identifier; sampleid: unique validation site identifier (each validation site consists of 9 pixels); subpixel_centroid_x ,subpixel_centroid_y: coordinates of one of the pixels, in lat lon WGS84; nsubs: number of experts who annotated the location; dom1: dominant class by first expert; dom2: dominant class by second expert; rev: a value of 1 if annotation was reviewed in Geo-Wiki; a value of 2 if the feedback tool was used; a value of 0 if the annotation was not reviewed revclass: the final class assigned if the annotation was reviewed (i.e., rev is either 1 or 2). Initial data seta and Revisions (reviewed in Geo-Wiki) contain such fields : submission_id: unique submission identifier for validation site; sampleid: unique validation site identifier (each validation site consists of 9 pixels); submission_item_id: unique per pixel submission indentifier (at 30mx30m); enhancement: unqiue legend item code; name: legend item name ( grassland type); "subpixel_centroid_x" "subpixel_centroid_y": coordinates of pixel centroids, in lat lon (WGS84). The classes, their definition and guidance for visual interpretation are provided below: Grassland/land-cover class Definition Visual interpretation guidance Natural/semi-natural permanent grasslands with no visible signs of livestock These grasslands are largely unaffected by direct human management and lack visible pasture-related infrastructure such as fencing, watering points, roads, paddocks, or animal enclosures. The vegetation is typically dominated by native plant species. The key interpretation criterion for this class is the absence of visible signs of livestock grazing or management infrastructure, although traces of wild animals may still be present. This class includes grassland without visible human presence, such as tundra grasslands and moss–lichen systems in northern Canada, Europe, and Russia; seasonally flooded grasslands; steppes savanna grasslands with dominant herbaceous cover; heathlands; and natural reserves and protected grassland systems. This class excludes: Closed shrublands (>75% shrub cover); Xeric shrublands in desert environments with NDVI values lower than 0.15. Visual interpretation criteria include heterogeneous and natural vegetation patterns, limited or absent visible management, and strong correspondence between vegetation distribution and natural hydrological organization. Natural permanent woody vegetation as part of open woodlands This class represents natural open shrublands and open woodland systems dominated by low-growing woody vegetation with herbaceous understory. Shrub cover generally ranges between 40% and 75%, while tree cover remains below 10%, preventing the formation of continuous canopies. These systems maintain natural vegetation organization and show no visible signs of livestock-related management infrastructure, or any other human-made disturbances. Semi-natural permanent grasslands with visible livestock presence These grasslands are dominated by native species and are not actively reseeded, fertilized, or irrigated. However, they contain visible signs of livestock grazing and low-intensity pasture management, including fencing, watering points, livestock tracks, or grazing pressure. The key feature of this class is the coexistence of predominantly natural vegetation structure with clear evidence of grazing activity and limited farming infrastructure. This class for example includes: Low-intensity grazing systems in the Pampas region of South America; Semi-natural pastures in Europe, Africa, and other regions; Silvopastoral systems and temperate semi-natural woodlands such as the Dehesa systems of Spain and Portugal. This class excludes: Extensive rangelands lacking visible livestock infrastructure. Temporary grasslands associated with other land use This class represents grass-dominated vegetation that occurs temporarily as part of broader land-use cycles rather than permanent grassland systems. Examples include: Forestry clear cuts and post-fire recovery areas; Logging and thinning areas; Shifting cultivation systems; Fallow land between cultivation cycles for less than four years; Construction and mining sites undergoing temporary grass regeneration. Improved permanent grasslands associated with livestock These grasslands have been intentionally planted or modified using introduced species, fertilizers, irrigation, mowing, and controlled grazing to enhance productivity and maintain vegetation structure. This class can also be described as cultivated grassland with the presence of livestock. Typical indicators include: Geometrically regular field patterns; Homogeneous vegetation texture; Fencing and paddocks; Watering ponds and livestock infrastructure; Roads and access paths; Intensive mowing or grazing patterns. This class includes for example: Intensively managed pastures in Brazil, Northern Europe, Australia, and the United States; Hayfields with intensive mowing used for livestock feed production. Where regional knowledge is limited, the interpretation primarily relies on visible livestock infrastructure and management patterns. Improved permanent grasslands associated with other land use These are intentionally planted or modified grasslands not primarily associated with grazing activities. This class includes: Sports fields and golf courses; Urban parks and roadside grasslands; Residential lawns and gardens. This class excludes: Ski slopes; Hay production systems linked to livestock feeding. Natural permanent open woodlands with visible livestock presence These landscapes consist of open woodland and shrubland systems characterized by natural woody vegetation patterns combined with visible signs of livestock grazing and associated infrastructure Other land cover This class includes all remaining land-cover and land-use types not belonging to the grassland categories above. Examples include: Croplands; Forests; Wetlands and peatbogs; Urban and built-up areas; Permanent snow and ice; Water bodies; Bare ground and rocky outcrops; Mixed cropland–pasture systems. In some regions, mixed cropland–pasture systems may alternate seasonally between grazing and crop production and are therefore classified as other land cover rather than permanent grasslands. Funded by World Resources Institute.