Get reliable science

Land cover maps are essential tools for tracking land surface changes, informing land management, monitoring environmental conditions, and assessing conservation efforts. �Current products, however, lack essential spatial or temporal resolutions required for such efforts. �The categorical classes in a majority of these maps also fail to capture the inherent heterogeneity of the land surface. �To address the need for high spatial and temporal resolution land cover maps that capture natural landscape variation, we produced annual, 30 meter, continuous land cover maps for U.S. rangelands from years 1984 to 2015. �We used over 23,000 vegetation plots from the NRCS National Resources Inventory and BLM Landscape Monitoring Framework spanning years 2004-2014 to train and validate an assortment of machine learning regression models. ��We capitalized on a cloud-based platform for planetary-scale geospatial analysis with massive computational capabilities (Google Earth Engine) to process and store over 250 spatially contiguous gridded climatic, biotic, and abiotic spatiotemporal variables (inclusive of satellite remote sensing data) to drive the model. �The resulting yearly maps provide percent cover of annual and perennial herbaceous vegetation, shrubs, bare ground, and litter for U.S. rangelands, inclusive of the Great Plains. �The modeling framework allows for annual generation of land cover maps into the future. �These maps provide essential information for rangeland management and conservation, and will be used to create a 30 meter spatial resolution primary productivity model optimized for rangeland vegetation.