Fusing Land Use Data and Population Density Estimates for High Resolution Population Modeling: LandScan HD

ORNL's Population Density Tables (PDT) project models population density for day and night scenarios across 50+ specific land use/facility types. The LandScan HD project in turn spatializes these density estimates by linking them to specific land use and facility polygons acquired from a variety of sources, and a flexible data fusion framework has been established to merge relevant land use layers together in any given country. The spatialized densities can then be combined with a building layer, generated via a deep learning model, also developed at ORNL, in order to generate gridded population estimates for daytime and nighttime at national level. These estimates are made at 3-arcsecond (approx. 90 m) spatial resolution. Using Yemen as a case study, we mapped PDT densities to vector land use layers (in this case, Built-up Terrain Zones (BTZ) data from the Army Geospatial Center (AGC) and a curated subset of polygons from OpenStreetMap). The spatialized densities, the building layer, and national census estimates were combined to create our final 3-arcsecond countrywide population raster.