GLOBUS: GLObal Building heights for Urban Studies

Urban weather and climate studies continue to be important as extreme events cause economic loss and impact public health. Weather models seek to represent urban areas but are oversimplified due to data availability, especially building information. This paper introduces a novel Level of Detail-1 (LoD-1) building dataset derived from a Deep Neural Network (DNN) called GLObal Building heights for Urban Studies (GLOBUS). GLOBUS uses open-source datasets as predictors: Advanced Land Observation Satellite (ALOS) Digital Surface Model (DSM) normalized using Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM), Landscan population density, and building footprints. The building information from GLOBUS can be ingested in Numerical Weather Prediction (NWP) and urban energy-water balance models to study localized phenomena such as the Urban Heat Island (UHI) effect. GLOBUS has been trained and validated using the United States Geological Survey (USGS) 3DEP Light Detection and Ranging (LiDAR) data. We used data from 5 US cities for training and the model was validated over 6 cities. Performance metrics are computed at a spatial resolution of 300-meter. The Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) were 5.15 meters and 28.8 %, respectively. The standard deviation and histogram of building heights over a 300-meter grid are well represented using GLOBUS.