Nationwide Footprint-Level Classification of Residential Building Subtypes for the United States

This dataset provides the nationally consistent, footprint-level classification of residential building subtypes across the United States. It contains over 140 million building footprints classified into four distinct structural residential categories: Single-Family Homes (SFHs), Townhouses (including duplexes), Apartments, and Mobile Homes. The dataset is organized geographically and stored as individual, high-performance GeoPackage (.gpkg) files for each of the 50 U.S. states and the District of Columbia. Data Schema (Attribute Fields) Each state-level GeoPackage contains the following standardized attribute columns: geometry (Polygon): The spatial boundary (polygon) of the building footprint. Res_type (Integer): The classified residential structural subtype code: 1 = Single-Family Home (SFH) 2 = Townhouse (explicitly includes laterally attached row houses and semi-detached duplexes) 3 = Apartment 4 = Mobile Home height (Double): The estimated physical height of the building in meters. Baseline heights are obtained from Microsoft Global ML footprints, with a minimum threshold of 4.2 m enforced (representing a typical one-story structure) and missing values imputed using local tract-level medians. type (String): Building type tags from OSM building footprint (Limited coverage and varies by region). Methodology and Classification Workflow The residential classifications were achieved using a novel, scalable hierarchical spatial workflow that integrates geometric building footprints with high-confidence administrative and demographic auxiliary data: Multi-Source Footprint Integration: Geometry was compiled by merging Microsoft US Building Footprints, Microsoft Global ML Building Footprints, and OpenStreetMap (OSM) geometries. Non-residential structures were purged during preprocessing using OSM land-use layers and manual reclassification of OSM building tags. Mobile Home Identification: Candidate footprints located within spatial buffers around documented mobile home parks were classified using state-specific Random Forest (RF) models trained on footprint geometry (area, aspect ratio, rectangularity, and local density). High-confidence "seeds" were subsequently expanded using context-aware spatial promotion rules. Single-Family Home (SFH) Calibration: High-confidence SFHs were initially mapped using administrative address attributes. To prevent over- or under-counting, a size-ranked rule-based spatial promotion algorithm was implemented to sequentially classify candidate footprints until the counts aligned with rigorous state-level housing stock benchmarks derived from the American Community Survey (ACS) 5-year estimates. Townhouse and Apartment Discrimination: The remaining unclassified structures were categorized using a sequential, rule-based physical geometry and address-point pattern-recognition workflow. Duplexes and linear townhouses were distinguished from multi-family apartment blocks by evaluating physical constraints (convexity, height, and structural width) paired with the spatial density and linear alignment of internal address points from the National Address Database (NAD). Spatial Context Completion: A final, iterative morphological distance-matching algorithm resolved any remaining unclassified structures based on local neighborhood class composition and physical architectural similarities (height, area, and aspect ratio).