This study introduces a novel dataset of parking lot boundaries covering fifteen US cities. We generate this dataset using a deep learning segmentation model described in Qiam et al. (2025), and a subsequent post-processing workflow. The dataset, publicly available in shapefile format, enables spatial analysis of parking land use at both inter- and intra-city levels. To estimate the share of off-street land used for off-street parking, we link these polygons with tax parcel datasets, in order to exclude streets and public sidewalks. Off-street surface parking accounts for as little as 3.4% of parcel land in Oakland and as much as 10.7% in Anaheim, with central business districts ranging from 2.3% in Boston to 31.7% in Tulsa.
Qiam et al. (Wed,) studied this question.