Evolving Cityscape: A Dataset for Building Footprints and Heights from Satellite Imagery in China

Sebastiano Papini, Susie Xi Rao, Peter Egger

Abstract: The study of cities faces a core challenge: the absence of data that are simultaneously high-resolution, large-scale, and longitudinal. Only combining these three aspects reveals detailed (almost building-level) changes while covering vast urban areas consistently over time and promises advancing our understanding of the driving mechanisms of spatial agglomeration. We present a novel approach that leverages computer-vision techniques on Sentinel satellite imagery to generate detailed building-volume data throughout 106 cities in China over a six-year period (2018-2023). We validate the model by assessing building-volume density in out-of-sample cities. Additionally, we compare our results to nightlight-luminosity data, a frequently utilized remote-sensing resource for tracking density and human activity, and demonstrate how the proposed method and data drastically improve the measurement of urban density. The proposed method provides researchers in the social sciences at large with access to large and exponentially growing archives of customary daylight-satellite imagery either through direct use of the provided dataset or through adaptation of the model with new data.