Mapping urban villages in Beijing-Tianjin-Hebei regions using Sentinel-2, SDGSAT-1 glimmer imagery, and individual building footprint

Accelerated urbanization has led to the widespread emergence of informal urban settlements (IUSs), particularly urban villages (UVs), posing significant challenges to social equity, public welfare, and sustainable urban development. Current research on UV identification often relies solely on highresolution satellite imagery, neglecting the unique spatial patterns and socio-economic characteristics of UVs, thus limiting the effective distinction between UVs and other confusing landuse types. Additionally, the high acquisition cost of highresolution imagery restricts its feasibility for large-scale UV mapping. To address these issues, this study proposes a dualstage “candidate-classification” method integrating multi-source remote sensing and geographic data. First, individual building footprints (IBFs) are analyzed using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm to identify potential UV candidate areas based on spatial clustering characteristics. Subsequently, spatial structural and socio-economic features are incorporated into a random forest (RF) classification model for accurate UV identification. Experiments conducted in the core cities of the Beijing-Tianjin-Hebei region (Beijing, Tianjin, and Shijiazhuang) demonstrate excellent classification performance, achieving Overall Accuracy and F1 scores of 91.65% and 90.21%, respectively. Furthermore, gradually incorporating socio-economic and building density features improves the average OA by 2.56% and 5.33%, respectively, confirming their significant contribution to UV identification accuracy. The proposed method effectively captures multi-dimensional UV characteristics, enabling accurate identification of urban villages, and shows strong potential for nationwide application in UV mapping.