Evaluating The potential of multi-source remote sensing and urban big data for urban land use classification

Accurate urban land use classification is critical for urban planning but is often hindered by spectral and morphological similarities inherent in single-source remote sensing imagery. This study provides a systematic evaluation of how the integration of multi-source remote sensing imagery and urban big data enhances the accuracy of urban land use classification. We utilized GF-1 high-resolution (HR), Sentinel-1 SAR, Sentinel-2 multispectral, and Luojia-1 nighttime light (NTL) imagery, along with Points of Interest (POI), Tencent location-based data (LocBigData), and electricity consumption records. An object-based random forest model was employed to classify 12 urban land use categories. The data fusion approach achieved an overall accuracy of 92.17% and a Kappa coefficient of 0.91, significantly outperforming the GF-1 baseline (82.76%, Kappa 0.80). Each data source made distinct contributions: Sentinel-2 enhanced grassland discrimination through red-edge spectral bands; Sentinel-1 SAR effectively distinguished water bodies from buildings due to surface roughness sensitivity; LocBigData captured spatiotemporal population patterns to distinguish commercial and residential areas; POI data provided semantic context to identify residential, educational, and industrial zones; and electricity consumption patterns highlighted commercial activity. These findings demonstrate the complementary strengths of diverse data sources and provide a foundation for optimizing data fusion strategies in urban land use mapping.