Deep learning–based mapping and spatial patterns of sustainable roofs in high-density urban CBDs: Evidence from Guangzhou and Shenzhen, China

Sustainable roofs, including green roofs (GR) and photovoltaic (PV) roofs, are increasingly used as essential components of urban green infrastructure and building-scale renewable energy systems that support climate resilience and environmental quality in high-density cities. However, large-scale, spatially explicit analyses of sustainable roofs in urban core areas remain limited due to data scarcity and the difficulty of reliably distinguishing roof types. Recent advances in deep learning (DL)-based remote sensing have enabled automatic mapping of sustainable roofs at the city scale, but empirical applications remain scarce in Chinese megacities, and systematic comparisons within and across cities are still rare. To address this gap, we adopted a DL-based framework for sustainable roof identification and applied it to eight representative central business districts (CBDs) in two major cities (Guangzhou and Shenzhen, China). High-resolution satellite imagery was used to automatically detect GR and PV roofs, and spatial statistical analyses were conducted to examine their distribution patterns, compositional characteristics, and differences both within and between cities. The results reveal significant variations in the spatial configuration and composition of sustainable roofs across CBDs, reflecting disparities in development intensity, functional structure, and architectural form. This study highlights intra- and inter-city differences in sustainable roof deployment in high-density urban cores and provides empirical evidence to support context-appropriate planning and implementation strategies for sustainable roofs amid rapid urbanization.