GABLE: A first fine-grained 3D building model of China on a national scale from very high resolution satellite imagery

Three-dimensional (3D) building models provide horizontal and vertical information of urban development patterns, which are significant to urbanization analysis, solar energy planning, carbon reduction and sustainability. Despite that many popular products on a global or national scale are proposed, these products usually focus on building extraction and height estimation at fairly coarse resolutions while building categories are not taken into consideration. In this study, we extend the previous work in two aspects involving the introduction of semantically fine-grained categories (i.e., 12 rooftop classes) and spatially fine-grained representations of individual buildings with compact polygons. Specifically, we develop a novel framework for the generation of the 3D building models, including developing a network for the joint rooftop extraction and classification, another parallel network for the height estimation, and a post-processing algorithm for the fusion of results from the two independent networks. To train the networks and improve the generalization, we construct two custom largescale datasets in addition to the existing Urban Building Classification (UBC) dataset and 2023 IEEE Data Fusion Contest (DFC 2023) dataset. Finally, the nation-scale fine-GrAined 3D BuiLding modEl (GABLE) product is derived based on Beijing -3 satellite images (0.5-0.8 m) with our proposed framework. GABLE provides a compact rooftop polygon, a category and a height value for each individual building instance. Further analyses are conducted to uncover the distribution of buildings on a national scale in terms of diversity, height and density. These analyses demonstrate the significance and values of GALBE, while the potentials are far beyond these.