Ye Tao, Shan Hongyu, Wu Jidong, Zhou Qiang, Ma Mingfu, Zhao Wenzhi, Ya Ru, Gao Yuan, Wu Lizheng
State Key Laboratory of Earth Surface Processes and Disaster Risk Reduction, Beijing Normal University, Beijing, 100875, China.
Key Laboratory of Environmental Change and Natural Disasters, Ministry of Education, Beijing Normal University, Beijing, 100875, China.
Sci Data. 2025 Jun 17;12(1):1013. doi: 10.1038/s41597-025-05266-4.
Large-scale high-precision building distribution data is important fundation for regional urban planning and resource allocation and disaster risk research. The Qinghai-Tibetan Plateau is the third pole of the world. Although understanding local human-environment interactions in the Qinghai-Tibetan Plateau is critically important, this has been hindered by a lack of high-resolution building footprint data due to the vastness and remoteness of the area. In this study, we generated the first vectorized building rooftop prints of the Qinghai-Tibetan Plateau and its surrounding areas by using high-resolution Google imagery and the building contour extraction algorithm of the AI Earth platform. Our results include 13.09 million buildings covering 6092.7 km, validated with a total of 250 × 1 km test samples. The data had an overall accuracy of 87%, a recall of 91.9%, and an F1 score of 64.8%, thus providing an advanced description of the building distribution of the study area as compared to CBRA. Our work has immense potential in facilitating exposure assessment for studies on disaster risk in this area.
大规模高精度建筑分布数据是区域城市规划、资源配置和灾害风险研究的重要基础。青藏高原是世界第三极。尽管了解青藏高原当地的人类与环境相互作用至关重要，但由于该地区地域辽阔且地处偏远，缺乏高分辨率的建筑足迹数据，这一工作受到了阻碍。在本研究中，我们利用高分辨率谷歌影像和人工智能地球平台的建筑轮廓提取算法，生成了青藏高原及其周边地区的首批矢量化建筑屋顶图。我们的结果包括1309万栋建筑，覆盖面积达6092.7平方公里，并通过总共250个1公里的测试样本进行了验证。数据的总体准确率为87%，召回率为91.9%，F1分数为64.8%，因此与CBRA相比，对研究区域的建筑分布提供了更详细的描述。我们的工作在促进该地区灾害风险研究的暴露评估方面具有巨大潜力。
