Replication Data for: Continental-scale mapping and analysis of 3D building structure

Urban land use is often characterized based on the presence of built-up land, while the land use intensity of different locations is ignored. This narrow focus is at least partially due to a lack of data on the vertical dimension of urban land. The potential of Earth observation data to fill this gap has already been shown, but this has not yet been applied at large spatial scales. This study aims to map urban 3D building structure, i.e. building footprint, height, and volume, for Europe, the US, and China using random forest models. Our models perform well, as indicated by R2 values of 0.90 for building footprint, 0.81 for building height, and 0.88 for building volume, for all three case regions combined. In our multidimensional input variables, we find that built-up density derived from the Global Urban Footprint (GUF) is the most important variable for estimating building footprint, while backscatter intensity of Synthetic Aperture Radar (SAR) is the most important variable for estimating building height. A combination of the two is essential to estimate building volume. Our analysis further highlights the heterogeneity of 3D building structure across space. Specifically, buildings in China tend to be taller on average (10.35 m) compared to Europe (7.37 m) and the US (6.69 m). At the same time, the building volume per capita in China is lowest, with 302.3 m3 per capita, while Europe and the US show estimates of 404.6 m3 and 565.4 m3, respectively. The results of this study (3D building structure data for Europe, the US, and China) are publicly available, and can be used for further analysis of urban environment, spatial planning, and land use projections.

现有城市土地利用表征往往仅以建成用地的存在性为依据，却忽略了不同区位的土地利用强度。这种研究视角的局限性，在一定程度上源于城市土地垂直维度数据的匮乏。地球观测（Earth observation）数据填补这一空白的潜力已得到证实，但目前尚未在大空间尺度上得到应用。本研究旨在借助随机森林（random forest）模型，绘制欧洲、美国与中国的城市三维建筑结构数据，具体包括建筑基底面积（building footprint）、建筑高度与建筑体量。经检验，本研究所用模型表现优异：三个研究区域合并后的决定系数（R²）分别为：建筑基底面积0.90、建筑高度0.81、建筑体量0.88。在多维度输入变量中，本研究发现：用于估算建筑基底面积的最重要变量为源自全球城市基底（Global Urban Footprint, GUF）的建成区密度，而用于估算建筑高度的最重要变量则为合成孔径雷达（Synthetic Aperture Radar, SAR）的后向散射强度。若要估算建筑体量，则需结合上述两类变量。本研究的分析还进一步揭示了三维建筑结构的空间异质性。具体而言，中国的建筑平均高度（10.35米）高于欧洲（7.37米）与美国（6.69米）。与此同时，中国人均建筑体量最低，仅为302.3立方米，欧洲与美国的人均建筑体量则分别为404.6立方米与565.4立方米。本研究生成的欧洲、美国与中国三维建筑结构数据集已公开共享，可用于城市环境、空间规划以及土地利用预测等后续研究。