Li Ziming, Chen Bin, Huang Yufei, Wang Han, Wang Yadian, Yuan Yiming, Li Xuecao, Chen Jing M, Xu Bing, Gong Peng
Future Urbanity & Sustainable Environment (FUSE) Lab, Division of Landscape Architecture, Faculty of Architecture, The University of Hong Kong, Hong Kong 999077, China.
Future Urbanity & Sustainable Environment (FUSE) Lab, Division of Landscape Architecture, Faculty of Architecture, The University of Hong Kong, Hong Kong 999077, China; Urban Systems Institute, The University of Hong Kong, Hong Kong 999077, China; Institute for Climate and Carbon Neutrality, The University of Hong Kong, Hong Kong 999077, China.
Sci Bull (Beijing). 2025 Sep 30;70(18):3029-3041. doi: 10.1016/j.scib.2025.07.006. Epub 2025 Jul 9.
Accurate, detailed, and up-to-date urban land use information plays a key role in understanding the urban environment, enhancing urban planning, and promoting sustainable urban development. Recent advancements have focused on refining urban land use classification methods and generating data products at various scales. However, detailed parcel-level urban land use mapping across China remains insufficient with low accuracy. To address this issue, we propose an enhanced mapping framework of essential urban land use categories by integrating multi-modal deep learning models and multi-source geospatial data. Utilizing complete, accurate land parcels derived from the combined OpenStreetMap and Tianditu road networks as the smallest classification units, we have developed an enhanced Essential Urban Land Use Categories (EULUC) map covering all cities in China for 2022, termed EULUC-China 2.0. The mapping results show that residential, industrial, and park and greenspace are the dominant land use categories, collectively accounting for nearly 78% of the urban area. Compared to its predecessor, EULUC-China 1.0, the updated 2.0 version offers more detailed, spatially explicit information that reveals distinct spatial patterns within diverse land use compositions of each city. Our evaluation demonstrates that the overall accuracies of Level-I and Level-II classification reach up to 79 % and 72 %, respectively, representing substantial enhancements across all categories over the previous product. These improvements are primarily attributed to the effectiveness of deep learning in processing multi-modal inputs, particularly through the graph modeling of Point-of-interest (POI) data. The publicly accessible product (https://zenodo.org/records/15180905) and the insights derived from this study offer a valuable dataset and references for researchers and practitioners addressing critical challenges in urbanization.
准确、详细且最新的城市土地利用信息对于理解城市环境、加强城市规划以及促进城市可持续发展起着关键作用。近期的进展主要集中在完善城市土地利用分类方法以及生成不同尺度的数据产品。然而，中国全国范围内详细的地块级城市土地利用制图仍然不足，精度较低。为解决这一问题，我们提出了一个通过整合多模态深度学习模型和多源地理空间数据来增强基本城市土地利用类别制图的框架。利用结合了开放街道地图（OpenStreetMap）和天地图道路网络得出的完整、准确的地块作为最小分类单元，我们绘制了2022年覆盖中国所有城市的增强版基本城市土地利用类别（EULUC）地图，即EULUC-China 2.0。制图结果显示，住宅、工业以及公园和绿地是主要的土地利用类别，总计占城市面积近78%。与前身EULUC-China 1.0相比，更新后的2.0版本提供了更详细、空间明确的信息，揭示了每个城市不同土地利用构成中的独特空间模式。我们的评估表明，一级和二级分类的总体准确率分别达到79%和72%，相较于之前的产品，所有类别都有显著提高。这些改进主要归因于深度学习在处理多模态输入方面的有效性，特别是通过兴趣点（POI）数据的图建模。这个可公开获取的产品（https://zenodo.org/records/15180905）以及本研究得出的见解为研究人员和从业者应对城市化中的关键挑战提供了有价值的数据集和参考。
