1.MNR Key Laboratory for Geo-Environmental Monitoring of Great Bay Area;2.Guangdong Key Laboratory of Urban Informatics;3.School of Architecture and Urban Planning, Shenzhen University
作为反映城市生态环境的重要指标，城市生态宜居性受到学术界越来越多的关注。首先结合遥感与POI数据生成分类特征，利用文档主题生成模型进行特征重表达，采用随机森林模型提取住区，为后续的生态宜居性评价提供基本单元；随后，依据多源数据生成多个生态宜居性评价指标，使用基于熵权的TOPSIS方法构建生态宜居评价模型，得出每个住区综合得分；以深圳市为实验区对本研究提出的方法进行了验证，并对深圳市的住区生态宜居性进行了空间分析。结果表明：(1) 机器学习方法与多源数据结合可得到精细的城市功能分区图，总体精度可达82.1%；(2) 基于TOPSIS方法构建的生态宜居评价框架能够对住区生态宜居性进行有效量化，综合得分高的住区多为片区绿化率高、空气质量好、建筑密度较小的住宅区，而得分较低的住区主要集中在城中村等生态环境较差的区域，结果符合客观事实；(3) 深圳市内的住区及其评价得分呈现出明显的空间分异，南山、福田、罗湖区域的住宅小区较多；宝安、龙华、龙岗区内部同时包含较多的住宅小区与城中村，导致三个区内部各评价单元的得分差异较为明显；因住区数量少、绿地面积大、空气质量高等客观条件，坪山、盐田和大鹏区生态宜居得分情况较好。以上结果展示了本研究方法的有效性，可为城市生态宜居建设及城市规划等提供案例参考和数据支撑。
As an important indicator reflecting urban ecological environment, urban ecological livability has aroused academia concerns. However, existing studies mainly focused on city-scale or region-scale ecological livability evaluation. With the increasing demands for fine-scale urban planning and management, small-scale livability evaluation is urgently needed, however, it is still conceptual and qualitative. In recent years, the availability of abundant, fine-grained, and multi-source data in the big data era laid the foundation for comprehensive and quantitative livability evaluation at much finer scales. In this study, we propose a framework to evaluate fine-scale livability using multi-source data. Firstly, we combined remotely sensed imageries and Point-of-interests (POI) data to generate multiple features. Subsequently, these features were represented via Latent Dirichlet Allocation (LDA) model. Random Forest model was then utilized to extract dwelling areas, namely, the basic evaluating units. Next, TOPSIS model was employed for evaluating ecological livability, by integrating indicators derived from the multi-source data. We also validated the developed methods in Shenzhen City. The results show that: (1) the accurate urban functional map was obtained by combining multisource data and machine learning methods with overall accuracy of 82.1%. Most of the residential zones and urban villages in Shenzhen City were distributed within the central and southern urban areas. (2) The TOPSIS can be used for quantitative evaluation of the ecological livability of the dwelling areas. Most of the dwelling areas with high scores had high proportion of green space, good air quality and low building density. while the opposite sides were mainly concentrated in urban villages. The score of urban ecological livability is consistent with actual conditions. The results proved the effectiveness of the five indicators in evaluating livability of Shenzhen City. However, the thermal comfort indicator which was reflected by Land Surface Temperature had a weak influence on livability score in our study. (3) The spatial distribution of dwelling areas is quite different over different district, as well as their livability scores. Nanshan, Futian and Luohu districts have more residential areas than others. Some districts (such as Baoan, Longhua and Longgang) have contrasting scores, because they have both many high-score residential areas and low-score urban villages. While Pingshan, Yantian and Dapeng districts receive higher ecological livability scores than others, due to the limited dwelling areas, more green space, and better air quality. The experimental results of this case study demonstrated the proposed study framework is effective and reasonable. These results can not only provide theoretical support for improving urban living environment, but also provide technical supports for the ecological livability renovation projects and urban planning.
董轩妍,胡忠文,吴金婧,王敬哲,杨超,张杰,夏吉喆,邬国锋.基于多源数据的住区生态宜居性评价——以深圳市为例.生态学报,,(). http://dx. doi. org/10.5846/stxb202104251087复制