基于Sentinel-2数据的郑州市中心城区遥感信息提取

Remote Sensing Information Extraction of Central City of Zhengzhou Based on Sentinel-2 Data

摘要: 随着城市化进程的不断发展，城市下垫面及自然水体受到了破坏，与人类居住用地不相协调，土地的利用方式对洪涝灾害具有放大化的影响，并逐渐成为城市的隐患问题。本文以郑州市中心城区为例，采用10 m分辨率的Sentinel-2遥感数据，对获取的MNDWI指数进行预处理，然后进行阈值分割提取水体面积分布，并在此基础上，分别用支持向量机分类法、决策树分类法、随机森林分类法进行城市土地利用类型信息的提取，通过Kappa系数和混淆矩阵对比其精度。分别提取城市土地利用和水体面积信息并进行对比，对郑州市中心城区的土地利用进行研究分析，为城市的合理规划提供建议。结果显示：1) 采用阈值分割法提取水体的总体精度百分比为96.98，Kappa系数为0.94，精度较高；2) 随机森林分类法、决策树分类、支持向量机分类法的土地利用分类的Kappa系数分别为0.91、0.79、0.76，总体精度分别为93.38%、84.25%、82.38%，其中随机森林分类法的土地利用信息提取的Kappa系数及精度水平最高，准确性最高；3) 郑州市中心城区的水体面积占城市土地利用总面积的3.79%，占比较低；而城市居民用地和其他用地两类面积的占比较高，超过75%，居民建设用地和生态用地较为矛盾，城区的土地规划较不合理。采用Sentinel-2遥感数据提取郑州市中心城区的数据可以有效分析土地利用面积占比，据此可以为郑州市的土地利用规划提供数据支撑，协调居住与自然用地的比例，有效改善下垫面对地表水资源的调节作用，从而减少城市内涝的发生，不断推进郑州生态宜居型城市的建设。

Abstract: With the continuous development of urbanization, urban underlying surface and natural water have been destroyed, which is not compatible with human residential land. Land use has a large impact on flood disaster, and gradually become a hidden problem in the city. In this paper, the central city of Zhengzhou is taken as an example, 10 m resolution Sentinel-2 remote sensing data is used to preprocess the obtained MNDWI index, and then threshold segmentation is carried out to extract water area distribution. On this basis, support vector machine classification, decision tree classification, and random forest classification are respectively used to extract urban land use type information. The accuracy was compared by the Kappa coefficient and confusion matrix. The information on urban land use and water body area is extracted and compared respectively, and the land use of the central urban area of Zhengzhou is studied and analyzed to provide suggestions for the rational planning of the city. The results showed that: 1) The overall accuracy percentage of water extraction by threshold segmentation method was 96.98 and the Kappa coefficient was 0.94, indicating high accuracy. 2) The Kappa coefficient of land use classification of random forest classification, decision tree classification, and support vector machine classification were 0.91, 0.79, and 0.76 respectively, and the overall accuracy was 93.38%, 84.25%, and 82.38% respectively. Among them, the Kappa coefficient and accuracy level of land use information extraction of random forest classification were the highest, indicating high accuracy. 3) The water body area in the central city of Zhengzhou accounts for 3.79% of the total land use area, which is relatively low; however, urban residential land and other land areas account for a relatively high proportion, more than 75%. Residential construction land and ecological land are contradictory, and urban land planning is not reasonable. Using Sentinel-2 remote sensing data to extract the data of the central urban area of Zhengzhou can effectively analyze the proportion of land use area, which can provide data support for the land use planning of Zhengzhou, coordinate the proportion of residential and natural land, effectively improve the regulation effect of underlying surface water resources, and thus reduce the occurrence of urban waterlogging and continuously promote the construction of an ecologically livable city in Zhengzhou.