Comparative of resident-information extraction accuracy from multi-scale remote sensing data in karst mountainous areas

Guizhou Province hosts a typical karst landform in China, where nearly 65% of the land is karst regions. It leads to the complexity, vulnerability and diversity of Guizhou’s ecological environment because of its unique mountainous plateau terrain and the existence of karst geological conditions. The spatial distribution of the karst mountainous areas is quite different from that of other plateaus mountainous areas, which seriously affects the survival and development of the mountainous residents. Therefore, studying the satellite images with suitable resolution in the karst areas to obtain the information of the residents is very significant. In this paper, the Qinglong county is selected as the study area. The county is under the jurisdiction of Qianxi’nan Prefecture, Guizhou Province. It is located in the middle of the Yunnan-Guizhou plateau and the west of the Miaoling mountains. The main karst forms include peak-cluster depression, sinkholes, dry gutters and karst caves in Qinglong. The author selected four different resolution images as the data source, which are Y3 image (2.1 m), sentinel-2 image(10 m), GF1 image(16 m) and Landsat8 image (30 m), respectively and extracted the residential area, township and rural regions of the study area using the object-oriented classification and human-computer interaction methods. Furthermore, the accuracy of the results that three kinds of residential information extracted from the four different resolution were compared with the geographical census data obtained from the 0.5 m resolution Pleiades image through visual interpretation and field verification. The results show that, (1) In the same resolution image, the extraction error of the urban residents is smaller than that of the rural residents. Among the different resolution images, the accuracy change of the resolution of urban residential areas is the smallest, which is 23.99%, whereas the accuracy change of the rural residential areas is the biggest, up to 35.3%; (2) The total commission errors of residential information increases rapidly from 2.1 m to 30 m resolution images, which are, in order, 2.56%, 15.58%,24.50%, 32.72%. The total omission errors in the urban residential areas are obviously smaller than that in the rural residential areas, and the most wrongly type divided into other classes in three residential areas is bare lands; (3) With the reduction of image resolution, the total omission errors of residential areas are 2.86%, 18.60%, 27.99%, 37.49%, respectively, among which non-central rural residential areas are more vulnerable to the environmental impact from their surroundings, and the total omission errors of residents increase significantly with the reduction of image resolution. The final results obtained are consistent with the conclusion that the spatial resolution of the satellite sensor is proportional to the spatial resolution of the satellite sensor.