LAND COVER CLASSIFICATION RESEARCH AT SCALE OF 1:50000 IN SANJIANGYUAN STUDY AREA BASED ON SPOT-5 IMAGES

摘要

1:50000 land cover classification and mapping program is an important part of the mapping project in western China, which is organized by Chinese Slate Bureau of Surveying and Mapping Organization and will finish 5032 maps at scale of 1:50000 until 2010. With an average altitude of 4400 m, Sanjiangyuan mapping district is located at the center of the Qinghai-Tibet Plateau. Because of complicated terrain and inconvenient traffic, field reconnaissance is difficult. Obviously, land mapping in Sanjiangyuan district is suitable for high-resolution remote sensing image and field survey. Therefore, it is essential to choose typical area and carry out the 1:50000 land cover classification study. At first, the original SPOT-5 images are corrected as high-precision ortho-rectification images, which are fused as 2.5 m resolution multispectral image; Secondly, the fused image is interpreted and the draft of 1:50000 land cover classification map is mapped by using computer software; Then, field surveys are carried out according to the draft; Finally, 1:50000 land cover thematic map is mapped over the repeated examination and revision, which are based on the field surveys combining with other related data in study area. The objective of the 1:50000 land cover classification experiment in the Sanjiangyuan study area is to establish typical interpretation sample image sets and provide necessary remote sensing classification guidance and technical support for a large area of Sanjiangyuan mapping district. Based on the sample images and field surveys conducted by the sample images, the identified sampling image sets are promoted to the entire Sanjiangyuan mapping district. According to Western Land Cover Classification System, the experiment establishes 106 representative sample images of 23 typical land cover types and interprets 266 plots based on SPOT-5 image in the area of a 1:50000 map. Especially, the smallest plot in the map is 1 mm2; the corresponding plot on the reality ground is 10000 m2, as much as the size of a football field. Through several experiments in Sanjiangyuan mapping district, the results show that the land cover classification map can achieve 80% overall classification accuracy. This has greatly narrowed the field survey time and workload, which is especially important to western mapping project in the extreme hard conditions. By experimenting in the typical Sanjiangyuan study areas, the results show it is feasible that using SPOT-5 remote sensing images and a spot of field surveys to implement 1:50000 land cover classification and mapping.