关键词: 土地覆被, 精度评价, 混淆矩阵, 空间一致性分析, 新疆地区

Abstract:

Xinjiang region is of strategic significance to China and Central Asia. This study aimed to effectively combine different data sources and classification systems to mitigate the lack of their interoperability regarding spatial distribution of land cover data. For this purpose, three types of land cover data were included. They were the visual interpretation of land use status in 2010 remote sensing monitoring data, GlobeLand30, and GlobCover2009. Four methods including category similarity analysis, category confusion analysis, confusion matrix analysis, and spatial consistency analysis were used to evaluate their accuracies and consistencies. We expect that this study would provide recommendations for the applicability of land cover data in the arid region of northwest China. The results showed that the three types of land cover data exhibited a good consistency for describing land cover categories in Xinjiang, with similarity higher than 0.9. Particularly, bare land identification demonstrated the highest consistency, followed by grassland, cultivated land, and forest. About 95% of the land area in Xinjiang showed a relatively high consistency, and the overall accuracy for land cover data ranged from 64.11% to 72.57%. Data from the group of visual interpretation/GlobCover2009 demonstrated the lowest accuracy, followed by the group of GlobeLand30/GlobCover2009. The group of visual interpretation data/GlobeLand30 had the highest accuracy, but it still had room for improvement. These results demonstrated that using the same satellite sensor plays an integral role in enhancing the accuracy of evaluation results. Moreover, classification systems, classification methods, spatial resolution, and satellite passage time used would also have a huge impact on the accuracy of evaluation results. In order to solve this problem more effectively, multi-source remote sensing data integration technology or deep learning will become more promising for accurately interpreting remote sensing image data in the near future, for further improving data accuracy in global land cover mapping and application fields. Depending on the distinctive landscape pattern of Xinjiang region, this research analyzed the accuracy of three different kinds of data for different land cover categories to provide reliable information which shall be proved to be useful in resource development, environment protection and sustainable development of Xinjiang. Additionally, it initiated a framework for providing basic data for China’s significant development strategy "the Belt and Road". Moreover, the results demonstrated the better performance of GlobeLand30 in accuracy assessment. As compared to other land cover data within the same category, the GlobeLand30 data is overwhelming in spatial resolution.

Key words: land cover, accuracy evaluation, confusion matrix, spatial consistency analysis, Xinjiang region