塔里木大学学报

本研究基于GEE云平台提供的Landsat-8 OLI数据，以NDVI、EVI、NDWI、DEM、SLOPE作为辅助数据，利用RF、CART和SVM分类算法，对新疆的土地覆盖类型进行分类，为开展长时序土地覆盖数据产品智能快速提取提供技术支撑。绘制近些年土地覆盖图为新疆土地资源的管理和生态环境保护策略制定提供参考。研究结果表明：1)在使用相同数据源和训练样本的情况下，RF算法优于CART算法，CART算法优于SVM算法。从新疆土地覆盖分类精度看，RF分类算法总体精度最高，为96.6%,Kappa系数为0.95。SVM分类算法总体精度最低，为84.2%,Kappa系数为0.81。2)自2016年以来新疆NDVI值呈降低趋势，主要因为准噶尔盆地大量草地的退化。研究表明想要推进生态保护修复工作，须防止草地退化，加强土地荒漠化治理。

This paper is based on Landsat-8 OLI data provided by the Google Earth Engine(GEE) cloud platform, utilizing NDVI, EVI, NDWI, DEM and SLOPE as ancillary data. By utilizing of RF, CART, and SVM classification algorithms, the research classifies land cover types in Xinjiang, to provide technical support for the intelligent and rapid extraction of long-term land cover data products. The creation of land cover maps in recent years serves as a reference for the management of land resources and the formulation of ecological environment protection strategies in Xinjiang. The experimental results indicate: 1) Under the same data source and training sample conditions, the RF algorithm outperforms the CART algorithm, and the CART algorithm outperforms the SVM algorithm. Regarding to the classification accuracy of land cover in Xinjiang, the RF classification algorithm achieves the highest overall accuracy at 96.6%, with a Kappa coefficient of 0.95. The SVM classification algorithm exhibits the lowest overall accuracy at 84.2%, with a Kappa coefficient of 0.81. 2) Since 2016, NDVI in Xinjiang has shown a decreasing trend primarily due to the degradation of a substantial amount of grassland in the Junggar Basin. The paper suggests that advancing ecological protection and restoration efforts requires preventing grassland degradation and strengthening the management of land desertification.