在气候变化和人类频繁活动等因素的共同驱动下，灌木在全球近一半的干旱半干旱草原区域扩张，表现为灌木覆盖面积和生物量逐渐增加的过程。灌木扩张过程改变了原有生态系统的结构和功能，重分配了资源环境的空间格局，给畜牧业经济和草原生态带来多方面的影响。因此在区域尺度研究干旱半干旱草原灌木的分布及其动态，是研究灌木扩张速度和方向、分析和理解灌木扩张的机制、评估其带来的经济生态影响的重要前提和基础。

内蒙古大草原面积辽阔且类型丰富，是我国重要的畜牧产品输出地和生态安全屏障。但近年来，以锦鸡儿属为主的灌木在草原大面积扩张。围绕内蒙古灌木扩张，相关研究大多数基于文献综述和野外实验，缺乏大范围的灌木盖度分布数据，因此限制了对区域灌木扩张机制和影响的理解。面对现势需求，本论文利用哨兵二号（Sentinel-2）遥感数据生成的植被指数时间序列，对内蒙古锡林郭勒草原灌木盖度遥感制图进行了研究。主要方法包括：首先，考虑到该地区灌木类型分布的规律，利用降水数据将研究区域分为西区、中区和东区；之后，逐区域对比干旱年和湿润年、不同类型植被指数、不同机器学习算法的精度，并且选择出重要时间段；最后，对比分区与不分区、完整时间序列与重要时期两两组合的4种制图策略的精度，生成了全区域的灌木盖度分布制图数据。主要结论如下：

（1）干旱年的灌木盖度提取精度要高于湿润年，随机森林的表现要优于支持向量机、偏最小二乘回归和高斯过程回归。

（2）三个子区域的最佳植被指数和重要时间段存在差异。西区的最佳指数为NDTI（Normalized Difference Tillage Index）和NDGI（Normalized Difference Greenness Index），而中区和东区的最佳指数为NDTI、水分指数和背景调节型指数。三个子区域的重要时间段主要集中在5-7月和9-10月，且它们的夏季重要时间段呈现从西到东提前的特征。

（3）制图策略方面，分区的制图精度高于不分区，而植被指数重要时期制图比完整时间序列在低值预测方面表现更佳。

（4）最终利用“分区+植被指数重要时期”制图策略生成了全区灌木盖度分布制图数据，RMSE为0.055，总体精度为80%，召回率为95%，灌木盖度在空间上从西到东逐渐增加。

本研究对影响灌木盖度遥感提取的关键因素进行了综合分析，首次完成了内蒙古锡林郭勒草原区域的高空间分辨率灌木盖度分布遥感制图数据，为该地区未来灌木动态监测和其他相关研究提供了重要参考。

Driven by multiple factors such as global climate change and human activities, shrubs have expanded in nearly half of the arid and semi-arid grasslands in the world, showing a gradual increase in shrub coverage and biomass. The process of shrub expansion changed the structure and function of the original ecosystem, redistributed the spatial patterns of resources and environment, and brought many impacts on animal husbandry economy and grasslands ecology. Therefore, studying the distribution and dynamics of shrub expansion in arid and semi-arid grasslands at regional scale is an important premise and basis of studies related to shrub expansion, including studying its speed and direction, analyzing and understanding its mechanism, and evaluating its economic and ecological impacts.

Inner Mongolia grassland is the main economic base of animal husbandry of China, and it is also an important ecological safety barrier in the north of China. However, in recent years, shrubs dominated byCaraganahave expanded in a large area in Inner Mongolia grassland. The related researches on shrub expansion in Inner Mongolia are mainly based on literature review and field experiments because of the lack of large-scale shrub coverage distribution data, which limits the researches on the mechanism and impacts of regional shrub expansion. Therefore, this study aims to map fractional shrub cover at regional scale across the Xilin Gol steppe in Inner Mongolia by using the time series of vegetation index generated by Sentinel-2 remote sensing data. Firstly, considering the distribution of shrub types in the region, we divided the whole study area into three subregions (West, Middle and East region) by using precipitation data; Secondly, the accuracy of drought- and wet- year data, different types of vegetation index and different machine learning algorithms were compared region by region, and the key phenological periods were also selected; Finally, we compared the accuracy of four mapping strategies, which were pairwise combination of zoning and non-zoning, full time series and important period, and generated the mapping data of shrub coverage distribution in the whole region using the best mapping strategy. Through the above steps, the study mainly draws four conclusions:

(1) The accuracy of shrub extraction with drought-year data is higher than that with wet-year data, and the performance of random forest is better than that of support vector machine, partial least square regression and Gaussian process.

(2) The optimal vegetation index and key phenological periods of the three subregions are different. NDTI and NDGI are the best vegetation indices in the West region, while NDTI, NDMI and background adjustment indices are the best vegetation indices in the Middle and East region. The key phenological periods of the three subregions are mainly concentrated in the rapid growth stage (June to July) and the senescence stage (September) of vegetation, and the key phenological periods of early summer in the three subregions shows the pattern of advancing from west to east.

(3) In terms of mapping strategy, the mapping accuracy of zoning is higher than that of non-zoning, and the mapping performance using vegetation index in important period is better than that using full vegetation index series in low value prediction of shrub coverage.

(4) Finally, we used the mapping strategy of "zoning plus important period of vegetation index" to generate the distribution data of shrub coverage in the whole region. RMSE is 0.055, the overall accuracy is 80%, and the recall rate is 95%. The shrub coverage increases gradually from west to east in space.

This study makes a comprehensive analysis of the key factors affecting the extraction of shrub coverage by remote sensing, and successfully extracts the distribution data of shrub coverage at high spatial resolution and large scale in Xilin Gol steppe of Inner Mongolia, which provides an important reference for future shrub dynamic monitoring and other related researches in this area.