森林是陆地生态系统的重要组成部分，在调节气候、碳汇碳库和水文循环等方面均发挥着至关重要的作用。2000年以来，我国大规模开展造林/还林等生态恢复工程，黄土高原地区植被退化严重，生态环境脆弱，是我国开展生态恢复工程的先锋和核心区域之一。生态恢复工程实施后，该地区的植被明显改善，但是对于森林覆盖和生长状况发生的各项变化仍待进行全面和系统的研究。本文借助遥感和地理信息系统技术，综合使用多种卫星遥感和野外样方调查数据，采用空间分析、时间序列分析及机器学习等方法，分析了该地区森林覆盖及生长状态变化情况；采用形态空间格局分析方法研究了该地区森林空间格局的动态变化特征；最后，结合自然和人类活动因素全面地分析了森林变化的驱动因素，并进一步从森林覆被变化的角度评估了生态恢复工程投入的有效性。 本研究所得的主要结果如下： 1. 2001~2016年间黄土高原地区的树木覆盖面积、平均树木覆盖率和树木覆盖区域的平均树木覆盖率均显著增加 (P < 0.05)。树木覆盖面积从64%增加至94%，平均树木覆盖率从7%变化至11%，树木覆盖区域平均树木覆盖率从约10%增长至12%。全区约55.72%的区域树木覆盖呈增加趋势，显著增加的区域主要分布在黄土高原地区的核心的山地和山地边缘地区（约占30.19%），树木覆盖减少的区域主要集中在平原地区（约占5.84%）。基于过去16年的树木覆盖特征将研究区划分为5个区域，该时空分区方式能够指示树木分布的适宜和限制区，为评估黄土高原地区树木和森林覆盖的时空变化提供了更全面和详尽的信息。 2. 比较四种卫星遥感数据获得的2001~2016年黄土高原地区森林覆盖的年际变化、变化趋势和森林覆盖增加与减少的差异性与一致性，结果显示研究区森林覆盖面积呈增加趋势，森林覆盖变化范围主要集中分布于研究区的东南部，且森林增加面积大于减少面积，森林增加比减少具有更多一致性。证实了该地区的森林覆盖范围正在稳步增加（新增森林面积约在35 km2与56,258 km2之间，分别约占该区面积的0.01%和8.79%）。然而，森林增加的数量和空间位置仍存在较大差异，这或许与采用的森林定义、数据的来源、选取的训练样本及算法等的差异均密切相关。 3. 基于野外样方调查得出的森林定义为：树木平均高度为9±3 m，平均盖度为66±18%。采用随机森林模型方法，结合野外样地和卫星遥感数据，以森林和非森林地物的物候特征差异为主要分类依据，获得了全区的森林和非森林分类结果（分类结果的总精度为96%，Kappa系数为0.91）。将2001~2016年间的森林变化分为森林净增加、森林净减少、森林扰动、持续森林和非森林五个类型。森林面积从原来占全区的约8.19%增长至15.82%。基于随机森林模型分类输出的森林概率时间序列能够反映出森林不同生长阶段的连续变化情况，从非森林变为成熟森林。这一变化显示了该地区的森林转型特征，从森林面积减少转变为增加。在森林净增加地区，森林概率阶段存在明显的从低变高的过程，证实了森林面积的增加；在森林持续区，高等级森林概率类型的增加指示出森林密度的增加，表明了森林生长状态的持续提升。随机森林模型输出森林概率与时间分割方法结合监测森林变化的研究框架为区域森林变化分析、制图及评估提供了新思路。 4. 基于原有森林和新增森林的不同生态价值，提出了一种分析原有森林和新增森林时空格局动态变化的评判方法，发现2001~2016年黄土高原地区84.21%的原有森林未发生覆盖变化。采用形态空间格局分析模型分析森林空间格局的结果显示：核心森林显著增加 (P < 0.01; 2,585 km2 a-1)，由2001年约占研究区的5.30%增长至2016年的6.39%。两项主要的生态恢复工程（“退耕还林”工程和“天然林保护”工程）在研究区的森林景观上留下了明显的印迹：从农田到森林的变化形成了许多小的森林岛屿斑块，在原有森林周边增加的森林提高了森林覆盖度和核心森林面积。 5. 2001~2016年间黄土高原地区的森林面积、累计造林面积、人口和GDP均显著增加 (P < 0.01)，但同期气温和降水的增加并不显著 (P > 0.05)。自然和人类活动因素均影响着研究区的森林分布，降水是其中最重要的影响因素，当年降水量大于400 mm，随着降水量的增加森林分布概率明显增大。巨大的造林投入驱动了该地区森林面积的增加，经济的快速增长为生态恢复提供了雄厚基础资金支撑。在GDP水平较高的县（区），造林投入亦相应较高，植被和树木覆盖增加明显，而森林覆盖增加并不明显，森林覆盖增加主要发生在自然条件较好并且人类活动较少的地区。

Forests, as one of the most essential components of the land ecosystem, play key roles in climate change mitigation, carbon sequestration and hydrologic recycle. Since the millennium, People’s Republic of China launched a number of large ecological restoration programmes relating to forestation. Loess Plateau located in the central north of China, is infamous for its fragile ecological environment and dedegraged vegetation, thus it is listed as the pioneer and core region for these ecological restoration programmes. After the implementation of these programmes, the vegetation of the Loess Plateau has been dramatically improved. However, forest change in this region is largely unknown, and further studies are needed. In this thesis, with the help of remote sensing and geographical information system technology, by the integrated use of satellite remote sensing data and field survey data, spatial analysis, time series analysis and machine learning methods were applied to analyse forest cover and growth condition change. Morphological spatial pattern analysis was further adopted to analyse forest spatial pattern dynamic and change. Based on those analyses, drivers for forest change in the Loess Plateau were distinguished both from biophysical and anthropogenic aspects, and the effectiveness of these ecological restoration programmes from forest change perspective were also evaluated. The results and conclusions of this research are as follows: 1. From 2001 to 2016, tree-covered area, cumulative tree cover and average annual tree cover of the Loess Plateau all significantly increased (P < 0.05). The area of tree-covered changed from 64% to 94% accounting for the study area, cumulative tree cover from 7% to 11% and average annual tree cover from 10% to 12%. Tree cover increase trend area were approximately 55.72% of the study area. Significant increase tree cover trend regions accounted for 30.19%, and mainly located in the core and the boundary of the mountains. While tree cover decrease regions principally concentrated in the plain area, and were about 5.84%. Based on the 16-year tree cover data, the study area was classified into five regions, which is able to indicate the suitable and restricted areas for tree distribution. This classification method offers useful information for tree cover evaluation for the Loess Plateau. 2. Forest cover change trend and change region and area acquiring from the four different satellite remote sensing data sets for the Loess Plateau from 2001 to 2016 showed that forest cover has an increasing trend; forest increasing areas were mostly located in the southeast; the area of forest gain was larger than forest loss, and more consistencies existed in forest gain than forest loss. These consistencies proved that forest cover of Loess Plateau had increased, and the increase area was between approximate 35 km2 and 56285 km2 (0.01% and 8.79% of the study area). However, the quantity and the spatial locations for these changes have large differences, and these differences may due to the selected forest definition, data source, training samples and algorithm of these data sets. 3. Based on the field survey, forests were defined as trees with average height: 9±3 m and average cover: 66±18%. The classification accuracy for forest and non-forest by the use of random forest method with the combination of field and satellite remote sensing data based on the phenological characteristics was 96%, and the Kappa coefficient was 0.91. Forest change from 2001 to 2016 was classified into five types, namely, forest net gain, forest net loss, forest fluctuation, persistent forest and non-forest. Forest area increased from about 8.19% to 15.82%. Forest probability acquired from remote sensing time series reflected different forest growth stages, from non-forest to mature forest. Such change shows forest transition from forest loss to forest gain has taken place in the study area. In the forest net gain region, the obvious increase from low to high forest probability proved the increase of forest. In persistent forest region, the increase of high forest probability levels indicated the increase of forest density, which reflected the improvement on forest growth condition. The frame for forest monitoring with the combination of the output probability from random forest model and time segmentation method provides a new way for forest change analysis, mapping and evaluation. 4. A spatial pattern analysis method for spatiotemporal dynamics of old and new forest based on different ecological values has been proposed. From 2001 to 2016, 84.21% of the old forests existed throughout the study period. Moreover, core forests (defined as a forest area which is surrounded by other forest areas) significantly increased (P < 0.01, 2585 km2 a-1), change from 5.30% in 2001 to 6.39% in 2016. Two main ecological restoration programmes have left clear footprints on the forest landscape of the Loess Plateau: (1) The Natural Forest Conservation Programme, aiming at expanding old forest, has resulted in the establishment of considerable areas of new forest surrounding old forest. This has promoted new core forest areas to emerge. (2) The Grain for Green Programme has mainly caused a fragmented landscape of forest islets which gradually connect to core forest areas. 5. From 2001 to 2016, forest area, cumulative forestation area, population and GDP in the Loess Plateau all significantly increased (P < 0.01), while the increase of temperature and precipitation were not significant (P > 0.05). Both of the biophysical and anthropogenic factors determine forest distribution of the study area. Precipitation is the most important factor, and when the annual precipitation is large than 400 mm, the probability of forest occurrence obviously increases with the increase of precipitation. Large forestation investment has driven forest area increase, and the rapid increased economy provides the basic financial support for the large ecological restoration. In the area with higher GDP, the forestation investment is also higher, and the gain of vegetation and tree cover are dramatic, but not for forest. Forest gain occurs in the region with better biophysical condition and less human activities.