<p>&nbsp; &nbsp; &nbsp; 黄土高原矿区是我国煤炭主要产区之一，大范围、高强度地下开采引起的地面塌陷不可避免地造成矿区耕地资源的持续损害，导致黄土矿区的耕地面积和质量降低，人-地矛盾日益突出。因此，定量研究煤炭开采区耕地资源的动态演变规律对于矿区可持续发展具有重要意义。目前，卫星遥感技术已成为土地利用变化监测的重要手段，但针对黄土高原复杂地貌的矿区耕地资源演化的时序遥感监测成果甚少。为此，本文以典型黄土矿区大佛寺煤矿为实例，基于长时间序列遥感数据，在GEE平台上采用时空分割算法提取大佛寺矿区的耕地变化信息，分析土地利用尤其是耕地利用的变化特征，探讨地形因子对耕地空间分布及演化的影响，揭示采煤沉陷变形对于矿区耕地时空演化特征的扰动效应。论文研究的主要内容及结果如下：</p> <p>&nbsp; &nbsp; &nbsp;（1）将遥感影像空间特征与耕地作物生长期的时间窗口相结合，构建了一种基于时空分割的多特征影像耕地提取技术方法。按作物生育关键期分阶段生成光谱特征中位数、标准差等统计量，加入纹理特征，并将主成分分析结果作为SNIC算法的输入，实现影像空间分割；引入地形因子，并将特征优选结果应用于随机森林分类算法。结果表明，在长时序、中小尺度复杂地形区域进行耕地空间分布提取中获得了较好的精度，分类结果精度较单时相影像提高了约9.81%，较基于像素分类法提高了约5.6%。地形校正则有效地解决了地形复杂地区光照不均所导致的像元类型混淆，降低了错分率。</p> <p>&nbsp; &nbsp; &nbsp; （2）采用土地利用动态指标对大佛寺矿区2005-2020年间耕地变化时空特征进行了定量分析，探讨了地形因子对耕地空间分布及演化的影响，并基于LandTrendr时间轨迹断点监测模型，逐年度获取了矿区耕地演化的空间位置及其发生时间。结果表明，矿区耕地利用过程阶段性变化特征明显，耕地利用结构整体上呈现出波动式上升趋势，且矿区内弃耕现象较为普遍，弃耕时长与弃耕耕地面积成反比。从时间变化上看，在研究时段内，耕地变化强度呈&ldquo;先增加后减少&rdquo;的趋势，主要转出方向为建筑用地、林草地以及裸地，主要转入方向以林地、草地以及裸地为主；耕地利用变化特征由高弃耕率、低恢复率向低弃耕率、高恢复率转变。从空间变化上看，耕地演化分布与地形因子呈现较强的空间相关性，具体表现为耕地普遍位于高程较高、坡度较缓、地形位较低的区域，弃耕耕地多位于地势变化明显的区域，其中，高海拔台塬边界陡坡区域是耕地最易发生转化的地区。与已有分类结果对比，时间轨迹模型提取准确度可达93.33%，体现了GEE平台在进行长时序小区域尺度上耕地演化特征提取的有效性。</p> <p>&nbsp; &nbsp; &nbsp; （3）结合采煤沉陷扰动范围及相关指标，分析了开采沉陷区耕地演化的时空特征及沉陷扰动对耕地变化的影响。首先采用InSAR技术以及开采沉陷计算模型获取了开采引起的地表下沉范围，并圈定耕地损毁影响范围。然后按开采扰动程度将研究区划分为不同的采煤影响区域，并对各影响区内耕地面积减少的时空分布情况进行分析统计，最后从耕地平整度和土壤特性变化两个方面分析了采煤沉陷变形对耕地的影响机理。结果表明，耕地面积减少幅度与采动影响程度呈正相关。其中，开采沉陷区、间接影响区、非影响区耕地面积的变化率分别为-24.27%、-19.39%、24.07%。开采沉陷区内地表不均匀沉陷改变了耕地的平整度，采动裂缝破坏了耕地的完整性，造成土壤水分和肥力过度流失，降低了矿区耕地的质量。</p>

<p>&nbsp; &nbsp; &nbsp; The Loess Plateau mining area is one of the major coal producing areas in China. The ground collapse caused by large-scale and high-intensity underground mining inevitably causes continuous damage to the arable land resources in the mining area, which in turn leads to the reduction of the arable land area and quality in the loess mining area, making the human-land conflict increasingly prominent. Therefore, it is of great significance to conduct quantitatively research on the dynamic evolution of arable land resources in coal mining areas for the sustainable development of mining areas. At present, satellite remote sensing technology has become an important means of monitoring land use changes. But the results of time-series remote sensing monitoring for the evolution of arable land resources in mining areas with complex terrain on the Loess Plateau are still insufficient. In this context, this paper takes Dafosi coal mine as the study area, which is a typical loess mining area. Based on the long time series remote sensing image data, the Spatio-temporal segmentation algorithm is used on the GEE platform to extract information on the change of arable land in the Dafosi mining area, analyze the change characteristics of land use, especially arable land use, and explore the influence of topographic factors on the spatial distribution and evolution of arable land to reveal the disturbance effect of coal mining subsidence deformation on the Spatio-temporal evolution characteristics of arable land in the mining area. The main contents and results of the paper are as follows.</p> <p>&nbsp; &nbsp; &nbsp; (1) A multi-featured image extraction technology method based on spatio-temporal segmentation of cropland is constructed, by combining the spatial features of remote sensing images with the time windows of cropland crop growing period. The specific elements of the method are to generate statistics such as median and standard deviation of spectral features in stages according to the critical crop fertility period, add texture features and use the results of principal component analysis as input to the SNIC algorithm to achieve image spatial segmentation; introduce topographic factors and apply the results of feature preference to the random forest classification algorithm. The results show that the method used in this paper achieves better accuracy in extracting the spatial distribution of cultivated land in long time series, small and medium scale complex terrain areas, and the accuracy of the classification results is improved by about 9.81% compared with single time phase images and about 5.6% compared with pixel-based classification method. The terrain correction effectively solves the confusion of image element types caused by uneven illumination in complex terrain areas and reduces the misclassification rate.</p> <p>&nbsp; &nbsp; &nbsp; (2) In this section, the spatial and temporal characteristics of arable land changes in the Dafosi mining area between 2005 and 2020 are quantitatively analysed using dynamic land use indicators, and the influence of topographic factors on the spatial distribution and evolution of arable land is explored. In addition, the spatial location and time of occurrence of arable land evolution in the mining area were obtained on an annual basis based on the LandTrendr time-trajectory breakpoint monitoring model. The results show that the phase change of the arable land use process in the mine area is obvious, and the overall arable land use structure shows a fluctuating upward trend, and the abandonment phenomenon is more common in the mine area, and the length of abandonment is inversely proportional to the area of abandoned arable land. In terms of temporal changes, the intensity of cultivated land transfer during the study period showed a trend of &quot;increasing before decreasing&quot;, with the main transfer directions being construction land, forest and grassland and bare land, and the main transfer directions being forest, grassland and bare land; the characteristics of change in cultivated land use changed from a high abandonment rate and a low recovery rate to a low abandonment rate and a high recovery rate. In terms of spatial changes, the distribution of cropland evolution and topographic factors show a strong spatial correlation, which is reflected in the fact that cropland is generally located in areas with higher elevation, gentler slope and higher topographic position, while abandoned cropland is mostly located in areas with obvious topographic changes, among which, the gently sloping areas of high-altitude plateau are the areas where cropland is most likely to be converted. In comparison with the existing classification results, the accuracy of the temporal trajectory model can reach 93.33%, which demonstrates the effectiveness of the GEE platform in extracting the evolutionary features of arable land on a small regional scale in a long time series.</p> <p>&nbsp; &nbsp; &nbsp; (3) The spatial and temporal characteristics of the evolution of arable land in mining subsidence areas and the impact of subsidence disturbance on the change of arable land are analysed by combining the extent of coal mining subsidence disturbance and related indicators.The InSAR technique and the mining subsidence calculation model were used to obtain the extent of mining-induced surface subsidence and to circle the area of arable land destruction. The study area was then divided into different coal mining impact areas according to the degree of mining disturbance, and the spatial and temporal distribution of the reduction in arable land area within each impact area was analysed and statistically calculated, and finally the mechanism of the impact of coal mining subsidence deformation on arable land was analysed in terms of the changes in arable land levelling and soil properties.The results show that the reduction of arable land area is positively correlated with the degree of mining impact. Among them, the change rates of arable land area in the mining subsidence area, indirectly affected area and non-affected area were -24.27%, -19.39% and 24.07% respectively. The uneven subsidence of the surface within the mining subsidence area changed the flatness of the arable land, and the mining cracks damaged the integrity of the arable land, causing excessive loss of soil moisture and fertility and reducing the quality of the arable land in the mine area.</p>