土壤是陆地生态系统的最大碳库，对全球碳循环具有重要影响。在全球气候变化及人类活动等多种因素的共同影响下，土壤有机碳（Soil organic carbon，SOC）的时空分布发生着显著变化。研究表明：土壤扰动造成的土壤有机碳库的微小变动即可引起大气CO₂浓度的较大变化，从而导致全球变暖趋势加剧。煤炭是我国的主体能源，煤炭开采作为人类主要生产活动之一，对SOC的影响越来越受到人们的关注。当前，我国重点建设的14个亿吨级大型煤炭基地，有9个基地位于生态脆弱的中西部地区。其中，神府东胜矿区（简称：神东矿区）地处我国西部水土流失最为严重的黄土高原生态脆弱区，被认为是我国煤炭开采强度最高的矿区之一，采煤与环境保护的矛盾尤为突出。因此，本文选取神东矿区为研究区，采用原位监测和遥感估算相结合的方法，获取煤矿区长时序大尺度SOC信息，研究SOC时空变化特征及其驱动机制，厘清SOC变化脉络，揭示煤炭开采对SOC变化的影响机理，为西部生态脆弱矿区环境保护及绿色低碳开发提供决策支持。本文主要研究内容及结论如下：

（1）实地采样获取神东矿区表层（0－20 cm）土壤有机碳密度（Soil organic carbon density，SOCD）数据，利用QUAC、FLAASH、6S和LaSRC等4种方法对Landsat遥感影像进行大气校正，提取采样点土壤光谱反射率信息并进行分析。结果表明，通过LaSRC大气校正的遥感影像区分SOCD能力最强。利用相关性分析选取SOCD估算模型的特征变量，分别使用多元线性回归（Multiple Linear Regression，MLR）和随机森林（Random Forest，RF）算法构建矿区不同土地利用类型的SOCD估算模型。结果表明，利用RF算法构建模型的决定系数R²和均方根误差（Root Mean Squared Error，RMSE）均优于利用MLR构建的模型。

（2）选用RF估算模型对神东矿区1990－2021年SOCD进行估算，计算矿区表层土壤有机碳储量（Soil organic carbon stocks，SOCS），分别使用Sen+Mann－Kendall分析、标准差椭圆和转移矩阵方法对矿区SOCS时空变化特征进行分析。结果表明，神东矿区SOCS空间分布以东南部最低，整体呈由东南向西北逐渐增强的格局, 32a间SOCS没有出现明显空间聚集或空间差异性增大的现象；1990－2021年间SOCS整体呈显著增加趋势（R² = 0.89，p < 0.01），由1990年的6.34 Tg增加至2021年的7.73 Tg，增长速度为0.038 Tg·a^-1。32a间SOCS与年平均气温、土壤侵蚀、海拔呈显著负相关； SOCS与年降水量、植被呈显著正相关。32a间不同土地利用类型的SOCS的积累速度有所不同，草地年均SOCS最高，耕地次之，再次为林地和未利用地。

（3）基于SBAS-InSAR，获取2016－2021年神东矿区地表沉陷数据。分别将矿区不同开采强度、开采工作面沉陷盆地作为驱动因子，分析煤炭开采对SOCS时空变化的影响规律，揭示其扰动机制。研究发现：低、中开采强度对SOCS造成的负面影响较大，相关开采区域会产生较多的沉陷坡面，土壤侵蚀加剧，SOC随土壤侵蚀发生迁移，并在迁移过程中矿化，SOCS减少；在高、极高开采强度区域，开采沉陷面积大，但减少了地面不均匀沉陷与拉伸范围，土壤侵蚀速率明显小于低、中开采强度区域，同时由于大面积沉陷后产生了汇水区，导致该区域开采沉陷盆底SOCS高于低、中开采强度区域。

（4）分别利用地理探测器和路径分析模型对神东矿区SOCS变化驱动机制进行研究。前者表明，土壤侵蚀是神东矿区SOCS变化解释力最强的单个因子，土地利用次之，再次为NDVI、降水、海拔、气温、开采强度；后者表明，对SOCS变化的总作用程度依次为：NDVI > 土壤侵蚀 > 开采强度 > 降水 > 海拔 > 土地利用 > 气温。两种结果存在一定差异，主要是由于路径分析模型将变量之间的作用分解为直接影响和间接影响，从而可以更全面地刻画变量间的关系。因此，本文选用路径分析模型对矿区SOCS变化驱动机制进行研究。结果表明，植被对SOCS变化的总作用最强，土壤侵蚀对SOCS变化的直接作用最强，煤炭开采不仅直接影响SOCS，还会因其对海拔、植被和土壤侵蚀造成扰动，从而间接驱动SOCS变化。

Soil functions as the preeminent carbon reservoir within terrestrial ecosystems, wielding a pivotal influence on the global carbon cycle. Under the concurrent influence of diverse factors such as global climate change and anthropogenic activities, the temporal and spatial distribution of soil organic carbon (SOC) undergoes pronounced alterations. Research findings underscore that even subtle perturbations in the soil organic carbon reservoir due to soil disturbance can precipitate considerable fluctuations in atmospheric CO₂ concentrations, thereby accentuating the prevailing global warming trajectory. With coal serving as the predominant energy source in China, heightened attention is directed towards the impact of coal mining on SOC, given its status as a principal anthropogenic activity. At present, China is strategically prioritizing the establishment of 14 large-scale coal bases, each with an annual production capacity of one billion tons. Notably, nine of these bases are situated in environmentally delicate regions within the central and western expanses of the country. The Shendong Mining Area, situated in the ecologically vulnerable Loess Plateau of western China, stands out as one of the coal mining regions characterized by the highest intensity in the nation. The conundrum of coal mining conflicting with environmental preservation is particularly conspicuous in these locales. Consequently, this study selects the Shendong Mining Area as its focal point, employing a hybrid approach that combines in-situ monitoring and remote sensing estimation to procure extensive, longitudinal data on SOC within coal mining areas. The study seeks to scrutinize the spatiotemporal attributes of SOC, delineate its driving mechanisms, disentangle the dynamics of SOC fluctuations, and expound upon the impact mechanisms of coal mining on SOC. The outcomes of this investigation are intended to furnish decision-making support for environmental conservation and the promotion of green, low-carbon development in ecologically fragile mining regions in western China. The ensuing sections explicate the principal research components and conclusions of this paper:

(1) Ground-based sampling was undertaken to acquire data on the surface (0-20 cm) soil organic carbon density (SOCD) in the Shendong Mining Area. Four methods, namely QUAC, FLAASH, 6S, and LaSRC, were applied for the atmospheric correction of Landsat remote sensing images. The soil spectral reflectance information from the sampling points was extracted and subjected to analysis. The outcomes reveal that, among the atmospheric correction methods, LaSRC demonstrates the highest efficacy in distinguishing SOCD in the remote sensing images. Characteristic variables for the SOCD estimation model were identified through correlation analysis. Subsequently, Multiple Linear Regression (MLR) and Random Forest (RF) algorithms were utilized to construct SOCD estimation models for different land-use types within the mining area. The findings indicate that models generated using the RF algorithm exhibit superior coefficients of determination (R²) and lower Root Mean Squared Error (RMSE) when compared to those constructed using the MLR algorithm.

(2) The RF estimation model was employed to assess the SOCD in the Shendong Mining Area for the period spanning 1990 to 2021. Subsequently, soil organic carbon stocks (SOCS) for the surface layer were computed, and an analysis of the temporal and spatial variations of SOCS in the mining area ensued. The analytical approaches included Sen+Mann-Kendall analysis, standard deviation ellipse, and transition matrix methods. The findings elucidate that the spatial distribution of SOCS in the Shendong Mining Area exhibits a nadir in the southeast, portraying an overarching trend of progressive augmentation from the southeast to the northwest. Notably, over the 32-year interval, there is an absence of conspicuous spatial clustering or an increase in spatial heterogeneity of SOCS. The aggregate trend in SOCS from 1990 to 2021 manifests a noteworthy increase (R² = 0.89, p < 0.01), escalating from 6.34 Tg in 1990 to 7.73 Tg in 2021, with a growth rate of 0.038 Tg·a^-1. The spatiotemporal analysis reveals a significant negative correlation between SOCS over the 32-year span and variables such as annual mean temperature, soil erosion, and elevation. Conversely, a positive correlation is observed with annual precipitation and vegetation cover. Furthermore, disparate rates of SOCS accumulation are noted among different land-use types during the 32-year period, with grassland exhibiting the highest annual average accumulation, followed by cropland, forestland, and unused land, in that order.

(3) Leveraging the SBAS-InSAR technique, subsidence data in the Shendong Mining Area spanning the period 2016 to 2021 were obtained. Diverse mining intensities and subsidence basins of mining faces were identified as driving factors to scrutinize the spatiotemporal variations in SOCS induced by coal mining and to unravel the associated disturbance mechanisms. The investigation discloses that lower and medium mining intensities exert a considerable adverse impact on SOCS, giving rise to the formation of numerous subsidence slopes in the corresponding mining regions. This phenomenon intensifies soil erosion, leading to the migration of SOC concomitant with the soil erosion process. Throughout this migration, mineralization occurs, resulting in a decrement of SOCS. In regions characterized by higher and extremely high mining intensities, despite the presence of extensive subsidence areas, there is a reduction in uneven subsidence and stretching ranges on the ground. The rate of soil erosion in these areas is markedly lower than in regions characterized by lower and medium mining intensities. Concurrently, the extensive subsidence gives rise to runoff areas, contributing to elevated SOCS levels in the subsidence basin compared to regions with lower and medium mining intensities. The outcomes of the study suggest that the impact of coal mining on SOCS is contingent upon the intensity of mining activities. Lower and medium intensities tend to induce more pronounced negative effects, while higher intensities exhibit distinctive patterns characterized by reduced soil erosion and enhanced SOCS in subsidence basins.

(4) The research investigated the driving mechanisms behind variations in SOCS in the Shendong Mining Area using Geographic Detector and Path Analysis models. The findings from Geographic Detector revealed that soil erosion emerges as the most potent single factor explaining the variability in SOCS, followed by land use, NDVI, precipitation, elevation, temperature, and mining intensity. Conversely, the Path Analysis model delineated the overall impact sequence on SOCS changes as follows: NDVI > Soil erosion > Mining intensity > Precipitation > Elevation > Land use > Temperature. Discrepancies between the two models arise due to the Path Analysis model's ability to decompose the effects between variables into direct and indirect impacts, offering a more comprehensive understanding of their relationships. Therefore, this study opted for the Path Analysis model to comprehensively explore the driving mechanisms of SOCS changes in the mining area. The results underscore that vegetation exerts the most substantial overall influence on SOCS changes, with soil erosion contributing the most direct impact. Importantly, coal mining not only directly impacts SOCS but also induces disturbances in elevation, vegetation, and soil erosion, thereby indirectly influencing changes in SOCS.