土壤呼吸是土壤碳库输出碳的主要途径，对陆地生态系统的碳循环起着关键的调控作用，即使其微小变动，也可能对大气中的二氧化碳浓度产生显著影响。在煤炭开采背景下，深入了解荒漠化露天矿的土壤呼吸昼夜变化特征及影响因子，及时准确监测荒漠化矿区土壤呼吸（Soil respiration, Rs），对评估生态脆弱的露天矿区生态系统碳循环至关重要。因此，本研究以红沙泉露天煤矿为研究区，获取排土场复垦区和人工林等5个地物不同深度的土壤昼夜呼吸速率及相关数据，以及无人机多光谱、热红外影像和哨兵2号卫星影像数据；通过统计分析得到土壤呼吸昼夜变化特征及其主要影响因素，并利用无人机和卫星影像波段计算光谱指数，采用多元线性回归（Multiple linear regression, MLR）、随机森林回归（Random forest regression, RFR）、支持向量回归（Support vector regression, SVR）、反向传播神经网络（Back-propagation neural network, BPNN）和粒子群优化支持向量回归（Particle swarm optimization support vector regression, PSO-SVR）构建样地及矿区尺度的表层土壤呼吸遥感估算模型，分别得到样地和矿区表层的土壤呼吸空间分布，并利用表层和深层实测土壤呼吸之间的线性拟合关系，分别得到样地和矿区的深层的土壤呼吸空间分布。主要结论如下：

（1）5种地物类型之间的昼夜土壤呼吸差异显著，10和30cm土壤呼吸速率日均值表现为人工林>红沙泉>排土场复垦区>柽柳林>南线，10cm分别为5.72、0.98、0.34、0.08、-0.08μmol·m^-2·s^-1，30cm分别为5.14、2.18、0.77、0.40、-0.05μmol·m^-2·s^-1，南线为碳汇，其余4种地物类型为碳源。土壤呼吸速率昼夜变化规律明显，排土场复垦区不同土层和红沙泉10cm土壤呼吸昼夜变化呈“双峰”趋势，人工林、南线不同土层及红沙泉30cm均呈“单峰型”，而柽柳林10和30cm土壤呼吸昼夜变化呈“多峰型”。在日尺度上，排土场复垦区、人工林和红沙泉10cm土壤呼吸速率与土壤温度（P<0.05）和土壤含水量（P<0.01）呈显著正相关；土壤温度和土壤含水量能较好的综合解释矿区内5种地物类型土壤呼吸速率50.5%~97.3%的变化，而南线10cm仅11.3%。柽柳林和南线土壤呼吸与土壤碳含量关系与其余3种地物类型不同，与无机碳（Soil inorganic carbon, SIC）呈显著负相关（r=-0.69, P<0.01），无机过程造成土壤呼吸速率为负值，SIC是土壤碳汇的关键影响因素；而排土场复垦区、人工林和红沙泉均与溶解性活性有机碳（Dissolved active organic carbon, DOC）呈较强负相关性，排土场复垦区和柽柳林土壤呼吸还与土壤有机碳（Soil organic carbon, SOC）存在较强正相关性，SOC和DOC是土壤碳源的重要影响因子。

（2）在构建样地尺度的土壤呼吸估算模型中，绿波段叶绿素植被指数（Green band chlorophyll vegetation index, CI_green）、红边波段叶绿素植被指数（Red edge band chlorophyll vegetation index, CI_rededge）和绿波大气抗阻指数（Green wave atmospheric resistance index, GARI）、地表温度（Surface temperature, T）、盐分指数（Salt index, SI）变量组合的PSO-SVR表层土壤呼吸估算模型精度最高（R²=0.959, RMSE=0.497μmol·m^-2·s^-1, AIC=-0.561）。除排土场复垦区10与30cm土壤间相关性较弱，其余地物类型10与30cm之间的土壤呼吸均表现为较强的性关系，决定系数R²分别为0.97、0.96、0.83和0.73。样地尺度的土壤呼吸空间分布上，排土场复垦区和人工林路面缺少植被和人为活动较多的土壤呼吸多为低值；而人工林和排土场，以及红沙泉径流两边植被较多的区域，土壤呼吸多为高值；远离径流的区域植被稀疏、裸土较多和盐渍化严重的区域，土壤呼吸多为低值；而南线的西南区域土质多为风沙土，土壤呼吸速率多为负值，存在“碳汇”现象。

（3）在构建矿区尺度的土壤呼吸估算模型中，基于CI_green、增强型植被指数（Enhanced vegetation index, EVI）、归一化差异植被指数（Normalized difference vegetation index, NDVI）、GARI、土壤调节植被指数（Soil adjusted vegetation index, SAVI）、SI和土壤湿度监测指数（Soil moisture monitoring index, SMMI）光谱指数，利用RFR构建的矿区表层土壤呼吸估算模型的精度最优（R²=0.757，RMSE=0.509μmol·m^-2·s^-1，AIC=-3384.33），并得到矿区表层土壤呼吸空间分布结果；利用表层和深层的土壤呼吸线性拟合方程（R²为0.663）得到矿区深层土壤呼吸空间分布结果。在空间分布上，土壤呼吸高值主要位于矿区内的中部和东部、植被覆盖较多和人类活动大的生活聚集区和煤炭开采区，其余大部分人迹罕至的荒漠区域的土壤呼吸速率为低值，碳汇区主要集中在南线的裸地。与周边未开采的区域相比，煤炭开采区的土壤呼吸速率更高，煤炭开采会导致区域性的土壤碳排放增加。

Soil respiration is the main pathway for carbon output from soil carbon pools and plays a crucial regulatory role in the carbon cycle of terrestrial ecosystems. Even minor changes in soil respiration may have a significant impact on the concentration of carbon dioxide in the atmosphere. In the context of coal mining, it is crucial to deeply understand the diurnal variation characteristics and influencing factors of soil respiration in desertified open-pit mines, and timely and accurately monitor soil respiration (Rs) in desertified mining areas, in order to assess the carbon cycle of ecologically fragile open-pit mining ecosystems. Therefore, this study takes the Hongshaquan open-pit coal mine as the research area to obtain soil diurnal respiration rates and related data at different depths of five land features, including reclamation area of dump and artificial forest, as well as unmanned aerial vehicle multispectral, thermal infrared images and Sentinel-2 satellite image data; By statistical analysis, the diurnal variation characteristics and main influencing factors of soil respiration were obtained, and spectral indices were calculated using drone and satellite image bands. Multiple linear regression (MLR), random forest regression (RFR), support vector regression (SVR), backpropagation neural network (BPNN), and particle swarm optimization support vector regression (PSO-SVR) were used used to construct remote sensing estimation models for surface soil respiration at the scale of sample plots and mining areas. The spatial distribution of soil respiration at the surface of sample plots and mining areas was obtained, and the linear fitting relationship between measured soil respiration at the surface and deep layers was used to obtain the spatial distribution of soil respiration at the deep layers of sample plots and mining areas. The main conclusions are as follows:

(1) There is a significant difference in soil respiration between the five types of land cover during the day and night. The daily average soil respiration rates at 10 and 30cm are as follows: artificial forest > Hongshaquan > reclamation area of dump > tamarix forest > southern line. The values at 10cm are 5.72, 0.98, 0.34, 0.08, -0.08μmol·m^-2·s^-1, and at 30cm are 5.14, 2.18, 0.77, 0.40, -0.05μmol·m^-2·s^-1, respectively. The southern line is a carbon sink, while the remaining four types of land cover are carbon sources. The diurnal variation pattern of soil respiration rate is obvious. The diurnal variation of soil respiration in different soil layers and 10cm of Hongshaquan in the reclamation area of dump shows a "double peak" trend. The artificial forest, different soil layers on the southern line, and 30cm of Hongshaquan all show a "single peak" pattern, while the diurnal variation of soil respiration in 10 and 30cm of tamarix forest shows a "multi peak" pattern. On a daily scale, there is a significant positive correlation between soil respiration rate and soil temperature (P<0.05) and soil moisture content (P<0.01) in the 10cm soil of reclamation area of dump, artificial forest, and Hongshaquan; Soil temperature and soil moisture content can comprehensively explain the changes in soil respiration rates of 50.5% to 97.3% for the five types of land cover in the mining area, while the southern line only has 11.3% at 10cm. The relationship between soil respiration and soil carbon content in the tamarisk forest and the southern line is different from the other three types of land cover, and is significantly negatively correlated with soil inorganic carbon (SIC) (r=-0.69, P<0.01). Inorganic processes cause negative soil respiration rates, and SIC is a key influencing factor of soil carbon sink; The reclamation area of dump, artificial forest, and Hongshaquan all show a strong negative correlation with dissolved active organic carbon (DOC). The soil respiration in the reclamation area of dump and the tamarisk forest also has a strong positive correlation with soil organic carbon (SOC). SOC and DOC are important influencing factors of soil carbon sources.

(2) In the construction of soil respiration estimation models at the sample plot scale, the PSO-SVR surface soil respiration estimation model with the combination of green band chlorophyll vegetation index (CI_green), red edge band chlorophyll vegetation index (CI_rededge), green wave atmospheric resistance index (GARI), surface temperature (T), and salt index (SI) variables has the highest accuracy (R²=0.959, RMSE=0.497μmol·m^-2·s^-1, AIC=-0.561）。 Except for the weak correlation between 10 and 30cm soil in the reclamation area of dump, the soil respiration between 10 and 30cm of other land types shows a strong sexual relationship, with determination coefficients R² of 0.97, 0.96, 0.83, and 0.73, respectively. In terms of spatial distribution of soil respiration at the sample plot scale, the soil respiration in reclamation area of dump and artificial forest road surfaces is mostly low due to the lack of vegetation and human activities; However, artificial forests, reclamation area of dump, and areas with more vegetation on both sides of the Hongshaquan runoff have high soil respiration values; Areas far from runoff with sparse vegetation, abundant bare soil, and severe salinization tend to have low soil respiration; The southwestern region of the southern line is mostly composed of sandy soil, with negative soil respiration rates and a phenomenon of carbon sequestration.

(3) In constructing a soil respiration estimation model at the mining area scale, based on CI_green, Enhanced Vegetation Index (EVI), Normalized Difference Vegetation Index (NDVI), GARI, Soil Adjusted Vegetation Index (SAVI), SI, and Soil Moisture Monitoring Index (SMMI) spectral indices, the RFR constructed surface soil respiration estimation model in the mining area has the best accuracy (R²=0.757, RMSE=0.509μmol·m^-2·s^-1, AIC=-3384.33), and obtain the spatial distribution results of surface soil respiration in the mining area; The spatial distribution of deep soil respiration in the mining area was obtained by using a linear fitting equation (R²=0.663) for surface and deep soil respiration. In terms of spatial distribution, the high values of soil respiration are mainly located in the central and eastern parts of the mining area, living areas with high vegetation coverage and human activities, and coal mining areas. The soil respiration rate in most of the sparsely populated desert areas is low, and the carbon sink areas are mainly concentrated in the bare land along the southern line. Compared to the surrounding unexplored areas, the soil respiration rate in coal mining areas is higher, and coal mining can lead to an increase in regional soil carbon emissions.