地表土壤水分指储存在土壤上层中的水含量，是地表能量平衡和水循环的关键变量，控制着陆地表面和大气之间的相互作用。准确且时空完整的土壤水分信息对于气候预测、洪水预报、干旱评估等应用至关重要。尽管卫星遥感已被广泛应用于全球土壤水分的获取，但由于轨道覆盖、传感器空间分辨率和反演算法的限制，现有的土壤水分产品通常时空覆盖不完整、空间分辨率较低，难以满足实际应用需求。已有的数据融合和降尺度方法虽然可以提高土壤水分数据的空间分辨率和时空完整度，但往往无法兼顾准确性、时空完整度和时空分辨率的需求。传统方法构建时空完整土壤水分数据需要大量辅助数据和复杂计算，效率低且结果不稳定。相比之下，时空统计方法根据变量内在的时空关系，有效处理缺失数据，提高计算效率，同时保持数据原有的时空分辨率和精度。因此，本文以时空统计理论为基础，利用精度较高的SMAP卫星产品，研究构建不同分辨率下时空完整土壤水分数据的方法。主要研究成果如下：

（1）提出了基于固定阶数滤波（Fixed Rank Filtering，FRF）的单一卫星土壤水分数据时空重构方法。首次将时空统计领域最新的FRF方法引入单一传感器的卫星土壤水分产品重构研究，充分发挥FRF空间维使用固定阶数进行空间降维提高计算效率、时间维使用Kalman滤波进行动态估计的优势。以美国区域的SMAP增强土壤水分数据（SM_P_E，空间分辨率为9km）为例开展了研究。结果表明，FRF重构方法显著提高了SM_P_E数据的时空完整度，平均时空完整度提高40%以上。利用FRF重构的土壤水分数据在有原始数据像元和无原始数据像元的精度都与原始数据有很高的一致性。FRF重构后的数据与原始数据在空间分布格局和时间序列变化趋势方面与原数据保持一致，但在局部空间变化方面，空间细节信息有所减少。

（2）针对大范围长时间连续缺失情况下时空统计插值难以高精度重构卫星遥感土壤水分小尺度时空变异特征的问题，将陆面模型再分析的土壤水分产品引入卫星土壤水分产品重构，提出了利用时间克里金融合方法（Temporal Kriging Fusion Method，TKFM），融合再分析产品和遥感产品生成时空完整的土壤水分产品。该方法综合了再分析产品和遥感产品在时空覆盖范围以及产品精度方面的互补特性，利用土壤水分数据的时间相关性，完成对未知点的最优估测。以SM_P_E遥感数据和ERA5-Land再分析数据为例开展了研究。结果表明，该融合方法在不牺牲产品时空分辨率和精度的条件下，生成了空间纹理特征精细的土壤水分数据。土壤水分产品的平均时间完整度由42.94%提高到85.44%，平均空间完整度由42.80%提高到85.65。融合结果精度稳定，原始数据、融合结果、融合结果在有原始数据像元、融合结果在无原始数据像元与实测数据的相关性分别为0.69、0.68、0.69和0.66。融合数据的空间分布和时间序列动态变化与原数据保持一致。与FRF重构方法相比，TKFM在保留空间细节方面表现出更显著的优势，但融合产品的时空完整度和精度稳定性略低于FRF重构结果。

（3）针对高空间分辨率土壤水分数据时空完整度低的问题，提出了迭代融合多源微波数据的时间贝叶斯最大熵（Temporal Bayesian maximum entropy，T-BME）方法，用以生成高空间分辨率（3km）时空完整的土壤水分数据。该方法对多源数据的不确定性进行量化，利用多源土壤水分数据的时间信息，采用迭代融合的思想，保证了融合后高分辨率土壤水分数据的时空完整性和准确性。对融合结果的分析评价表明，通过T-BME方法，细分辨率土壤水分数据时空完整度显著提升，平均时间完整度由10.07%提高到92.56%，平均空间完整度由10.07%提高到92.27%。融合方法改善了数据的精度，融合前后高空间分辨率数据与实测数据的相关系数分别为0.34和0.62。融合后的数据与原始数据在空间分布和时间序列变化方面保持一致，但融合后数据的空间纹理特征更精细。当时间相关性范围内缺乏高空间分辨率输入数据时，融合结果的局部空间细节信息有所减少。

（4）针对辐射计和雷达数据融合后空间细节信息较少的问题，提出了融合主动微波、被动微波和光学/红外数据信息的方法。在前一部分研究的基础上，加入高分辨率的光学/红外数据源，进一步改善融合结果的空间特征。结果表明，该融合方法不仅增强了融合结果的空间特征，还进一步提高了数据的完整度，平均时间完整度和平均空间完整度相比于迭代融合的结果分别提高了2.03%和1.29%。融合结果精度可靠，与实测站点数据的相关性为0.59。融合结果的其他评价指标，如RMSE、Bias、ubRMSE，也均与高精度的被动微波数据相一致。融合结果的空间分布格局和时间变化趋势与原始数据相一致。

Land surface soil moisture refers to the water content stored in the upper layers of soil and is a crucial variable in surface energy balance and water cycling, controlling the interactions between the land surface and the atmosphere. Accurate and spatiotemporally complete soil moisture information is crucial for applications such as climate prediction, flood forecasting, and drought assessment. Although satellite remote sensing has been widely used to obtain global soil moisture, due to limitations of orbit coverage, sensor spatial resolution and retrieval algorithms, existing soil moisture products usually have incomplete spatiotemporal coverage and low spatial resolution, making it difficult to meet the requirements. Although existing data fusion and downscaling methods can improve the spatial resolution and spatiotemporal completeness of soil moisture data, they often cannot take into account the requirements of accuracy, spatiotemporal completeness, and spatiotemporal resolution. Traditional methods to construct spatiotemporally complete soil moisture data require a large amount of auxiliary data and complex calculations, which are inefficient and result in unstable results. In contrast, spatiotemporal statistical methods effectively handle missing data and improve computational efficiency based on the inherent spatiotemporal relationship of variables, while maintaining the original spatiotemporal resolution and accuracy of the data. Therefore, this paper is based on the theory of spatiotemporal statistics and uses the high-precision SMAP satellite products to study methods of constructing complete spatiotemporal soil moisture data at different resolutions. The main research results are as follows:

(1) A spatiotemporal reconstruction method of single satellite soil moisture data based on Fixed Rank Filtering (FRF) was proposed. For the first time, the latest FRF method in the field of spatiotemporal statistics is introduced into the reconstruction of satellite soil moisture products from a single sensor. The method takes advantage of the fixed rank of FRF to reduce spatial dimensions for improved computational efficiency in the spatial dimension and uses Kalman filtering for dynamic estimation in the temporal dimension. The study focused on SMAP enhanced soil moisture data (SM_P_E) with a spatial resolution of 9km in the United States as a case study. The results show that the FRF reconstruction method significantly improves the spatiotemporal completeness of SM_P_E data, and the average spatiotemporal completeness is increased by more than 40%. The accuracy of the soil moisture data reconstructed using FRF is highly consistent with the original data in both pixels with and without original data. The FRF reconstructed data is consistent with the original data in terms of spatial distribution patterns and temporal trends. However, there is a reduction in spatial detail information in localized spatial variations after FRF reconstruction.

(2) Aiming at the problem that spatiotemporal statistical interpolation is difficult to reconstruct the small-scale spatiotemporal variation characteristics of satellite remotely sensed soil moisture with high accuracy in the case of large-scale and long-term continuous missing, the soil moisture product of the land surface model reanalysis is introduced into the satellite soil moisture product reconstruction. We proposed the Temporal Kriging Fusion Method (TKFM) that combines reanalysis products and remote sensing data to generate spatiotemporally complete soil moisture products. This method combines the complementary characteristics of reanalysis products and remote sensing products in terms of spatiotemporal coverage and product accuracy, and utilizes the temporal correlations in soil moisture data to provide optimal estimates for unknown points. This study was conducted using SM_P_E remote sensing data and ERA5-Land reanalysis data in the United States as a case study. The results show that this fusion method generates soil moisture data with refined spatial texture features without sacrificing the spatial and temporal resolution and accuracy of the product. The average temporal completeness of soil moisture products increased from 42.94% to 85.44%, and the average spatial completeness increased from 42.80% to 85.65. The accuracy of the fusion results is stable. The correlations between the original data, the fusion results, the fusion results in pixels with original data, and the fusion results in pixels without original data and the in situ data are 0.69, 0.68, 0.69 and 0.66 respectively. The spatial distribution and temporal dynamics of the fused data are consistent with the original data. Compared with the FRF reconstruction method, TKFM shows more significant advantages in retaining spatial details, but the spatiotemporal completeness and accuracy stability of the fusion product are slightly lower than those of the FRF reconstruction results.

(3) Aiming at the problem of low spatiotemporal completeness of high spatial resolution soil moisture data, a temporal Bayesian maximum entropy (T-BME) method that iteratively fuses multi-source microwave data is proposed to generate high spatial resolution (3km) spatiotemporally complete soil moisture data. This method quantifies the uncertainty of multi-source data, makes full use of the temporal information from multi-source soil moisture data, and employs an iterative fusion approach to ensure the spatiotemporal completeness and accuracy of the fused high-resolution soil moisture data. Analysis and evaluation of the fusion results show that through the T-BME method, the spatiotemporal completeness of high resolution soil moisture data is significantly improved, with the average temporal completeness increasing from 10.07% to 92.56%, and the average spatial completeness increasing from 10.07% to 92.27%. The fusion method improves the accuracy of the data. The correlations between the original data, the fused data and the in situ data are 0.34 and 0.62 respectively. The fused data is consistent with the original data in both spatial distribution and temporal variations, but the spatial texture features of the fused data are more refined. When there is a lack of high spatial resolution input data within the temporal correlation range, the fused results exhibit a reduction in local spatial detail information.

(4) In response to the issue of reduced spatial details after the fusion of radiometer and radar data, we propose a method that combines active microwave, passive microwave, and optical /infrared data. Building upon the previous research, we introduce high spatial resolution optical /infrared data sources to further enhance the spatial characteristics of the fusion results. The results show that this fusion method not only enhances the spatial characteristics of the fusion results, but also further improves the completeness of the data. In comparison to the results of iterative fusion, the average temporal completeness and average spatial completeness have increased by 2.03% and 1.29%, respectively. The fusion results exhibit reliable accuracy, with a correlation of 0.59 when compared to in situ data. Additionally, other evaluation metrics such as RMSE, Bias, and ubRMSE are consistent with high precision passive microwave data. The spatial distribution pattern and temporal trends of the fusion results are consistent with the original data.