土壤水分（Soil Moisture, SM）是影响水文循环过程、生物生态过程和生物地球化学循环过程的关键地表状态变量，是全球水循环、碳循环及能量循环的基本组成部分。当前业务化发布的SM中低分辨率产品难以满足中小尺度上的应用需求。为获取高空间分辨率、高精度的SM产品，进一步提高SM产品的应用价值，SM产品的降尺度模型、真实性检验、业务化生产等正成为当前遥感领域的研究热点。

青藏高原（Qinghai-Tibet Plateau，QTP）及其周边地区气候条件与地表水热状况复杂，在该地区已有一些降尺度SM方法与真实性检验的研究。然而，这些研究多基于已有的降尺度模型，对模型适用性和模型在区域尺度上的优化探讨不足，进行区域尺度高精度SM估算难度较大。同时，当前降尺度SM产品真实性检验多数直接移植了业务化中低分辨率SM产品的验证方法，缺少配套的高精度SM参考产品与高密度SM地面观测，在策略上缺少定量分析，对已有的SM产品在QTP上的可能误差或者不确定性的认识较为不足。另外，相关研究普遍在验证中以像元内有限的点尺度观测代表像元尺度真值，忽略SM产品、SM地面观测在遥感像元尺度上的空间异质性和对真值的空间代表性。

基于此现状，本研究提出了一种新的SM降尺度探索性策略——耦合小波分析的优化SM降尺度（optimized wavelet-coupled fitting method, OWCM），以提高四种现有通用SM降尺度方法（弹性网络回归、面到面回归克里金、随机森林和神经网络，general model，GM）的模型拟合能力。具体地，基于2016年全球陆表特征参量（GLASS）叶面积指数（LAI）、反照率（Albedo）、植被覆盖度（LAI）产品数据以及中国西部1-km全天候地表温度数据，以QTP粗分辨率SM产品（RFSM, 0.25°×0.25°）为降尺度源，基于四种通用拟合模型与对应的四种耦合小波分析模型，生成8个0.01°×0.01°降尺度高分辨率SM数据集（GM, 基于原始通用拟合模型生产的数据集; OWCM, 基于耦合小波分析的通用拟合模型生产的数据集）。之后，与地面SM实测数据、原始粗分辨率RFSM进行真实性检验，同时在时间序列上使用探索性的数据分析方法（EDA）与RFSM进行比较。结果发现，在与地面观测数据的直接验证中，OWCM结果集在选定的低空间异质性地面观测节点上的R 和 ubRMSE 分别介于 0.6734 和 0.9703 和 0.0165 m³/m³ 和 0.0895 m³/m³ 之间，这些指标优于对应的GM结果集——分别介于 0.0054 和 0.8591 和 0.0300 m³/m³ 和 0.1209 m³/m³ 之间；与原始RFSM进行间接验证的结果表明，OWCM 通常比相应的 GM 更有效地捕捉SM的时空动态，并更好地保持降尺度源SM产品的准确性。基于广义可加模型（GAM）的分析结果表明，空间异质性仍对拟合后的精细分辨率SM数据集精度产生明显影响，在空间异质性较高的区域拟合精度难以保持在较高水平。总的来说，耦合小波变换可以在SM拟合模型中作为提高粗分辨率SM遥感产品时空缺失填补、预测和降尺度模型性能与精度的通用框架。

进一步地，优化降尺度SM的真实性检验过程，基于QTP祁连山地区天峻加密SM观测网（TJ-WSN）非加密区与加密区SM观测数据，结合SM时间序列分解与误差分析模型，直接验证、TC模型与ETC模型验证的结果均表明增加地面观测点数量与密度、合理利用稀疏地面观测的降尺度SM真实性检验策略与方案可获取更具真值代表性的像元尺度地面SM观测值，从而提升降尺度产品真实性检验结果准确性。在不同目标分辨率上，随着地面测点数量的增加，多数评估像元的均方根误差RMSE不断减小，最终稳定在约0.03-0.05 m³/m³之间；相关系数R不断增加，最终稳定在约0.8以上；误差标准差不断减小，最终稳定在约0.04 m³/m³。进一步对SM产品精度与不确定性的影响来源分析讨论的结果发现，相比于降尺度过程，降尺度目标分辨率的差异影响更大，为降尺度产品引入了更多不确定性，但这部分影响总体不大。GAM模型结果表明，各因子空间异质性在较小时能够减小不确定性，而较大时能增加产品的不确定性，并且发现植被与地形因子的空间异质性对产品不确定性的影响更大。同时，验证策略能够在一定程度上减轻由TC模型中参考SM产品分辨率不足带来的负面影响。总的来说，在降尺度SM的真实性检验中，适当增加地面测点数量、合理使用像元内的稀疏测点的验证方案与策略能够改善由于地面测点代表性不够造成的评价指标低估或高估现象，提高SM真实性检验结果的准确性，并且在不同分辨率、不同参考SM数据下这一策略都适用，可用于不同地面观测环境下的降尺度SM产品的真实性检验研究中。

Soil moisture（SM）is a key surface environmental variable and fundamental component, affecting the hydrological cycle and bioecological process. In recent years, released commercial SM products with coarse resolution are challenging to meet SM application requirements at medium and small scales. In order to obtain SM products with high spatial resolution and precision, researches on SM downscaling model, evaluation, and commercialization are becoming hotspots.

Qinghai-Tibet Plateau（QTP）and its surrounding areas have complex climatic conditions and vegetation coverage. Some SM downscaling researches have been studied in this region. However, current SM analyses generally rely on existing SM fitting methods; few studies have focused on general method optimization, which is directly related to the subsequent accuracy of resulting SM products. At the same time, most of the current downscaled SM validations directly transplant the verification methods of coarse-scale SM products, for the lack of supporting high-precision SM, high-density SM ground observations and quantitative validation strategy. There is limited awareness of possible errors or uncertainties of downscaled SM products. In addition, researchers generally use limited point-scale observations to represent the pixel-scale true value in verification, ignoring the spatial heterogeneity at pixel scale and the spatial representation of ground SM observations.

Based on the question, this work presents a novel strategy（i.e., optimized wavelet-coupled fitting method, OWCM）to enhance the fitting accuracy of general methods（GMs）by introducing a wavelet transform（WT）technique. Four separate GMs are selected, i.e., elastic network regression, area-to-area regression kriging, random forest regression, and neural network regression. The fitting procedures are then tested within a downscaling analysis implemented between aggregated Global Land Surface Satellite products（LAI, FVC, Albedo）, Thermal and Reanalysis Integrating Medium-resolution Spatial-seamless LST and Random Forest Soil Moisture（RFSM）data sets in both the WT space and the regular space. Then, eight fine-resolution SM datasets mapped from the trained GMs and OWCMs are then analyzed using direct comparisons with in-situ SM measurements and indirect intercomparisons between the aggregated OWCM-/GM-derived SM and RFSM. Direct verification with ground observation data shows that the R and ubRMSE of OWCM-derived SM results with the selected low spatial heterogeneity ground observation nodes are between 0.6734 and 0.9703, 0.0165 m³/m³ and 0.0895 m³/m³, respectively. These metrics outperformed the corresponding OWCM-derived SM results, which are between 0.0054 and 0.8591, 0.0300 m³/m³ and 0.1209 m³/m³, respectively. Moreover, OWCM-derived SM products represent a significant improvement in terms of their ability to spatially and temporally match RFSM. Although the results of generalized additive model（GAM）show that spatial heterogeneity still substantially impacts the fitting accuracies of both GM and OWCM SM products, the improvements of OWCM over GM are significant. In general, OWCM can be used in SM fitting as a general framework to improve the performance and accuracy.

Further, this work optimizes the verification process of downscaled SM from direct and indirect verification, combining SM time series decomposition and error analysis model（Triple collocation, TC and Extended triple collocation, ETC）, with SM datasets of Tianjun observation network（TJ-WSN）in the QTP Qilian Mountains. The results confirm that the verification strategies of increasing the number and density of ground observation nodes and rationally using sparse ground observations have positive effects on obtaining more representative SM true values and improving the accuracy of downscaled SM verification results. At different target resolutions, with the increase of nodes, the RMSE of most evaluation pixels decreases continuously, and finally stabilizes between about 0.03-0.05 m³/m³; the R increases continuously, and finally stabilizes above about 0.8; the standard deviation of the error continued to decrease, and finally stabilizes at about 0.04 m³/m³. Then, the results of further analysis on the influence sources of SM product accuracy and uncertainty find that compared with the downscaling process, the resolution difference of downscaling target has greater impacts and introduce more uncertainty, but this effect is generally insignificant. GAM model shows that the spatial heterogeneity of factors can reduce the uncertainty when it is small, while it can increase the uncertainty when it is large, and the spatial heterogeneity of vegetation and terrain factors has a greater impact. Moreover, the verification strategy can mitigate the negative effects caused by insufficient resolution of the reference SM products in TC model to some extent. Based on the above process, validation strategies to downscaled SM products are constructed to further improve the accuracy of SM downscaled product evaluation.