青藏高原玛曲网络日尺度多层土壤水分验证数据集（2021-2025）

基于玛曲网络测5、10、20、40cm土壤水分观测数据，结合与土壤水分相关的地表特征因子，本研究以随机森林（RF）为预测模型发展了长时序土壤水分廓线观测的时间序列外推方法，并将现有的2008年-2019年观测进行时间外推。具体而言，将来自MODIS数据的地表温度、归一化植被指数，来自ERA5-Land的降水和蒸发，来自SRTM DEM的坡度、坡向、海拔和经纬度，以及来自SoilGrids的土壤有机质、黏土和砂土作为输入特征，将15分钟的土壤水分观测做日平均处理后作为模型训练目标训练模型。完成模型训练之后，用2021年11月至2025年10月的特征输入模型最终生成对应时间的多层土壤水分数据。与地面观测相比，本数据与具有非常高的相关性且时序完整，可作为验证卫星、模式土壤水分产品的基准。

Based on the soil moisture observation data at 5, 10, 20 and 40 cm depths from the Maqu Network, combined with surface characteristic factors related to soil moisture, this study developed a time series extrapolation method for long-term soil moisture profile observations using Random Forest (RF) as the predictive model, and conducted time extrapolation on the existing observations from 2008 to 2019. Specifically, land surface temperature and Normalized Difference Vegetation Index (NDVI) derived from MODIS data, precipitation and evaporation from ERA5-Land, slope, aspect, elevation, latitude and longitude from SRTM DEM, as well as soil organic matter, clay and sand from SoilGrids were used as input features. The 15-minute interval soil moisture observations were processed via daily averaging to serve as the training targets for model training. After completing the model training, the features from November 2021 to October 2025 were input into the model to finally generate multi-layer soil moisture data corresponding to the target time period. Compared with in-situ ground observations, this dataset exhibits extremely high correlation and complete time series, and can serve as a benchmark for validating satellite and model-based soil moisture products.