Soil moisture (SM) is a critical component of the land surface hydrological cycle, significantly impacting various sectors such as hydrology, meteorology, and agriculture. Accurate, high-resolution SM data are essential for effective flood forecasting, water resource management, and understanding the soil freeze-thaw processes in cold regions. This study aims to generate 1 km resolution liquid surface SM (SSM) data with a twice-daily update frequency by downscaling SMAP Level-4 SSM data using random forest (RF) and multiple linear regression (MLR) in the source region of the Yellow River (SRYR), by considering the differences in SM changes between freezing and thawing periods. To obtain the SSM data, 16 downscaling schemes of both RF and MLR were designed for each of the three scenarios. In each downscaling process, both land surface temperature (LST) and normalized difference vegetation index (NDVI) were utilized in MLR and RF models, alongside various combinations of additional variables such as albedo, elevation, leaf area index (LAI), soil texture. Results showed that during the freezing period, RF produced superior SSM estimates when supplemented with NDVI, LST, albedo, elevation, LAI, and soil texture. MLR was more effective during the thawing period when paired with NDVI, LST, elevation, LAI, and soil texture. During the freezing period, the downscaled SMAP SSM exhibited average R, RMSE, ubRMSE of 0.76, 0.029 m3·m-3, and 0.023 m3·m-3, respectively, when compared with in-situ measurements. During the thawing period, the average R, MAE, RMSE, and ubRMSE between the downscaled SMAP SSM and in-situ measurements were 0.52, 0.057 m3·m-3, 0.067 m3·m-3, and 0.054 m3·m-3, respectively, compared to 0.45, 0.070 m3·m-3, 0.083 m3·m-3, and 0.060 m3·m-3 for the original SMAP SSM. Thus, the research significantly enhances both the accuracy and spatial resolution of SMAP SSM estimations, underscoring its vital role in advancing hydrological studies within the SRYR.

中文翻译：

---

通过考虑黄河源区冰冻期和解冻期之间的差异，从 SMAP SSM 生成高分辨率（1 公里）表层土壤水分


土壤水分 （SM） 是地表水文循环的关键组成部分，对水文、气象和农业等各个部门产生重大影响。准确、高分辨率的 SM 数据对于有效的洪水预报、水资源管理和了解寒冷地区的土壤冻融过程至关重要。本研究旨在通过考虑冻结期和解冻期之间 SM 变化的差异，在黄河源区 （SRYR） 使用随机森林 （RF） 和多元线性回归 （MLR） 缩小 SMAP Level-4 SSM 数据，生成每天更新两次的 1 公里分辨率液面 SM （SSM） 数据。为了获得 SSM 数据，针对 3 种情景中的每一个场景设计了 16 种 RF 和 MLR 降尺度方案。在每个降尺度过程中，MLR 和 RF 模型中都使用了地表温度 （LST） 和归一化差值植被指数 （NDVI），以及其他变量的各种组合，例如反照率、海拔、叶面积指数 （LAI）、土壤质地。结果表明，在冰冻期间，当补充 NDVI 、 LST 、 反照率、 海拔、LAI 和土壤质地时，RF 产生了更好的 SSM 估计值。MLR 与 NDVI 、 LST 、 海拔 、 LAI 和土壤质地配对时，在解冻期更有效。在冻结期间，与原位测量相比，缩小的 SMAP SSM 的平均 R、RMSE、ubRMSE 分别为 0.76、0.029 m3·m-3 和 0.023 m3·m-3。在解冻期间，缩小的 SMAP SSM 和原位测量之间的平均 R、MAE、RMSE 和 ubRMSE 分别为 0.52、0.057 m3·m-3、0.067 m3·m-3 和 0.054 m3·m-3，而 0.45、0.070 m3·m-3、0.083 m3·m-3 和 0。060 m3·m-3 为原始 SMAP SSM。因此，该研究显著提高了 SMAP SSM 估计的准确性和空间分辨率，强调了它在推进 SRYR 内水文研究方面的重要作用。