Monitoring the spatio-temporal variation in soil moisture content (SMC) of the surface soil layer is essential for agriculture and water resource management activities, especially in regions where the socio-economic condition and livelihood depend upon agriculture and allied sectors. In the present study, we have compared different machine learning (ML) and linear regression models to estimate the SMC integrating field observed soil moisture and Sentinel-1 SAR data. Total 56 soil samples were collected from the surface soil layer (~ 5 cm) in correspondence with the passing date of the Sentinel-1 sensor over the study area. The surface SMC was estimated for bare soil areas, which was extracted by applying the threshold values on vegetation and water index maps derived from the Sentinel-2 multispectral data. The univariate linear regression with the co-polarized VV band provided higher accuracy compared to the cross-polarized VH band. However, the multiple linear regression with VV and VH bands indicated similar accuracy as obtained by the VV band alone. The random forest model was observed as the best performing ML model for soil moisture estimation (R2 = 0.87 and 0.93 during modeling and validation, respectively; RMSE: ~ 0.03). The obtained results indicate well accurate surface soil moisture verified with in-situ information collected during the dry rabi crop season (January to March 2019). The maximum SMC was observed for March, followed by February and January, that corroborated with the total monthly precipitation and irrigation activities. The study highlights the potentiality of ML models and Sentinel-1 SAR data for soil moisture estimation, which is useful for policy-level implications and decision making in agriculture and water resource management activities.

中文翻译：

---

通过机器学习模型使用 Sentinel-1 C 波段 SAR 数据估计表面水分含量

监测表层土壤水分含量 (SMC) 的时空变化对于农业和水资源管理活动至关重要，尤其是在社会经济条件和生计依赖于农业和相关部门的地区。在本研究中，我们比较了不同的机器学习 (ML) 和线性回归模型，以估计 SMC 集成现场观察到的土壤水分和 Sentinel-1 SAR 数据。根据 Sentinel-1 传感器在研究区域上的通过日期，从表层土壤层（~ 5 厘米）收集了总共 56 个土壤样品。裸土区域的地表 SMC 是通过对源自 Sentinel-2 多光谱数据的植被和水指数图应用阈值来提取的。与交叉极化 VH 频段相比，具有共极化 VV 频段的单变量线性回归提供了更高的精度。然而，具有 VV 和 VH 波段的多元线性回归表明与单独通过 VV 波段获得的精度相似。随机森林模型被认为是土壤水分估计的最佳 ML 模型（在建模和验证期间，R2 分别为 0.87 和 0.93；RMSE：~ 0.03）。获得的结果表明，地表土壤水分准确度通过在干燥狂犬病作物季节（2019 年 1 月至 3 月）收集的现场信息进行了验证。3 月份观察到最大 SMC，其次是 2 月和 1 月，这与每月总降水量和灌溉活动相吻合。该研究强调了 ML 模型和 Sentinel-1 SAR 数据用于土壤水分估计的潜力，