Considering variations in surface soil moisture (SSM) is essential in improving crop yield and irrigation scheduling. Today, most remotely sensed soil moisture products have difficulties in resolving irrigation signals at the plot scale. This study aims to use Sentinel-1 radar backscatter and Sentinel-2 multispectral imagery to estimate SSM at high spatial (10 m) and temporal resolution (at least 5 days) over an agricultural domain. Three supervised machine learning algorithms, multilayer perceptron (MLP), a convolutional neural network (CNN), and linear regression models, were trained to estimate changes in SSM based on the variation in surface reflectance and backscatter over five different crops. Results showed that CNN is the best algorithm as it understands spatial relations and better represents two-dimensional images. Estimated values for SSM were in agreement with in-situ measurements regardless of the crop type, with $RMSE=0.0292 ({\mathrm{cm}}^3/{\mathrm{cm}}^3)$ and $R^2=0.92$ for the Sentinel-2 derived SSM and $RMSE=0.0317 ({\mathrm{cm}}^3/{\mathrm{cm}}^3)$ and $R^2=0.84$ for the Sentinel-1 soil moisture data. Moreover, a time series of estimated SSM based on Sentinel-1 (SSM-S1), Sentinel-2 (SSM-S2), and SSM derived from SMAP-Sentinel1 was compared. The developed SSM data showed a significantly higher mean SSM state over irrigated agriculture relative to the rainfed cropland area during the irrigation season. The multiple comparisons (fisher LSD) were tested and found that these two groups are different ($p_{value}=0.035$ in 95% confidence interval). Therefore, by employing the maximum likelihood classification on the SSM data, we managed to map the irrigated agriculture. The overall accuracy of this unsupervised classification is 77%, with a kappa coefficient of 65%.

Accurate knowledge of surface and subsurface soil moisture (SM) is essential for hydrologic modeling, weather forecasting, and agricultural water management. NASA’s Soil Moisture Active Passive (SMAP) satellite (level 3) provides ‘surface’ SM with 2–3 days temporal resolution, hence lacks daily and subsurface SM information. This study developed a convolutional neural network–long short-term memory (ConvLSTM) deep learning model to produce ‘daily’ surface (5 cm) and subsurface (25 cm) SM products (9 km) by integrating SMAP level 3 ancillary data, North American Land Data Assimilation System (NLDAS-2; 12 km) SM, and Soil Landscapes of the United States (SOLUS100) digital maps across the contiguous U.S. Two input scenarios were evaluated: scenario 1 used only SMAP ancillary data, while scenario 2 included both SMAP ancillary data and SOLUS100 soil maps. Model evaluation with in situ SM data showed higher accuracy for scenario 2, indicating the importance of soil properties (texture and bulk density) in SM estimation. Coarse-textured soils showed the highest estimation accuracy, followed by medium- and fine-textured soils. The model also performed in estimating subsurface SM than surface SM for most land-cover types. Incorporating SMAP ancillary data and SOLUS100 digital soil maps into the ConvLSTM improved the spatial and temporal estimation of surface and subsurface SM. The results highlight the potential of deep learning for integrating multi-source multi-scale observations for improving SM estimation at large scale.

Accurate real-time information about soil moisture (SM) at a large scale is essential for improving hydrological modeling, managing water resources, and monitoring extreme weather events. This study presents a framework using convolutional long short-term memory (ConvLSTM) network to produce short- (1, 3, and 7 days ahead) and mid-term (14 and 30 days ahead) forecasts of SM at surface (0–10 cm) and subsurface (10–40 and 40–100 cm) soil layers across the contiguous U.S. The model was trained with five-year period (2018–2022) datasets including Soil Moisture Active Passive (SMAP) level 3 ancillary covariables, North American Land Data Assimilation System phase 2 (NLDAS-2) SM product, shortwave infrared reflectance from Moderate Resolution Imaging Spectroradiometer (MODIS), and terrain features (e.g., elevation, slope, curvature), as well as soil texture and bulk density maps from the Soil Landscape of the United States (SOLUS100) database. To develop and evaluate the model, the dataset was divided into three subsets: training (January 2018–January 2021), validation (2021), and testing (2022). The outputs were validated with observed in situ data from the Soil Climate Analysis Network (SCAN) and the United States Climate Reference Network (USCRN) soil moisture networks. The results indicated that the accuracy of SM forecasts decreased with increasing lead time, particularly in the surface (0–10 cm) and subsurface (10–40 cm) layers, where strong fluctuations driven by rainfall variability and evapotranspiration fluxes introduced greater uncertainty. Across all soil layers and lead times, the model achieved a median unbiased root mean square error (ubRMSE) of 0.04 cm³ cm⁻³ with a Pearson correlation coefficient of 0.61. Further, the performance of the model was evaluated with respect to both land cover and soil texture databases. Forecast accuracy was highest in coarse-textured soils, followed by medium- and fine-textured soils, likely because the greater penetration depth of microwave observations improves SM retrieval in sandy soils. Among land cover types, performance was strongest in grasslands and savannas and weakest in dense forests and shrublands, where dense vegetation attenuates the microwave signal and reduces SM estimation accuracy. These results demonstrate that the ConvLSTM framework provides skillful short- and mid-term forecasts of surface and subsurface soil moisture, offering valuable support for large-scale drought and flood monitoring.

Accurate knowledge of surface and subsurface soil moisture (SM) is essential for hydrologic modeling, weather forecasting, and agricultural water management. NASA’s Soil Moisture Active Passive (SMAP) satellite (level 3) provides ‘surface’ SM with 2–3 days temporal resolution, hence lacks daily and subsurface SM information. This study developed a convolutional neural network – long short-term memory (ConvLSTM) deep learning model to produce daily surface (5 cm) and subsurface (25 cm) SM products by integrating SMAP Level 3 ancillary data, North American Land Data Assimilation System Phase 2 (NLDAS-2) SM, and Soil Landscapes of the United States (SOLUS100) digital maps across the contiguous U.S. Two input scenarios were evaluated: scenario 1 used only SMAP ancillary data, and scenario 2 included both SMAP ancillary data and SOLUS100 soil maps. Model evaluation with in-situ SM data showed higher accuracy for scenario 2, indicating the importance of soil properties (texture and bulk density) in SM estimation. Coarse-textured soils showed the highest estimation accuracy, followed by medium- and fine-textured soils. The model also performed in estimating subsurface SM than surface SM for most land-cover types. Incorporating SMAP ancillary data and SOLUS100 digital soil maps into ConvLSTM improved surface and subsurface SM estimation. The results highlight the potential of deep learning for integrating multi-source multi-scale observations for improving SM estimation at large scale.