A method is proposed for the production of downscaled soil moisture active passive (SMAP) soil moisture (SM) data by combining optical/infrared data with synthetic aperture radar (SAR) data based on the random forest (RF) model. The method leverages the sensitivity of active microwaves to surface SM and the triangle/trapezium feature space among vegetation indexes (VIs), land surface temperature (LST), and SM. First, five RF architectures (RF1–RF5) were trained and tested at 9 km. Second, a comparison was performed for RF1–RF5, and were evaluated against in situ SM measurements. Third, two SMAP-Sentinel active–passive SM products were compared at 3 km and 1 km using in situ SM measurements. Fourth, the RF5 model simulations were compared with the SMAP L2_SM_SP product based on the optional algorithm at 3 km and 1 km resolutions. The results showed that the downscaled SM based on the synergistic use of optical/infrared data and the backscatter at vertical–vertical (VV) polarization was feasible in semi-arid areas with relatively low vegetation cover. The RF5 model with backscatter and more parameters from optical/infrared data performed best among the five RF models and was satisfactory at both 3 km and 1 km. Compared with L2_SM_SP, RF5 was more superior at 1 km. The input variables in decreasing order of importance were backscatter, LST, VIs, and topographic factors over the entire study area. The low vegetation cover conditions probably amplified the importance of the backscatter and LST. A sufficient number of VIs can enhance the adaptability of RF models to different vegetation conditions.

NDVI (Normalized difference vegetation index) is a critical variable for monitoring climate change, studying ecological balance, and exploring the pattern of regional phenology. Traditional neural network models only consider image features in time series prediction, while historical data and its changes play an important role in time series forecasting. For this study, we proposed convolutional neural networks (CNN) combined feature engineering forecasting model (SF-CNN), which integrated both the advantages of image characteristics learned from CNN and statistic characteristics calculated by historical data related to the forecast period to improve the accuracy of NDVI predictions in the next 3 months with 30-day interval at multiple complex areas. To intuitively show the performance of SF-CNN, it was compared with CNN using the same parameters. Results mainly showed that (1) in terms of visual analysis, the texture, pattern, and structure of predicted NDVI using SF-CNN are similar to the observed NDVI, and SF-CNN exhibits strong generalization ability; (2) in terms of quantitative assessment, SF-CNN generally outperforms CNN, and it can improve the reliability and robustness for predicting NDVI through simple statistical characteristics while reducing the uncertainties; (3) SF-CNN can learn seasonal and sudden changes in four different and complex study areas with considerable accuracy and without extra data.

In current upscaling of in situ surface soil moisture practices, commonly used novel statistical or machine learning-based regression models combined with remote sensing data show some advantages in accurately capturing the satellite footprint scale of specific local or regional surface soil moisture. However, the performance of most models is largely determined by the size of the training data and the limited generalization ability to accomplish correlation extraction in regression models, which are unsuitable for larger scale practices. In this paper, a deep learning model was proposed to estimate soil moisture on a national scale. The deep learning model has the advantage of representing nonlinearities and modeling complex relationships from large-scale data. To illustrate the deep learning model for soil moisture estimation, the croplands of China were selected as the study area, and four years of Visible Infrared Imaging Radiometer Suite (VIIRS) raw data records (RDR) were used as input parameters, then the models were trained and soil moisture estimates were obtained. Results demonstrate that the estimated models captured the complex relationship between the remote sensing variables and in situ surface soil moisture with an adjusted coefficient of determination of ${\overline{\mathrm{R}}}^2$ = 0.9875 and a root mean square error (RMSE) of 0.0084 in China. These results were more accurate than the Soil Moisture Active Passive (SMAP) active radar soil moisture products and the Global Land data assimilation system (GLDAS) 0–10 cm depth soil moisture data. Our study suggests that deep learning model have potential for operational applications of upscaling in situ surface soil moisture data at the national scale.

China is frequently subjected to local and regional drought disasters, and thus, drought monitoring is vital. Drought assessments based on available surface soil moisture (SM) can account for soil water deficit directly. Microwave remote sensing techniques enable the estimation of global SM with a high temporal resolution. At present, the evaluation of Soil Moisture Active Passive (SMAP) SM products is inadequate, and L-band microwave data have not been applied to agricultural drought monitoring throughout China. In this study, first, we provide a pivotal evaluation of the SMAP L3 radiometer-derived SM product using in situ observation data throughout China, to assist in subsequent drought assessment, and then the SMAP-Derived Soil Water Deficit Index (SWDI-SMAP) is compared with the atmospheric water deficit (AWD) and vegetation health index (VHI). It is found that the SMAP can obtain SM with relatively high accuracy and the SWDI-SMAP has a good overall performance on drought monitoring. Relatively good performance of SWDI-SMAP is shown, except in some mountain regions; the SWDI-SMAP generally performs better in the north than in the south for less dry bias, although better performance of SMAP SM based on the R is shown in the south than in the north; differences between the SWDI-SMAP and VHI are mainly shown in areas without vegetation or those containing drought-resistant plants. In summary, the SWDI-SMAP shows great application potential in drought monitoring.

Land surface temperature (LST) is one of the most valuable variables for applications relating to hydrological processes, drought monitoring and climate change. LST from satellite data provides consistent estimates over large scales but is only available for cloud-free pixels, greatly limiting applications over frequently cloud-covered regions. With this study, we propose a method for estimating all-weather 1 km LST by combining passive microwave and thermal infrared data. The product is based on clear-sky LST retrieved from Moderate-resolution Imaging Spectroradiometer (MODIS) thermal infrared measurements complemented by LST estimated from the Advanced Microwave Scanning Radiometer Version 2 (AMSR2) brightness temperature to fill gaps caused by clouds. Terrain, vegetation conditions, and AMSR2 multiband information were selected as the auxiliary variables. The random forest algorithm was used to establish the non-linear relationship between the auxiliary variables and LST over the Tibetan Plateau. To assess the error of this method, we performed a validation experiment using clear-sky MODIS LST and in situ measurements. The estimated all-weather LST approximated MODIS LST with an acceptable error, with a coefficient of correlation (r) between 0.87 and 0.99 and a root mean square error (RMSE) between 2.24 K and 5.35 K during the day. At night-time, r was between 0.89 and 0.99 and the RMSE was between 1.02 K and 3.39 K. The error between the estimated LST and in situ LST was also found to be acceptable, with the RMSE for cloudy pixels between 5.15 K and 6.99 K. This method reveals a significant potential to derive all-weather 1 km LST using AMSR2 and MODIS data at a regional and global scale, which will be explored in the future.