Soil salinization is an important factor that jeopardizes agricultural production and ecological environment. Rapid and accurate acquisition of soil salinity information in farmland is instructive for sustainable agricultural development and land resource management. In order to improve the accuracy of soil salinity prediction under vegetation cover conditions by satellite remote sensing, the eighth Agricultural Division of Xinjiang Production and Construction Corps was taken as the study area in this study. The soil surface (0-20 cm) samples were collected under high fractional vegetation cover conditions in July and August 2023, respectively, and synchronized satellite images were acquired. Sentinel-1, Sentinel-2 and environment variables provide 3 different types of explanatory variables. The dataset A (polarization indices, spectral indices), dataset B (polarization indices, environment variables), dataset C (spectral indices, environment variables), and dataset D (polarization indices, spectral indices, environment variables) were constructed separately from different combinations of Sentinel-1 radar information, Sentinel-2 multispectral information and environment variables. Then, three integrated machine learning algorithms, namely adaptive boosting (AdaBoost), gradient boost regression Tree (GBRT) and eXtreme gradient boosting tree (XGBoost), were applied to construct soil salinity inversion models based on different datasets. The results showed that Models constructed from dataset B (polarization indices and environmental variables) and C (spectral indices and environmental variables) achieved higher prediction accuracies compared to dataset A (polarization indices and spectral indices). It is shown that when environmental variables are involved in the prediction of soil salinity, the model effect is more effective than the model constructed by polarization and spectral indices suggesting that the model effects are more effective than those constructed from polarization and spectral indices. When environmental variables were applied to dataset D together with polarization indices and spectral indices, the prediction accuracy of all models constructed based on dataset D are generally higher than those constructed on dataset A, B, and C, and that the synergy of environmental variables with radar data and multispectral data can effectively improve the model accuracy. Radar information, spectral information and environmental variables are complementary in soil salinity prediction. Based on the correlation analysis, it can be seen that radar information, spectral information and environmental variables can be used as effective characteristic variables for soil salinity prediction in the study area. It was worth noting that the correlation between topographic factors and land surface temperature with soil salinity is relatively high, with the highest correlation between elevation and surface soil salinity (r = 0.52). Considering the spatial characteristics of soil salinity distribution in the study area can provide effective characteristic variables for soil salinity prediction under vegetation cover condition. In all datasets, the XGBoost had the best performance, followed by GBRT, and the AdaBoost had a large validation error. The D-XGBoost model having the highest accuracy with a validation set R² of 0.72, an RMSE of 2.40 g/kg, and an MAE of 1.29 g/kg. The integrated learning algorithms based on the combination of multiple source variables has a strong nonlinear fitting ability. XGBoost can better model the complex nonlinear relationship between soil salinity content and remote sensing information, environmental factors, and obtain ideal fitting results. The joint application of multi-source remote sensing data and integrated learning algorithms can obtain the ideal soil salinity inversion accuracy under vegetation cover conditions. This study provides an effective technical means for real-time dynamic monitoring of soil salinity by satellite remote sensing in farmland to optimize irrigation strategies and manage saline soils comprehensively in Xinjiang.

Water stress has been one of the most serious threat to the crop growth, development and yield quality in agricultural fields. Timely and accurate diagnosis of crop water stress can greatly contribute to the precision irrigation for the crop resilience and yield. In this study, the research object was taken from the summer maize in the typical dryland agricultural area of northwest China. A six-channel multispectral sensor was mounted on a drone to obtain the remote sensing image data of summer maize at the nodulatione and staminate pulling stage in 2022. At the same time, the stomatal conductance and phenotypic parameters of summer maize were also collected. The background was removed by supervised classification. The canopy vegetation index and image texture were obtained using the gray-scale covariance matrix. The sensitive vegetation index, image texture and phenotypic parameters and their combinations were screened out by the Bayesian information criterion and full subset filtering. The summer maize stomatal conductance estimation model was constructed to combine the three types of machine learnings: the extreme learning machine, the random forest, and the back-propagation neural network. The optimal model was mapped to estimate the stomatal conductance. The Pearson correlation coefficient of vegetation index and stomatal conductance were significantly positively correlated, whereas, the canopy reflectance of summer maize was weakly negatively correlated. Different types of image textures at different wavelengths were correlated with the stomatal conductance, and the highest correlation was found in the 550 nm band. The Pearson correlation coefficients between morphological structure phenotypes (plant height, stem thickness and leaf area) and stomatal conductance of summer maize were 0.72, 0.58 and 0.69, respectively, where the three types of phenotypic parameters data were correlated well with stomatal conductance. Vegetation indices with spectral reflectance data were used to assess the overall health and moisture status of the vegetation. Image texture was used to capture the spatial distribution, texture and structural features of crops. Crop phenotypic parameters were then used to reflect the physiological and morphological responses of the crop in a three-dimensional manner, providing visual information about the growth and moisture of the vegetation. The decision coefficients of the crop water stress diagnostic models that constructed from the three information sources increased from 0.728 and 0.750 to 0.841, respectively, compared with the single or two combinations, indicating the great potential to stomatal conductance prediction. The optimal combination of indicators was screened by Bayesian information criterion and full subset screening: DWSI, NDVI, MEA, ENT, plant height and leaf area. The back-propagation neural network model with the three complementary information sources was the optimal model for the water stress diagnosis of summer maize (coefficient of determination of 0.841, root mean square error of 0.043 mol/(m²·s), and mean absolute error of 0.034 mol/(m²·s)). The underestimation of stomatal conductance was significantly improved, compared with the rest models. The inverse map with the optimal model was widely applied to easily and accurately diagnose the crop water stress for the purpose of irrigation strategies and resource allocation. The finding can provide a feasible and accurate diagnosis of water stress in summer maize.

Thermal infrared imaging can be expected to rapidly and cost-saving monitor the soil moisture content in the large-scale farmland. But it is still unclear on the impact of lighting conditions on thermal infrared images. This study aims to explore the impact of direct sunlight or shadow occlusion of the crop canopy and soil on the UAV thermal infrared remote sensing, in order to diagnose the crop water stress and monitor soil moisture content. Summer corn with different irrigation treatments was taken as the research object. The thermal infrared images were divided into four parts: illuminated canopy, shaded canopy, illuminated soil, and shaded soil. The light temperature was extracted from the higher temperature, whereas, the shadow temperature was assumed as the lower temperature. Temperature extraction was used to calculate the 11:00, 13:00, and 15:00 canopy temperature difference (difference between canopy temperature and atmospheric temperature, ΔT), crop water stress index (CWSI), and evaporative fraction (ratio of latent heat flux to effective energy, EF). A comparison was made on three changes in the monitoring effect of the index on soil moisture content after using light temperature (ΔT_L, CWSI_L, EF_L) and shadow temperature (ΔT_S, CWSI_S, EF_S) at different times. The results show that: 1) There was the variation in the monitoring effects of the three indices over time. The EF monitoring effect was better at 11:00 and 15:00, the crop water stress index monitoring effect was better at 13:00. There was the less change in the ΔT monitoring over time. The temperature index should be selected for monitoring soil moisture content using field conditions and the time of flying the drone; 2) The monitoring effect was improved the most at 11:00 in the jointing period, after distinguishing light temperature and shadow temperature. The R²of EF, EF_S, and EF_L were 0.54, 0.65, and 0.78, respectively. The R² of CWSI, CWSI_S, and CWSI_L were 0.47, 0.64, and 0.70, respectively. There was no significant light temperature in the period of tasseling and filling, whereas, the index monitoring effect was significantly reduced using shadow temperature. There was the largest decrease in the CWSI_S at 13:00, compared with the CWSI, where the R² decreases were 0.11 and 0.06, respectively. Therefore, it was very necessary to choose the clear and cloudless weather and avoid cloudy days, when shooting thermal infrared images. Furthermore, the impact of lighting conditions on thermal infrared images was also change with the growth period; 3) The best monitoring soil moisture content was obtained using EF_L at 11:00 in the jointing and tasseling period, and CWSI at 13:00 in the filling period, where the R² of predicting soil moisture content were 0.75, 0.75, and 0.89, respectively. In addition, the soil moisture content was also dominated the monitoring effect. When the soil volumetric moisture content was around 0.23, both EF_L and CWS_I shared the more accurate prediction. Different time points, light temperature and shade temperature were utilized to analyze the monitoring effect of three temperature indexes on soil moisture content. The better monitoring time and index were determined in the three growth periods of summer corn. The finding can provide a strong reference for the UAV thermal infrared monitoring of soil moisture content.