Soil salinization, one of the most severe global land degradation problems, leads to the loss of arable land and declines in crop yields. Monitoring the distribution of salinized soil and degree of salinization is critical for management, remediation, and utilization of salinized soil; however, there is a lack of thorough assessment of various data sources including remote sensing and landscape characteristics for estimating soil salinity in arid and semi-arid areas. The overall goal of this study was to develop a framework for estimating soil salinity in diverse landscapes by fusing information from satellite images, landscape characteristics, and appropriate machine learning models. To explore the spatial distribution of soil salinity in southern Xinjiang, China, as a case study, we obtained 151 soil samples in a field campaign, which were analyzed in laboratory for soil electrical conductivity. A total of 35 indices including remote sensing classifiers (11), terrain attributes (3), vegetation spectral indices (8), and salinity spectral indices (13) were calculated or derived and correlated with soil salinity. Nine were used to model and estimate soil salinity using four predictive modelling approaches: partial least squares regression (PLSR), convolutional neural network (CNN), support vector machine (SVM) learning, and random forest (RF). Testing datasets were divided into vegetation-covered and bare soil samples and were used for accuracy assessment. The RF model was the best regression model in this study, with R² = 0.75, and was most effective in revealing the spatial characteristics of salt distribution. Importance analysis and path modeling of independent variables indicated that environmental factors and soil salinity indices including digital elevation model (DEM), B10, and green atmospherically resistant vegetation index (GARI) showed the strongest contribution in soil salinity estimation. This showed a great promise in the measurement and monitoring of soil salinity in arid and semi-arid areas from the integration of remote sensing, landscape characteristics, and using machine learning model.

Automatic road extraction from very-high-resolution remote sensing images has become a popular topic in a wide range of fields. Convolutional neural networks are often used for this purpose. However, many network models do not achieve satisfactory extraction results because of the elongated nature and varying sizes of roads in images. To improve the accuracy of road extraction, this paper proposes a deep learning model based on the structure of Deeplab v3. It incorporates squeeze-and-excitation (SE) module to apply weights to different feature channels, and performs multi-scale upsampling to preserve and fuse shallow and deep information. To solve the problems associated with unbalanced road samples in images, different loss functions and backbone network modules are tested in the model’s training process. Compared with cross entropy, dice loss can improve the performance of the model during training and prediction. The SE module is superior to ResNext and ResNet in improving the integrity of the extracted roads. Experimental results obtained using the Massachusetts Roads Dataset show that the proposed model (Nested SE-Deeplab) improves F1-Score by 2.4% and Intersection over Union by 2.0% compared with FC-DenseNet. The proposed model also achieves better segmentation accuracy in road extraction compared with other mainstream deep-learning models including Deeplab v3, SegNet, and UNet.

Accurate estimates of the spatial distribution of total nitrogen (TN) in soil are fundamental for soil quality assessment, decision making in land management, and global nitrogen cycle modeling. In China, current maps are limited to individual regions or are of coarse resolution. In this study, we compiled a new 90-m resolution map of soil TN in China by the weighted summation of random forest and extreme gradient boosting. After harmonizing soil data from 4022 soil profiles into a fixed soil depth (0–20 cm) by equal area spline, 18 environmental covariates were employed to characterize the spatial pattern of soil TN in topsoil across China. The accuracy assessments from independent validation data showed that the weighted model averaging gave the best predictions with an acceptable R² (0.41). The prediction map showed that high-value areas of soil TN were mainly distributed in the eastern Tibetan Plateau, central Qilian Mountains and the north of the Greater Khingan Range. Climate factors had a considerable influence on the variation of the soil TN, and land-use types played a pivotal part in each climate zone. This high-resolution and high-quality soil TN data set in China can be very useful for future inventories of soil nitrogen, assessments of soil nutrient status, and management of arable land.

Soil salinization is one of the major degradation processes threatening food security and sustainable development. Detailed soil salinity information is increasingly needed to tackle this global challenge for improving soil management. Soil-visible and near-infrared (Vis-NIR) spectroscopy has been proven to be a potential solution for estimating soil-salinity-related information (i.e., electrical conductivity, EC) rapidly and cost-effectively. However, previous studies were mainly conducted at the field, regional, or national scale, so the potential application of Vis-NIR spectroscopy at a global scale needs further investigation. Based on an extensive open global soil spectral library (61,486 samples with both EC and Vis-NIR spectra), we compared four spectral predictive models (PLSR, Cubist, Random Forests, and XGBoost) in estimating EC. Our results indicated that XGBoost had the best model performance (R² of 0.59, RMSE of 1.96 dS m⁻¹) in predicting EC at a global scale, whereas PLSR had a relatively limited ability (R² of 0.39, RMSE of 2.41 dS m⁻¹). The results also showed that auxiliary environmental covariates (i.e., coordinates, elevation, climatic variables) could greatly improve EC prediction accuracy by the four models, and the XGBoost performed best (R² of 0.71, RMSE of 1.65 dS m⁻¹). The outcomes of this study provide a valuable reference for improving broad-scale soil salinity prediction by the coupling of the spectroscopic technique and easily obtainable environmental covariates.

Due to the importance of soil organic carbon (SOC) in supporting ecosystem services, accurate SOC assessment is vital for scientific research and decision making. However, most previous studies focused on single soil depth, leading to a poor understanding of SOC in multiple depths. To better understand the spatial distribution pattern of SOC in Northeast and North China Plain, we compared three machine learning algorithms (i.e., Cubist, Extreme Gradient Boosting (XGBoost) and Random Forest (RF)) within the digital soil mapping framework. A total of 386 sampling sites (1584 samples) following specific criteria covering all dryland districts and counties and soil types in four depths (i.e., 0–10, 10–20, 20–30 and 30–40 cm) were collected in 2017. After feature selection from 249 environmental covariates by the Genetic Algorithm, 29 variables were used to fit models. The results showed SOC increased from southern to northern regions in the spatial scale and decreased with soil depths. From the result of independent verification (validation dataset: 80 sampling sites), RF (R²: 0.58, 0.71, 0.73, 0.74 and RMSE: 3.49, 3.49, 2.95, 2.80 g kg⁻¹ in four depths) performed better than Cubist (R²: 0.46, 0.63, 0.67, 0.71 and RMSE: 3.83, 3.60, 3.03, 2.72 g kg⁻¹) and XGBoost (R²: 0.53, 0.67, 0.70, 0.71 and RMSE: 3.60, 3.60, 3.00, 2.83 g kg⁻¹) in terms of prediction accuracy and robustness. Soil, parent material and organism were the most important covariates in SOC prediction. This study provides the up-to-date spatial distribution of dryland SOC in Northeast and North China Plain, which is of great value for evaluating dynamics of soil quality after long-term cultivation.

Climate change has proven to have a profound impact on the growth of vegetation from various points of view. Understanding how vegetation changes and its response to climatic shift is of vital importance for describing their mutual relationships and projecting future land–climate interactions. Arid areas are considered to be regions that respond most strongly to climate change. Xinjiang, as a typical dryland in China, has received great attention lately for its unique ecological environment. However, comprehensive studies examining vegetation change and its driving factors across Xinjiang are rare. Here, we used the remote sensing datasets (MOD13A2 and TerraClimate) and data of meteorological stations to investigate the trends in the dynamic change in the Normalized Difference Vegetation Index (NDVI) and its response to climate change from 2000 to 2019 across Xinjiang based on the Google Earth platform. We found that the increment rates of growth-season mean and maximum NDVI were 0.0011 per year and 0.0013 per year, respectively, by averaging all of the pixels from the region. The results also showed that, compared with other land use types, cropland had the fastest greening rate, which was mainly distributed among the northern Tianshan Mountains and Southern Junggar Basin and the northern margin of the Tarim Basin. The vegetation browning areas primarily spread over the Ili River Valley where most grasslands were distributed. Moreover, there was a trend of warming and wetting across Xinjiang over the past 20 years; this was determined by analyzing the climate data. Through correlation analysis, we found that the contribution of precipitation to NDVI (R² = 0.48) was greater than that of temperature to NDVI (R² = 0.42) throughout Xinjiang. The Standardized Precipitation and Evapotranspiration Index (SPEI) was also computed to better investigate the correlation between climate change and vegetation growth in arid areas. Our results could improve the local management of dryland ecosystems and provide insights into the complex interaction between vegetation and climate change.