Soil salinization significantly jeopardizes agricultural productivity and ecological stability in southern Xinjiang’s oasis regions, highlighting the urgent need to examine its spatial–temporal trends and driving mechanisms for improved resource management. Utilizing soil salinity measurements collected in 2010 and 2023, the current research applied multiple environmental variables processed via the Google Earth Engine (GEE) platform to evaluate the predictive capability of four machine learning algorithms—random forest (RF), Gradient Boosting Decision Tree (GBDT), Classification and Regression Tree (CART), and Support Vector Machine (SVM)—for accurate large-scale salinity mapping. Subsequently, a piecewise structural equation model (piecewiseSEM) was employed to quantitatively analyze the driving factors of soil salinization. Correlation analysis revealed seven critical variables—Red, NDSI, kNDVI, SDI, ET, elevation, and SM—as the most influential among the 41 environmental factors assessed for their impact on soil salinity. The performance evaluation ranked the models as follows: RF > GBDT > SVM > CART, with RF achieving the highest predictive accuracy (R2 = 0.756, RMSE = 2.265 g·kg−1, MAE = 1.468 g·kg−1). Between 2010 and 2023, soil salinization severity in the region exhibited a slight overall decrease; however, the extent of this reduction was relatively modest. The proportion of moderately and severely salinized areas declined, accompanied by reduced spatial variability, whereas the extent of mildly salinized soils increased markedly. These findings imply that soil salinity primarily experiences internal redistribution within the surface layers, with limited downward leaching. Evapotranspiration (ET) and soil moisture (SM) were identified as the dominant drivers affecting salinity dynamics during both periods, with the influence of SM becoming more pronounced over time. This trend highlights that in conditions of limited natural variability, human-induced irrigation practices have emerged as the primary regulator of soil salinity levels. The findings of this study provide novel methodologies and data support for the monitoring and prevention of soil salinization in arid regions.
Soil bulk density is a crucial indicator for assessing soil matter storage and soil quality. Due to the complexity of sampling soil bulk density, particularly in deeper layers, it is essential to study the spatial distribution patterns of soil bulk density and their influencing factors. To address the gap in large-scale studies of vertical (from surface to deeper layers) and horizontal (across a broad area) variations in soil bulk density in arid regions, this study focuses on Changji Prefecture, located in the central northern slope of the Tianshan Mountains and characterized by typical vertical zonation. By integrating classical statistics, geostatistics, and geographic information systems (GISs), this study investigates the spatial distribution patterns and driving factors of soil bulk density. The results indicate that soil bulk density in Changji Prefecture increases with soil depth, with significantly lower values in the surface layer than in deeper layers. Spatially, despite minimal variation in latitude, there is considerable elevation difference within the study area, with the lowest elevations in the central region. Soil bulk density exhibits a spatial distribution pattern of higher values in the northeast (desert areas) and lower values in the southwest (forest areas). The nugget effect in the surface layer (0–20 cm) is substantial at 44.9%, while the deeper layers (20–100 cm) show nugget effects below 25%, suggesting that the influence of both natural and anthropogenic factors on deep soil bulk density is limited and mainly affects the surface layer. Stepwise regression analysis indicates that among topographic factors, slope and elevation are the primary controls of spatial variability in soil bulk density across layers. This research demonstrates that, in arid regions, soil bulk density is influenced primarily by natural factors, with limited impact from human activities. These findings provide valuable data support and theoretical guidance for soil management, agricultural planning, and sustainable ecosystem development in arid regions.
Climate change drives land use intensity changes in Xinjiang, a typical inland arid region. There are relatively few studies on the changes in land use intensity under future climate change. For this purpose, this study adopts the Patch-level Land Use Simulation (PLUS) model and the Markov chain model, combined with shared socioeconomic pathways (SSPs). This study uses the PLUS model to make projections of land use/land cover (LULC) in Xinjiang under different climate scenarios for 2025–2060, constructs a land use intensity atlas to visualize regional spatial patterns, and analyzes the driving factors. The results show that under the SSP126 scenario, the cropland area decreases sharply while the forest, grassland, and water area expand rapidly. However, under the SSP245 and SSP585 scenarios, this trend is obviously reversed; the cropland area expands quickly, and the area of grassland and water decreases. In addition, under the SSP126 scenario, the management and control of LULC are strict, and it may be significantly affected by the conversion of cropland to forest, and the change of forest is relatively active. Under the SSP585 scenario, productivity increases, which may exacerbate the use of constructed land, and the change of constructed land is relatively active. Land use intensity may not significantly promote changes in land type proportions in the region. Population density and GDP are key drivers of land use intensity, showing relatively significant spatial heterogeneity. This study conducts research on the trend of LULC changes under different future climate scenarios, providing data support for the sustainable development of LULC and helping the government formulate different policies to cope with future LULC changes.
High-resolution mapping of soil organic carbon (SOC) in arid regions remains challenging. Using Xinjiang as a case study, this research constructed a prediction framework integrating Boruta feature selection with the Random Forest (RF) algorithm to achieve refined mapping of topsoil SOC. Results indicated that: (1) Among the tested machine learning models, the Boruta–RF framework achieved the highest predictive performance (R2 = 0.48, with the lowest RMSE); (2) Evapotranspiration (ET) and Vapor Pressure Deficit (VPD) were dominant drivers, with the stepwise increase in ET and negative inhibition of VPD confirming the decisive role of hydrothermal fluxes in regulating carbon input; (3) The total SOC storage was estimated at approximately 3.20 Pg C. Despite low carbon density, the desert ecosystem contributed 44.33% of the total storage, constituting a massive Sparse Carbon Pool. This study confirms the necessity of incorporating hydrothermal parameters and highlights that neglecting desert ecosystems leads to a significant underestimation of regional carbon storage.