Spatiotemporal Dynamics and Driving Mechanisms of Soil Salinization in Northwest China's Oasis–Desert Regions: A 20‐Year Remote Sensing and Machine Learning Analysis (2000–2020)

Abstract

Soil salinization is a major threat to land productivity, food production, and ecosystem balance in Northwestern China. However, in the oasis–desert transition zone, research on the remote sensing monitoring of saline–alkaline land over longer time scales that integrate natural and anthropogenic factors is relatively lacking. Moreover, the application of machine learning for the quantitative analysis of the multifactor interactions of salinization remains limited. In this study, a soil salinization detection index was constructed, and a random forest (RF) algorithm was used to systematically investigate the spatiotemporal evolution patterns and dominant driving mechanisms of regional soil salinization. From a methodological perspective, we innovatively fused spectral feature indices with machine learning regression algorithms and employed multidimensional data analysis to quantify the effects of various environmental factors on salinization processes. The results revealed that severely salinized soils dominated the study area in 2000, covering 167,445.8 km2. The extent of salinized areas decreased from 2000 to 2010 but increased from 2010 to 2020. During 2000–2010, the overall salinization level improved, with increases in nonsalinized, lightly, and moderately salinized areas and a decrease in severely salinized areas. However, from 2010 to 2015, salinization significantly deteriorated in the northern region, with a rise in severe salinization. From 2015 to 2020, the area of severe salinization continued to increase, while the areas of other salinization levels decreased, indicating an overall increasing trend. Spatial analysis revealed divergent trends across geographic sectors, with the western and central regions demonstrating significant improvement in soil quality metrics, whereas the southern and northern zones exhibited progressive degradation patterns. Despite these changes, the overall salinization level improved due to the reduction in moderately and severely salinized areas. The RF model identified sunshine hours as the primary driver of salinization, followed by temperature, evaporation, relative humidity, and precipitation. In contrast, GDP, wind speed, and population density had relatively minor effects.