Dynamic processes of modern desert contraction: a case study of Buguli and Toklak deserts on the southwestern margin of the Tarim Basin, China

Accurate delineation of desert boundaries and characterization of their spatiotemporal dynamics are essential for understanding desertification processes and evaluating ecological restoration in arid regions. However, conventional large-scale assessments often fail to capture fine-scale boundary dynamics and transitional uncertainties. To address this limitation, this study develops an integrated framework combining multi-source remote sensing, machine learning, and geographical detector analysis to map desert boundaries and quantify their evolution in the Buguli and Toklak deserts on the southwestern margin of the Tarim Basin, China. A high-resolution baseline boundary was established through visual interpretation of sub-metre imagery and extended to multi-temporal analysis using consistent classification of Landsat and Sentinel-2 data. For the 2025 desert boundary extraction, the Random Forest classifier achieved the highest accuracy (overall accuracy = 0.96; Kappa = 0.91) and was applied to reconstruct desert extent for six periods from 2000 to 2025. Results reveal a persistent net contraction of desert areas, with a total decrease of 322 km2. Spatially, boundary changes were concentrated in marginal zones, particularly along the eastern, southern, and southwestern edges, while interior regions remained relatively stable. Land cover transition analysis shows that contraction areas were mainly converted from sandy land to grassland and cropland. However, some areas initially classified as unused land were associated with early-stage cropland reclamation and shelterbelt establishment, reflecting classification uncertainties in transitional landscapes. Geographical detector results indicate that human-related factors exerted the strongest influence on desert boundary dynamics (population density: q = 0.57; GDP: q = 0.49), substantially exceeding the effects of climatic variables. These findings suggest that recent desert contraction has been primarily driven by intensified human activities, while climate plays a secondary but synergistic role. Overall, this study improves understanding of desert contraction processes and provides a scientific basis for evaluating and optimizing desertification control strategies in arid regions.