Objective Soil salinization limits ecosystem stability in arid regions. This study aims to characterize soil salinization dynamics, assess restoration outcomes, and identify response thresholds of key factors, providing a scientific basis for salinization remediation in the lower reaches of the Tarim River, northwestern China.

Method Six sets of Landsat images for 2000, 2005, 2010, 2015, 2020, and 2023, and annual ecological water conveyance data were used to derive the soil salinity index (S) and normalized difference vegetation index (N). The standard deviational ellipse and S-weighted spatial center were used to analyze the orientation and migration of salinization. Theil-Sen trend analysis, the Mann-Kendall test, and spatial overlay analysis were used to identify areas of joint salinization and vegetation change. Geodetector assessed the explanatory power and interactions of environmental factors. A random forest-SHAP model was constructed, and the variable value at which the smoothed SHAP effect crossed zero was defined as the response boundary within the model framework.

Result (1) Salinization decreased overall from 2000 to 2023. The area proportion of severely salinized land and saline soil declined from 80.57% to 20.61%, and the S-weighted spatial center of salinization migrated a cumulative distance of 2 412.64 m upstream. (2) Joint improvement in soil salinity and vegetation accounted for 16.42% of the study area, mainly along both sides of the river and within the areas affected by ecological water conveyance. (3) Vegetation cover was the leading factor explaining the spatial differentiation of salinization, with q of 0.36. Its interaction with ecological water conveyance increased q to 0.45, indicating a strong coupling between vegetation recovery and artificial hydrological recharge. (4) Salinization showed nonlinear model responses to distance from the river, vegetation cover, and annual ecological water conveyance. A salinization-reducing effect occurred within 0–166 m of the river, whereas relatively higher salinization risk occurred within 166–1835 m; the lateral hydrological influence of the river weakened beyond approximately 1 835 m. When N exceeded 0.09 and annual ecological water conveyance exceeded 4.00 × 10⁸ m³, their associations with a lower salinity index strengthened in the model.

Conclusion Salinization in the lower reaches of the Tarim River is improved overall during 2000–2023. Vegetation cover, ecological water conveyance, and distance from the river are important factors associated with its spatial differentiation. The RF-SHAP thresholds are statistical response boundaries under the current data and model framework and provide a reference for evaluating ecological water conveyance and salinization management by zone.