Analyzing the changes in vegetation under different factors is crucial for ecological protection in arid areas. The spatial-temporal variations of vegetation in the lower reaches of the Tarim River (LRTR) from 2000 to 2020, were analyzed using the Theil-Sen estimator and the Mann-Kendall test. The future trends of NDVI are projected to use the Hurst exponent method. The driving mechanisms of vegetation changes were analyzed using the GeoDetector method and multivariate residual analysis. The NDVI values in the LRTR significantly increased during the study period, indicating good vegetation recovery. The overall vegetation level remains poor and was primarily concentrated around the riverine areas. There is still a risk of vegetation degradation in most areas of the future LRTR. Compared to climate change, vegetation was more affected by human activities. Human activities have helped restore the riparian vegetation and prevented the degradation of vegetation far from the river. Therefore, distance from river channels is the strongest explanatory factor (q = 0.078) for vegetation changes, followed by precipitation, and temperature, while changes in slope have minimal impact on vegetation. Statistics have found that when two factors are combined, their impact on vegetation change is stronger. These findings are beneficial for identifying vegetation evolution patterns in LRTR and providing theoretical support for the government to carry out ecological restoration.

Globally, Central Asian wild fruit forests are critical repositories of wild fruit germplasm resources and provide essential ecosystem services. However, their habitats are facing escalating degradation risks driven by climate warming, shifting precipitation regimes, and intensifying anthropogenic disturbances. Accurately quantifying climate-driven spatiotemporal variations in habitat suitability for keystone wild fruit tree species is therefore an essential prerequisite for formulating targeted conservation and management strategies in arid and semi-arid landscapes. In this study, we applied the maximum entropy (MaxEnt) model to simulate the current (2000–2020 baseline) and future (2030s, 2050s, 2070s) potential suitable habitats of two dominant wild fruit tree species, Malus sieversii (Ledeb.) M.Roem. and Prunus cerasifera Ehrh., in the Ili Valley, a core distribution area of Central Asian wild fruit forests in northwestern China. We integrated rigorously screened species occurrence records with key environmental predictors and characterized future climate conditions using three Shared Socioeconomic Pathways (SSPs; SSP126, SSP245, and SSP585) spanning low to high radiative forcing levels. The model exhibited excellent predictive performance (AUC > 0.85), confirming the robustness and reliability of our habitat suitability simulations. Elevation and annual precipitation were identified as the dominant environmental variables governing habitat suitability for both species, highlighting the critical role of terrain–hydroclimate interactions in maintaining viable dryland refugia for wild fruit forests. Under the baseline climate scenario, the total area of suitable habitats reached 24.014 × 10³ km² for Malus sieversii and 18.990 × 10³ km² for Prunus cerasifera. Future climate projections revealed a consistent and significant contraction trend in suitable habitats for both species, with the magnitude of habitat loss escalating with increasing radiative forcing and longer projection time horizons. Specifically, under the high-emission SSP585 scenario by the 2070s, the suitable habitat area is projected to decline by 7.579 × 10³ km² for Malus sieversii and 9.883 × 10³ km² for Prunus cerasifera relative to the baseline. Our findings delineate climate-vulnerable hotspots of wild fruit forests and provide a robust spatial scientific basis for prioritizing in situ conservation, targeted habitat restoration, and anthropogenic disturbance regulation to support the long-term persistence of these irreplaceable wild fruit germplasm resources under accelerating global climate change.