The spatiotemporal pattern and asymmetry characteristics of the normalized difference vegetation index (NDVI) in Xinjiang were analyzed on multiple scales. A multi-model attribution analysis framework that combined a geodetector model (GD), geographically weighted regression (GWR), and random forest (RF) was constructed, since previous efforts using these approaches individually were not able to capture both linear and nonlinear effects. The action laws of contribution degree identification, spatial non-stationarity analysis, and response threshold exploration of NDVI driving factors were also analyzed. The results showed that: (1) the annual mean NDVI in Xinjiang from 2000 to 2021 was 0.106, and overall macroscopic pattern was high in mountainous areas and low in basins. The interannual NDVI exhibited a fluctuating and slightly increasing trend, while the summer NDVI increased the fastest. The asymmetric change trend of the NDVI between seasons was the strongest in the Altay Mountains and Yili River Valley. (2) The NDVI first increased and then decreased with increasing elevation, reaching a peak at a height ranging 2-3 km. The NDVI was highly heterogeneous in mountainous and oasis areas and relatively homogeneous in basins. (3) A scale effect was observed. The detection results of the GD model differed between the Xinjiang and mountain scales. (4) Temperature (Tem), relative humidity (Rh), and precipitation (Pre) had positive effects on NDVI changes, whereas land surface temperature (LST) and summer temperature had negative effects. The threshold of LST was 9 degrees C in summer, and the temperature threshold was 25 degrees C. Our results provide guidance for analyzing the causes and ecological effects of vegetation growth.