What Dominates the Variation in Habitat Quality from a “Future” Perspective Based on Interpretable Machine Learning: Evidence from the Mid-Section of the Tianshan Mountains (MSTM), China

Exploring future habitat quality changes in the Mid-Section of the Tianshan Mountains (MSTM) is crucial for regional biodiversity conservation. This study utilizes climate projection data from CMIP6 and integrates the SD-PLUS-InVEST analytical framework to simulate future LULC and habitat quality under three distinct future scenarios. Additionally, the XGBoost-SHAP model is applied to identify and interpret the key regulatory factors within the modeling framework that influence habitat quality spatial heterogeneity. The results show the following: (1) the projections under the three 2035 scenarios generally follow the development trend of 2020, with continued spread of dry land and construction land, but general reduction in the ecological land, reflecting an intensifying conflict between land development and ecological preservation. (2) Habitat quality varies significantly across scenarios, generally exhibiting a “U-shaped” distribution pattern characterized by larger areas of high and low quality and smaller areas of moderate quality. Within the SSP5–8.5 scenario, habitat quality is relatively poor, accompanied by pronounced spatial heterogeneity and imbalance. (3) NDVI is identified as the dominant factor influencing habitat quality spatial heterogeneity, followed by GDP, TEM, and DEM. Although the influence of these factors varies slightly across scenarios, their relative importance remains generally consistent, reflecting the structural stability and response coherence of the ecosystem.