Understanding the impacts of ecological engineering-driven land use and land cover (LULC) redistribution on ecosystem landscape patterns and habitat quality (HQ), and their key drivers at the basin scale remains limited. In this study, Fragstats and the HQ module of InVEST were employed to evaluate LULC change effects on landscape patterns and HQ in the Qinhe River Basin (QRB), a critical transition zone between the Loess Plateau and the North China Plain with vigorous implementation of the Grain for Green Program, utilizing the XGBoostSHAP method to identify the key drivers of HQ. Key findings reveal that during 1990-2022, LULC transitions were characterized by a 9.32 % increase in forest, alongside a 6.09 % reduction in cropland and 6.39 % decrease in grassland, establishing forest as the dominant cover type. Landscape patterns showed reduced fragmentation, with concurrent increases in aggregation, connectivity, and simplification. Over 33 years, 54.6 % of the basin achieved statistically extremely significant improvement, with substantial decline reduction in habitat degradation. HQ enhancement was strongly correlated with cropland/grassland-to-forest conversion. Notably, the XGBoost-SHAP framework uncovered that HQ spatiotemporal variations were jointly driven by natural and anthropogenic factors, with land use intensity and elevation as the main contributing factors. This method further revealed nonlinear relationships: land use intensity exerted a significant negative impact, while elevation, vegetation coverage, and slope showed positive effects with threshold-dependent influences. These outcomes provide critical empirical evidence for evaluating ecological conservation effectiveness in the QRB and offer a transferable framework for global watersheds managing the landscape and habitat impacts of large-scale ecological restoration initiatives.