Under global climate change, extreme climate events are important factors for vegetation dynamics in arid and semi-arid regions. However, a systematic understanding of the response patterns and nonlinear characteristics of vegetation to extreme climate across multi-temporal scales and geomorphic subregions remains lacking. Taking the Yellow River's 'Ji Zi Bend' as the study area, we used 23-year MODIS NDVI data and station-based extreme climate indices (2000–2022), combined with trend analysis, correlation analysis, lag analysis, and the interpretable machine learning model XGBoost-SHAP, to systematically analyze vegetation response characteristics at annual, seasonal, and monthly time scales, including lagged responses across seven geomorphic subregions. The results showed that: (1) NDVI in the study area increased at a rate of 0.004 a⁻¹ from 2000 to 2022, with significant increases in 69.65% of the area and significant decreases in only 7.85%; the regional climate concurrently exhibited warming-drying and intensifying extremes, with both Consecutive dry days (CDD) and Maximum five-day precipitation (RX5day) showing upward trends. (2) Vegetation responses to extreme climate exhibited significant temporal scale dependence: annual responses showed stronger associations with extreme precipitation, monthly responses displayed significant associations with combined water and temperature conditions, together with lagged responses. (3) XGBoost-SHAP indicated that CDD and RX5day had the largest negative and positive contributions, respectively, to model-predicted vegetation variation, with both showing clear predictive nonlinear relationships and model-derived response turning points of approximately 69.91 days and 78.63 mm, respectively. These findings improve our understanding of vegetation sensitivity to climate extremes across multiple temporal scales in dryland ecotones.