The influences of vegetation cover and climate change on runoff may have scale dependence. However, few studies have been carried out in the southwest karst region with vulnerably ecological environment and complex geological structure. Therefore, we selected the study area in the Xijiang watershed with four sub-watersheds (Yujiang, Hongshuihe, Xunjiang, and Wuzhou), which is characterized by typical karst landforms in southwest China. The objective of this study is to investigate the multi-scale responses of runoff to vegetation and climate change in karst watersheds. The Mann-Kendall trend test is used to analyze the trends in runoff, Normalized Difference Vegetation Index, precipitation, potential evapotranspiration, and temperature in these watersheds from 1982 to 2015. Furthermore, the correlation of runoff with vegetation and climate factors at different scales was quantified. Additionally, runoff was predicted by using the multivariate empirical mode decomposition method. The results indicated that the runoff of the Hongshuihe watershed showed a significant decreasing trend (P<0.05). The main representative scales for runoff were 3 years and 5 years in the Yujiang watershed, 10 years and 22 years in the Hongshuihe watershed, and 3 and 22 years in Xunjiang and Wuzhou watersheds. It was found that scale-dependent relationships between runoff and its influencing factors. At different scales, precipitation and potential evapotranspiration were always significantly correlated with runoff (P<0.05), while the temperature and NDVI were no significantly correlated with runoff at some scales in four watersheds. The multivariate empirical mode decomposition method (R²: 0.81-0.86) was more accurate at predicting runoff than the stepwise multiple linear regression (R²: 0.69-0.78). The study can provide the scientific basis for the rational utilization of water resources, the sustainable development of social economy, and ecological environment protection in karst watersheds.