Taking mountainous area of the upper Heihe River as study area, this paper uses Mann-Kendall test and Yamamoto mutation test to identify trend and mutation characteristics of hydrometeorological factors based on hydrometeorological observation data from 1980 to 2022. ABCD-snow model coupled with a snowmelt module is established. Combined with Budyko coupled hydrothermal equilibrium framework at arbitrary time scales and considering coupling effect of multiple factors, an improved sensitivity attribution method is adopted to quantify impacts of climatic and underlying surface factors on multi-time-scale (annual, seasonal and monthly) evolution of runoff in the upper Heihe River before and after the mutation. The results show that runoff in the upper Heihe River experienced an abrupt change in 1997 over past 40 years. Before the mutation, annual runoff presented an insignificant decreasing trend with the largest decline in summer and autumn. After the mutation, annual runoff increased significantly; all seasons showed a remarkable growth trend except summer with an insignificant increase. ABCD snow model effectively improves simulation performance of key hydrological processes in cold regions. Its efficiency coefficient during calibration period rose from 0.75 to 0.90, which well reflects snowmelt recharge characteristics of spring runoff. From 1980 to 1997, decrease of rainfall and rise of reference evapotranspiration in summer and autumn led to runoff reduction, especially in June, August and September. From 1998 to 2022, increased rainfall and snowmelt dominated runoff growth, while slowdown of reference evapotranspiration and improvement of normalized difference vegetation index (NDVI) further promoted watershed runoff generation. Runoff became more sensitive to climate change, with relative contribution rate rising from 56.9% to 70.7% before and after mutation, and synergistic effect of multiple factors was generally lower than 1.1%.