The control basins of 8 major hydrologic stations (Zhimenda, Shigu, Pingshan, Gaochang, Beibei, Cuntan, Wulong, Yichang) in the upper reaches of the Yangtze River are selected as sub basins. Using their runoff and meteorological data from 1979 to 2015, the machine learning method was used to calculate the sensitivity of runoff to changes in precipitation and temperature, compared with the multiple linear regression method and the Budyko framework method, and further analyzed the contribution of precipitation and temperature changes to runoff changes. The results showed that the change trends of runoff and precipitation were not significant (p=0.01), and the temperature showed a significant upward trend (p<0.01) from 1979 to 2015. The average range of sensitivity coefficients of runoff to precipitation and temperature was 0.37~0.76mm/mm and -4.77~-33.53mm/℃, respectively. The sensitivity coefficient of runoff to precipitation and temperature decreases with the increase of drought index. The impact of precipitation change on runoff change is greater in the sub basin with large drought index, and the impact of temperature change on runoff change is greater in the sub basin with small drought index.