To analyze the spatiotemporal changes of watershed characteristic parameter in the Budyko Framework in the middle reaches of the Yellow River and capture the impact of different factors on the watershed characteristic parameter, multiple linear regression (MLR), gradient boosting (GB), and random forest (RF) models were constructed based on runoff, meteorological, and human activity data from eight subbasins in the middle reaches of the Yellow River. The watershed characteristic parameter ω in the Fu Baopu equation was simulated. By cross validation, select the model with the best performance, identify the main control factors that significantly affect ω, and further incorporate the optimal model into the water heat coupling equilibrium equation to construct a timevarying Budyko framework, quantifying the contribution of climate change and underlying surface changes to runoff. The results showed that among the three models, the RF model outperformed the MLR and GB models in simulating ω. From 1980 to 2019, the ω values of each subbasin showed an increasing trend, mainly controlled by human activities such as impermeable area, population, and regional GDP. Potential evapotranspiration is an important controlling factor in climate factors. The change of underlying surface is the main driving factor for runoff changes in most subbasins of the middle reaches of the Yellow River, but the impact of climate change on the Qinhe River subbasin is slightly stronger than the change of underlying surface.