The peak river flow for the Mackenzie River is modelled using GRACE satellite observations and temperature data, which advances the applications of space-based time-variable gravity measurements in cold region flood forecasting. The model estimates peak river flow by simulating peak surface runoff from snowmelt and the corresponding baseflow. The modelled results compared fairly well with the observed values at a downstream hydrometric station. The results also revealed an average 22-day travel time for the snowmelt water to reach the hydrometric station. The major driver for determining the peak flow was found to be the temperature variations. Compared with the Red River basin, the results showed that the Mackenzie River basin has relatively high water storage and water discharge capability, and low snowmelt efficiency per unit temperature. The study also provides a GRACE-based approach for basin-scale snowfall estimation, which is independent of in situ measurements and largely eliminates the limitations and uncertainties with traditional approaches. The model is relatively simple and only needs GRACE and temperature observations for peak flow or flood forecasting. The model can be readily applied to other cold region basins, and could be particularly useful for regions with minimal data.