Abstract. Most models of glacier melt are forced by observed meteorological data in the vicinity of the glacier in question. In the absence of these observations, the forcing is commonly derived from statistical or dynamical downscaling of low resolution reanalysis datasets. Here we perform dynamical downscaling via the Weather Research and Forecasting (WRF) model in order to evaluate its performance against the observations from automatic weather stations (AWSs ) for three mountain glaciers in the interior of British Columbia, Canada over several summer seasons. The WRF model, nested within the ERA-Interim global reanalysis, produced output fields at 7.5 km and 2.5 km spatial resolution for all glaciers, as well as 1 km resolution for one of the glaciers. We find that the surface energy balance (SEB) model, forced by WRF at 2.5 km, adequately simulates the AWS-derived seasonal melt rates despite large biases in the individual SEB components. Overestimation of the number of clear sky days in WRF at 2.5 km explains the positive bias in the seasonal incoming shortwave radiation. This positive bias, however, is compensated by a negative bias in the seasonal incoming longwave radiation, and by an underestimation of sensible and latent heat fluxes. The underestimation of sensible heat fluxes down to −80 % of AWS-derived fluxes, as calculated by the bulk aerodynamic method, is due to the underestimated near-surface wind speeds. An increase of WRF spatial resolution from 7.5 to 1 km does not improve the simulation of downscaled variables other than near-surface air temperature. For relatively small glaciers (< 7 km length along the flowline), the grid spacing of ≥ 1 km is not fine enough to simulate the local cloud convection and topographic wind effects (katabatics). Since the incoming radiative fluxes, as dominant drivers of seasonal melting, are relatively well simulated (within 10 % difference from observed fluxes) by ERA-Interim at 80 km spatial resolution, there is no need for further downscaling of these radiative fluxes. Temperature and precipitation downscaling remains an important step for the simulation of turbulent fluxes and surface albedo.