We apply a physically based coupled surface energy balance and snowpack model to a site close to the equilibrium line on Austfonna ice cap, Svalbard, over the 2004-08 melt seasons, to explain contributions to the energy available for melting and to quantify the significance of refreezing. The model is forced using in situ meteorological measurements and precipitation downscaled from ERA-Interim reanalysis. Applying a Monte Carlo approach to determine the tunable parameters of the model, we estimate the uncertainty related to the choice of parameter values. Multiple criteria are evaluated to identify well-performing parameter combinations, evaluating the model performance with respect to longwave outgoing radiation, snow and ice temperatures and surface displacement. On average, over the investigated melt seasons (1 June to 15 September) net radiation and sensible heat contributed 90 ± 2% and 10 ± 2%, respectively, to the mean energy available for melting snow and ice. The energy consumed by subsurface heat exchange reduced runoff by 15±2% in 2004 and 49±3% in 2008. Refreezing of meltwater and rain was estimated to be 0.37 ± 0.04 m w.e. a-1 on average over the five seasons, which represents a considerable reduction of mass loss during summer. Our findings suggest that refreezing potentially exerts a decisive control on glacier mass balance in persistently snow- or firn-covered areas.