The mesoscale atmospheric model WRF is used over three Svalbard glaciers. The simulations are done with a setup of the model corresponding to the state-of-the-art model for polar conditions, Polar WRF, and it was validated using surface observations. The ERA-Interim reanalysis was used for boundary forcing and the model was used with three nested smaller domains, 24 and 8 km, and 2.7 km resolution. The model was used for a two-year period as well as for a more detailed study using 3 summer and winter months. In addition sensitivity tests using finer horizontal and vertical resolution in the boundary layer and using different physics schemes were performed. Temperature and incoming short- and long-wave radiation were skillfully simulated, with lower agreement between measured and modelled wind speed. Increased vertical resolution improved the frequency distributions of the wind speed and the temperature. The choice of different physics schemes only slightly changed the model results. The polar-optimized microphysics scheme outperformed a slightly simpler microphysics scheme, but the two alternative and more sophisticated PBL schemes improved the model score. A PBL scheme developed for very stable stratifications (QNSE) proved to be better in the winter.