The snow melt from the High Atlas represents a crucial water resource for crop irrigation in the semiarid regions of Morocco. Recent studies have used assimilation of snow cover area data from high‐resolution optical sensors to compute the snow water equivalent and snow melt in other mountain regions. These techniques however require large model ensembles, and therefore it is a challenge to determine the adequate model resolution that yields accurate results with reasonable computation time. Here we study the sensitivity of an energy balance model to the resolution of the model grid for a pilot catchment in the High Atlas. We used a time series of 8‐m resolution snow cover area maps with an average revisit time of 7.5 days to evaluate the model results. The digital elevation model was generated from Pléiades stereo images and resampled from 8 to 30, 90, 250, 500, and 1,000 m. The results indicate that the model performs well from 8 to 250 m but the agreement with observations drops at 500 m. This is because significant features of the topography were too smoothed out to properly characterize the spatial variability of meteorological forcing, including solar radiation. We conclude that a resolution of 250 m might be sufficient in this area. This result is consistent with the shape of the semivariogram of the topographic slope, suggesting that this semivariogram analysis could be used to transpose our conclusion to other study regions.