The temperature distribution in permafrost soils is affected by a wide variety of parameters, which can vary over small distances and on short timescale. An adequate representation of these small-scale heterogeneities in permafrost models remains a challenging task. Energy balance models calculate the surface temperature based on the partitioning of energy at the surface. The surface temperature is then projected into deeper soil layers. In principle, this class of permafrost models can account for small-scale spatial heterogeneity, if only a sufficiently resolved set of all input parameters is provided. In practice, such data sets rarely exist, so it is necessary to identify the crucial parameters and the spatial and temporal scales, over which they must be accounted for to achieve a satisfactory accuracy of the model. For this purpose, a detailed understanding of the surface energy budget is indispensable.We present continuous measurements of all components of the surface energy budget at a high-arctic permafrost site on Svalbard over the course of one year. An eddy covariance system is used to determine the turbulent land-atmosphere exchange processes. The results not only illustrate the annual transition from long-wave radiation forcing during the polar night to forcing by solar radiation during the summer, but also highlight the importance of sensible and latent heat fluxes for the formation of the surface temperature.During the snow-free period, the surface temperatures of an area of about 100 x 100 m² have been monitored at spatial resolutions below one meter using a thermal camera system. Strong temperature differences between wet and dry areas are found on short timescales of a few hours. Using an energy balance model, this can be explained by different surface resistances to evaporation and hence a different energy partitioning between the sensible and the latent heat flux. However, on timescales of one week to one month, the differences between wet and dry areas widely average out and are hence negligible in the context of subsurface temperature evaluation.During winter, the temperature at the snow-soil interface and the temperature profile to a depth of 1.5 m have been monitored at 14 different locations within an area of half a square kilometre. In contrast to summer, sustained average temperature differences of up to 6 K between different locations are found at the snow-soil interface, although energy balance calculations and direct measurements suggest little spatial variation of the temperature of the snow surface. The temperature differences can be directly related to the thickness of the snow cover and possibly also its history of formation. They result in strong site-to-site variations of the soil temperatures at 1.5 m depth, which range from -6°C to -0.3°C in March. The snow cover is therefore found to be the prime source of spatial variability of the permafrost temperatures at the study site.