Automated measurements of snow physical properties in the remote Arctic are scarce, which poses a challenge not only for the investigation of the snow cover evolution throughout the season and years, but also for climate and permafrost modelling, as snow is a crucial parameter for the ground thermal regime. Here, I present the first-time analysis of a sophisticated automated snow measurement data set which was obtained in a Low Centre Polygon (LCP) complex on Samoylov Island, a permafrost site in the Lena River Delta, Siberia. My work focused on analysing the inter-annual variability and seasonal evolution of the snow physical properties depth, density and temperature. I investigated the influence of on the ground thermal regime as well as snow-soil interactions within a time period of four years from 2014 to 2018.Furthermore, I used the data to validate the new ‹CryoGrid (CG) Community› version of the CG permafrost model for Samoylov Island with regard to the snow physical properties. My processing routine included the quality check of the snow depth, snow temperatures as well as density time series and I compared the automated measurements with field observations to evaluate the sensor performances. Furthermore, I generated ‹CG Community› model runs with two versions of the Crocus snow scheme (Standard and Arctic) to validate the models’ ability to reproduce the snow and ground thermal regime for the time span of the observations (2014 to 2018). My results revealed great inter-annual variations in the snow cover extent and internal layering of the snowpack on Samoylov Island within the analysed time period. I found a great spatial variability of the snow depth, which was depended on the micro-topography of the LCP complex. The mean end-of-season (EOS) snow depth was the highest in the polygon centre (0.46 to 0.74 m) and the lowest on the polygon rims (0.32 and 0.53 m). The EOS density for the snowpack up to 0.3 m in the polygon centre was between 204 and 236 kg m−3. The snowpack was characterized by the typical Arctic stratigraphy, consisting out of a basal low density layer (depth hoar) overlain by layers with a higher density (wind slab). The ratio of these layers was connected to the snow cover build-up from October to November, especially the timing and intensity of snowfall. The depth hoar layer evolution started in the early snow season in December and took place until January. It showed a connection to the soil freeze-back, as it was a concomitant of the decrease in soil volumetric water content. The density profile, especially the depth hoar fraction, showed a yearly variability which impacted the ground thermal regime. A higher depth hoar layer thickness (about 0.15 m) was accompanied by stronger ground cooling compared to a year with lower depth hoar layer thickness (about 0.05 m). The ‹CG Community› simulations were able to generate the snow depth variability and evolution. The simulations could not reproduce the measured density profile and had difficulties to generate its high inter-annual and seasonal variability. I found the results of the ‹CG Community› simulations using the Standard Crocus snow scheme to be in better agreement with the automated measurements on Samoylov Island. In combination with the Arctic Crocus snow scheme, the ‹CG Community› model run showed an underestimation of snow depth due to wind effects, which led to an overcompaction of the snowpack. Furthermore, the model runs revealed a high sensitivity to the snowfall rate in the forcing data. My work highlights the complexity of the snowpack and how its inter-annual variability influences the ground thermal regime. I provide a validation of the ‹CG Community› model and outline the need to improve the implementation of snow physical processes in snow schemes and snow-soil models, respectively.