Climate models use input parameters from early aggregated measurement data to simulate arctic scenarios. Using coarsly resolved input causes a statisticial uncertainty which gets passed through the rest of the computation process by error propagation. To identify this extent of this error with regard of thermal properties, soil samples were gathered from three locations on Kurungnakh island in the Lena River Delta, Siberia/Russia. Comparable soil layers were investigated concerning the variability of the share of their soil constituents (air, ice, organic matter, solid matter, water) as well as their thermal properties (volumetric heat capacity, volumetric thermal conductivity, thermal diffusivity). As measure of variability, the coefficient of variation CV was used. The sites vary more in the composition of their constituents than in their thermal properties. 56.25% of soil content CVs but only 16.6% of thermal property CV’s lie above 25 %. Furthermore, it was investigated if input parameters like thermal diffusivity differ depending on if the moment of data aggregation is hastened. A t-test was conducted to examine the likeliness of a match of the mean of thermal diffusivity calculated for individual samples and a value that is calculated from means as intermediate result. When values vary strongly at the moment of aggregation, it was more likely that the hypothesis of that the resulting thermal diffusivity value equals the mean value of thermal diffusivity of individual samples could be rejected on the 5-% significance level. This is the case when averaging soil content shares. These findings indicate that resolving soil constituents on a fine scale is important to accurately model thermal properties. Errors made by those models which assume homogeneous soils over large grid-cells might be significant.