Mountain environments represent heterogeneous environments with shallow soils that are sensitive to human impact and climate change. Despite the thin soil cover, high soil organic carbon content of mountain soils may provide a major source of atmospheric CO2, if released. However, the importance of mountain soils remains controversial, largely due to insufficient information on the spatial variability of mountain SOC stocks. Here, we study the spatial variability of soil properties and SOC stocks in a changing mountain environment in the Bernese Alps (Switzerland) and the methodologies to assess them. We use different interpolation techniques (averaging, inverse distance, ordinary-, block- and regression-kriging) and sampling densities and analyze the sources of uncertainty using a nested sampling approach and the Gaussian and Taylor error propagation. We found a low sensitivity of the median SOC stocks of the study area (ranging between 8.1 and 8.6 kg C m(-2) in the upper 30 cm), the general patterns of the predicted stocks and the explanatory power with respect to the utilized interpolation techniques. In contrast the small-scale SOC pattern fluctuates strongly between different interpolation techniques. All interpolation techniques, except regression kriging, show a low variability of the calculated root mean square errors of the predicted SOC stocks in terms of variable sampling densities. To improve spatial prediction using regression kriging, which combines the kriging approach with multiple linear regression based on factors controlling the SOC variability (e.g. soil type, land use and topography), large sampling density (>35 samples per km(2)) is required in alpine environments. This is especially true for the coarse mineral fraction, which introduces the largest source of uncertainty. Nested sampling designs seem to provide an efficient tool to study SOC inventories and their associated sources of uncertainties in mountain environments. (C) 2014 Elsevier B.V. All rights reserved.
