Spatially-explicit regional-scale prediction of soil organic carbon stocks in cropland using environmental variables and mixed model approaches

The effects of soil redistribution on the carbon (C) cycle and the need for spatially and depth-explicit C estimates at large scales have recently been receiving growing attention. In eroding agricultural landscapes, C gets transported from erosional to depositional landscape elements forming a heterogeneous pattern in quantity and quality of the distributed carbon. At present, methods and research to characterize this horizontal and vertical variability are either limited to local slope scales or, if applied to larger scales, to surface soil horizons with large uncertainties when extrapolated to deeper layers. In this study, we used soil profile data collected in two zones of differing soil texture (loam and clay-rich soils) in Luxembourg, to calibrate a linear mixed-effect model to predict the 3D soil C stock distribution on a regional scale for cropping systems using a set of spatially-explicit hydrologic, climatic, pedologic and geomorphologic variables. We demonstrate that due to a high spatial variability of C stocks it is mandatory to consider various environmental processes to predict C accurately on a regional scale, especially in deeper soil layers, and to avoid simple depth extrapolation of topsoil C data as has been done earlier in flat landscapes. Using estimates of topsoil C contents derived from hyperspectral remote sensing, we predict spatial patterns of C stocks for cropland on a regional scale and provide new insights into the spatial heterogeneity of soil C storage covering a large area. The variability of C stocks in the two texture zones expressed as values larger or smaller than the mean ± standard deviation is hereby lower in the loam zone (26.2%) than in the clay zone (38.7%). We estimate a mean C stock (to 100 cm soil depth) of 9.4 ± 3.1 kg/m2 for the clay-rich soils and 11.3 ± 2.4 kg/m2 for loamy soils. This represents the first regional estimate for C stocks for the research area using continuous spatial explicit datasets.