Evaluation of the use of soil depth and parent material data when predicting soil organic carbon concentration from LOI values

A shortage of information on the percentage of organic carbon in soil, especially for soils in lower horizons, makes calculation of soil profile carbon storage difficult at regional and national scales. Previous studies paid little attention to how spatial information, such as soil layer position in the soil profile, soil parent material, and type of soil horizon, could best be taken into account when predicting organic C% from loss on ignition (LOI). This issue is addressed in this paper. Statistical tests indicated that spatial variables significantly influence the relationship between organic C%, as measured by the Walkley–Black method, and LOI. Model selection based upon the Bayesian Information Criterion favoured relationships between C% and LOI that took soil depth in the profile into account. However, validation and cross-validation exercises indicated that taking spatial information into account reduces the overall prediction error only slightly (1.8%). Nevertheless, these validation results showed that prediction error in LFH and A horizon soils was reduced by 2.2% and 2.8%, respectively, when information on soil layer positions was used. In addition, prediction error for C horizon soils was reduced by 23% when information on both soil layer positions and parent materials was simultaneously taken into account.