Developing statistical models to estimate the carbon density of organic soils

Carbon density is a key variable in assessments of local or regional soil carbon (C) stocks, but its direct measurement on large numbers of samples is both time-consuming and expensive. To assess whether the C density of organic soils can be inferred from other parameters, we examined the ability of field- (stratigraphic depth and material type) and lab- (bulk density and ash content) based variables to predict the C density of organic soil samples. Candidate models given three different levels of a priori information about samples were developed from data for continental western Canada and examined using Akaike's information criterion (AIC). Models at each level were then used to predict profile-level C storage in cores from three different regions (continental western Canada, Ontario, and the Northwest Territories). In profiles from western Canada, predictions were unbiased, with mean prediction errors of 0-7% and local precision depending on the amount of a priori information available. Application of models to other regions yielded mixed results, probably reflecting both differences in site characteristics and classification/analytical methods used. Since these error sources are impossible to separate given available data, we recommend that models for C density prediction should be tailored to a given research question and region. The results suggest that simple, field-based variables are sufficient to predict C density for the purpose of regional surveys. To obtain accurate estimates at the profile level, bulk density (and ash or C content) have to be measured in the lab.