Soil Carbon Analysis

Extract of methodology supporting: Barson, M.M, Muir, T., Malafant, K.W.J., Skjemstad, J.O. and Janik, L. How Big is Australia’s Soil Carbon Store? Poster presentation at: 5th International Carbon Dioxide Conference, Cairns, Australia, 8-12, September, 1997.


Soil organic matter is the largest global terrestrial carbon pool, and can be a source and a sink of atmospheric CO2. The need to quantify the impact of land clearing and agricultural land management techniques on soil carbon stores for Australia’s National Greenhouse Gas Inventory has focussed attention on the size of this pool, and its spatial distribution across the continent. Preliminary estimates suggest that Australia’s soil carbon sink is of the order of 48 Gt (Gifford et al. 1992). About 75 percent of Australia’s soils have organic carbon contents of less than 1 percent (Spain et al.) The more productive soils have levels greater than 5 percent, but their areal extent is small. The spatial distribution of soil carbon is poorly known.

Precipitation and temperature, soil texture, chemical and vegetation properties and land use history are thought to be the most important factors affecting soil carbon storage. Analysis of the variation in soil carbon along the 1000 km Moomba pipeline (Queensland border to Gunning, New South Wales) demonstrated that soil carbon content was more highly correlated with annual average rainfall and temperature than soil type or texture.

Soil Carbon Analysis

A data base of 30,000 soil profiles is being used to determine whether the relationship between environmental variables, pH, clay content, vegetation type, land use and other variables can be used to predict both the spatial distribution and change with depth of soil carbon on a continental scale.

Generalised additive modelling (GAM) techniques (Hastie and Tibshirani, 1990) have already been used to model and predict the surface soil carbon (0-10cm) for the continent based on a subset of the available data (1059 profiles). However, this modelling did not consider the change in carbon percentage with depth or allow for the estimation of total organic carbon (TOC) within the profile.


The collection of additional profiles (sites) with observations on soil carbon at varying depths has allowed the models/methods to be expanded to account for changes with depth in the profile. An initial investigation of the change in OC with depth for selected profiles indicated that an exponential decay model of the following form might adequately describe this change:

Y = a e-b d

Where Y is the observed OC percentage and d is the depth at which the observation was taken. The value of a can be interpreted as the initial surface value of the OC %, while b is the rate of decay with depth and the asymptotic OC % value. Two other aspects of this model also are of relevance:

    1. The model can be linearised and the parameters estimated via linear regression. The linearised form of the model:
    2. Log(Y) = Log(a ) - b depth

      can be fitted to any site profile with more than 3 measurements and can be automated for all the sites. The behaviour of OC with depth at each site can be summarised by considering the a and b values.

    3. The area under the curve is the total %OC in the profile between any two depths. The model function can be integrated analytically and the TOC calculated. This value combined with bulk density can then be used to estimate the TOC in tonnes for the continent.


The available site data consisting of profiles of %OC and depth were used to fit the linearised version of the model to all data sets with more than 3 observations. Any missing values were eliminated from the analysis. The actual data used is the observed %OC and the average of the upper and lower profile depths between which the %OC measurement was taken. Each site was summarised in terms of the a and b estimates from the fitted regression.

The summarised values for each profile were then combined with the following list of explanatory variables:

    1. Elevation from a DEM (e)
    2. Annual mean temperature (t)
    3. Mean temperature of the coldest quarter (c)
    4. Annual precipitation (p)
    5. Precipitation in the wettest quarter (w)
    6. A factor describing pH (< 6.5, 6.5 - 7.5, > 7.5) (pH)
    7. Clay content (cl)
    8. Vegetation class (v)
    9. Principle soil profile (pp)

A GAM analysis was used to provide independent predictive models for the variation in a and b values across the sites. A stepwise regression approach was adopted to "reduce" the number of explanatory variables using a combination of forward selection and backward deletion. All observed units with any missing values were eliminated and no interaction terms have been included in the current analysis. The GAM and stepwise analysis were done using the Splus system.

The analysis resulted in the following model describing the variation in a :

a = constant + s(e) + s(t) + s(c) + p + v

where s(x) indicates a smooth function of X based on a simple smoothing spline. This model accounted for 53% of the variability in a . The model for the variability in b only accounted for 21% of the variability with a simple model of:

b = constant + s(e) + t + s(c) + p + s(w) + v

Clearly the variability in a is more readily associated with environmental variables (not surprising given its interpretation), than is the variability in the decay parameter b .

This analysis allows us to "estimate" the parameters of the decay model from the more easily measured explanatory variables. From these predictive equations an estimate of the form/shape of the %OC change with depth can be made for any site on the continent. Using these models an indicative analysis for the continent was undertaken by "griding" information layers using a GIS to a 2 km x 2.7 km basis. The continental information, combined with the model, was then used to "predict" the OC on a grided basis for the whole continent.

Further Work

Future work will focus on analysing the responses of a and b more closely, extension of the model to handle duplex soils, an investigation of the model behaviour, simplification of parameter sets for specific soil types/areas and the investigation of the behaviour of the models under fine scale conditions. One limitation of the approach is the development of models for a and b which assume that they behave independently. This is unlikely to be a true reflection of their behaviour and extending the models to consider their interaction may improve the reliability of predicting the decay parameter, b .


Gifford, R. M., Cheney, N. P., Noble, J. C., Russell, J. S., Wellington, A. B. and Zammit, C. 1992. Australian land use, primary production of vegetation and carbon pools in relation to atmospheric carbon dioxide concentrations. pp 151-187 In: (eds) Gifford, R. M. and Barson, M. M. Australia’s renewable resources; sustainability and global change. Bureau of Rural Resources and CSIRO Division of Plant Industry. Bureau of Rural Resources Proceedings No. 14.

Spain, A. V., Isbell, R. F. and Probert, M. E. 1983. Soil organic matter. In: Soils: an Australian viewpoint. CSIRO Academic Press, pp 551-563.

Hastie, T.J. and Tibshirani, R.J. 1990. Generalized Additive Models. Chapman and Hall, London.

|| Home page || Resume || Projects || Publications ||
Web site established: 20 September 1998       Last updated: 20 September 1998
Site designer and maintainer: Kim Malafant (

Copyright © 1998 by Kim Malafant. All rights reserved. This Web page may be freely linked to by other Web pages. Contents may not be republished, altered or plagiarized. compleXia does not control or endorse the content of third party Web Sites.