The depth distribution of organic carbon in the soils of eastern Australia

Abstract

Abstract Subsurface soil organic carbon (SOC) is a large but still poorly understood component of the global carbon cycle. We investigated the depth distribution of SOC in eastern Australia, testing the hypotheses that SOC content near the surface is linked with water availability, whereas the distribution of SOC with depth is linked with land use, site factors and temperature. To do this, we measured SOC concentration to 1 m at 100 sites across eastern Australia, and fitted three parameter exponential depletion models to the results. Three machine learning algorithms were used to identify predictors important to the model parameters. Multiple regression models were then created based upon the machine learning results using bootstrapped stepwise regressions and the relative importance of the selected variables was assessed using proportional marginal variance decomposition. Surface SOC concentration was influenced predominantly by climate variables, of which seasonal rainfall was by far the most important. At depth, SOC storage was most influenced by site factors (mainly bulk density and soil type), and both land use and climate contributed similar amounts to model explained variance. The depth distribution of SOC was most influenced by land use, which accounted for ~60% of model explained variance, with site and climate factors being approximately equally important. These results support our hypotheses regarding the drivers of SOC depth distribution in eastern Australia and can be used to identify regions with the potential for additional subsurface soil carbon storage.

Keywords

• climate
• conditional inference trees
• cropping
• datamining
• gradient boosting machines
• grazing