Development and calibration of a soil carbon inventory model for New Zealand

A soil organic carbon (SOC) and SOC change model for New Zealand is developed for use in national SOC inventory reporting. The foundation for the model is a generalised least-squares regression, based on explanatory variables of land use, soil–climate class, and erosivity. The SOC change model is based on the assumption that changes in SOC over a decadal timescale are usually restricted to transitions in land use. Improvements to the model are then considered that are intended to reduce the uncertainty of SOC changes through reduction of the standard error of the land-use effects. Stochastic gradient boosting is used to find data layers most strongly associated with SOC. The most influential of these were then used in a general least-squares model after stepwise refinement. The stepwise-refined model significantly reduced the standard error for SOC, but did not result in a consistent reduction in the standard error for land-use classes, nor did it result in an improvement in the SOC change model. The method of calculating SOC change from a transition between two land-use classes is described, along with the significance of the transition, by use of a multi-comparison significance procedure.