Whole Soil Bulk Density
Figure 1. Distribution of site data with some measure of whole soil bulk density. Measures were wither performed in field or lab, or inferred from pedotransfer function.
Deriving pedotransfer function of whole soil bulk density
Figure 2. Goodness of fit model evaluations and modelled vs. observed plots for both pedotransfer function types to predict whole soil bulk density using soil carbon and clay content and with or without environmental covariates.
Predictive spatial soil modelling and uncertainty quantification
Figure 3. Table at top of image collection shows the model evaluations from external data for both the updated (SLGA Version 2) and original (SLGA Version 1) whole soil bulk density models for each of the specified depths. The middle xy-plots of predicted vs. observed is shown for the 0-5cm depth interval. A similar visual pattern is observed for the other depths and thus not shown in this document. Similarly, the prediction interval coverage probability plot is for the 0-5cm depth interval showing a well constructed quantification of model uncertainties and also similar for the other depth intervals.
Figure 4. National digital soil mapping of whole soil bulk density (0-5cm) show both the mean prediction and upper and lower prediction limits. The maps on the left column are for SLGA Version 1, while to ones on the right are for SLGA Version 2. These maps show similar general spatial pattern in the mean prediction at the national scale. The range of the uncertainties do appear smaller in general for SLGA Version 2, improving the use and reliability of the mapping.
Quantification of model extension limits
Figure 5. Assessment of model data coverage, and by association, model extrapolation risk for whole soil bulk density for the 0-5cm and 100-200cm depth intervals. With increasing soil depth, the risk of extrapolation errors increases due to an increasing sparsity of observed data.