This dataset shows predicted inherent plant available soil P across Australia. By inherent we mean natural soil P without the influence of fertilizers or cropping. The dataset was made using existing government soil P data. Only P data from sites that were not cultivated/cropped were used to make the dataset. “Disturbance of site” (NCST, 2009) was used where recorded to determine if the P result was from an uncultivated field / site. Otherwise, current land use was used to filter out results from cultivated or disturbed sites.
P measured using the Colwell P method (9B1 & 9B2, Rayment & Lyons, 2011) has been used as the basis for this dataset. The units are mg / kg of P and there are no significant figures (P predictions are integers from 1).
Demand for this dataset has originally come from the grazing industry in northern Australia. P is seen as the most limiting nutrient for livestock and affects animal growth, breeding and lactation. The Australian continent is also considered to have the lowest P concentrations when compared to the rest of the world. P supplementation in northern Australia is not as widespread as it is in the southern livestock industries.
Maps of total P have been available for a while now and are used for environmental purposes, however total P does not correlate well with plant and animal nutrition due in part to soil P sorption capacity.
Data was sourced from government soil databases using the Soil Data Federator. Extra historical data has been supplied by WA, NSW & Qld Governments. Less than 600 new sites in remote areas have been analysed using samples from the TERN Surveillance Archive and from current projects in the Roper catchment and with the Meat & Livestock Australia. Their support is greatly appreciated. In total there are 13948 sites with Colwell P used to fit the model that produces this dataset (Figure 1.)
Figure 1 - Sites used in AVP model
Most historical P data is for the surface layer (0 – 0.1 m) only. Less than 1000 sites had more than two samples taken down the soil profile. The specification for this product was to map all global soil depths (0 – 0.05, 0.05 – 0.15, 0.15 – 0.3, 0.3 – 0.6, 0.6 – 1, 1 – 2 m). Before model fitting, sites with only a surface P result, were augmented with interpolated P. Analysis of 997 sites with P data down the profile showed that P decreases toward zero rapidly with depth. Also no significant relationships were found when sites were stratified with geology or soil parent material. A power decay function was fitted to the 997 sites and this relationship used to interpolate P down the profile given a surface P concentration in the remaining sites with only surface P concentrations.
68 environmental covariates from the SLGA were used as predictors of P. The top ten predictors were Maximum difference between monthly precipitation (PTRX) (Harwood, 2019), Lithology, Slope (Gallant & Austin, 2012b), Radioactive Thorium (Nakamura & Milligan, 2015), Topographic Wetness index (Gallant & Austin, 2012c), Mean elevation with 1000m (Gallant & Austin, 2012a), Plan curvature (Gallant J.P., 2000), Landscape roughness (Gallant J.P., 2000) , Landscape weathering index (Wilford, 2012) and Minimum difference between monthly temperature (TNRI) (Harwood, 2019).
The site and covariate data were used with a random forest machine learning algorithm (Liaw & Wiener, 2018) using the caret (Kuhn, 2008) implementation of ranger (Wright et al., 2020). 10 K – folds with cross validation were used to fit the model. The final model was fitted to 70% of the site data. 30% of the remaining data was used for internal validation of the model. Model performance is shown in Table 1.
Table 1. Model validation statistics
The dataset was created from the fitted model using the predict function in the raster package (Hijmans et al., 020). 50 bootstraps were run to determine the 90% confidence interval for each prediction. Figure 2. shows surface (0 – 0.05 m) available P prediction across Australia.
Figure 2 - Surface (0 – 0.05 m) available P prediction across Australia
Figure 3. shows uncertainty calculated as,
Figure 3- Uncertainty of the surface (0 - 0.05 m) layer
Gallant J.P., J. C. and W. (2000). Chapter 3. Primary topographic attributes. In J. P. and G. Wilson J.C. (Ed.), Terrain Analysis: Principles and Applications. John Wiley and Sons.
Gallant, J. C., & Austin, J. M. (2012a). Relief - elevation range over 1000m derived from 1" SRTM DEM-S. v2. In CSIRO Data Collection. https://doi.org/http://doi.org/10.4225/08/56E9C8DDD664D
Gallant, J. C., & Austin, J. M. (2012b). Slope derived from 1" SRTM DEM-S. v4. In CSIRO Data Collection. https://doi.org/http://doi.org/10.4225/08/5689DA774564A
Gallant, J. C., & Austin, J. M. (2012c). Topographic Wetness Index derived from 1" SRTM DEM-H. v2. In CSIRO Data Collection. https://doi.org/http://doi.org/10.4225/08/57590B59A4A08
Harwood, T. (2019). 9s climatology for continental Australia 1976-2005: BIOCLIM variable suite. In CSIRO Data Collection (No. v1). CSIRO. https://doi.org/https://data.csiro.au/collection/csiro:42020v1
Hijmans, R. J., Etten, J. Van, Sumner, M., Cheng, J., Baston, D., Bevan, A., Bivand, R., Busetto, L., Canty, M., Forrest, D., Golicher, D., Gray, J., Greenberg, J. A., Karney, C., Mattiuzzi, M., Mosher, S., Shortridge, A., & Wueest, R. (2020). Package ‘ raster .’ https://cran.r-project.org/web/packages/raster/raster.pdf
Kuhn, M. (2008). Building predictive models in R using the caret package. Journal of Statistical Software. https://doi.org/10.18637/jss.v028.i05
Liaw, A., & Wiener, M. (2018). Breiman and Cutler’s Random Forests for Classification and Regression, Version 4.6-14. CRAN. https://cran.r-project.org/web/packages/randomForest/randomForest.pdf
Nakamura, A., & Milligan, P. R. (2015). Ternary Radiometric Image of Australia 2015 (Vol. 3rd, Issue Australia). Geoscience Australia.
NCST. (2009). Australian soil and land survey field handbook (3rd ed.). CSIRO Publishing.
Rayment, G. E., & Lyons, D. J. (2011). Soil Chemical Methods - Australasia. Australian soil and land survey handbooks. CSIRO Publishing.
Wilford, J. (2012). A weathering intensity index for the Australian continent using airborne gamma-ray spectrometry and digital terrain analysis. Geoderma, 183�184(0), 124–142. https://doi.org/http://dx.doi.org/10.1016/j.geoderma.2010.12.022
Wright, M. N., Wager, S., & Probst, P. (2020). Package “ranger.” https://cran.r-project.org/web/packages/ranger/ranger.pdf