Soil thickness, as defined in Australia (National Committee on Soil and Terrain, 2009), is the length of distance from the soil surface to para-lithic or lithic contact (i.e. the A and B soil horizons). The term is often used synonymously with soil depth. Maps of soil thickness have a wide range of uses as inputs to land capability and crop/species suitability assessments, in models of biophysical dynamics such as carbon and water storage and balance, or infrastructure planning, to name just a few. However, accurate mapping of soil thickness is fraught with numerous technical difficulties
The current digital soil mapping of thickness (Viscarra Rossel et al. 2014) predicts a maximum thickness across Australia as 1.84m. The same model predicts a minimum soil thickness of 0.1m. The mean soil thickness is 0.82m. Compared to other parts of the world, Australian soils are relatively deep because of the age and weathering processes these landscapes have undergone (Young and Young 2002). For example, some alluvial floodplain soils west of the Australian Great Dividing Range in the Riverina region (eastern Australia) have been measured at more than 20m thick (Chen 1997). While similar observations of soil thickness are known anecdotally in numerous other places, in different settings and contexts, this knowledge is not well represented in this current version soil thickness mapping across Australia.
Biological systems interact predominantly within the solum, which provides a foundation for agricultural productivity and ecosystem diversity. Given this importance of the solum, there is a clear need to firm up our understanding of its variation thickness across the Australian landscape, using the best available data, knowledge and tools.
How does Version 2 differ to what was done in Version?
Leveraging three large, in situ observation datasets and a wide range of spatial environmental variables, we developed three models depicting rock outcrops, intermediate and deep soils respectively. Our modelling approach addressed right-censored data, which is a common attribute of soil thickness data, and we applied an iterative, data re-sampling framework to quantify prediction uncertainties. We integrated the three models to create soil thickness maps and associated products of soil thickness exceedance probabilities.
Read on for a more detailed description of methods and summary of results.
Brief Methodological Description
We harmonised three point observation datasets in this study.
1. The Australian National Soil Site Collation (NSSC, Searle 2015). The database has information for 277,943 soil profile observations, describing 1,019,823 horizons. These data are distributed across Australia, with the compilation being resulting from collaboration between State and National agencies, and universities.
2. National Groundwater Information System (NGIS) database of bore hole data. This spatial database holds nationally consistent information about bores that were drilled as part of the Bore Construction Licensing Framework. The database contains 357,834 drill hole locations with associated lithology, bore construction and hydrostratigraphic records.
3. The Rock Properties database provided by Geoscience Australia give the locations of sampled rock outcrops across Australia . Filtering this dataset on the sample types of “outcrop sample” resulted in 14616 rock outcrop locations within areas where relief > 300m .
A significant amount of data processing was needed to harmonise and extract observed soil thickness values from both the NSSC and NGIS databases. The common goal for both datasets was to identify the depth of the bottom boundary of the soil before it transitions into consolidated or semi-consolidated lithic material or rock. For the NSSC database this identification process was done by analysing the horizon codes that were assigned to layers within each recorded soil profile. Processing of the NGIS database followed a similar sequence to that described in Wilford et al. (2016). That is, the boundary between regolith and fresh bedrock was designated using a query routine applied to more-or-less free-form text of bore hole layer descriptions which searched for the boundary between soil material and consolidated or semi-consolidated materials.
Raster data and predictive covariates
Most covariate data are the same as used by Viscarra Rossel (2015) in generating the Soil Landscape of Australia (SLGA) digital soil information infrastructure. We substituted the climatic data used in Viscarra Rossel et al. (2015) with a new suite of surfaces and associated solar energy and topo-climatic information as described in Harwood et al. 2014.
In order to get a more balanced or even contribution of SCORPAN variables in the spatial modelling, while also aiming for a simpler model configuration with minimal redundant covariates, we opted to implement a principal component analysis (PCA) of the available continuous covariate data. In our study all but one of the environmental covariates was continuous which was the geomorphons terrain classification layer which was a categorical variable of landform characterisations. The PCA was performed for each grouping of covariates for each factor of the SCORPAN function. In our case this meant performing PCA for the 47 climatic variables in one workflow, 21 organism variables in another, and 16 relief variables in another. We combined the 14 soil and parent material layers for the last PCA workflow. For each PCA, 500,000 randomly allocated point locations were distributed within the spatial data extent of the rasters. We performed a raster data extraction at each point location which resulted in a 500,000 x N (number of variables) matrix, which were subjected to PCA after the data were centred and scaled. We selected the number of PCs that cumulatively summarised at a minimum 95% of the data variation. Once done we mapped the PCs using the PCA equation and the stacked raster layers. These workflows resulted in 12, 4, 9 and 10 PCA layers for the climate, organism, relief and parent material + soil SCORPAN variables.
Rather than fitting a single model of soil thicknesses we went for a nuanced approach which entailed three separate models for:
Model 1. Predicting the occurrence of rock outcrops.
Model 2. Predicting the thickness of soils within the 0-2m range
Model 3. Predicting the occurrence of deep soils (soils greater than 2m thick)
Models 1 and 3 used the categorical model variant of the Ranger RF which was preceded by distinguishing; for Model 1, the observations that were deemed as rock outcrops from soils. And for Model 3, distinguishing soils that were less than 2m thick (and not rock outcrops) from soils greater than 2m thick. Ultimately both Models 1 and 3 were binary categorical models. 50 repeats of 5-fold CV (cross-validation) iterations of the Ranger RF model were run for each Model variant.
Model 2 used the regression form of the random forest model. After removing from the total data set the observations that were regarded as rock outcrops and soil greater than 2m, there were 111,302 observations available. Of these, 67,698 had explicitly defined soil thickness values. The remaining 43,604 were right-censored data and were treated as follows. For each repeated 5-fold iteration, prior to splitting the data in calibration and validation datasets, values from a beta function were drawn at random of length 43,604. This value (between 0 and 1) was multiplied by the censored value soil thickness and then added to this same value, creating a simulated pseudo-soil thickness. Once the simulated data were combined with actual soil thickness data, the values were square-root transformed to approximate a normal distribution. Ranger RF modelling proceeded after optimising the Hyperparameter settings as described above for the categorical modelling. Like the categorical modelling, 50 repeated 5-fold CV iterations were computed.
All three model approaches were integrated via a simple ‘if-then’ pixel-based procedure. At each pixel, if Model 1 indicated the presence of rock outcrops 45 times or more out of 50 (90% of resampling iterations), the estimated soil thickness was estimated as rock outcrop, or effectively 0cm. Similarly, for Model 3 which was the model based on prediction of deep soils (soils >2m deep). In no situations did we encounter both Models 1 and 3 predict in the positive on 90% or more occasions simultaneously. If Model 1 or 3 did not predict in the positive in 90% of iterations, the prediction outputs of Model 2 were used.
After model integration, we derived a set of soil thickness exceedance probability mapping outputs. These were derived simply by assessing the empirical probabilities (at each pixel) and then tallying the number of occasions the estimated soil depth exceeded given threshold depths of 10cm, 50cm, 100cm, and 150cm. This tallied number was divided by 50 to give an exceedance probability for each threshold depth.
Goodness of fit statistics of the 3 individual models based on data excluded from model fitting (validation data). Values represent the median of 50 model iterations and values in square brackets represent the 5th and 95th percentiles.
Acknowledging recent and historical approaches for mapping soil thickness around the world this study sought to develop an approach customised to Australian conditions and available data sources with which to derive inferences. This was achieved by separate modelling of rock outcrops, intermediate and deep soils that were then integrated into one output with associated quantified uncertainties. Each of the spatial models were calibrated using a suite of environmental covariates for which climatic themed variables consistently came up as the most important predictor variables. We deduce this because such variables have direct and indirect effects via regulating biota and the weathering of parental materials which ultimately drives spatial heterogeneity of soil thickness across Australian landscapes. This updated soil thickness mapping for Australia is an improvement on previous efforts and will provide better information to inform end-users in applications such as estimating soil and carbon stocks and soil water balance modelling and monitoring.
1. Biggs, A.J.W., Philip, S.R., 1995. Soils of Cape York Peninsula. Queensland Department of Primary Industries, Mereeba, Queensland.
2. Breiman, L., 2001. Random Forests. Machine Learning 45(1), 5-32.
3. Chen, S., Mulder, V.L., Martin, M.P., Walter, C., Lacoste, M., Richer-de-Forges, A.C., Saby, N.P.A., Loiseau, T., Hu, B., Arrouays, D., 2019. Probability mapping of soil thickness by random survival forest at a national scale. Geoderma 344, 184-194.
4. Chen, X.Y., 1997. Quaternary sedimentation, parna, landforms, and soil landscapes of the Wagga Wagga 1 : 100 000 map sheet, south-eastern Australia. Soil Research 35(3), 643-668.
5. Costin, A.B., 1954. A Study of the Ecosystems of the Monaro Region of New South Wales: With Special Reference to Soil Conservation. Government Printer, Sydney, Australia.
6. Harwood, T., Ferrier, S., Harman, I., Ota, N., Perry, J., Williams, K., 2014. Gridded continental climate variables for Australia. CSIRO Land and Water, Canberra.
7. Kempen, B., Brus, D.J., de Vries, F., 2015. Operationalizing digital soil mapping for nationwide updating of the 1:50,000 soil map of the Netherlands. Geoderma 241-242, 313-329.
8. Lacoste, M., Mulder, V.L., Richer-de-Forges, A.C., Martin, M.P., Arrouays, D., 2016. Evaluating large-extent spatial modeling approaches: A case study for soil depth for France. Geoderma Regional 7(2), 137-152.
9. Ma, Y., Minasny, B., Welivitiya, W.D.D.P., Malone, B.P., Willgoose, G.R., McBratney, A.B., 2019. The feasibility of predicting the spatial pattern of soil particle-size distribution using a pedogenesis model. Geoderma 341, 195-205.
10. McBratney, A.B., Mendonça Santos, M.L., Minasny, B., 2003. On digital soil mapping. Geoderma 117(1), 3-52.
11. McKenzie, N., Gallant, J., Gregory, L., 2003. Estimating Water Storage Capacities in Soil at Catchment Scales. Cooperative Research Centre for Catchment Hydrology.
12. Minasny, B., McBratney, A.B., 1999. A rudimentary mechanistic model for soil production and landscape development. Geoderma 90(1), 3-21.
13. Moore, I.D., Gessler, P.E., Nielsen, G.A., Peterson, G.A., 1993. Soil Attribute Prediction Using Terrain Analysis. Soil Science Society of America Journal 57(2), 443-452.
14. National Committee on Soil and Terrain, 2009. Australian Soil and Land Survey Field Handbook (3rd Edition). CSIRO Publishing, Melbourne, Australia.
15. Odeh, I.O.A., Chittleborough, D.J., McBratney, A.B., 1991. Elucidation of soil-landform interrelationships by canonical ordination analysis. Geoderma 49(1), 1-32.
16. Patton, N.R., Lohse, K.A., Godsey, S.E., Crosby, B.T., Seyfried, M.S., 2018. Predicting soil thickness on soil mantled hillslopes. Nature Communications 9(1), 3329.
17. Pelletier, J.D., Rasmussen, C., 2009. Geomorphically based predictive mapping of soil thickness in upland watersheds. Water Resources Research 45(9).
18. Pemberton, M., 1989. Land Systems of Tasmania Region 7: South west. Department of Agriculture, Tasmania, Hobart, Tasmania.
19. Perrr, J.A., Sleeman, J.R., Twidale, C.R., Prichard, C.E., Slatyer, R.O., Lazarides, M., Collins, F.H., 1964. General Report on lands of the Leichhardt-Gilbert Area, Queensland, 1953-54. Commonwealth Scientific and industrial Research Organisation, Melbourne. Australia.
20. Searle, R., 2015. The Australian site data collation to support the GlobalSoilMap. In: D. Arrouays, A.B. McBratney, J. Hempel, A.C. Richer-de-Forges, A.B. McBratney (Eds.), GlobalSoilMap: Basis of the global spatial soil information system. CRC Press, London,UK, pp. 127-133.
21. Searle, R., Grundy, M.J., McBratney, A.B., Gregory, L., Wilson, P., Malone, B.P., Stenson, M., 2019. Phased development of digital soil infrastructure for Australia, Joint Workshop for Digital Soil Mapping and GlobalSoilMap, University of Chile, Santiago.
22. Shangguan, W., Hengl, T., Mendes de Jesus, J., Yuan, H., Dai, Y., 2017. Mapping the global depth to bedrock for land surface modeling. Journal of Advances in Modeling Earth Systems 9(1), 65-88.
23. Tille, P.J., 2006. Soil-landscapes of Western Australia's rangelands and arid interior. West Australian Department of Agriculture and Food., Perth, WA.
24. Viscarra Rossel, R., Chen, C., Grundy, M., Searle, R., Clifford, D., Odgers, N., Holmes, K., Griffin, T., Liddicoat, C., Kidd, D., 2014. Soil and Landscape Grid National Soil Attribute Maps - Soil Depth (3" resolution) - Release 1. v3. CSIRO. Data Collection. https://doi.org/10.4225/08/546F540FE10AA.
25. Viscarra Rossel, R.A., 2011. Fine-resolution multiscale mapping of clay minerals in Australian soils measured with near infrared spectra. Journal of Geophysical Research: Earth Surface 116(F4).
26. Viscarra Rossel, R.A., Chen, C., Grundy, M.J., Searle, R., Clifford, D., Campbell, P.H., 2015. The Australian three-dimensional soil grid: Australia’s contribution to the GlobalSoilMap project. Soil Research 53(8), 845-864.
27. Wilford, J.R., Searle, R., Thomas, M., Pagendam, D., Grundy, M.J., 2016. A regolith depth map of the Australian continent. Geoderma 266, 1-13.
28. Wilson, P., Philip, S.R., 1999. An Assessment of Agricultural Potential of Soils in the Gulf Region, North Queensland. Queensland Department of Natural Resources, Merreba, Queensland.
29. Young, A., Young, R., 2001. Soils in the Australian Landscape. Oxford University Press, South Melbourne, Victoria.