Updating the Australian digital soil texture mapping (Part 2*): spatial modelling of merged field and lab measurements

Malone and Searle (2021) described a new approach to convert field measured soil texture categories into quantitative estimates of the proportion of clay, silt and sand fractions. Converted data can seamlessly integrate with laboratory measured data into digital soil mapping workflow. Here, we describe updating the Australian national coverages of clay, sand and silt content. The approach, based on machine learning, predicts each soil texture fraction at 90 m grid cell resolution, at depths 0-5 cm, 5-15 cm, 15-30 cm, 30-60 cm, 60-100 cm and 100-200 cm. The approach accommodates uncertainty in converting field measurements to quantitative estimates of texture fractions. Existing methods of bootstrap resampling were exploited to predict uncertainties, which are expressed as 90% prediction intervals about the mean prediction at each grid cell. The models and the prediction uncertainties were assessed by an external validation dataset. Results were compared with Version 1 Soil and Landscape Grid of Australia (v1.SLGA) (Viscarra Rossel et al. 2015). All predictive and functional accuracy diagnostics demonstrate improvements compared with v1.SLGA. Improvements were noted for the sand and clay fraction mapping with average improvement of 3% and 2%, respectively, in the RMSE estimates. Marginal improvements were made for the silt fraction mapping, which was relatively difficult to predict. We also made comparisons with recently released World Soil Grid products (v2.WSG) and made similar conclusions. This work demonstrates the need to continually revisit and if necessary, update existing versions of digital soils maps when new methods and efficiencies evolve. This agility is a key feature of digital soil mapping. However, without a companion program of new data acquisition through strategic field campaigns, continued re-modelling of existing data does have its limits and an eventual model skill ceiling will be reached which may not meet expectations for delivery of accurate national scale digital soils information.

Automated summary

The study introduces a new method to convert field-measured soil texture categories into quantitative estimates of clay, silt, and sand fractions, with a focus on updating Australian national soil coverages. Through machine learning, the approach predicts soil texture fractions at different depths and addresses uncertainties in the conversion process. The results show improvements in predictive accuracy compared to existing mapping products, highlighting the importance of continually revisiting and updating digital soil maps with evolving methods and efficiencies.