Lithology as a powerful covariate in digital soil mapping

Lithology can provide a powerful and easily used covariate to complement other parent material related covariates and improve the statistical performance of Digital Soil Models and Maps (DSMM). However, it appears that the use of lithology is not fully utilised as a covariate in digital soil mapping projects around the globe. It appears only 25% or less of DSMM studies contain parent material covariates in any form. There is a strong reliance on radiometric and other geophysical data, but their relationship with lithology is not always strong. A scheme is presented that encourages the effective incorporation of lithology into DSMM. It involves the classification of parent material into eleven different classes based on broad chemical composition. Eight of these are ordinal classes specifically based on silica and base content, and range from extremely siliceous (>85% silica) to ultra-mafic (<45% silica). Classes are allocated to all geological units described in the attribute table of the geology covariate layer, using a manual or semi-automated method. Lithology then becomes available as an ordinal or nominal categorical covariate for use in spatial models. Lithology was incorporated into multiple linear regression models for predicting soil distribution in eastern Australia. It was shown to have a major influence in the models for all soil properties examined, being the dominant or second dominant factor out of the six key factors considered at the scale of this modelling project. Inclusion of lithology in models increased R-2 from 0.43 to 0.50 for organic carbon, 0.38 to 0.51 for pH, 0.16 to 0.48 for sum of bases and 0.10 to 0.50 for sand. A comparison of the effectiveness of lithology versus radiometric and other geophysical spatial data is proposed.