Premium
Building and testing conceptual and empirical models for predicting soil bulk density
Author(s) -
Tranter G.,
Minasny B.,
Mcbratney A. B.,
Murphy B.,
Mckenzie N. J.,
Grundy M.,
Brough D.
Publication year - 2007
Publication title -
soil use and management
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.709
H-Index - 81
eISSN - 1475-2743
pISSN - 0266-0032
DOI - 10.1111/j.1475-2743.2007.00092.x
Subject(s) - pedotransfer function , bulk density , soil science , soil water , linear regression , empirical modelling , soil carbon , conceptual model , environmental science , mathematics , mineralogy , chemistry , statistics , computer science , hydraulic conductivity , database , programming language
The development of pedotransfer functions offers a potential means of alleviating cost and labour burdens associated with bulk‐density determinations. As a means of incorporating a priori knowledge into the model‐building process, we propose a conceptual model for predicting soil bulk density from other more regularly measured properties. The model considers soil bulk density to be a function of soil mineral packing structures ( ρ m ) and soil structure (Δ ρ ). Bulk‐density maxima were found for soils with approximately 80% sand. Bulk densities were also observed to increase with depth, suggesting the influence of over‐burden pressure. Residuals from the ρ m model, hereby known as Δ ρ , correlated with organic carbon. All models were trained using Australian soil data, with limits set at bulk densities between 0.7 and 1.8 g cm −3 and containing organic carbon levels below 12%. Performance of the conceptual model ( r 2 = 0.49) was found to be comparable with a multiple linear regression model ( r 2 = 0.49) and outperformed models developed using an artificial neural network ( r 2 = 0.47) and a regression tree ( r 2 = 0.43). Further development of the conceptual model should allow the inclusion of soil morphological data to improve bulk‐density predictions.