ANN Classification of Biological Measures of Soil Quality in S.E. Australia
David E Crowley, Pauline M Mele, Helen Hayden, Darryl R. Nelson
Last modified: 2008-09-13
Soil quality assessment for field crop production is typically determined using chemical and physical variables that are known to influence plant growth, but that are relatively static indicators of soil quality as compared to biological variables that are much more dynamic in response to changes in soil management practices. Currently, databases are being established for on-line monitoring of soil quality and yield prediction using statistical models that include mostly chemical and physical variables along with rudimentary measures of soil biomass and respiration as biological variables. Given the availability of new high resolution methods to characterize microbial communities, there is interest in determining how profiles of microbial community structure might be used to assess soil quality. Here we examine the use of ANN models for clustering of multiparametric variables that use molecular and biochemical data to characterize soil microbial community structures. Data were collected from a wide range of sites representing a variety of soil types that are under wheat production in S.E. Australia. Microbial community structures were determined using terminal restriction fragment linked polymorphisms (TRFLP) of 16S rRNA genes in DNA extracts and by biochemical profiles of phospholipid fatty acids (PLFA) that serve as quantitative markers for different groups of microorganisms. For inclusion in the ANN analysis, both TRLFP and PLFA data were first reduced in dimensionality using principle components analysis. Application of Kohonen SOM to the resulting data sets revealed significant linkages between many disparate single parameter chemical and physical variables that appear to drive changes in soil microbial community structure. The visual output of the SOM analysis provides a rapid and intuitive means to examine covariance between these variables and with minimal training could be useful for assisting land managers with interpretation of multiparametric soil analyses. Further development of these tools should help soil scientists identify key biological variables that can be used in soil quality monitoring across different landscapes and management systems.