Application of self-organizing maps for assessing soil biological quality

Agricultural ecosystems can be described by many different variables that include soil chemical, physical, and biological data. Whilst most of the chemical and physical properties variables that are relevant to soil quality are well understood, measures of soil biological properties so far have been much more difficult to use as decision support tools for monitoring soil quality. Descriptions of soil biological properties can range from single parameter variables such as microbial biomass or respiration to multiparametric data that describe biochemical profiles, measurements of enzyme activities, and molecular analyses of microbial communities. With the aim of developing practical measures of soil quality, integrative approaches are now being explored to sift out interrelationships between various types of variables. Among the different statistical tools applied, an increasing number of studies have used artificial neural networks (ANNs) to probe complex data sets. As an example of how ANN can be used, we provide an example analysis of soils from two different regions of Southeast Australia using Kohonen self-organizing maps (SOM) using data sets containing biochemical signatures of microbial communities determined by phospholipid fatty acid analysis (PLFA), genetic signatures obtained by terminal restriction fragment length polymorphisms (TRFLP), and a range of single parameter soil chemical, physical, and biological variables. The visual output of the SOM analysis provides a rapid and intuitive means to examine covariance between variables and with minimal training could be useful for assisting land managers with interpretation of multiparametric soil analyses. Further development of these tools should also help soil scientists to identify novel relationships and devise research to explore linkages between the biological, chemical, and physical properties of soils.