Artificial neural network modelling: a summary of successful applications relative to microbial water quality

Abstract

Artificial neural networks (ANN) are modelling tools that can be of great utility in studies of microbial water quality. The ability of ANNs to work with complex, inter-related multiparameter databases and provide superior predictive power in non-linear relationships suits their application to microbial water quality studies. To date ANNs have been successfully applied (a) for the prediction of peak microbial concentrations, (b) to sort land use associated faecal pollution sources and relative ages of runoff and (c) towards the selection and study of surrogate parameters. Predictions of peak microbial contamination or faecal pollution sources have been greater than 90% accurate. The importance of a subgroup of organisms that are isolated by the total coliform membrane filter test on m-Endo media in defining faecal sources was revealed through parameter selection exercises. The result is the definition of a new bacterial ratio that can be directly related to the age of faecal contamination in animal impacted runoff.