Predicting freshwater habitat conditions by the distribution of macroinvertebrates using artificial neural network

Abstract

Artificial Neural Networks were used to model the relationship between habitat conditions and macroinvertebrate assemblages in the Queensland river and stream system. The clean water approach was used to determine the relationship between presence and absence of macroinvertebrate taxa and physical predictor variables, studying data from reference sites in near pristine conditions. Trained models were applied to assess habitat conditions of impacted and test sites. The dirty water approach studied interrelationships between physical, chemical and biological processes. Sensitivity analyses were carried out by manipulating the values of input parameters and assessing the resulting changes in outputs. Artificial Neural Networks proved an effective computational approach to support bioassessment.