Neural networks for predicting seawater bacterial levels

Abstract

In this study artificial neural networks (ANNs) have been applied to predict faecal coliform concentration levels at compliance points along bathing water zones situated in the south west of Scotland, UK. Hydrological parameters, such as river discharges, sunshine, rainfall and tidal conditions, were used as the input data for these networks. Data collected at seven locations during the period 1990–2000 were used to train and verify the neural networks. A novel technique called the gamma test was used for data analysis to aid in the construction of ANN models. In general, the river discharges and tidal range were found to be the most important variables affecting the level of bacteria concentration at the compliance points. For compliance points close to the meteorological station, the amount of rainfall was found to be relatively significant in the model results. Relatively good correlation coefficients were obtained for the learning and verifying process for all of the ANNs and these networks confirmed that the samples failed to comply with the standard values specified in the European Union Bathing Water Directive in 57·4% of the cases.