Predicting near-shore coliform bacteria concentrations using ANNS B. Lin, S.M. Kashefipour* and R.A. Falconer School of Engineering, Cardiff University, PO Box 925, Cardiff CF24 0YF, UK (E-mail: linbl@cardiff.ac.uk; falconerra@cardiff.ac.uk) * Irrigation Department, Shahid Chamran University, Ahwaz, Iran Abstract Details are given of the application of Artificial Neural Networks (ANNs) to predicting the compliance of bathing waters along the coastline of the Firth of Clyde, situated in the south west of Scotland, UK. Water quality data collected at 7 locations during 1990–2000 were used to set up the neural networks. In this study faecal coliforms were used as a water quality indicator, i.e. output, and rainfall, river discharge, sunlight and tidal condition were used as input of these networks. In general, river discharge and tidal ranges were found to be the most important parameters that affect the coliform concentration levels. For compliance points close to the meteorological station, the influence of rainfall was found to be relatively significant to the concentration levels. Keywords ANNs; bathing waters; coastal basin; faecal indicators; prediction Introduction Bathing water quality has increasingly become one of the main concerns of coastal envi- ronmental managers in the U.K. Pathogenic bacteria in bathing waters have been always the most important way of distributing dangerous and epidemic diseases. Individual pathogens are generally difficult and expensive to measure and therefore in water quality studies it is a common practice to measure and/or model the levels of related indicator organisms (Thomann and Muller, 1987). The faecal coliform (FC) bacterial indicator is widely used in assessing bathing water quality. The European Union (EU) has published several standards regarding guideline and mandatory concentration levels of pathogen indicators for bathing water (Council of the European Communities, 1976). Currently numerical models based on solving solute transport and kinetic equations have been the main tool used to predict pathogen bacterial concentrations in coastal waters. The main advantage of using numerical models is that they are able to predict distributions of bacterial concentration at any time during the simulation. However, they need time series of hydrodynamic and water quality variables to drive and calibrate, and also require detailed bathymetric data, which are usually expensive and time consuming to obtain. Applying these models for predicting flow and water quality may also be restricted by the boundary conditions around the model domain (Falconer et al., 2000). In recent years artificial neural networks (ANNs) have been increasingly applied to a wide range of problems in various scientific fields (Widrow et al., 1994). In this paper ANNs are applied to predict FC concentrations at several compliance sites along a coastline located at the south west of Scotland, UK, from Girvan in the south to Ardrossan in the north. The effect of the individual variables on the FC concentrations at those sites is dis- cussed later in this study, and the failure of the bacterial concentrations in complying with the EU standards is also investigated. Artificial neural networks ANNs are a powerful tool in simulating dependent variables for a wide range of scientific Water Science and Technology Vol 48 No 10 pp 225–232 © IWA Publishing 2003 225