Predicting Bacterial Levels in Recreational Beach Waters along U.S. Gulf Coast

Abstract

A Bayesian model was proposed for daily predictions of the probability of water quality standard violation of enterococci (ENT) levels in recreational beach waters. The Bayesian model consisted of a prior distribution and a likelihood function, which were constructed using seven years of environmental and bacteriological data collected at six recreational beach sites along the U.S. Gulf coast. The likelihood function followed best a normal distribution while the prior distribution was found to be best fitted with a Nakagami distribution. Modelling results showed that the Bayesian model was capable of explaining 86.13% of recreational water quality advisories issued by the U.S. Louisiana Beach Monitoring Program with a false positive rate of 7.24% and a false negative rate of 6.23%, indicating an excellent performance of the Bayesian model. The Bayesian model, presented in this paper, is unique and novel in terms of (1) the integration of a deterministic model and a probabilistic model for the prediction of recreation beach water quality; (2) the identification of important hydrodynamic processes controlling the source and transport of bacteria in coastal beach waters; and (3) the identification of key sinks of bacteria in coastal beach waters. It was found that the tidal washing process plays the most important role in causing the violation of ENT water quality standard for coastal beach waters, followed by the wash-off process of up to four-day antecedent rainfall.