Combining real-time bacteria models and uncertainty analysis for establishing health advisories for recreational waters

Tools are needed to allow accurate and timely prediction of water quality so that recreational users can make informed decisions about the safety of water, and beach managers can post updated advisories. In this paper, we describe the development of a health advisory system based on the probability that real-time estimates of sewage indicator bacteria levels exceed water quality standards. Real-time estimates of bacteria concentrations were made using multivariate linear regression models and real-time input data. Probability distribution functions based on the model results and their associated variances were used to determine the probability that predicted bacteria levels exceed water quality standards. The models were developed for the Mystic River watershed, an urban watershed near Boston, using Enterococcus bacteria data measured during the summers of 2002 and 2003. The linear regression models had adjusted-R2 values of 0.55-0.82 for two river sites, and 0.42 for both a beach on a lake and a boathouse near a dam. Independent variables with predictive power included precipitation and the time since the last rainfall. The probabilistic models for the beach and the boathouse sites correctly predicted water quality exceedances and nonexceedances with > 85% accuracy.