You are here:
Predicting Fecal Indicator Bacteria Concentrations in the South Fork Broad River Watershed Using Virtual Beach
Spidle, D., M. Molina, AND Mike Cyterski. Predicting Fecal Indicator Bacteria Concentrations in the South Fork Broad River Watershed Using Virtual Beach. Upper Oconee Watershed Network Annual Summit, Athens, GA, September 30, 2016.
Poster presented at Upper Oconee Science and Policy Summit, University of Georgia, September, 30, 2016.
Virtual Beach (VB) is a decision support tool that constructs site-specific statistical models to predict fecal indicator bacteria (FIB) at recreational beaches. Although primarily designed for making decisions regarding beach closures or issuance of swimming advisories based on exceedance to the FIB criteria, VB can also be used for studying relationships between any water quality indicator and ambient environmental conditions. Our objective was to evaluate the effectiveness of statistical models developed using VB for predicting the impairment of inland rivers and streams. An intensive field study was conducted from 2012-2015 in which water samples were collected during rainfall events and baseflow conditions from two sites on the South Fork Broad River watershed located in Madison and Oglethorpe County, Georgia . Samples were analyzed for E. coli and Enterococci along with other water quality parameters. Specifically, data from October 2012 to September 2014 were used to develop a multiple linear regression (MLR) model describing the relationship between E. coli and Enterococci and a set of independent variables (IVs): turbidity, total suspended solids (TSS), rainfall and water temperature. IV data from the third year of the study, October 2014 - September 2015, were used to assess the predictive capacity of the MLR model. R-squared values for the MLR models developed by VB using the first two years of data were 0.82 for E. coli and 0.82 Enterococci at the Clouds Creek site (N=284). For the Carlton site, R-squared values were 0.81 for both E. coli and Enterococci (N=190). Model accuracy (i.e., the model prediction agrees with the observation, in terms of the predicted response variable being above or below the regulatory standard) ranged from 96-99% for E. coli and 100% for Enterococci. For model assessment purposes, 126 CFU (E. coli) and 35 CFU (entero) were used for both the Decision Criteria (DC) and Regulatory Standard (RS). Predictions from the selected models using the IV data from the third year of the field study produced an accuracy of 96-98% for E. coli and Enterococci using the RS and DC values set for model development. Our preliminary results show statistical models developed with VB can be effective for predicting the impairment of inland rivers and streams.