Related Experiment Video
Updated: May 27, 2026
12:14
A Duplex Digital PCR Assay for Simultaneous Quantification of the Enterococcus spp. and the Human Fecal-associated HF183 Marker in Waters
Published on: March 9, 2016
Development of predictive models for determining enterococci levels at Gulf Coast beaches
Zaihong Zhang1, Zhiqiang Deng, Kelly A Rusch
1Department of Civil & Environmental Engineering, Louisiana State University, Baton Rouge, LA 70803-6405, USA. zzhan15@tigers.lsu.edu
Water Research
|December 2, 2011
Summary
This study compared artificial neural network (ANN) and US EPA Virtual Beach (VB) models for predicting enterococci levels at beaches. The ANN model showed superior performance in reducing errors, aiding beach water quality management.
Area of Science:
- Environmental science and engineering
- Water quality monitoring and modeling
- Public health and beach safety
Background:
- The US EPA BEACH Act mandates swimming advisories when water quality standards are exceeded.
- Existing methods for predicting water quality lack systematic performance comparisons.
- Enterococci levels are key indicators of fecal contamination and potential health risks at beaches.
Purpose of the Study:
- To systematically compare the performance of different models for nowcasting and forecasting enterococci levels.
- To evaluate an artificial neural network (ANN) model against two US EPA Virtual Beach (VB) models.
- To assess the utility of these models for improving beach water quality management and public health protection.
Main Methods:
- Developed and compared an ANN model and two VB models (linear and nonlinear) using 944 data sets from Holly Beach, Louisiana (2005-2010).
- The ANN model incorporated 15 environmental variables, while VB models used 5-6 input variables.
- Model performance was evaluated using linear correlation coefficient (LCC) and Root Mean Square Error (RMSE) through training, validation, testing, and hindcasting.
Main Results:
- The ANN model demonstrated superior performance in terms of RMSE (0.803 adjusted) compared to the linear (1.815 adjusted) and nonlinear (1.961 adjusted) VB models during hindcasting.
- VB models showed better performance in terms of LCC (0.354 for linear, 0.521 for nonlinear) compared to the ANN model (0.320 adjusted).
- The ANN model, with more parameters, achieved a lower RMSE, indicating better accuracy in predicting enterococci levels.
Conclusions:
- The ANN model performs better than VB models in reducing prediction errors (RMSE) for enterococci levels.
- Predictive models, particularly ANN and nonlinear VB, can be combined with real-time data for effective beach water quality nowcasting and forecasting.
- These modeling approaches can significantly reduce health risks associated with contaminated beach waters and enhance beach management strategies.
