Skip Links

USGS - science for a changing world

Scientific Investigations Report 2013–5166

Coastal and Marine Geology Program
Prepared in cooperation with the U.S. Environmental Protection Agency, Great Lakes Restoration Initiative

Developing and Implementing Predictive Models for Estimating Recreational Water Quality at Great Lakes Beaches

By Donna S. Francy, Amie M.G. Brady, Rebecca B. Carvin, Steven R. Corsi, Lori M. Fuller, John H. Harrison, Brett A. Hayhurst, Jeremiah Lant, Meredith B. Nevers, Paul J. Terrio, and Tammy M. Zimmerman

Thumbnail of and link to report PDF (2.2 MB) Abstract

Predictive models have been used at beaches to improve the timeliness and accuracy of recreational water-quality assessments over the most common current approach to water-quality monitoring, which relies on culturing fecal-indicator bacteria such as Escherichia coli (E. coli.). Beach-specific predictive models use environmental and water-quality variables that are easily and quickly measured as surrogates to estimate concentrations of fecal-indicator bacteria or to provide the probability that a State recreational water-quality standard will be exceeded. When predictive models are used for beach closure or advisory decisions, they are referred to as “nowcasts.” During the recreational seasons of 2010–12, the U.S. Geological Survey (USGS), in cooperation with 23 local and State agencies, worked to improve existing nowcasts at 4 beaches, validate predictive models at another 38 beaches, and collect data for predictive-model development at 7 beaches throughout the Great Lakes. This report summarizes efforts to collect data and develop predictive models by multiple agencies and to compile existing information on the beaches and beach-monitoring programs into one comprehensive report.

Local agencies measured E. coli concentrations and variables expected to affect E. coli concentrations such as wave height, turbidity, water temperature, and numbers of birds at the time of sampling. In addition to these field measurements, equipment was installed by the USGS or local agencies at or near several beaches to collect water-quality and metrological measurements in near real time, including nearshore buoys, weather stations, and tributary staff gages and monitors. The USGS worked with local agencies to retrieve data from existing sources either manually or by use of tools designed specifically to compile and process data for predictive-model development.

Predictive models were developed by use of linear regression and (or) partial least squares techniques for 42 beaches that had at least 2 years of data (2010–11 and sometimes earlier) and for 1 beach that had 1 year of data. For most models, software designed for model development by the U.S. Environmental Protection Agency (Virtual Beach) was used. The selected model for each beach was based on a combination of explanatory variables including, most commonly, turbidity, day of the year, change in lake level over 24 hours, wave height, wind direction and speed, and antecedent rainfall for various time periods. Forty-two predictive models were validated against data collected during an independent year (2012) and compared to the current method for assessing recreational water quality—using the previous day’s E. coli concentration (persistence model). Goals for good predictive-model performance were responses that were at least 5 percent greater than the persistence model and overall correct responses greater than or equal to 80 percent, sensitivities (percentage of exceedances of the bathing-water standard that were correctly predicted by the model) greater than or equal to 50 percent, and specificities (percentage of nonexceedances correctly predicted by the model) greater than or equal to 85 percent. Out of 42 predictive models, 24 models yielded over-all correct responses that were at least 5 percent greater than the use of the persistence model. Predictive-model responses met the performance goals more often than the persistence-model responses in terms of overall correctness (28 versus 17 models, respectively), sensitivity (17 versus 4 models), and specificity (34 versus 25 models). Gaining knowledge of each beach and the factors that affect E. coli concentrations is important for developing good predictive models. Collection of additional years of data with a wide range of environmental conditions may also help to improve future model performance. The USGS will continue to work with local agencies in 2013 and beyond to develop and validate predictive models at beaches and improve existing nowcasts, restructuring monitoring activities to accommodate future uncertainties in funding and resources.

First posted November 21, 2013

Revised March 27, 2014

  • Downloads Directory
    Contains: PROCESSNOAA executables, source code, and data-sample example files. Refer to the Readme file for more information. Files can be accessed individually or by Zip (515 KB).

For additional information, contact:
Director, Ohio Water Science Center
6480 Doubletree Ave
Columbus, OH 43229-1111
http://oh.water.usgs.gov/

Part or all of this report is presented in Portable Document Format (PDF). For best results viewing and printing PDF documents, it is recommended that you download the documents to your computer and open them with Adobe Reader. PDF documents opened from your browser may not display or print as intended. Download the latest version of Adobe Reader, free of charge.


Suggested citation:

Francy, D.S., Brady, A.M.G., Carvin, R.B., Corsi, S.R., Fuller, L.M., Harrison, J.H., Hayhurst, B.A., Lant, J., Nevers, M.B., Terrio, P.J., and Zimmerman, T.M., 2013, Developing and implementing predictive models for estimating recreational water quality at Great Lakes beaches: U.S. Geological Survey Scientific Investigations Report 2013-5166, 68 p., http://dx.doi.org/10.3133/sir20135166/.

ISSN 2328-031X (print)

ISSN 2328-0328 (online)



Contents

Abstract

Introduction

Study Areas

Methods

Development of Predictive Models

Data Management and Analysis

Quality Assurance and Quality Control

Identification of Variables and Development of Predictive Models

Validation of Predictive Models and Implementation of Nowcast Systems

Future Work

Summary

References Cited


Accessibility FOIA Privacy Policies and Notices

Take Pride in America logo USA.gov logo U.S. Department of the Interior | U.S. Geological Survey
URL: http://pubs.usgs.gov/sir/2013/5166/
Page Contact Information: GS Pubs Web Contact
Page Last Modified: Friday, March 28, 2014, 01:16:09 PM