Predictive Models

PREDICTIVE MODELS FOR RECREATIONAL WATER QUALITY

Swim advisories or closings are issued by beach managers on the basis of standards for concentrations of bacterial indicators—Escherichia coli (E. coli) or enterococci for freshwaters and enterococci for marine waters.  The analytical methods for these organisms, however, take at least 18–24 hours to complete.  Recreational water-quality conditions may change during this time, leading to erroneous assessments of public-health risk.  As a result, some agencies have turned to predictive modeling to obtain near-real-time estimates of recreational water quality. Predictive models, developed through statistical techniques such as multiple linear regression (MLR), use easily measured environmental and water-quality variables to estimate bacterial-indicator concentrations or the probability of exceeding target concentrations.

Predictive modeling at Five Lake Erie beaches
NOWCASTING: an operational system at Huntington, Bay Village, Ohio
How can we develop a predictive model for our beaches?
Collecting better data for predictive models

Predictive modeling at Five Lake Erie beaches: 
The USGS Ohio Water Science Center (Ohio WSC) has been working to develop predictive models for five Lake Erie beaches in Ohio:  Lakeview (Lorain, Ohio), Huntington (Bay Village, Ohio), Edgewater Park and Villa Angela (Cleveland, Ohio), and Lakeshore Park (Ashtabula, Ohio).  See a detailed report at: https://webarchive.library.unt.edu/eot2008/20081107222810/http://pubs.usgs.gov/sir/2006/5192/.

NOWCASTING: an operation system at Huntington, Bay Village, Ohio
At Huntington, investigations are further along than at other beaches. Predictions based on the Huntington model have been available to the public through an Internet-based NOWCASTING system since May 30, 2006. 

The NOWCAST is like a weather forecast, in that it provides the probability (in percent) that the bathing-water standard for E. coli will be exceeded.  (The Ohio single-sample bathing-water standard for E. coli is 235 colony-forming units/100 milliliters).  So on any given morning, there could be from a 1- to 100- percent probability that the standard would be exceeded.  How does one know when the probability presents too great a risk to go swimming?  Would you go swimming if there was an 80-percent probability that the standard would be exceeded?  What about a 25 percent chance?  To help out, beach managers established a threshold probability of 27 percent for Huntington based on historical data.  If the probability was greater than or equal to 27 percent, than the beach was posted with an advisory on the NOWCAST.

How did the NOWCAST system perform in 2006?  NOWCASTS were provided to the public for 85 days during the recreational season of 2006.  The NOWCAST provided a correct response, 80 percent of time.  False positive responses were provided 10 percent of time; that means that the NOWCAST incorrectly predicted that the standard was exceeded on 6 out of 59 days that the standard was actually NOT exceeded.  False negative responses were higher – 42 percent.  That means that the NOWCAST incorrectly predicted that the standard would NOT be exceeded on 11 out of 26 days that the standard was actually exceeded.

Although the false negative rate for the NOWCAST is higher than we would like, the NOWCAST still provides more accurate information and better estimates of public health risk than the use of the previous day’s E. coli (the current method used by most beach managers.).  During 2006, the previous day’s E. coli provided only 57 percent correct responses.   False positives were provided 30 percent and false negatives 72 percent of time.

We will continue to work to improve the predictive ability of the NOWCAST in 2007.

How can we develop a predictive model for our beaches?
To find out, in a step-by-step fashion how to develop predictive models for your beaches, click on the techniques report.  The steps to develop predictive models are data collection; exploratory data analysis; model development, selection, and diagnosis; determination of model out values; and model validation and refinement.

Collecting better data for predictive models:   
Predictive modeling is a dynamic process meant to augment existing beach-monitoring programs. Models should be continuously validated and refined to improve predictions.

The USGS Ohio Water Science Center is working with Northeast Ohio Regional Sewer District (NEORSD) to collect more accurate wave height data.  In the past, wave heights were estimated by field crews based on visual determinations and placed into one of four categories; they are currently measured by use of a wave measuring stick. To further improve wave-height measurements, the USGS installed a buoy at Edgewater, Cleveland, Oh.  The buoy is equipped with instrumentation to measure wave heights and store the data.  Data from the buoy are transmitted hourly to a wireless access point.