Nowcasts were provided to the public at 8 beaches during the recreational season of 2014. Performance of the nowcast model was compared to using the persistence method (using the previous day’s E. coli to predict the current day’s E. coli concentration) at 6 of those beaches..
- At 5/6 beaches, the nowcast model performed better than the persistence method in terms of overall percent correct and specificity. Specificity is the percentage of non-exceedances of the E. coli bathing-water standard (235 colonies per 100 milliliter) that were correctly predicted.
- At 4/6 beaches, the nowcast model performed better than the persistence method in terms of sensitivity. Sensitivity is the percentage of E. coli exceedances of the bathing-water standard that were correctly predicted.
- The model performed especially well at Huntington, where three subseason models were used.
- Because sampling was not done on any consecutive days at Fairport Harbor and Headlands, comparison to the persistence method could not be done.
- The Maumee Bay State Park model sensitivity was lower than expected because the Berger Ditch stream gage was removed mid-season and a backup model was used.
- The model performed poorly at Nickel Plate. Work will continue in 2015 to test new models at this beach.
At Huntington, we have been providing nowcasts to the public the longest— 756 days during the recreational seasons of 2006-14.
- The nowcast provided a correct response 85.4 percent of time.
- The sensitivity of the nowcast was 61.9 percent. That means that the nowcast correctly predicted that the standard was exceeded on 83 out of 134 days that the standard was actually exceeded.
- The specificity of the nowcast was 90.5 percent. That means that the nowcast correctly predicted that the standard was NOT exceeded on 563 out of 622 days that the standard was actually NOT exceeded.
The Huntington nowcast continues to provide 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.), especially in regard to sensitivity. In addition, nowcast results have been consistent from year-to-year.
Having many years of data at Huntington afforded the opportunity to test a non-observation model. A non-observational (NOD) model is a model that can be run remotely with available data and doesn’t require a visit to the beach. During 2014, performance of the NOD model was compared to the standard model and prior day E. coli method. The standard model was the model used for the Huntington nowcast and requires a visit to the beach to measure model parameters. Although the NOD model did not perform as well as the standard model, it still performed better than using the prior day E. coli method (see donut graphs below).