Towards automating measurements and predictions of Escherichia coli concentrations in the Cuyahoga River, Cuyahoga Valley National Park, Ohio, 2012–14

Nowcasts are systems that can provide estimates of the current bacterial water-quality conditions based on predictive models using easily-measured, explanatory variables; nowcasts can provide the public with the information to make informed decisions on the risk associated with recreational activities in natural water bodies. Previous studies on the Cuyahoga River within Cuyahoga Valley National Park (CVNP) have found that predictive models can be used to provide accurate assessments of the recreational water quality. However, in order to run the previously developed nowcasts for CVNP, manual collection and processing of samples is required on a daily basis to acquire the required explanatory variable data (laboratory-measured turbidity). The U.S. Geological Survey and the National Park Service collaborated to develop a more automated approach to provide more timely results to park visitors regarding the recreational water quality of the river.

In May 2012, an in-stream water-quality sensor was installed by the U.S. Geological Survey at Jaite, Ohio (a site centrally located in CVNP on the Cuyahoga River), to provide near-real-time measurements of turbidity and water temperature. To transition from methods used during previous studies at CVNP, a relation between laboratory- and in-stream measured turbidity was developed after the recreational season of 2012. During the recreational seasons of 2012 through 2014, discrete water samples were collected and processed to determine Escherichia coli (E. coli) concentrations at Jaite and one site upstream of Jaite (Lock 29) within CVNP. Predictive models, using in-stream turbidity measurements, were developed for the recreational seasons of 2013 and 2014 to estimate recreational water quality in regards to Ohio’s single-sample water-quality standard for primary-contact recreation.

A computer program was developed to manage the nowcasts by running the predictive models and posting the results to a publicly accessible Web site daily by 9 a.m. The nowcasts were able to correctly predict E. coli concentrations above or below the water-quality standard at Jaite for 79 percent of the samples compared with the measured concentrations. In comparison, the persistence model (using the previous day’s sample concentration) correctly predicted concentrations above or below the water-quality standard in only 68 percent of the samples. To determine if the Jaite nowcast could be used for the stretch of the river between Lock 29 and Jaite, the model predictions for Jaite were compared with the measured concentrations at Lock 29. The Jaite nowcast provided correct responses for 77 percent of the Lock 29 samples, which was a greater percentage than the percentage of correct responses (58 percent) from the persistence model at Lock 29.