EPA Research Partner Support Story: Predicting water quality at beaches

Partners: Local and regional beach managers across states that border the Great Lakes, as well as other states
Challenge: Predicting water quality at beaches
Resource: Virtual Beach software
Project Period: 2007 – Present

To protect public health, beach managers need to continually assess the level of potentially harmful microbes (primarily bacteria) in the water. However, traditional culture-based testing methods can take 24 hours to get results – preventing same-day, proactive beach closures and leaving many recreational swimmers open to sickness or infection, or potentially close a beach needlessly and incur economic losses. EPA’s Virtual Beach software offers a solution.

“This reliable, predictive water quality model is key to protecting health and promoting recreational enjoyment of our beaches. The model provides same-day public notifications of beach conditions at a lower cost than traditional monitoring. Communities that use Virtual Beach can dedicate more of their resources to locating and correcting sources of contamination and improving local beaches. The (Wisconsin DNR’s) partnership with EPA in the development of this practical scientific tool offers a great pay off.” – Wisconsin DNR former Secretary Cathy Stepp

Virtual Beach facilitates efforts to support the local economy while protecting the health of residents. Virtual beach is used to assist in advisory issuances in the Great Lakes states and to forecast water conditions in numerous locations in Illinois, Indiana, Maryland, Michigan, Minnesota, New York, Ohio, Pennsylvania, Rhode Island, South Carolina, and Wisconsin. In recent years, the software has been used for modeling water quality at water intake pipes (due to harmful algal bloom concerns) and shellfish harvesting areas (fecal coliform of typical concern) along the southern Atlantic coast.

An updated, web-based version of Virtual Beach is currently being developed. It will provide cutting-edge analytical tools, like ensemble modeling using a variety of machine-learning algorithms, as well as methods for handling non-detects and missing data, to maximize the predictive accuracy of water quality models while not over-fitting the training data which leads to poorer predictions. This package will be a general analytical tool that can be used for a large variety of site-specific modeling questions, e.g., at recreational beach sites, shellfish harvesting areas, or public drinking water intake locations.