The current shift from persistence models to predictive tools is a major move in making beach water quality information more timely for swimmers. Predictive tools, such as statistical models, allow experts to predict the current quality of water like they predict today’s weather. For beach goers, this means water quality information is becoming increasingly reflective of the water they’re about to jump into, rather than retrospective of the beach they visited yesterday.
In a previous article we outlined how environmental health officers and other beach managers measure beach water quality. Currently, persistence modelling is the most common and accessible method used around the world to monitor the risk of illness for bathers. Persistence models use culture-based testing to measure the concentration of fecal indicator bacteria (FIB) in the water. Worldwide, E.coli and enterococci are the standard FIB for recreational water (you can learn more about them, and other causes of waterborne illness, here.).
Persistence models can also include more rapid DNA extraction methods, such as quantitative real-time polymerase chain reaction (qPCR) to enumerate indicator bacteria. Rapid methods, such as qPCR testing, can generate results in as quick as 3 hours (not including collection and transport time), while culture-based methods require a lab and on average 18-24 hours to get results. In fact, inoculation or membrane filtration can take anywhere from 24 to upwards of 48 hours from collection of sample to reporting of results.
This means if a sample is taken at 8:00am on Monday morning results will typically not be available until at least 8:00am Tuesday, and possibly as late as Wednesday. Water conditions can change dramatically in the span of 24 hours and the results from Monday morning being used to inform beach goers on Tuesday or Wednesday may not be reflective of the current conditions.
So, while effective in measuring FIB in water, the time-delay inherent in the persistence approach prevents up-to-date, informed decision-making for beachgoers. In other words, beachgoers base their decision whether or not to swim today based on retrospective water quality information. That’s where predictive tools come in. Predictive methods provide water quality information for beachgoers before they hit the beach by estimating water quality and the risk of illness for recreational water users in a timely manner.
In a nutshell, predictive modelling is like weather forecasting for the water.
The primary objective of recreational water quality monitoring is to protect public health and safety. That is, results are used to inform the public whether water is clean and the level of risk of illness it poses to recreational water users. The US Environmental Protection Agency (EPA) realized that the current approach was preventing them from providing the public with accurate and timely information. The EPA pioneered a shift to predictive modelling as a way to address that issue.
There are a number of predictive tools used in estimating water quality. The most common of these tools, predictive modelling combines historical and current FIB data with other environmental variables to build a statistical model that can estimate current concentrations of FIB in the water. Inputs include variables such as day of year, turbidity, change in water level, wave height, antecedent rainfall, bird counts, and the direction and speed of the wind. Often referred to as “nowcasting”, these predictive models make day-to-day estimates about concentrations of FIB in the water. While less common, some more advanced forecast models can actually predict conditions up to two days in advance.
Other predictive tools include rain threshold models, deterministic models, decision tree, and binary models. You can learn more about them here.
Results from predictive models are presented in the same way as those from persistence models so they can still be measured against existing local standards of water quality, like the EPA’s beach action value. The difference, however, is it that results for Tuesday mornings water quality are actually reflective of Tuesday mornings water.
Not only are predictive models more timely, but they can also more accurately estimate recreational water quality and health risks to swimmers.
Predictive models are constructed using statistical software. The EPA’s Virtual Beach, for example, is widely used. With the exception of “white box” models, predictive models are constructed on a beach by beach basis to account for site-specific environmental variables. This allows experts to constantly test the strength of their model against itself and make frequent adjustments to improve accuracy. Further, like most good things in life, statistical models only get better with age as they continue to accumulate data.
As with any type of statistical modelling, there is room for error. But so far the results from several predictive monitoring pilot projects have been overwhelmingly positive, for both their accuracy and timeliness. For example, in collaboration with experts from UCLA and Stanford, Los Angeles based Heal the Bay’s pilot project found that predictive modelling accurately predicted 25% more days where water quality was unfit for recreational usage at a select group of L.A. beaches. Similarly, the United States Geological Survey (USGS)’s Great Lakes pilot project has found predictive models to be consistently more accurate than persistence models across a number of categories, including overall correctness, sensitivity, and specificity. Both of these projects “nowcast” daily results.
In New Zealand, the Auckland Council’s Safeswim program has launched its own predictive model platform. The modelling tool provides water quality estimates for over 50 swimming locations in the region. Safeswim currently employs two different predictive tools: a “black box” model and a proof of concept “white box” model. “Black box” models are more simple linear models that use statistics to predict FIB. They are achieving similar results to the projects in California and the Great Lakes. “White box” models are more complex, and require knowledge of local contamination sources and the environmental relationships and processes that influence the movement and abundance of the contaminants. While very much in their pilot phase, the “white box” models are proving to provide more nuanced and accurate information for beachgoers before they get in the water.
Like the weather, water changes quickly and constantly, and a wide range of environmental factors can impact its quality in a short period of time. Predictive models offer more timely and accurate information, allowing beachgoers to be informed and aware of the water they are getting into. Predictive modelling is the new gold standard for recreational water quality monitoring, and soon when you go to the beach you will be able to check the forecast for both the weather and the water.
© Lake Ontario Waterkeeper, 2011 - 2018