Forecasting Beach Water Quality Like the Weather

“One hundred years ago, we could not predict whether it would be sunny or rainy the day after tomorrow. Now we can predict the weather as much as 10 days in advance. By the middle of the 21st century, we ought to be able to predict the weather at the beach…both above and below the water line.”

What could fundamentally improve our chances of having a safe, fun summer swimming at the Jersey shore – but without the manpower and laboratory costs of taking water samples every day?

When it rains, stormwater outfalls discharge high levels of enterococcus into the ocean that causes most beach advisories. Especially when the beach is near a stormwater outfall (map on page 20).

How can you guess your risks when it rains but no water samples are being taken? Fundamental change can come by using predictive models to forecast beach water quality.

Forecasting water quality every day of the week is a cost-effective supplement to sampling. It's been done for years at beaches along the Great Lakes. The Ohio Nowcasting model dates from 1998. The United States Geological Survey has been partnering with local and state agencies since 2006 – now in Ohio, Wisconsin and Illinois.

California began testing their Water Quality Nowcast at three marine beaches this summer.

Nowcasting is a “predictive model”. It uses environmental and hydrodynamic conditions to predict bacteria levels at bathing beaches - more accurately than just sampling once a week, using day-old sample results (page 3). Nowcasting accurately predicted water quality at two Ohio beaches 80% of the time, while sampling alone was 62% accurate.

Sampling is a "persistence model" - today equals tomorrow. It assumes that yesterday's bacteria levels can be used to estimate today's (page 2). But a lot can change in 24 hours – the wind can shift and blow stormwater away from the swimming area where it is diluted offshore. A study of beaches in 8 states along the Great Lakes showed that advisories based on day-old sample results were wrong about 2 out of 3 times (Table 1, page 2).

That 24-hour wait causes false positive errors - posting an advisory on Tuesday based on Monday's results, only to find out on Wednesday that Tuesday's bacteria levels had already dropped below the standard. Taking one sample a week causes false negative errors - missing potential exceedances during the other 6 days of the week.

Additional sampling needs to be done so that each beach can be assigned its own unique “threshold probability” for issuing an advisory. The model is essentially a compromise between false positive and false negative errors.

The goals for Ohio's Nowcasting model are to be 5% more accurate than advisories that are just based on sampling, more than 80% accurate for predicting high levels of bacteria, and 85% accurate for predicting acceptable bacterial levels.

You can find their forecasts on their website, on Twitter at @NEORSDbeaches, or by using the myBeachCast mobile app.

As you would expect, a model that forecasts water quality at a lake beach will be different from one developed for an ocean beach.

One big difference is due to the grain size of the sediment in lakes. Nowcasting found that turbidity from stirred-up lake sediments is actually better than rain at predicting bacteria levels in the water. That's because enterococcus and E. coli can thrive in fine sediments. Lake sediments have a lot more silt and clay fines than sand beaches pounded by ocean waves.

Now Standford is testing their predictive model at three marine beaches in Southern California this summer.

They partnered with UCLA's Institute of the Environment and Sustainability, and Heal the Bay, and are funded by the California State Water Resources Control Board. Their initial research is described in these papers: “Sunny with a chance of gastroenteritis: predicting swimmer risk at California beaches”, and “Predicting water quality at Santa Monica Beach: evaluation of five different models for public notification of unsafe swimming conditions”.

Their goals are to develop a model that is 10% more accurate than advisories that are just based on sampling, 30% accurate for predicting high levels of bacteria, and more than 90% accurate for predicting acceptable bacterial levels (pps. 425 and 428).

So far, they found that the most important environmental predictors of water quality at their beaches were rainfall and tide. They have found that when there is more sunshine there are lower levels of bacteria, since sunlight kills bacteria. And they found that models miss unusual events, like a sewage spills.

Their research has also revealed that marine forecasting models will be driven by the regional climate. For example, rainfall was not the most critical factor for predicting water quality at the beach in Santa Monica (page 113).

How could that be? Because: “The summer dry weather in California also contributes to the weaker dependence of [bacteria] concentrations on rainfall; there is rarely measurable rainfall in the summer season” (page 113). And: “Rainfall in the summer is usually due to trace rainfall events due to the passing of the monsoonal storms” (page 429). Los Angelos gets a little more than 15 inches a year of rainfall - NJ gets 40-51 inches.

You can find their forecasts on their website, on Twitter at @BeachReportCard, or or by using their mobile app.

Weekly sampling is expensive. Public Health has not done well since the recession. More sampling would mean higher state and local taxes, mostly for manpower.

That's why the EPA has been nudging states to use forecasting models to supplement their water sampling since 2012.

Previous blogs about forecasting marine water quality:

It's Not Just the Latest Rainfall that Closes Beaches

Wind and Currents Push Stormwater Bacteria Offshore – or into the Swimming Zone