Granite Geek: Predicting Dangerous E.coli Levels in N.H. Lakes

Aug 10, 2016

It’s easy enough to check the weather before you head out for a hike. Maybe you click on the weather app on your smartphone, scan for thunderstorms, and plan accordingly. But when it comes to going for a swim, real-time information on water conditions is not just a click away.

After several outbreaks of the bacteria E. coli in New Hampshire lakes this summer, the state has been looking into using a software program that could predict dangerous water conditions. Granite Geek David Brooks, a reporter for The Concord Monitorspoke with NHPR’s Peter Biello.

How does this software work?

It’s something called “Virtual Beach” that was developed by the EPA, and I believe it was developed initially for salt-water oceans. The U.S. Geological Survey—the office in Pembroke—is working with a couple of state agencies, seeing if they can use it with some specialty buoys they have built to predict outbreaks of E.coli on beaches. They’re testing with buoys on Lake Winnepesaukee on Weirs Beach and Pawtuckaway State Park.

You write in your column that some other parts of the country have already been using similar software in their lakes, but that New Hampshire lakes might be trickier to monitor—why is that?

It’s nice and flat in the Midwest and conditions tend to be relatively uniform over very large areas, so when you’re doing environmental monitoring, it’s a little bit easier. But to make it work here, it would have to be lake-specific. There would have to be one set of parameters for Lake Winnepesaukee and another one for Pawtuckaway and Silver Lake and Hollis.

Is that because maybe one lake is more shaded and doesn’t get direct sunlight, and would be different from another lake when it comes to bacteria growth?

Yeah, and underlying geology, and what’s leeching into the water, and where the wind is coming from…

And you write that these buoys are going to cost $20,000 per year. Wouldn’t it be more efficient to just count the bacteria?

That’s what I asked, and that’s what I thought, frankly, that the buoys were doing. The problem is that in order to get a sense of how much E.coli is in it, you’ve got to culture it, which takes a minimum of 24 hours. Therefore, you can’t get real-time data by just sucking in the water and counting the E.coli and getting a number. It just doesn’t work that way.

So the software is trying to predict it indirectly. This summer, the buoys are gathering data on all sorts of factors, things you and I could understand, like the water temperature. They’ve got on-shore meteorological stations that are gathering air temperature and wind direction at the same time. The buoys are measuring dissolved oxygen and the conductivity of the water (which apparently makes quite a difference).

So they’re using all these indirect factors to correlate them to when there’s an outbreak. So we’ll have all these large sets of data: here’s what happened with all these factors, and here’s what happened with E.coli all summer long at this location. Based on all of those factors, they’ll try to predict when an E. coli outbreak might happen.

So they’re solving for X, in other words. If they have conditions A, B, and C, then it follows that X will be the condition, and therefore the lake might not be safe.

Yes. This is classical mathematical modeling. It’s a big, complicated equation, where you put in lots of variables. Researchers perform what they call “multivariate analysis” to find out which variables are more important in correlation to which other ones.

So more complicated than what I just spelled out, which begs the question: how effective are these models in the physical, unpredictable world?

You mentioned weather forecasting. The reason weather forecasting is five gazillion times better than it was when I was a kid is partly because we have better machines to measure it but mostly because we have better models to understand what’s happening in the atmosphere.

They’re mathematical models. They’re like giant spreadsheets, but they’re not that simple, of course. All sorts of variables get put in and because of experiments and data that has been gathered over the past, you can give correlations and probabilities of certain things happening as a result.

Mathematical modeling is one of the backbones of modern life and sometimes we tend to sort of sneer at it—“oh, it’s just a model, models can tell you anything you want”—but that’s not really true. A good model has predictive power. People have always wanted models to tell them the future. Well, a mathematical model can do a better job of telling the future than anything human beings have ever developed. And that’s what they’re trying to do here in New Hampshire lakes.