USGS Multimedia Gallery: (Video)--Bayesian Modeling, Fly Fishing, and Effects of Urbanization on Stream Ecology

Song: All right Tom, now you're going to teach me fly fishing.

Tom: Yep. I'm gonna teach you a little bit

about fly fishing. This is a really fun

way to fish. It's a little different than

most forms because what you use for bait

are artificial lures called flies. They're

very light and usually for the kind of

fishing we're doing they're very small.

You usually have – there are two types:

there's a dry fly which floats on the

surface and that's always fun because you

see things coming up and grabbing it off the

surface. And then there are wet flies which

are weighted flies like this little one here.

And these you fish under the water. This

particular one is meant to look like a midge.

Then we have these other ones that

look like worms. Fish will pick those up under the water.

Song: I'm Song Qian with the USGS.

Tom: And I'm Tom Cuffney with the USGS.

And for the last 10 years now, we've been

studying the effects of urbanization

across the United States. We've looked now

in nine major metropolitan areas. The Raleigh

area, Boston, Birmingham, Milwaukee,

Green Bay, Dallas, Fort Worth, Denver,

Portland, and Salt Lake City.

Song: Since three years ago, I started working

with Tom and the USSS and colleagues on

this EUSE project and my role here is

mostly data analysis and modeling.

Tom: Yeah, the EUSE project that we're

working on is the effects of urbanization

on stream ecosystems which is where the

EUSE acronym comes from. We've really been

looking at the effects of urbanization

on fish, invertebrates, algae, habitat,

and chemistry. We've really been using

Song a lot. He's helped us in developing

models to help explain what is happening

in these streams and what helps us compare

among metropolitan areas and also to

predict possible changes in the effects

of possible mitigation procedures.

The flies themselves are so light that they

really can't be cast. So you have to use

the cast in a fly rod is the fly line itself.

The fly line is a colored line that you

have here and these are weight forward

fly lines. That means that most of the

weight is in the first section of the

fly line. You use this to lure the fly.

The clear mono filament part, this is

the leader and this is what connects

the fly to your fly line. You're

constantly replacing this and you use

different weights and leaders depending on

what type of fishing you're doing. If you're

fishing for very small trout, you may

use very light line. If you're fishing

for bass, you might use a much

heavier monofilament line.

Song: One particular thing we did was a

multi-level model and another thing we

did was Bayesian network modeling to

understand how the land-use changes

affect the stream ecosystems.

Tom: Now the multi-level models were

very important for me at least because it

helped us explain why we were seeing

different relationships between urbanization

and stream effects in different parts of

the country and that method really helped

us explain why we were seeing these different results.

The trick to fly fishing is that the fly

rod itself is a big spring and you need

to use that spring to cast a line.

The nice thing about the multi-level

models from the perspective of application

is that it showed that there were other

interferences in terms of factors that

were affecting urbanization like agriculture

in Milwaukee, Green Bay, and Denver,

and Dallas, Fort Worth which turned

out to be, were masking the effects we

saw of urbanization as well as the

ability to incorporate the climatic

factors of moisture and precipitation

which varied among the metropolitan area.

Song: These are examples about variables

operating at very different spatial-temporal

scales. We're looking at land use cover,

urban cover, agricultural coverage

which is something that varies at the

watershed scale but when we talk about

the background or antecedent agricultural

land use, that's in the regional scale.

Tom: The other thing that we did was looking

at the Bayesian Network models. Those are

very interesting models because they

employed a lot of prior knowledge and

that's something I think was rather

unique in that – was the ability to go

out and find experts and talk to them

about what should be happening in these

systems and what they expected to happen;

and then actually looking at the data.

So would you talk a little bit about how that process works?

Song: All right, so this is the one big

difference between Bayesian statistics

and classical statistical inference.

In the classical statistical inference

we based our inference purely on data

and all the other previous experience

and all this doesn't count. Essentially

every time when you go out and collect

data and try to develop a model, you are

trying to reinvent the wheel. With

Bayesian statistics, you can actually

logically combine information from various different sources.

Tom: There are other types of casts

where you can roll cast like this, which

is where you just don't have the line

behind you. You just pull it back and

roll it over. That can be used a lot

when you have something behind you that

prevents you from false casting like this.

All right, should be go upstream and give her a try?

Song: Yeah.

Tom: All right. From an ecological standpoint

in terms of just trying to understand

ecology, we've also been working on a

number of models looking at understanding

specific elements of the biology.

Song: That's right so the statistics

fundamentally is a tool for learning.

Science is used in two different ways.

One is for helping management and helping

support decision making. The other is

learning and making inferences about

what's behind the natural phenomena.

From the statistical perspective,

statistics is always a tool for learning from data.

I could really get used to this!

Tom: Oh yeah man! You gotta get a rod!

So Song, over the last three years we've

done a lot of modeling that's helped us

both understand how urbanization changes

across the country, as well as to look

at possible things that we can do to

mitigate those effects. In a nutshell

what do you think are the three most

important things that we've come up

with in the last couple of years?

Song: I would say the first thing is

that complicated systems need a

little bit more complicated models.

Tom: And I think the fact that we're

seeing different responses to urbanization

in different regions of the country and

that the antecedent land use has a

strong effect on how urbanization

impacts systems is important.

Song: That's one thing that we feel

is very interesting, an interesting point;

the most interesting outcome from the multi-level analysis.

Tom: And then the third thing, I think,

that's been really important is the

Bayesian network models that have

allowed us to make relatively simple

models so managers can make adjustments

to kind of get some perspective on

what management decisions would do

in terms of changing the resources.

Song: That's right and the Bayesian

network model allows us to pull

information from various sources

and make a sound scientific judgment and management strategy.

Tom: So working with you in the last

few years, we've come a long way from

those simple regression models that I did initially.

Song: And to me working with you guys is,

I get to come out and actually look at

how the data was collected and

the real stream, the real ecosystem.

Tom: So it's been good talking to

you but let's get back to fishing.

Song: Yeah, let's go. This is just like learning statistics.

Tom: So what's the probability that you're gonna catch a fish?

Song: Well so far, zero!

Tom: That's the same as my probability. I think this is very classical.

Song: Yeah.

[End of Audio]