Hurricanes, Hypercanes, and the Limits of Prediction
Release Date: 08-21-2009
Math professor Christopher Danforth's research aims to improve our ability to make accurate predictions of physical systems — like the paths of hurricanes — with the use of numerical models.
Today, Hurricane Bill continues on its projected path to pass between Bermuda and the U.S. East Coast. It's one of the roughly ninety hurricanes that form each year around the world. Although only ten or so hurricanes arise in the Atlantic annually, they receive a great deal of attention because in the United States people continue to flock to vulnerable coastal locations — despite thousands of deaths and multi-billion-dollar damages. Recall Katrina.
And in Caribbean countries poverty and inappropriate building materials make many populations vulnerable to a hurricane's ferocious winds and punishing storm surge.
Making matters worse, storm forecasting remains a humbling science. Predicting hurricane tracks has slowly improved over recent decades — but much remains unclear about their chaotic physics. Just hours before a strong hurricane strikes land, forecasters may have a hard time knowing whether it will grow to become a monstrous "category five" storm or diminish to something more mild.
Earlier this year, one of the world's leading meteorologists, Kerry Emanuel, director of the Program in Atmospheres, Oceans and Climate at MIT, made a visit to the University of Vermont to speak on Divine Wind: the History and Science of Hurricanes, his remarkable book combining art and physics.
The day after his lecture, UVM Today sat down with Emanuel and two UVM scientists who are working to better understand hurricanes: physical geographer and Vermont state climatologist Lesley-Ann Dupigny-Giroux, and applied mathematician Christopher Danforth. I wanted to know how their divergent fields of research were shining light on the dark and stormy mess we call a hurricane. Some selections from the conversation follow:
UVM TODAY: Since Hurricane Katrina, there has been more public discussion about the connection between climate change and hurricanes. Why has there been so little scientific research on this problem until the last few years?
KERRY EMANUEL: Just as Katrina excited a lot of interest in the public, it also excited a lot of interest among scientists. Before Katrina there was almost nothing written about this connection, just a handful of people working on it.
I published a paper in 1987 in Nature, the first one in the modern era to even address the question of how the destructive potential of hurricanes could rise with a warming climate. But it just lay dormant for a long time. Since Katrina, a lot more scientists are working on the issue. So progress in that corner of science has been much more rapid.
What have you been learning?
EMANUEL: Well, I think we now understand that it's a complex problem! Not everyone admitted that until recently.
Chris, you recently told me that if we had billions of probes every foot of the atmosphere and ultra-powerful computers, we'd still be unable to predict the weather beyond two weeks. And guessing the track of a hurricane is hard even a few days in advance. Why is forecasting so difficult?
CHRIS DANFORTH: Given a perfect computer, complete knowledge of the atmosphere, and a thermometer everywhere you go, in two weeks your weather forecast is no better than a coin flip.
This limit of predictability is a function of the chaotic stuff that happens inside those square-foot boxes you mentioned. That variability starts to matter on the timescale of a few days and it spreads. So, we're never going to forecast well beyond that point. But we are making incremental progress toward making a five-day weather forecast better than it is today.
But the resolution required to forecast hurricanes is much finer than what we have for observing the atmosphere now. The first problem is defining where the hurricane is now and what are the conditions surrounding it — that's your guess.
Then, taking that guess, you're trying to extrapolate into the future using equations that describe fluids. The goal is to ensure that this "cone of uncertainty" that the National Weather Service issues ends up containing the actual path of the storm, say, four days before it hits the coast. You want that part of the coast to be in the cone!
It doesn't always happen. And that's related to the second issue — modeling error — more than mistakes in measuring the initial state. That's why my research is focused on mathematical techniques for reducing these modeling errors.
EMANUEL: It's hard to track hurricanes, but forecasting their intensity may be even more daunting. With today's technology and models, people have a sense of the limits of predictability for a hurricane track. But intensity forecasting hasn't even progressed enough that we understand the basic limits of predictability.
We might be able to say: that hurricane will go into New Orleans tomorrow, with a fifty-kilometer margin of error. But we might not able to say very much about its intensity — ever! The feeling right now is that we just don't understand the physics--but even if we did, there may be fundamental limits to predicting storm intensity.
Imagine I'm a Hollywood filmmaker and I come to you and say, "I want to make a movie about the most ferocious possible hurricane that there could actually be on the planet; how big should I make this storm?"
EMANUEL: People have thought about making films of "hyerpcanes," which might result if you had an asteroid impact in the ocean that superheated a patch of ocean water. Then you get something that isn't terribly big geometrically but that was very, very strong, like Mach 1 winds — and going up very far into the stratosphere.
Is there evidence that this happened in history?
EMANUEL: Nope. Nor is it likely that there would be such evidence because these would likely be open-ocean events and I can't think of a proxy recorder of them. This is a theoretical entity, the hypercane.
OK, asteroid impact aside, what are some worst-case scenarios for hurricanes?
EMANUEL: One place that we have never, to my knowledge, recorded a hurricane is the Persian Gulf. It's dry there. It doesn't generally rain at all. But when we do downscaling of climate simulations, even in the current climate, and produce hundreds of thousands of artificial hurricane tracks, once in a while one will wander into the Persian Gulf from the Arabian Sea.
I think we have 240-mile an-hour winds forecast in some of those events. It's an outside possibility, but it almost happened a few years ago: one got into the entrance to the Persian Gulf. If a hurricane gets into the Persian Gulf in July, all hell would break loose because the Persian Gulf surface waters can heat up to about 35 degrees centigrade. You can imagine what would happen in Dubai.
I saw a picture in National Geographic of a new skyscraper in Dubai where each floor turns separately, moved by the wind.
EMANUEL: Those people are going to get centrifuged out to Oman someplace!
Is there frustration on the part of policymakers about the intrinsic uncertainty in hurricane forecasts?
LESLEY-ANN DUPIGNY-GIROUX: I don't know that this frustration is specific to hurricanes. It's more general. When you show graphs that have error bars, people are always surprised to see the size of the error bars.
Explaining error bars and uncertainty makes them understand a little bit better why some models don't always work. Many people are more familiar with the idea: "this goes in, that goes out." But in atmosphere/ocean/land systems, it doesn't necessarily work that way. There are a lot of feedback loops. When people learn about them, they start to say, 'Oh, that's why science can't always give a definitive answer."
It's important to realize that uncertainty is not a bad thing. It's just the nature of some systems. I work with the public all the time. When I say "uncertainty," a lot of people, who are not scientists, think it means: "well, she doesn't know what she is saying." But to a scientist or engineer, having a clear sense of uncertainty means you know what you don't know!
EMANUEL: A lot of this problem has to do with the fact that our K-12 science education is stuck in the nineteenth century. I took courses from Edward Lorenz [a pioneer of chaos theory] in the 1970s and there was wholesale denial of the ideas of chaos — in the physics department at MIT! So it's not surprising that we don't teach science well in our K-12 schools.
Science education is supposed to prepare the average adult to cope with the modern world. To the extent that it doesn't prepare students to think about uncertainty, it's a terrible disservice.
So how do you help the public understand why they should be skeptical of a four-day hurricane forecast, but confident in a hundred-year climate change model?
EMANUEL: There are even a lot of scientists — skeptics — going around saying, "if we can't forecast the weather ten days from now, what business do we have making a climate forecast for fifty years?"
To which an answer is: OK we can't predict the weather ten days from now, but we sure can tell you that next January will be colder than this July. Now why is that? People start scratching their heads.
It's because there is both chaotic variability and there is a forced variability having to do with the season. People understand that you can't forecast the weather in ten day and that you can sure bet that January will be colder than July. You can use that understanding to say this is what we are talking about in the case of climate change — it's forced variability.
DANFORTH: When I go to talk to people who aren't mathematicians — like middle-schoolers — I show them Bob Barker from the Price Is Right standing next to this Plinko board. Have you seen this game?
A guy drops a little chip at the top and it bounces off a series of pegs and lands in a bin at the bottom. I ask the students: what do you think the chances are that — if I measured the starting point really, really carefully — I could predict which bin it's going to land in?
The answer is: I don't have a chance in hell of doing so no matter how careful I am about my initial measurements. There's a coin flip taking place every time this round disk hits a round peg. By the time the chip is hitting its fourteenth peg, I'd need to know where Bob Barker is standing because his gravitational attraction on that chip starts to matter!
But if you ask me what the probability distribution is of where the chips are going to land if I play this game ten million times, we can be very sure about that and make a good guess. Examples like that help people contrast predictability with statistical averages.
EMANUEL: That's a great example. And climate change is just tilting the whole Plinko game a little bit, or a lot!
What does the historical record reveal about past hurricanes?
DUPIGNY-GIROUX: There is still a lot to learn about historical hurricanes. Some of the older hurricanes are still being discovered and the tracks being cleaned-up: getting the frequency right, getting the location clear, understanding the impacts by going back through diaries and so on.
EMANUEL: But in many parts of the world it's pretty slim pickings. [To Dupigny-Giroux] Have people gone through ships' logs as carefully in the Pacific as they have in the Atlantic?
DUPIGNY-GIROUX: The Atlantic is the hotbed right now. But, you're right, some of the Pacific records and ships logs are a gold mine waiting to be found.
EMANUEL: Also, the field of paleo-tempestology is promising for trying to get much longer records of hurricanes in certain parts of the world. The problem is that techniques have been developed to maturity, like studying overwash deposits, but the folks that fund technique development, like the National Science Foundation, aren't interested in it anymore. There isn't an agency that will just pick up these techniques and apply them.
What do you see as the most important problems or unanswered questions in hurricane science?
DANFORTH: We've gone a long way toward getting a better sense for what the atmospheric state is at any given time and for what it's been like in the past. But that doesn't address the fact that the physical models we build of the atmosphere are very sensitive to the choices we make for how we represent it. When we "draw" the atmosphere, how many grid points should there be? A billion or a trillion?
This is important because when you change that resolution you dramatically change the output of the model.
In some of the best climate models out there today, if you double the resolution — presumably resolving more phenomena, maybe even hurricanes — all hell breaks loose. The models break! With higher resolution, they don't do as well in simulating the past 100 years of climate. Why?
It's because there are hundreds of thousands of "knobs" in the model — choices, really: should there be a cloud in this box? Should it be raining in this box? There are thresholds, based on the values in the corners of these boxes. As the box gets smaller, the values that we've tuned in the model to match the last 100 years aren't perfect anymore — and that's where problems show up.
I think the most fundamental problem left in the work I do is representing how models and reality are two different beasts. How do we find the projection of reality onto the model that most accurately reflects the differences between the two and how do we use that mathematical object to make our forecasts better?
DUPIGNY-GIROUX: I'm concerned about hurricane mitigation. How do you get people to understand that hurricanes are a force to be reckoned with and that patterns change over time?
In the Caribbean, hurricane tracks have been moving. For about seventy years of records, there were mostly northerly tracks moving across the Caribbean, but then in the last ten or fifteen years, they seem to have shifted further south.
That is an unanswered question: why has there been this shift in the latitudinal track that hurricanes are taking across the Caribbean?
And then, following up on that, how do you warn the populations in those regions about this change?
Coming from the Caribbean, we have a saying: "June too soon; July stand by; August come it must; September remember; October all over." Now, if things are changing so that the timeframe within which hurricanes are striking is extended, or the latitudinal boundary changes, how does that factor into what has become a cultural norm? How difficult is that going to be to dislodge that perception?
EMANUEL: I see there are two big problems.
On the physical side we're beginning to recognize that hurricanes may have a very important feedback on the climate itself through their effect on the ocean.
It's only recently that we've started to recognize that hurricanes' absence from climate models may be more serious than just "oh, isn't that too bad." Instead, the fact that they don't represent hurricanes may actually limit their ability to predict the overall climate.
On the political side, in the United States we have the worst conceivable set-up that will guarantee a succession of Katrinas regardless of whether the climate changes. The problem is the massive underwriting of hurricane risk. It just encourages more coastal development, more set-ups for Katrinas, and more ruinous effects on coastal ecology.
It's like if you got a big discount on your health insurance if you smoked more than a pack of cigarettes a day.
Have any of you ever been in a hurricane?
DUPIGNY-GIROUX: I only watch them from a distance!
EMANUEL: I do a lot of experimental work in hurricanes. It's not bad at all. I've had much worse flights on United Airlines. Seriously. It's not so bad to fly airplanes into hurricanes. It's been done for almost every sizeable Atlantic hurricane since the 1940s. I'm planning to start a hurricane safari operation when I retire, taking paying customers.
So people who don't have enough money to go out in space can take a trip into a hurricane?
DUPIGNY-GIROUX: Make sure it's a cat five storm!
DANFORTH: And who's going to underwrite your safari insurance?