More In This Article
The tornado that leveled much of Joplin, Mo., Sunday evening gave little warning. Although a watch had been in effect for a broader region for much of the day, some locals had as few as 20 minutes' notice that a tornado that would ultimately span as much as three quarters of a mile was about to touch down on top of them.
Major advances have been made in the science of issuing storm watches, thanks to improvements in computing power that allow meteorologists to run multiple computer simulations of the weather simultaneously. But the local conditions that drive thunderstorm—and subsequent tornado—formation still defy prediction.
Scientific American spoke with warning coordination meteorologist Greg Carbin of the National Weather Service Storm Prediction Center to better understand the challenges of tornado prediction.
[An edited transcript of the interview follows.]
What is the state of the science with regards to tornado prediction?
What we deal with at the Storm Prediction Center, we don't issue warnings, we issue watches. We have been able over the last few years to make some strides with respect to predictions on a larger scale. We can't give you hours of lead time prior to a single tornado but we can give you days of lead time prior to [favorable] conditions [for storm formation] becoming established. Something we saw back in April with the widespread tornadoes in Alabama and Mississippi, we had a large outlook five days in advance circling states from Arkansas and Louisiana up to Tennessee and painting those regions as having a greater risk for a severe storm developing. Our ability to predict the scenarios over several states has improved.
When you get down to the scale of a tornado event, you're dealing with the actual formation of a thunderstorm which may cover five to 10 kilometers in space. That is the elusive part. It's such a small scale. The physics at that scale, we don't understand all the physics of how tornados initiate or even how a thunderstorm forms and then a tornado forms from those.
How do we make these predictions?
In the case of the Joplin tornado there may have been upwards of 30 minutes' warning. There is a critical threshold passed in tornado forecasting: before and after a storm forms. Once an individual storm forms, there is a huge increase in our ability to predict given what we know about the ambient environment. When you're talking 13-, 15-, 20- or 30-minute lead times, that's only possible with a storm showing up on radar. Prior to the storm actually forming, it's not possible.
The National Weather Service system has 120 offices across the country with highly sophisticated Doppler radar and meteorologists watching the radar to interpret the danger of a storm. But prior to an actual thunderstorm forming, the uncertainty of a tornado threat is so much higher that we can only address over a larger area and a longer time span.
What technologies have allowed these advances in larger scale prediction?
The most revolutionary aspect of meteorology since radar is the ensemble forecast—models to predict the future state of the atmosphere several days in advance. Twenty years ago we had a handful of computer models up to five days out at a hemispheric scale. Maybe there were two or three models to choose from…Now, with ensembles and advances in computing technology, we can run many models—20 at a time and, for the first time as meteorologists, gain insight not only into predictions down the road but also confidence in predictions. We can begin to talk about [atmospheric] patterns that are more predictable than others. We can have more confidence in this forecast than another.
The atmosphere doesn't always have the same predictability. Some patterns are more predictable than others. Ensembles can give you a large spread or the solutions can be clustered. If you see clustering, you gain confidence in a particular scenario because the models all seem to agree. Now, more often than not, there's a lot of spread. There's still the dilemma of uncertainty. But you can begin to say that I'm more confident in this pattern than that one.
What improvements do we need to make to enable better local tornado prediction?
It all comes back to dealing with nonlinearity, which is what we see the atmosphere behave like a lot. The atmosphere is nonlinear—it's dynamic, it's dependent on initial conditions and, at some point, unpredictable…Temperatures are a variable in the atmosphere that we can measure quite accurately. We understand the processes involved. Tornadoes are a completely different animal. Their formation is tied to initial conditions—how that environment comes together, aspects of which we don't understand. The "butterfly effect" [seemingly insignificant perturbations leading to large-scale impacts] definitely applies in this environment.
We will never have the degree of resolution we need in atmospheric variables and measures in order to achieve some perfect forecast. It's just not possible…We might be able to get, say, an hour lead-time on a tornado. In that realm, prior to the storm actually forming, we're running what we call storm-scale models. We run them in ensembles. Even then we can't provide a deterministic forecast. We can give the probability that something could happen over some area.
We're already doing that with respect to watches and outlooks, say a 20 percent chance of a tornado over this 25-mile area…We're moving toward a storm-scale ensemble to better understand the potential for tornadic thunderstorms to form over a region smaller than a state but larger than a county.
What specifically do we not understand that makes such tornado predictions so difficult?
We don't understand positive and negative feedback mechanisms in the atmosphere…There are so many mild adjustments, slight adjustments that can make a huge difference in whether you end up getting the formation of storms. The sensitivity the atmosphere has to ingredients in the formation of tornados and magnifying that slight change in something we can't even observe can have a dramatic impact on the forecast.
What are the key ingredients?
The basic ingredients for severe weather are the amount of moisture content [and] instability, meaning a change in temperature [higher or lower in the atmosphere], like cooler air aloft than at the surface. Basically, the question is if you displace an air parcel at the surface will it continue to rise or [will it] cool and drop back down? A stable atmosphere [means] rising air will settle back down. Then there is lift, or ascent, usually on a large scale but also on a small scale. Then, for tornadoes or organized severe weather, there is wind shear—changing wind speeds and/or direction with height. It's not uncommon to see higher winds higher up, but the character of that wind speed and change with height plays a role in the ability of storms to persist, rotate and produce tornadoes and hail.
The old advice I heard growing up in Missouri is take cover if you see green skies or if the wind blows from two directions—true?
The storm is almost on you if you see green skies; predicting from there, that's what we call "dumb man meteorology." If the sky is green, there's a tremendous amount of moisture and rain or hail aloft. To hold that amount of rainwater and hail aloft requires a tremendously strong updraft. There's an intense thunderstorm nearby holding that massive amount of water and hail in the air. If there's a strong updraft, you're not far from a strong downdraft.
A tornado, if it's going to occur, occurs on the interface of the updraft and downdraft, the rapidly rising air and the rapidly sinking air…The cool air associated with the descending air interfaces with the rising air in front—that's a zone of potential spin in the atmosphere. You take that potential spin which exists in the horizontal plane and you pull it vertical—it's like the classic angular momentum of a figure skater. By pulling rapidly spinning air upward and together, it will spin even faster. That's the dynamic in which a tornado will form.
Are we seeing the formation of bigger tornadoes, stronger storms?
If we are, I wouldn't be able to tell you. The long-term record is neither long enough or of high enough quality for us to make a determination about that. It only really goes back to 1950, or maybe we can pull together data into the 1800s. But as you go back in time there are fewer tornadoes reported because there were fewer people around. The quality of the long-term record is not very good and not long enough for us to say.
Remember, these are extreme, rare events. The return frequency is such a length of time that you need a very long record to be able to say there are more intense or violent tornadoes than before.
To give you an example from the short-term record, just the last two months, [we had] an April with a very large number of tornadoes. If it was busy in April, it might stay busy in May since that's the most active month for tornadoes. [But] we had the opposite. We entered the first three weeks of May with hardly any tornadoes. We can say that there is no correlation between activity in April and what happens in May. We can't use that as a forecasting tool. That's just the nature of this game and points to the underlying nonlinearity of meteorology that we talked about with regard to tornado forecasting.
Are tornadoes correlated with any larger atmospheric pattern, like La Nina?
I just put a chart on the Web site yesterday, the number of daily tornadoes back to 1950.…If you look at El Nino and La Nina, there's not a very high correlation, though a La Nina pattern for March and April may favor more severe weather activity. Also in neutral years, when we're transitioning between La Nina and El Nino, we're in a regime that would support above normal storm activity…El Nino tends to be hostile to severe weather…It's not as easy or clean as I wish it was.