Twisters in
the Crosshairs


getmedia/b88b33c5-dcb4-4c35-9d2f-984b2681c950/Twisters-in-the-Crosshairs_thumb.jpg.aspx?width=60&height=60&ext=.jpg

Even if they spoil your picnics every now and again, when it comes to rain and snow, storms and heat waves, the weathermen have usually seen it coming, days in advance.

Not so the tornado. As was evidenced this May in Oklahoma, if you get any warning at all, it's not likely to be more than a quarter of an hour's notice.

What makes such a forceful weather event so hard to predict? Part of it is the small size of twisters, whatever their power. Another part is the large size of data crunching that has to go on to forecast their appearance.

Weather balloons remain the best tracker for potential tornadoes. Image: NOAA

"Despite all these gee whiz satellites and technology, the observing that contributes the most to forecast and models is still from weather balloons," says Neil Jacobs, Chief Atmospheric Scientist for the Panasonic Avionics Corporation. Satellites aren't too good at measuring vertical data. Though they can read a license plate from outer space, they can't tell you how far that plate is from the ground (though you might be able to guess, if you've ever seen a car). Similarly, they can't tell if a cloud is 20,000 feet above ground or 50,000 feet.

Weather balloons pass through the part of the atmosphere that concerns us most, the troposphere, and they do a pretty good job of telling us what's going on, vertically and otherwise, every inch off the ground. The problem is that there are only 69 launching spots in the U.S. The balloons are sent up only twice a day and they only send their data to NOAA when they've reached their zenith and pop, when data from the start of the journey is already two hours old. "Back in the 60s that was smoking fast, but the technology hasn't changed," says Jacobs.

Tornado prediction won't make any leaps and bounds until there's an increase in the amount of data that drives atmospheric models.

What has changed are the models of the weather. But they're only as good as the data they process. "The equations in the models are theoretically perfect. If you want to make the weather models better, you've got to build the initial conditions that feed those equations. It's garbage in, garbage out."

To lighten the garbage load, and increase the load of data, Jacobs and his colleagues turned to those other vehicles that travel through that crucial first 20,000 feet of atmosphere: regional airplanes. They've attached sensors to some 400-odd aircraft.

The data from cruising altitude, largely horizontal, is not so useful, so they don't bother with planes meant for longer flights. But regional planes make half a dozen flights a day, and twice each flight they pass that lowest part of the atmosphere. The data from the front end of these smaller aircraft is relayed in less than 15 seconds.

So why not just put even more sensors on more planes, so our hatches can be battened with a little more readiness? Unfortunately, planes are used in such heavy rotation that they are rarely available for such modifications. It's only during their annual heavy inspection that Panasonic can pop in their sensors. But even if every plane in the world had the sensors in the nose, there's not yet enough computational power to make use of it.

Tornadoes, the smallest of big weather events, are "sub grid scale," says Jacobs. "To properly resolve an atmospheric feature, you need a minimum of three data points inside it. If you're talking about a thunderstorm cloud, maybe five kilometers across, if you're running a one-kilometer model, it will resolve that cloud. But if it's a funnel 200 meters wide, the model won't know it's there. There are models that can run down to five meters resolution that can resolve spinning funnel formations. But it will be a long time before they'll be operational—the computational resources aren't there."

The electric bill for Panasonic's computing is nearly $500,000 per month. They had to move their hardware to a liquid-cooled facility in Orlando to avoid tripping the circuit for the Raleigh-Durham area where the company is located.

But the fact that giving people in the plains more time to scurry into their basements is largely dependent on faster chips should not be a cause for despondence.

"Every time I turn around IBM has a faster machine," says Jacobs. "In terms of seeing a time when we can predict tornadoes—I'll see that in my lifetime for sure."

Michael Abrams is an independent writer.

There are models that can run down to five meters resolution, that can resolve spinning funnel formations. But it will be a long time before they’ll be operational.

Neil Jacobs, Panasonic Avionics Corporation

getmedia/b88b33c5-dcb4-4c35-9d2f-984b2681c950/Twisters-in-the-Crosshairs_thumb.jpg.aspx?width=60&height=60&ext=.jpg

July 2013

by Michael Abrams, ASME.org