Select your localized edition:

Close ×

More Ways to Connect

Discover one of our 28 local entrepreneurial communities »

Be the first to know as we launch in new countries and markets around the globe.

Interested in bringing MIT Technology Review to your local market?

MIT Technology ReviewMIT Technology Review - logo

 

Unsupported browser: Your browser does not meet modern web standards. See how it scores »

{ action.text }

Sensing Rain

As computers become faster and cheaper and weather observation instruments improve, researchers promise even higher-resolution forecasts. In two or three years, some predict, resolutions will drop down to one kilometer. Getting there, however, will be an uphill battle. The first problem is the huge increase in computing power necessary to account for the additional variables of weather on such a small scale. “The topography of the local terrain, the presence of bodies of water, vegetation, cloud formation-all this has to be taken into account,” says Joel Myers, founder and president of AccuWeather. In fact, going down to one kilometer from four kilometers requires about 16 times the number-crunching muscle. 

Despite the heavy computational price, Lloyd Treinish and his colleagues at IBM’s Yorktown Heights, NY, research facility are working on one-kilometer-resolution weather forecasts. The so-called Deep Thunder project is part of a larger IBM effort known as Deep Computing, which is concerned with analyzing large amounts of data and solving complex computational problems. Like many companies working in the area of high-resolution weather forecasting, IBM foresees business opportunities in providing better prediction models to weather-sensitive companies. Treinish and his team have modified a standard regional model, developed at Colorado State University, and adapted it to the terrain, wind flow, and ocean-driven moisture patterns of the New York area. They use data received from the National Weather Service and then verify their forecasts using nearby weather stations-even one installed in Fishkill, NY. Later this year, IBM will install five more stations at its facilities in southeastern New York to help further fine-tune the models. Deep Thunder generates an updated 24-hour forecast two to four times a day, which requires almost two hours of computer time per forecast.

The modified model has proven its worth several times-as when it recognized the severity of a February 2003 blizzard some nine hours before the Weather Service. Treinish attributes the model’s accuracy to its fine resolution, which in addition to providing a more detailed look at local weather often leads to better forecasts for an entire region. “By getting at the physics behind smaller-scale weather events, you can get a better picture of what’s going on with larger-scale events like storms,” he says.

Treinish even talks of getting down to 500-meter resolution in the future, though he points out it would require modeling wind turbulence around New York City’s tall buildings-and it would demand about 100 times as much number crunching. “But that’s not an unheard-of increase in computing power, if you’re willing to wait a few years,” he says. 

Such accurate, high-resolution weather forecasts depend on models that are regularly recalibrated by comparing their predictions against actual observed weather. But trying to tell whether or not forecasts are improving is harder than it sounds, because the differences can be subtle and complicated: perhaps wind speed predictions are getting a bit more accurate, while temperature predictions are getting a bit worse. David Stensrud of the National Oceanic and Atmospheric Administration’s Severe Storm Laboratory in Norman, OK, notes that improving the models is slow going. “These models are so complicated that if you can correct for one problem you can easily cause another one,” Stensrud says. “You have to run the new model over a large number of cases to check it, and that makes it a huge, labor-intensive effort.”

And of course, a forecasting model is only as good as the instruments feeding it data. Eventually, meteorologists may be able to access finely detailed weather data from vast networks of sensors spaced just tens of meters apart over many parts of the country. A group led by Deborah Estrin, a computer scientist and director of the Center for Embedded Networked Sensing at the University of California, Los Angeles, is already embedding wireless sensor networks designed to monitor microclimate data-including, eventually, carbon dioxide levels-around small patches of trees and plants. “We want to explore the relationship between monitoring weather on a regional scale and on a microscale,” she says. 

Gathering such specific data may be in our future, but is it practical? “Putting out forecasts at the level of city blocks definitely makes it seem as if the forecasts are more precise,” says Craig Edwards, chief meteorologist with the Minneapolis office of the National Weather Service. “But the forecasts for one block would probably be the same as for other blocks.” There’s also a trade-off between resolution and how far into the future a model can make accurate predictions, says Young, because of how quickly small-scale weather phenomena change. “Today’s high-resolution forecasts are useful for a day or so,” he says. “We’re rapidly heading to resolutions that won’t buy you anything beyond six hours.”

Regardless of the new technology’s utility, meteorologists will be able to show off more of it in their forecasts as the public gets comfortable with weather jargon and maps. “The weather IQ of the public has increased tremendously over the last 10 years,” says AccuWeather’s Myers. “Instead of saying there’s a chance of rain today, we could say there’s a 20 percent chance of rain between 10:00 a.m. and noon, a 40 percent chance between noon and 2:00 p.m., and a 20 percent chance after that. That’s the sort of information you could use to schedule your golf game.”

Others in High-Resolution Forecasting
COMPANY GRID RESOLUTION SERVICE
Meteo Consult
(Wageningen,
Netherlands)
Site-specific, based on location of weather stations Weather maps and satellite and radar images delivered to Web-enabled mobile phones
MyWeather
(Madison, WI)
110 meters Text message forecasts sent to alphanumeric pagers and mobile phones
AWS Convergence
Technologies
(Gaithersburg, MD)
Site-specific, based on location of proprietary weather stations Web-based application that provides forecasts for customers in the energy industry
Weather Services
International

(Billerica, MA)
Eight to 10 kilometers; point-specific forecasts Weather maps and satellite and radar images delivered to people working in the media, energy, marine, and aviation industries (pilots can receive data in-flight on personal digital assistants)

0 comments about this story. Start the discussion »

Tagged: Computing

Reprints and Permissions | Send feedback to the editor

From the Archives

Close

Introducing MIT Technology Review Insider.

Already a Magazine subscriber?

You're automatically an Insider. It's easy to activate or upgrade your account.

Activate Your Account

Become an Insider

It's the new way to subscribe. Get even more of the tech news, research, and discoveries you crave.

Sign Up

Learn More

Find out why MIT Technology Review Insider is for you and explore your options.

Show Me