Facebook Twitter



When a blizzard pounded the East Coast last year, spinning off tornadoes and snowstorms that killed more than 100 people, it evolved on a path and a timetable precisely forecast days in advance by meteorologists.

Satellite photos showed three squalls merge into a single superstorm in the Gulf of Mexico, and meteorologists knew immediately that an unusual bend in the jet stream would sweep it up the eastern seaboard and into an area of record low pressure over New England.On a Thursday, the National Weather Service warned that a "major winter storm" was coming, and Pennsylvania State University meteorologists tipped New York Times readers the next day that the storm "will produce weather of memorable proportions in the East this weekend."

The only people surprised by the blizzard were those who hadn't seen the forecasts - or chose to ignore them.

But if this storm amply demonstrated just how accurate weather forecasting can be, it also pointed out the inherent limitations of the system. Although forecasters were able to give a couple of days' warning of trouble, they did not anticipate this century's greatest storm until after it formed and was just starting its deadly rampage in the Gulf of Mexico.

Powerful computers and sophisticated computer models now routinely permit National Weather Service meteorologists to anticipate the movement of weather fronts and atmospheric pressure 36 hours in advance with 95 percent accuracy.

Local forecasters, in turn, can use that information to tell if the weather generally will be sunny, rainy, overcast or clear six or seven days into the future - and do it as reliably as the puny three-day "outlooks" of just a generation ago.

But even the best forecasters are not much better than their grandfathers at predicting the weather farther ahead - even next month. And if they forecast rain tomorrow, they still cannot say for sure if we will get a bothersome drizzle or disastrous downpour.

"If all you care about is whether to carry an umbrella to work tomorrow, they're very good at that," said Dean Churchill, assistant professor of meteorology at the University of Miami. "But if you're a commodities broker and you want to find out what crop yields might be in the Midwest grain belt three months from now, you're out of luck."

The reason is simple: complexity.

Weather is governed by well-understood physical laws. But those laws govern an almost unimaginably vast and complex set of variables, from the evaporation rate over Antarctica to the cloud cover over the local forecast office.

Currently, only the most obvious and easy-to-measure variables, such as the atmosphere's temperature, moisture content and density, are used in computer models at the National Meteorological Center in Camp Spring, Md.

Wayman Baker, who helps develop the models, said the simulations start with current global conditions. They then mathematically project how such phenomena as convection and condensation will move warm, moist air masses, cold, dry-air masses and other major weather factors in 15-minute increments. It repeats the same calculations hundreds of times until a desired 48-hour or 10-day model is complete.

The models, which forecast weather influences only in broad terms and are not yet useful in predicting the processes that can create a superstorm, are distributed electronically to regional weather offices.

While this process produces reasonably accurate forecasts for a few days, scientists said it is practically useless to look ahead a few weeks. This is because models can only consider what meteorologists can tell them. After several days, they said, the influence of subtle phenomena - such as mid-ocean sea-surface temperatures and the gravity-wave drag that slows wind flow over mountains - begin to compound one another.

These influences are not included in current computer models, because they are difficult to measure and would greatly complicate the calculations. Eventually such factors multiply to the point where they significantly distort the expected behavior of storms and other obvious weather compo-nents.

"Errors multiply, and those small errors you made early on - which don't really matter in short-term forecasts - get compounded and multiply to the point where you can't . . . do better than to guess based on the season," said Dan Cayan of the Scripps Institute at University of California, San Diego.

To get around this problem, some meteorologists are looking for new ways of making forecasts that might avoid the mathematical morass of current models.

One theory, by David Baumhefner of the National Center for Atmospheric Research in Boulder, Colo., is called ensemble forecasting. To factor out "observational errors," he assumes that field reports only approximate actual conditions. He makes a range of 10 to 40 different 30-day forecasts by plugging in slightly different values for each condition. He then looks for a consensus.

Others hope for help from the new science of chaos theory, which seeks to show that mathematical order underlies such apparently random phenomena as curling smoke, turbulent rivers or, well, weather.