Science / Medicine : A Global Calculation : Computers Work to Beat the Odds in Forecasting the Weather

In 1952, when John von Neumann was developing the modern computer, one of the first tasks he attempted was forecasting the weather.

The computer was the final link needed to begin a new era in forecasting that could make use of new methods of measuring atmospheric conditions with balloon-borne instruments developed during World War II and new mathematical theories of atmospheric physics that came into use at about the same time.

Ever since then it has been a matter of refining the process--gathering more data, developing more complex mathematical models and using ever more powerful computers to predict what it all means.

Of course, one way to predict the weather is simply to say that tomorrow will be a repeat of today. You don’t even have to look out the window. That forecasting method has an official name at the National Meteorological Center here--persistence--and it is reckoned to be accurate about half of the time.

Thus the nation’s top forecasters figure that they’ve got to be at least 60% accurate before they can claim to beat simple persistence.

How they do it involves considerably more than looking out the window, according to William Bonner, center director.

Difficult to Project

To visualize the complexity, imagine yourself standing at the beach, watching the surf break against the sand. A short way out a plastic bottle bobs in the waves. Will the bottle wash ashore? When? Where? All you can do is guess because you have no way of knowing or measuring or predicting all of the forces that are affecting the motion of that bottle.

Predicting the weather at any given location is akin to forecasting when and where the bottle will wash ashore. And the Earth’s atmosphere is much more complex than a small stretch of ocean. It is a sea of gases in constant movement, churned by 8the rotation of the planet, the heat of the day, the cool of the night, the seasonal angles of the sun, evaporation, condensation, the temperature differentials of ocean currents, plowed fields, glacial mountains, city pavement and tropical rain forests and a jillion or so other forces large and small.

Forecasting starts by taking a three-dimensional global snapshot of atmospheric conditions twice daily.

Meteorologists release more than 700 balloons carrying “radiosonde” measurement devices to radio back temperature, humidity and barometric pressure and allow tracking of wind speed and direction as the balloons ascend. About two-thirds of the balloons are sent up in the Northern Hemisphere, the rest in the Southern, at about the same time, midnight and noon Greenwich Mean Time (4 p.m. and 4 a.m. Pacific Standard Time.)

The resulting data are shared worldwide by government-operated forecasting services under auspices of the 160 nation members of the World Meteorological Organization, an arm of the United Nations. Here at the National Meteorological Center, other data also is added--interpretations of winds and temperatures from satellite photos and weather observations reported by ships at sea, ocean buoys, aircraft and at thousands of ground reporting stations.

That information is funneled into the center’s supercomputer housed in nearby Suitland, Md., a Cyber-205 made by Control Data Corp. and one of the most powerful machines available. It makes the millions of calculations needed to complete a numerical forecast model of the atmosphere.

There are several forecast models, differing by the geographic area they cover and the level of detail within that area. But all go about the task in a similar way--they surround the Earth with a mathematical three-dimensional grid and solve equations at specified points within that grid that predict the physics of the atmosphere based on the data available.

North America Most Detailed

The most detailed grid covers North America with calculation points 60 kilometers apart at 18 vertical slices of the atmosphere. The lowest resolution model covers the Northern Hemisphere in 16 vertical layers at grid intersections 360 kilometers apart.

Even on a supercomputer, the calculations required are so huge that the computer runs virtually 24 hours a day to keep up with the task.

The day’s runs actually begin at 8:30 p.m. EST here each night--an hour and a half after the midnight Greenwich Mean Time launch of radiosonde balloons--and continues more or less nonstop for the next 12 hours. Then the cycle starts over again with data from the noon GMT balloon launch.

First comes an early prediction of barometric pressure patterns, winds and temperatures for the next 48 hours to give forecasters their first look at the next two day’s weather. That is followed by a more detailed look at the United States along with less detailed studies of the North American and Northern Hemisphere weather.

Next there is a worldwide forecast model for aviation weather after sufficient data has been received from the Western Pacific and Southern Hemisphere balloons and, during hurricane season, a special hurricane tracking forecast computer run follows the aviation run.

Otherwise, the fourth computer run of the cycle is the “medium-range forecast,” which makes predictions out to 10 days and is started six hours after the balloons fly. It, too, is a global model and is used for the extended forecasts issued by local forecast offices.

Last run of the cycle is a repeat of the medium-range forecast model with the latest available data, but projected out only six hours into the future. The resulting data is used to test information received from the next balloon launch for errors and to fill in the gaps where no observations are available.

The end result of all this computing are more than 500 computer-generated maps each day that depict the pressure patterns, winds and temperatures all around the globe.

Once the computer finishes a map, human forecasters take over. They can interpret the maps and draw in the cold fronts and warm fronts and the jet stream--features the computer does not identify. And they can modify the patterns if their experience indicates that the computer is wrong.

Those forecast maps are distributed by computer to all the regional and local national weather service offices throughout the country and by private data networks to all of the private forecasting services, including WeatherData Inc., the firm that produces forecasts published by The Times. Thus all forecasters, public and private alike, have equal access to the raw data and computer modeling upon which final forecasts are based.

Sometimes, the computer is a better predictor than human forecasters.

Such was the case last November, for instance, when the center here was rendered nearly inaccessible by a record 15-inch snowfall that surprised the public and stranded thousands of motorists for most of the night on the Beltway. Subsequent analysis indicated that the computer models showed the possibility of five inches of snow or more but forecasters successively downgraded that, first here at the meteorological center and then at the regional National Weather Service forecast office in the same building. The public was told merely that there was a chance of snow showers.

Although the center has steadily improved the accuracy of its forecast models as increased computer power has allowed it to run increasingly complex and higher resolution forecast models, Bonner explained, it is still a long way from being able to pinpoint the effects of individual storm cells.

When you consider that the most detailed forecast model has grid points 80 kilometers apart while a storm cell may cover only a few square miles, you can understand why.

Bonner said the center has improved its accuracy about as much as it can with present equipment and is planning to install a computer about 10 times more powerful than its Cyber-205 by 1990. It would be the eighth new generation computer installed by the National Meteorological Center since the first was acquired in 1955. Each has brought about a 10-fold increase in power over its predecessor, which in turn has enabled a 5% to 10% increase in forecasting skill.

The added power and speed will allow the processing of more sophisticated forecast models, according to Francis J. Balint, deputy chief of the automation division. For every added day that a forecast is carried into the future, a corresponding increase in the number of mathematical calculations must be made. Since the current Cyber-205 works virtually full-time to compute the forecasts now utilized, its replacement with a faster computer means that forecasts further into the future could be done in the same amount of time, Balint explained. That suggests that a new computer would allow the forecast 7.5 days out to be as accurate as it is now for six days out, Balint said.

Two Weeks Theoretical Limit

Center director Bonner said the theoretical limit of the usefulness for numerical forecasting is about two weeks out. Beyond that too many new variables will have been introduced into the actual physics of the atmosphere for the models to predict.

Bonner said they have experimented with 30-day forecast models, but for the present and certainly for the coming decade, long-range forecasts will be statistical probabilities based on historical climatological data plus the forecaster’s experience.

Bonner said that once the center gets its next-generation supercomputer, he believes its forecasting will be as good as any in the world, but he acknowledged that the European Center for Medium-Range Forecasting in Redding, England, financed by a consortium of 17 nations, has a slightly better track record than his center based on some measures of forecast accuracy recognized by meteorologists.

He attributes that to the more narrow mission of the European Center to forecast in the three-day to 10-day range (compared to the 12-hour to 10-day forecasts at his center). And, he said, since its inception in the 1970s the European consortium has had more powerful computers than the National Meteorological Center.

Expects Other Tools

In addition to a faster computer, the center here looks forward to other tools in the coming decade that will help it make more accurate forecasts, according to Harlan Saylor, deputy director of the center. Doppler radar, which gives a direct measure of wind velocity precipitation at various altitudes--and thus storm movement--is one such tool that is now undergoing development at a test installation in Denver. Another tool is the better atmospheric temperature measurements that the next generation of weather satellites will bring.

Bonner says he expects in the coming decade to predict stormy conditions up to a week in advance with more and more detail becoming available for nearer-term weather. Unseasonable storms should be predictable one or two days in advance and reasonably accurate precipitation predictions should become available about six hours in advance of a storm.

1. Gathering the Information

Weather Balloons--More than 700 weather balloons are launched simultaneously twice daily around the world in member nations of the World Meteorological Organization. During ascent, radiosondes attached to the balloons send reports of air pressure, temperature and humidity at various altitudes. Ground tracking equipment measures wind direction and speed. At about 90,000 feet, the $60 balloon bursts and a parachute carries the radiosonde to Earth with a prepaid mailbag and instructions to the finder to return it for reconditioning and re-use up to six times.

Weather Satellites--They orbit the Earth about 22,000 feet high synchronized with the planet’s daily spin so they always remain above the same point on the surface. Satellite photos reveal cloud patterns and through interpretation also show air temperatures above 10,000 feet and wind patterns.

Aircraft and Ships--Commercial airliners and ships radio periodic reports of weather conditions at their locations. During hurricane season, the National Weather Service sends specially equipped planes to track tropical storms and hurricanes.

Ocean Buoys--Radio transmitters send data on air temperature, wind speed, air pressure, humidity and wave action.

Ground Stations--Airports and numerous other points around the world distribute hourly reports of weather conditions at their locations.

Not to scale.

2. Processing the Information

National Meteorological Center in Camp Springs, Md. operates a nearby computer center at Suitland, Md. into which all the weather observation data is funneled. A supercomputer analyzes the information. By calculating formulas based on atmospheric physics, the computer turns out maps predicting weather for up to the next 10 days. Forecasts are shared worldwide through the World Meteorological Organization. Individual forecasters then use that information to devise their own predictions and show the results in maps like that at right.

3. Producing the Forecast