Introduction to The Forecast: Better Weather Prediction Ahead
Most of the time, local weather forecasts are pretty close to the mark. If the month is July and the forecast is for a high of 35 degrees C (95 degrees F), you can probably leave your jacket at home. And if it's March and the chance of rain is pegged at "near 100 percent," you would be smart to carry an umbrella. As we all know, however, weather predictions do not always match up with the weather that actually occurs. Perhaps you have been planning a weekend outdoor party, based on a forecast of sunny skies and warm temperatures. Then Saturday morning dawns overcast and chilly.
Because forecasts are usually fairly reliable, the ones that err tend to be the ones we remember, and they can leave us with the impression that forecasts today are no better than they ever were. In fact, meteorology (the study of the atmosphere and weather phenomena) has made steady progress since the mid-1900's. When averaged across the country, three- to four-day forecasts issued by the United States National Weather Service (NWS) in 2002 were as accurate as the two-day forecasts issued in the 1980's.
Much of the improvement in weather forecasting has come about because of the development of forecasting software, computer programs that enable supercomputers to peer into the future more and more accurately. This advanced software includes computer models (simulations) that show how a particular weather system is likely to act. The global network of weather observations used to provide data for these models has also improved. In the United States, for example, the NWS in 1999 completed a 12-year, $4.5 billion modernization program. The project included the purchase of new forecasting equipment and training for staff members to learn how to use it.
Nevertheless, forecasting remains an imperfect science. The longest-range forecasts—those covering about 14 days—are only a little better than climatology, a description of the average weather conditions in an area over an extended period of time. Even forecasts for a day or two into the future can sometimes be dramatically off the mark, when local weather features develop unexpectedly. The good news is that improvements in software and observing tools continue to be made, helping forecasters to spot local and regional weather developments and track their progress in more detail than ever. Longer-range forecasts are improving also, as meteorologists develop increasingly complex models and manipulate them with ever more powerful supercomputers.
Exactly what is this ever-changing phenomenon we call weather? Simply stated, weather is what's going on in the atmosphere at a given place on Earth at a particular time. Although the atmosphere has many different layers, nearly all weather forms in only the bottommost one, called the troposphere, which extends from the ground up to an altitude of about 16 kilometers (10 miles).
How Weather Forms
The primary cause of all weather phenomena is the sun. Days are warmer than nights because as the Earth spins, the sun heats one side while the other side is in darkness. The Earth is also tilted slightly, and that produces seasons. As the Earth rotates around the sun, the Earth's tilt on its axis causes different parts of the planet to be exposed to the sun for longer periods of time and at greater intensity. In June, the tilt points toward the sun, which produces summer in the Northern Hemisphere. In December, the tilt points away from the sun, so the Southern Hemisphere experiences summer. On average, the equator (an imaginary line that circles the middle of the planet) gets more intense sunlight than any other region, because it is facing the sun most directly. As the air over and near the equator is heated, it expands and rises. Cooler air from the polar regions replaces it and is heated in turn.
Such changes in air temperature also cause changes in air pressure (the weight of a column of air). In areas where warm air is rising, the pressure of the atmosphere upon the Earth is low. Colder, denser air exerts a higher pressure. Large bodies of warm and cold air are called air masses, and their movement around the Earth contributes to changes in the weather. This movement is called wind.
When two air masses of different temperatures and densities meet, they do not usually mix. They remain divided by a boundary called a front. If cold air is pushing warmer air forward, the boundary is called a cold front. If cold air is retreating and being replaced by warmer air, then the transition zone is known as a warm front. Where fronts intersect, the warm air is forced upward because it is less dense. As it rises, it cools. When it cools below a certain point, water vapor in the air condenses into liquid water and forms clouds. If the water drops become large enough, they fall as rain, snow, or other forms of precipitation.
Fronts are pushed along the Earth's surface by fast-moving winds called jet streams, most of which are located in the troposphere. Like ribbons of air, jet streams wrap around the globe from west to east. They also bend north or south like a meandering river.
Weather Features of Varying Sizes
As fronts and air masses move, they produce weather features of varying size and duration. Forecasters refer to large-scale weather features as synoptic, a term that comes from a Greek word meaning “a general view.” Synoptic features can span more than 1,600 kilometers (1,000 miles), covering sizable portions of a continent or an ocean. Such features, which include summer heat waves and winter cold spells, can last as long as a week or more in a particular location. Although synoptic features dictate the overall pattern of weather in an extended area, many other factors come into play in generating local weather conditions within that area. Mesoscale features dominate regions about the size of a county or a small state. “Meso” means “middle” in Latin. Mesoscale features are those between the synoptic scale and the local scale and are about 16 to 160 kilometers (10 to 100 miles) across. They include thunderstorms, blizzards, hurricanes, and other dramatic weather phenomena that generally last no more than a few hours.
The smallest weather features are local, or microscale, features and are less than 16 kilometers (10 miles) across. Mountains and large bodies of water can often generate microscale features that occur time and again. For instance, a strong wind called a katabatic or drainage wind regularly flows down many mountain canyons at night as the air at the top of the mountain cools. Other microscale features, such as tornadoes and snow squalls, are one-time-only events triggered by fast-moving weather systems and often lasting less than an hour.
The History of Forecasting
Our understanding of how weather works, and our ability to predict it, is a fairly recent development. Several of the basic instruments used to measure atmospheric conditions, including the first modern thermometer and the barometer for measuring air pressure, were invented in the 1600's. Weather observing became a popular pastime in the 1700's. The third president of the United States, Thomas Jefferson, kept records of daily weather conditions for almost 50 years, beginning in 1776.
When the telegraph was invented in the 1840's, meteorologists were able to communicate with one another quickly over great distances. They could transmit temperature and barometric pressure readings from their area and, by compiling information from different regions, develop weather maps. These maps used standard symbols to represent such features as cloud cover and types of precipitation and showed each day's major weather features across a wide area. By studying a number of maps compiled over a length of time, meteorologists learned to forecast weather early enough to warn of storms or predict favorable winds. The forecasts quickly became in great demand, especially by mariners.
It was not until World War I (1914–1918), however, that a small group of scientists in Bergen, Norway—inspired by the battle lines of the war—developed the concept of fronts. The leaders of the group, known as the Bergen School, were a father-and-son team of physicists, Vilhelm F. Bjerknes and Jacob A. Bjerknes. In a series of studies published during the early 1920's, the Bergen School explained how fronts move and how they affect the weather.
In 1922, a British physicist, Lewis Fry Richardson, became one of the first scientists who attempted to forecast the weather by creating a mathematical description of the atmosphere. Richardson developed a series of equations based on physical laws and on weather features that were starting to be regularly measured at the Earth's surface and aloft, including temperature, air pressure, humidity, and wind speed. Richardson believed that if one had precise and complete information about the present state of the weather, the future of the atmosphere could be predicted accurately. Even though such a level of data gathering could not be achieved in Richardson's time, his equations pointed the way toward much better forecasts. What was lacking was a way to handle the massive amount of calculation needed.
The computing power that would make Richardson's vision a reality began to emerge in 1950. That year, Hungarian-born mathematician John von Neumann of Princeton University in Princeton, New Jersey, and his staff developed the first weather-predicting software. They tested the program on ENIAC (Electronic Numerical Integrator And Computer), the first general-purpose electronic digital computer. Although it took ENIAC 24 hours to complete its calculations, the computer managed to provide a reasonably accurate forecast of where a few features of the upper atmosphere over North America would be on the following day. Meteorologists quickly embraced the idea of mathematical weather prediction—using computers to create models of weather systems.
In the early 1960's, the work of Edward N. Lorenz, a meteorologist at the Massachusetts Institute of Technology in Cambridge, put a slight damper on such enthusiasm. Lorenz discovered through computer simulations that there may be a natural limit to the accuracy of weather forecasting. An old folk saying proclaims, “For want of a nail, the shoe was lost; for want of the shoe, the horse was lost” until eventually, “for want of a battle, the kingdom was lost.” Similarly, tiny shifts in local weather conditions can cause ever-larger effects hours or days later. In 1972, Lorenz developed the concept of “the butterfly effect” to describe the problem, proposing—with some exaggeration to make his point—that a butterfly flapping its wings in Brazil might produce a tornado in Texas many days later. Of course, nobody could ever trace such a cause-and-effect pattern. The point was that small, unmeasurable aspects of weather can lead to powerful consequences over time.
Collecting Weather Data
In spite of Lorenz's conclusion, meteorologists continued to improve computer weather-forecasting models. They also sought ways to collect more extensive and precise data. For many years, human-staffed weather stations around the world had compiled hourly weather reports that were entered into the global network of atmospheric information. The reports included measurements of the type and amount of precipitation, visibility, air pressure, wind strength and direction, humidity, air temperature, cloud cover, and cloud types. Beginning about 1990, many of these stations were replaced by automated ones. By 2002, the United States had about 1,000 automated weather sites, reporting conditions minute by minute.
Meteorologists have also improved their monitoring of the upper levels of the atmosphere, including the jet streams, with weather balloons, satellites, radar stations, and other tools. Weather balloons, for example, carry aloft radiosondes, packages of electronic instruments connected to a radio transmitter that sends data to a receiver on the ground. Radiosonde-equipped weather balloons are released into the atmosphere twice a day at more than 900 sites around the globe to measure air temperature, humidity, and pressure at altitudes up to 30 kilometers (19 miles).
Although weather balloons have been used to gather data since the 1930's, modern balloons are considerably more advanced than earlier ones. The instruments are more compact and lightweight, and digital technology allows for more frequent high-precision sampling of the atmosphere along the balloon's route. As scientists track a balloon's location using the Global Positioning System (GPS)—a network of Earth-orbiting satellites that beam continuous radio signals to the ground—they can also deduce the speed and direction of winds at various heights along the balloon's path.
The Role of Satellites and Radar
Because roughly 70 percent of the Earth is covered by oceans, meteorologists also rely heavily on reports from weather stations aboard ships, buoys, and aircraft. Overall, though, the data coverage at sea is very sparse compared with that from the ground-based network. Weather satellites help to fill this gap. The first U.S. weather satellite was TIROS (Television and Infra-Red Observation Satellite), launched in April 1960. TIROS carried two miniature black-and-white television cameras that provided photographs of cloud cover as the satellite circled the Earth once every 99 minutes. As of 2002, the United States, Japan, France, India, Russia, and the European Space Agency were using a total of about a dozen weather satellites for daily monitoring.
There are two main types of weather satellites: geostationary and polar-orbiting. A geostationary satellite is positioned at a height of 35,900 kilometers (22,300 miles) above the equator, an altitude at which it completes one orbit in the same amount of time that the Earth rotates once on its axis—24 hours. The satellite thus seems to remain above the same spot on the equator. Because of its high altitude, a geostationary satellite can monitor the weather over about one-third of the globe. On the other hand, the great height at which a geostationary satellite orbits limits the amount of detail it can provide about weather features. Nevertheless, geostationary satellites survey vast areas of ocean from which meteorologists would not otherwise receive data.
A polar-orbiting satellite circles the globe at a much lower altitude, 800 kilometers (500 miles), making a complete circuit over the poles about every 100 minutes. Because the Earth rotates beneath a polar-orbitting satellite, the satellite crosses a different part of the planet with each pass, allowing it to cover the entire globe about 14 times each day.
Cameras aboard weather satellites take photographs of the parts of the Earth that are in sunlight. These pictures capture cloud patterns, some of which are related to storms. Many satellites also carry instruments that can monitor infrared radiation (energy from just beyond the red end of the visible-light spectrum) emitted by clouds and water vapor. Infrared radiation can be detected 24 hours a day, so infrared images make it possible to track storms even when they are shrouded by night.
For certain tasks, such as tracking precipitation over large areas, radar works better than weather balloons or satellites. A number of ground-based radar stations are scattered throughout the world. Each weather radar contains a transmitter that emits short, frequent pulses of micro-waves (very short radio waves). Some of the pulses bounce off raindrops or hailstones before returning to their source. By measuring the strength of the reflected signal and how long it took to return, meteorologists can determine how intense and how far away the precipitation is.
A system of weather radar was first installed across the United States in the 1950's. Beginning in the 1980's, the original system was replaced with Doppler radar, which gets its name from a phenomenon known as the Doppler effect or Doppler shift. Microwaves returning from a distant object that is moving have a slightly different frequency than the original signal. If the object is moving toward the radar transmitter, the microwaves will be closer together, or higher in frequency. If the object is moving away, the waves will be farther apart, or lower in frequency.
Since precipitation is blown by winds, Doppler radar signals bouncing off raindrops or other precipitation can be used to infer wind speed and thus can determine the intensity as well as the location of tornadoes, downbursts, and other destructive weather phenomena. By 1999, when the NWS completed its Doppler network, the average lead time for tornado warnings had doubled. On average, forecasters were able to warn people of impending tornadoes more than 10 minutes in advance, enough time for most people to seek shelter.
Using Computers to Prepare A Forecast
All of the weather data gathered by ground stations, balloons, and satellites must be organized in order to produce a forecast. In the past, the primary method of preparing a forecast was to draw a weather map by hand. Although hand-drawn weather maps are still used by some meteorologists, they have largely been replaced as the backbone of modern forecasting by maps generated by computers as parts of models. Meteorologists have developed several models, each with its own strengths and weaknesses, depending on the way it simulates the atmosphere.
Two to four times a day, meteorologists launch their chosen model, using as a starting point the most recent weather data received from all sources. The model then simulates the movement of the atmosphere forward in time—usually in 5- to 15-minute intervals—for a number of different points in a geographical area.
Each point represents a block of air in the computer model's atmos-phere. The points are spread out in a three-dimensional mesh or grid. The grid can cover a small region or an area as large as a hemisphere or even the entire globe. The points are separated horizontally by distances of 30 to 100 kilometers (18 to 60 miles) and vertically by 20 to 2,000 meters (66 to 6,600 feet). Within several hours, having performed trillions of mathematical computations, a model can produce a simulation of the weather for up to 15 days into the future.
Because of the problem recognized by Lorenz—that tiny, unnoticed changes in local conditions can dramatically affect weather conditions days later and hundreds of miles away—the accuracy of computer models decreases as the forecast period increases. However, meteorologists have developed some clever ways to get around the “butterfly effect.” One of these is known as ensemble modeling. With a powerful enough computer, forecasters can operate not just one version of a model, but an ensemble, or group, of 10, 20, or more. Each version of a model is identical but is programmed with slightly different starting conditions.
By randomly changing the initial conditions in each variant of the model being used in an ensemble and then comparing the results, forecasters can get a better sense of how predictable the atmosphere is at the starting point. If all of the ensemble members produce similar results, meteorologists can be reasonably confident that they will be able to predict changes in the weather accurately. If the ensemble members produce widely differing results, meteorologists know that the atmosphere on that particular day is less predictable.
Maps called spaghetti plots allow meteorologists to visualize the agreement or disagreement within ensembles. Spaghetti plots show the results of a number of ensemble forecasts on a single map depicting a particular weather feature. When the lines depicting the feature are packed closely together, the forecaster understands that the various versions of the model agree on the location or movement of that feature. When the lines are tangled like spaghetti, the map indicates that there is little agreement between the different versions of the model. In this way, meteorologist James Goerss of the Naval Research Laboratory in Monterey, California, and his colleagues demonstrated in 2000 that an ensemble of three different hurricane forecasts can improve the prediction of a hurricane's track by at least 20 percent over a single model.
Better and Better Models
In spite of the many improvements and innovations in weather forecasting, in the early 2000's researchers were working on attaining even greater accuracy through even better data gathering and computer modeling. In particular, researchers have been striving to improve the starting points of short-range computer models. To further this goal, many states have installed mesonets, collections of weather stations 10 to 20 times more dense than the usual observing network. While a typical standard observing network has stations separated by roughly 160 kilometers (100 miles), in a mesonet the stations might be separated by only 32 to 48 kilometers (20 to 30 miles). Mesonets are helping forecasters to spot mesoscale features that can influence local weather. The Oklahoma Mesonet, for example, has revealed that large belts of winter wheat in that state humidify the air above them during the growing season.
Another factor that has enhanced weather observations is a network of more than 30 wind profilers and acoustic sounders that has been tested across the central United States. The two instruments work together to give a more complete picture of conditions aloft. Wind profilers and acoustic sounders both point upward. Wind profilers send microwaves into the atmosphere in three to five different directions. When the echoes bounce off small variations in air density and return, the profilers compare the Doppler shifts from each of the directions to gauge the wind speed and direction. Acoustic sounders measure air temperatures at various altitudes every few minutes. The sounders broadcast sound waves into the atmosphere and track them with radar. Because the speed of sound varies with air temperature, the profilers can determine the temperature.
A new innovation to improve computer modeling is called nested modeling. The smaller the distance between two grid points in a model, the greater the resolution and the more detailed the information. However, since the atmosphere occupies three dimensions in space, it takes an eightfold increase in computer power simply to cut the distance between model points in half and another doubling in power to cut the time step in half. Computer time is often too expensive to run the highest-resolution model over a large area. When a potentially dangerous weather feature—such as a hurricane in the Atlantic Ocean—is developing and forecasters want to know how it will affect a particular county in Florida, meteorologists can now link, or nest, a smaller, high-resolution model within a larger-area, low-resolution model. In this way, the model can simulate short-term weather across the United States in low resolution but at a higher resolution for the county in Florida and still be affordable.
Improving the Forecast
In addition to improving the accuracy of weather forecasts, meteorologists are working on enhancements to the presentation of the forecast itself. The shortest-term forecasts are sometimes called nowcasts, because they predict the behavior of local weather features for the next several hours. In the early 2000's, some personal weather services, such as those offered on the Internet and through pagers and cellular phones, provided hour-by-hour forecasts. Private firms produce nowcasts by using general NWS data and then updating it for specialized audiences more frequently than the NWS can afford to.
Most people rely on the one- to two-day outlook to plan everyday activities. Computer power is helping the NWS, as well as national weather services in other countries of the world, to make these outlooks more detailed than ever. The Interactive Forecast Preparation System, implemented throughout the United States in the early 2000's, takes the output from computer models and translates it into everyday language for presentation on television or in newspapers. A meteorologist reviews the automatic translation, watching for possible errors and adjusting for local features, such as a high mountain range or a lake, that were too small to be included in the computer model.
Many private companies, such as AccuWeather, Inc., of State College, Pennsylvania, issue their own forecasts and sell them to radio and television stations or to newspapers. Like the forecasts from the NWS, these outlooks draw from large-scale models—either the private company's own or those of the NWS—and add the forecasters' expertise. Such companies may tailor their forecasts to specialized clients, such as windsurfers, farmers, shippers, or other people who may need more detailed information than the NWS can provide.
Currently few weather forecasts are very accurate beyond about 8 to 10 days, though longer-term forecasts have become common on television and on the World Wide Web since 2000. The best of the computer prediction models of the early 2000's spot many of the large-scale shifts in jet-stream behavior with some accuracy as many as 10 days before they happen. However, suspecting there will be a cold front in some part of the country a week from now is not the same thing as providing a precise temperature forecast for next Saturday in a particular part of a city. A more useful type of long-range outlook may be the kinds of regional maps the NWS produces for periods of 6 to 10 days and 8 to 14 days. These maps show the probability that temperatures and precipitation will be above, below, or near normal for the regions outlined. Ensemble modeling is expected to play an increasing role in creating such outlooks. Such techniques also hold promise for predicting climate variations for a season or even for several years.
Weather will always bring us surprises—some pleasant, some not so nice—and we are unlikely to see perfect accuracy in weather prediction in our lifetimes. But it will certainly get increasingly better. While a forecaster might now say that there is a 70 percent chance that the coming weekend will be warm and sunny, a future forecaster, using the same data, might be able to predict beautiful weather 10 days in advance with 90 percent accuracy. Even with that degree of accuracy, however, some uncertainty would remain. Still, having great picnic weather 9 times out of 10 would be good enough for most people.