Weather Forecasting
A policy statement of the American Meteorological Society as adopted by the Council on 13 January 1991

Weather forecasts and warnings are the most important services provided by the meteorological profession. Forecasts are used by government and industry to protect life and property and to improve the efficiency of operations, and by individuals to plan a wide range of daily activities. This summary of present-day weather forecasting capabilities is intended to provide general guidance to a broad constituency of users. The American Meteorological Society Statements on Flash Floods (Bull. Amer. Meteor. Soc., 66, p. 858), Hurricane Detection, Tracking, and Forecasting (Bull. Amer. Meteor. Soc., 67, p. 1508), and Tornado Forecasting and Warning (Bull. Amer. Meteor. Soc., 72, p. 1270) assess the current state-of-the-art pertaining to some of the most hazardous weather events that require warnings to protect people, commerce, and industry. Topics related to prediction of climate change are addressed in the Statement on Global Climate Change (Bull. Amer. Meteor. Soc., 72, p. 57).
As practiced by the professionally trained meteorologist, weather forecasting today is a highly developed skill that is grounded in scientific principle and method and that makes use of advanced technological tools. The notable improvement in forecast accuracy that has been achieved since the 1950s is a direct outgrowth of technological developments, basic and applied research, and the application of new knowledge and methods by weather forecasters. High-speed computers, meteorological satellites, and weather radars are tools that have played major roles in improving weather forecasts.
The most impressive gain in forecast accuracy in recent years has been in prediction for the 1 to 5 day range. A number of factors have contributed to the increase in accuracy. Foremost among these has been the further development of numerical prediction models, based on the laws of physics, that are able to forecast the formation and movement of the large high and low pressure systems that govern day-to-day weather changes in middle and high latitudes. These models have undergone steady improvement since their introduction more than a quarter century ago. The improvement has been made possible in large measure by the growth in the capacities of the computers that are required for carrying out the vast number of calculations involved in a numerical forecast.
Several other factors have contributed significantly to this increase in forecasting accuracy. One is the development of statistical methods for enhancing the scope and accuracy of model predictions. Another is the improved observational capability afforded by meteorological satellites. A third primary reason for the increase in accuracy is the continued improvement of the initial conditions prepared for the forecast models. Statistical methods allow a wider variety of meteorological elements to be predicted than do the models alone, and they tailor the geographically less precise model forecasts to specific locations. Satellites now provide the capability for nearly continuous viewing and remote sensing of the atmosphere on a global scale. The improvement in initial conditions is the result of an increased number of observations and better use of the observations in computational techniques.
The numerical prediction models that have proved successful in 1 to 5 day prediction have at present only limited, and as yet unproved, capabilities for making predictions for periods beyond 10 to 14 days. Forecast methods and tools for monthly outlooks have been a mixture of dynamical and statistical techniques. Recent improvements in numerical weather prediction models have led to a greater emphasis on the dynamical part of the forecast input, especially for the first 10 days of the month. For seasonal outlooks, no dynamical technique is used and the forecast is based entirely on a variety of statistical tools. At monthly and seasonal ranges, day-to-day weather changes are not predictable, either in theory or in practice, and forecasts are concerned instead with the probability that average temperatures and total precipitation for the forecast period, typically a month or a season, will be above or below normal. Improvements in statistical forecasting methods have resulted in some skill in monthly and seasonal prediction thus making the information potentially beneficial to certain users. The skill of both temperature and precipitation forecasts varies with season, location, and situation, however. In some regions, the seasonal forecasts remain no better than a forecast based on a continuation of the weather observed at the beginning of the season or a forecast based on average conditions.
The cause of the large circulation anomalies that produce persistent pattens of abnormal warmth and cold and wetness and dryness are not well-understood. There is, however, growing evidence that surface conditions (i.e., snow and ice cover, sea surface temperature, and soil moisture) play an important role in these anomalies. Numerical models now under development provide the basis for an improved understanding of long-range weather variations and for an enhanced ability to predict them.
For very short prediction periods (0 to 6 hours) principle interest centers on small-scale, short-lived, often violent phenomena, such as thunderstorms, tornadoes, as flash floods, and on localized nonviolent weather such as dense fog and freezing rain. Although a growing capability exists for mathematically modeling small-scale weather systems, practical application of the models in forecasting specific events is not yet feasible because of the difficulty and expense of observing the atmosphere in the required detail. An exception to this statement is the prediction of local wind systems, such as sea breezes and mountain winds, that are induced by topographical features. Numerical prediction of such systems is feasible and is presently under development.
In view of the difficulty in applying physical methods in the forecasting of small-scale weather, the emphasis at present is on detecting and tracking these features and determination by forecasters of their development and movement for short intervals into the future. Weather radar and geostationary satellites are particularly valuable tools for this purpose, and their capabilities and usefulness have been and are being steadily improved. Rapid communication of warnings and the establishment of adequate warning networks are also essential elements in very short- range prediction. Progress has been made in these areas too.
The tropical storm or hurricane is a phenomenon that is of special importance in short- range prediction. With the help of weather satellites these systems can now be detected and tracked over data-sparse ocean regions and estimates of their intensities can be made. Aircraft reconnaissance and ground radar permit increasingly accurate tracking and a reliable determination of intensity as the storms approach inhabited coastal areas.
Progress in forecasting the movement and changes in intensity of tropical cyclones has not kept pace with advances made in forecasting the behavior of extratropical cyclones, although predictive skill has begun to increase in recent years through the use of combined statistical and numerical prediction methods. A principle reason for the slow progress has been the lack of adequate data for use in numerical prediction models. Because of the limited application of these models, human judgement also weighs heavily in hurricane prediction.
The methods currently employed yield useful warnings of storm danger a day or two in advance. They are less successful in providing timely warnings of narrow within which major damage is likely to occur and within which special precautionary measures should be taken.
The usefulness of the present imperfect forecasts has been increased by application of sophisticated statistical methods to aid decision making when evaluating various preparedness actions. Additional information regarding current skill in tropical cyclone forecasting can be found in the AMS Statement on Hurricane Detection, Tracking, and Forecasting (Bull. Amer. Meteor. Soc., 67, p. 1508).

The foregoing overview offers a brief account of the present state of the forecasting practice and highlights some of the important accomplishments of recent years. It is now desirable to state more fully the current levels of accuracy and skill. The statements that follow pertain to land areas in temperate and high latitudes of the Northern Hemisphere. Skill levels are lower over the northern oceans and the Southern Hemisphere owing to the reduced numbers of observations in these areas. Forecasting skill can be determined objectively by comparing the accuracy of a given set of forecasts with the accuracy of a corresponding set produced by simple procedures, such as predicting that the weather will remain unchanged (persistence or assuming that the weather will correspond to average conditions (climatological forecast). Unless forecast accuracy exceeds levels achieved by these simple methods, predictive skill cannot be said to exist.

1) For the period 0 to 12 hours: The accuracy of weather forecasts in this time range depends very much on the specificity required in the forecast and varies with the weather situation. Forecasts of general weather conditions and trends in this time periods show considerable skill and utility. However, the spatial and temporal detail that can be included in the forecast decreases as the period increases. The behavior of small, short-lived, severe local storms is predictable only for periods of the order of several minutes to an hour. Recent observational studies of severe convective storms combined with numerical simulations of mesoscale systems (approximately 1 to 100 km in horizontal extent) have provided the forecaster with an increased ability to anticipate thunderstorm development and make very short-range forecasts of the evolution of the convective system. The behavior of larger features, such as squall lines, fronts, and organized areas of heavy precipitation, is often predictable for periods of up to 6 to 12 hours. Weather changes associated with large cyclonic storms are well forecast for this time range. Strong downslope winds, gorge winds, and other features induced by irregular terrain and surface inhomogeneities can often be predicted for periods up to several hours ahead or longer.
The ability to provide forecasts and warnings for flood events and tornadoes is discussed in the Statements on Flash Floods (Bull. Amer. Meteor. Soc., 72, p. 1270).
Major improvements in forecast accuracy in this time range are anticipated in the 1990s as a result of new discoveries and increasingly sophisticated data processing systems.
2) For the period 12 to 48 hour: Skillful predictions of the development and movement of large extratropical weather systems, and of the associated day-to-day variations in temperature, precipitation, cloudiness, and air quality, can be made throughout this period. During the decade of the 1980s, accuracy in numerical weather prediction continued to improve. For example, the 48-hour forecasts of sea level pressure in 1987 were as accurate as 24-hour forecasts in 1977. Useful predictions of tropical storm movement can also be made, although rapid changes in destructive potential are not well predicted. In addition, general areas within which severe storms and thunderstorms are likely to occur can often be specified up to 48 hours in advance. However, as indicated above, the exact times of occurrence and locations of individual local storms cannot be predicted at this range.
3) For the period 3 to 5 days: Large-scale circulation events such as major storms and cold waves usually can be anticipated 3 to 5 days in advance. Daily temperature forecasts decrease in skill, relative to climatological forecasts, from good at 3 days to fair by 5 days. Forecasts of precipitation occurrence show fair skill at 3 days and marginal skill at 5 days. The accuracy of numerical weather forecasts for this time period also has continued to improve. Five-day forecasts are now roughly as accurate as 3-day forecasts made a decade ago. A similar increase in accuracy has occurred in forecasts of mid-level atmospheric circulation patterns and in temperature predictions derived largely from numerical model forecasts.
4) For the period 6 to 10 days: Mean temperatures and precipitation for the period can be predicted with some skill. Temperature can generally be predicted with greater skill than precipitation. Daily maximum temperatures can be forecast with modest skill for the first two days of the period.
5) Monthly and seasonal forecasts: Slight skill exists in forecasting average temperatures and precipitation for the month or season. The skill in forecasting average temperature is greater than that in forecasting average precipitation. No verifiable skill exists in forecasting day-to-day weather changes a month or a season in advance.

The utility of forecasts can be increased by informing the user of the estimated or otherwise-determined probability of occurrence of an event as is done, for example, in daily precipitation forecasts and in some monthly and seasonal outlooks. Probability statements offer potential benefits to those engaged in weather-sensitive operations, if weather-related costs and losses can be evaluated.
Opportunities exist for increasing forecast accuracy and capabilities in all of these time ranges. It is known from theoretical studies that the limit for which useful forecasts of daily weather can be made is of the order of 10 to 14 days ahead. This theoretical limit considerably exceeds the present practical limit of 5 to 7 days. To close the gap between what has been achieved and what is achievable will require: 1) improvements in the global observing network through increased utilization of satellite, aircraft, and ocean buoy systems; 2) improvements in model physics through interpretation of datasets obtained in special field experiments, and through other theoretical approaches; and 3) improvements in the computational procedures employed in the models and in the speed and memory capacities of the computers. As far as very short-range prediction is concerned, there have been a number of recent technological developments which, if fully exploited, will lead to a substantial improvement in capabilities for forecasting and warning of severe weather events. These developments include: 1) Doppler radars that have the ability to detect, among other things, the parent mesoscyclones that spawn tornadoes, and the thunderstorm microbursts that have on occasion caused fatal aircraft accidents, and have the capacity to provide detailed rainfall information to help the forecaster issue more site-specific and timely flash flood warnings; 2) satellite imaging and sounding systems that provide almost continuous surveillance of the severe storm environment and detection and tracking of the storms themselves; 3) advanced ground-based sounding systems that yield wind, temperature, and humidity measurements of unprecedented temporal resolution; 4) interactive computers that make it possible for the forecaster to display, manipulate, and rapidly digest the great quantity and variety of data that severe storm forecasting entails; and 5) devices and systems for the rapid transmittal of data and the timely dissemination of warnings. Advances in prediction at very short ranges can also be anticipated as a result of recently completed, and proposed future, field investigations of severe storms and of recent progress in numerical modeling of intermediate or mesoscale weather systems, such as tropical storms, squall lines, and large thunderstorm complexes. A better understanding of mesoscale features embedded within large-scale circulations, such as winter storms, has been gained during recent years. This area of research, involving interactions among various scales of motion in the atmosphere, offers a significant opportunity for the advancement of forecasting capabilities.
Regarding long-range prediction, there are signs that some improvement in predicting monthly or seasonal averages will emerge from present research on large-scale, ocean- atmosphere interactions and related topics; from expanded, global observations of the ocean- atmosphere system; and from the application of suitable numerical prediction models. However, until predictive capabilities at these ranges can be better demonstrated theoretically, only modest expectations are warranted. Even small improvements should be of substantial economic benefit.  

Top of Document
AMS Home Page

© 1996 American Meteorological Society