An Information Statement of the American Meteorological Society
(Adopted by AMS Council on 25 March 2015)
This Information Statement describes the current state of the science of weather analysis and forecasting from short-term severe weather events to monthly and seasonal forecasts.
The nation’s Weather and Climate Enterprise is typically grouped into three sectors: government agencies, academic institutions, and the private sector. These three sectors play vital roles in providing products and services to the user community, which includes the general public as well as weather-sensitive government agencies (e.g., the military, Homeland Security, Environmental Protection Agency, numerous state and local offices, etc.) and private industries (e.g., energy, agriculture, transportation, etc.).
Forecasters are tasked with synthesizing available observations from multiple platforms (including surface observations, weather balloons, radar, and satellite), numerical guidance from computer model forecasts, scientific theory, and experience-based intuition to arrive at a forecast. The forecast process is often collaborative, with teams of meteorologists routinely integrating new information into the forecast as the event approaches. Increasingly, forecasters are also responsible for effectively communicating the forecast and its anticipated impacts upon life and property to various stakeholders, including the public.
The impacts of meteorological phenomena upon both life and property, whether short-lived or long-lasting in nature, are substantial:
- Nearly 90% of the emergencies declared by the Federal Emergency Management Agency are weather-related.
- More than 7000 road fatalities per year can be directly or indirectly attributed to weather.
- Approximately 70% of air traffic delays are caused by weather, at a cost of about $6 billion per year.
- Heat waves kill an average of about 175 people each year in the U.S.
- U.S. utilities save more than $150 million per year using 24-hour temperature forecasts to meet electricity demands most efficiently.
- Reducing the length of coastline under hurricane warnings involving the need for evacuations saves up to $1 million per coastal mile in evacuation and other preparedness costs.
How are forecasts made?
Analyses of Weather Data
Forecasting the weather begins by continuously observing the state of the atmosphere, the ocean, and land surface. The World Meteorological Organization provides the framework for an evolving worldwide suite of observing systems, such as satellites, radars, and surface weather observations that aid in monitoring these conditions. Observations from private citizens, particularly of precipitation type and severe weather, are increasingly available through social media, platforms such as the mobile Precipitation Identification Near the Ground (mPING) initiative, and through organized efforts such as the National Weather Service’s Cooperative Observer program and the Community Collaborative Rain, Hail and Snow (CoCoRaHS) network. Although major research challenges remain, scientists have made considerable progress in developing mathematical techniques to integrate these observations into snapshots of the land surface and atmospheric state at any given time. These analyses serve as the foundation for weather prediction on scales from individual clouds to regional severe weather events and global patterns. Analyses of current and past weather support many diverse environmental applications, including fundamental scientific investigations of the climate system.
For time scales on the order of a few minutes to a few hours, forecasters rely heavily on an extrapolation of current weather trends. Such forecasts are often called “nowcasts.” For instance, radar animations may be examined to predict the timing of a squall line while also taking into account changes in the environmental conditions with which the storm may interact. Nowcasting techniques often rely upon extrapolation, statistics, and experience-based intuition rather than sophisticated atmospheric models, yet they can be highly accurate on short timescales.
At time scales of a few hours to a month or more, numerical weather prediction (NWP) is the dominant forecasting technique. NWP, or computer-based modeling of the atmosphere, involves representing the current atmospheric state on a three-dimensional grid, applying the physical and dynamical equations that govern how the atmosphere will change in time at each grid point, and repeating this process to generate a forecast of desired length. Computer memory and processing limitations dictate the number of grid points and complexity of small-scale physical process parameterizations that can be reasonably used, with more points generally leading to a better solution. Modern NWP models are developed and maintained in collaboration among multiple agencies and are often coupled with models of the land surface and ocean.
Despite their sophistication, models suffer from shortcomings, such as an incompletely observed atmosphere, insufficient computing resources, and imperfect simulation of small-scale phenomena, such as clouds and precipitation. Systematic forecast evaluation and verification is conducted to quantify model behavior and to identify specific model capabilities and shortcomings. Meteorologists use two approaches to address these shortcomings:
- Application of statistical techniques to correct systemic biases. Past model forecasts are compared to observations over a long period of time to quantify expected errors. The result of this analysis is used to improve output from future model forecasts of similar situations.
- Using an ensemble of models to explore forecast uncertainty that results from the chaotic nature of the atmosphere and errors in the modeling system. Even with a perfect model, very similar initial states of the atmosphere will sometimes become very different in time. To capture this possible variability in the solutions, an ensemble is comprised of many slightly different forecasts, with the differences in how each forecast is obtained meant to mimic modeling and observing system error. If each individual forecast is equally likely and, if enough sufficiently variable forecasts are obtained, the probability of different forecast outcomes can be quantified.
Role of Forecasters
Even as models and objective guidance tools have improved in recent years, highly skilled forecasters remain an integral part of the current state of forecasting. Forecasters are continually trained on how best to use the latest data, model outputs, and forecast tools. Forecasters continue to add value to NWP guidance for daily forecasts, watches, and warnings, especially for high-impact events, and specifically in the 12–48 hours of the forecast period. From 1992–2012, National Weather Service forecasters achieved a 20%–40% improvement over two commonly used numerical models for a 24-hour forecast of one inch of precipitation over one day, even as the NWP guidance continually improved.
Communication of Weather Forecasts
Methods for the delivery of weather information continue to evolve with advances in technology and social science research, which includes the disciplines of communication, psychology, geography, economics, and anthropology. Private citizens, businesses, and institutions have become far more sophisticated in their receipt and use of weather information due in part to the constant expansion of public and commercial weather products through technologies such as the Internet, wireless communication devices, electronic message signs, and other media. For example, location-specific information including hourly forecasts, storm-based severe weather warnings, and radar data is delivered directly to smartphones.
Most forecast products and services are based on discrete values or thresholds. These services provide users with the best estimate of what will happen. However, for optimal decision-making, users need to consider the range of possible events beyond the most likely outcome in determining the appropriate action. They also need information about the likelihood and probable strength of potential high-impact events as early as is practical, given that the increase in uncertainty with time is a complicating factor. This expression of forecast uncertainty is an area warranting improvement and will require increased communication between users and producers of weather forecasts, as well as research on how best to represent this uncertainty (e.g., graphically, with percentages, through text explanations, etc.).
How reliable are today’s forecasts?
Skill, predictability, and lead time
The skill of a forecast refers to how accurate the forecast is compared to some reference or baseline prediction, such as a forecast compared against climatology or persistence of current conditions. The predictability of meteorological events differs based on the size and timing of the event. Larger systems are inherently more predictable at a given lead time than are smaller ones. Predictability also decreases as the lead time — the amount of time between the present and when the phenomenon is expected to occur — increases.
Predictability is further limited by an incomplete representation or, at times, full understanding of the relevant physical processes, inadequate observations and methods for their incorporation into numerical model forecasts, and limitations in computational power. Nevertheless, as the examples below illustrate, advancements in all of these aspects in recent years have resulted in more skillful forecasts across diverse meteorological phenomena and forecast lead times.
For lead times of approximately twelve hours to two days, short-range forecasts are typically issued for meteorological phenomena, such as tropical storms, hurricanes, and frontal systems and their accompanying sensible weather elements (e.g., temperature, wind, and precipitation). Many of these forecasts are significantly improving: two-day National Hurricane Center hurricane track forecasts issued in 2012 had an average error of 79 miles as compared to 140 miles in 2002 and 192 miles in 1992. Likewise, two-day NOAA Weather Prediction Center forecasts of 24-hour accumulated precipitation issued in 2012 were as accurate as one-day forecasts in 2006.
Defined as forecasts with lead times of two to seven days, medium-range forecasts are most successful for meteorological phenomena that stretch across areas of a thousand miles or more, or for larger-scale conditions that set the stage for development of smaller phenomena, such as severe thunderstorms. Over the past three decades, the skillful range of medium-range forecasts has been extended by roughly one day per decade. Specifically, five- and six-day surface temperature forecasts issued by the National Weather Service had the same level of accuracy in 2012 as did three- and four- day surface temperature forecasts, respectively, in 1992.
Extended-range forecasts are typically issued for meteorological phenomena that cover areas ranging from thousands of miles to the size of a continent and involve lead times of one to two weeks. Presently, forecasts of daily or specific weather conditions do not exhibit useful skill beyond eight days, meaning that their accuracy is low. However, probabilistic forecasts issued to highlight significant trends (e.g., warmer than normal, wetter than normal) can be skillful when compared to a baseline forecast. For example, the NOAA Climate Prediction Center operational 8–14 day temperature forecast skill in 2013 was approximately equal to that of operational 6–10 day temperature forecasts from the late 1990s, again demonstrating an increase in forecast success over time.
Monthly and longer-range forecasts
Finally, monthly and seasonal forecasts are typically issued for meteorological phenomena that cover areas ranging from the size of a continent to the planet as a whole. Skill in monthly and seasonal forecasts is extremely variable from period to period, but the skill of NOAA Climate Prediction Center one- and three-month forecasts of temperature and precipitation increased by more than 25% between 2006 and 2013. Increases in forecast skill at these lead times can largely be attributed to improved understanding of and ability to forecast major modes of large-scale climate variability such as the El Niño-Southern Oscillation and Madden-Julian Oscillation.
Users of Weather Information and Forecasts
Government agencies, businesses, academia, and the general public have developed innovative ways to use forecast products to increase economic efficiency and productivity, advance scientific research, and reduce exposure to weather risks. Across the entire spectrum of users, there are growing requirements for accurate forecasts with greater temporal and spatial specificity. The demand for accurate, specialized forecasts from various economic sectors has led to the continued growth of the U.S. weather industry, particularly private forecast services.
During the past ten to fifteen years, improvements in forecast quality have enabled the development of advanced and customized applications to suit the needs of various user groups. The need for different levels of detail is dictated by how the information is to be applied. For example, during a winter storm, different users need specific information relevant to their business operations. The general public is typically most interested in weather conditions that may impact their activities and safety. Specialized users, such as utility companies, public health/air quality sectors, road crews, and airlines need much more specific forecasts (e.g., where will the snow fall, how much, and when) to help determine the need for changes in their operations. These users require a level of specificity not available from general-purpose forecasts. Adjustments in forecasts for those specifics require concentrated, continuous monitoring of many parameters by highly skilled forecasters as well as involving those users in collaborative design of future forecast products.
Opportunities for Future Improvement
Opportunities exist for increasing forecast skill at all time ranges. However, realizing these opportunities will require further research, close international cooperation and coordination, improved observations of the atmosphere, ocean, and land surface, and the incorporation of these observations into numerical models. Also, benefit will be derived from higher spatial resolution of numerical models; increasingly powerful supercomputers; wider use and improvement of model ensembles; the development of data mining and visualization methods that enable forecasters to make better use of model guidance; and collaborative forecast development activities among operational forecasters and researchers.
Beyond improving the forecast itself, improvement in the communication and best use of forecast information is also needed. Research integrating social science is key in identifying opportunities for future advances. For example, research conducted by social scientists across multiple disciplines has found that delivering weather warnings across multiple media increases the likelihood that people will get and act upon this information. Scholars have conducted numerous studies on different public groups about perceptions of risk and uncertainty. They are also working to explore the relative value of effective communication of accurate weather forecasts to appropriate decision-makers. Collaborative research with social scientists will also enable forecasters to codify best practices in forecasting philosophy, communication, and training amidst rapid technological change. An increase in the presence and use of social media is contributing to additional avenues for providing weather information and collecting real-time observations.
In summary, weather forecasts are increasingly accurate and useful, and their benefits extend widely across the economy. While much has been accomplished in improving weather forecasts, there remains much room for improvement. The forecasting community is working closely with multiple stakeholders to ensure that forecasts and warnings meet their specific needs. Simultaneously, they are developing new technologies and observational networks that can enhance forecaster skill and the value of their services to their users.
[This statement is considered in force until March 2020 unless superseded by a new statement issued by the AMS Council before this date.]