Weather Forecasting Accuracy for FAA Traffic Flow Management: A Workshop Report (2003)

Historically, only a small fraction of the resources allocated to weather forecasting by federal agencies (other than the Federal Aviation Administration) have been focused on the development of weather guidance that supports the needs of the aviation system, and there is no reason to think this will change. Therefore, the FAA and the commercial airlines will have to take the lead if they want to see development and implementation of the type of operational products needed to improve the safety and efficiency of the aviation weather system. The technology and knowledge to significantly improve the 2- to 6-hour convective forecast products for aviation exist now. In particular, recent advances in understanding subsynoptic-scale meteorology, high-resolution observation capabilities, computer power, communication systems, and software systems make it possible to dramatically improve weather forecasting products for the aviation community. Such products do not have to be part of the National Weather Service suite of operational products; adequate electronic processing power and communications are already available for disseminating any new forms of guidance that would be developed. Users of the national airspace can simply decide what types of products are needed and build them. During the final workshop session, summarized in this chapter, participants considered, in the light of the workshop presentations, what research activities are necessary to move toward the next generation of convective forecasting products.

Many workshop participants stated that probabilistic guidance from ensemble model calculations combined with improved high-resolution observations may hold the most promise for improving convective forecast

products in the next 5 to 10 years. Current limitations in predictive capabilities due to uncertainties in initial conditions and model formulation will likely require a longer research effort to provide marked improvements. In the meantime, probabilistic approaches can provide meaningful information for managing the nation’s airspace.

The aviation traffic flow system is essentially a never-ending sequence of decisions. For such systems there is a wealth of information available on how to optimize decision making in a manner that minimizes costs, maximizes benefits, or both. For example, the utility and insurance industries routinely make decisions based on reliable¹ probabilistic information. With probabilistic guidance and estimates of the respective costs of yes-or-no decisions, it is possible to determine threshold probabilities above which, when averaged over many events, the ratio of costs to benefits is optimized. Such threshold probabilities transform probabilistic guidance into optimal “black-white” decisions (e.g., whether to expect having to take an alternate route to avoid expected adverse weather and therefore having to load extra fuel). Moreover, reliable probabilistic guidance makes it possible to define a uniform and consistent set of criteria on which to base operational decisions.

For example, suppose the aviation traffic flow system mandated that pilots do not try to navigate through areas of thunderstorms once the percent-area coverage exceeds a certain threshold.² If such a guideline were in place, it would be highly desirable to have aviation weather guidance that provides reliable probabilities of the critical percent-area coverage. This type of system is possible with current technology, though increased spatial resolution of the next generation of numerical models likely will allow much better guidance by better resolving the location, organization, and orientation of convection exceeding the critical percent-area coverage. With such

guidance it is possible to calculate a threshold probability of the percent-area coverage above which, for example, the frequency of having to divert will, when evaluated for many events, cost the airline more money than if other possible strategic actions were taken to avoid the critical area. This approach can be applied to all sorts of decisions in the traffic flow system. Traffic flow managers would have quantitative information to help guide decisions in a manner that reduces overall costs.

Much discussion during the workshop focused on the issue of “strategic” versus “tactical” decisions. Several individuals argued that a continuum of guidance is necessary to optimize the decision-making process. Such guidance would be based primarily on observations for very short-term forecasts (0 to 2 hours) and mostly on model output for forecasts of 6 hours or more. The period in between (2 to 6 hours) will most likely require an optimal blend of both observations-based forecasts and model forecasts. In all instances, statistical techniques will be necessary to generate reliable probabilistic guidance.

One proposal discussed by the workshop participants was that strategic plans be structured in such a way that they can be readily modified (in a tactical framework) to adjust for changing conditions. This approach can result in shorter flight distances. It was also noted that terminal forecasts must be an integral part of such a system.

Several workshop participants mentioned that, ideally, traffic flow managers and pilots would like to have forecasts of the radar reflectivity field. While in the past such a request was dismissed out of hand, it is now possible to generate guidance of this sort using cloud-scale resolution numerical models. However, there are several points to bear in mind. First, although the models have algorithms that can convert model parameters into reflectivity, there is currently very little skill at forecasting timing, location, and intensity of individual convective storms. On the other hand, we do have meaningful skill forecasting the timing and location of mesoscale areas of convection and the organizational mode of the convection in those areas. For example, high-resolution models can distinguish between organized lines of storms and scattered air mass convective cells. Such forecasts would be of value for strategic decisions. Moreover, since we now have a WSR-88D archive of observed reflectivities, it is possible to statistically postprocess model forecasts to correct for model bias and to create categorical reliable probabilistic forecasts of parameters such as percent-area coverage. It is even possible to create probabilistic forecasts of percent-area coverage of convective tops over selected values (e.g., over 12 km altitude). For line-type situations, such forecasts would show elongated contiguous areas of high probability of high percent-area coverage. Likewise, it is

possible to construct the same type of guidance using statistical models that use current conditions and archives of historical observations. An optimal blend of the two approaches (model output and observations) would mathematically select and weight the best predictors from both approaches to produce the guidance with the least error. This hybrid approach can be designed to provide a continuum of guidance for forecasts as short as a few minutes out to 6 or more hours.

It is important to emphasize that this forecasting system can be built with today’s technology. As observations become more plentiful, models increase their resolution, multimodel ensembles are created, and archives of observations and model forecasts lengthen, the skill of such a system can only improve. A number of workshop participants expressed the view that this approach is the best foundation on which to base a research and development plan. Other than a commitment of resources, there is no reason why such a system cannot be built. Many workshop participants noted that the success of more advanced forecast products will rely in part on effective training of those who use the convective weather products to make aviation decisions.

In a couple of instances during the workshop, representatives of various components of the FAA traffic flow management system and the commercial airlines indicated a willingness to transform their operations to take advantage of guidance in probabilistic form. This sentiment is consistent with that of professionals working in other areas affected strongly by convective storms (e.g., hydrometeorology, quantitative precipitation forecasting, and severe storm forecasting) who have already recognized the advantages of forecasting in this manner and have taken significant steps to transform their guidance into probabilistic form. Therefore, experience and expertise for producing probabilistic guidance for forecasting aviation convective weather are available.

Workshop participants suggested that improving the time and space resolution of observations (especially in the boundary layer) is critical for improving convective weather forecasts. Utilizing all available Doppler weather radars, which provide high-quality measurements of boundary layer winds, could be a particularly useful first step to enhancing the coverage of surface observations. Fortunately, the cost of automated surface observing systems has decreased dramatically during the past decade, portending the availability of national coverage on the mesoscale as individual states install networks. Likewise, the soundings obtained from commercial aircraft will provide a wealth of new upper-air data, as will the next generation of satellite observations and ground-based remote sensing. These new observations, coupled with the increase in numerical model resolution and

the creation of archives of WSR-88D data, can provide a foundation for an advanced short-term forecasting system. There is cause for considerable optimism in improving skill at forecasting convection, especially as we leave behind the era of having to parameterize convection and switch to models that capture much of the nonhydrostatic processes that characterize convective events.

The workshop concluded with a discussion of critical tasks and future directions to address the issue of improving operational convective weather forecasting. These include:

• Defining probabilistic forecasting and determining how it could best be applied in air traffic management.

• Identifying how the FAA could best utilize available weather forecast products by incorporating them into its current operational activities.

• Establishing predictability confidence limits for all convective regimes, defining key convective regimes and model capabilities in those areas, and characterizing the impact of convective forecasts on air traffic control decision making.

• Identifying and evaluating the various means and mechanisms for generating probabilistic forecasts.

• Clarifying concepts of accuracy, verification, and reliability of forecasts.

• Describing the attributes of convection most relevant to the FAA operationally.

• Identifying the best approaches for conveying convective forecasts and products to air traffic controllers and pilots.

• Outlining needed research to improve the reliability and utility of 2 to 6 hour convective forecasts, especially probabilistic forecasts.