Help with Optimization and Decision Support Modeling for Business HW5

Reducing Flight Delays through Better Traffic Management

Author(s): Ved P. Sud, Midori Tanino, James Wetherly, Michael Brennan, Miro Lehky, Ken Howard and Rick Oiesen

Source: Interfaces , Jan. - Feb., 2009, Vol. 39, No. 1, 2008 Franz Edelman Award for Achievement in Operations Research and the Management Sciences (Jan. - Feb., 2009), pp. 35-45

Published by: INFORMS

Stable URL: https://www.jstor.org/stable/25622772

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms

INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Interfaces

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Interfaces infjiMJ Vol. 39, No. 1, January-February 2009, pp. 35-45 DOI I0.1287/inte.l080.0417 issn 0092-21021 eissn 1526-551X1091390110035 ? 2009 INFORMS

THE FRANZ EDELMAN AWARD Achievement in Operations Research

Reducing Flight Delays Through Better Traffic Management

Ved P. Sud, Midori Tanino, James Wetherly Federal Aviation Administration, Washington, DC 20591

[email protected], [email protected], [email protected]}

Michael Brennan, Miro Lehky Metron Aviation, Sterling, Virginia 20166 [email protected], [email protected])

Ken Howard, Rick Oiesen Volpe Center, Cambridge, Massachusetts 02142 [email protected], [email protected])

As air traffic in the United States has grown over the last several years, traffic demand has begun to outstrip capacity. As of 2005, the Federal Aviation Administration (FAA) had no effective approach for strategically man aging a weather event that has been very disruptive to the national aviation system?large-scale thunderstorms that block the major flight routes in the northeastern United States. The operations research team that supports the FAA's efforts to provide innovations in air traffic management, led by researchers at Metron Aviation, Inc. and the Volpe Transportation Center, recognized the consequence of this operational deficiency and set out to resolve it. In this paper, we show how this team (1) developed and applied system-simulation models to quantify the extent of the traffic flow management

problem and convey its magnitude to the FAA and to the aviation industry; (2) designed the Airspace Flow Program (AFP), a new approach to managing air traffic that could correct

the problem within the limitations of a short development cycle and a change-resistant culture; (3) designed and developed an interactive simulation system that could be and was used to refine and perfect

this concept prior to deployment by developing policies on the use of a decision support system; (4) engaged FAA and airline traffic management experts in a series of interactive exercises using the simu

lation system to develop the final software design, operational procedures, and decision rules for deployment and use; and

(5) provided a clear and convincing postdeployment benefits assessment for the new traffic management approach.

The deployment of this new capability was an enormous success that both the FAA and the airline community heralded widely. The postdeployment impact assessment showed benefits to the aircraft operators and the flying public of almost $190 million in 2006 and 2007, the first two years of use, compared to less than $5 million in design and development costs. Broader usage of AFPs and new applications for them show a projected 10-year benefit of approximately $2.8 billion.

Key words: simulations: applications; transportation: models, assignment, scheduling, vehicle routing.

Amajor responsibility of the Federal Aviation Administration (FAA) is to provide air traffic management services for the national airspace. Air traffic management fills the real-time role of ensur

ing that airplanes and passengers travel safely and

efficiently from the departure airport, through the airspace, to their destination.

Air traffic management has two interrelated func tions: (1) air traffic control and (2) traffic flow manage

ment (TFM). The better-known function is air traffic

35

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management 36 Interfaces 39(1), pp. 35-45, ?2009 INFORMS

control, a role that the familiar air traffic controller

fills largely by using a radar scope and a headset to manage aircraft within a defined volume of airspace. Each controller is responsible for keeping each flight in his or her volume safely separated from every other flight.

TFM's role, and the subject of this article, is to keep the amount of traffic each controller must direct to a manageable level by anticipating future traffic demand and strategically controlling aggregate flows of flights to keep the demand within tolerable bounds. To support its TFM function, the FAA has developed the Enhanced Traffic Management System (ETMS), a software and communications system that collects and integrates real-time data from FAA and airline sources to identify future demand and capacity imbal ances at airports and in the airspace. ETMS includes decision-support tools that help FAA traffic man agers prevent these imbalances by issuing structured directives known as traffic management initiatives. These actions redistribute traffic demand over time

and space by delaying and rerouting flows of traffic. Metron Aviation, Inc. and the Volpe Transportation

Center support the FAA's System Operations Services Programs Office by helping to sustain and improve existing ETMS software, and by designing, develop ing, and deploying new concepts and approaches for the next generation of TFM. The operations research (OR) practitioners at Metron Aviation and Volpe, the 'TFM OR team," work with the FAA and the airlines to identify TFM operational problems and to solve them by developing new concepts and approaches.

In this paper, we will describe how this team used the principles and practices of OR to design and help deploy a new type of traffic management initiative, and thus identified and solved a major TFM problem in the national airspace.

This new type of traffic management initiative, the Airspace Flow Program (AFP), is a powerful TFM capability that was introduced in June 2006 with extensive public exposure from the most senior execu tives in both the FAA and the aviation industry (Levin 2006). The innovation is projected to save aircraft operators $1 billion to $3 billion in operating costs by reducing delays and cancellations over the next decade, and is projected to reduce passenger delays by more than a million hours each year.

Identifying and Illustrating the Key TFM Problem

Traffic-Flow Management Before Airspace Flow Programs As of 2004, the FAA had developed and deployed effective ETMS-based solutions for two key TFM prob lems (1) controlling high-arrival demand at airports

with limited capacity and (2) managing moderate scale thunderstorm systems.

To monitor and control high-arrival demand at airports, traffic managers use the Flight Schedule

Monitor (FSM), an ETMS decision support tool first deployed in 1998. When the FSM projects a future demand or capacity imbalance at any major US or Canadian -airport, managers can use algorithms in the tool to compute and assign a delayed departure time for each flight; this traffic management initiative, known as the Ground Delay Program (GDP), extends the arrival demand safely and fairly. ETMS commu nicates the assigned departure times, which the FSM has computed, to the airlines for planning and to air port control towers for enforcement.

To manage moderate-scale convective weather (thunderstorms) in the en route portion of a flight's path, the FAA developed the ETMS-based Flow Con strained Area (FCA) tool (Figure 1), which was intended to be used for rerouting traffic. Using this tool, traffic managers can geographically define con gested areas in the airspace, such as a region of thun derstorm activity. ETMS will then produce a list of

(1) FAA creates FCAs over small-scale weather systems. f \

f (2) ETMS generates lists \ Jl^^^flp^^fu I / ?f fights in the FCAs. l^^^^^^^y \ \s. (3) Airlines route flights on ,mmm"^^^^>^^^^^

I \f the lists around FCAs. 1 St^J^x^C^^l

Figure 1: The FAA uses FCAs to manage traffic during small-scale en route weather events.

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management Interfaces 39(1), pp. 35-45, ? 2009 INFORMS 37

flights expected to traverse each area. Traffic man agers use these lists to decide which flights must be routed around the weather and congestion.

As of 2004, the FAA had no effective approach for managing one key TFM problem: wide-scale connec tive weather fronts, i.e., extended lines of thunder storms blocking major flight routes. These weather systems typically occur during the busy summer travel season and cause severe problems when they occur over the heavily traveled Northeast. This traffic management problem regularly resulted in enormous, system-wide disruptions, leading to billions of dollars annually in increased operating costs and revenue loss to the airlines and to general aviation operators, and inconvenience to the flying public.

Such severe weather systems substantially reduce en route capacity and are too large for much of the traffic to fly around. In the early 2000s, to reduce the traffic flow through these systems, the FAA began to employ a traffic management approach that could be implemented using the then-available ETMS tools. The FAA recognized that holding flights on the ground had to be part of reducing demand. It had only one tool that could impose ground delay, namely, airport GDPs. Therefore, it began to issue GDPs to manage en route problems. Although they were designed as an airport tool, GDPs were drafted into use for en route severe weather under the theory that reducing the flows into and out of a number of

major airports on the perimeter of a weather system would sufficiently reduce the flow of traffic through the system to make it manageable. Figure 2 shows an example of multiple airport

GDPs used in support of severe weather en route. Under this approach, all traffic destined for a num ber of airports around the weather system (from 6 to 12 airports, depending on the severity of the system) would be delayed to reduce traffic demand.

The FAA and the airlines recognized that applying GDPs for en route weather problems was an imperfect solution; however, they believed that GDPs helped reduce demand at a reasonable cost to the aircraft

operators. Lacking any better alternatives, and in the absence of any insight about their effectiveness or effi ciency, GDPs were routinely employed as part of the en route weather management plan.

(1) For large-scale weather events, the FAA ran GDPs at perimeter airports to slow traffic through the weather.

Figure 2: A large weather system, such as this, might have been managed by 9 or 10 airport GDPs.

The TFM OR team at Metron Aviation and Volpe were not convinced that GDPs were useful in manag ing large-scale en route weather events. The TFM OR team members wanted to propose and implement a

method of managing these events that would provide better control at less cost; however, they knew that to overcome the institutional resistance to interfering

with a working system they would have to prove the need for change and show that an effective alternative was possible and implementable with an acceptable level of risk.

Showing the Need for a Better Solution Although GDPs had been used for en route weather events for several years, no appropriate TFM impact modeling tools existed; therefore, no analyses of the effectiveness of this practice had been made. To show the need for an improved approach, the TFM OR team developed a new modeling tool that would help eval uate the effects of a number of traffic management ini tiatives, such as applying GDPs at various locations, on traffic demand elsewhere in the national airspace. The TFM OR team used this modeling tool to mea sure the impact of multiple airport GDPs on the flight demand through the congested area and generated a series of figures and statistics to clearly illustrate the results.

Figure 3 shows output from the new modeling tool. The left side of Figure 3 shows the modeled

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management 38 Interfaces 39(1), pp. 35-45, ?2009 INFORMS

* * %

Unmanageable demand

I ^^Beibre GDPs . f After GDPs | Maximum capacity

Figure 3: The graphs show modeled demand within the weather system before and after the application of 10 GDPs.

traffic-demand profile within the thunderstorm sys tem based on scheduled traffic demand. The horizon

tal line indicates the maximum number of flights the region can accommodate during the thunderstorms; the plot shows high levels of demand above capacity. The right side of Figure 3 shows the modeled demand profile after the application of 10 GDPs that were intended to reduce demand within the thunderstorm

system to meet the maximum capacity. The lighter portion of each bar reflects flights assigned delay by the airport GDPs. We see that the GDPs have hardly improved the problem; although they have shifted the demand a little, they have barely reduced it. The bur den of bringing the demand under the capacity line will fall to overworked air traffic controllers using air borne holding, diversions, and other high-cost, less safe methods.

Figure 4 helps explain why using airport GDPs to manage traffic during en route weather is both ineffi cient and ineffective.

GDPs are not selective about which flights into the airport they control. Flights into GDP airports that do not pass near the weather, such as Los Angeles to Chicago, are delayed unnecessarily. Flights pass ing through the weather but not going to one of the GDP airports, such as Atlanta to Washington, are not delayed, and so contribute to the excess demand. Additionally, because the excess demand in the weather has not been reduced sufficiently, New York departures have to be held on the ground, leading to system gridlock.

The modeling tools developed by the TFM OR team were able to quantify these effects, which had not

been fully understood. In a typical en route weather problem, the models showed that 60 percent of the flights that fly through the weather system are not controlled at all because they do not fly to one of the GDP airports. At the same time, up to 80 per cent of the flights delayed by the GDPs were delayed unnecessarily because they did not fly through the weather. All told, nearly 70,000 minutes of wasted delay were imposed on aircraft on a typical severe weather day. At an airline operating cost of $31 per minute of ground delay, this suggested that a possi ble cost reduction of more than $2 million for each thunderstorm day could be achieved.

These statistics provided convincing evidence that GDPs were an inadequate tool for managing severe

(1) Before AFPs, FAA ran GDPs at perimeter airports.

(2) Many flight, needlessly delayed (D). / \ y\

\ \^ / (3) Too many flights passed through -^jj^Sjft y the weather without delay (ND). ?^.r \

*v (4) Too many flights already within the weather so controllers had to delay NY departures tactically (D).

Figure 4: This map shows an example of GDPs controlling the wrong flights to solve the en route weather problem.

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management Interfaces 39(1), pp. 35-45, ?2009 INFORMS 39

weather en route, and that the FAA needed an alter native that was more effective, efficient, and equitable.

Designing the Solution: AFPs In designing an alternative, the TFM OR team at

Metron Aviation and Volpe considered several con cepts involving radical new (for operational aviation) approaches to the traffic management solution, includ ing network flow models and integer programming solutions. However, in practice, an operation as vast and complex as civilian aviation can absorb only a limited amount of change at a time. To be accepted by the community and to be realistically deployable in a short period, a solution would have to be as small a change in concept and require as little new software as possible, while still providing an ambitious new capability.

The team's objective then became finding a way to provide a new means of managing en route weather by building on existing tools and procedures, adding and changing established practices only where and only as much as necessary. In time, a broad outline of an approach came together. The new traffic manage ment tool would need to have the following:

A user interface to let traffic managers define a region of constrained airspace, based on weather and demand, and specific traffic flows through that space. This would be provided by the existing FCA geo graphic interface.

Automation to identify each flight currently planning to pass through the designated airspace and meeting the criteria that the traffic managers speci fied, and to provide that information in a regularly updated list that would be useable by new decision support tools. The existing ETMS capability already provided part of this; the rest would have to be built.

A tool to graphically compare capacity to demand at a constrained resource and to allow traffic managers to make demand-control decisions. This could be pro vided by extending the airport display tools in FSM to include FCA flight lists.

Allocation algorithms to compute the required ground delay to assign to each flight to keep demand within the capacity constraints. Although the require ments for assigning delay for the en route prob lem are different from the airport problem, there is enough commonality that the FSM algorithms could be extended such that they could be suitable.

A communications infrastructure to distribute

the delay information to the flight operators and all FAA field locations. The infrastructure in place to sup port GDPs provided almost all the requirements for the new tool.

A set of tested procedures to ensure that both the traffic managers and flight operators use these capa bilities effectively. Our experience with GDPs helped in developing these requirements; however, there

were many critical issues and dozens of important issues that would have to be successfully resolved for a deployment of the new approach to succeed.

Figure 5 illustrates how this new approach, the AFP, affects traffic. In the presence of a large-scale weather event that reduces air capacity, traffic managers create an FCA across the front. ETMS software identifies the

flights going through the FCA; these flights, and only these flights, are assigned ground delays to reduce demand. Flights that are close enough to the edge of the weather to route around can do so and thus

have their assigned delay lifted. For reasons discussed below, some flights are allowed to pass through the weather without assigned ground delay; the tool can support this flexibility.

This solution approach had the two key character istics needed to ensure a successful deployment. First, because it is built on well-known concepts, it gave the team confidence that the FAA and flight operators

would be comfortable with the changes and would be able to develop and manage AFP procedures. Sec

(1) FAA creates an FCA and runs an AFP.

(2) Incidental flights are not delayed (ND). \ y\

(3) All flights heading NE into thTweather are delayed i^^^^^^^^/ \

Some flights can still route around the AFP (ND)T| /

(5) Demand is properly controlled and NY flights can flow free (ND).

Figure 5: The AFP approach integrates existing tools to create a new type of traffic management initiative.

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management 40 _ ' Interfaces 39(1), pp. 35-45, ?2009 INFORMS

ond, because the AFP solution extensively used exist ing infrastructure, development cost and risk were

minimized. In the fall and winter of 2004, the TFM OR team

constructed the impact-modeling tool that illustrated the need for a new approach to the large-scale thun derstorm problem and developed the AFP concept. In February 2005, it presented the results and pro posed solution at a joint meeting of FAA and airline representatives. In April 2005, based on the team's analysis, the FAA decided to proceed with the devel opment and deployment of the AFP capability.

Using Interactive Simulations to Ensure a Successful Deployment

The Deployment Challenge After securing the FAA's decision to proceed, the TFM OR team faced the challenge of ensuring a suc cessful deployment. Although the AFP concept was simple, it was also very general. It left open ques tions about how the concept should be applied; these

would have to be resolved before the new capabil ity could be deployed. These questions included the following:

For what weather conditions or other congestion events should AFPs be employed?

For a given event, what specific FCA should be used? How should it be shaped, where should it be located, and what flights should be excluded from controls?

How should the capacity for an AFP (the rate at which it can accept flights) be established?

How should unexpected flights that occur in the middle of the day and flights that route into an AFP be treated? How should cancelled flights, or flights routed out of an AFP, be handled?

How should AFPs interact with GDPs? How should AFPs interact with each other?

What rights should airlines have to substitute one flight for another in an AFP? How would exercis ing those rights affect the integrity of the program?

What are the software requirements and best pro cedures for revising and cancelling active AFPs?

There was an additional component to this chal lenge: to make this program compatible with other inflexible commitments that the FAA had made, the

AFP capability had to be fully deployed and oper ational in just over a year. A major development program to replace all ETMS software by 2009 was already under way; therefore, the ETMS system would be subject to a software freeze after the spring of 2006. The last chance to improve TFM to meet rapidly rising traffic demands and expected delay increases during the next four to five years depended on the success ful deployment and effective operations of AFPs. Any level of failure would be an enormous setback for both

the FAA and the OR community in aviation. The FAA made the decision to commit to the AFP

program in April 2005 and set a deployment date of May 2006. This meant that the software require ments had to be finalized within five months to allow

time for coding and testing, and the operational pro cedures had to be established within nine months to

allow time for training the FAA and airline traffic managers. To understand the magnitude of the chal lenge posed, one must consider the complexity and criticality of the system that the AFPs were designed to change.

The US national airspace includes tens of thousands of flights per day, hundreds of controllers, dozens of airline operations centers, and a team of FAA traf fic managers distributed over the Air Traffic Con trol System Command Center in Herndon, Virginia, and 20 en route traffic centers across the country. The ETMS software and TFM procedural rules must be able to anticipate and properly handle an enormous range of possible actions by personnel at any of these locations, under any situation that might arise. If the

AFP capability is to be successful, and not break an operational system that is critical to the national and worldwide economies, AFPs must:

Interact properly with other existing traffic man agement initiatives, such as GDPs, ground stops, and reroutes.

Accept changes in plans by flight operators, such as cancelling flights, diverting flights to other airports, changing flight routes and departure times, and sub stituting flights without compromising the integrity or effectiveness of the overall TFM solution.

Deal intelligently and safely with unexpected changes in system capacity when weather becomes other than predicted.

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management Interfaces 39(1), pp. 35-45, ?2009 INFORMS 41

Communicate all relevant information to all affec

ted parties in a reliable and timely manner, includ ing aircraft operators, traffic managers, and departure controllers at airport air traffic control towers.

Be able to withstand, generally, all unexpected events and perturbations that characterize the national airspace on a bad weather day. None of these requirements could be developed or

tested under operational conditions. All of these capa bilities had to be perfected and in place on the first day of AFP operations.

Designing a Simulation Model to Meet the Deployment Challenge The TFM OR team recognized that the best way to develop the software specifications and the proce dures that would meet all these requirements in the available time was through system modeling. The team laid out a plan to develop a realistic, interactive simulation environment that modeled the national

airspace and the impact of operational decisions on its outcome. FAA and airline experts could then use this test environment to identify the system-software needs and to develop feasible and effective opera tional procedures for all the decision makers in the system. The software and procedural requirements perfected in the simulations would be the model for the operational system, thereby reducing the deploy

ment risk substantially. The principal components of the simulation system

were (1) a flight event simulator that modeled the actions and interactions of all flights in the simulation, (2) an ETMS emulator that modeled the communica tions and control infrastructure used to monitor and

control the national airspace, and (3) the decision sup port tools that the FAA traffic managers, airlines, and

other aircraft operators used to monitor and modify the behavior of flights under their control. These com ponents could all be extended to reflect new software

and procedural concepts as they evolved. Figure 6 shows a schematic of the simulation system. Design and development of this nationwide simula

tion model began in May 2005; a workable capability that met all the high-level requirements was available for use in late August 2005.

Flight status Flight status and situational and situational

information 3 information^ ???-?

Flight I % \ Flight 3^^' controls F^ght nfo \ controls *09&

^^^^^^ database

Figure 6: An interactive real-time simulation system was the key to pre deployment planning.

Using Interactive Simulations to Capture Expert Knowledge The FAA designated a team of aviation experts to sup port the successful development and deployment of the AFP capability. This team included experienced FAA traffic managers, representatives from both the airline industry and general aviation community, the TFM OR team from Metron Aviation and Volpe, and other research organizations. This group was responsi ble for transforming the AFP operational concept into software requirements and operational procedures.

For six months starting in August 2005, when the simulation system was ready for operation, simula tion exercises were the focus of the AFP development effort. For 8 to 10 days every month, a dozen or more air traffic managers, airline personnel, and representa tives of general aviation traveled to Metron Aviation's facility in Herndon, Virginia, to use the simulation system to work on the problems and solutions for a successful deployment.

A typical AFP simulation experiment would start with a common traffic management problem and apply a proposed combination of operational proce dures and software rules to resolve the problem. The FAA and airline managers would interact with the system and with each other to evaluate the feasibility and effectiveness of the candidate approach applied to the hypothesized problem. Figure 7 illustrates the flow of a simulation.

Feasible Procedures and Software Requirements Developed Through Simulations The interactive simulation approach allowed the FAA and airlines to assess factors that could never have

been gauged using other approaches to developing

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management 42 Interfaces 39(1), pp. 35-45, ?2009 INFORMS

Demand/capacity problem Procedure/software approach Demand scenario: Capacity scenario: Procedural approach: Software rules: e.g., e.g., pre-holiday e.g., thunderstorms e.g., a series of AFPs each new AFP applied

rush w/ many popup over Ohio moving shifting eastward with takes precedence over flights. into New York. the weather. previous ones.

-Simulation input In the simulation, FAA applies Airlines respond to changes in traffic management controls to their flights by canceling, delaying, balance capacity and demand. ^ f and rerouting flights.

I ^HKhKi T*-[ emulator J~H^^^"flcte^^B / m*.

_Y_ Evaluate results and apply conclusions

Revise procedures: Revise software: e.g., minimize number of e.g., new AFP controls AFPs that are applied. should not override

existing ones.

Figure 7: Each simulation exercise allowed traffic managers to evaluate software and decision options for a given traffic management problem.

such a capability. Factors such as workload distri bution, predictability, and the interaction of multiple decision makers, all pursuing their own objectives, were clearly evident by using the simulation exer cises. This enabled the system experts to quickly elim inate operationally impractical approaches and focus on those that would be manageable under real-time conditions.

As an example of how the simulation system led to key deployment decisions, consider the general case of extended thunderstorms in the Midwest blocking air traffic to and from the New York/New England area. In early experiments, FAA traffic managers created AFPs that captured all traffic through the affected area and applied departure delays to reduce demand to the available capacity (Figures 8(a) and 8(b)). In addition, the simulation environment showed the impact of this decision on other parts of the airspace system. Figure 8(c) shows the flights sched uled to depart from Newark airport after the AFP has been applied. The lighter portions of the bars

represent flights that have been assigned departure delays by the AFP. These delays would need to be enforced by the already overworked tower controllers at Newark; this would be an unacceptable increase in their workload. This suggested that there was a fatal flaw in one of the main motivations for deploying the AFP capability.

However, the flight modeling in the simulation went farther. To provide an aircraft, every departing flight requires an arriving flight, as anyone who has ever waited by an empty gate at an airport is aware. The simulation included software agents that acted as airline dispatch centers, matching arriving aircraft at each airport to the most appropriate departing flight. These flights could not depart until their aircraft had arrived. Therefore, simply delaying arrivals into the New York airport would lead to natural departure delays. The traffic managers used the simulation to evaluate whether this effect could solve their prob lem; they ran another experiment in which the AFP applied delays only to the inbound flights?those

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management Interfaces 39(1), pp. 35-45, ?2009 INFORMS 43

(a) Midwest demand before AFP

Be ?w CW? HWp _ . [?.?. S.S1 a

FCAA95 05/17/2005 14:012 ENTRY

i s S P n |:-^ n ^' I ^ i"" i ^| i iiiiiiiiiiii i 8 1111 T1?J^ JL i I I 1 1

(b) Midwest demand after AFP

^mIt J*^*'

FCAA95 05/17/2005 15:202 ENTRY

i \ -1-1-j-j-j-j-j-1-j-j-' :

'' Illllllllll: i ! I I ! I i 8 8 8 8 I !

_m?.!^Mi?m}m*m!$*._._ ._ \ '<i ' ' ' . ?:

(c) Newark departures after AFP

^ afm] wj? a isj

EWR 05/17/2005 15:202 ETA

^^^i^^^^S ^ Figure 8: The simulation showed that assigning delays to all flights in the

weather system would have too much negative impact on New York-area airports.

heading into the Northeast from the west and from the south (Figure 9). When the AFP assigned delays to only some flights

in the airspace (Figure 9(a)), the preexecution model ing tools showed that demand remained substantially

(a) FAA managers delayed only inbound flights with the AFP.

:_i_^.^__.1 (b) Initial impact assessment showed airspace

demand over capacity.

a & ta Wso *i & a

FCAA95 05/17/2005 15:202 ENTRY |

jj : (c) End-to-end simulation showed the airspace

problem solved.

-^j^j? -g^j rgj

FCAA95 05/18/2005 01:552 ENTRY

?00i-j--j--~T~-1-1 " ''"j ??r~-r?-1 [-r-? ?r4Q0

300 j-[MD

^I^B ^B ^B^B ^B^B^H^B ^B ^B ^B ,^Brw

mimmmmmm'1,:/ , T^T'.: ~~???

Figure 9: The simulation showed that a key airspace capacity problem could be solved without explicitly delaying flights leaving the New York area airports.

over capacity (Figure 9(b)). However, when the sim ulation played out, and the downstream departure delays caused by late-arriving flights were properly reflected in the postexecution demand profile, it was clear (Figure 9(c)) that the demand through the AFP

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management 44 Interfaces 39(1), pp. 35^5, ?2009 INFORMS

could be kept within capacity without requiring New York tower controllers to enforce departure delays. This observation led to the policy of one-way AFPs for Midwest weather; the AFP concept might have failed without this policy. Other examples of the many procedural and soft

ware decisions that developed using the simulation system include the following:

For the first year of deployment, AFPs would be used at only six predefined locations, address ing major thunderstorm systems in the northeast (Brennan 2007).

Controls from GDPs should take precedence over controls from AFPs for any flights subject to both types of controls.

The ground delay should be canceled for a flight that routes out of an AFP, providing an incentive to reduce demand in the congested area.

Airlines should be allowed to substitute one delayed flight with another to best manage their schedules. This minimizes the AFP's effect on their

schedules while still meeting the FAA requirements. Without the simulation environment to identify and

resolve these operational issues in advance and to build their solutions into the software and procedures, the deployment of the AFP capability would not have been as successful as it was and might have failed completely.

Assessing the Postdeployment Benefits of AFPs The new AFP capability was operationally ready on June 6, 2006; the first AFPs were executed on June 9. The FAA and the airlines considered the program an immediate success (Blakey et al. 2007, Keyes 2006). For the FAA traffic managers, AFPs provided more control. For the airlines, AFPs imposed less delay and provided more flexibility and predictability. For the flying public, AFPs made it more likely that flights would eventually get through.

The FAA used AFPs to manage thunderstorms on 18 days during the summer of 2006. Their success, as reflected by the positive feedback from the avi ation community, was good but not sufficient. An FAA requirement is that a postimplementation review must be performed to justify the deployment of new

capabilities. The TFM OR team's final task was to produce a well-supported benefits assessment using operational results.

To assess the benefits that the AFPs provided, the TFM OR team considered the benefits both to flight operators and to passengers. It took the same funda mental approach for each group: use recorded opera tional data to estimate the costs incurred by the flight operators on those days when AFPs were used, and on similar days before AFP technology was available. After applying a regression model (see AFP Benefits to Airlines for details) to adjust these costs for additional, relevant conditions, the reduction in costs incurred

when using AFPs provides an estimate of the opera tional benefit of the program.

AFP Benefits to Airlines

The principal benefit to airlines of the new capa bility is the reduction in additional operating cost caused by disruption on severe weather days. The TFM OR team designed a regression model predict ing expected cost as a function of the severity of the

weather (Klein et al. 2007), the total demand on the system, and whether AFPs or GDPs were used in sup port of severe weather.

The regression model indicated a reduction in oper ating cost to the airlines of more than $2.7 million on each of the days AFPs were used, consistent with and slightly better than the predeployment analyses. The model considered all days in the severe weather seasons (June 1 to August 30) for 2005-2007. When applied to the 43 days in 2006 and 2007 that AFPs were used, the total estimated airline AFP benefit for the first two years of use was $118.5 million, with a 95 percent confidence interval from $12.5 million to $223.6 million. We can compare this to the total one-time cost of deploying AFPs, including the OR, the software development, the simulation exercises, and the traffic manager training, of approximately $5 million.

AFP Benefits to Passengers The use of AFPs provides greater efficiency to the airlines, other flight operators, and to passengers. To compute the effect of AFP use on passengers, we first ran a separate model to compute the effective passen ger delay on each day in the three summer seasons,

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms

Sud et al.: Reducing Flight Delays Through Better Traffic Management Interfaces 39(1), pp. 35-45, ?2009 INFORMS _ 45

which also modeled the rebooking of passengers on later flights if their scheduled flights were cancelled (Wang 2007). This effective delay statistic was used to estimate passenger benefits. The regression is similar to the one above; however, it used estimated passen ger delay rather than airline costs; it also showed that using AFPs instead of GDPs will save passengers more than a million hours of unnecessary delays per year. This equates to a total estimated benefit of $1.2 million to passengers on each day that AFPs were used.

Conclusion: The Future of AFPs The addition of the AFP capability has transformed how the FAA thinks about TFM problems. The first-year application of AFPs was limited to prede fined configurations used to manage severe en route

weather in the Northeast. However, flexibility was built into the initial AFP design such that the capabil ity would be applicable to many other TFM needs.

The early success of the initial AFP application inspired many new uses. Additional preplanned

AFPs and procedures for dynamic AFPs, to be applied wherever convective weather develops, were devel oped for use in the summer of 2007 and beyond. AFPs have been used to jointly control arrivals at closely grouped airports, such as Dallas-Fort Worth and Dallas-Love airports. They have been used to manage high-departure volume from the New York area (Levin 2007, Blakey et al. 2007).

Based on this much broader use of the AFP con

cept, current estimates of AFP benefits to aviation in the United States over the next 10 years range from $2.5 billion to $3.4 billion. Additionally, there are sub stantial intangible benefits. AFPs, by giving airlines options to route around weather or wait it out, pro vide more flexibility and predictability to airlines than the alternatives do. Airlines are less likely to cancel flights in AFPs, thus benefiting passengers. AFPs are

more equitable because they include all flights that contribute to the problem.

AFP use is also spreading internationally. AFPs have been used for nearly a year to manage traf fic bound for resort destinations in Mexico. Canadian

traffic managers have considered using them to con trol end-of-weekend private and charter flights into the extended mining and energy regions of northwest Canada. AFPs will be an important part of the air traffic management system currently under develop

ment for South Africa, and are being planned for use in China, Australia, and other countries developing air traffic management solutions.

The success of AFPs provides a high-profile success story for the FAA that shows its willingness to aggres sively and effectively pursue new solutions for both present and future aviation problems.

References Blakey, M., J. May, R. Chew, M. Sammartino. 2007. Summer

delays. Press conference, May 23, Federal Aviation Admin istration, Washington, D.C. Retrieved November 7, 2008, http: / / www.metronaviation.com / media.php.

Brennan, M. 2007. Airspace flow programs?A fast path to deploy ment. /. Air Traffic Control 49(1) 51-55.

Keyes, A. 2006. FAA seeks to cut air traffic congestion. National Pub lic Radio (July 14), http://www.npr.org/templates/story/story. php?storyld=5558360.

Klein, A., R. Jehlen, D. Liang. 2007. Weather index with queuing component for National Airspace System performance assess

ment. Retrieved September 18, 2008, http://www.atmseminar. org/past-seminars/7th-seminar-barcelona-spain-july-2007/ papers/ paper_024/view.

Levin, A. 2006. New system could reduce flight delays during storms. USA TODAY (May 14), http://www.usatoday.com/ travel/news/2006-05-14-flight-delays_x.htm.

Levin, A. 2007. FAA says it has plan to minimize summer storm delays. USA TODAY (May 23), http://www.usatoday. com/travel/flights/2007-05-23-faa-delays_N.htm.

Wang, D. 2007. Methods for analysis of passenger trip performance in a complex networked transportation system. Dissertation,

George Mason University, Fairtax, VA. Retrieved September 24, 2008, http://catsr.ite.gmu.edu/pubs/WangDissertation.pdf.

This content downloaded from �������������73.24.246.41 on Sat, 13 Feb 2021 01:32:29 UTC��������������

All use subject to https://about.jstor.org/terms