Cover Image

Not for Sale



View/Hide Left Panel
Click for next page ( 91


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 90
90 CHAPTER 5 Airport Demand Management The research has concluded that the current system suffers from unclear responsibility: no one has the authority and accountability for the management of congestion at mega-region airports. The management of existing resources could be improved: this chapter builds the case that capacity in the mega-regions can be increased only when the all the major players are empowered to solve the problem. Opportunities to reduce mega-region airport congestion and improve the overall cost and quality of passenger service exist; what would be beneficial are policies and programs that encourage key decisionmakers to grasp such opportunities. When the system fails, a trigger mechanism could be set off; with the responsibilities of each party clearly specified, the goals of accountability and transparency could be met. There are roles for both the national and local levels in defining these roles and procedures. The responsibility of those in charge is to make air travel reliable for passengers; this is a form of accountability beyond making the airport available for all classes of aeronautical activities. A way to do this is to focus on the passenger experience. A congested airport does not necessarily make the airport rea- sonably available nor are delays arguably nondiscriminatory from the passenger perspective. Exhibit 5.0. Highlights and key themes included in Chapter 5. 5.1 Introduction and their customer service values (see Exhibit 5.0 for highlights and key themes in Chapter 5). From the research undertaken to date on this project, it is As used in this report, the term "demand management clear that the scarce resource of capacity is not allocated effi- program" is one that limits delays that occur if too many air- ciently. Chapter 5 investigates methods in which such capacity craft are scheduled to arrive at an airport during a particular could be allocated in a way that balances passenger service from time. Under this use of the term, demand management is two perspectives: flight frequency and service reliability. The not meant to refer to any program specifically designed to balance of stakeholder roles is explored in this chapter, with the decrease the number of air trips made.33 goal of developing approaches that are agreeable to all stake- holders and fit the individual needs of a congested airport. The chapter examines alternatives to the current congestion and 5.2 The Promise of Demand demand management structure in which the roles at the fed- Management: A Case Study eral and local levels are unclear. It reviews a wide variety of can- The same quantity of air transport payload capacity can didate strategies and actions. Chapter 5 further develops several be provided with larger numbers of small aircraft flights or strategies to increase airport throughput capacity, examining smaller numbers of large aircraft flights. It has long been the barriers and constraints that impact their implementation. The research explores the idea that more attention should be paid to studies at individual congested airports to prioritize the 33 See Chapter 2 for a discussion of strategies that are designed to decrease the value of individual flights, based on their contribution to delay number of passengers flying.

OCR for page 90
91 recognized that the decisions of air carriers about what recipe gers based on the aircraft type and the great circle distance of to use have important ramifications for the quality of service the flight route. The seat information is based on U.S. averages and level of accessibility provided by the air transportation obtained from the DOT's T100 database, when available, and system on the one hand and for the amounts of flight traffic, company websites in other cases. Hereafter, the term "aircraft levels of congestion and delay, and infrastructure require- size" is used to mean the number of seats. ments on the other. To explore these trade-offs, the research The average size of an SFO arrival flight over the four team analyzed June 2008 schedules for several days at one selected days was 135 seats, whereas the standard deviation is coastal mega-region airport, SFO. The aim was to document 80 seats, reflecting the diverse size of the fleet serving SFO. To and analyze the wide diversity of aircraft sizes contained in the examine the size distribution in more detail, a cumulative dis- SFO fleet mix in order to identify situations where a different tribution function was constructed (Figure 5.1), which indi- choice of aircraft size could substantially reduce delay with a cates, for any given size, the proportion of flights with aircraft minimal loss or (taking the reduced delay into account) even at or below that size. On the small end of the distribution, an improvement in the level of service provided. about 3% of the flights are 13 seats or below. These include a smattering of corporate jets. Next there are a sizable number of regional jets, of sizes ranging from 30 to 80 seats. Altogether, 5.2.1 SFO Fleet Mix aircraft 80 seats and smaller account for 26% of the fleet mix. The research team examined SFO arrivals on four days in The biggest portion of the fleet--about 60%--is in the June 2008: the 5th, 13th, 18th, and 25th. These days were cho- 100180 seat range. These include the large jet mainstays of the sen because they feature varying levels of congestion and delay, domestic airline fleet, such as the Boeing 737, MD 80 series, as measured by on-time performance. Flight information was and A320 series. Widebodies of 200 seats or more--including downloaded from the FAA Aviation System Performance Met- Boeing 767s, 777s, and 747s along with Airbus 340s--account rics (ASPM) database. In theory, the database includes all for the remaining 14% of the SFO fleet mix. flights, including air carrier, general aviation, and cargo, that The diverse fleet mix at SFO means that the vast majority were actually flown. The database does not include cancelled of total seats are provided by a relatively small proportion of flights. Altogether, there were 2,165 arriving flights on these flights, as shown in Figure 5.2. This figure was constructed by days, of which 10 were cargo flights. In the fleet-mix analysis, sorting the 2,155 flights from largest to smallest, and then the team focused on the 2,155 passenger flights. computing the fraction of total seats provided by the cumula- The ASPM passenger flight data was supplemented with tive sum of the seats with the largest aircraft. To aid with inter- two other variables: the estimated seats available for passen- pretation, the aircraft size for each of these flights is also plotted Figure 5.1. Aircraft size distribution, SFO (June 2008).

OCR for page 90
92 Figure 5.2. SFO fleet mix profile (June 2008). using the secondary vertical axis. Figure 5.2 reveals that 87% ranging from around 100 to several thousand miles, while the of the seat capacity is provided by the 66% of the flights using handful of non-scheduled large jet flights--often diversions Airbus 319 (124 seats) or larger aircraft. Conversely, 10% of or ferries--are on average somewhat shorter. the SFO flights using the smallest aircraft account for less Aside from distance, aircraft size is related to segment traf- than 2% of the total seats. fic density--the quantity of passenger traffic per unit time. If One could argue that the "value" of a flight depends on not the density is low, smaller aircraft are needed to attain an only its size but also its distance. Longer distance flights gen- acceptable level of flight frequency. As traffic increases, airlines erally have higher fares and serve trips of longer duration. can use larger aircraft, exploiting economies of scale while still Moreover, the time savings from making the trip by air instead maintaining a convenient number of daily flights. of by surface mode is roughly proportional to distance. Figure Figure 5.4 depicts this phenomenon. Based on the June 18, 5.2 therefore contains a second share curve that is based on 2008, SFO arrival schedule, Figure 5.4 summarizes the service seat-miles instead of seats. This curve is generally higher than provided by individual passenger carriers on individual flight the seat-share curve. For example, the 40% of the flights flown segments in terms of the number of flights (plotted on the hor- with the largest aircraft generate 60% of the seats but 66% of izontal axis) and the average seats per flight (plotted on the the seat-miles. This difference reflects the positive correlation vertical axis). Different symbols are used to differentiate the between aircraft size and flight distance. The only exception is segments according to length. Seats per day for a segment, for the smallest aircraft in the fleet, bizjets of 15 seats or fewer, which is directly related to traffic density, is the product of the on which many of the flights are quite long, which accounts two coordinates. A series of isoquants indicate combinations for the sharp up-tick in the seat-mile share curve at the far of aircraft size and flight frequency that yield the same quantity right of the figure. of seats per day. Segments on which small (<100 seats, in this The relationship between size and distance is shown more discussion) aircraft are used have low densities, almost always directly in Figure 5.3, which plots aircraft size against flight less than 300 seats/day. Within this set of segments, the key length on a log-log scale. The data in this figure are differen- determinant of aircraft size is distance, with smaller Embraers tiated according to whether the flight was a scheduled flight assigned to segments 300 miles or less, larger Canadairs serv- appearing in the Official Airline Guide. The correlation for ing the 601- to 1,200-miles segments, and a mixture of the two the scheduled flights is evident, with the trend-line indicating types employed on the 301- to 600-miles segments. that aircraft size increases proportionally with the square root Although all the segments served by small aircraft are low of flight distance. No such relationship is evident for the non- density, not all low-density segments are served by small air- scheduled data. The small corporate jet flights have lengths craft. The variability is particularly notable for segments of

OCR for page 90
93 Figure 5.3. Aircraft size versus segment length, SFO arrivals (June 2008). 300600 miles. For example, one airline provides 280 seats per 5.2.2 Economies of Aircraft Size day from Portland (a 551-miles segment) with two MD80s, whereas another airline uses five Canadair flights to provide The fleet-mix behaviors observed in the previous discussion 270 seats from Boise (522 miles). Similarly, an airline flies are shaped by two main economic factors: economies of scale eight Embraers a day from Medford, a distance of 329 miles, in the cost of operating aircraft and the service advantages of whereas another provides almost as many seats (239 vs. 260) higher flight frequency. Cost economies are illustrated in Fig- with two 737 flights from Burbank, which is 326 miles away. ure 5.5, which plots aircraft direct operating cost per seat Figure 5.4. Daily seat capacity production, SFO (June 18, 2008).

OCR for page 90
94 Figure 5.5. Operating cost per seat, fuel: $4.30/gal. against segment distance for two aircraft types, the 144-seat This does not mean that such a choice is a bad one, but it does Boeing 737-400 and the 50-seat Embraer 145. Unit costs for imply that the service benefits of operating small aircraft must the regional jet are consistently higher, but the ratio increases be weighed against the congestion costs. with stage length, from 1.7 at 100 miles to 1.9 at 1,000 miles. These trade-offs were analyzed using the four June 2008 days More important, however, the absolute difference in cost per described, based on a deterministic queuing analysis. The seat increases rapidly with distance. Therefore, the cost of approach can be visualized using a queuing diagram, as shown increasing schedule convenience by offering more flights is the in Figure 5.6, which is based on June 5th operations. The hori- lowest on short-haul flights. On the other hand, the benefits zontal axis is the time of day; the vertical axis is the cumulative of high frequency are probably greater for these flights, as they count. There are two count curves, one for the schedule and often are used for short-duration business trips and also must one for actual arrivals. The schedule curve gives the number of compete with the automobile. Finally, short-haul segments flights that are scheduled to arrive at or before a given time. It have traditionally been served by commuter airlines, which in is constructed by sorting the flights in order of scheduled arrival the past were subject to less stringent safety regulation if they time. The horizontal coordinate of the point corresponding to operated aircraft of 60 seats or fewer. Pilot contracts with large the nth flight is the time when it is scheduled to arrive, and the jet carriers have also limited the sizes of aircraft that can be vertical coordinate is n. The actual curve is constructed in a operated by their lower-paid counterparts working for com- similar way, except in this case the flights are sorted in order of muter affiliates. actual arrival time. Looking at Figure 5.6, one can observe that the two curves virtually overlap during the early part of the day. This means 5.2.3 Operational Impacts of Up-gauging that at the time when n flights were scheduled to have arrived, At SFO, as in most airports, small aircraft use the same run- very close to n flights had arrived, implying very little delay. ways as large ones and occupy them for about the same length Later on, the curves separate. For example, the 500th arrival of time. Thus, when the airport is congested, the operational was scheduled to occur around 21:20, but it was not until more impact of a small flight is no less than that of a large one. than an hour later that the 500th flight actually did pull in. This Indeed, the slower approach speeds and longer in-trail separa- implies that arriving flights at SFO were delayed during the tion requirements of small aircraft can result in longer effective latter part of the day. The total amount of this delay can be service times. Thus, when airlines and other airport users pro- obtained by subtracting the sum of the scheduled arrival times vide capacity with more small flights rather than fewer large from the sum of the actual arrival times. On June 5, it was ones, the result can be higher levels of congestion and delay. 12,790 min, or an average of 24 min per flight.

OCR for page 90
95 Figure 5.6. Queuing diagram, SFO arrivals (June 5, 2008). Figures 5.75.9 show the queuing diagrams for the remain- is a very bad day, with substantial delays beginning at 8 AM in ing three days. Figure 5.7, for June 13, shows slight delays over the morning. Average arrival delays on these three days are, much of the day, but no high-delay periods such as seen in the respectively, 15, 8, and 45 min per flight. The research team later part of June 5. June 18 is free of significant delays, except wanted to estimate how arrival delays on these days would be for a very small amount at the end of the day. Finally, June 25 different if certain flights were removed from the arrival traffic Figure 5.7. Queuing diagram, SFO arrivals (June 13, 2008).

OCR for page 90
96 Figure 5.8. Queuing diagram, SFO arrivals (June 18, 2008). and developed an algorithm for doing so. The details of the Removing a flight can never make the arrival time of process are not important, but the basic principles are very another flight later. straightforward: If a delayed flight is removed, the delay incurred by that flight is (of course) eliminated. If the actual arrival time of the removed flight was during a If the actual arrival time of a flight is during a high-delay period with no delay, removing it will make no difference. period, removing the flight enables subsequent flights to Figure 5.9. Queuing diagram, SFO arrivals (June 25, 2008).

OCR for page 90
97 move up and incur less delay, until there is a gap in the haul flights are most easily substituted by surface transport. traffic stream large enough to make a trailing flight's arrival Thus, eliminating short-haul flights could be an efficient way time independent of the time of the flight in front of it. to reduce congestion and delay at SFO. In assessing the operational impacts of eliminating short- Among these principles, the last is the most ambiguous, as haul flights or any other strategy, it is useful to quantify delay in one must determine whether another flight could move up if seat-minutes rather than aircraft-minutes. The costs of delay to a preceding flight were eliminated or landed earlier. If the time airlines increase with aircraft size, as do (on average) the num- between the two successive arrivals is, say, 60 min, there is ber of passengers affected by a delay. Thus, the operational clearly no interaction between them, but if it is 1 min, there impacts will be calculated in units of seat-minutes. Seat-minute almost certainly is. The question is where to draw the line. The delay can be calculated from a queuing diagram in which one 90th percentile of the observed inter-arrival times was used counts seats on the vertical axis instead of counting flights. (i.e., time between successive arrivals), conditional on airport To predict the seat-delay impact of eliminating short-haul capacity, in the data. This turned out to be 4 min for high- flights, a set of hypothetical "cut-off" distances (80, 150, 200, capacity conditions and 4.4 min for other conditions. With and 300 miles) was chosen. For a given distance, the research these assumptions, about one third of the total delay--or team predicted how seat-delay would change if all of the flights 8 min per flight--incurred by SFO arrivals can be attributed within that distance were removed from the arrival stream. to arrival capacity constraints. The remainder is due to prob- Also, to put these results in perspective, the additional line-haul lems at other airports and airline internal malfunctions such time was calculated if these aircraft seats were transformed into as maintenance problems. car seats--that is, the passenger on these flights drove to SFO Using the ability to predict the delay impacts of removing instead of flying. This was done by comparing the scheduled flights from the arrival stream, the research team considered flight time with the driving time estimated from Yahoo maps. three up-gauging strategies. This additional line-haul time is also expressed in seat-minutes, based on the sizes of the aircraft used for the eliminated flights. Although the units are the same, the unit values may be differ- 5.2.4 Up-gauging Through Elimination ent, depending on the relative seat-minute cost of vehicle oper- of Short-haul Flights ation, aircraft operation, and aircraft delay, as well as the fact In the first strategy, short-haul flights are eliminated. As that driving times are more predictable than flight delays. observed, short-haul flights generally use smaller aircraft, so Results averaged over all four days are shown in Figure 5.10. this strategy implicitly involves up-gauging. In addition, short- Eliminating flights shorter than 150 miles saves more in delay Figure 5.10. Time impacts of eliminating short-haul flights, by cut-off distance, 4-day average.

OCR for page 90
98 Figure 5.11. Time impacts of eliminating short-haul flights, by cut-off distance (June 25, 2008). than it costs in additional line-haul time. As the cut-off dis- for more frequent commuter service. As discussed, there are tance increases, more flights are eliminated and the delay sav- situations in which segments of comparable length and total ings increases, but the extra line-haul time increases much seat capacity vary differently--for example, Boise to SFO with faster. The cross-over point, assuming equal valuation of the five small jet flights a day on one airline versus Portland to two forms of time, is somewhere between 150 and 200 miles, SFO with two large jet flights on another. In the flight consol- and probably closer to the former. If, as may well be the case, idation approach, the migration of services from the former the unit cost of flight-delay seat-minutes is considerably greater model to the latter one is encouraged. than that of extra flight time, eliminating flights of 200 miles, Like the short-haul flight elimination strategy, this one or even 300 miles, or less may be cost-beneficial. involves trade-offs. The cost of flight consolidation is less fre- The greatest operational benefit from eliminating short- quent service and diminished schedule convenience. To make haul flights occurs on highly congested days, such as June 25 flight consolidation as painless as possible, it is desirable to in the sample. Figure 5.11 therefore shows results for that day identify situations in which the elimination of a flight through only. Although it displays the same pattern as in Figure 5.10, consolidation has the least impact on convenience. To quan- the delay savings curve is shifted up, so that, for short distances, tify the effect of consolidation, it is imagined that if a given delay savings are double the line-haul time increase. Moreover, flight is eliminated, the passengers on that flight would be if the unit cost of flight delay were more than 1.6 times that of forced to take the next earlier flight on the same airline from extra driving time, eliminating all flights less than 300 miles the same origin airport. This is somewhat arbitrary, as passen- on such a highly congested day would be justified. The day-to- gers could respond in other ways, such as taking the next later day differences found in comparing Figures 5.10 and 5.11 flight, switching airlines, or going to a different airport. How- point to the promise of having a flexible strategy for serving ever, the assumption has the virtue of simplicity, and for most short-haul trips, using flights on good days and surface passengers, the assumed response is probably the least disrup- modes on congested days. This strategy is referred to as real- tive one. It respects customer brand (and airport) loyalty, and, time intermodalism. because it assumes early arrival, does not disrupt passengers' planned activities in the Bay Area.34 5.2.5 Up-gauging Through Flight Consolidation 34 On the other hand, schedules at the origin may be disrupted because passen- gers must depart earlier. For this reason, some passengers would opt to take the A second approach to up-gauging is to encourage, when next later flight, but modeling this mixed response is not complex and probably appropriate, the substitution of less frequent large jet service not worthwhile.

OCR for page 90
99 With this assumption, one can evaluate the loss of conven- uled to arrive 180 min--or 3 hours--earlier.) The portion of ience from eliminating a particular flight by finding the differ- the distribution corresponding to these flights is shown in Fig- ence between its scheduled arrival time and the scheduled ure 5.13. It shows that a small but non-zero fraction of flights arrival time of the previous flight from the same origin oper- have an SDI of zero. These are cases in which airlines inten- ated by the same airline and multiplying this difference by size tionally schedule two arrivals from the same origin at exactly of the aircraft serving that flight. A metric is obtained with units the same time. The data contain five such cases, all involving of seat-minutes, which is traded against the seat-minutes of major carriers operating large equipment from distant hubs. delay that would be saved if the flight did not take place. This Carriers do this presumably to provide sufficient capacity metric is termed "schedule delay impact" (SDI). while maintaining the ability to cancel flights without disrupt- SDI was evaluated for each SFO arrival in the four June 2008 ing passengers when traffic or capacity is low. Aside from these days in the sample. Figure 5.12 shows the cumulative distribu- cases, the lowest SDI values are on the order of 1,000-seat min- tion of the SDI obtained, using a log scale. It is apparent from utes, the equivalent of a 33-seat flight whose predecessor is the figure that flights fall into three categories. First, there is a 30 min earlier. set of "one-off" passenger and cargo flights for which this met- The research team used the SDI metric to identify the best ric is meaningless. These were all assigned an arbitrary, large flights to eliminate in pursuing the flight consolidation strat- SDI value and correspond to the vertical part of the distribu- egy. Analogous to the short-haul elimination strategy, a mini- tion on the right of the figure. Next, there is a set of flights that mum SDI value was set and eliminated all flights below that are the first flights of the day for a given airline and origin. value. The impact on queuing delay at SFO was then assessed. Given assumptions, elimination of these flights would force The procedure is somewhat complicated by the fact that when passengers to travel on the previous day. For all intents and one removes a flight, the SDI values of other flights may change, purposes, these flights are "off the table" as far as consolidation as the removed flight is no longer available to receive passen- is concerned. Flights in this category appear in the s-shaped gers from some other flight. The team therefore updated the portion of the curve to the left of the vertical portion. SDIs after each flight consolidation. The results for June 25, the The remaining flights--about 65% of the total--are the worst day in the sample, appear in Figure 5.14. ones that can be considered for elimination through flight Queuing-delay savings of a magnitude greater than schedule- consolidation. Among these, the most promising are those delay savings are obtained for SDI cut-offs up to 4,000 seat- with the lowest SDI values--say, 10,000 seat-min or less. (To min. Queuing delay is clearly more expensive than schedule put this figure in perspective, a flight using a 56-seat aircraft delay, as it ties up aircraft and forces passengers to wait in would have an SDI of 10,000 if the previous flight was sched- planes and airport terminals, whereas schedule delay can be Figure 5.12. Cumulative distribution of SDI, SFO arrivals (June 2008).

OCR for page 90
100 Figure 5.13. Cumulative distribution of SDI, SFO arrivals (June 2008). anticipated and incorporated into passengers' activity sched- set 15 seats as the threshold for small aircraft, which would ules. For these reasons, it is not unreasonable to assume a unit eliminate all bizjets but no commuter flights. To assess the cost ratio of 2:1 or more. Figure 5.14 suggests that a consider- mobility impacts of this strategy, one must assume a time able number of flights could be eliminated through consolida- penalty for diverting a flight (or a seat) from SFO to some tion before the optimal trade-off point is reached. other alternative. That penalty reflects the additional travel time from being forced to fly into a less accessible airport. Depending on one's point of view, it may also be increased to 5.2.6 Up-gauging by Diverting capture the greater value of time of bizjet travelers as com- Very Small Aircraft pared to the rest of us. Finally, the strategy of diverting small aircraft from SFO to Figure 5.15 summarizes the impacts of this strategy. On some other local airport was considered. The research team June 25, diverting aircraft of 15 seats or fewer saves over Figure 5.14. Time impacts of eliminating short-average; June 25, 2008).