Analysis of Gate-Hold Delays
At the OEP-35 Airports
John F. Shortle
Jianfeng Wang
Juan Wang
Lance Sherry
Center for Air Transportation Systems Research
Department of Systems Engineering and Operations Research
George Mason University
Fairfax, Virginia 22030
Email: jshortle@gmu.edu, jwang5@gmu.edu, jwangl@gmu.edu, lsherry@gmu.edu
Abstract—One point of congestion in the air transportation
system is the set of gates at a terminal. When an inbound
flight is unable to pull into its assigned gate, the flight, its
connecting passengers, and its flight crews experience delays and
missed connections. Previous research has focussed on optimizing
gate assignments, both scheduled and disrupted, and optimizing
surface flow. This paper analyzes the degree to which gate hold
is a problem and the functional causes of gate hold. Analysis of
flight performance data for 35 OEP airports for the summer of
2007 identifies that: (i) Significant gate-hold delays, in which more
than 30% of arriving aircraft are delayed, occurred at 11 of the
OEP-35 airports, (ii) major gate-hold delays are rare events (e.g.,
once a month at ATL), (iii) on days when there is a major gate
hold, large delays are experienced by all major carriers at the
airport, (iv) the primary causes of gate-hold delays are increased
turnaround times and/or disrupted arrival/departure banks. The
methodology for analysis, the results, and the implications of
these results are discussed.
Keywords—gate; congestion; delay; disruption; gate hold
I. I NTRODUCTION
The objective of this paper is to evaluate the degree to
which airport gates are a limiting resource in the flow of
airplanes arriving to and departing from an airport. As an
airplane arrives to and departs from an airport, it passes
through several potential choke points. These include runways,
taxiways, ramps, gates, and so forth. Depending on the demand
and number of resources, different points in this process may
become constrained and act as choke points.
The runways (and associated constraints) are usually the
limiting resource in the flow through the airport. This is
because various separation requirements – for example, wakevortex separation requirements – limit the maximum number
of operations that can be safely handled in a given time
interval. Once an airplane has landed and has taxied to its
gate, it can often pull directly into the gate without further
delay, but sometimes it must wait for a departing aircraft to
leave the gate. When the shortage of gates is more severe, an
arriving aircraft may be sent to a “penalty box” where it must
wait until a suitable gate becomes available (Figure 1).
While gates are not usually thought of as a choke point,
this paper aims to evaluate the extent to which this can be the
Penalty Box
Fig. 1.
Penalty box at Chicago O’Hare airport
case. Conditions under which gates act as a limiting resource
may increase over time if capacity is added to other parts of
the system (say, through capabilities proposed in NextGen)
without an appropriate analysis of the matching gate capacity.
At a high level, we can view this problem from the
perspective of queueing theory [1]. In a theoretical queue,
delays are related fundamentally to one or more of three
issues: (a) a large arrival rate, (b) a low service rate, or (c) a
small number of servers. Here, the gates are the servers and
the service rate is the rate that aircraft can be turned at the
gate. For example, if an aircraft has a longer turnaround time
than what is scheduled, this is effectively a reduction in the
service rate of the gate. For a queueing system operating near
capacity, reducing the service rate – even by a little bit – can
lead to significant delays [1]. In summary, gate-related delays
are attributable in some way to either (a) too many arriving
aircraft, (b) long turnaround times, or (c) insufficient gates. A
second goal of this paper is to identify functional origins of
gate-related delays related to these three fundamental causes.
This paper is organized as follows. Section II summarizes
literature related to gate delay. Sections III and IV discuss
several methods for estimating gate delay using the BTS and
ASPM databases. Section V presents results for the OEP35 airports as well as detailed results for Atlanta Hartsfield
International Airport (ATL) and John F. Kennedy International
Airport (JFK).
The results indicate that on most days, gate delays do
not significantly limit throughput of the airport. However,
on certain bad days, the limited number of gates leads to
extreme delays. The problematic days appear to be linked
to disruptions in the schedule. Disruptions in the schedule
have the tendency of keeping aircraft on the ground longer,
effectively reducing the gate “service rate.” For example, a
disruption in the schedule may require using a different crew
on a given flight. In a similar manner, if a flight is cancelled,
then the aircraft may either be held at the gate or reassigned to
a later flight, thus increasing the time at the gate. Ground delay
programs for outbound flights may also increase the number
of gates required as airlines choose not to enplane or not to
push back these flights. Finally, disruptions in the schedule
may lead to sets of aircraft arriving at the same time.
II. BACKGROUND AND RELATED LITERATURE
A. Gate Assignment
Gate assignment is usually handled in three phases [2].
The first planning phase occurs several months before the day
of operation. Ground controllers check that a feasible gate
assignment can be made with the proposed flight schedule,
without making an actual gate assignment. The second phase
involves the development of a single-day plan prior to the start
of the actual day of operation. A gate assignment tool (similar
to what is shown in Figure 2) can be used to get the initial gate
assignment plan at the start of the day, or the initial gate board.
The third phase revises these daily plans throughout the day
of operation due to irregular operations such as delays, bad
weather, mechanical failures and maintenance requirements
[3]. The same gate assignment tool can be used to create
the initial gate board at the beginning of the day (phase
2) and to update the gate assignment throughout the day
(phase 3). These updates affect gate hold, and hence flight ontime performance, as well as airport manpower and passenger
relocations.
Previous research on gate assignments has stressed improving the performance of initial gate assignments. The problem
has been modeled as an integer program [4], [5], a mixed
integer program [3], [6] and a network flow problem [7],
[8]. Recently, some models have been developed to focus
on gate changes [9]. Some research has even evaluated the
robustness of the initial gate assignment plan and the real-time
gate changes necessary to meet the stochastic flight delays
that occur in real operations [8], [10], [11]. Such research
0:00
3:00
6:00
12:00 15:00 18:00 21:00 24:00
Gate 1
Gate 2
G 3
Gate
……
Gate n
Fig. 2.
Notional Gate Assignment GUI
has typically considered small stochastic disruptions in the
schedule and not major disruptions. This paper shows that
major schedule disruptions can have a significant impact on
gate congestion, and hence corresponding gate assignment
strategies need to be developed for these scenarios.
Other research has given benchmarks for the operational
efficiency of airports based on input measures such as the
number of gates, the number of runways, airport operating
costs, the number of airport employees, and so forth [12], [13].
Research in [13] shows that terminal efficiency is improved
by expanding the number of gates and managing them in
a way to ensure their effective utilization. This can best be
accomplished by placing them in common or exclusive use
but not preferential use.
B. Airport Surface Operations
Queueing models and integer programming models have
been used to model the taxi-in process [14], [15], [16]. The
minimum service time under ideal turnaround operations has
also been modeled [17]. But the impact of schedule disruptions
on turnaround operations has not been studied thoroughly.
Modeling turnaround times in irregular operations could lead
to a better understanding of gate utilization.
More accurate departure demand prediction is another
potential benefic of improved ground-operations modeling.
Departure demand is based on the push-back time, which
is determined by gate arrival time, the turnaround time, and
the schedule. Departure demand prediction is important to air
traffic controllers for runway and taxiway scheduling. But it
has been found that the prediction of departure demand is
not accurate because the turnaround time can not be predicted
well [18]. Also, a gate shortage can result in pushing back
aircraft from the gate, even though the downstream departure
runway is a constraint [18]. This has implications on matching
the runway and gate capacity under disruption scenarios. If
departure runway capacity drops while arrival runway capacity
remains the same, gate demand is higher because holding at
the gate is preferred since it is more expensive to hold on the
taxiway due to fuel burn, crew cost, aircraft maintenance, and
taxiway congestion.
TABLE I
C OVERAGE OF BTS AND ASPM DATABASES
Carrier
Airport
Cancelled flights
International flights
Aircraft type
Tail number
BTS
20 US major
32 US major
Yes
No
No
Yes
ASPM
All
77
No
Yes
Yes
BTS Flights Only
III. DATA SOURCES
Gate-hold delay is not recorded in actual operations and
thus is not directly available in any database. In particular,
although several databases record the taxi-in time of individual
flights, they do not break this time into its component parts,
such as the delay specifically due to waiting for a gate. Thus,
gate-hold delay must be inferred or approximated using other
information available.
This section discusses several data sources – in particular,
the BTS and ASPM databases – and associated fields that can
be used to infer or approximate gate-hold delay. The precise
algorithms used to to estimate gate-hold delay are discussed
in Section IV. Key differences between the ASPM and BTS
databases are the following [19], [20] (see also Table I):
• BTS includes data for air carriers that have at least
one percent of total domestic scheduled-service passenger revenues (20 carriers); ASPM includes data for all
carriers.
• BTS includes data for operations to and from airports
that account for at least one percent of the nation’s total
domestic scheduled-service passenger enplanements (32
airports); ASPM includes data for 77 airports.
• BTS includes cancelled flights; ASPM does not.
• BTS does not include international flights; ASPM does.
(BTS covers nonstop scheduled-service flights between
points within the United States, including territories.)
• BTS does not include information regarding aircraft type;
ASPM does.
• BTS contains the tail number of flights; ASPM only
contains the tail number of flights that are also in BTS.
Table II shows the number of flights recorded in the data
sources for two example days at two airports. In both days,
ASPM records more flights than BTS, due to the inclusion of
international flights. Also, the gap between ASPM and BTS is
larger for JFK than for ATL, since JFK has a higher percentage
of international flights. As a reference, the third line gives the
average number of daily flights for each airport (averaged over
one year, 2007) obtained from Airport Council International.
The last line shows the number of flights observed from the
public website www.flightstats.com.
In ASPM, key fields that are relevant to this paper are the
wheels-on time (the time that an aircraft lands on the runway),
the gate-in time (the time that an aircraft pulls into a gate), the
actual taxi-in time, the taxi-in delay, and the unimpeded taxiin time; these fields are available in the “Individual Flights”
table. Other key fields are airport, carrier code, season, and
TABLE II
F LIGHTS RECORDED IN VARIOUS DATA SOURCES
Source
BTS
ASPM
Average Daily Flights (ACI)
Flightstats.com
ATL
6/5/07
2239
2690
2724
2794
JFK
7/3/07
675
1175
1223
1324
unimpeded taxi-in time; these fields are available in the “Taxi
Times” table. The actual taxi-in time is the gate-in time minus
the wheels-on time. The taxi-in delay is actual taxi-in time
minus the unimpeded taxi-in time. The unimpeded taxi-in
time is an estimated value. It represents the time it would
take for an aircraft to taxi in the absence of delay-causing
factors such as congestion or weather [20]. The unimpeded
taxi-in time is broken down by calendar year, by season, by
carrier, and by airport (for all carriers and airports reporting
in the ASQP and Aeronautical Radio, Incorporated [ARINC]
data). In 2001, there were 14 reporting carriers and about 300
reporting airports.
In BTS, key fields that are relevant to this paper are flight
date, destination airport, carrier, scheduled arrival time, and
actual taxi-in time. These fields are available in the “On-Time
Performance” table.
This paper also uses information from the public website
www.flightstats.com. This site provides flight information by airport and day, including the flight’s final gate
assignment, flight number, arrival time, departure time, carrier,
and aircraft type. For example, we found that 179 different
gates were used at ATL on June 11, 2007. We use this value
as an estimate of the number of gates at ATL in the summer
of 2007. However, the present number of gates is different,
since new gates have been added [21].
IV. E STIMATION OF G ATE -H OLD D ELAY
This section describes several methods to infer gate-hold
delays based on information in the ASPM and BTS databases.
We also describe a method to estimate gate-occupancy time
by tracking the tail number of aircraft.
Roughly speaking, we approximate gate-hold delay as a
certain portion of taxi-in delay, where taxi-in delay is the
difference between the actual taxi-in time and the unimpeded
taxi-in time. More specifically, we assume that, at most, 5
minutes of the taxi-in delay is comprised of non-gate-related
delay such as taxi-way and ramp congestion. That is,
Gate-hold delay ≈ max(Taxi-in delay − 5 min, 0).
(1)
The maximum function ensures that the estimated delay is
non-negative. For example, if the actual taxi-in time is 7
minutes, and the unimpeded taxi-in time is 6 minutes, then
the estimated gate-hold delay is set to 0. If the actual taxi-in
time is 12 minutes, then the estimated gate-hold delay is set
to 1 minute.
Of course, this is an approximation, since the method assumes that non-gate-related delays comprise exactly 5 minutes
ASPM
Database
Taxi-in
Taxi
in
Delay
Fig. 3.
ASPM
Database
• Date
• Carrier
• Airport
BTS
Database
Gate Hold = Max
(Taxi-in Delay – 5 min, 0)
Gate-Hold
Gate
Hold
Delay
Gate-hold estimation, method 1a
• Same
S
ttail
il number
b
• Closest connection with
turnaround time < 1 day
• Non-paired flights eliminated
Unimpeded
Taxi-in Time
Taxi-in
Delay
+
Gate Hold = Max
(Taxi-in Delay – 5 min, 0)
Departure
Arrival
Gate-Hold
Delay
Paired
Actual
Taxi-in Time
ATL, 6/5/07
No. Arrivals
Before pairing 1120
Fig. 4.
Gate-hold estimation, method 1b
After pairing
Fig. 5.
(or less) of the total taxi-in delay. The approximation may
be partially justified, since it has been observed that gaterelated delay is a dominant contributor to the total taxi-in
delay [15]. Also, we can vary the 5-minute parameter to check
the sensitivity of results to this value. However, the overall
implication of the approximation is that the gate-delay statistic
does not provide an exact magnitude. Thus, it is generally
more useful as a relative metric to identify qualitative trends
and/or differences between airports, carriers, and so forth.
We now describe two specific methods for calculating gatehold delay from the ASPM and BTS databases and one method
for calculating gate-occupancy time.
A. Method 1a: ASPM Database
The first method is based on data found in the ASPM
database in the “Individual Flights” table (Figure 3). Here,
taxi-in delay is a specific field in the database (equal to the
taxi-in time minus the unimpeded taxi-in time). The method
extracts the taxi-in delay from the database and substitutes
it into (1) to obtain the gate-hold delay. A limitation of this
method is that it does not track data associated with cancelled
flights.
B. Method 1b: ASPM and BTS Databases
Figure 4 shows a modified version of the basic method that
combines information from the ASPM and BTS databases.
First, data on each flight is obtained from the BTS database
(from the on-time performance table), including actual taxi-in
time, date, carrier, and airport. The last three fields are used to
look up the unimpeded taxi-in time from the ASPM database.
We assume that any date between June 1 and August 31 is
mapped to the summer season for the purpose of looking up
the unimpeded taxi-in time in the ASPM database. The taxi-in
delay is then calculated as the difference between the actual
taxi-in time and the unimpeded taxi-in time. The rest of the
procedure is similar to method 1a.
C. Method 2: Arrival-Departure Pairing
Figure 5 shows a method to estimate the gate-occupancy
time of a flight using its tail number. The aircraft tail number
is usually available for a flight in the BTS database but not for
a flight in the ASPM database. Each arriving flight is paired
with its departing flight as follows. For each arriving flight,
1075 (96%)
Arrival-departure pairing, method 2
the method searches for the earliest subsequent departure with
the same tail number. If no such departure exists, or if the
next available departure is more than 24 hours later, then the
arriving flight is termed “unpaired” and is eliminated from the
set. If a match is found, the gate occupancy time is the “out”
time of the departing flight minus the “in” time of the arriving
flight.
There are several limitations of this method. First, because
of missing tail numbers, not all of the arrivals can be paired,
resulting in missing turnaround times. For example, for the
data at ATL on 6/5/2007, 2.5% of the flights are unpaired.
Secondly, a missing departure can result in more than one
arriving flight being paired with the same departing flight,
resulting in an abnormally long gate-occupancy time for the
earlier arrival. Finally, the method is limited to flights in the
BTS database, so international flights are excluded.
V. R ESULTS
The results in this paper are based on data collected from
the summer of 2007, June 1 through August 31.
A. OEP-35 Airports
We first give an overview of gate-hold severity at the major
US OEP-35 airports. These airports serve major metropolitan
areas and also serve as hubs for airline operations. More than
70 percent of passengers move through these airports. Delays
at the OEP-35 airports have a ripple effect to other locations.
Key FAA performance measures are based on data from this
set of airports [22].
Figure 6 shows the gate hold severity at the OEP-35 airports,
calculated using Method 1a in the previous section. The bars
in the graph, corresponding to the left-hand side of the yaxis, show the fraction of days during which 30% or more
of the flights experienced a strictly positive gate hold. Out of
the OEP-35 airports, 11 had at least 1 such day; the other 24
airports did not have any such days, so they are not shown
on the x-axis. The right-hand side of the y-axis shows the
estimated average daily gate-hold delay in minutes for the
gate-congested days considered (that is, the days in which
30% or more of the flights experienced a gate-hold delay). The
figure shows that gate-hold delay can be a significant problem
for some (but not all) of the major airports in the NAS. For
16000
Estimated number of days Estimated average daily
in which 30% (or more)
gate-hold delay (min)
of flights
during
g have ggate hold
g these days
y
60
50
Estimated gate-hold
percentage
50%
12000
4000
40%
10000
3000
Days 40
60%
6/11
14000
8000
30%
Min
30
20
10
0
2000
6000
1000
4000
0
2000
8/24
Estimated daily gate-hold
delay in minutes
20%
7/29
6/5
10%
ATL JFK DFW PHL LAX DTW MIA MSP EWR IAH LGA
Fig. 6.
0
OEP-35 airports gate delays, summer 2007
0%
6/1
6/11
6/21
7/1
7/11
7/21
7/31
8/10
8/20
8/30
Day of year
Fig. 8.
Estimated number of days Estimated average daily
in which 20% (or more)
gate-hold delay (min)
of flights have gate hold
during these days
100
Days
AirTran Airways
Atlantic Southeast Airlines
4000
80
60%
3000
Min
60
1000
20
0
0
Fig. 7.
OEP-35 airports gate delays, summer 2007
Delta Air Lines
Three worst days
50%
2000
40
ATL gate hold by day, summer 2007
Gate hold
percentage
40%
30%
sample
"good" day
20%
10%
example, ATL had 50 days out of 92 in which 30% of arrival
flights had gate hold.
Figure 7 shows the same analysis when the threshold is
reduced from 30% of flights with gate hold to 20%. In general,
the airports from the previous figure appear in this figure in
roughly the same order (plus or minus a few places). In both
Figures 6 and 7, the magnitude of delay appears to be at least
somewhat correlated with number of congested days. There
are a few notable exceptions. For example, in Figure 7, ORD
and DEN have a small number of gate-congested days. But
when these days occur, the total daily gate-hold delay is high
relative to the other airports. This is due to the large volume
of traffic at the two airports.
B. ATL
This section gives a more detailed analysis of ATL airport,
since it was among the highest in terms of gate delay (Figures 6 and 7).
Figure 8 shows a break-down of the data for each day of the
summer of 2007, calculated using method 1b. The bars in the
graph (corresponding to the left-hand side of the y-axis) show
the total daily gate-hold delay in minutes. Visual inspection
shows that most days experience a relatively mild gate-hold
delay around 2,000 minutes. However, a few days stand out
as having exceptionally high delays. For example, on June 11,
50% of flights experienced a gate hold and the total delay
was 14,000 minutes. The three worst days – June 11, July 29,
and August 24 – will be examined in more detail shortly. The
general observation is that gate-hold delay is not a problem
on most days. However, when it is a problem, it tends to be
0%
2007-6-5
Fig. 9.
Summer
Average
2007-8-24
2007-7-29
2007-6-11
ATL gate-hold delay by carrier
a big problem.
Figure 9 shows a break-down of the data by carrier. The
figure shows the average gate-hold percentage for the three
major carriers at ATL on various days, as well as a summer
average. The three major carriers are Delta Air Lines, Atlantic Southeast Airlines, and AirTran Airways, where Atlantic
Southeast is a feeder carrier for Delta and also uses Delta’s
gates. Each of the three carriers has at least 200 arrivals per
day while no others carriers at ATL have more than 35 arrivals
per day. Although the figure shows that there are differences
between the major carriers, no major carrier is exempt from
gate-hold delay on the “bad” days. That is, when it is bad for
the airport, it is bad for all the carriers.
We now give a more careful comparison of the differences
between the carriers. Table III shows the results of a t-test (at
the 95% confidence level) comparing the delay between the
carriers. (In all cases, the assumption that the carriers have
equal delays is rejected if the t-statistic is greater than 1.96
in absolute value.) There is no statistical difference between
Delta and AirTran. However, there is a statistical difference
between Delta and its feeder carrier, Atlantic Southeast. This
may indicate that Delta gives gate preference to its main flights
over the feeder flights. Also, the differences between AirTran
and the combined operations of Delta and Atlantic Southeast
TABLE III
ATL GATE - HOLD DELAY BY CARRIER
Group
Delta vs. AirTran
Delta vs. Atlantic Southeast
Atlantic Southeast vs. AirTran
Delta & Atlantic Southeast vs. AirTran
t-statistic
-1.03
19.56
-16.48
9.10
70%
60%
Difference
Not significant
Significant
Significant
Significant
50%
Gate-hold
Percentage
40%
2007-06-05 (good day)
30%
Summer Average
g
2007-08-24 (bad day)
20%
2007-07-29 (bad day)
10%
2007-06-11 (bad day)
0%
Aircraft Type
(sorted by number of seats in ascending order)
Gate-Hold
Delay (min)
Fig. 11.
Aircraft Seats
Fig. 10.
ATL gate-hold delay by aircraft seats
may indicate that a higher fraction of common gates would
reduce gate hold.
In general, when two aircraft compete for one gate, preference may be given to the larger aircraft since it carries more
passengers and the delay cost is higher. Figure 10 shows the
scatter plot of gate-hold delay by aircraft number of seats
for all flights in the summer of 2007 (using method 1a). It
shows that aircraft with a higher number of seats are less
likely to get a very high gate hold delay. A t-test shows
that gate-hold-delay and aircraft-seat-number are correlated
(with a confidence level greater than 99.99%). The correlation
coefficient is -0.026, meaning that 100 more seats bring down
the gate-hold delay by 2.6 minutes per flight on average.
Figure 11 shows gate hold percentage for different major
aircraft types in different days (using method 1a). All major
aircraft types were affected on the worst days, despite the
difference among their number of seats. So even though
larger aircraft get some preferential treatment when a gate is
assigned, this treatment did not exempt larger aircraft from
experiencing gate-hold delay on the worst days.
C. Bad Days at ATL
We now examine more closely the system behavior on the
three worst summer days. It is helpful in this regard to also
examine the system behavior on a good day for comparison.
Figures 12-15 show system characteristics on four days: June
5 (a sample good day), August 24 (the second-worst day), July
29 (the third-worst day), and June 11 (the worst day).
We first compare the first two figures, Figures 12 and 13.
Each figure has three graphs. The first graph shows the
percentage of arriving flights that experience a non-zero gatehold delay, grouped by hour of the day according to the actual
wheels-on time of the arriving flight (method 1b). On the
good day (Figure 12), the gate-hold percentage is typically
around 20% or less. There are three mild peaks in the morning,
afternoon, and evening. In contrast, on the bad day (Figure 13),
the gate-hold percentage climbs steadily throughout the day to
ATL gate-hold delay by aircraft type
values near 100%. The morning peak at 9am is still visible. In
addition, a number of late arriving flights from the previous
day contribute to a peak shortly after midnight.
The second graph shows a comparison of scheduled arrivals with actual arrivals (BTS data). (The scheduled arrivals
correspond to arrivals at the gate while the actual arrivals
correspond to arrivals at the runway.) On both days, the
schedule has peaks roughly at 8am and 7pm, with perhaps
an intermediate peak around 4pm. These peaks correspond
with the three peaks in the gate-hold percentage graph on the
good day (Figure 12). The main difference between the two
days is that on the good day, the actual operations roughly
follow the scheduled operations; on the bad day, there is a
significant schedule disruption at 7pm, right at the evening
peak. In particular, the arrival rate drops to near zero during
this hour. The incoming flights are delayed and the peak is
shifted later by about three hours.
The third graph shows the turnaround delay experienced by
each aircraft, by time of day (method 2). Turnaround delay
is defined as the actual turnaround time minus the scheduled
turnaround time. (Based on this definition, it is possible for the
turnaround delay to be negative. This typically occurs when
an aircraft arrives late, but leaves on time. In particular, an
aircraft that stays overnight, arriving late, but departing ontime the next morning can have a large negative turnaround
delay.)
On the good day (Figure 12), turnaround delays are typically
no more than ±60 minutes, though there are some aircraft
that experience larger deviations from the schedule. On the
bad day (Figure 13), the situation is somewhat different. First,
there are significantly more aircraft with turnaround delays
that exceed 60 minutes. Second, there is a gap in the arrival
process around 7-8pm (consistent with the schedule disruption
discussed previously). These effects are related.
Considering that both major carriers, Delta and AirTran,
operate their flights in a hub and spoke network, this type of
network relies heavily on predictability of its schedule so that
crew and passengers can transfer efficiently between the arrival
and departure banks. A disruption during this time can cause
significant problems. For example, if half of the flights in the
bank are delayed, then some of the flights that have arrived
may wait for the delayed flights in order to establish continuity
of passengers and crew. This causes the aircraft to remain at
June 5, 2007 (sample good day)
Gate-Hold Percentage
Arrivals
100%
120
80%
100
Turnaround Delay
Scheduled gate
arrivals
(
(operations
ti
/ hr)
h)
Actual landings
240
120
80
60%
60
0
40
-120
20%
20
-240
0%
0
40%
0
3
6
9
12
15
18
21
( i )
(min)
360
-360
24
0
3
6
9
Hour of day (actual wheels on time)
12
15
18
21
0
24
3
6
Fig. 12.
9
12
15
18
21
0
21
0
6
9
12 15 18 21
Hour of day (actual arrival time)
0
Hour of day (actual arrival time)
Hour of day
ATL gate-hold delay on 6/5/2007 (a sample good day)
August 24, 2007 (2nd worst day)
Gate-Hold Percentage
100%
120
80%
Scheduled Gate
Arrivals
Actual Landings 360
Arrivals
(
(operations
i
/ hr)
h)
100
240
80
120
60%
60
0
40
-120
20%
20
-240
0%
0
40%
0
3
6
9
12
15
18
21
Turnaround Delay
(min)
-360
0
24
3
6
9
12
15
18
21
0
24
3
Hour of day
Hour of day (actual wheels on time)
Fig. 13.
6
9
12
15
18
Hour of day (actual arrival time)
ATL gate-hold delay on 8/24/2007 (the second-worst day)
July 29, 2007 (3rd worst day)
Gate-Hold Percentage
Scheduled Gate
Arrivals
Arrivals
((operations
p
/ hr))
Turnaround Delay
100%
120
80%
100
240
60%
80
120
40%
20%
0%
60
0
40
-120
20
-240
0
0
3
6
9
12 15 18 21
Hour of day (actual wheels on time)
24
0
Fig. 14.
3
6
(min)
Actual Landings360
9
12 15
Hour of day
18
21
-360
0
24
3
ATL gate-hold delay 7/29/2007 (the third-worst day)
June 11, 2007 (worst day)
Gate-Hold Percentage
Arrivals
Turnaround Delay
(operations / hr)
100%
120
80%
100
240
80
120
60%
60
0
40
Scheduled gate
arrivals
-120
20%
20
Actual landings
240
-240
0%
0
40%
0
3
6
9
12
15
18
21
24
Hour of day (actual wheels on time)
-360
0
3
6
9
12
15
18
21
24
Hour of day
Fig. 15.
(min)
360
ATL gate-hold delay 6/11/2007 (the worst day)
0
3
6
9
12
15
18
Hour of day (actual arrival time)
21
0
30
Cancellations
16
(count / hr)
Arrival
Cancellation
10
8
2007-07-29 (bad day)
2007-08-24 (bad day)
15
Departure
Cancellation
6
2007-06-11 (bad day)
Cancellations 25
per hour
(arrivals and
departures) 20
14
12
2007-06-05
2007
06 05 (good day)
10
4
2
5
0
0
3
6
9
12
15
18
21
0
24
0
3
6
Hour of day
Fig. 16.
Fig. 17.
ATL cancellations on 6/11/2007
60%
their gates longer, thereby reducing the number of available
gates and contributing to gate-hold delay. Disruptions can also
require crew changes which can delay aircraft at their gates.
In summary, the good day (June 5) is characterized by
a close adherence to the schedule without much disruption.
The bad day (August 24) is characterized by a significant
schedule disruption, leading to long turnaround times. The
long turnaround times limit gate availability and contribute
to gate-hold delays.
Now we consider the other two bad days at ATL: July 29
(the third-worst day) and June 11 (the worst day). From a
qualitative perspective, July 29 (Figure 14) is very similar
to August 24. Specifically, there is a significant drop in
arrival capacity in the late afternoon (this time around 4-6pm).
This leads to significant deviations from the schedule where
the disrupted arrival bank comes in several hours after its
planned arrival time. This disruption leads to large increases
in turnaround times, which in turn leads to large gate-hold
delays.
The system behavior on June 11 (the worst day, Figure 15),
on the other hand, is qualitatively different than the other two
bad days. In particular, June 11 does not appear to have a
significant schedule disruption, except for perhaps a modest
drop in arrivals around 7-8pm. One possible explanation is
the number of cancellations (Figure 16, BTS data). This
graph shows the number of arrival and departure cancellations
by hour of the day. There is a large spike in departure
cancellations around 5pm, followed by a large spike in arrival
cancellations around 8pm. When departure flights are cancelled, the airplanes remain on the ground at the gate. Because
the departure cancellations precede the arrival cancellations,
there is a period of time when there are more airplanes on the
ground, but the arrival rate into the airport has not yet been
reduced, leading to a shortage of gates.
In a similar manner, Figure 17 shows the hourly cancellation rate for the four focused days. The cancellation rate
is low on June 5, 2007 when the gate-hold performance is
good. The cancellation rate is high on the three worst days,
especially from 3-10pm, coinciding with the periods of high
gate hold. Cancellations can have a negative effect on gate
9
12
15
Hour of day
18
21
24
ATL cancellations on 4 focused days
12000
Estimated gate-hold
percentage
50%
10000
40%
8000
30%
6000
20%
Estimated daily gate-hold
delay in minutes
6/10
6/11
7/3
6/1
10%
4000
2000
0%
0
6/1
6/11
6/21
7/1
7/11
7/21
7/31
8/10
8/20
8/30
Day of year
Fig. 18.
JFK gate hold by day of year
delay, because a cancellation grounds an airplane for a period
of time typically longer than it would otherwise remain at a
gate, thus reducing overall gate capacity. In another words,
flight cancellations reduce the number of active flights. This
decreases the overall fleet’s airborne time, which increases the
overall fleet’s ground time, which has the potential to increase
gate delay.
Moreover, departure cancellations often come before arrival
cancellations, as in Figure 16. When a flight scheduled to
depart late is cancelled, its aircraft may be reassigned to a
later departure whose incoming flight is scheduled to arrive
later and is also cancelled. This reduces flight delay by
providing additional robustness in the schedule, but creates
longer turnaround times which reduces the number of available
gates.
Although the three worst gate-delay days have high cancellation rates, we have also observed many days with high
cancellation rates and low gate-delay days. Thus, while high
cancellation rates seem to occur on the worst days, they do
not necessarily imply high gate delays.
D. JFK
Figure 18 shows the daily gate-hold delay and daily gatehold percentage at JFK for each day in the summer of 2007
(similar to Figure 8 for ATL, using method 1a). The total
gate-hold delay is very high on a few isolated days. The three
highest days are June 1, June 10, and June 11.
6/1/2007 (bad day)
7/3/2007 (good day)
60
60
50
50
40
40
30
30
20
20
Arrival Count
10
0
20070703
Gate Hold Count
0
23 1 3 5 7 9 11 13 15 17 19 21 23 1
200707
02
Arrival Count
10
Gate Hold Count
0 3 5 7 9 11 13 15 17 19 21 23 1
20070704
20070601
Actual wheels on date and hour
6/10/2007 (bad day)
60
20070602
Actual wheels on date and hour
6/11/07 (bad day)
60
50
50
40
40
30
30
20
Arrival Count
10
Gate Hold Count
0
20
Arrival Count
10
Gate Hold Count
0
23 1 3 5 7 9 11 13 15 17 19 21 23 1
23 1 3 5 7 9 11 13 15 17 19 21 23 2
200706
09
20070610
20070611
2007061
0
Actual wheels on date and hour
20070611
20070612
Actual wheels on date and hour
Fig. 19.
JFK gate hold by time of day
Figure 19 shows a more detailed picture of the three worst
days, as well as a sample “good” day, July 3. The figure shows
the total arrival count (by hour) and the total number of arrivals
(by hour) that experience gate hold. On the bad days, almost
all flights experience gate hold between 5pm and midnight.
Figure 20 shows the actual wheels-on-time profile for the
four days (ASPM data). There is a close match until about
4pm. After that, there are significant differences between the
four days. The actual wheels-on profile on the good day is
similar to the schedule (not shown), while the actual wheelson profile on the bad days is typically lower than the schedule
between 5pm to midnight. This is the key difference between
a good day and a bad day: The schedule disruption is more of
a problem than the volume of traffic because (i) the volume of
traffic from 5pm to midnight in a good day is no worse than
that in a bad day, and (ii) gate hold is not a problem in the
arrival peak hour, 3pm. This demonstrates that the wheelson delay is a reason that the same schedule can perform
differently.
Table IV compares the gate-hold percentage among the
major carriers at JFK – JetBlue Airways, Delta Air Lines,
and American Airlines. There are statistically significant differences between each pair of carriers. (In all cases, the
assumption that the carriers have equal gate-delay percentages
is rejected if the t-statistic is greater than 1.96 in absolute
value.) The differences illustrate that the individual carriers
have different levels of over-scheduling relative to the number
of gates. A higher fraction of common gates may reduce gate
hold.
TABLE IV
JFK GATE HOLD DELAY BY CARRIER
Group
JetBlue vs. Delta
JetBlue vs. American
Delta vs. American
t-statistic
29.77
4.91
24.89
Difference
Significant
Significant
Significant
VI. C ONCLUSIONS
This paper developed several methods to approximate gatehold delay using data from the ASPM and BTS databases. We
applied the methods to the OEP-35 airports using data from
the summer of 2007.
Out of the OEP-35 airports, the analysis identified 11
airports with one or more days in which 30% or more of
arriving flights experienced a gate delay. The three worst
airports (based on this metric) were ATL, JFK, and DFW.
Specifically, they experienced a daily gate-hold percentage of
30% or more on 50, 38, and 28 days, respectively (out of
92 days). Airports with moderate gate-hold delays were PHL,
LAX, DTW, MIA, and MSP. Airports with light gate-hold
delays were EWR, IAH, and LGA. Other airports did not
experience any significant gate-hold problems. Out of these
11 airports, the average daily gate hold ranged from 904 to
3670 minutes.
Then, we applied a more detailed analysis to ATL and
JFK, the two worst airports. On most days, gate hold was not
an excessive problem. However, on a few isolated days, the
problem was very bad. We examined hour-by-hour statistics
for the worst days and compared them with typical “good”
days. One common behavior of the worst days was the
figGateHoldJFKWheelsOnDelayIssue
60
R EFERENCES
20070601 (bad day)
50
20070610 (bad day)
20070611 (bad day)
20070703 (good day)
40
Wheels-on
30
Count (actual)
20
10
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
H
Hour
off dday
Fig. 20.
JFK wheels on time profile
disruption of the schedule, leading to less efficient operations,
including increased cancellations, longer turnaround times
and position times, resulting in more arriving flights than
open gates. Schedule disruptions appeared to be more of a
significant factor on the worst days than the absolute number
of arrivals (as one might expect).
For both ATL and JFK, the worst delays occurred during the
evening hours. Also, at both airports, there were statistically
significant differences between the gate delays experienced
by the major carriers. This suggests that more common gates
could improve gate-hold delay. Aircraft type was also somewhat correlated with gate-hold delay at both airports. That is,
larger aircraft experience slightly less gate delay, on average.
Overall, issues identified related to gate-hold delay included
the gate assignment policy, gate-carrier compatibility (common gates), wheels-on-time delay, number of gates, cancellations, aircraft swapping, minimum service time, service-time
delay, and aircraft type. Many of these issues are related to
schedule disruptions which are the main common factor identified in the worst days. Other issues are strategic problems, such
as the common-gate issue. Future work will study whether
the unexpected events found in this paper can be taken into
account in the decision making process and used to mitigate
delays.
ACKNOWLEDGMENTS
This research was funded by grants from the NASA
Aeronautics Program, the FAA, and the internal Center
for Air Transportation Systems Research Foundation. The
authors appreciate technical contributions from Tony Diana, Akira Kondo, Stephanie Chung, Midori Tanino (FAA),
Maria Consiglio, Brian Baxeley, Kurt Nietzche (NASA),
Chris Smith, Ben Levy, Jeff Legge, George Hunter (Sensis), Norm Fujisaki, Terry Thompson, Mark Klopfenstein
(Metron Aviation), Ed Stevens, Mary-Ellen Miller (Raytheon),
Guillermo Calderon-Mesa, Vivek Kumar, John Ferguson, Abdul Kara, George Donohue, Karla Hoffman, and Rajesh Ganesan (CATSR/GMU).
[1] D. Gross, J. F. Shortle, J. M. Thompson, and C. M. Harris, Fundamentals
of Queueing Theory, 4th ed. John Wiley & Sons, Inc, 2008.
[2] M. Bazargan, Airline Operations And Scheduling. Ashgate Publishing
Ltd., 2004.
[3] A. Bolat, “Procedures for providing robust gate assignments for arriving
aircrafts,” European Journal of Operational Research, vol. 120, pp. 63–
80, 2000.
[4] A. Haghani and M.-C. Chen, “Optimizing gate assignments at airport
terminals,” Transportation Research Part A: Policy and Practice, vol. 32,
pp. 437–454, 1998.
[5] S. Yan and C.-M. Huo, “Optimization of multiple objective gate assignments,” Transportation Research Part A: Policy and Practice, vol. 35,
pp. 413–432, 2001.
[6] A. Bolat, “Assigning arriving flights at an airport to the available gates,”
The Journal of the Operational Research Society, vol. 50, pp. 23–34,
1999.
[7] S. Yan and C.-M. Chang, “A network model for gate assignment,”
Journal of Advanced Transportation, vol. 32, pp. 176–189, 1998.
[8] S. Yan and C.-H. Tang, “A heuristic approach for airport gate assignments for stochastic flight delays,” European Journal of Operational
Research, vol. 180, p. 547567, 2007.
[9] Y. Gu and C. A. Chung, “Genetic algorithm approach to aircraft gate
reassignment problem,” Journal of Transportation Engineering, vol. 125,
pp. 384–389, 1999.
[10] S. Yan, C.-Y. Shieh, and M. Chen, “A simulation framework for
evaluating airport gate assignments,” Transportation Research Part A:
Policy and Practice, vol. 36, pp. 885–898, 2002.
[11] U. Dorndorf, F. Jaehn, C. Lin, H. Ma, and E. Pesch, “Disruption
management in flight gate scheduling,” Statistica Neerlandica,
vol. 61, no. 1, pp. 92–114, 2007. [Online]. Available: http:
//ideas.repec.org/a/bla/stanee/v61y2007i1p92-114.html
[12] J. Sarkis, “An analysis of the operational efficiency of major airports
in the united states,” Journal of Operations Management, vol. 18, pp.
335–351, 2000.
[13] D. Gillen and A. Lall, “Developing measures of airport productivity and
performance: an application of data envelopment analysis,” Transportation research. Part E, Logistics and transportation review, vol. 33, pp.
261–274, 1997.
[14] K. Andersson, F. Carr, E. Feron, and W. D. Hall, “Analysis and
modeling of ground operations at hub airports,” in 3rd USA/Europe Air
Traffic Management R&D Seminar, Napoli, 13-16 June 2000, 2000.
[Online]. Available: http://hdl.handle.net/1721.1/37315
[15] H. R. Idris, I. Anagnostakis, B. Delcaire, R. J. Hansman, J.-P. Clarke,
E. Feron, and A. R. Odoni, “Observations of departure processes at
logan airport to support the development of departure planning tools,”
Air Traffic Control Quarterly, vol. 7, pp. 229–257, 1999. [Online].
Available: http://ntrs.nasa.gov/
[16] P. C. Roling and H. G. Visser, “Optimal airport surface traffic planning using mixed-integer linear programming,” International Journal of
Aerospace Engineering, vol. 2008, 2007.
[17] H. Fricke and M. Schultz, “Improving aircraft turn around reliability:
Specific aircraft body design parts hamper ground handling and airport
performance,” in Third International Conference on Research in Air
Transportation, 2008, pp. 335–343.
[18] H. R. Idris, “Observation and analysis of departure operations at boston
logan international airport,” Ph.D. dissertation, Massachusetts Institute
of Technology, 2001.
[19] BTS, “Airline on-time performance and causes of flight delays,” 2003.
[Online]. Available: http://www.bts.gov/help/aviation/index.html
[20] FAA, “Data reference guide,” 2008. [Online]. Available: http:
//aspm.faa.gov
[21] R. T. Ramos, “Airtran doesn’t control new gates, airport says,”
Atlanta Business Chronicle, February 2007. [Online]. Available:
http://atlanta.bizjournals.com/
[22] FAA, “OEP frequently asked questions - OEP 35 airports,” July 2007.
[Online]. Available: http://www.faa.gov/