FLIGHT TIME COMPONENTS AND THEIR DELAYS ON US DOMESTIC ROUTES
Gerasimos Skaltsas
G
i
Sk l
Dr. Peter Belobaba
Prof. Amy Cohn
MIT GLOBAL AIRLINE INDUSTRY PROGRAM
IAB/Industry Consortium Joint Meeting
IAB/Industry Consortium Joint Meeting
November 4, 2010
Database
D t b
Database




FAA Aviation System Performance Metrics
detailed flight
g data for 77 Airports
p
80% of all scheduled domestic commercial flights in 2009
88% of all domestic enplanements (based on the T‐100 Domestic Segment )
Final analyzed sample
 2276 directional non‐stop routes between 72 ASPM Airports
 operated by 40 US carriers in daily basis
 54% of the all scheduled domestic commercial flights in 2009.
2
Carrier‐Routes in the Sample
1600
Nu
umber of routes
1400
1336
59%
1200
1000
599
26%
800
250
11%
600
400
200
62
3%
27
1%
4
5
0
1
2
3
Number of carriers
2
6

3679 distinct directional route‐carrier pairs.

59% of the routes are served only by one carrier, and 26 % are served by
two In the remaining 15% operate more than 2 carriers.
two.
carriers

We expect that competition affects the schedule padding practices,
because the presence of the carriers in the ticket distribution systems
depends often on their scheduled block time.
time
3
Schedule Padding by Time of Day
(ORD – JFK, 2009)
(ORD JFK, 2009)
Schedule
ed Block Time (m
minutes)
160
150
25 min
140
130
120
110
100
09‐10
10‐11
12‐13
13‐14
14‐15
15‐16
16‐17
17‐18
18‐19
19‐20
20‐21
21‐22
22‐23
23‐24
Scheduled Arrival Time
 Airlines lengthen the scheduled block times of their flights to improve
• the reliability of their schedules
• their “on‐time
on time arrival
arrival” statistics (a flight that arrives at the gate more
than 15 minutes later than scheduled is considered delayed)
 The time that airlines add to their schedules is called BUFFER.
 BUFFER = Scheduled Arrival Time – Delay‐Free Arrival Time.
4
Understanding the
Measurement of Delay
Measurement of Delay
5
Variability in Gate Delays
Newark (EWR) – Los Angeles (LAX), 2009
25
10%
Relative Frequency
6%
17.4
5
41.2
Gate Delaay (minutes)
Average:
Median:
St. Deviation:
8%
4%
2%
20
15
10
5
0
0%
‐20
‐10
0
10
20
30
40
50
60
70
JAN FEB MAR APR MAY JUNE JULY AUG SEP OCT NOV DEC
Gate Delay (minutes)
Gate Delay = Actual Gate Departure – Scheduled Gate Departure Time
Causes:
•
•
•
•
Airport congestion at origin or destination (ceiling, visibility, winds)
Airspace congestion Propagated delay (aircraft, crew)
Airline operations ( boarding, catering, fueling etc.)
6
Variability in Taxi‐out Delays
Newark (EWR) – Los Angeles (LAX), 2009
14.8
Average:
Median: 10.8
St. Deviation: 16.9
8%
6%
Taxi ‐ out Tim
me (minutes)
Relative
e Frequency
10%
4%
2%
0%
‐20
‐10
0
10
20
30
40
50
60
70
45
40
35
30
25
20
15
10
5
0
JAN FEB MAR APR MAY JUNE JULY AUG SEP OCT NOV DEC
Unimpeded Taxi‐out
Average Taxi‐out
Taxi‐out Delay (minutes)
Taxi‐out Delay = Actual Taxi‐out Time – Unimpeded Taxi‐out Time
Causes: • Airport congestion at origin or destination (ceiling, visibility, winds)
• Airspace congestion • Delays during pushback
y
gp
7
Variability in Taxi‐in Delays
Newark (EWR) – Los Angeles (LAX), 2009
15
Average:
2.5
1.6
Median:
St. Deviation: 5.0
12%
Relative Frequency
Taxi ‐ in TTime (minutes)
14%
10%
8%
6%
4%
2%
0%
10
5
0
‐20
‐10
0
10
20
30
40
Taxi‐in Delay (minutes)
50
60
70
JAN FEB MAR APR MAY JUNE JULY AUG SEP OCT NOV DEC
Unimpeded Taxi‐in
Average Taxi‐in
Taxi‐in
Taxi
in Delay = Actual Taxi
Delay = Actual Taxi‐in
in Time Time – Unimpeded Taxi
Unimpeded Taxi‐in
in Time
Time
Causes: • Weather conditions at airport of destination, Runway configuration
• Gate unavailability
8
Variability in Airborne Delays
Newark (EWR) – Los Angeles (LAX), 2009
360
10%
Relative Frequency
6%
Airborne TTime (minutes)
Average:
22.2
Median: 20
St. Deviation: 19.2
8%
4%
2%
0%
‐20
‐10
0
10
20
30
40
50
60
70
Airborne Delay based on Nominal Flight Time (minutes)
350
340
330
320
310
300
290
JAN FEB MAR APR MAY JUNE JULY AUG SEP OCT NOV DEC
Nominal Airborne Time
Average Airborne Time
Airborne Delay = Actual Airborne Time – Nominal Airborne Time
Actual Flight Time = Wheels On – Wheels Off
Nominal Airborne Time = 10th percentile of Actual Airborne Time Distribution
Causes: • Airspace congestion
• Winds (strength and direction)
9
Variability in Block Delays
Newark (EWR) – Los Angeles (LAX), 2009
Blockk Time (minutes)
Relatiive Frequency
10%
Average:
‐3.8
Median: ‐6
St. Deviation: 24.1
8%
6%
4%
2%
0%
‐70
‐50
‐30
‐10 0 10
30
Block Delay (minutes)
50
70
400
390
380
370
360
350
340
330
320
JAN FEB MAR APR MAY JUNE JULY AUG SEP OCT NOV DEC
A
Average Scheduled Block Times S h d l d Bl k Ti
A
Average Actual Block Times A
l Bl k Ti
 Huge variability in block delays
 The average block delay is very close to zero (most often negative) The average block delay is very close to zero (most often negative)
 The average block delay is larger than the median, because there is a small percentage of flights with excessive delays that moves the average to the right.
 The block delay distribution is almost symmetric around the median.
10
Percentaage of route‐carrrier pairs
Distribution of Block Delays
((January 2009)
a ua y 009)
8%
7%
6%
5%
4%
3%
2%
1%
0%
Average Block Delay (minutes)
Average Block Delay (minutes)




74% of the 3679 studied route‐carrier pairs have negative average block delays
20% have average block delays between zero and five minutes,
only
l 6% have
h
d l
delays
l
longer
than
h five
fi minutes
i
In absence of gate delays, 92% of the flights would arrive on time (with gate
delays 80%)
 Flights on most routes suffer lengthy gate delays (shown next),
next) that are difficult
to be predicted. This results to significant arrival delays
11
Percentage of route‐carrrier pairs
Correlation between Block Delay and Delay Components
y
p
35%
Taxi Out and Block Delay
Airborne and Block Delay
Taxi In and Block Delay
Taxi In and Block Delay
30%
25%
20%
15%
10%
5%
0%
Pearson product‐moment correlation coefficient
 Very strong positive correlation between block delays and taxi‐out delays
– Taxi‐out delays responsible for the biggest portion of the block delays
 Strong positive correlation between block delays and airborne delays
gp
y
y
– Smaller than the correlation of the taxi‐out delays
 Small correlation between block delays and taxi‐in delays
– Taxi‐in delays are usually very small compared to the taxi‐out and the airborne delays
12
Arrival and Gate Delays
Gate Delay
ATL ‐ DFW
Arrival Delay
15
Gate Delay (minutes))
20
D
Delay (minutes)
20
10
15
10
5
5
0
0
‐5
18
16
14
12
10
8
6
4
2
0
Gate Delay
FLL ‐ ATL
Arrival Delay (min
nutes)
25
Arrival Delay
13
Percentagge of route‐carriier pairs
Distribution of Gate Delays
(January 2009)
(January 2009)
8%
7%
6%
5%
4%
3%
2%
1%
0%
Average Gate Delay in January 2009 (minutes)
Average Gate Delay in January 2009 (minutes)
 The average gate delays (Jan 09) of 17% route‐carriers are longer than 15 min
 The annual range of the monthly average gate delays can be very large and
there is no seasonality
 It is very difficult for airlines to predict the gate delays
 We expect that this will result to a strong correlation between the gate delay
and the arrival delay
14
Percentage of route‐ccarrier pairs
Strong Correlation between Gate and Arrival Delays
Gate and Arrival Delays
1573
45%
40%
35%
30%
25%
713
20%
15%
317
10%
5%
842
38
67
129
0%
[0.3, 0.7) [0.7, 0.75) [0.75, 0.8) [0.8, 0.85) [0.85, 0.9) [0.9, 0.95) [0.95, 1)
Pearson product‐moment correlation coefficient
 Very strong positive correlation
 A delayed pushback results most often to a delayed arrival
 Gate delays are caused mostly by stochastic factors, such as propagated
delays (aircraft, crew) and airline operations (boarding, catering, fueling
etc.). Therefore, airlines can not predict them and schedule without taking
into account the
h variability
b l in gate delays
d l
15
Taxi‐o
out Time (minute
es)
Destination can affect the Taxi out Times
45
40
35
30
25
20
15
10
5
0
ORD – LGA
Unimpeded Taxi‐out
ORD – MCO
Average Taxi‐out
 Flights from the same airport (ORD) and by the same carrier (American
Airlines) but with different destinations (LGA and MCO) do not follow the same
distribution of taxi‐out times
The annual range of monthly average taxi‐out times is
• 11.3 minutes on the ORD‐LGA route
• 2.6 minutes on the ORD‐MCO route
 We expect that this happens due to the existence of Ground Delays Programs
that hold the flights that are destined to LGA on the ground
16
45
Taxi‐ou
ut Time (minute
es)
40
ORD – LGA by AAL
35
70%
60%
50%
30
25
40%
20
30%
15
20%
10
5
10%
0
0%
Unimpeded Taxi‐out
Unimpeded Taxi
out
Percenttage of flights th
hat were under a Ground Delay Program
Destination can affect the Taxi out Times
Average Taxi‐out
g
 The taxi‐out times of the flights from ORD to LGA are affected by the GDPs
 Mayy and June have the highest
g
average
g taxi‐out times and the highest
g
percentage of flights that were held on the ground by GDPs
 September has the smallest average taxi‐out times and the smallest
percentage
p
g of held flights
g
 Only 3 flights that were destined to MCO were held in 2009
17
Buffer (minute
es)
Buffer in Schedule increases with Distance
with Distance
January 2009
100
90
80
70
60
50
40
30
20
10
0
y = 0.0878x + 27.387
R² = 0.3767
0
50
100
150
200
Nominal Airborne Time (minutes)
250
300
350
400
 Each point corresponds to the average buffer time and nominal airborne
time for a route‐carrier pair in January 2009
 Some linearityy between buffer and distance.
 Carriers hide more delays in the long‐haul flights to handle the increased
uncertainty in the airborne times (winds).
 We expect that the extensive padding of some short‐haul
short haul flights
compensates the gate delays and the taxi‐out, taxi‐in delays rather than the
airborne delays.
18
Conclusions
 The average block delay is most of the times negative, and close to zero.
Taxi out delays,
Taxi‐out
delays airborne delays and taxi‐in
taxi in delays are very effectively hidden
in the schedule
 Very strong Correlation between gate delays and arrival delays. Carriers
schedule without taking into account the variability in gate delays because
its difficult to predict them
 Large seasonality in taxi‐out and airborne times. This makes necessary a
month basis analysis of the schedule padding practices
 Limited seasonality and variability in taxi‐in times
 The congestion in the arrival airport affects the gate delays and the taxi‐
out times through the Ground Delay Programs
 Linearity between buffer and distance. Long‐haul flights have in average
larger delays than short‐haul flights
19
Future Research
1.
2.
For two selected months , use
• linear regression methods • non‐parametric regression trees
non parametric regression trees
to study the relationship between the buffer time and the
• fight components • distance • ground hold times
• time of the day
• route competition • carrier type (LCC vs. NLC)
Measure quantitatively and qualitatively the benefits and the costs of
padding for the carriers and the airports.
For airlines:
• Crew costs
• Utilization costs
• Recovery costs/ delay propagation
• Presence in ticket distribution systems
• Reliability – on time performance For airports:
• Delays
y
• Level of Service
• Traffic
• Revenues
• Cost for investments in terminals, runways, ATC technologies
20