The Derivation and Analysis of The Passenger Peak Hour: An Empirical Application To Brazil
The Derivation and Analysis of The Passenger Peak Hour: An Empirical Application To Brazil
The Derivation and Analysis of The Passenger Peak Hour: An Empirical Application To Brazil
to Brazil
P.T.Wang
Infraero, SCS, Quadra 04, Bloco A, N0 58, Edifício Infraero, Brasilia - D F, CEP 70.300-
500, Brazil.
D.E.Pitfield,
Abstract
Many different definitions are currently used to define a design ‘peak’ hour at airports,
such that the majority of passengers receive adequate service levels and only an acceptably
small proportion experience the impact of congestion. The evaluation of level of service
provision depends on this definition. Previous definitions have either used a nominal hour
below the absolute peak or have taken a representative hour from the schedule of flights.
This paper endeavours to define this level of traffic empirically. Data for 48 Brazilian
determine the appropriate peak hour for designed decisions. This empirically derived
traffic level is regressed against annual traffic levels to establish a relationship that should
enable the forecast of design peak hour demand with changing annual throughputs.
1 Introduction
A variety of definitions are used to define the peak hour at airports. It is not the objective
to provide facilities that result in a perfect match of unrestrained service levels for this
peak as this may result in a waste of resources at other times. There is a consensus that
planning should aim to satisfy demand at some level below this peak, with the result that
the majority of passengers receive adequate service levels and only a small proportion
experience the impact of congestion. Planning for this traffic level, for example in the
design of passenger terminals, also means that the evaluation of the level of service
There have been few previous studies designed to empirically derive or confirm the
forecast levels of peak hour demand for British Airport Authority (BAA) airports by
correlating them with hourly, monthly and day of the week patterns and in Brazil a number
of studies have been prompted by the realisation that many airports were designed for too
high a level of peak flows, thus wasting resources, and that a lack of consistency meant
that service provision standards could not be consistently appreciated by the public.
Initially, Andrade (1993) attempted to address this problem, but data shortcomings
prevented his work reaching fruition. In this paper, an improved data-base is used and a
methodology to identify the design peak hour is suggested, applied and tested.
2
2 Passenger Peak Hour Definitions
The definition of the passenger peak hour is fundamental for a facility’s design purposes.
It is not sensible to use the highest hourly demand of the year for design purposes because
level of demand at prevailing pricing policies that the capacity of the airport can handle
with an acceptable level of service, with just a few hours of operation during the year in
There are many definitions of the peak hour for design purposes that are in use by airport
authorities. All of them try to define an acceptable portion of the users that will receive
not less than an adequate level of service during one year of operation. The main ones are
described in Ashford and Wright (1992). The standard busy rate (SBR), in the past used
by BAA in the UK, is the 30th highest hour of annual passenger flow, or the rate of flow
SBR is the busy hour rate (BHR) where the hours concept is replaced by the highest 5
percent of the annual traffic. Both methods use the same principle of ranking all the hours
of the year in decreasing order of passenger flow and after that select a fraction (hours or
percentage) when it is considered that some degree of congestion can be expected and
tolerated.
3
The Federal Aviation Administration (FAA) in the US uses typical peak hour passengers
(TPHP) where values are related to annual throughputs (Federal Aviation Administration,
1976). These are shown in Table 1. The busiest timetable hour (BTH), based on timetables
and load factors of flights, can be used for small airports where there is little historical data
available.
The peak profile hour (PPH), also used by the FAA, is the highest hourly passenger flow
in the average peak day of the peak month. This measure has been used by the Brazilian
Civil Aviation Authority. For many larger airports the PPH is close to the SBR. This is not
the case with the Brazilian domestic airports examined here where it can range from the
busiest to the 100th busiest hour because the level of demand is low on average and is
In practice, while the International Civil Aviation Organisation (1977) states that it is
common practice to use the 30th or 40th busiest hour of the year, Dutch airports use the
6th busiest hour and the Aeroport de Paris uses 3 percent overload standard. In fact, the
relationship between demand and capacity depends not only on the peaking characteristics
4
but also on the level of service to be provided at the nominal peak. The overall level of
service may be adjusted according to the chosen design hour and its level of service. The
essence of this choice of design hour is that it be representative of some relatively stable
condition1.
3.1 Data
Airlines using Brazilian airports are required to provide throughput statistics to the
Departamento de Aviação Civil (DAC). This information is gathered for regional, national
and international airlines. The reliability of the level of information, however, and the ease
of access differs for each sector as does the degree of currency of the information. This
study uses domestic traffic data for the national airlines sector comprising, VARIG,
Data is available for these airlines for all 48 airports owned by Infraero which handled
domestic traffic of these carriers for period 1990-1994. Information relates to daily
the data but it is felt that the integrity of the subsequent analysis is unaffected2. From this
1
The level of service and its relation to capacity is dealt with in Mumayiz (1985) and standards are dealt
with in Airports Associations Co-ordinating Council and International Air Transport Associatin (1990).
2
The main difficulties are, first, that for 18 of the 48 airports, regional airlines have a market share of
over 20 percent, so the peak hour derived from national airlines data may be irrelevant for these airports.
However, in the pooled analysis, these airports are not important. Second, aggregated data from these
tapes does not match official airport statistics despite the same source of information, the airlines. See
Wang (1995) for full details.
5
data, for each airport, the 200 busiest hours were found for each year and the throughput
3.1 Methods
It has been suggested by Horonjeff and McKelvey (1994) that design peak hour traffic
should be related to annual throughput. If this is the case then not only is the definition of
the peak hour useful, but changes in annual throughput can allow it to be redefined using
Before a regression can be estimated, however, it is necessary to define the traffic level
associated with the design peak hour. The busiest hour will not be chosen because it
would waste resources and because abnormal factors may result in it reaching different
levels during different time periods. With such variability a useful definition of the peak
cannot be obtained nor will any attempt to explain traffic in terms of annual throughput be
successful. For each airport differences in the traffic levels for the 200 busiest hours
covering the five years study period can be observed. The question then is, ‘At what
empirical level does the lack of variability in peak hour traffic allow a reasonable and
useful definition of the design peak hour?’ Two methods are utilised to throw light on this.
Firstly, the standard deviation for each of the 200 busiest hours is calculated for each year
and divided by the mean traffic level for that hour. Plots of this for each of the 48 airports
are shown in Wang (1995) and the most useful are shown in section four below. Secondly,
6
the coefficient of determination (R2) is calculated for the same 5 year traffic levels against
the relevant annual throughput. This shows a stronger correlation implying the confidence
with which a design peak hour can be chosen, and then successfully regressed against
Figure 1plots the standard deviations for the 200 busiest hours of departing traffic for the
period 1990 to 1994, for São Paulo International Airport-Guarulhos. There is, after a
fluctuations at the beginning, a steady decline in this statistic until stability is reached about
the 80th busiest hour. Figure 2 shows the value of R2 for the same airport where again
stability is reached around the 80-100th busiest hour. Figure 3, shows the relationship
between the 200 busiest hours and the cumulative percentage of departing traffic for the 5
year period and it can be seen that the definition of the design peak hour suggests a
respectively the standard deviation, the R2 and the cumulative traffic-busiest hour
relationship. The standard deviation suggests a design peak hour as low as the 20th busiest
hour, but the R2 statistic suggests a more appropriate use of the 60th busiest hour. This
7
Figure 1: São Paulo International Airport-Guarulhos: Standard Deviation/Mean, 1990-4
0 .2 5
0 .2 0
0 .1 5
Standard Deviation/Mean
0 .1 0
0 .0 5
0 .0 0
0 50 100 150 200
8
1 .0 0
0 .9 0
0 .8 0
0 .7 0
0 .6 0
R squared
0 .5 0
0 .4 0
0 .3 0
0 .2 0
0 .1 0
0 .0 0
0 50 100 150 200
Figure 3: São Paulo International Airport-Guarulhos: Rank hour against BHR, 1990-4
9
9 .0 0
8 .0 0
7 .0 0
6 .0 0
Cumulative percentage (BHR) (%)
5 .0 0
4 .0 0
3 .0 0
2 .0 0
1 .0 0
0 .0 0
0 5 0 1 0 0 1 5 0 2 0 0
0.30
0.25
0.20
0.15
0.10
0.05
0.00
0 50 100 150 200
10
Figure 5: Rio de Janeiro Galeão International Airport: Coefficient of Determination, 1990-
1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
0 50 100 150 200
Figure 6: Rio de Janeiro Galeão International Airport: Rank hour against BHR, 1990-4
11
1 0 .0
9 .0
8 .0
7 .0
6 .0
5 .0
4 .0
3 .0
2 .0
1 .0
0 .0
R a nk B us y Hou r
Similar graphs for the other 46 airports (Wang, 1995), all of which are smaller, do not
yield such clear conclusions. Brasilia, the third largest airport, shows an R2 that rises
steadily over most of the range of the busy hours, exhibiting some stability at about the
80th hour, which is equivalent to about 5 percent BHR. However, the standard
deviation/mean statistic, after initial fluctuations, steadies at the 60th hour (equivalent to 4
percent BHR). The evidence for some smaller airports, in particular for SBCT, SBCY and
SBGO, shows that both statistics are stable over the whole range of the busy hours rank.
This is dude to a lack of variability in busy hour traffic determined by a sparse yet
predictable and peaked timetable with steady load factors. This lack of variability means
that the methodology advocated here is not necessary for such airports in determining the
design peak hour. In conclusion, it would be sensible, especially for larger airports
experiencing variability in their busiest hours, to allow the BHR to vary between 3.0 and
12
4.5 percent, with the lower figure appropriate for larger airports and the higher percentage
It can be established that regressing a design peak hour of 3.0-4.5 percent yields little
difference in the resulting equations and since a figure of 3.5 percent is indicated for the
main airports this is used in the remainder of this section. The pooled time series-cross
uses regression involving 236 data points as three airports had no reported traffic for some
of the years covered and were thus omitted. Equation 1 and Figure 7 shows the results for
a simple regression of design peak hour traffic (DPT) on annual throughput for departing
passengers (AT). The log of throughput (LAT)is added to the equation 2 and in Figure 8.
13
1400
1200
1000
800
600
400
200
Annual Throughput
Actual Linear
Figure 8: Regression of Design Peak Hour on Annual Departures Throughput and Log of
Throughput
1400
1200
1000
800
600
400
200
0
0 500000 1000000 1500000 2000000 2500000 3000000
Annual Throughput
Actual Non-Linear
In both sets of results the size and sign of the coefficients are as expected and they are
statistically significant. The efficiency of the model is checked by stratifying the sample of
14
airports to look at smaller airports and at the sample without the biggest airports. These
Annual Embarkation throughput less All airports, with the exception of the
than top
30,000 100,000 200,000 300,000 3 2 0
Annual Embarkation throughput less All airports, with the exception of the
than top
30,000 100,000 200,000 300,000 3 2 0
With the exception of the results for the smallest airports, both models give a reasonable
level of explanation, although the robustness of the simple linear model seems stronger as
the slope coefficient appears reasonably stable. Table 4 shows the simple linear model of
the design peak hour for arriving passengers. Although smaller airports are again not well
explained, the level of explanation for airports with throughputs of 200,000 per annum or
more are reasonable. It is noteworthy, that the slope coefficient seems to be different for
15
the sample including the largest airports and excluding the largest two or three from the
Annual Disembarkation throughput less All airports, with the exception of the
than top
30,000 100,000 200,000 300,000 3 2 0
Finally, Table 5 reports the application of the same simple model to total traffic. Again,
explanation is poor for smaller airports and there are marked differences in the slope
Total Annual throughput less All airports, with the exception of the
than top
60,000 250,000 500,000 1,000,000 3 2 0
16
6 Conclusions
This paper has adopted an empirical approach to the determination of the design peak
hour at airports. Using data on departing domestic traffic for 48 airports in Brazil shows
that the busiest hours have different traffic levels for the five years studied. Consequently,
it seems to make little sense to attempt to design for such peaks but rather to identify a
busy hour that has more predictabl traffic levels. Statistics have been used to determine
this design peak hour in terms of the Busy Hour Rate (BHR). The evidence for the larger
airports indicates that this is between 3.5 and 4.5 percent with percentages at the top of
This conclusion has been examined using a regression analysis of the chosen design peak
hour on annual throughput. It is noted that design peak hours in the range suggested
produce virtually identical results, so 3.5 percent is adopted as the design peak hour in the
subsequent work. This shows that annual throughput gives a good explanation of the
design peak hour. Disaggregating the results by airport size suggests a simple linear model
Examining a linear model of arriving traffic supports the design peak hour derivation but
offers a poorer explanation for small airports although a better one for larger airports.
These results are mirrored when the methodology is applied to total traffic.
17
References
Graduacao.
Ashford, N. and Wright, P. (1992) Airport Engineering, 3rd ed., John Wiley, New York
Horonjeff, R. and McKelvey, F. (1994) Planning and Design of Airports, 4th ed.,
55-72.
18