Modelling future airport capacity and capacity utilisation

Chapter 7

Modelling future airport capacity and capacity utilisation For assessing the effects of limited airport capacity on global air traffic, we ideally need to assign capacity values to each airport, at least to those airports which may reach their capacity limit before the forecast horizon. This is an enormous task and a major difference to pure demand forecasts, which often neglect potentially negative effects of limited airport capacity. Naturally, a global capacity assessment cannot be conducted with the same precision as a detailed capacity study for a particular airport. Here, we have detailed information about many influencing factors, such as the runway and terminal layout, fleet mix, weather and terrain and air traffic management. It would be an insurmountable task to gather and process such an amount of information for all airports worldwide, and to make projections about their development for the next 20 30 years. Even limiting this approach to only such airports that may reach their capacity limit before the forecast horizon, and collecting capacity relevant information for each airport, generates a sample that is most likely too large for a detailed airport capacity assessment within a reasonable time frame. Furthermore, while a number of airports will reach their capacity limit before the forecast horizon, some airports may be successfully enlarged, typically by one or more new runways. Therefore we not only need a method that is able to assess current airport capacity on a global level, but also one that is able to assess future airport capacity. Whether adding a new runway is a realistic option for enlarging airport capacity will be the topic of Chapter 8, Modelling future airport capacity enlargements and limits. As a result, we need a method that produces robust results with input information which is generally available. We focused therefore on the runway system, as this typically limits overall airport capacity in the long term. As we will see later in more detail, it is not necessary (and hardly possible on a global level) for the model to produce current and future capacity values that are as ‘exact’ as possible. Future development of global air traffic depends among other things on the availability of future airport capacity, as we allow for capacity enlargements in our model. Therefore if the airport capacity model under- or overestimates current airport capacities, future Airport Capacity Constraints and Strategies for Mitigation. DOI: https://doi.org/10.1016/B978-0-12-812657-8.00007-5 © 2020 Elsevier Inc. All rights reserved.

161

162

PART | II Models for assessing mitigation strategies

airport capacity enlargements will be easier or harder to realise, respectively, until the forecast horizon. One major factor which determines the probability of a successful airport capacity enlargement is the number of aircraft movements that the airport has to handle. The more aircraft movements, the harder airport capacity enlargements become. Thus under- or overestimating airport capacity means more or fewer steps of increasing capacity, respectively. Any inaccuracies in assessing airport capacity that still remain typically play only a minor role in the strategic planning process. Nevertheless, for our generic airport capacity model we aimed for an as-high-as-possible degree of accuracy, but we do not pretend it is as accurate as detailed airport capacity analyses for specific airports.

7.1

Background

In Chapter 3, Capacity utilisation at airports worldwide, we introduced the analytical tool of air traffic ranking curves and the capacity utilisation index (CUI) concept. The CUI is an indicator to describe capacity utilisation of airports by relating the average hourly air traffic volume to the 5% peak hour volume. However, this relationship only works for airports in a meaningful way if an airport is operating close to its capacity during the 5% peak hour. As we have seen, this is not the case for many smaller airports which do not reach traffic volumes near capacity during peak hours. For such airports a value of practical capacity has to be found in another way for computing the capacity utilisation of today and for the future. Another main benefit of the CUI concept is the deduction of a so-called annual airport capacity or annual service volume. These two terms are used interchangeably in this book. This is in contrast to a naı¨ve approach, whereby the annual capacity is calculated as maximum hourly capacity multiplied by the number of operating hours of the airport per day multiplied by the number of days per year, whereby the CUI takes the demand side into account. For example, there is less demand during night hours or weekends; thus the technical capacity will not be fully used because of a temporary lack of demand. This is even the case for highly utilised airports such as London Heathrow (LHR). However, it is difficult to define universally applicable CUI values that describe maximum airport capacity utilisation considering the demand side. While such values vary from airport to airport and even from runway system to runway system, maximum CUI values are in a range from 0.75 to 0.86 and might be even lower for airports comprising of many runways in a complex layout. Asian airports, for example, tend to have more belly-shaped ranking curves compared to their US and European counterparts, resulting in slightly higher CUI values, while very large airports with a high portion of domestic traffic such as Atlanta Hartsfield Jackson (ATL) achieve more pronounced peaking as the annual air traffic ranking curves clearly indicate. However, while the CUI is a valuable tool for analysing traffic structures at airports and deriving a

Modelling future airport capacity and capacity utilisation Chapter | 7

163

meaningful concept of annual airport capacity or service volume, the large number of airports operating well below their capacity during peak hours makes it difficult to develop a robust and simple econometric model of generic airport capacity on the basis of the CUI concept. As we have already outlined before, the two requirements of the model to be robust as well as simple to use are a conditio sine qua non.

7.2

Model theory and parameter estimation

We have chosen an approach for estimating future airport capacity which is based on data envelopment analysis (DEA) and regression analysis. DEA is a non-parametric empirical method of operations research to estimate production frontiers employing linear programming techniques. This approach was first described in Charnes et al. (1978). A good overview of DEA can also be found in Cooper et al. (2007). Today, DEA is a standard tool for efficiency analyses and benchmarking of so-called decision-making units (DMUs). DEA allows us to compare DMUs which differ in their input and output structure. Examples of such DMUs comprise hospitals, energy production or cost-/profit-centres of large organisations. A major advantage of DEA is that there is no need to explicitly describe the production function, not even the functional form; hence, it is a non-parametric method. Instead, the production function is unveiled by empirical data. However, therein lies a major weakness of DEA: the results depend very much on the DMUs and the selection of inputs and outputs upon which the DEA is based. Increasing the number of inputs and outputs results in an increase of the number of efficient DMUs, and it is difficult to select between different DEA setups (Berg, 2010). Thus we have chosen an effective DEA setup. We have taken two inputs and one output and a constant returns to scale output-oriented DEA formulation. Instead of measuring the efficiency of DMUs, we analyse airport capacity utilisation: which airport achieves what output given a particular input structure? This is analogous to measuring efficiency. Inputs are the number of runways [Digital Aeronautical Flight Information Files (DAFIF), 2016] and the number of operating hours of an airport [Official Airline Guide (OAG), 2016]. The number of runways is simply added with the exception that pure independent runway systems are counted as single runway systems with the appropriate fraction of their annual aircraft movements. For example, London Heathrow (LHR) is represented as an airport with one runway and half the aircraft movements. Beijing (PEK) is also counted as an airport with one runway but only a third of its annual aircraft movements because of three independent runways. Operating hours of an airport are defined as hours with more than five aircraft movements in the case of larger airports to reflect, for example, night flight restrictions. In the case of small airports, all hours with at least one

164

PART | II Models for assessing mitigation strategies

aircraft movement are counted as operating hours, as these hours contribute to a significant part of their total air traffic volume. The output is given by the number of annual aircraft movements of an airport [Official Airline Guide (OAG), 2016] or an appropriate fraction in the case of independent runways. The idea behind the ‘fraction’ approach is that we make the simplifying assumption that the capacity of an independent runway system equals the capacity of a single runway multiplied by the number of runways. This model assumption enables us to use the information in the data more efficiently by imposing constraints based on prior knowledge. It should be added that airports with two independent runways have a slightly higher capacity per runway than single runway airports, since often aircraft can be assigned to each runway so as to optimise the runway utilisation, for instance, by sequencing following aircraft by their weight. To conduct DEA, we employed the Robust Data Envelopment Analysis (DEA) for R package (rDEA) (Simm and Besstremyannaya, 2016). The output/input structure, that is the ratio of the number of annual aircraft movements per runway and of average number of aircraft movements per operating hour, is assumed to be constant over time for each airport. This ratio varies between airports and describes their production characteristics. For example, Chicago O’Hare (ORD), an airport with a complex system of eight runways and rather small aircraft, with on average of 90 passengers per flight, achieves on average about 109 aircraft movements per operating hour, but only about 107,000 annual aircraft movements per runway. Chicago O’Hare has a high average hourly throughput due to pronounced peaking characteristics, the high number of runways and rather small aircraft, but because of its complex runway system, it has a rather small annual throughput per runway. On the other hand, London Heathrow, an airport with two independent runways and much larger aircraft, with on average of 159 passengers per flight, achieves on average about 72 aircraft movements per operating hour, but around 238,000 annual aircraft movements per runway. Hourly throughput is limited by the rather low number of runways and larger aircraft, but the independent runway system allows for a high annual throughput per runway, even though the airport is used primarily during the daytime and not at night because of a curfew from 22 p.m. to 5 a.m. These examples show that there is a great deal of implicit information in the data which makes this problem well suited for DEA. Furthermore, the assumption of a constant input structure per airport for the forecast period seems to be reasonable. Before we discuss the results of the DEA in more detail, Fig. 7.1 gives an overview of the generic airport capacity model estimation of which the DEA is a part. The first step is the aforementioned DEA to estimate current airport capacity for a sample of airports. For this, we have chosen the largest 200 airports in terms of 2016 aircraft movements, because they belong to the

Modelling future airport capacity and capacity utilisation Chapter | 7

165

Performing data envelopment analysis (DEA) of the 200 largest airports in terms of aircraft movements

Current capacity utilisation per airport

Computing average number of aircraft movements per runway and operating hour at highest possible capacity utilisation

Mean values per runway class at highest possible capacity utilisation

Estimating functional relationship between number of runways and average number of aircraft movements per runway and operating hour at highest possible capacity utilisation FIGURE 7.1 Estimation of the generic airport capacity model.

large number of airports already suffering from a lack of capacity or are likely to do so within the next 20 30 years, as described in Chapter 3, Capacity utilisation at airports worldwide, and Chapter 4, Constrained and under-utilised airports. Furthermore, including smaller airports does not change the efficient frontier which we want to estimate. In a second step, we calculated the average number of aircraft movements per runway and operating hour at the highest possible level of capacity utilisation for each airport. This was deduced from the annual distribution of hourly volumes as given by the volume ranking curves of those airports which have been operating at their capacity limit for many years, such as London Heathrow. The CUI at these airports reaches a maximum of about 0.86, depending on the complexity of operations at multi-runway airports. The so-called maximum average number of aircraft movements stands, thus, for a practical or sustainable capacity which can be achieved over several hours and may be exceeded in peak hours such as the 5% peak hour. The hourly capacity of a runway, which may serve as a short-time capacity of say one hour, is thus higher than the maximum average number of aircraft movements. In fact, this capacity value is defined by the ratio of the maximum average number of aircraft movements and the highest possible CUI. In the case of a single runway or two independent parallel runways the

166

PART | II Models for assessing mitigation strategies

capacity is thus 15% 20% higher than the maximum average number of aircraft movements. This defined capacity also represents a practical capacity, but without any indication of the average delay per aircraft movement, as is normally the case when we speak of the practical capacity in contrast to the theoretical capacity (see Chapter 2: Concepts of capacity and methods of estimation). As has been mentioned, comparable delay statistics are not available for a global capacity analysis. To compute the maximum average number of aircraft movements, we first calculated the annual service volume per airport by dividing the actual number of annual aircraft movements by the capacity utilisation as estimated by the DEA from step one (see Fig. 7.2). Then, the annual service volume is divided by the number of runways and operating hours of the airport. Based on this, mean values per airport class are calculated. Airport classes are defined by the number of runways, whereby airports with two runways are further subdivided into those with parallel independent, parallel dependent and crossing runways. Two runways are defined as independent if the two parallel runways are not closer than 1500 m. The last step is to perform a regression analysis based upon the mean values from step two. Thereby, it is possible to generalise the results from step two and extrapolate to runway systems with more than eight runways. So far, there are only three airports with six runways [Denver (DEN), Detroit Metropolitan Wayne County (DTW) and Amsterdam Schiphol (AMS)], one with seven runways [Dallas/Fort Worth (DFW)] and one with eight runways [Chicago O’Hare (ORD)]. Thus, performing the regression analysis on top of 300,000 London Gatwick (LGW): 0% capacity reserve

Annual aircraft movements per runway

100% LGW

250,000 Jakarta SoekarnoHatta (CGK): 19% capacity reserve PEK

200,000

97% LGW and 3% ATL London Heathrow (LHR): 9% capacity reserve *

83% LGW and 17% ATL

Dubai (DXB): 10% capacity reserve

HKG

31% LGW and 69% ATL

SIN

Atlanta HartsfieldJackson (ATL): 0% capacity reserve

Los Angeles (LAX): 13% capacity reserve

150,000

100% ATL

100,000 Chicago O'Hare (ORD): 10% capacity reserve DEN

50,000

DFW

AMS

* Excluding annual movement cap of 480,000

0

0

20

40

60

80

100

Average aircraft movements per operating hour

FIGURE 7.2 Efficient or capacity frontier of the 200 sample airports.

120

140

Modelling future airport capacity and capacity utilisation Chapter | 7

167

the mean values yields more robust results, as we will see later in this chapter in more detail. Fig. 7.2 shows the efficient frontier, or what we call a capacity frontier, of the 200 sample airports for estimating the actual capacity utilisation. The number of annual aircraft movements represents the output and the number of runways and operating hours are the inputs for the DEA model. Typically, the two axes of a DEA diagram with one output and two inputs are defined as ‘input 1/output’ and ‘input 2/output’, resulting in an efficient frontier which is orientated to the origin of the diagram (see, e.g., Cooper et al., 2007). However, for better interpretation of the model results we have chosen to invert the axes’ definition, that is output to input, which means ‘annual aircraft movements per runway’ and ‘average aircraft movements per operating hour’. Such a modification of a standard DEA diagram results in a nonlinear middle part of the efficient frontier in Fig. 7.2. This affects the presentation of the model results but has no effect on their computation. The capacity frontier is defined by London Gatwick (LGW, one runway) and Atlanta Hartsfield Jackson (ATL, five runways) airports, which are the reference airports. London Gatwick has the highest number of annual aircraft movements per runway, and ATL has the highest average aircraft movements per hour. Therefore these two airports have 100% capacity utilisation in the sense that, according to the DEA employed, the actual traffic volume represents the maximum throughput among the airports of the sample of the 200 biggest airports. The capacity utilisation of the remaining 198 airports is computed on the basis of virtual reference airports on the capacity frontier which have the same input structure. This is illustrated in Fig. 7.2 by projections from the origin through the airport considered onto the capacity frontier. The ratio of the dashed part of the arrow to the full arrow, that is the dashed plus the solid part, is equal to the capacity utilisation of the airport. Virtual reference airports are a linear combination of the airports that form the capacity frontier: LGW and ATL. For illustration purposes, we have highlighted five example airports: G

G

Jakarta Soekarno Hatta (CGK), an airport with two independent runways which is benchmarked against 100% LGW. From this comparison, CGK has a capacity reserve of 19%. London Heathrow (LHR), an airport with two independent runways which is benchmarked against a linear combination of 97% LGW and 3% ATL. From this comparison, LHR has a capacity reserve of 9%, if we do not account for the annual movement cap of 480,000. However, due to the movement cap, LHR is currently virtually fully utilised. Nevertheless, LHR potentially could handle more flights. LHR managers have proposed to raise the cap to about 500,000 movements per year (Evening Standard, 2016).

168 G

G

G

PART | II Models for assessing mitigation strategies

Dubai (DXB) airport, an airport with two dependent parallel runways which is benchmarked against a linear combination of 83% LGW and 17% ATL. From this comparison, DXB has a capacity reserve of 10%. Los Angeles (LAX) airport, an airport with four runways which is benchmarked against a linear combination of 31% LGW and 69% ATL. From this comparison, LAX has a capacity reserve of 13%. Chicago O’Hare (ORD) airport, an airport with eight runways which is benchmarked against 100% ATL. From this comparison, ORD has a capacity reserve of 10%.

Fig. 7.2 displays a cone which is defined by a dashed line from the origin to LGW and ATL, respectively. Most airports are within the cone, which basically means that the reference airports cover a wide range of output/input combinations. This is helpful for identifying virtual reference airports that are combinations of LGW and ATL. However, there are some airports a little outside the cone. Beijing (PEK), Hong Kong (HKG) and Singapore Changi (SIN) airports are examples that are to the left of the cone and have a high value of annual aircraft movements per runway compared to their value of average aircraft movements per operating hour. This is a result of their above-average number of operating hours (more than 8700 hours per year) and rather small number of runways (two for HKG and SIN and three for PEK). On the other hand, Amsterdam Schiphol (AMS), Denver (DEN) and Dallas/Fort Worth (DFW) airports are examples that are to the right of the cone and have a low value of annual aircraft movements per runway compared to their value of average aircraft movements per operating hour. This is a result of their below average number of operating hours (less than 7200 hours per year for DEN and DFW and less than 7400 per year for AMS) and rather high number of runways (six for AMS and DEN and seven for DFW). The approach bears a resemblance to the capacity envelopes discussed in Chapter 2, Concepts of capacity and methods of estimation (Gilbo, 1993, and Fig. 2.2 for Munich airport). However, while the capacity envelopes are based upon output measures, that is arrivals and departures, our DEA approach employs relative performance indicators of annual aircraft movements per runway and average aircraft movements per operating hour. Furthermore, capacity envelopes help to assess hourly capacity and the relationship with the arrival/departure mix, while our model aims more at the long-term capacity and focuses on the average hour volume at full capacity utilisation and a balanced arrival/departure mix. Table 7.1 shows the top 50 airports ranked by the degree of capacity utilisation derived from the DEA. As has been mentioned, based on the capacity utilisation we can eventually calculate an annual service volume per airport (column 9) and a maximum average capacity per runway per operating hour (column 10), that is at the highest possible annual capacity

TABLE 7.1 Results from the data envelopment analysis (DEA) for the top 50 airports in terms of 2016 capacity utilisation. 1

2

3

4

5

6

7

8

9

10

11

12

No.

IATA code

Airport name

ACM 2016

CUI 2016

OH 2016

RWYs 2016

CU DEA2

eASV (DEA)

MAC/RWY/ OH (DEA)

CU DEA3

Cols. 8 11

1

ATL

Atlanta Hartsfield Jackson

875,211

0.67

7223

5

100

875,211

24.2

100

0

2

LGW

London Gatwick

273,606

0.69

7334

1

100

273,606

37.1

100

0

3

MEX

Mexico City Benito Juarez

406,022

0.68

8083

2

93

436,778

26.9

93

0

4

LHR

London Heathrow

476,226

0.82

6645

2

91

523,302

39.3

91

0

5

ORD

Chicago O’Hare

855,194

0.63

7811

8

90

946,134

15.1

90

0

6

DXB

Dubai

403,542

0.82

8784

2

90

447,947

25.5

90

0

7

LGA

New York LaGuardia

372,399

0.74

6827

2

90

415,506

30.4

99

29

8

LAX

Los Angeles

635,265

0.69

7753

4

87

733,198

23.6

89

22

9

SAW

Istanbul Sabiha Gokcen

219,326

0.63

7161

1

81

270,751

37.6

81

0

10

CGK

Jakarta Soekarno Hatta

440,707

0.71

8024

2

81

547,212

34.0

100

219

11

IST

Istanbul Ataturk

457,446

0.76

8734

3

76

598,129

22.8

76

0

12

DUB

Dublin

201,368

0.67

7122

1

75

270,049

37.7

75

0 (Continued )

TABLE 7.1 (Continued) 1

2

3

4

5

6

7

8

9

10

11

12

No.

IATA code

Airport name

ACM 2016

CUI 2016

OH 2016

RWYs 2016

CU DEA2

eASV (DEA)

MAC/RWY/ OH (DEA)

CU DEA3

Cols. 8 11

13

DFW

Dallas/Fort Worth

646,079

0.68

7189

7

74

872,443

17.3

74

0

14

PEK

Beijing

606,105

0.79

8757

3

74

820,818

31.2

74

0

15

HKG

Hong Kong

399,489

0.69

8739

2

73

547,212

31.3

73

0

16

CAN

Guangzhou Baiyun

430,241

0.74

8374

3

73

592,467

23.5

73

0

17

XMN

Xiamen Gaoqi

194,030

0.80

6950

1

73

267,205

38.2

73

0

18

CTU

Chengdu Shuangliu

310,453

0.86

7653

2

72

430,010

27.9

72

0

19

CLT

Charlotte Douglas

511,794

0.75

6664

4

72

714,935

26.7

72

21

20

MUC

Munich

374,718

0.64

6842

2

71

529,674

38.7

71

0

21

EWR

Newark Liberty

400,757

0.72

7337

3

70

575,353

26.0

73

24

22

FUK

Fukuoka

167,574

0.85

5490

1

69

242,227

44.1

69

0

23

BOM

Mumbai Chhatrapati Shivaji

307,071

0.77

8694

2

69

446,555

25.7

69

0

24

SEA

Seattle Tacoma

395,526

0.68

7387

3

69

576,423

25.9

72

24

25

PHX

Phoenix Sky Harbor

381,678

0.69

7129

3

67

571,770

26.6

70

23

26

SAN

San Diego

173,266

0.71

6589

1

66

260,870

39.5

82

216

27

SIN

Singapore Changi

357,945

0.69

8751

2

65

547,212

31.3

65

0

28

BLR

Bengaluru Kempegowda

176,468

0.67

8442

1

64

273,606

32.3

80

216

29

DEL

Delhi Indira Gandhi

383,259

0.70

8784

3

64

598,905

22.7

64

0

30

CSX

Changsha Huanghua

166,711

0.78

6639

1

64

262,093

39.2

64

0

31

FRA

Frankfurt

454,775

0.69

6749

4

63

716,242

26.5

63

0

32

DEN

Denver

545,318

0.59

7173

6

63

868,642

20.1

63

0

33

TAO

Qingdao Liuting

167,266

0.81

6919

1

63

267,187

38.2

63

0

34

SUB

Juanda

162,080

0.66

6607

1

62

261,603

39.3

77

215

35

CDG

Paris Charles de Gaulle

448,506

0.67

7225

4

62

724,247

25.0

62

0

36

URC

¨ ru¨mqi Diwopu U

165,285

0.74

6935

1

62

266,905

38.3

62

0

37

HND

Tokyo Haneda

450,207

0.64

7730

4

61

733,002

23.6

61

0

38

BKK

Bangkok Suvarnabhumi

331,898

0.71

8760

2

61

547,212

31.2

61

0

39

KMG

Kunming Changshui

328,953

0.82

7287

2

60

544,909

37.3

60

0

40

KUL

Kuala Lumpur

356,414

0.70

8378

3

60

592,694

23.5

60

0 (Continued )

TABLE 7.1 (Continued) 1

2

3

4

5

6

7

8

9

10

11

12

No.

IATA code

Airport name

ACM 2016

CUI 2016

OH 2016

RWYs 2016

CU DEA2

eASV (DEA)

MAC/RWY/ OH (DEA)

CU DEA3

Cols. 8 11

41

CKG

Chongqing Jiangbei

255,462

0.83

7369

2

60

425,531

28.6

60

0

42

LIM

Lima Jorge Chavez

164,080

0.60

7837

1

60

273,606

34.3

75

215

43

JFK

New York John F. Kennedy

440,398

0.70

7924

4

60

736,280

23.1

60

0

44

STN

London Stansted

154,785

0.59

6508

1

59

260,224

39.6

59

0

45

MNL

Manila Ninoy Aquino

261,597

0.72

8398

2

59

442,076

26.2

59

0

46

SVO

Moscow Sheremetyevo

258,837

0.69

8681

2

58

446,434

25.7

58

0

47

SFO

San Francisco

419,364

0.68

7540

4

57

729,765

24.1

59

21

48

GVA

Geneva

148,053

0.66

6405

1

57

257,814

40.1

57

0

49

SZX

Shenzhen Bao’an

314,042

0.79

8095

2

57

547,212

33.6

57

0

50

SYD

Sydney Kingsford Smith

317,815

0.68

6222

3

57

556,976

29.6

57

0

362,892

0.71

7533

2.5

70

513,323

30.2

72

22

Ø

Cols. 8 11, difference between the values of columns 8 and 11; CU, capacity utilisation in %; CUI, capacity utilisation index; DEA, data envelopment analysis; DEA2, DEA with two inputs; DEA3, DEA with three inputs; MAC/RWY/OH, maximum average capacity per runway per operating hour; OH, operating hours; RWYs, runways.

Modelling future airport capacity and capacity utilisation Chapter | 7

173

utilisation. The values of column 10 form the input for estimating the generic airport capacity model in step three of Fig. 7.1. Before we proceed, we want to discuss briefly an alternative DEA setup. As Berg (2010) pointed out, it is difficult to decide between different DEA setups. We have seen in Chapter 3, Capacity utilisation at airports worldwide, that slot coordination is an influencing factor for modelling the relationship between 5% peak hour volume and average hour volume. Thus slot coordination affects airport capacity and capacity utilisation. We have to check, therefore, whether including slot coordination improves our model significantly. Including more input variables leads to a further discrimination between airports and an increase in the number of airports which have 100% capacity utilisation. This ultimately leads to lower capacity values in the generic airport capacity model for a given runway system. Slot coordination has been considered in two categories in the alternative DEA setup (IATA, 2016): G G

Airport is not Level 3 slot coordinated Airport is Level 3 slot coordinated

Columns 11 and 12 of Table 7.1 present the results. Column 11 shows the new capacity utilisation values and column 12 the difference in capacity utilisation between the original and the alternative DEA setup. The alternative DEA setup yields three airports with 100% capacity utilisation: G G G

LGW (capacity utilisation in the original DEA setup: 100%) ATL (capacity utilisation in the original DEA setup: 100%) Jakarta Soekarno Hatta (CGK, capacity utilisation in the original DEA setup: 81%)

There is a significant increase of capacity utilisation, greater than 10% points, at Lima Jorge Chavez (LIM, 115% points), Juanda (SUB, 115% points), San Diego (SAN, 116% points), Bengaluru Kempegowda (BLR, 116% points) and CGK (119% points) airports. This is a result of a different capacity frontier, which is now defined by LGW, ATL and CGK. The major difference between the original and the alternative DEA setup is a lower annual service volume for a number of airports with one or two runways. This value drops from between 262,000 and 274,000 aircraft movements to between 210,000 and 220,000 aircraft movements per runway. From our point of view, these values seem to be rather low. As the difference in capacity utilisation is otherwise very small, we have chosen the original setup for the capacity forecast. However, a model assessment with actual airport data will be carried out at the end of this chapter. Nevertheless, this shows how data-sensitive DEA is. Including more inputs or outputs increases the number of efficient DMUs. We further checked for setups with variable returns to scale or splitting the dataset completely according to slot coordination status but this leads to increasing numbers of airports which were supposed to operate at their capacity limit. The results

174

PART | II Models for assessing mitigation strategies

became, thus, more unreasonable when checking key statistics such as 5% peak hour volume, CUI and annual aircraft movements. Considering air traffic control in the DEA, in other words, whether instrument flight rules as for example in Europe or visual flight rules as in the United States normally apply, yields similar results in terms of annual service volume. Therefore adding features such as inputs and outputs or more flexible production frontiers have to be carried out in a meaningful way, and even then, more complexity is not always better. Nevertheless, this clearly illustrates how difficult the choice is between different setups. For our case, it is essential for the setup to remain effective and to choose a smart problem representation, such as the ‘fractional’ approach, to use the information in the dataset as efficiently as possible and to keep problem complexity as low as possible. Table 7.2 displays the airport class means of the maximum average number of aircraft movements per runway and operating hour and their variance based upon the results of the DEA (see Table 7.1, column 10, all 200 airports), which is the basis for step three of the generic airport capacity model (see Fig. 7.1), that is the regression analysis. As expected, the maximum average number of aircraft movements per runway and operating hour— representing a practical capacity which can be maintained over several hours—declines gradually with increasing numbers of runways. The variance tends to decline with increasing numbers of runways as well but is only available for airports of up to six runways. The reason for the latter is that

TABLE 7.2 Maximum average aircraft movements per runway and operating hour by airport class. Variance (σi2)

Number of runways

Maximum average aircraft movements per runway and operating hour based on DEA

1

37.9

9.18

2, independent

37.9

9.18

2, dependent

29.8

20.99

2, crossing

29.8

5.72

3

26.3

6.18

4

25.3

2.50

5

23.3

3.07

6

20.1

4.48

7

17.3

8

15.1

DEA, data envelopment analysis.

Modelling future airport capacity and capacity utilisation Chapter | 7

175

Maximum average aircraft movements per runway and operating hour

40

35

30

25

y = 47.708 – 11.439x0.5 R 2 = 97.36%

20

15

10

5

0 0

1

2

3

4

5

6

7

8

9

Number of runways per airport

FIGURE 7.3 Regressing maximum average aircraft movements per runway and operating hour on the number of runways.

there is only one airport with seven and one with eight runways. The airport class of six runways consists of three airports (DEN, AMS and DTW). Fig. 7.3 displays the values of Table 7.2 in a chart which visualises the functional relationship between maximum average aircraft movements per runway and operating hour and the runway system of an airport as represented by the number of runways. In Fig. 7.3 the value for airports with two runways is an average of dependent and crossing systems, while the capacity value per runway for two independent runways is equal to the value for single runway airports. The regression line in Fig. 7.3 has been estimated by ordinary least squares (see, e.g., Greene, 2011). For airports with two dependent and crossing runways the two original data points were considered for model estimation, so that we have in total nine data points. Table 7.3 presents the estimation results of the generic airport capacity model. The square root of the number of runways per airport is the independent variable. We have taken this figure because the maximum average aircraft movements per runway and operating hour declines slightly less than linearly as depicted in Fig. 7.3. The model has been estimated on class mean values; nevertheless, both variables are highly significant. R2 is very high with 97.36%. However, this should not be overemphasised as this is a rather parsimonious model approach. The model allows us to calculate annual service volumes for a generic airport with a given number of runways and operating hours. The annual service volume is thereby the product of the hourly capacity per runway

176

PART | II Models for assessing mitigation strategies

TABLE 7.3 Estimation results of the generic airport capacity model with square root transformation. No.

Variable

Coefficient

1

Constant

2

Square root of number of runways R2

Standard error

p-value

47.70753

1.63136

0.00000

2 11.43949

0.76903

0.00001

97.36%

(column 3 of Table 7.4), the number of runways and the number of operating hours per year. We call the application of the model in this way the base mode, since the model yields the same results of annual airport capacity for airports with the same number of runways (in the case of two runways differentiated by independent and dependent or crossing runways) and operating hours. Table 7.4 provides the base case capacities for runways of airports with up to eight runways (but not limited to) and 18 operating hours per day year-round (column 4 of Table 7.4). Furthermore, column 5 displays the relative capacity gain by adding one runway at a time. As expected, adding more and more runways to an airport yields typically decreasing capacity gains, and significant capacity gains in terms of annual service volume can be only achieved with up to five to six runways if operating hours are held constant (this is not the case for peak hour volumes, as we will see later in this chapter). However, in the next chapter, we will show that expanding an airport gradually from one or two runways to up to six runways or more is, for our analyses in Part III of the book, of lesser importance, because such an enlargement will be in many cases an insurmountable task. Column 6 displays the annual service volume per runway and airport class. Again, this value declines with increasing numbers of runways. There are only five airports with six or more runways in the data sample of 200 airports. Thus a closer analysis of their modelled capacity values seems appropriate. Dallas/ Fort Worth (DFW), an airport with seven runways, had 646,079 aircraft movements in 2016 and on average about 20 operating hours per day (Table 7.1). Based on the DEA (Table 7.1 and Fig. 7.1), the annual service volume was estimated to be 872,443 aircraft movements, thus a capacity utilisation of 74%. A generic airport with seven runways and 20 operating hours per day has an annual service volume of 802,134 3 20/18 5 891,260 aircraft movements (based on Table 7.4 and corrected for operating hours). Chicago O’Hare (ORD), an airport with eight runways, handled 855,194 aircraft movements in 2016 during on average about 21 operating hours per day (Table 7.1). Based on the DEA, we calculated an annual service volume

Modelling future airport capacity and capacity utilisation Chapter | 7

177

TABLE 7.4 Applying the generic capacity model in base mode. 1

2

3

4

5

6

Number of runways

MAACM/ RWY/OH (DEA)

MAACM/ RWY/OH (RA)

ASV (18/ 365)

Increase of ASV by adding one RWY (%)

ASV/ RWY (18/ 365)

1

37.9

36.3

238,281

2, independent

37.9

36.3

476,562

100

238,281

2, dependent

29.8

31.5

414,300

74

207,150

2, crossing

29.8

31.5

414,300

74

207,150

3

26.3

27.9

549,786

15a; 32b

183,262

4

25.3

24.8

652,494

19

163,124

5

23.3

22.1

726,906

11

145,381

6

20.1

19.7

776,046

7

129,341

7

17.3

17.4

802,134

3

114,591

8

15.1

15.4

806,888

1

100,861

238,281

ASV, annual service volume; DEA, data envelopment analysis; MAACM/RWY/OH, maximum average aircraft movements per runway per operating hour; RA, regression analysis; RWY, runway; 18/365, 18 hours 365 days a year. a Adding a third runway to an airport with two independent runways. b Adding a third runway to an airport with two dependent/crossing runways.

of 946,134 aircraft movements, thus a capacity utilisation of 90%. A generic airport with eight runways and 21 operating hours per day has an annual service volume of 806,888 3 21/18 5 941,369 aircraft movements (based on Table 7.4 and corrected for operating hours). Finally, based on the DEA, the average annual service volume of airports with six runways is estimated to be around 878,000 aircraft movements, and for the three airports with six runways ranges from almost 869,000 to more than 892,000. Average operating hours per day are nearly 20 hours. The generic airport with six runways of Table 7.4 results in an annual service volume of about 862,000 aircraft movements for 20 operating hours per day. Thus we conclude from the discussion in this chapter and the brief model assessment for the five largest airports in terms of their number of runways that the modelled values are sufficiently precise for a global or network encompassing analysis in Part III of the book. Fig. 7.4 illustrates the incremental mode of the generic capacity model, which is typically employed for already existing airports. After performing a

178

PART | II Models for assessing mitigation strategies

Perform data envelopment analysis (DEA) for airports under study

Current capacity utilisation

Calculate annual service volume for the highest possible capacity utilisation of the current runway system

Estimate of currrent airport capacity

Calculate the capacity difference between the current and future runway system based on the generic capacity model

Estimate of capacity improvement

Add the generic capacity difference to the current annual service volume of the airports under study FIGURE 7.4 Applying the generic capacity model for forecasting in incremental mode.

DEA, only the capacity gain by adding one runway at a time is added to the estimated current capacity limit of an airport. Under the assumption of a constant input structure, as discussed earlier, the multiplier for the capacity gain can be retrieved from column 5 of Table 7.4. For example, upgrading a single runway airport to two independent runways yields a multiplier of 2 (100% capacity gain), while upgrading an airport with three runways to four runways has a multiplier of just 1.21 (21% capacity gain).

7.3 Model application: comparison of model results with actual traffic data The question we want to answer in the following section is how do the model results compare with actual peak hour capacities of airports? With that we want to present a deeper insight into the capabilities of the chosen approach.

Modelling future airport capacity and capacity utilisation Chapter | 7

179

This comparison has been carried out for those airports in Tables 2.2 and 2.3 of Chapter 2, Concepts of capacity and methods of estimation, which are among the 200 airports the analysis is based upon. As can been seen from these tables, airport capacity is typically presented as some kind of sustained or practical capacity (‘declared capacity’ if the airport is slot coordinated). The key concept of the model aims to identify the maximum average number of aircraft movements per runway and operating hour. Therefore we have tried to connect both concepts by means of the CUI. As described in Chapter 3, Capacity utilisation at airports worldwide, the CUI is defined by the ratio of average hourly traffic volume to the 5% peak hour volume. If the airport is at its capacity limit, the average hourly traffic volume is basically equal to the maximum average number of aircraft movements per runway and operating hour multiplied by the number of runways. Therefore dividing the latter metric by the CUI value yields a decent approximation of 5% peak hour volume and, thus, of practical capacity. However, the question remains which CUI value should be taken as the appropriate one when we want to estimate 5% peak hour volume? As can be seen from Table 7.1, the maximum attainable CUI value tends to decrease with the number of runways, in other words, with the complexity of the runway system of an airport. Fig. 7.5 displays the CUI values of those airports in Table 7.1, which have the highest CUI values within each airport capacity class. These

Maximum CUI value of top 50 airports in terms of DEA capacity utilisation

1 0.9 FUK

0.8

DXB CLT

y = 0.8686 – 0.0319x R² = 89.44%

IST

0.7

ATL

DFW

AMS

ORD

0.6 DEN

0.5

y = 0.8704 – 0.0337x R² = 81.51%

0.4 0.3 0.2 0.1 0 0

1

2

3

4

5

6

7

8

9

Number of runways FIGURE 7.5 Reference CUI values in relation to the number of runways (two runways applies to dependent and crossing runways, two independent runways is equal to single runway). CUI, capacity utilisation index.

180

PART | II Models for assessing mitigation strategies

represent the top 50 airports in terms of capacity utilisation of the sample of 200 airports for the DEA. Here, fully independent runway systems such as London Heathrow or Beijing airports are assigned to single runway airports, and the category of two-runway airports comprises only those with dependent and crossing runways. As can be seen, there is a distinct decrease of the maximum CUI value with an increasing number of runways. The value of Denver (DEN) airport seems to be rather low; therefore we have chosen the CUI value of Amsterdam Schiphol (AMS) as an alternative, in spite of the fact that AMS is ranked 60th in terms of capacity utilisation. The approach chosen ensures that the maximum CUI value for each generic airport can be found as given by the number of runways, and airports with high CUI values, but sub-maximum peak hour volumes, that is a lack of capacity utilisation, can be avoided. Nevertheless, this procedure is only as good as the data that is employed, and there are, in general, only a few airports with five or more runways worldwide. Therefore the results should be interpreted with caution and should not be overemphasised. Fig. 7.5 shows two linear regression lines. The lower dashed line is the original regression (with Denver airport, called the DEN model), while the upper solid line is the alternative model (with Amsterdam Schiphol airport, called the AMS model). Table 7.5 displays detailed results of both regressions. While both models perform very well in terms of goodness-of-fit (R2 . 80% in both cases), and the estimated coefficients are significant (p-value ,0.01 for the DEN model and even ,0.001 for the AMS model), we have chosen to retain the AMS model for further analysis, as it is superior in terms of R2 and significance of coefficient estimates. Besides, the regression results tend to support the hypothesis that the CUI value for

TABLE 7.5 Estimation results of the model ‘capacity utilisation index (CUI) by number of runways’. Model

No.

Variable

AMS

1

Constant

2

Number of runways

Coefficient

Standard error

p-value

0.86857

0.02260

0.00000

2 0.03190

0.00448

0.00038

R2

89.44% DEN

1

Constant

2

Number of runways

0.87036

0.03308

0.00000

2 0.03369

0.00655

0.00213 81.51%

Modelling future airport capacity and capacity utilisation Chapter | 7

181

TABLE 7.6 Modelled maximum 5% peak hour volume for comparison purposes (two runways applies to dependent and crossing runways, two independent runways is equal to two times single runway values). Number of runways per airport

Reference CUI

Maximum average aircraft movements per operating hour

Modelled maximum 5% peak hour volume

1

0.84

36.3

43.3

2

0.80

63.1

78.4

3

0.77

83.7

108.3

4

0.74

99.3

134.0

5

0.71

110.6

156.0

6

0.68

118.1

174.4

7

0.65

122.1

189.2

8

0.61

122.8

200.2

CUI, capacity utilisation index.

Denver airport may be an outlier, which we have corrected by taking Amsterdam Schiphol instead. Based upon the estimated AMS model, we derive so-called reference CUI values for each airport capacity type as given by the number of runways (see Table 7.6). These reference CUI values represent maximum values of airports by the number of runways and decrease with an increasing number of runways. Values start at 84% for single runway airports and airports with independent runway systems, such as London Heathrow or Beijing airports, and reach a value of 61% for airports with eight runways, such as Chicago O’Hare. Furthermore, as a result of the model approach, values for airports with more than eight runways can be computed as well. This is particularly useful in a forecast scenario, where adding runway capacity seems to be appropriate. This will be the topic of Chapter 8, Modelling future airport capacity enlargements and limits. Based upon the maximum or reference CUI values and maximum average number of aircraft movements per operating hour, we can calculate the modelled maximum 5% peak hour volumes by airport capacity class, which serve as a proxy for the hourly capacity of the runway system, as already discussed in Chapter 2, Concepts of capacity and methods of estimation, and Chapter 3, Capacity utilisation at airports worldwide, in this book. Thereby, it is possible to conduct a comparison of the model results of the DEA approach and the capacity values of Tables 2.2 and 2.3 of Chapter 2,

182

PART | II Models for assessing mitigation strategies

Concepts of capacity and methods of estimation [excluding Memphis airport (MEM), as it is mainly a cargo airport]. Tables 7.7 and 7.8 show the comparison results. For this purpose, we have added three columns to Tables 2.2 and 2.3. One for modelled 5% peak hour volumes and another for the current capacity utilisation as obtained from the DEA (Table 7.1 and Fig. 7.2). The last column is a more or less subjective evaluation of the results. To calculate the 5% peak hour volume for each airport, we divided the estimated annual airport capacity, which we have obtained from the DEA, by both the number of operating hours and the reference CUI value. Comparing observed with estimated volumes and capacity utilisation rates, we have grouped modelled 5% peak hour volumes and capacity utilisation based on the DEA in Tables 7.7 and 7.8 according to the following rather ‘soft’ rules: G

G

G

Good: Modelled 5% peak hour volumes are more or less within the capacity ranges and close to the declared capacity, respectively. Sufficient: Here, modelled 5% peak hour volume is significantly higher than the capacity ranges and declared capacity. However, these values are still acceptable, as capacity utilisation is also rather low. Therefore we assume that there are still long-term capacity reserves which the airport can use in the future. Low capacity ranges and declared capacity ranges may be the result of issues such as a lack of terminal or apron capacity, environmental constraints or just a current lack of demand. Examples include airports such as San Francisco (SFO, four crossing runways) and Du¨sseldorf (DUS, two dependent runways, with administrative restrictions). Modest: In the case of Denver airport (DEN), modelled 5% peak hour volume is significantly lower than compared to the capacity ranges of Table 7.7, even if we take the rather low capacity utilisation (63%) into account. As we have already seen in Fig. 7.5, the actual CUI value of DEN is rather low, which results in a larger spread between 5% peak hour and average hour volume.

Overall, the model produces quite reasonable 5% peak hour volume values compared to the capacity ranges of Table 7.7 and declared capacities of Table 7.8. Here, we have to consider that the 5% peak hour volume is a good proxy of hourly airport capacity, but not the highest peak hour volume an airport can handle; the 5% peak hour volume is rather a good representation of the practical hourly capacity (see Chapter 2: Concepts of capacity and methods of estimation). Furthermore, peak hour volumes show a high degree of variation, depending on which peak hour we choose, as we have seen from the analysis of traffic ranking curves. This variation increases with decreasing CUI values, as the traffic ranking curve becomes steeper. This is a particularly important point for large airports with a high number of

TABLE 7.7 Comparison of the modelled 5% peak hour volumes with the hourly capacity rate ranges of the airports of Table 2.2. Airport identifier and name

ATL

Atlanta Hartsfield Jackson

Aircraft operations (arrivals and departures) per hour Visual

Marginal

Instrument

216 226 (AP)

201 208 (AP)

175 190 (AP)

206 (DP)

166 169 (LIMC AP)

219 222 (DP)

Modelled 5% peak hour volume

Capacity utilisation based on DEA (%)

Result evaluation

170

100

Good

168

41

Sufficient

183 186 (DP)

168 179 (LIMC DP) BOS

Boston Logan

116 125

109 112

84 86

BWI

Baltimore Washington

68 80

64 80

62 64

104

39

Sufficient

CLT

Charlotte Douglas

176 182

161 162

138 147

144

72

Good

DCA

Ronald Reagan Washington

69 72

69 72

54 64

109

53

Sufficient

DEN

Denver

262 266 (AP)

224 279

224 243

178

63

Modest

266 298 (DP) DFW

Dallas/Fort Worth

226 264

194 245

170

187

74

Good

DTW

Detroit Metropolitan Wayne County

178 184

163 164

136

178

43

Good (Continued )

TABLE 7.7 (Continued) Airport identifier and name

Aircraft operations (arrivals and departures) per hour Visual

Marginal

Instrument

EWR

94 99 (AP)

76 84

Newark Liberty

Modelled 5% peak hour volume

Capacity utilisation based on DEA (%)

Result evaluation

68 70

101

70

Good

94 100 (DP) FLL

Fort Lauderdale Hollywood

74 82

66 72

56 66

93

44

Sufficient

HNL

Honolulu

117 120

91 105

60 77

135

23

Sufficient

IAD

Washington Dulles

150 159 (AP)

112 120 (AP)

108 111 (AP)

142

32

Good

156 164 (DP)

136 145 (DP)

125 132 (DP)

IAH

Houston George Bush

172 199

152 180

144 151

170

52

Good

JFK

New York John F. Kennedy

84 87 (AP)

85 86

74 84

125

60

Sufficient

90 93 (DP)

LAS

Las Vegas McCarran

122 128

106 111

78 83

134

48

Sufficient

LAX

Los Angeles

167 176

147 153

133 143

127

87

Good

LGA

New York LaGuardia

80 86

76 77

74 76

76

90

Good

MCO

Orlando

160 171

148 161

144

135

41

Good

MDW

Chicago Midway

64 84

64 74

52 70

170

23

Sufficient

MIA

Miami

132 150

132 148

100 104

122

47

Good

MSP

Minneapolis Saint Paul

156 167

142 151

114 141

136

52

Good

ORD

Chicago O’Hare

214 225

194 200

168 178

197

90

Good

PHL

Philadelphia

120 126

94 96

84 88

135

50

Sufficient

PHX

Phoenix Sky Harbor

138 145

108 109

96 101

103

67

Good

SAN

San Diego

48 57

48 52

48

47

66

Good

SEA

Seattle Tacoma

100 112

86 100

76 78

100

69

Good

SFO

San Francisco

100 110

90 93

70 72

130

57

Sufficient

SLC

Salt Lake City

148 150

138 140

114 120

106

42

Good

TPA

Tampa

113 150

95 115

90 95

105

27

Good

AP, arrival priority configuration; DEA, data envelopment analysis; DP, departure priority configuration; LIMC, low instrument.

186

PART | II Models for assessing mitigation strategies

TABLE 7.8 Comparison of the modelled 5% peak hour volumes with the declared capacities of the airports of Table 2.3. Airport (IATA code)

Declared capacity (aircraft movements per hour)

Modelled 5% peak hour volume

Capacity utilisation based on DEA (%)

Result evaluation

Single runway airports London Stansted (STN)

36 50

47

59

Good

Dublin (DUB)

38 46

45

75

Good

Stuttgart (STR)

42

48

38

Sufficient

London Gatwick (LGW)

38 50

44

100

Good

Geneva (GVA)

36

48

57

Sufficient

Airports with two intersecting runways Hamburg (HAM)

48

81

35

Sufficient

Warsaw (WAW)

38 42

79

33

Sufficient

Perth (PER)

38

72

24

Sufficient

Lisbon (LIS)

35 40

75

42

Sufficient

Airports with two close parallel runways Dubai (DXB)

57 65

63

90

Good

Manchester (MAN)

44 56

74

43

Sufficient

Berlin Tegel (TXL)

52

81

45

Sufficient

Nice (NCE)

40 50

81

41

Sufficient

43 45

79

50

Sufficient

a

Du¨sseldorf (DUS)

Airports with two independent parallel runways Munich (MUC)

90

92

71

Good

London Heathrow (LHR)a

79 89

94

91

Good

Oslo (OSL)

76 80

94

44

Good

Palma de Mallorca (PMI)

66

95

33

Sufficient

a Du¨sseldorf capacity is limited by administrative rules to a single runway capacity, London Heathrow has an annual movement cap of 480,000.

Modelling future airport capacity and capacity utilisation Chapter | 7

187

runways, as in these cases, CUI values tend to be lower, and thus the variation of peak hour volumes increases. When comparing modelled 5% peak hour volume with declared capacities of Level 3 airports, we see a rather good approximation with single runway airports and airports with two independent runways, however, rather high estimates for airports with intersecting and two close parallel runways. It seems that declared capacities in these cases represent the rather cautious results of capacity finding discussions in coordination committees, as can be seen by comparing declared capacities with those of single runways. The values of airports with intersecting runways are barely higher and even the values of airports with two close parallel runways do not differ greatly from single runway airports. Annual airport capacities as a means of annual service volume are computed on the basis of maximum average hourly volume (per airport or per runway for a given number of runways) in our approach. Since traffic ranking curves typically run much flatter around the average hourly volume of an airport than around peak hour volumes, even for lower degrees of capacity utilisation, we expect the model to perform better than Tables 7.7 and 7.8 suggest.

7.4

Conclusion

While a detailed airport capacity assessment produces more precise results, but lacks forecast and large-scale application capabilities, the purpose of this chapter was to present a robust approach to compute annual capacities as annual service volumes of airports worldwide that have an adequate precision. Thus, the approach presented is highly problem-specific and cannot be a substitute for a detailed airport capacity assessment in airport-specific studies. However, obtaining annual service volume of airports worldwide is a prerequisite for reflecting limited airport capacity in a global air traffic forecast. The next step is to consider how airport capacity evolves in future, which is the topic of the next chapter. The chosen approach is based upon key ideas of Chapter 3, Capacity utilisation at airports worldwide, and Chapter 4, Constrained and under-utilised airports, such as ranking traffic curves, 5% peak hour and average hourly volume, which lead to the concept of the CUI. If an airport operates close to its maximum technical capacity during peak hours, the CUI can serve as an indicator of what share of that capacity (approximated by the 5% peak hour volume) can be utilised realistically over the course of a year on average, taking the demand side and complexity factors, such as air traffic control and interactions between runways into account. Utilisation rates have been observed to lie between 59% and 86%, depending on the runway system. As a result, the CUI is a link between a rather technical view of airport capacity,

188

PART | II Models for assessing mitigation strategies

that is peak hour volume, and the demand side of air traffic and complexity factors, that is average hourly volume. However, we had to extend these ideas, because in many cases, 5% peak hour volumes have been found to be well below peak hour capacity. Furthermore, the maximum value of the CUI tends to depend on the runway system. This makes classical regression analysis difficult. Therefore we expanded our approach by incorporating DEA and conducted a regression analysis based upon the DEA results. In contrast to the CUI, capacity utilisation rates in the DEA are based on average hourly traffic volumes per airport and runway, respectively; hence, values of highest possible capacity utilisation can be reached. Nevertheless, 5% peak hour volumes can be modelled additionally for comparison purposes. This model approach enables us to compute current and future annual service volumes of airports worldwide. To assess the model, we have calculated modelled 5% peak hour volumes and compared them with capacity rates of US airports and declared capacities of mainly European airports. Here, the model performed satisfactorily. Furthermore, while the analysis of Chapter 3, Capacity utilisation at airports worldwide, and Chapter 4, Constrained and under-utilised airports, is about the current traffic situation, we can now forecast 5% peak hour volume and average hourly volume in a situation of the highest possible capacity utilisation.

References Berg, S., 2010. Water Utility Benchmarking: Measurement, Methodology, and Performance Incentives. International Water Association. Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the efficiency of decision-making units. Eur. J. Oper. Res. 2 (6), 429 444. Cooper, W.W., Seiford, L.M., Tone, K., 2007. Data Envelopment Analysis—A Comprehensive Text With Models, Applications, References and DEA-Solver Software. Springer, New York. Digital Aeronautical Flight Information Files (DAFIF), 2016. US Government, Washington D.C. Evening Standard, 2016. Heathrow Airport Bosses Propose Raising Cap on Flights to 500,000 a Year. Evening Standard, 9/29/2016. ,https://www.standard.co.uk/news/politics/heathrowbosses-propose-raising-cap-on-flights-to-500000-a-year-a3357071.html#comments. (accessed 09.07.18.). Gilbo, E., 1993. Airport capacity: representation, estimation, optimisation. IEEE Trans. Control Syst. Technol. 1 (3), 144 154. Greene, W.H., 2011. Econometric Analysis. Pearson Education, Harlow. IATA, 2016. Worldwide Slot Guidelines (WSG)—Annex 11.6. Montreal. Official Airline Guide (OAG), 2016. Market Analysis. Reed Travel Group, Dunstable. Simm, J., Besstremyannaya, G., 2016. Robust Data Envelopment Analysis (DEA) for R. CRAN. ,https://cran.r-project.org/web/packages/rDEA/rDEA.pdf..