An Operations Research Model For The Evaluation Of An Airport Terminal: SLAM (Simple Landside Aggregate Model)

Lorenzo Brunetta Dipartimento di Elettronica e Informazione, Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milano ITALY Tel. 39-02-2399-3509, Fax. 39-02-2399-3412 email: brunetta@elet.polimi.it Luca Righi and Giovanni Andreatta Dipartimento di Matematica Pura ed Applicata, Universit degli Studi di Padova, via Belzoni 7, 35131 Padova, ITALY Tel. 39-049-827-5983, Fax. 39-049-8758596 email: righi@ math.unipd.it , andreatta@math.unipd.it The Simple Landside Aggregate Model (SLAM) is a model for estimating capacity and delays in airport passenger terminals. SLAM is designed to answer what if questions about alternative configurations of the various processing and holding facilities in a terminal. It consists of a network of modules, one for each facility of the terminal. These modules are based on a set of quite simple mathematical formulas to be used for the estimation of the capacity of each facility (in terms of passengers per hour) and the level of service (LOS) associated with it. LOS is quantified both in terms of space available per facility occupant and waiting time for being processed. Key Words: Strategic, tactical and aggregate models, level of service. The modelling of airport terminal operations has advanced significantly over the last 15 years (Tosic, 1992). Available models have improved in detail and fidelity, as well as user friendliness. As a result, their use as decision support aids or design tools in terminal development projects has been steadily increasing. Some existing models are strategic in nature sacrificing level of detail in exchange for speed and flexibility, while others are primarily tactical incorporating high levels of detail in data and system definition. The lack of a satisfactory strategic model of landside operations motivated the creation of the Simple Landside Aggregate Model (SLAM), see (Odoni and de Neufville, 1992). In this paper we describe the creation and the application of the new model SLAM to the airport terminals of Milan. SLAM is an analytical aggregate model for estimating capacity and delays in airport passenger terminals. SLAM consists of a network of modules, one for each facility of the terminal: these modules are based on a set of quite simple mathematical formulas. Their objective is not to provide a thorough analysis of a given facility, but to help in the estimation of the capacity of the facility (in terms of passengers per hour) and the level of service (LOS) associated with it (compared to internationally accepted standards as those, for example, in (Ashford, 1988) or in the International Air Transport Association manual (IATA, 1995)). SLAM is equipped with an interactive GUI and provides, among its outputs, colour-coded displays of LOS achieved at each landside facility over any specified period of time. SLAM is part of an integrated model called TAPE (Total Airport Performance Evaluation). The TAPE model is an integrated model of both landside and airside, and includes aggregate and detailed models.

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 2

The TAPE concept has been applied to the Linate airport and to the scenarios of the new Malpensa 2000 airport that we will describe in the next sections.

1. Basic Definitions The identification and classification of landside elements refers mainly to the Special Report of the Transportation Research Board (TRB, 1987). Tosic gives an excellent review of airport passenger models (Tosic, 1992). Odoni and De Neufville propose methodological issues in passenger terminal design (Odoni and De Neufville, 1992). Both papers were published in a special issue of Transportation Research in 1992, which contained other useful references. The airport landside includes the passenger terminal with all its components. We consider only functional components, i.e., elements providing services or amenities directly related to a passenger boarding or unboarding an aircraft. Non functional components such as concession areas, rest rooms, and telephones, although important passenger amenities, are not a basis for defining airport landside capacity. In particular, we examine all facilities and services associated with an air passengers ground based journey from entrance to the terminal building to boarding aircraft (departing passenger) and from getting off a plane to leaving the airport. Terminal curb and parking are often, but not here, considered being parts of landside. The ground access system is simply modelled here as a set of sources and sinks of passengers from the terminal (arriving passenger) or to boarding another aircraft (transit passenger). The passengers perception of the quality and conditions of service of one or a set of functional components constitutes the service level. Standard measures of the service level of components are waiting time, processing time, walking time, crowding, and availability of passenger amenities for comfort and convenience. A high level of service may be provided if the airport landside has ample capability to accommodate passengers, baggage, and airport visitors. This airport landside capability is, of course, influenced by the capacity (in terms of persons processed per unit of time) of the facilities in the terminal. Capacity can be evaluated for each individual functional component of the airport landside. One or more of these components are likely to become the bottlenecks of landside capacity, i.e., the major constraints on serving additional passengers at the terminal. Let us introduce some basic definitions. We define the dwell time as the average time a person is in a place or in a process. A principal component of dwell time is the amount of slack time that airport users spend in the various parts of the terminal. We call slack time (or discretionary time) the free time a passenger has to spend at a terminal, i.e., the time not spent being processed or waiting to be processed at functional components or moving from one functional component to another. We will often speak of the peak hour, i.e., a representative hour of busy conditions within a functional component. A peak hour is typically defined from historical records by frequency of occurrence. In fact, it may be the thirtieth or fortieth busiest hour of the year, or the average daily peak hour of the peak month, or the peak hour of the 95-percentile busy day. Following the description of each facility, we will deal with the demand patterns, i.e., the number of passengers and characteristics of their behaviour that materially influence the ability of a functional component (or group of components) to accommodate them. Such factors, as the time when passengers arrive at the airport, their age, their trip purpose, the fare paid, the baggage carried or checked, and whether the passenger has a ticket and boarding pass, are often important.

3 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

2. Classification of Landside Elements Following the analysis presented in (TRB, 1987), landside elements may be subdivided into three classes: Processing facilities: they process passengers and their luggage. Holding facilities: areas in which passengers wait for some events (as the check-in opening for a flight, the start of flight boarding, etc). Flow facilities: the passengers use them to move among the landside elements.

2.1. Identification of LOS The level of service (LOS) represents the quality and conditions of service of one or more facilities as experienced by passengers. Interrelationships exist among the typical measures of service level such as waiting time, processing time, walking time, and crowding. Service level targets are important because of their serious implications for airport costs and economics as well as for the image of the airport. In fact, maintaining a particular level of service at an airport may contribute to attracting new business and is also a reflection of the local or national communitys goals. Each component of an airport landside has its own unique operating characteristics and demands; hence it is hard to define service level in a unique way. Research conducted by the International Air Transport Association (IATA) on Traffic Peaks led to the need of standard definitions for evaluating levels of service and airport capacity (IATA, 1981). In order to specify the LOS standards, the working group suggested that potential congestion should be measured in different ways, according to the type of airport landside facility involved, i.e., depending on whether one is dealing with processing facilities, holding facilities or flow facilities. According to the facility being analyzed, three fundamental measures of capacity can be used to estimate potential congestion. These are respectively static, dynamic and sustained capacity. They are defined as follows (Svrcek, 1994): Static capacity is used to describe the storage capability of a holding facility or area, and it is usually expressed as the number of occupants that a given area may hold at any one moment. It is a function of the total usable space available and the level of service provided, i.e., the amount of space each occupant needs to have. Dynamic capacity refers to the maximum processing rate or flow rate of pedestrians (i.e., occupants) through a subsystem per unit time. The actual time unit selected as the measurement index (e.g., minutes, hours) depends on the nature of the facility and operation involved. Sustained capacity is used to describe the overall capacity of a subsystem to accommodate traffic demand over a sustained period within the time and space standards of a particular level of service. It is thus a measure of the combined static and dynamic capacities of a facility.

To specify the LOS, a set of letters from LOS = A (best) to LOS = F (unacceptable), are used. In table 1 (IATA, 1981) the levels of service are described in terms of flow, delays and level of comfort.

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 4

Table 1. IATA LOS standards Los A B C D E F Level Excellent High Good Adequate Inadequate Unacceptable Description Free flow, no delays, excellent comfort level Stable flow, very few delays, high comfort level Stable flow, acceptable delays, good comfort level Unstable flow, passable delays, adequate comfort level Unstable flow, unacceptable delays, inadequate comfort level Cross-flow, system breakdown, unacceptable comfort level

Note that although the description of each individual service level remains the same, subsystems have different spatial requirements. System managers and designers should specify the desired or required level of service. Usually, level C is recommended as a minimum and level D is considered tolerable for crash periods. Multidimensional LOS standards should be specified for processing facilities; for example, in the IATA manual (IATA, 1982) for a check-in area LOS standards are specified both in terms of time (maximum waiting time acceptable) and of space (the area that should be available per waiting passenger). In table 2 we provide the LOS landside parameters used by Societ Esercizi Aeroportuali (SEA, the Milan Airport Authority) that are common to the three scenarios analyzed in Section 7. The IATA standards are used by many Airport Authorities, but are not universal, as some airport authorities have developed their own standards, while others (especially in developing countries) utilize no LOS standards at all. For instance, the British Airports Authority and Aroports de Paris have developed their own space standards for holding facilities and processing facilities (Ashford, 1988). Table 2. LOS parameters LOS AREA Baggage Claim (sq. m.) Flow-space (p. p. sq. m.) Check-In (sq. m.) Holding area (sq. m.) Generic area (sq. m.) A 2.00 20.00 1.80 2.70 1.40 B 1.80 25.00 1.60 2.30 1.20 C 1.60 40.00 1.40 1.90 1.00 D 1.40 57.00 1.20 1.50 0.80 E 1.20 75.00 1.00 1.00 0.60

The following pages contain a description of all the airport facilities and a survey of the various techniques proposed to address the simulation of their operations: a reader that is already familiar with the problem and related literature can skip them. 2.2. Processing facilities 2.2.1. Ticket counter and baggage check-in The operation begins when a passenger enters the queue to obtain the boarding pass and checks his baggage, and ends when the passenger leaves the ticket counter area.

5 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

It has to be noted that the (average) processing time at any particular airport depends on many factors (staff experience, flight market, and passenger characteristics) as well as on airline operating policies (i.e., number of active counters). Processing time variance can also be large. In particular, the processing time can become much longer than average if any problem arises, and, for this reason, most airlines introduce a separate desk for passengers with particular problems. Capacity of check-in processing facilities is judged by considering the average service time and by comparing the number of passengers in a terminal holding area with the size of that area. Some of the analytical models proposed in the literature for check-in counters belong in the class of Queuing Theory Models. This is also the case for most of the other processing facilities in airport operations. Lee proposed a pioneering application of M/M/n queuing systems to check-in procedures (Lee, 1966). Newell initially proposed a deterministic approach (Newell, 1971). This model had a strong influence on further developments in this area, and it has applications in modeling several types of facilities where service is provided to individuals by a processor of some kind; see (Piper, 1974), for an example of practical application. Basically, this is a graphical model that computes approximately the total waiting time of passengers, given the cumulative arrival function at the check-in counter and the service rate for each time period. This simple and effective model has also been extended (to representing more than one flight) in (Tosic et al., 1983). A simulation model based on the Monte Carlo method has been presented in (Tosic et al., 1983). Being a simulation model it needs detailed data, and provides quite realistic information on the behaviour of check-in counters. 2.2.3. Passenger security screening Originating passengers must undergo a security screening operation. Sometimes transfer passengers also have to pass through security screening while moving to a connecting flight. For this reason, security-screening areas are often elements of queuing and delay for passengers. The security screening, among the many components of the airport landside, is the one that most closely fits the assumptions of a simple queuing model. The average time required for clearance of a passenger, the variability of that time, and the rate of passenger arrival at the security screening area are key variables for its capacity assessment. The security screening is typically a single (or multiple if there is more than one channel) service counter facility and can be simply modelled by a queuing model. The level of service of the passenger security screening area is mainly influenced by the delays associated with waiting in queue for clearing security screening. Both stochastic and deterministic queuing models have been proposed in the literature. Examples of application of the stochastic models are in (Rallis, 1958, 1963, 1967). In particular, the Copenhagen terminal building was analyzed by applying M/D/n queuing systems. Newell proposed a deterministic model by means of graphical analysis using cumulative diagrams of number of passengers versus aircraft departure time (Newell, 1971). 2.2.4. Passport Control Generally, passengers departing for, or arriving from, an international flight must clear Passport Control and Customs formalities. Typically, all passengers go through passport inspection (and sometimes through customs) both on departure and on arrival; furthermore, on arrival, foreign citizens

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 6

must first clear immigration, while other passengers proceed directly to baggage claim and then to customs. Immigration and passport inspection can be combined, however procedures differ from airport to airport within the European Union. Usually separate facility inspections are provided for Schengen (SC) and Non-Schengen (non- SC) citizens. Walking speeds and distances from check-in to inspection areas and from arrival gates to the inspection areas determine the distribution of actual passenger arrivals. According to these aspects, it is possible to identify the typical demand and operating factors influencing service level and capacity. These are: number of channels, personnel and amount of space; the average processing time per passenger and the efficiency rate of selection for close inspection policy; the passenger characteristics, that is, the fraction of foreign passengers departing/arriving on international flights, baggage loads, the available queue space. The capacity value for passport control facilities is judged by considering the average time passengers must wait to have their passport checked (and to clear immigration formalities), and by comparing the number of people in the queue area with the size of that area. The limited availability of inspectors may produce low level of service; in such cases the airport operator is requested to provide additional space for waiting passengers and to reduce passenger inconvenience. The inspection service system is reasonably well represented by a mathematical multichannel queuing model and, for most purposes, direct observation of conditions and simple calculations of average delays and queue sizes are adequate for capacity assessment.

2.2.5. Customs Generally, passengers arriving from an international flight must clear Customs formalities, after Passport Control. These include baggage inspection and collection of duties on certain exported (imported, on arrival) items. The limited availability of inspectors may produce low level of service; in such cases the airport operator is requested to provide additional space for waiting passengers and to reduce passenger inconvenience. Continuing efforts to streamline arrival procedures and to improve speed and accuracy in customs screening may relieve problems at airports experiencing rapid growth in international traffic. Mathematical multichannel queuing models can well represent the inspection service system and general purpose simulation procedures may be applied to specific cases. For capacity assessment direct observation of conditions and simple calculations of average delays and queue sizes are appropriate.

2.2.6. Gates Airports in Europe usually operate gates on a common-use basis, in which the assignment of aircraft to gates is entirely an airport operators decision, while, in the US, aircraft gates may be operated on an exclusive-use basis. A lot of models for gate assignment have been proposed. Some of them take into account both the type of aircraft and the passenger walking distances. Basically, they are based on a gate assignment with first in - first out (FIFO) rule (Le et al., 1978) and (Hamzawi, 1986). Babic et al. proposed a method to minimize passenger walking distances by properly assigning aircraft to gates every day, taking into

7 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

account passenger flows on that particular day (Babic et al., 1984). Mangoubi and Mathaisel incorporated transfer passengers in their formulation of the flight-to-gate assignment problem (Mangoubi and Mathaisel, 1985). Both approaches assume that a specific configuration is given so that walking distances are known and fixed, and, therefore, these models are appropriate at the tactical level. Wirasinghe and Vandebona proposed a long-term planning model (Wirasinghe and Vandebona, 1987). As for gate position requirements, Bandara and Wirasinghe proposed a way for determining the gate position requirements based on a deterministic model (Bandara and Wirasinghe, 1989). Edwards and Newell investigated stochastic models of gate utilization (Edwards and Newell, 1969). Steuart proposed a different stochastic model (Steuart, 1974).

2.2.7. Baggage claim facilities Baggage claim is the most critical step of the inbound baggage system. The number of passengers waiting in the baggage claim depends on the rates at which passengers arrive from the gate and luggage is processed. In general, the maximum demand levels occur when larger aircraft arrive. The baggage claim area capacity can be measured considering the average time passengers must wait to retrieve their checked baggage and comparing the number of people in the claim area with the size of that area. The number of passengers claiming baggage must be calculated from schedule forecasts. In general, the linear dimension of the device is determined on the basis of the number of passengers, rather than of baggage, except in some cases in which baggage ratio is very high. The expected average time passengers have to wait for bags and the number of waiting passengers in the claim area can be computed by simple queuing models. In the literature, mathematical queuing and simulation models have been developed to predict the arrival (of deplaning passengers and baggage) to baggage claim areas, and to forecast possible future conditions. In (Horonjeff, 1969) and (Barbo, 1967) a deterministic queuing model was developed to relate the arrival distributions of passengers (and the arrival distributions of baggage) to the number of passenger bags that are on the carousel at a given time. Browne et al. studied the baggage claim areas of the JFK airport in New York (Browne et al., 1970). Their objective was to compute the expected maximum inventories of passengers and bags using inventory type models. Newell analyzed a baggage claim device and proposed a two queues system, one for passengers waiting for bags, the other for bags waiting for their owner (Newell, 1971). The problem was to estimate the number of passengers waiting in front of the devices for their bags. Tosic et al. proposed a Monte Carlo type simulation model to evaluate the elements of the baggage claim area (Tosic et al., 1983). In this model each passenger and all his/her bags are treated individually.

2.3. Passenger holding areas Passenger holding areas are spaces where passengers move around and wait for flight departures and arrivals. These facilities include lobbies, gate lounges, transit passenger lounges, baggage claim area, the arrival area, the area set aside for ancillary facilities, etc. The number of waiting passengers is a function of the number of aircraft served by the holding area, and their functional characteristics, including capacity and loading factors. The number of passengers simultaneously waiting in the terminal is also influenced by other important factors, such as passenger

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 8

arrival time at the airport, degree to which passengers are accompanied by family or friends, and the length of time between commencement of boarding of a flight and its departure. The amount of time spent in a particular area, that is a fraction of passenger dwell time, is central to determine the number of simultaneous occupants of a given area (Odoni and de Neufville, 1992). Dwell time is mainly caused by the amount of slack time that passengers spend in the various parts of the terminal building. This slack time is in turn allocated among the terminal holding areas. Clearly the loading, that is the number of simultaneous occupants, depends on the fraction of the slack time spent in that area. This discussion applies both to departing and transit; for arriving passengers the concept of slack time is less important because they try to leave the airport as soon as possible. Stochastic models for estimating dwell time are presented in (Odoni and de Neufville, 1992). 2.4. Flow facilities The total time spent by a passenger to cross the terminal building from its entrance point to the gate is the sum of the waiting and service times in the processing facilities plus the sum of the times required to move from a service station to another. Large airport terminals with multiple gate positions necessarily involve large internal transfer distances. Mechanized circulation aids are commonly used to improve circulation in large terminal buildings. In airports with multiple terminal designs (e.g., Paris Charles de Gaulle), and remote satellites (i.e., London Gatwick), the distances can be so large that mechanized movement becomes essential. The terminal circulation component may be seen as a flow pedestrian problem and analyzed by using procedures and standards such as those suggested in (TRB, 1985). The time required to travel from the curb to the gate is the most important measure of service level.

3. Models for Landside We will next review some well known simulation tools for modeling landside airport operations. We should mention that all these tools have large data requirements and lack of flexibility, they need powerful hardware platform and a lot of CPU time in order to complete an analysis. Mumayiz has provided an overview of several major airport landside simulation models (Mumayiz, 1990). Discreteevent, time-dependent simulations are most commonly used. They are primarily developed by universities for academic research, by industry for proprietary purposes and by the government. At the University of Texas at Austin, Dunlay et al. developed simulation models to evaluate landside terminal capacity, which led to the ACAP airport simulation model (Dunlay et al., 1975). In Florida, the AIRSIM event-oriented simulation model was developed to deal with individual passenger and baggage flows according to a fixed flight schedule. Other simulation models were developed by universities: e.g., in Denmark, for the quantitative evaluation of basic design strategies in the expansion of the Copenhagen Kastrup Airport (Mumayiz, 1990). Finally, government agencies such as the Federal Aviation Admistration (FAA), the Canadian Air Transport Administration and the British Airports Authority sponsor the development of their own simulation models. For example, the FAA participated in the development of ACAP (FAA, 1988). SABRE Decision Technologies is a consulting company that has participated in terminal design projects at several airports in different countries. The methodology proposed by SABRE is based on very

9 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

detailed simulation models. The company has developed a series of tools for the evaluation of various landside operations (SABRE, 1994). The applications are customized in order to model individual airport landside facilities, and by simulation it is possible to obtain quantitative measures of queue lengths and of passenger waiting time for all service areas (ticket, check-in, security, passport, immigration, customs, etc.). Corridors, people mover systems and baggage handling systems can also be modelled. Various statistics can be used for the evaluation of the capacities of the elements under different scenarios. The documentation available to us specifies neither the requirements for implementing the models nor the technical characteristics of the input and output tools of these packages. Detailed simulation analysis tools have been developed by SABRE for the San Francisco International Airport, for the Greater Orlando Aviation Authorithy, for the new International Airport Irkutsk-Novy in Russia, for London Heathrow, for Berlin Schonefeld Airport, etc. The Airport Terminal Building Simulator (ARTS) is a discrete-event simulation model of the actions and decisions made by individual arriving, departing and transfer passengers in the airport terminal building, developed by Argo Research (Pararas, 1995). The basic feature of ARTS is the flexible implementation of behavioural models representing the way passengers make decisions in the terminal building on one hand, and of the decision policies related to the operations in the terminal facilities on the other. ARTS is a new generation model that overcomes, on a low-cost platform, typical problems of the old packages, such as large data requirements and lack of flexibility. ARTS stresses flexibility in the definition and implementation of alternative terminal building operating policies and in the modeling of passenger behaviour. ARTS has been designed to be extensible (so that, in addition to the set of built-in behavioural models provided with the system, new policies can be incorporated seamlessly) and, at the same time, flexible enough to allow multiple, and possibly conflicting, policies to co-exist. Unfortunately it still takes many months of training to produce an input that leads to a successfull simulation. The formulas presented in the IATA manual (IATA, 1982) are contained in an aggregate model that have been recently implemented and made available to users by IATA (IATA, 1995). 4. Proposed Aggregate Models In the following outline of aggregate models for each facility of the terminal, we will refer to the classification proposed in section 2, presenting first the models for the processing facilities, then those for the waiting facilities, and finally those for the flow facilities. We must point out that the models of this section are intentionally simple: the output produced by an aggregate model must be easy to understand and very fast to obtain. This choice is reasonable, since a detailed analysis can always be provided by a detailed model. The input requested by the models is extracted from the data, usually collected by every airport Authority, that are typically provided to a detailed model. For evaluating a processing facility we need a criterion that is bidimensional, i.e., a criterion that simultaneously takes into account both time and space. Time standards refer to the time spent in the facility by a given percentage of the passengers, while the space standards consider the amount of space per person that is available. For evaluating a holding facility only space standards are used and finally, for evaluating a flow facility, one has to consider the number of passengers that can cross a section of the facility per unit of time. Let us introduce a variable that we will call Index of Service (IOS), strictly related to the level of service (LOS). The LOS is a qualitative statement, represented by a single letter (A to F). To most of the LOS there correspond internationally accepted standards (quantitative measurements). We will call

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 10

index of service (IOS) these quantitative measurements. For example, in a waiting lounge the LOS = B corresponds to 2.3 < IOS < 2.7 (m2 per person). Typically, the aggregate model for a specific facility will consist of a simple formula, like the following: IOS = Area AP ADT

that says that the index of service (IOS) for that facility can be computed dividing the Area by the product of the number of Arriving Passengers (AP) at that facility during one hour (the Peak Hour) times the Average Dwell Time (ADT) spent by a passenger in the facility. The IOS can then be used to obtain the LOS of that facility. For example, if the Area in front of the check-in is 1500 m2, the number of passengers arriving at the check-in during the Peak Hour is 3600, and the average Dwell Time is 0.15 (hours), then the IOS for that facility is 2.78 (m2 per person), which means that the corresponding LOS is A. The example illustrates how to obtain the LOS for a given facility. The same formula can be used to answer other questions, like: Given AP and ADT, and assuming a specific target LOS, what is the (minimum) Area that should be reserved at that facility? Given the Area and AP, and assuming a specific target LOS, what is the (maximum) allowable ADT? For example, if at a Passport Control facility AP = 2100, ADT = 2 minutes (0.033 hours), and we want to achieve a LOS = C, then the Area is given by: Area = IOS AP ADT = 1.0 2100 0.033 = 70.0 m2. In the sequel, when we present a formula that gives the IOS as a function of other variables, it should be understood that from that same formula any variable could be obtained as a function of the others.

4.1.1. Computing Dwell Times in a processing facility In this subsection we describe quick and dirty methods to compute the Dwell Time (both its average and its distribution) at a processing facility. We recall that the Input required by our model can be extracted from the statistical data that are typically available to an airport manager and that our analysis refers to the peak hour (PH). However, the time window to consider is typically greater than one hour, since we have to take into account all the flights departing or arriving that can possibly interact with the PH; for example, a check-in counter at the Linate Airport is usually opened two hours and fifteen minutes before the scheduled departure time. In order to estimate the Average Dwell Time (ADT) spent by a passenger in a processing facility, we consider two different approaches. The first one is based on classic Queuing models (M/M/s or similar) and provides a reasonable approximation of ADT under the assumptions that AP, the average number of customers arriving to the processing facility, and the average potential service volume of that same facility (let it be s ) can both be considered approximately constant over a significant period of time. The main drawback of this approach is that it is difficult to obtain the steady state, i.e., AP must be

11 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

strictly lower than s . Of course, this approach will not be able to take into account the dynamic effects of variations over time of AP or s . The second approach is suggested when these dynamic effects are too important to ignore. It utilizes a deterministic equivalent approximation that will follow exactly the evolution over time of AP and s . Basically, this is a graphical model, that computes approximately the total waiting time of passengers, given the cumulative arrival function at the check-in counter and the service rate for each time period. As we said, it was initially proposed by Newell (Newell, 1971) and extendedto representing more than one flightby Tosic et al. (Tosic et al., 1983). In the second approach (let us call it Deterministic Equivalent Approach), we may (under)estimate the Dwell Time for each processing facility by considering the passenger arrival profile and the profile of the number of passengers served, as functions of time. In the following, for the sake of clarity, we shall refer to the check-in facility, instead of considering a generic processing facility. For each flight, the passenger arrival profile (which must be given as input) is a function of time that provides the number of passengers that have already arrived in the system (i.e., the check-in facility). The profile of the passengers that have been served by the system (and therefore have left it) is again a function of time, but it also depends on the number of servers; this profile is not given as input, but can be inferred from the number of servers which are open and from the mean service time. The number of servers opened by a given air carrier is sometimes conditioned upon the carriers target level-of-service standards.

Number of passengers

A(t)

time spent

D(t)

150

120

90

60

30

Minutes before departure of flight

Figure 1. Processing facility queuing

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 12

Let A(t) be the number of passengers that have arrived at the facility up to time t, and D(t) the overall number of passengers that have already left the facility by time t. Of course, A(t) and D(t) are nondecreasing functions. Passenger profiles can be properly approximated by piece-wise linear functions (we represent time on the x axis and number of passengers on the y axis). Furthermore, the combined arrival profiles of the passengers of all flights assigned to the same Check-In counter (or block of counters) can be summed up by using the arithmetic of the piece-wise linear functions, thus producing an overall piece-wise linear profile. It follows that we can approximate A(t) and D(t) by piece-wise linear functions. In figure 1 we represent a hypothetical A(t) and D(t) in the case where a single flight is assigned to a given counter. If a passenger is the n-th passenger to enter the system (let us call him/her passenger n), then his/her Dwell Time DT(n) can be computed as follows, under the natural assumption of a FIFO discipline: DT(n) = D-1(n) -A-1(n), where A-1(n) and D-1(n) are the inverse functions of A(t) and D(t). Considering A(t) and D(t) as piecewise linear functions, their inverses are again piece-wise linear functions (and so is their difference). Since in our simulation we realized that the dynamic effects of the system are too important to ignore and that it is difficult to obtain the steady state in a terminal system, we adopted the Deterministic Equivalent Approach in the software implementation of the SLAM model. The previous paragraph illustrated some situations where the arrival pattern of patterns at certain facilities cannot be considered as random. The random arrival pattern (or at least something closely approximating to it), however, is the on most commonly met, and we investigate it here in more detail. How then can we calculate what the peak is likely to be for such a portion of the peak hour? Random arrival patterns will often follow a Poisson distribution, which, when large numbers are involved can be approximated by a normal distribution. Therefore, we firstly establish the mean . In the normal distribution the mean and the variance are equal by definition, the standard deviation is the square root of the variance, . A further property of the Normal distribution is that (approximately) 95% of all observations will be within the limits of the mean 2 times the standard deviation. This means that there is only the 2.5% chance that the throughput during any peak period will exceed +2 . By fully exploiting the geometry of piece-wise linear functions and finite automata, SLAM extends the results by Tosic et al. It can provide information (rather unusual for aggregate models) such as the graphs, as functions of time, of the following quantities: cumulative number of served passengers, number of passengers in queue, number of passengers in queue per counter, number of counters together with number of passengers in queue per counter, number of counters together with expected queue time and an optimal allocation of the facility resources to reach a required level of service. Of course we also estimate, during each period of interest, the average dwell time, the average waiting time, and the space and time LOS.

13 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

4.1.2. Computing Minimal Landside Times for passengers and baggage For a departing passenger, the minimal landside time is given by the sum of the times (dwell time plus service time) spent at all processing facilities, plus the time needed to move along the various paths from the terminal entrance to the gate exit through the required facilities, plus the time needed to get from the gate to the airplane and board it. This minimal landside time can therefore be easily computed. It may vary with the time of the day because of congestion periods at some facilities and it may be different according to the particular flight of interest. If it is too long, a delay may result for the corresponding flight. We consider separately departing baggage and transfer baggage. In the case of departing baggage, the minimal landside time is given by the sum of the following quantities: time to go from the terminal entrance to the appropriate check-in counter, dwell time spent at check-in, service time spent at the check-in, transport time from check-in to the outbound baggage hall, dwell time plus service time to be security screened (this is zero if no security screening is in place), dwell time plus service time to be properly sorted (this time may be negligible if a highly automated handling system is utilized and can be of significant magnitude if a manual or other slow handling system is utilized), transport time from the baggage hall to the apron stand and finally the time required for loading the baggage into the aircraft. Again this minimal landside time can be easily computed. It may vary during the day and it may depend upon the particular flight considered. If it is too long, a delay may result for the corresponding flight. In the case of transfer baggage, let us consider a banking situation where a number of flights arrive in a narrow time window, passengers and baggage are transferred and then another set of flights will depart. In this case, the minimal landside time required by baggage arriving with flight A and departing with flight D is the sum of: time for unloading from aircraft A; transport time from the apron of arriving flight A to the baggage handling center; sorting time (including security screening if needed); transport time from the baggage handling center to the apron of departing flight D; and time for loading the baggage into aircraft D. The sorting time needs some further explanation. In fact, aircraft D will usually take off only if all its baggage is on board. This baggage may come from any one of the arriving flights in the bank, which means that all baggage of all arriving flights has to be sorted in order to be sure to carry on board all the proper baggage. This implies that in order to check if a flight can take off on time or if it has to suffer a delay due to late baggage handling, we can proceed as follows: let TS be the time when the last piece of baggage from the bank has been sorted. This can be computed by first considering the curve of the cumulative number of bags arriving at the baggage handling center as a function of time and then subtracting from it the curve of the cumulative number of bags that have been sorted (again as a function of time). TS is the first moment when the difference of the two curves permanently vanishes. A departing flight D in the bank will have to be delayed if and only if: TTO(D) - TS < TL(D) Where TTO(D) indicates the planned time for take off (leaving the blocks) of flight D, and TL(D) indicates the time required to transport the baggage from the handling center to the appropriate apron plus the time needed to load the baggage into the aircraft D.

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 14

4.1.3. Computing space requirements in a processing facility The input data for our model are: number of departing flights in the PH, time of departure of each flight, aircraft types, flight types, number of passengers on board, passenger arrival profiles (for each flight type), number of counters and service time. The number of persons arriving (AP) at the check-in area can be easily estimated considering the index of the last passenger minus the index of the first passeger arrived at the check-in during the interval under consideration (usually the check-in peak hour). The ticket counter can be modelled using the model discussed for the check-in facility. In particular, the time and space standards are assumed to be the same as for check-in. The model for the Security Check facility is like the one proposed for the check-in with the only difference that the average service rates are different. The input data of our model for Passport Control facility are: passenger arrival profiles (A(t)), percentage of passengers not belonging to Schengen Union countries (PNSC), number of counters (s(t)), that may vary over time, and service time distribution for non SC passengers (TNSC). The average service time (1/) of each server is obtained by computing the mean of the service distribution, 1/ = PNSC E(TNSC) and then obtaining the global average service time (ATg(t)) by: ATg(t) = 1/ [s(t) ]. The average service time computed with one of the above formulas, can be used to provide the profile of the passengers served up to time t ( D(t) = t / ATgSC(t), or D(t) = t / ATgNSC(t)). The Dwell Time can now be computed as in Section 4.1.1 while the space IOS can be calculated with the usual formula. The input data for Customs facility are: customs area; passenger arrival profiles (A(t)); percentage of passengers (Pgreen) going through green channel, i.e., nothing to declare; percentage of passengers (Pred) going through red channel; percentage of passengers (PCgreen) going through green channel, that are checked; service time distribution for green passengers (Tgreen) and for red passengers (Tred). Notice that the only passengers that experience any delay are those included in Pred and PCgreen. All the others suffer no delay at all. Since we have only one type of facility, the average service rate is obtained by computing the mean of the service distribution, times the percentage of passengers going through the red channel plus the percentage of passengers going through the green channel, that are checked, and then taking its reciprocal: = 1 / [ PCgreen E(Tred) + Pred E(Tred)]. The average service rate computed with the above formula, can be used to provide the profile of the passengers served up to time t (D(t) = t ). The Dwell Time can now be computed and therefore we may compute the average dwell time (ADT) as usual.

15 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

No default time standard is provided for customs. The default minimum space standard for the customs facility is 2 m2 per passenger interviewed. The Baggage Claim facility is the most difficult facility to model analytically. However this facility is extremely important in shaping passenger impressions of the level of service provided. We follow the model outlined by Odoni and de Neufville. The model is based on the assumptions that the passenger arrival time distribution (TP) and the baggage arrival time distribution (TB) at a baggage claim device is uniform (TP ~ U(tP1, tPL) and TB ~ U(tB1, tBL)). We recall that the mean of a uniform distribution (T ~ U(t1, tL)) is E(T) = (t1 + tL) / 2, and its variance is V(T) = (tL - t1)2 / 12. The input parameters for these models are: area, the passenger arrival time distribution (TP), the baggage arrival distribution (TB), the time of arrival of the first and the last passenger at the baggage claim device (tP1, tPL); the time of arrival of the first and the last baggage at the baggage claim device (tB1, tBL); average number of pieces of baggage per passenger with at least one piece of baggage (nbag). We have to point out that the time of arrival of the last passenger and that of the last baggage are parameters that may be estimated by considering the service rate, the number of passengers (with and without bags), and nbag. Let us assume that all the pieces of baggage belonging to a given passenger show up together at the baggage claim and that the passenger and baggage arrival time distributions are independent. We now have six different ways of estimating the dwell time a passenger spends in the baggage claim area, according to each combination of (tP1, tPL, tB1, tBL). For example, if tPL< tB1 holds, the waiting time distribution WT can be obtained as follows WT = TB -TP (see figure 2).
t1
P

tL

t1

tL

Figure 2. Combination of (tP1, tPL, tB1, tBL). if tPL< tB1

Then the mean is: E(WT) = E(TB) - E(TP) = (tBL + tB1 ) / 2 - (tPL + tP1 ) / 2. The formula for computing the IOS is the usual one, where we indicate with AP the number of passengers arrived and ADT = E(WT).

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 16

4.1.4. Aggregate models for holding facilities The Average Dwell Time (ADT) for those facilities dedicated to the holding of passengers has to be given as input together with area, number of passengers (Pax), and number of well-wishers (NWW). The number of persons entering (AP) a holding facility has to be estimated considering the number of passengers and that of the well-wishers: AP = Pax + NWW. The space IOS can be computed as usual. In lounges and waiting or assembly areas, the model is the general one proposed above. Notice that, in computing the number of persons in the area (AP = Pax + NWW), the number of well wishers is equal to zero if the area is placed after the security check on departure, or before customs on arrival. Of course, if more detailed information is available, the above formula can be further elaborated.

4.1.5. Aggregate models for flow facilities In this section we analyze the models related to flow facilities. We are interested in the pedestrian density in a flow area as suggested in (Fruin, 1971). The number of persons going through a flow area has to be estimated considering the number of passengers and that of well wishers. Of course, the number of well wishers is zero if the flow facility is beyond the security check (on departure), or before customs (on arrival). The input parameters for the level of service are the volume of passenger traffic and the corridor width. By corridor width we mean the effective corridor width, which is the corridor width reduced by 1.5 meters (i.e., we do not consider the space near the edges of the facility). The IOS for flow facilities can be computed in terms of persons per meter width by: traffic volume corridor width.

IOS =

5. Program Structure SLAM is made of a graphical user interface, called SLAM-Workbench (SLAM-Wkb for short) and by an engine (SLAM-Solver). The task of SLAM-Wkb is to assist the user in providing to SLAM the input data, then to start an elaboration, and finally to present graphical and textual output. The SLAM Solver is implemented in ANSI C, while the SLAM Wkb in JAVA. Both input and output files are text files composed by tables, so a person can read and manipulate them via any software package (e.g., Excel or Access) that can write a TAB spaced table; due to this fact both SLAM-Wkb and SLAMSolver perform a specific input check. The input of the program is composed by tables that contain: scheduling of the flights, terminal physical configuration, allocation of the terminal resources to manage the flights (policy data). A SLAM input has no large data requirements. A SLAM simulation (with over 800 flights) requires approximately 6 seconds of CPU on a PC Pentium 133 running under Windows 95. SLAM output is divided into 2 files: a textual and a graphical output file. In the textual output file there are the results of SLAM elaboration for each of the facility considered, while in the graphical output file there are the

17 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

graph points and the LOS levels (where required) for plotting facility charts. In figure 3 there is an example of a graphical output of SLAM.
Check-In
4 10 9 8 3 7 Open Counters 2.5 Passengers 6 5 4 3 1 2 0.5 1 0 10:00 10:10 10:20 10:30 10:40 10:50 11:00 11:10 11:20 11:30 11:40 11:50 12:00 9:00 9:10 9:20 9:30 9:40 9:50

3.5

Open Counters Passengers

1.5

Time

Figure 3. Open counters and passengers versus time In both output files a summary table with the LOS provided by each facility is recorded. Results of a SLAM elaboration are provided for each facility or facility component. The facilities considered are: Departure Concourse, Ticketing, Check-in, Security, Passport control, Flow, Gate Lounge, Baggage claim, Customs, Arrival Concourse. Minutes of delays usually shown by airside models are used as input by SLAM. The integrated model has been applied to a situation of airside congestion in Linate, i.e., a scenario with fog between 7:30 and 9:15. SLAM correctly pointed out that landside was mainly affected in the gate lounges and that the airside congestion propagated its negative effects to the landside component well after 9:15 till about noon. 6. The Milan Airport System The MILAN Airport system is managed by SEA. This system underwent to a major reorganization on October 1998. During the study reported in this work (January 96 - September 98) the Milan airports managed by SEA were: Milan Linate International Airport (7 km from the city centre, open 24 hours a day, used for short and medium haul domestic and international flights);

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 18

Milan Malpensa Intercontinental Airport (48 km from the city centre, open 24 hours a day, principally used for long haul intercontinental and international flights and for charter flights); Malpensa 2000 Airport (this terminal opened October 25, 1998, and currently serves 67% of all flights and is expected to augment its share in the future).

In the past 16 years Milans airports have recorded very high growth rate (a yearly average of about 5,7%) with a shift from approximately 6 million passengers in 1979 to 15 million in 1995. In 1998 Linate was by far the busiest airport in todays Milan system, with about 75% of the Milan total air traffic. In 1996 the Milan Airport system registered one of the highest growth rates in Europe and is now the seventh busiest airport system in Europe, in terms of annual number of passengers. The airside of Linate airport until 1998 was generally recognized as one of the most congested and capacity-constrained in Europe today. Moreover, the capacity and congestion problem at Linate had characteristics which were almost unique among European airports. Unlike most other congested airports in Europe, where runway capacity acted as the single major bottleneck of the airport, Linate was characterized by a double bottleneck situation: apron capacity imposed a constraint just as important as runway capacity. The Linate apron had 32 stands of which 5 were nose-in (aviobridges) and the others remote. It had one runway 2,440 m x 60 m. During peak traffic periods the capacity of the apron was often exceeded. When this happened aircraft had to be towed to a small number of reserve stands on the helicopter apron. The Linate scenario used for preparing the data for the runs is based on a historic busy day (in terms of aircraft movements per day): November 27, 1995. The airport landside configuration is the one in use since November 1995. Since that date, the check-in islands have been managed directly by the Air Companies. The Malpensa 2000 project was centered on the construction of the new terminal located at a distance of approximately 3 km west of the current Malpensa Nord airport site. The resulting new airport includes the two existing runways (respectively 3,950 m and 3,500 m long) and a series of new buildings that will all be located west of the airport site. The new airport area, totaling approximately 12 sq. km, is basically a rectangle 4.5 km long and 2.5 km wide. The core structure of the new Malpensa 2000 airport, around which all the other buildings and functions will be developed, is the passenger terminal (a linear building about 600 m long, 25 m high and 80 m wide) which is connected by three roofed skywalks to three aircraft satellites (only two were considered to be in operation during the busy days used in this study, which is the now current situation); it will also be connected to the new railway station. The four-floors above-ground terminal also offers large areas for shops and other businesses. By October 25, 1998, Malpensa 2000 was opened. It includes the first operational segment of the new airport, also known as Polo Funzionale: two thirds of the new passenger terminal, the aircraft apron and two satellites with a capacity of handling at least 12 million passengers/year. There are twenty fly-bridges in the satellites, besides remote airplane parking areas on the edges of the apron. The project completion is due by the year 2000. The new Malpensa 2000 airport will have a capacity of handling 18 million passengers/year in addition to the 6 million passengers/year of Malpensa Nord operating as a remote satellite. In terms of accessibility Malpensa 2000 airport aims to become a hub of the European intermodal transportation network, integrated with international road and railway networks, so that it will be serving more than just the metropolitan area of Milan. The Malpensa 2000 scenarios used for preparing the data of the runs were based on two possible busy days (in terms of aircraft movements per day); in the first we considered commercial, cargo and charter flights; in the other we investigated the hypothesis of the national air company using the airport as a

19 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

hub. These two scenarios differ mainly in the fact that in the latter there are three peak periods of congestion and, of course, more flights. The airport landside configuration is basically the one in use since October 1998. Beginning with that date, the terminal are organized in three main areas, each one dedicated to the type of flights it is serving: Schengen, Non-Schengen, North American flights. 7. Slam Models of Linate and Malpensa 2000 We should begin this section by mentioning the fact that no simulation was ever previously done for the landside of the Linate airport and of the Malpensa 2000 airport. All the data were collected for preparing the runs for this paper. The only studies we could compare with is the one contained in the recent report I Livelli di Servizio e la Capacit Strutturale (Sustained Capacity) dellAeroporto di Linate (SEA, 1995) and Malpensa 2000 - La Capacit Strutturale dellAerostazione Passeggeri (SEA, 1996). In table 3 we have the check-in opening profiles. In table 4 we list the processing time and available space for processing facilities. Table 3. Check-in opening profiles First check-in opening (time before take-off) 1h:45 2h:15 2h:45 3h:00 Last check-in closure (time before take-off) 15 15 Number of counters 1 1 (during first 30) 2 (thereafter) 2 (during first 30) 3 (during the next 90) 2 (thereafter) 3 (during first 60) 4 (during the next 60) 3 (thereafter)

Aircraft type Small Narrow body

Wide body

15

Large body

15

Table 4. Processing facilities: processing time and available space Facility Check-in counters Express Check-in (no bags) Security Ticketing Passport control (Non Schengen European) Passport Control (Extra-European) available area in front (in sq. m.) 16 16 60 5 20 20 processing time (in sec.) 90 (120 for USA) 60 5 4 20 40

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 20

7.1. Results obtained for Linate In the Linate Airport Terminal the registration desks are divided in 13 groups, each group serving one or more airlines (see table 5). There are 2 security checks, 20 gates, and 2 Baggage claim units. Table 5. LOS of Linate international check-in counters Counters dedicated to the following airlines BA-BQ ME-LZ MS-YE-BM-AP-KM AT-TK IJ-SK IB-TP AZ LH KL-UK-others AF OA-FU SR-SN-OS EI-AY Number of counters 6 2 3 3 3 4 12 4 4 4 2 4 2 LOS reached 8:30-9:30 D A D E F A E LOS reached 9:30-10:30 A E F A A A A A F

During the busy day 428 movements were recorded (213 arrivals and 215 departures) with a total of 28,964 passengers (14,635 arriving to Linate and 14,329 leaving from Linate; transit passengers counted both as arriving and as departing passengers). Congestion at the terminal is detected by SLAM during the two peak hours (8:30 - 9:30 A.M.; 9:30 10:30 A.M.). The area in front of each check-in counter is assumed to be equal to 16 sq. m. If we consider all 53 international counters as part of a unique common check-in facility (i.e., a passenger can check-in for her/his flight at anyone of the open counters), then the resulting LOS is A during both Peak Hours. However, if we consider the more realistic situation where international check-in counters are dedicated to specific airlines, then the LOS varies from A to F. More specifically, if we group the 53 international counters into 13 clusters, we obtained the LOS reported in table 5. For domestic check-in, there are 28 counters 4 of which are reserved for VIP passengers. The remaining 24 counters are split in the following way: 15 are positioned before the security check and 9 after. If we consider all 24 counters to be in common use, then the resulting LOS is A in both Peak Hours. The same A LOS is observed if we cluster the counters according to the dedicated airlines. The LOS at Baggage Claim areas is always A for both domestic and international areas. 7.2. Results obtained for Malpensa 2000 There are 6 security checks, the gates are grouped in four clusters (North, South, Satellite, Remote Satellite) for an overall sum of 36, and there are 2 Baggage claim units (Schengen, Non-Schengen), each with 5 devices. The Central Body Registration Desks is for passengers with bag and it is divided in 6 islands that can be grouped in 2 main clusters each one dedicated to the type of flight it is serving:

21 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

Schengen, Non-Shengen (North American and Other Continents). The Central Body Registration Desks are divided in 13 clusters: AA, AF, AP, AZ, BA, DL, IB, KL, LASI, SR, TW, UA, JOLLY. During the non hub busy day 614 movements are considered (312 arrivals and 302 departures) with a total of 37,928 passengers (18,964 arriving to Malpensa and 18,964 leaving from Malpensa; transit passengers counted both as arriving and as departing passengers). During the hub effect busy day 838 movements are considered: 434 arrivals (62 cargoes) and 404 (32 cargo) departures, with a total of 69,915 passengers: 34,398 arriving to Malpensa and 35,517 leaving from Malpensa of which 6,035 transit (23% of commercial flight passengers). The level of service provided by the Malpensa 2000 airport is computed by SLAM: the level of service provided by the processing facilities is in general A for the case without the hub effect, while it goes from A to F in the scenario with the Hub effect. This is due not only to a higher number of flights (from 614 to 838) and higher number of passengers (from 38 to 70 thousands), but also to a more detailed analysis of the passenger flows. Results for the hub case can be read in table 6. Table 6. Hourly LOS of Check-In at Malpensa 2000 with hub effect (Column headings: hourly window starting time)
Check-in: UA DL TW AP IB LASI SR AF KL AZ JOLLY BA AA Passport control 7:30 8:30 9:30 10:30 11:30 12:30 13:30 14:30 15:30 16:30 17:30 18:30 19:30 20:30 21:30 A A A A C A A A D C A A A A F A A A A A A A A A A A A A F A A A A A A A A A E A A A D A A A B A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A C F A A A F A F A A F C A A A A A A A A A A A A F A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A D A A A A A A F A A A A A A A A F A A A A A A A A A B A A A A A A A A A A A F A A A A

8. Final Remarks In this paper we first identify the models of landside elements that are capable of measuring the capacity of an element under different operating conditions. Generally, the efficiency of the element is evaluated by comparing its capacity values with standard measures of the level of service provided to passengers. Thus, our main goals were to select adequate tools to measure capacity and to identify reference values for level-of-service standards. Unfortunately, many of the existing tools are not publicly available. In addition the existing level-of-service standards are not accepted universally. However, the models presented here can work with any specified level-of-service standard that uses the same performance metrics. We also examined the scientific literature and the software on terminal design and

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 22

management. For a given facility, we adopt the models and formulas found in the literature, whenever available, and we provide them ourselves when they are missing. The aggregate models proposed are based on relatively simple formulas: their objective is not to provide a thorough analysis of a given facility, but to be used for the estimation of the capacity of the facility by specifying a limited set of parameters and operating conditions. The application of the SLAM approach to Linate airport has demonstrated the fact that the Airport Authority was collecting a great amount of data and statistics that were recorded in various ways often with duplication and sometimes with inconsistencies among the data themselves. As a first useful result SLAM has made clear the need of validating the data in order to obtain a globally correct and consistent database. In any event, the SLAM model has correctly pointed out the weaknesses of Linate, that is the fact that the check-in area is not always satisfactory for all companies. At the operational level the application of the SLAM approach to Malpensa 2000 airport has pointed out some critical factors: the check-in configuration for some airlines is not sufficient in certain time slots; the passport control configuration provides a level of service C from 12:30 to 14:30. As expected, when all elements of the airport scenario are working properly, the level of service provided to passengers is reasonable. In the scenarios we examined in this paper no serious consequences were propagated from the airside to the landside or viceversa. Other scenarios could be conceived where this interaction could have significant effect: to obtain this interaction one can directly incorporate these delays into the flight schedule provided in input. SLAM will be applied in the next future to the terminal of other airports. Acknowledgements

The research was supported by the European Union research grant TAPE (Total Airport Performance Evaluation), DG VII R95 - B67161 - SIN000326, by the Italian National Research Council (CNR), Progetto Finalizzato Trasporti 2, by the Italian Ministry of University and Scientific and Technological Research (MURST) project MOST, and by SEA. We thank Professor Amedeo Odoni of MIT for his support during all our research. We are grateful to Fabio Degli Esposti, Giovanni Viscovi, Luca Silari, Paolo Sordi and Paolo Morghen of SEA for their useful comments in the making of SLAM and for providing us the input data. Special thanks for their help in creating and implementing the model go to Denise Beltramin, Elvira Fiascone, Guglielmo Guastalla, Massimo Morin and Sabina Piccolo. References Ashford N., 1988. Level of service design concept for airport passenger terminals - a European view. Transportation Planning and Technology 12, 5-21. Babic O., Teodorovic D., and Tosic V., 1984. Aircraft stand assignment to minimize walking. Journal of Transportation Engineering 110 (1), 55-66. Bandara S. and Wirasinghe S.C., 1989. Airport Gate Position Estimation Under Uncertainty, in: Transportation Research Record 1199, Transportation Research Board, National Research Council, Washington D.C., pp. 41-48.

23 / L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model)

Barbo W.A., 1967. The Use of Queuing Models in Design of Baggage Claim Areas at Airports. Graduate report, Institute of Transportation and Traffic Engineer, University of California, Berkeley, CA. Browne J.J., Kelly J.J., and Le Bourgeois F., 1970. Maximum inventories in baggage claim: a double ended queuing system. Transportation Science 4 (1), 64-78. Dunlay W., McCullogh F. et al., 1975. A System Analysis Procedure for Estimating the Capacity of an Airport: System Definition, Capacity Definition and Review of Available Models. Council for Advanced Transportation Studies, University of Texas at Austin. Edwards J. V. and Newell G. F., 1969. A study of gate use at Honolulu international airport. Transportation Science 3 (3), 183-191. FAA, 1988. Planning and Design Guidelines for Airport Terminal Facilities. Advisory Circular AC 150/5360-13, Federal Aviation Administration, Washington, D.C. Fruin J.J., 1971. Pedestrian Planning and Design. Metropolitan Association of Urban Designers and Environmental Planner, New York. Hamzawi S.G., 1986. Management and planning of airport gate capacity: a microcomputer-based gate assignment simulation model. Transportation Planning and Technology 11 (3), 189-202. Horonjeff R., 1969. Analysis of passenger and baggage flows in airport terminal buildings. Journal of Aircraft 6 (5), 446-451. IATA, 1981. Guidelines for Airport Capacity/Demand Management. Geneva, Switzerland. IATA\ATAA, 1982. IATA Airport Terminal Capacity Analysis. Internal Report, Air Transport Association of America, Washington, D.C. IATA, 1995. Airport Development Reference Manual. International Air Transport Association, Montreal, Quebec, Canada. Le C. D., Mangano F. and Nixon P., 1978. Determine gate capacity by computer. Airport Forum 8 (5), 67-70. Lee A. M., 1966. Applied Queuing Theory. MacMillian and Co. Ltd., London. Mangoubi R.S. and Mathaisel D.F., 1985. Optimizing Gate Assignments at Airport Terminals. Transportation Science 9 (2), 173-187. Mumayiz S.A., 1990. An Overview of Airport Terminal Simulation Models, presentation to the 69 Annual Meeting of Transportation Research Board. Newell G.F., 1971. Application of Queuing Theory. Chapman and Hall, London. Odoni A.R. and de Neufville R., 1992. Passenger Terminal Design. Transportation Research 26A (1), 27-35.

L. Brunetta, L. Righi and G. Andreatta / SLAM (Simple Landside Aggregate Model) / 24

Pararas J.D., 1995. ARTS: Airport Terminal Building Simulator. ARGO Technical Report. Piper H.P., 1974. Design principles for decentralized terminals. Airport Forum 4 (3), 39-51. Rallis T., 1958. The capacity of airports. Ingenioren International Edition 2, 89-102. Rallis T., 1963. Terminal Transportation Engineering I, Airports. Report 27 of the Department for Road Construction, Transportation Engineering and Town Planning, The Technical University of Denmark, Copenhagen. Rallis T., 1967. Capacity of Transport Centers, Railroad Stations, Road Haulage Centers and Airports. Report 25 of the Department for Road Construction, Transportation Engineering and Town Planning, The Technical University of Denmark, Copenhagen. SABRE, 1994. Airport Landside Consulting. Brochure, SABRE Decision Technologies. Steuart G. N., 1974. Gate position requirements at metropolitan airports. Transportation Science 8 (2), 169-189. SEA, 1995. The Level of Service and the Sustained Capacity of the Linate Airport (In Italian). Internal Report DAA/DGA/DPC. SEA, 1996. Malpensa 2000 - The Sustained Capacity of the Airport Passenger Terminal (In Italian). Internal Report DAA/DGA/DPC. Svrcek T., 1994. Planning level decision support for the selection of robust configurations of airport passenger buildings. Ph.D. Thesis, Dept. of Aeronautics & Astronautics, Flight Transportation Laboratory Report R94-6, MIT, Cambridge, Ma, USA. Tosic V., Babic O. and Janic M., 1983. Airport passenger terminal simulation (in Serbo-Croatian). Annales of Operations Research in Air Transportation, Faculty of Transport and Traffic Engineering, University of Belgrade, 83-100. Tosic V., 1992. A review of airport passenger terminal operations analysis and modeling. Transportation Research 26A (1), 3-26. Transportation Research Board (TRB), 1985. Special Report 209: Highway Capacity Manual. TRB, National Research Council, Washington, D.C. Transportation Research Board (TRB), 1987. Special Report 215: Measuring Airport Landside Capacity. TRB, National Research Council, Washington, D.C. Wirasinghe S.C. and Vandebona U., 1987. Passenger Walking Distance Distribution in Single-and Dual-Concourse Centralized Airport Terminals, in: Transportation Research Record 1147, Transportation Research Board, National Research Council, Washington D.C., pp. 40-45.