1 AN INTEGRATED VERTIPORT PLACEMENT MODEL CONSIDERING VEHICLE

2 SIZING AND QUEUING.

3
4
5
6 Carl Khalife
7 Research Assistant
8 Centre for Transport Studies
9 Department of Civil and Environmental Engineering
10 Imperial College London
11 SW7 2BU, UK
12
13 Joseph Slim
14 Masters in Engineering Student
15 Centre for Transport Studies
16 Department of Civil and Environmental Engineering
17 Imperial College London
18 SW7 2BU, UK
19
20 Jose Escribano Macias, Ph.D., Corresponding Author
21 Research Associate
22 Centre for Transport Studies
23 Department of Civil and Environmental Engineering
24 Imperial College London
25 SW7 2BU, UK
26 jose.escribano-macias11@imperial.ac.uk
27
28 Panagiotis Angeloudis, Ph.D.
29 Reader in Transport Systems and Logistics
30 Centre for Transport Studies
31 Department of Civil and Environmental Engineering
32 Imperial College London
33 SW7 2BU, UK
34
35
36 Word Count: 6306 words + 4 table(s) × 250 = 7306 words
37
38
39
40
41
42
43 Submission Date: July 30, 2021
 Khalife, Slim, Escribano and Angeloudis 2

1 ABSTRACT
2 The increasing levels of congestion and infrastructure costs in cities have created a need for more
3 intelligent transport systems. Urban Air Mobility (UAM) offers a solution by introducing intra-
4 urban aerial transport to overcome the existing congested infrastructure. The performance of UAM
5 systems are highly dependent on vertiport locations, vehicle sizing and infrastructure specifica-
6 tions. This study takes a holistic approach to UAM network optimisation by considering the inter-
7 relatedness of these decisions. A vertiport placement model with vehicle sizing constraints is de-
8 veloped to determine the optimal vertiport configuration while considering eVTOL performance.
9 The resulting configuration is used to model waiting times depending on infrastructure specifica-
10 tions. These waiting times are incorporated in the vertiport placement and vehicle sizing models.
11 An iterative approach is undertaken to find the network configuration that balances the infrastruc-
12 ture, operational costs as well as passenger waiting times. The purpose of the study is to inform
13 policy-making by proposing a holistic approach to UAM network design.
14
15 Keywords: vertiport placement, urban air mobility, queuing theory, vehicle sizing
 Khalife, Slim, Escribano and Angeloudis 3

1 INTRODUCTION
2 Urban Air Mobility (UAM) as a concept has been pursued by research and industry since the
3 mid-1900s, yet only recently has technology reached the level of maturity required to make urban
4 air travel economically feasible. By September 2018, approximately $1 billion was invested in
5 UAM, and over 70 electrical vertical take-off and landing (eVTOL) manufacturers were founded
6 (1), highlighting the industry’s potential market value.
7 While UAM includes varied operation types, one of its most promising aspects is On-
8 Demand Air Mobility (ODAM), which envisions a demand-responsive operational model to serve
9 intra- and inter-urban transportation, as well as short-haul flights that are not sustainable using
10 traditional aviation (2). ODAM seeks to solve two of the main problems that are expected to
11 shape the future of the aerospace industry: demand for increased travelling speeds and the need for
12 reduced emissions (3). Its potential market size is currently valued at $2.5 billion in the first years
13 of operations and increase to up to $500 billion (1).
14 ODAM is expected to operate between a network of vertiports, which have take-off, landing
15 and charging capabilities. These can be retrofitted into existing infrastructure such as unused
16 helipads or highway cloverleafs, to significantly reduce the initial investment cost compared to
17 traditional modes of transport (4).
18 Recently, several initiatives have launched to further develop the field of ODAM. Uber
19 expressed its aspiration to launch an air-taxi service in the near future through the Uber Elevate
20 initiative (4). Additionally, alongside NASA, Uber Elevate organises yearly conferences and work-
21 shops to form a community centred around building solutions for UAM development (4, 5). NASA
22 is also developing new air-traffic management methods that accommodate ODAM operations into
23 the urban airspace (6).
24 (7) reported that achieving a return on invested capital in the ODAM market is possible
25 even for small networks, but highlight that the cost of charging and refuelling will significantly
26 affect the business case. At the same time, the study stresses the importance of ensuring very fast
27 turnaround times in minimising the total operating costs. Turnaround time is governed by many
28 factors, such as the number of pads at each vertiport, battery capacity, charging rate, length of trips
29 as well as the dispatch strategy.
30 These findings suggest vertiport placement models that determine the optimal configuration
31 of supporting infrastructure, as well as vehicle parameters, are essential in ensuring the viability
32 of ODAM systems. Failing to do so results in network configurations which are infeasible from
33 an operational perspective. For example, neglecting vehicle sizing constraints in the vertiport
34 placement model might lead to configurations where the operational requirements are not satisfied.
35 Further, neglecting the effects of the number of pads at each vertiports on waiting times, might
36 lead to underestimating turnaround times. However, our review of current literature suggest that
37 an integrated solution that encompass these aspects is yet to be developed (see Section 3).
38 To address this gap, this paper proposes an approach to designing air-taxi networks by
39 considering the optimisation of three inter-related decisions: vertiport locations, vehicle and in-
40 frastructure specifications. Each decision is captured in a stage, with vertiport locations being
41 determined through a hub-location problem, and infrastructure specification being obtained using
42 Jackson open network theory. An iterative algorithm is developed that solves three stages sequen-
43 tially, improving the solution through a feedback mechanism that updates problem parameters until
44 an optimal solution is achieved.
45 The paper’s contribution is threefold:
 Khalife, Slim, Escribano and Angeloudis 4

1 1. To the authors’ knowledge, it is the first study to propose a methodology to design a

2 UAM network considering the interrelationships between vehicle sizing, vertiport in-
3 frastructure and network design.
4 2. It models vertiport operations using open network theory, instead of relying on network
5 simulation usually found in literature.
6 3. It presents a computationally efficient solution heuristic that can accommodate large
7 problem instances. A case study based on the hypothetical design of a London UAM
8 system, in which our algorithm shows significant improvement in all metrics compared
9 to our benchmark.
10 The remainder of the paper is structured as follows. An analysis of the current state of
11 the art is presented in the next section, followed by a description of the mathematical model and
12 formulation. After, the case study based on the city of London is introduced and the results of our
13 algorithms are presented and analysed. Finally, our concluding section suggests further works.

14 CURRENT APPROACHES IN ODAM NETWORK DESIGN

15 This section reviews the state-of-the-art in the field of ODAM vertiport location planning and
16 vehicle sizing. A summary of the literature reviewed is presented in Table 1, highlighting the
17 features unique to our methodology.

18 Vertiport Location
19 The vertiport placement problem is generally structured as a facility or hub location problem,
20 in which the geographical position of hubs is determined to optimise a specific objective. The
21 literature of the facility location problem is extensive (8), with one of the first formulations being
22 proposed by (9), known as p-median problem.
23 Extensions to the original problem have been proposed for multiple applications (10). How-
24 ever, the p-median formulation provides an adequate framework to model vertiport placement mod-
25 els, as it allows the positioning of hubs to be governed by economical metrics, such as the travel
26 time savings, weighted demand distance, and infrastructure costs.
27 Within this framework, (4) proposed an initial network configuration model applied to
28 Los Angeles and London case studies. A clustering algorithm is used to group the demand into
29 discrete facility candidate locations, and a facility location algorithm is developed to maximise trip
30 coverage.
31 Another vertiport placement model is presented by (11) that assumes short-term eVTOL
32 demand is driven by high income car users with large travel times. Consequently, the objective
33 is to maximise the total travel time savings relative to driving, but ignores the investment costs
34 relating to infrastructure development and costs of operations. In fact, operational constraints are
35 not evaluated, as vertiports are given unlimited capacity to allow the optimisation not to focus too
36 many ports in areas of large demand.
37 A capacitated p-median problem instance is explored by (12). A neighbouring searching
38 algorithm is used to partition service region into catchment areas for each supporting facility and
39 select the optimal placement that minimise weighted demand distance.
40 (13) presented an uncapacitated p-median formulation to solve the vertiport placement
41 problem for airport strips, where demand is estimated using airport incoming hourly trips. Selected
42 vertiports are then modelled as an M/M/c queue system during post-processing, to determine the
43 potential market penetration at each port.
 Khalife, Slim, Escribano and Angeloudis 5

1 Operational parameters are incorporated to the vertiport placement problem by (14). The
2 study proposed a two stage model for vertiport placement and operation scheduling problem, where
3 the latter optimises the charging scheduling of vehicles to minimise total delay for all passengers.
4 However, given that the stages are produced in sequence without feedback the algorithm will not
5 produce optimal results, as the operations model is executed as a post-process.
6 Deviating from the original hub location formulation, (15) and (16) both propose a vertiport
7 location problem using a k-means clustering approach. Without applying the hub location problem
8 formulation, this approach is unable to consider factors relating to the operation, cost, or capacity
9 of the infrastructure.
10 To date, no vertiport placement study considers vehicle sizing, charging and operational
11 constraints. Without notion of such concepts, the resulting vertiport configuration could lead to
12 vehicle designs with unreasonable battery masses, charging requirements, and costs.

13 Vehicle Sizing
14 This stream of research aims to determine the optimal power and battery requirements for vehicles
15 to sustain operations, which requires uses set mission parameters as input, namely the flight time
16 requirements and charging time requirements.
17 Among the most widely employed tools to design air vehicles is NASA’s Design and Anal-
18 ysis Rotorcraft (17), which has already been used to design air-taxi vehicles (18, 19).
19 (20) proposed a cruise depletion rate parameter to encompass the energy performance into
20 a single parameter. The approach is refined in (21) in order to incorporate relationships with
21 other critical vehicle properties, such as vehicle mass. The iterative process in (21) assumes a
22 power loading parameter for each mission segment to determine the battery size that achieves the
23 specified mission requirements.
24 Another vehicle design tool is presented by (22) that compares the performance of VTOL
25 and STOL aircraft in terms of maximum take-off weight given varying flight ranges and cruise
26 speeds. (23) developed a vehicle sizing model focusing on noise emission, power/energy con-
27 sumption rates, and costs in the context of UAM.
28 Deviating from the previous literature, (24) performed a weight-based optimisation with
29 consideration of range, speed, rotor size, wing loading and battery energy density parameters.
30 Weight is assumed to act as a proxy for the direct operational cost.
31 This section reveals that vehicle sizing is mainly driven by cost, which is dependent on the
32 required levels of cruise and hover specified by the mission profile. Nevertheless, the provision
33 of insufficient supporting infrastructure in the form of vertiports or landing areas may lead to
34 significant loiter times, which would require a larger battery size, therefore driving up costs, while
35 also increasing turnaround times.
36 However, current research does not consider the effect of loiter time on vehicle perfor-
37 mance, vertiport placement and costs. These can be estimated using UAM network simulation,
38 which have been developed with varying level of detail (20, 25–27).
39 While simulation-optimisation frameworks can be used to optimise network design using
40 agent-based models, the approach will never yield an optimal solution, and an analytical approach
41 is nevertheless required to benchmark the output of the simulation. Furthermore, the simulation
42 speed becomes the limiting factor of the optimisation, which increases as the scope increases in
43 size and complexity.
 Khalife, Slim, Escribano and Angeloudis 6

TABLE 1 Literature summary

Focus Author Approach Objective Features
(4) C, EA WD Demand aggregation

Vehicle Sizing Vertiport Location

(11) EA TS Traffic; Demand
(12) EA WD Vertiport capacity
(13) EA TS Transfer; Queue post-process
(14) C TT Charging; Scheduling
(15) C D Case Study
(16) C D Time Savings
(18) - W Aircraft layout
(19) - W Aircraft layout
(20) - W Energy source
(21) - C Battery size
(24) - W Aircraft layout
This paper EA TS Queuing; Battery size
Legend:
Approach: C - K-means clustering, EA - Exact algorithm.
Objective: C - Cost, D - Distance, TT - Travel Time, TS - Travel Savings, WD - Weighted
Demand, W - Weight.

1 Contribution
2 The infrastructure limitations in vertiports (number of charging, take-off and landing pads) sug-
3 gests that vehicle fleets will experience loitering during operation. The available loiter time of
4 vehicles is determined through vehicle sizing. As such, the vertiport placement models that ignore
5 the loitering and infrastructure requirements will result in operationally infeasible configurations.
6 While network simulations can be used to model operations in simulation-optimisation frame-
7 works, they lack practicality due to the large run times of existing simulations.
8 To solve the integrated network design and vehicle sizing problem, this paper proposes a
9 three stage iterative approach. In the first instance, a modified p-means hub location model is used
10 to determine the optimal geographical position of vertiports. The outputs of this stage serves as
11 the inputs to the vehicle sizing model that calculates the vehicle battery size requirements given
12 the operational requirements of the proposed UAM network. The final stage uses a queuing model
13 to estimate loiter times and define the infrastructure requirements of the vertiports. A feedback
14 loop collects the results of the infrastructure model and modifies the initial vertiport placement
15 model. This process is repeated until the marginal cost of operating an additional pad outweighs
16 the marginal benefit of doing so. The proposed approach allows us to model the inter-dependencies
17 between between various UAM decisions, while also finding the configuration that optimises op-
18 erational, infrastructure costs as well as service times.

19 METHODOLOGY
20 This paper proposes a holistic approach to UAM network optimisation, the objective is to deter-
21 mine the optimal location of vertiports given pre-defined demand intensity, as well as determining
22 the appropriate vehicle battery size and vertiport properties that minimise overall implementation
 Khalife, Slim, Escribano and Angeloudis 7

FIGURE 1 Solution method workflow. Highlighted in green are the inputs, processes and
outputs of the vertiport placement model, in blue the vertiport sizing model, and in orange
the infrastructure model.

1 costs. A three-stage algorithm is presented composed of a linear vertiport placement model, a

2 vehicle sizing stage, and an infrastructure model.
3 The vertiport placement model focuses on selecting the vertiport locations that maximises
4 travel savings to customers provided. In doing so, the proposed model ensures that selected links
5 between the vertiports conform with the operational constraints defined by the aircraft and infras-
6 tructure properties.
7 The vehicle sizing stage determines the battery size that satisfies the energy requirements
8 outlined by the placement model. Once the battery size is determined, the charging requirements
9 are calculated based on the C-rate and discharge rates of the battery, which vary based on its mass
10 and the specific energy of the vehicle.
11 Finally, the infrastructure model is applied to each vertiport to determine their service rate
12 requirements. Utilising open network theory, the congestion at the vertiports is modelled, and the
13 allowed loiter times based on the number of landing, charging and take-off pads throughout the
14 network is calculated. As network theory requires arrival demand rates, these are obtained from
15 agent-based modelling tools at the beginning of the optimisation run. The input and output for
16 each stage are presented in Figure 1, which outlines the problem framework.
17 The stages are executed in sequence and continuously through a feedback loop until the
18 marginal cost of operating an additional pad outweighs the marginal benefit of doing so. At this
19 point a solution is reached that outlines the optimal state of infrastructure, operational costs and
20 passenger waiting times.

21 Formulation
22 The problem formulation of the solution method components: vertiport placement, vehicle
23 sizing, and vertiport infrastructure, are presented separately. Throughout the remainder of this
 Khalife, Slim, Escribano and Angeloudis 8

1 section, the variable and parameter definitions used are outlined below.

Indices Sets
i, j = nodes V = node set
u = segment U = route segment set
k = vertiport pad type K = pad set

Link Parameters Vehicle Parameters

Gi j = driving time from i to s p,u = specific power [W/kg]
j [h] se,u = specific energy [Wh/kg]
Li j = expected trip time from i N = flight cycles [-]
to j [h] mp = vehicle payload [kg]
Di j = relative demand from i mb = battery mass [kg]
to j [trip] BU , BL = upper/lower battery level bound [%]
Hi j = haversine demand from i C = battery C-rate [%/h]
to j [km] E = battery energy [J]
ps = flight segment power r = reserve battery [%]
requirement [W/kg] Rr , Rd = recharge/discharge rate [%/h]
ls = flight segment power loa- Q = required recharge [%]
ding requirement [W/kg] r = recharge time [s]
ts = flight segment travel time M = max take-off mass [kg]
Q = battery capacity [mAh]

Decision Variables Queuing Parameters

xi j = Boolean: open or closes Qi j = arrival rate [W/kg]
link between two vertiports P0i j = probability of no queue [-]
oi = Boolean: open or closes βc , βl = charging and VTOL pads bound [-]
vertiport i λqi j = queue length [-]
cik = Integer: number of servers pikl = probability of changing
per type k at vertiport i from state k to l at i [-]
zi j = Boolean: flight link Wqi j = waiting time of queue q [s]
requirements satisfied

2 Vertiport Placement
3 The vertiport placement model finds the most optimal vertiport locations in a city given relative
4 demand levels between candidate locations. The model is based on a variant of the uncapacitated
5 p-hub location problem and aims to maximise the total travel time saved in the system relative
6 to driving to quantify the overall benefit to its users. The following modelling simplifications are
7 adopted:
8 Assumption 1: Driving is the main competitor of air-taxi services. Qualitative air-taxi
9 demand studies (28, 29) found autonomous cars to be the biggest competitor to air-taxi services.
10 As such, the objective maximises the total time saved in the system relative to driving.
11 Assumption 2: Trips are only operated when the factored routing distance lies within a
12 specified bound Hmin , Hmax . Vehicles are assumed to fly at a average speed of 242km/h as required
13 by (30), and a safety routing routing factor µ = 1.42 is assumed.
14 Assumption 3: Relative demand levels between vertiports are considered. As (31) postu-
15 lates, absolute demand levels will vary as the service is introduced and are challenging to model.
16 Thus, relative demand levels provide a more accurate depiction of the operational service require-
17 ments between vertiports.
 Khalife, Slim, Escribano and Angeloudis 9

1 Assumption 4: Following (4), air taxi trips must provide over 40% travel time savings to
2 be considered relative to driving. Furthermore, travel time savings constitute the main benefit of
3 UAM.

maximise (Gi j − Li j )Di j xi j (1)

∑ oi = n (1.1)
i∈V
xi j ≤ o j ∀i, j ∈ V (1.2)
xi j ≤ oi ∀i, j ∈ V (1.3)
zi j Hmin ≤ zi j Hi j ∀i, j ∈ V (1.4)
zi j Hi j ≤ zi j Hmax ∀i, j ∈ V (1.5)
0.6zi j Di j ≤ zi j Li j ∀i, j ∈ V (1.6)
∑Uu pu tu M
zi j ≤ zi j semax ∀i, j ∈ V (1.7)
mb
∑U putu M
zi j semin ≤ zi j u ∀i, j ∈ V (1.8)
mb
zi j M max(lu ) ≤ zi jCmax se,i mb ∀i, j ∈ V (1.9)
u∈U
Q
zi j ≤ zi jCmax se,i mb ∀i, j ∈ V (1.10)
r
∑U putu M Emax − Emin
zi j u − ≤ zi jCmax se,i mb ∀i, j ∈ V (1.11)
semax N
xi j ≤ zi j ∀i, j ∈ V (1.12)
oi = {0, 1} ∀i ∈ V (1.13)
xi j , zi j = {0, 1} ∀i, j ∈ V (1.14)
4 The objective is outlined by equation (1), which maximises the savings of using the UAM
5 service compared to driving between each origin destination pair. Constraints (1.1-1.3) ensure that
6 n vertiports are opened, and only links between opened vertiports are activated. Note that variables
7 oi , zi j and xi j are set as Boolean by (1.13) and (1.14).
8 Constraints (1.4-1.11) outline the operational conditions that must be met for a link to
9 be usable in the network. As specified by assumption 2, (1.4) and (1.5) state that the length of
10 activated paths are bounded by [Hmin , Hmax ] bounds. Constraints (1.6) constitutes assumption 4,
11 ensuring that links activated provide at least 40% travel time savings with respect to driving.
12 The remaining constraints relate to the vehicle size requirements. The specific energy of the
13 battery must lie within the bounds semin and semax given (1.7) and (1.8). Finally, (1.8-1.11) ensure
14 that the battery size satisfy the charging and discharge requirements for each vertiport i. The origin
15 of these relationships are described in the next section.

16 Vehicle Sizing
17 The vehicle sizing stage aims to minimise the battery mass required to undertake a trip defined
18 by its mission profile. As observed in equations (1.8-1.11), the size of the battery and its recharge
19 capability determine the viability of the UAM network. The methodology developed in this section
 Khalife, Slim, Escribano and Angeloudis 10

1 assumes that 50% of the vehicle mass is attributed to structure and the remaining 50% to battery
2 and payload as informed by (18). Thus, the maximum take-off mass (M) is expressed as follows:

M = 2(mb + m p ) (2)
3 Determining the battery mass requires the estimation of the operational power require-
4 ments. Using the output specified by the vertiport location algorithm, we select the trip with
5 highest power requirements and model the battery to ensure operations are feasible for this trip.
6 Thus, the energy requirement for a trip is calculated using equation (3).
U
Et = ∑(putu ) (3)
u
7 Given a maximum specific energy esmax parameter which determines the amount of energy
8 produced per unit mass of battery, the minimum battery mass mb that satisfies the design mission
9 profile is given by 4.

M maxu∈U (lu )
mb = (4)
esmax
10 Rearranging this equation leads to the following expression for battery mass:

2m p · maxu∈U (lu )
mb = (5)
esmax − 2 maxu∈U (lu )
11 Consequently, the specific energy es required to undertake a trip can be written as:
Et
es = (6)
mb
12 Uber’s operational requirements (30) requires vehicles to be able to undertake the largest
13 trip in the system for 3 hours continuously while only charging for 5 minutes between trips. This
14 minimises the opportunity costs associated with vehicles servicing less demand due to charging.
15 Consequently, loiter times would lead to increased recharge rates as the vehicle maintain
16 flight for longer periods. If the recharge rate exceeds to maximum allowable C-rate for Li-Ion
17 batteries, a vehicle will not satisfy Uber’s requirement. As such, minimising loiter time not only
18 minimises lost opportunity costs, but also reduces the peak time requirement for each vehicle. The
19 battery level En+1 after charging Echarge following a trip En that requires an energy of Etrip can be
20 obtained using the following arithmetic sequence:

En+1 = En − Etrip + Echarge (7)

21 Let ∆t be the duration of one cycle during rush hour, Emin be the required battery level at
22 the end of rush-hour and Emax be the battery level at the beginning of rush-hour. The number of
23 cycles during that window is given by N = ∆t T cycles, where T is the time to undertake the largest
24 trip in the system. To reach a desired battery level Emin at the end of rush hour, the required energy
25 recharge R is given by:
Emax − Emin
Rr = Et − (8)
N
Emax = 2(mb + m p )es BU max (lu ) (9)
u∈U
 Khalife, Slim, Escribano and Angeloudis 11

Emin = 2(mb + m p )es BL max (lu ) (10)

u∈U
1 As the battery discharges rapidly and unpredictably when it falls below its 10% threshold
2 (BL = 0.1), this limit constitutes a lower bound that should never be reached, in addition to any
3 charge reserves required. Furthermore, the top 20% take significantly longer to charge and are
4 usually ignored in UAM models (BU = 0.8) (32). Thus, 30% of the battery capacity is to never
5 be consumed, which is represented mathematically by scaling the battery capacity by a factor
6 B f = 0.7.
7 The required recharge rate R to satisfy Uber’s requirement given a charging time tr is given
8 by:
Rr
R= (11)
r
9 Li-Ion’s batteries dictate the maximum charge/discharge rate a battery can sustain. This is
10 added as a hard constraint in the vertiport placement model. The maximum discharge rate given a
11 battery capacity Bc is given by:

2 maxu∈U (lu )(mb + m p )

Rd = (12)
Q

12 Infrastructure Specifications
13 Given a vertiport configuration, the queuing theory model finds the number of landing, charging,
14 storage and take-off pads which balance the waiting times in the system, as well as the operating
15 and infrastructure costs. These waiting times are fed back into the vertiport placement model until
16 the marginal cost of removing an additional vertiport outweighs the marginal benefit.
17 Vertiports can be modelled as open-network multi-server queuing systems using Jackson’s
18 theory (33), which assumes independence of arrival rates at each server in the steady state. How-
19 ever, waiting times at each pad are not independent of each other in congested scenarios. In prac-
20 tice, congestion will propagate upstream, from storage pads to charging pads, and from charging
21 pads to landing pads. This scenario infeasible from an operational and service level perspective as
22 loiter times will exceed the predefined battery specifications and passenger waiting times will not
23 be tolerable.
24 To avoid the aforementioned scenario, the service rate at each pad must exceed the vehicle
25 arrival rate. Assuming a Poisson distributed arrival rate, and a service rate that exceeds arrival rate
26 rate, a steady state configuration can be reached. In this case, the arrival rates and queue sizes at
27 each server can be assumed to be independent of the other servers in the system.
28 However, UAM demdand is expected to have a bi-modal daily distribution, suggesting that
29 the system will not operate in a steady state configuration. Nevertheless, designing the system
30 to a steady state configuration with a similar behaviour to rush hour conditions will lead to a
31 configuration that is feasible for lower demand levels. Consequently, queuing theory can be used
32 to optimise the number of landing, charging, storage and take-off pads assuming a steady state,
33 rush-hour configuration.
34 The arrival rates at each vertiport, transition probabilities between pads of a given vertiport
35 and the service rate of each pad are obtained by running an agent-based model for a very large
 Khalife, Slim, Escribano and Angeloudis 12

1 number of pads. In this scenario, all demand is satisfied. This enables us to study the behaviour
2 of each vertiport independently. At each vertiport, the vehicles are routed to different pads in a
3 probabilistic manner.
4 Given the rush hour arrival rate Qi j , the service rate µi j of pad j at vertiport i and the set of
5 probabilities pi jz of a vehicle in vertiport i follows:
Qi j
<1 (13)
µi j mi j
6 If steady state conditions are not satisfied, the queue size is expected to increase indefinitely,
7 leading to the aforementioned congested scenario. Given an arrival rate Qi1 at the landing pads of
8 vertiport i, the transitional probabilities pi jk of a vehicle transitioning from pad type j to k at
9 vertiport i, the arrival rate Qi j at each pad j of vertiport i is calculated as follows:
Z
Qi j = ∑(pik j )Qi1 (14)
z
For each pad j of vertiport i, the utilisation ρi j is defined as:
Qi j
ρi j = (15)
µi j
The probability P0i j of an empty queue at pad j of vertiport i is calculated as follows:
c−1 ρ n ρicj
ij −1
P0i j = ∑ [ + ρi j ] (16)
n=0 n! c!(1 − c )
The average length of the queue λqi j is given by:
ρ
P0i j ρi j c ci j
λqi j = ρ (17)
c!(1 − ci j )2
Finally, the waiting time Wqi j at pad j of vertiport i is given by:
λqi j
Wqi j = (18)
Qi j
10 Waiting times directly affect the vertiport placement and vehicle sizing models. Larger
11 wait times requirements reduce the travel time savings relative to driving and overall benefit of
12 the UAM network. Longer loiter times lead to larger battery sizes, which can make some trips
13 unfeasible due to the recharge requirements. Larger vehicle size and increased energy depletion
14 due to loiter increase the cost of vehicle procurement, charging and maintenance. Consequently,
15 waiting times heavily influence the operational feasibility of the UAM network and the system
16 costs.

17 Operational and Infrastructure costs

18 The operating costs in this study are composed of: vehicle maintenance, vehicle insurance, pilot
19 salary, energy costs, battery procurement and life cycle costs, indirect costs, carbon tax and op-
20 portunity costs. The values and reference for all the parameters introduced in this section will be
21 outlined in the Case Study section.
22 Maintenance costs are calculated assuming a specific rate of maintenance requirements per
23 flight-hour. Vehicle acquisition costs are based on the vehicle maximum take off weight, with
24 insurance consisting of a percentage of the total vehicle acquisition cost. Crew costs include all
25 pilot salaries given the number of vehicles required. Energy and battery costs are determined based
26 on the estimated average flight time which includes loitering, with battery life-cycles obtained as
 Khalife, Slim, Escribano and Angeloudis 13

1 per (28).
2 Indirect costs include credit card fees, overhead for commercial aviation, taxing, landing
3 fees and others and are assumed to represent 10% of the total costs. Opportunity costs are calcu-
4 lated based on the proportion of waiting time Tw from the total trip time Tt :

Tw
5 O=α Co (19)
Tt
6 where α corresponds to the operating margin (29), and Co to the operational costs.
7 The main infrastructure cost components are associated with the landing, charging, storage,
8 and takeoff pads. This includes the cost of piling, composite decking as well as the carrier terminal
9 cost. These costs are used to determine the infrastructure cost of different vertiports deployment
10 scenarios. Additionally, a high voltage charger cost of is included for each charging pad. The
11 take-off and landing pads for eVTOL vehicles are comparable to a helicopter pad.

12 Feedback Loop
13 Increasing the number of landing, charging and storage pads will lead to improve waiting times
14 and reduce opportunity costs provided the operational requirements are satisfied. Despite these
15 improvements, it will also increase the infrastructure costs.
16 This interrelationship is captured using a feedback mechanism, which collects the output
17 of the vertiport infrastructure model and updates the constraints of the vertiport placement model
18 based on the requirements of the vertiport infrastructure. The feedback loop aims to find a balance
19 between the infrastructure, operational costs and passenger waiting times. The number of pads is
20 decreased with every iteration until the marginal cost exceeds the marginal benefit or the problem
21 becomes infeasible.

22 CASE STUDY
23 The method described in the methodology is applied to find the optimal system configuration for a
24 potential deployment scenario in the city of London. This study assumes a five vertiport configu-
25 ration for the short-term UAM application. To showcase the study’s contribution, a baseline model
26 based on current literature is developed which ignores the interdependencies between vertiport
27 placement, vehicle sizing and infrastructure specifications. This baseline is used as a comparison
28 to the integrated approach proposed in this paper. All cost parameters used are presented in Table
29 2.

30 Demand Generation
31 As stated by (31), UAM is still in its early stages of development to forecast absolute demand
32 levels for trips within a city. However, using existing transportation data, one can estimate relative
33 demand levels between different areas for the purpose of vertiport placement.
34 Assumption 1: TfL’s Rolling Origin and Destination Survey (RODS) is assumed to give an
35 adequate representation of London’s underground travellers’ behaviours
36 The RODS captures important statistics on trips undertaken in the London Underground
37 Limited (LUL). It provides an origin-destination matrix classified by station, zone, and time of day.
38 The data is based on November 2017 counts and abnormal fluctuations in demand due to unusual
39 conditions are neglected (40). As such, the RODS is considered to give an adequate representation
40 of London’s underground travel behaviours on a typical weekday and will be used to infer air-taxi
41 demand.
 Khalife, Slim, Escribano and Angeloudis 14

TABLE 2 Cost parameters.

Operating costs
Parameter Value Reference
Mechanic salary [$/h] 60 (28)
Maintenance time [h] 0.68 (28)
Vehicle acquisition [$/kg] 333 (28)
Pilot salary [$/h] 100 (23)
Pilot yearly flight-hours [h] 500 (28)
Energy costs [$/MJ] 0.0492 (34)
Emissions rate [kg/MJ] 0.0786 (35)
Battery acquisition [$/MJ] 111 (28)
Carbon tax [$/kg] 0.0198 (36)
Operating margin [-] 0.3 (7)
Infrastructure costs
Vertiport costs [$/m2 ] 4122 (37–39)
High-voltage charger [$] 453 (4)

1 Assumption 2: Each borough can hold at most one vertiport

2 Most vertiport placement studies have been undertaken in the United States, where the
3 demand is usually aggregated on a census tract level (31), (11), equivalent in size to London wards.
4 Considering the differences between the United Kingdom’s and United States’ potential air-taxi
5 market (41), it is more adequate to discretise London into larger geographical entities. Further,
6 there are more wards than London underground stations which will result in many OD pairs not
7 having demand associated to them. As a result, a coarser, borough level aggregation was chosen,
8 which produced the candidate presented in Figure 2 and Table 3. One limitation of this approach
9 is that the boroughs of Bexley, Bromley, Croydon, Kingston upon Thames and Sutton, which are
10 not served by underground stations, do not have demand associated to them.

FIGURE 2 London borough map

11 Assumption 3: Applying (20)’s distance weighing factor to the aggregated RODS demand
12 gives us a good indication of relative UAM demand between boroughs
 Khalife, Slim, Escribano and Angeloudis 15

1 To represent realistic air-taxi demand levels based on existing transport datasets, (21) define
2 a distance weighing function, which is calculated based on the haversine distance d as follows:

1 2
3 wd = d 2 e−0.001d (20)
367.8791
4
5 The distance weighing factor centres demand around distances of 8 to 20 km: trips that are
6 too short or too long to be attractive by air-taxi have been de-emphasised. The weighted demand
7 distribution is thus more suitable for representing potential air-taxi demand and will be used as an
8 input to the vertiport placement model.

9 Results
10 To reflect the current state-of-the-art, the baseline model developed for this study is defined by
11 equations 1-1.7 and 1.12-1.14. Rather, vehicle sizing is undertaken by post-processing the vertiport
12 placement’s outputs. The effects of waiting times on the system will also be ignored.
13 Our holistic model proposed in Section 4, is also used to incorporate the effects of waiting
14 times and vehicle performance into the original vertiport placement model, as well as determining
15 the optimal pad configuration at each vertiport. An initial upper boundary pad configuration of
16 16 pads per vertiport is used to ensure operational feasibility. This number is reduced until the
17 marginal cost of removing a certain pad outweighs the marginal benefit of doing so, provided the
18 operational constraints are satisfied. To quantify the effects of increasing demand on the system,
19 the model is run for three rush-hour demand scenarios: 50, 150 and 250 vehicles/hour.

FIGURE 3 Vertiport placement results for the baseline and optimal case. Haringey, West-
minster and Newham are selected in both cases.

20 All demand scenarios lead to the same optimal vertiport configuration and vehicle size. In
21 comparison to the baseline model the longest trip in the system is 39% shorter, that being the 14km
22 trajectory connecting Barnet to Westminster. The baseline selects the Brent-Newham connection
23 of 23km as its longest trajectory.
24 In spite of the shorter travel distances, the battery size in the holistic model is 6% larger
25 than in the baseline case. This suggests that the vehicle sizing model is unable to produce a feasible
26 battery size for this trip, highlighting a key limitation of existing methods.
27 Figure 5 highlights that, although both vehicle configurations are effectively able to fly the
28 largest trip in the system, the baseline configuration does not satisfy Uber’s operational require-
Khalife, Slim, Escribano and Angeloudis 16

TABLE 3 Borough candidate vertiport location.

Borough Latitude Longitude
Enfield 51.6538 -0.0799
Barnet 51.6252 -0.1517
Haringey 51.6 -0.1119
Waltham Forest 51.5908 -0.0134
Harrow 51.5898 -0.3346
Havering 51.5812 0.1837
Brent 51.5588 -0.2817
Redbridge 51.559 0.0741
Hackney 51.545 -0.0553
Hillingdon 51.5441 -0.476
Islington 51.5416 -0.1022
Camden 51.529 -0.1255
Ealing 51.513 -0.3089
Tower Hamlets 51.5099 -0.0059
Newham 51.5077 0.0469
Southwark 51.5035 -0.0804
Kensington and Chelsea 51.502 -0.1947
Westminster 51.4973 -0.1372
Hammersmith and Fulham 51.4927 -0.2339
Greenwich 51.4892 0.0648
Hounslow 51.4746 -0.368
Lambeth 51.4607 -0.1163
Wandsworth 51.4567 -0.191
Bexley 51.4549 0.1505
Richmond upon Thames 51.4479 -0.326
Lewisham 51.4452 -0.0209
Kingston upon Thames 51.4085 -0.3064
Bromley 51.4039 0.0198
Merton 51.4014 -0.1958
Croydon 51.3714 -0.0977
Sutton 51.3618 -0.1945
City of London 51.5155 -0.0922
Barking and Dagenham 51.5607 0.1557
 Khalife, Slim, Escribano and Angeloudis 17

FIGURE 4 Battery state of charge variation during peak-time.

1 ments under reasonable C-rate assumptions. In fact, battery levels deplete below reserve after only
2 undertaking 2 trips while charging at a 5C rate. The holistic model provides sufficient capacity to
3 undertake the longest trip in the system continuously for 3 hours while only charging for 5 minutes
4 at a 4.8 C-rate.
5 With notion of the effects of the pad configuration on waiting times, and hence operational
6 costs, the holistic model can also be used to determine the optimal pad configuration at each verti-
7 port.

FIGURE 5 Yearly operational costs.

8 The yearly operational cost is £198 million for the 250 vehicle/hour scenario, compared
9 to £119 and £40 million in the 150 and 50 vehicle/hour scenario, respectively. Nevertheless, the
 Khalife, Slim, Escribano and Angeloudis 18

TABLE 4 Vertiport infrastructure requirements and costs.

Scenario Low-Demand Medium-Demand High-Demand
Number Landing Pads [-] 16 25 34
Number Charging Pads [-] 16 29 38
Number Take-off Pads [-] 11 22 29
Waiting Time Proportion [%] 3.1 0.91 0.56
Landing Pads Cost [$ million] (% of total) 4.39 (47) 6.86 (42) 9.33 (43)
Charging Pads Cost [$ million] (% of total) 1.85 (20) 3.35 (21) 4.39 (20)
Take-off Pads Cost [$ million] (% of total) 3.02 (32) 6.04 (37) 7.96 (37)
Total Infrastructure Cost [$ million] 9.26 16.25 21.69

1 operational cost breakdown is similar for all scenarios, with batteries, vehicle maintenance and
2 crew incurring the largest costs.
3 It is also interesting to note that opportunity costs are minimal in all scenarios. In fact,
4 total waiting times remain below 5% of the total flight time in all optimal configurations, with 150
5 and 250 vehicle per hour demand scenarios yielding waiting times that constitute under 1% of the
6 total flight time. Waiting times increase the energy, opportunity and maintenance costs, leading to
7 greater marginal costs that out weight the benefit of removing a vertiport pad.
8 Due to the Li-Ion’s battery life and price ranges, the battery costs are the highest opera-
9 tional cost component, which is in line other studies (7). Improvements in battery performance
10 thus offers significant cost reduction opportunities. (28) assumption of 33 minutes per flight hour
11 for maintenance makes associated costs significant. Further, automation has the opportunity to
12 eliminate pilot costs, but will lead to additional costs which should be quantified.
13 Increasing peak hour demand from 50 to 150 and 250 vehicles/hour leads to a 76 and 130%
14 increase in the total number of pads. Under all conditions, a larger number of charging pads are
15 required compared to the other types. This makes intuitive sense due to the longer service time of
16 charging pads. Further, there are more landing pads than take-off pads in all scenarios, suggesting
17 that marginal cost of removing a landing pad is larger than the one associated with removing a
18 take-off pad.
19 The yearly infrastructure cost is £21.7 million for the high demand configuration, 134%
20 larger than the lower demand scenario (£9.5 million). Although all configurations require more
21 charging pads than any other pad type, landing pads account for the highest proportion of infras-
22 tructure costs as shown in Table 4.
23 It is interesting to note that operating costs are orders of magnitude larger than infrastructure
24 costs in both scenarios, strengthening the notion that, if retrofitted into existing infrastructure,
25 UAM offers significant cost advantages compared to other transport modes.

26 DISCUSSION
27 The study demonstrates the importance of adopting a holistic approach to UAM network design.
28 Its major contribution lies in the development of a method that optimises the main components of
29 UAM systems, while also considering the inter-dependencies between them. By integrating vehicle
30 sizing and performance constraints into the vertiport placement model, the proposed methodology
31 generates a vertiport configuration that satisfies Uber’s peak time requirements.
32 In contrast, the baseline model yields an infeasible solution that is only detected during
 Khalife, Slim, Escribano and Angeloudis 19

1 post-processing. A potential solution is to reduce the number of opened links between the network
2 in order to reduce the peak demand at the vertiports, but this solution will result in a suboptimal
3 UAM network.
4 Our results show that under low waiting times, battery acquisition and replacement cost
5 are the largest operating cost component. Nevertheless, opportunity costs quickly dominate when
6 waiting times increase, making UAM networks unsustainable. Thus, UAM operations should be
7 designed to operate under minimal turnaround times.
8 Another interesting insight is that operating costs grow at a faster rate than infrastructure
9 costs when waiting times increase. Therefore, it is advisable to design vertiports to accommodate
10 near 0 theoretical average waiting times at rush-hour. Loiter times lead to the largest increases
11 in operational costs. Not only do they lead to opportunity cots, but they also deplete additional
12 energy, therefore requiring larger batteries and larger charging times.
13 Sub-system level interactions largely affect the wider UAM network. Increases in waiting
14 times lead to changes in network configuration, operating costs and battery requirements which
15 can make the system unfeasible. Waiting times are heavily dependent on the configuration of pads
16 at each vertiport. Consequently, feedback loops can be used to model the effects of sub-system
17 interactions on the wider UAM system to a reasonable degree.
18 In spite of these findings, there are several limitations with the approach in this study. The
19 demand modelling is on RODS (40) which only consider trip undertaken in the London Under-
20 ground. The model can be improved by considering other modes of transport and including a logit
21 decision model based on the utility parameter of each mode (42). This includes analysing the
22 effects of different pricing schemes in UAM demand as well as queue management.
23 Furthermore, multi-mode trips, where UAM can be used to complement existing transport
24 modes, are ignored in this study. The safety routing factor mu specified in Assumption 2 simplifies
25 the potential effects of air traffic management and noise in the routing of eVTOLs and the con-
26 figuration of vertiports. Finally, incorporating stochastic travel times, waiting times, and demand
27 levels would provide a more robust network, particularly when incorporating weather effects on
28 demand and flight times.

29 CONCLUSIONS
30 This study proposes a holistic approach to optimising UAM networks by considering sub-system
31 interactions on the wider UAM network. The methodology contains three main components: a
32 vertiport placement model, a vehicle sizing process, and a infrastructure queuing model. The latter
33 can be modelled using multi-server open network theory by considering a rush-hour, steady state
34 demand configuration, and its outputs are used to modify the constraints of the vertiport place-
35 ment model. This approach overcomes the speed limitations related to simulation-optimisation
36 approaches that could be used with the existing agent-based models in the literature.
37 The method is applied to find the configuration that optimises operational, infrastructure
38 costs as well as service times for the city of London under different demand levels. Results show
39 that UAM systems could effectively be deployed in small scale, with low turnaround times. The
40 findings show that operational costs significantly increase with waiting times, and as such these
41 should be reduced under all circumstances, particularly at higher demand levels. Thus, vertiport
42 design approaches not considering vertiport congestion will lead to suboptimal or infeasible con-
43 figurations, as was the case with the baseline approach used in this study.
44 Further work is required to improve the demand modelling to include a logit decision model
 Khalife, Slim, Escribano and Angeloudis 20

1 that considers competing transport modes and multi-modal trips. Moreover, travel times, waiting
2 times, and demand could be modelled stochastically. With the addition of pricing models, these
3 changes would allow the determination of expected profit margins and optimal investment strate-
4 gies.

5 AUTHOR CONTRIBUTION STATEMENT

6 The authors confirm contribution to the paper as follows: Khalife, Slim, Escribano Macias and An-
7 geloudis carried out the study conception and design, Escribano Macias, Slim and Khalife devel-
8 oped the models and analysed the results. Escribano Macias and Khalife prepared the manuscript.
9 All authors have reviewed the results and approved the final version of the manuscript.
 Khalife, Slim, Escribano and Angeloudis 21

1 REFERENCES
2 1. Goyal, R., Urban Air Mobility (UAM) Market Study, 2018.
3 2. Holmes, B. J., A vision and opportunity for transformation of on-demand air mobility. In
4 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016, p. 3465.
5 3. NASA Aeronautics Research Mission Directorate, Strategic Implementation Plan. NASA,
6 Washington DC, 2019.
7 4. Holden, J. and N. Goel, Fast-forwarding to a future of on-demand urban air transportation,
8 2016.
9 5. Parker, R. A., NASA Strategic Framework for On-Demand Air Mobility A Report for NASA
10 Headquarters, 2017.
11 6. Chan, W. N., B. E. Barmore, J. L. Kibler, P. U. Lee, C. J. O’Connor, K. Palopo, D. P. Thip-
12 phavong, and S. J. Zelinski, Overview of NASA’s Air Traffic Management-eXploration
13 (ATM-X) Project, 2018.
14 7. Johnston, T., R. Riedel, and S. Sahdev, To take off, flying vehicles first need places to land.
15 McKinsey & Company, Boston, MA, USA, 2020.
16 8. Klose, A. and A. Drexl, Facility location models for distribution system design. European
17 Journal of Operational Research, Vol. 162, 2005, pp. 4–29.
18 9. Hakimi, L., Optimum Locations of Switching Centers and the Absolute Centers and Me-
19 dians of a Graph. Operations Research, Vol. 12, No. 3, 1964, pp. 450–459.
20 10. Campbell, J. F., Hub location and the p-hub median problem. Operations Research,
21 Vol. 44, No. 6, 1996, pp. 923–935.
22 11. Daskilewicz, M., B. German, M. Warren, L. A. Garrow, S.-S. Boddupalli, and T. H.
23 Douthat, Progress in Vertiport Placement and Estimating Aircraft Range Requirements
24 for eVTOL Daily Commuting. In 2018 Aviation Technology, Integration, and Operations
25 Conference, 2018, p. 2884.
26 12. Wei, L., C. Y. Justin, and D. N. Mavris, Optimal Placement of Airparks for STOL Urban
27 and Suburban Air Mobility. In AIAA SciTech Forum, Orlando, FL, USA, 2020, pp. 1–16.
28 13. Rath, S. and J. Y. J. Chow, Air Taxi Skyport Location Problem for Airport Access. arXiv
29 preprint arXiv:1904.01497, 2019.
30 14. Wu, Z., Exploration of On-Demand Urban Air Mobility : Network Design, Operation
31 Scheduling and Uncertainty Considerations, 2020, pp. 1–4.
32 15. Lim, E. and H. Hwang, The Selection of Vertiport Location for On-Demand Mobility and
33 Its Application to Seoul Metro Area. International Journal of Aeronautical and Space
34 Sciences, Vol. 20, No. 1, 2019, pp. 260–272.
35 16. Rajendran, S. and J. Zack, Insights on strategic air taxi network infrastructure locations us-
36 ing an iterative constrained clustering approach. Transportation Research Part E: Logistics
37 and Transportation Review, Vol. 128, No. July, 2019, pp. 470–505.
38 17. Johnson, W., NDARC-NASA Design and Analysis of Rotorcraft Theoretical Basis and
39 Architecture, 2010.
40 18. Silva, C., W. R. Johnson, E. Solis, M. D. Patterson, and K. R. Antcliff, VTOL Urban Air
41 Mobility Concept Vehicles for Technology Development. In 2018 Aviation Technology,
42 Integration, and Operations Conference, 2018, p. 3847.
43 19. Johnson, W., C. Silva, and E. Solis, Concept vehicles for VTOL air taxi operations. Pro-
44 ceedings of the AHS International Technical Meeting on Aeromechanics Design for Trans-
45 formative Vertical Flight 2018, 2018.
 Khalife, Slim, Escribano and Angeloudis 22

1 20. Kohlman, L. W. and M. D. Patterson, System-Level Urban Air Mobility Transportation

2 Modeling and Determination of Energy-Related Constraints. 2018 Aviation Technology,
3 Integration, and Operations Conference, 2018.
4 21. Kohlman, L. W., M. D. Patterson, and B. E. Raabe, Urban Air Mobility Network and
5 Vehicle Type-Modeling and Assessment, 2019.
6 22. Courtin, C. and R. J. Hansman, Model Development for a Comparison of VTOL and STOL
7 Electric Aircraft Using Geometric Programming. , No. June, 2019.
8 23. Brown, A. and W. Harris, A vehicle design and optimization model for on-demand avia-
9 tion. In 2018 AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Con-
10 ference, 2018, p. 105.
11 24. Kadhiresan, A. R. and M. J. Duffy, Conceptual Design and Mission Analysis for eVTOL
12 Urban Air Mobility Flight Vehicle Configurations. In AIAA AVIATION Forum, Dallas, TX,
13 USA, 2019, pp. 1–19.
14 25. Melton, J., D. Kontinos, S. Grabbe, J. Alonso, J. Sinsay, and B. Tracey, Combined elec-
15 tric aircraft and airspace management design for metro-regional public transportation.
16 NASA/TM, Vol. 216626, 2014, p. 2014.
17 26. Patterson, M. D., K. R. Antcliff, and L. W. Kohlman, A Proposed Approach to Studying
18 Urban Air Mobility Missions Including an Initial Exploration of Mission Requirements,
19 2018.
20 27. Rothfeld, R., M. Balac, K. O. Ploetner, and C. Antoniou, Initial Analysis of Urban Air
21 Mobility’s Transport Performance in Sioux Falls. 2018 Aviation Technology, Integration,
22 and Operations Conference, 2018.
23 28. Booz Allen Hamilton, UAM Market Study - Technical Out Brief, 2018.
24 29. NASA Urban Air Mobility, Urban Air Mobility (UAM) Market Study, 2018.
25 30. Uber, eVTOL Vehicle Requirements and Missions, 2018.
26 31. German, B., M. Daskilewicz, T. K. Hamilton, and M. M. Warren, Cargo Delivery in by
27 Passenger eVTOL Aircraft: A Case Study in the San Francisco Bay Area. In AIAA SciTech
28 Forum, Kissimmee, FL, USA, 2018, 2018 AIAA Aerospace Sciences Meeting.
29 32. McDonald, R. and German B., eVTOL Stored Energy. In 2017 Uber Elevate Summit,
30 2017.
31 33. Chen, H. and D. Yao, Jackson Networks. In Fundamentals of Queueing Networks. Stochas-
32 tic Modelling and Applied Probability, vol 46, Springer, New York, NY, 2001, pp. 15–35.
33 34. PowerCompare, Compare Electricity Prices: Average UK Rates Tariffs Per kWh, 2019.
34 35. Kohlman, L. W., M. D. Patterson, and B. E. Raabe, Urban Air Mobility Network and
35 Vehicle Type—Modeling and Assessment. , No. February, 2019, p. 43.
36 36. HM Revenues and Customs, Carbon Emissions Tax, 2018.
37 37. Turner & Townsend, International construction market survey 2016. Global rebalancing:
38 a changing landscape, 2015, pp. 1–76.
39 38. Taylor, M., A. Saldanli, and A. Park, Analysis of Alternate Vertiport Designs. In Proceed-
40 ings IEEE ICNS Conference, IEEE, Dulles, Virginia, 2020.
41 39. Fixr, Compare costs estimates, 2021.
42 40. Transport for London, TfL Rolling Origin And Destination Survey, 2019.
43 41. Moore, M. D., Uber elevate: evtol urban mobility. In Rotorcraft Business and Technology
44 Summit, 2017.
 Khalife, Slim, Escribano and Angeloudis 23

1 42. Fu, M., A. Straubinger, and J. Schaumeier, Scenario-based Demand Assessment of Urban
2 Air Mobility in the Greater Munich Area, 2020, pp. 1–16.