Sage Journals: Discover world-class research

Abstract

Urban air mobility (UAM) is a concept envisioning safe, sustainable, affordable, and accessible air transportation for passenger mobility, cargo delivery, and emergency management within or traversing a metropolitan area. While the deployment of UAM will require a network of infrastructure (i.e., vertiports), this emerging concept could face a variety of barriers, such as local opposition, infrastructure costs, and multimodal integration. This paper presents a novel simulation methodology using k-medians clustering and sensitivity analysis to model potential vertiport locations in the San Francisco Bay Area. When controlling for land use characteristics (i.e., parcel size and zoning), the simulation and sensitivity analysis finds that as the number of vertiports increases, the average travel times of trips taken by UAM decrease. The analysis also identifies an inflection point of the number of vertiports at which UAM becomes competitive with driving. The analysis concludes that with respect to size, parcels of 2 acres or smaller are the only locations that allow UAM travel times to be competitive with driving. The paper concludes that neither quantitative nor qualitative factors alone will be sufficient to determine the feasibility of vertiports and overcome community concerns associated with vertiport siting. Metropolitan planning organizations, local governments, and other public agencies will likely need to augment technical analysis with robust stakeholder and community engagement throughout the planning process.

Keywords

urban air mobility advanced air mobility modeling vertiports network planning land use

Urban air mobility (UAM) is a concept that envisions safe, sustainable, affordable, and accessible air transportation for passenger mobility, cargo delivery, and emergency management within or traversing a metropolitan area ( 1 – 5 ). In recent years, developments in electric and hydrogen propulsion, vertical and short takeoff and landing (VTOL/STOL) aircraft, and other technologies have created renewed interest in on-demand urban aviation ( 1 , 2 ).

While UAM may be enabled by the convergence of many factors, several challenges, including community acceptance, safety, equity, planning and implementation, airspace management, and operations, could create barriers to mainstreaming ( 5 , 6 ). The deployment of UAM will require a network of landing facilities, charging and/or refueling infrastructure, and digital infrastructure for advanced air traffic management. While UAM may be able to use existing helipads initially, growth of the sector will likely require additional facilities for aircraft (e.g., takeoff and landing, parking, and maintenance) and passengers (e.g., waiting, boarding, access, and egress) ( 2 , 7 –10). However, establishing new facilities could face a variety of barriers, such as local opposition, infrastructure costs, and multimodal integration ( 5 , 11 , 12 ).

Several market studies conducted before the global pandemic estimate a market potential for UAM passenger service between US$2.8 and US$4 billion in the United States by 2030 ( 13 , 14 ). However, Johnston et al. ( 13 ) and Lineberger et al. ( 14 ) found that limited infrastructure has the potential to constrain the growth of UAM.

Several studies have attempted to develop a standardized classification for UAM takeoff and landing facilities ( 13 , 14 ). Generally, these studies classify three types of facilities, from largest to smallest: 1. vertihubs; 2. vertiports/vertibases; 3. vertipads/vertistations. Table 1, adapted from Cohen et al. ( 5 ), provides additional information on these facilities. Other terms include skyports and aerodromes. UAM may also be able to use existing infrastructure such as heliports, airports, and waterways (for amphibious operations). In this paper, UAM infrastructure will be referred to as vertiports ( 15 ).

Table 1.

Taxonomy and Definitions of Urban Air Mobility (UAM) Infrastructure.

Type of UAM infrastructure	Description
Vertihub	A larger facility (possibly with multiple floors) to accommodate numerous landing pads with parking for multiple aircraft. Johnston et al. ( 13 ) estimate vertihubs would be approximately 70,000 ft², spread across multiple floors, and cost between US$6 and $7 million to build.
Vertiport/vertibase	A medium-sized facility intended to accommodate up to three landing pads and up to six parked aircraft. Johnston et al. ( 13 ) estimate vertiports/vertibases would be approximately 23,000 ft² and cost between US$500,000 and $800,000 to construct.
Vertipad/vertistation	A single landing pad and parking stall intended to accommodate one or two parked aircraft. Johnston et al. ( 13 ) estimate that vertipads would be approximately 6,000 ft² and cost between US$200,000 and $400,000 to construct.

Source: Adapted from Cohen et al. ( 5 ).

The deployment of UAM will require a growing network of vertiports as services are deployed. While UAM may be able to use existing aviation facilities such as helipads and airports, new infrastructure close to traveler origins and destinations will likely be needed to reduce the number of modal connections and travel times associated with first- and last-mile connections to vertiports. In urban centers, this can present several notable challenges caused by limited land for new vertiports, heights of existing buildings that could affect airspace to and from new vertiports, availability of supporting infrastructure (e.g., charging/fueling, information technology, etc.), and the impacts of vertiport operations on surrounding communities. These issues could be magnified if vertiports are exclusively available to one service provider (i.e., exclusive-use) in markets with multiple services. As such, the construction of an urban vertiport network may face several challenges.

In a fully mature network, Johnston et al. ( 13 ) estimate that large cities (e.g., London, New York City, and Tokyo) could require 85 to 100 takeoff and landing pads comprised of three to five vertihubs, 10 to 15 vertiports, and five to 10 vertipads. The study concludes that UAM infrastructure will likely be smaller scale during initial service deployment and will require a combination of private sector investment, public sector subsidies, and non-aeronautical revenue sources (similar to revenue generated by airports from retail, landside access, and other sources) ( 13 ).

Vertiport Siting and Network Design

Several studies qualitatively and quantitatively examine vertiport siting and network design. Fadhil et al. ( 16 ) utilized a weighted linear combination method to identify potential vertiport locations in both Los Angeles and Munich using qualitative data from subject matter experts. Rothfeld et al. ( 17 ) used a similar weighted combination based on qualitative location suitability and a location-allocation optimization algorithm ( 9 , 17 , 18 ). The study used MATSim, an agent-based simulation framework for travel behavior analysis. Daskilewicz et al. ( 19 ) analyzed the spatial distribution of jobs, income, and population in the San Francisco Bay Area and Los Angeles to maximize travel time savings. The study concluded that optimal vertiport locations would likely be in higher income census tracts.

Uber Elevate ( 2 ) developed a k-means clustering method using long-distance Uber rides to identify potential vertiport locations in Los Angeles and London. However, because the study uses proprietary data from one service provider, the results are limited. Bulusu et al. ( 20 ) also used a k-means clustering model to explore the San Francisco Bay Area. However, demand data for the study were not comprehensive for the entire metropolitan region, an important consideration to understand the broader impacts across the transportation network ( 21 ). Lim and Hwang ( 22 ) and Syed et al. ( 23 ) also used k-means clustering in Seoul and Washington D.C., respectively. Both studies found that the location of vertiports is more important than the number of vertiports. One notable limitation of k-means clustering is that vertiport locations are susceptible to outliers, which may place vertiports further away from the majority of demand. If vertiports are further away, then demand for UAM may change altogether. Additionally, some vertiport locations considered suitable in the model may be impractical in the real world when considering noise, equity, and other concerns.

Several studies have found that UAM adoption could be influenced by vertiport placement and feasibility constraints, such as zoning and parcel areas of potential locations ( 4 , 18 , 24 ). Robinson et al. ( 25 ) used vacant land parcel information in South Florida to understand the potential for STOL aircraft. However, the larger parcels required for STOL operations may not be needed for VTOL aircraft ( 25 ). Wu and Zhang ( 26 ) developed a single-allocation hub-and-spoke optimization formulation for Tampa, Florida. The study used LiDAR and geographic information systems (GIS) to identify potential vertiport locations based on available ground spaces and land use restrictions ( 26 ).

Antcliff et al. ( 27 ) suggested that the San Francisco Bay Area could be an early adopter market for UAM because of the high percentage of long-distance or “super” commuters, weather, and geography. In addition, the existence of multiple urban centers in the region tends to result in bidirectional traffic patterns that could reduce “deadhead” flights without paying passengers and improve the operational efficiency and profitability of early operations ( 28 ).

The remainder of this paper is organized into three sections. The next section introduces a novel modeling methodology that could be used to estimate the number of recommended vertiports using fast, high-fidelity, regional-scale microsimulation. The third section highlights the results of the model. The final section concludes with a discussion of limitations of the study and proposes recommendations to help advance UAM infrastructure planning, implementation, and community integration.

Methodology

Vertiport network design requires determination of the location of every vertiport in the metropolitan area with all the other locations of vertiports in mind. This paper presents a clustering algorithm for optimal vertiport locations, based on existing travel demand in the San Francisco Bay Area. These locations are made feasible by introducing land use and zoning constraints. In this section, the microsimulator, street network, demand, network optimization, land use constraints, and overarching simulation procedure are described.

Microsimulator

This paper uses Microsimulation Analysis for Network Traffic Assignment (MANTA), a fast, parallelized, GPU-based regional-scale traffic microsimulator. MANTA can simulate millions of trips along a given street network in only a few minutes, using the intelligent driver model, gap acceptance, and lane changing algorithms to move the vehicles on each edge as they interact with other vehicles ( 29 ). This version of MANTA also leverages a priority queue Dijkstra’s algorithm for the routing of trips. MANTA has been calibrated and validated with open source Uber Movement data ( 29 ).

MANTA is used over existing simulators for several reasons. In transportation simulation, tradeoffs exist across granularity, spatial scale, computational time, and simulator costs. While many simulators exist that satisfy several of these, none is able to satisfy all four to enable quick scenario planning. A high granularity is necessary for UAM analysis because of the sensitivity of UAM demand to variations in travel time ( 2 ). The spatial scale also needs to encompass greater metropolitan regions since UAM is expected to address longer trips by distance and time ( 2 , 4 ). The high granularity and large spatial scales necessary for analysis means that computational times of existing simulators are too long—in the order of days—for fast sensitivity analysis ( 29 ). Since MANTA is a GPU-based simulator, the cost of the processors is also more affordable than more powerful machinery such as high-performance computing architectures, which allows it to be accessible to researchers ( 29 ).

Network

The street network is constructed from the OpenStreetMap (OSM) network of the San Francisco Bay Area using the OSMnx library. The network is cleaned to remove nodes at curves or bends in edges in OSM that are not topologically accurate ( 30 ). The resulting network contains 224,223 nodes and 549,008 edges. The metropolitan region’s nodes are shown in Figure 1.

Figure 1.

Demand

This study uses the San Francisco Bay Area Metropolitan Transportation Commission’s (MTC) Travel Demand Model One. The MTC model uses a population synthesizer that locates households based on the 2000 Decennial Census Public Micro-sample, where each traffic analysis zone (TAZ) typically has a similar population and can range from the size of a block to several square miles ( 31 ). It is then validated across years 2000, 2005, 2010, and 2015. Using this behavioral data, this study uses a granular microsimulation that randomly assigns origins and destination, sampled from a uniform distribution, within a TAZ to a network node in that TAZ ( 32 ).

A simplified method is used to determine demand for UAM, by filtering automobile trips with travel times $\geq 30$ min, known as Marchetti’s Constant ( 33 ). Marchetti’s Constant is limited by its lack of true behavioral realism, but can be a useful tool because of its simplicity ( 34 ). A preliminary run in MANTA was performed using 3.2 million trips from San Francisco Bay Area MTC between 5:00 a.m. and midnight on a typical day, and the resulting travel times were then filtered to ≥30 min ( 29 ).

Network Optimization

The vertiport network design problem can be viewed as an uncapacitated variant of the facility location problem (FLP), a common formulation in transportation and operations research ( 35 ). This problem is uncapacitated because the maximum number of aircraft is not specified for each facility. The objective is to minimize the sum of the distances traveled to and from every traveler’s nearest vertiport, based on every traveler’s origin and destination. The uncapacitated FLP is shown in optimization.

minimize \sum_{i \in V, j \in D} d_{ij} x_{ij}

(1a)

subject to \sum_{i \in V} x_{ij} = 1 : \forall j \in D,

(1b)

\sum_{i \in V} y_{i} \leq k,

(1c)

\forall i \in V, \forall j \in D : x_{ij} \in {0, 1},

(1d)

\forall i \in V : y_{i} \in {0, 1}

(1e)

where $V$ is the number of vertiports, $D$ is the number of travelers (demand), $x_{ij}$ is whether traveler $j$ is connected to vertiport $i$ , $d_{ij}$ is the distance between vertiport $i$ and traveler j’s origin or destination point, and $y_{i}$ is if the vertiport is open to use. The first constraint in Equation 1b indicates that every demand point is assigned to only one vertiport. The second constraint in Equation 1c requires that the number of active vertiports must be less than or equal to the specified $k$ , the maximum number of vertiports. The third and fourth constraints in Equations 1d and 1e indicate that $x_{ij}$ and $y_{i}$ , respectively, are binary variables. The assumption in this formulation is that every traveler takes only one UAM leg in their multimodal journey, the most likely scenario in scaled UAM operations ( 2 ).

In this study, this linear programming problem is reformulated using an unsupervised k-medians clustering approach. Thus, for each cluster, the objective function transforms into Algorithm 1. The first step of the algorithm is to select $k$ points randomly from the demand as cluster centroids, sampled from a uniform distribution. Then, at every iteration $t$ , each node is added to a cluster based on its minimum distance to each of the $k$ centroids. The next step recalculates the median of that cluster, and every new median cluster point has its cluster points $d$ recalculated, again based on the minimum distance of each demand point to the nearest of the $k$ clusters. Finally, the algorithm converges to a set of cluster centroids with its associated demand points.

Algorithm 1 k-Medians Clustering for Vertiport Location
1: procedureK-MEDIANS( $D$ ) 2: $D = {d_{1}, d_{2}, . . ., d_{n}}$ is the set of input demand points 3: Randomly initialize $K$ clusters, $C = {η_{1}, η_{2}, . . ., η_{k}}$ according to uniform distribution 4: while $\| η_{k, t + 1} - η_{k, t} \| >$ do 5: for ${n = 1, . . ., N}$ do 6: Find center $η_{k} \in C$ closest to point $d_{n}$ 7: Assign demand point $d_{n}$ to center $η_{k}$ 8: for ${k = 1, . . ., K}$ do 9: Calculate median of all points in cluster $k$ 10: Assign new center $η_{k}$ as median

Algorithm 1 k-Medians Clustering for Vertiport Location

1: procedureK-MEDIANS(

D

)
2:

D = {d_{1}, d_{2}, . . ., d_{n}}

is the set of input demand points
3: Randomly initialize

K

clusters,

C = {η_{1}, η_{2}, . . ., η_{k}}

according to uniform distribution
4: while

| η_{k, t + 1} - η_{k, t} | >

do
5: for

{n = 1, . . ., N}

do
6: Find center

η_{k} \in C

closest to point

d_{n}

7: Assign demand point

d_{n}

to center

η_{k}

8: for

{k = 1, . . ., K}

do
9: Calculate median of all points in cluster

k

10: Assign new center

η_{k}

as median

The $k$ -medians algorithm calculates the L1 norm absolute (Manhattan) deviation between points on a linear scale ( 36 ). This contrasts with the squared deviations computed in a $k$ -means problem, which increases the cost of outliers on an exponential scale.

Land Use Constraints and Parcel Assignment

As previously discussed, land use and zoning requirements can limit vertiport placement ( 32 ). To plan a vertiport network, additional factors should be considered in addition to UAM demand. A holistic optimization must be executed to understand not only what is efficient, but also what is possible.

Using parcel level data from the San Francisco Bay Area, land use is constrained to parcels that are zoned office or mixed-use office space, light and heavy industrial (including warehouses), retail, and parking structures ( 2 , 12 , 37 ). These parcels are believed to have compatible zoning for vertiports and minimum requirements that allow for takeoff and landing operations at a feasible scale. The nearest vertiport parcels that satisfy these land use constraints are assigned using a fast BallTree nearest neighbor search. This technique recursively divides the data into nodes defined by a centroid to perform a single distance calculation for lower and upper bounds ( 38 ). The decision was made to exclude constrained parcels in the linear program for clarity of the formulation’s results, specifically that the parcels chosen are simply the closest in distance to the nodes that are output from the formulation; however, additional research could consider this different formulation. Additional enhancements could include the inclusion of available parcels and their associated land uses, and other constraints, such as noise, airspace, and infrastructure availability, in the linear program.

Additionally, the parcels that support larger vertiports are also considered. Cohen ( 39 ) suggests that the area needed for helicopter takeoffs and landings should be $\geq 14$ acres ( 39 ). However, literature on eVTOL aircraft suggests that vertiports may require a considerably smaller footprint of approximately 1.4 acres (or 60,000 ft²) ( 40 ). Because of the diversity of aircraft currently under development, the model considers parcels between 1.4 and 14 acres as potentially suitable for vertiport infrastructure ( 37 ).

Simulation

After potential eligible vertiport parcels are identified, the model simulates how the parcel locations may affect travel behavior. The nearest node to each vertiport parcel is used as a proxy point for later input into the MANTA traffic microsimulator; refer to Figure 2 for an example. Trips that have the same origin and destination vertiports are eliminated and assumed to be a driving trip (e.g., the trip is too short to be served by UAM). The subsequent analysis assumes that the remaining eligible trips have been assigned as UAM trips.

Figure 2.

The demand used in the original ground travel simulation run in MANTA is parsed into trips from (i) the origin to its nearest origin vertiport, (ii) the origin vertiport to the destination vertiport, and (iii) the destination vertiport to the destination. The travel time of the first leg from the origin to the nearest vertiport is computed with an initial run in MANTA. The travel time from vertiport to vertiport is calculated using the Haversine distance, which is the great-circle distance between two points on a sphere given their longitudes and latitudes. The aerial travel time is then calculated assuming a flight speed of 150 mph ( 2 ). The travel time from the destination vertiport to the final destination is then simulated in MANTA. To calculate an accurate departure time for those trips requires both the previous legs’ travel times and estimated transfer delays at the vertiports, as shown in Equation 2.

T_{d} = T_{o} + T_{o, o_{vert}} + T_{o_{vert}, d_{vert}} + 2 T_{vert}

(2)

where $T_{d}$ is the departure time at the destination vertiport for the final destination, $T_{o}$ is the departure time of the traveler from the origin to the origin vertiport, $T_{o, o_{vert}}$ is the travel time via automobile from the origin node to the origin vertiport, $T_{o_{vert}, d_{vert}}$ is the aerial travel time between vertiports, and $T_{vert}$ is the transfer delay at each vertiport. $T_{vert}$ is assumed to be 10 min at each vertiport, based on the maximum acceptable waiting times for transit, security, embarkation, and disembarkation ( 41 ).

Modeling Results

This section details the results from both the clustering and the simulation processes. The clustering process determines the location of the vertiports from the k-medians algorithm, and then maps that parcel to the nearest feasible parcel—and subsequently, node—based on the land use constraints. The regional-scale simulation uses the MANTA traffic microsimulator to calculate the final congested travel times of each driving and multimodal UAM trip.

Clustering

Because of the randomization of the initial points, several trials are run to determine the final convergence points for the $k$ -medians clustering algorithm. Convergence occurs between three and 15 iterations, depending on $k$ . At k = 10, the cluster points, in radians, are shown in Figure 3.

Figure 3.

When $k$ is low, the mean distance between the calculated cluster point and the nearest feasible vertiport parcel is high, but as $k$ increases, the mean distance of parcels to the initially calculated median points stabilizes between 15 and 23 km, as shown in Figure 4. This large distance suggests that feasible vertiport sites may be challenging to find, and thus limit the overall value of UAM altogether.

Figure 4.

While a greater number of vertiports may reach more potential users within the shortest distance from a vertiport, many of the initial cluster points found are similar but simply farther from usable parcels. This suggests that land use patterns may be similar across localities. Resulting vertiport parcels are shown in Figure 5 for k = 10 and k = 50.

Figure 5.

In Figure 5, k = 10 creates a sparse network while k = 50 creates a much denser network of vertiports. When $k$ increases, the vertiports appear in the major low-density suburbs adjacent to San Francisco Bay itself, often near a large thoroughfare or highway (e.g., CA-82 in the Peninsula, Interstate 880 in Oakland, and Interstate 280 in San Jose). Interestingly, San Francisco, with over 800,000 residents and a dense financial district, shows only two feasible parcels in the k = 50 design and only one feasible parcel in the k = 10 design. This suggests zoning constraints could have a notable impact on the potential number of UAM vertiports ( 42 ).

From these parcels and the calculated departure and travel times for UAM trips, the model simulates the UAM trips using MANTA.

Simulation

Several analytical assessments are made from the simulator outputs. The first explores average UAM travel times across $k$ compared with non-UAM driving trips. The mean travel times of UAM and non-UAM across $k$ vertiports are shown in Figure 7. UAM shows benefit over driving the same trip at $k \geq 40$ , but shows benefit across both modes together in the transportation network only when $k \geq 100$ .

The kernel density estimation of the time deltas between the UAM and non-UAM scenarios across $k$ are shown in Figure 6. In this plot, a negative number indicates that a UAM trip time is longer than the same trip time when driving. A positive number indicates that driving takes longer than UAM for the same trip. The mean delta in travel time for $k = 10$ is −32 min, $k = 20$ is −13 min, and $k = 30$ is −7 min, showing that at low $k$ , the travel times from origin to destination are significantly faster when driving compared with using UAM. However, the mean delta in travel time becomes positive at $k \geq 40$ , suggesting there must be at least 40 vertiports for the average UAM trip to be faster than the average driving trip. A significant decrease in UAM travel time—nearly 20 min—occurs from $k = 10$ to $k = 20$ and from $k = 60$ to $k = 70$ vertiports. At $k \geq 70$ , the rate of change in travel time begins to plateau again, suggesting that $k = 70$ may be a good compromise for the number of vertiports in the San Francisco Bay Area to provide travel time benefit. $k = 100$ vertiports is an inflection point in which the mean travel time for trips with both UAM and driving becomes lower than the mean travel time for only driving across all trips.

Figure 6.

Figure 7.

Even using Marchetti’s Constant, not all trips would logically be assigned to UAM. Figure 8, where $k = 10$ , shows an example of a trip where UAM travel time is longer than driving. In this figure, the vertiports (red and light blue, respectively) are out of the way from the origin and destination (peach and green, respectively), increasing the total travel distance and time.

Figure 8.

On the other hand, an example of a UAM trip with a travel time shorter than driving is shown in Figure 9, where $k = 70$ . In this figure, the vertiports are along the route of a vehicle trip, reducing the travel time that would otherwise be required to deviate to the nearest vertiport.

Figure 9.

The second assessment is to understand how area constraints of the parcels affect vertiport locations and subsequent travel times. Given $k = 70$ , Figure 10 shows the mean travel times of UAM compared with the non-UAM scenario when adjusting for area of the parcel needed, $A$ , from 2 to 14 acres by increments of two. Intuitively, the mean travel times with UAM increase as the area constraints increase, confirming that land parcels of large areas such as 14 acres are less accessible and less prevalent. Only if the acreage constraint is below 2 acres does the mean travel time of UAM remain competitive with driving. This result suggests that a larger network of smaller vertiports makes more sense than a smaller network of larger facilities, as described in Table 1. It may be necessary to build vertically or create seamlessly multiplexed FATOs and TOLAs in small spaces. With the expectation that UAM would use VTOL instead of STOL aircraft, parcel areas of $A \leq$ 2 acres are feasible if vertiports are designed efficiently ( 43 , 44 ).

Figure 10.

Limitations and Future Work

Several limitations exist in the model. The first is the aerial travel time assumption, which currently is computed by a Haversine distance metric and a static assumption of aircraft speed at 150 mph. As of May 2021, there were over 200 aircraft under development, and the average speeds of these aircraft are unknown. The precise times of each component of the flight, such as takeoff, cruise, and landing are also not considered ( 45 ). In addition, precise flight trajectories that consider collision avoidance and weather are not incorporated. Ongoing work includes the integration of accurate uncrewed aerial simulation ( 46 , 47 ).

Another limitation is the assumption of fixed 10 min transfer times at the vertiports. In actuality, transfer times will likely vary based on the size and location of a vertiport. For example, it may take passengers longer to board and disembark at larger and more congested facilities. Because this could affect the comparative travel times between UAM and other modes of transportation, methodological advancements should attempt to estimate the time of the boarding, disembarkation, and transfer process more accurately based on the number of flights and passenger volumes at proposed vertiports ( 9 ).

The third limitation is the assumption of Marchetti’s Constant, which may not always accurately reflect actual traveler behavior ( 48 ). Efforts to replace Marchetti’s Constant involve UAM activity-based specifications to reflect more accurate individual behavioral patterns. This will allow for more accurate multimodal mode shares, as the model currently compares UAM with driving rather than with all modes in the transportation network (e.g., public transit, biking, transportation networking companies, etc.). Several studies have already looked into this ( 9 , 26 , 49 ).

Another limitation is that the travel time analysis does not assume dynamic traffic assignment, where vehicles may change their routes in the middle of the trip or because of more recent congestion before embarking on the trip. This is currently an ongoing improvement in MANTA.

An important consideration for future work is to generalize the methodology for other metropolitan areas. The San Francisco Bay Area has a strong central business district in San Francisco as well as several other large economic centers throughout the region (e.g., San Jose/Silicon Valley, Oakland, Tri-Valley, etc.). These economic centers contribute to several region-specific multidirectional travel patterns. As a result, the results of this study could be applied to metropolitan areas with both one large center (i.e., monocentric regions) and many multiple centers (i.e., polycentric regions) ( 50 , 51 ). The San Francisco Bay Area also has several air traffic control and infrastructure restrictions, which were not considered in this study but could be analyzed as part of future research.

Additional work should also consider the cost of UAM which can affect mode choice ( 26 , 52 , 53 ). Currently, the model may overestimate demand if UAM is more expensive than driving. Although more complex models are needed to compare UAM mode share alongside not only privately-owned vehicles but shared vehicle fleets (e.g., transportation network companies and automated taxis), the current model provides a methodological framework for commencing exploratory analysis.

Conclusion

While federal preemption currently governs most aspects of UAM regulation, establishing new UAM infrastructure is one area where local and regional governments will play an important role. While local governments cannot limit where UAM operates in the airspace, they can effectively limit where UAM aircraft take off and land through the zoning and siting of vertiports. As such, land use planning, zoning, and regulation represent important tools that local governments can use to either support or hinder the growth of UAM.

Using a case study of the San Francisco Bay Area, this paper presents a vertiport design and network simulation tool to understand the travel time effects of UAM vertiport networks in the San Francisco Bay Area under specific land use constraints. Given demand data at the zone level (further granularized to the node level) and extensive land use data at the parcel level, different UAM networks are developed with a varying number of vertiports and parcel constraints. These different configurations are developed using k-medians clustering and graph theoretic nearest neighbors. The vertiport locations are then integrated into the fast, parallelized GPU-based MANTA traffic simulator to understand travel times of UAM compared with driving. At $k < 40$ vertiports at feasible locations, the same trips using UAM compared with driving have considerably longer travel times, but at $k \geq 40$ , the benefit of UAM in travel time emerges. $k = 100$ vertiports is an inflection point at which the mean travel time for trips by both UAM and driving becomes lower than the mean travel time for only driving across all trips. In addition, as the area required for each parcel increases, travel times increase, implying that greater focus must be placed on efficient vertiport designs in smaller parcel areas. Although future research should incorporate additional modal connections (i.e., shared mobility and active transportation) to vertiports, intuitively a larger network of smaller vertiports could help reduce first- and last-mile travel distances and encourage non-vehicular connections, such as walking and micromobility.

While quantitative methods are an effective tool for assessing the technical viability of vertiport concepts, community engagement and local policy will be important to help understand and address potential community concerns, such as noise, aesthetics, traffic congestion, equity (i.e., displacement and gentrification), safety, and others. Community engagement will be an essential part of understanding and mitigating the adverse impacts of vertiports on vulnerable populations and communities. As part of the vertiport planning process, communities may consider embedding public participation and engagement at multiple stages. As part of long-range planning, public engagement could help communities understand if UAM is an appropriate strategy. Communities could conduct exploratory scoping to understand potential UAM demand and locations that could be compatible for vertiports. As the planning process progresses, communities may pivot toward more in-depth engagement processes that empower stakeholders to understand local concerns and use a variety of techniques to vet vertiport concepts and alternatives. If or when a developer decides to build a vertiport, additional community engagement is recommended and may be statutorily required by local, state/provincial, and federal laws.

Given the potential for UAM to grow over the next 10 to 20 years, communities will need to consider multimodal integration, surface transportation access, and the potential impacts of UAM. In the future, a network of vertiports has the potential to decentralize the impacts of aviation that have typically been limited to airport facilities and their immediate surroundings. Long-range planning and community engagement will likely require close collaboration between aviation and community stakeholders. Understanding the role of the built environment and its relationship to vertiport placement may also be important to integrating with nearby land uses and prioritizing sustainable mobility. As the UAM marketplace continues to evolve, community planning, public outreach, and continued research will be important to understanding UAM’s impacts on the transportation ecosystem and to balance community goals with commercial interests.

Footnotes

Acknowledgements

The authors would like to thank Dr Paul Waddell and Max Gardner of the Urban Analytics Lab at University of California, Berkeley for their assistance with advising and data procurement, respectively. The authors would also like to thank Dr Raja Sengupta and Emin Burak Onat of the Cal Unmanned Lab at University of California, Berkeley for their support. Finally, the authors would like to thank NASA and the NASA market study strategic advisory group for their support of this work.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: P. Yedavalli, A. Cohen; data collection: P. Yedavalli; analysis and interpretation of results: P. Yedavalli; draft manuscript preparation: P. Yedavalli, A. Cohen. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Pavan Yedavalli

References

Lawson

Flight Path: Taking to the Skies to Solve Congestion. Traﬃc Technology International, 2007, pp. 52–53.

Uber Elevate. Fast-Forwarding to a Future of On-Demand Urban Air Transportation. Urban Air Transportation, 2016.

Plaut

P. O.

Non-Motorized Commuting in the US. Transportation Research Part D: Transport and Environment, Vol. 10, 2005, pp. 347–356.

Balac

Rothfeld

R. L.

Hörl

The Prospects of On-Demand Urban Air Mobility in Zurich, Switzerland. Proc., IEEE Intelligent Transportation Systems Conference (ITSC), Auckland, New Zealand, IEEE, New York, 2019, pp. 906–913.

Cohen

A. P.

Shaheen

S. A.

Farrar

E. M.

Urban Air Mobility: History, Ecosystem, Market Potential, and Challenges. IEEE Transactions on Intelligent Transportation Systems, Vol. 22, No. 9, 2021, pp. 6074–6087.

Cohen

Shaheen

Urban Air Mobility: Opportunities and Obstacles. International Encyclopedia of Transportation, 2021, pp. 702–709.

Holmes

B. J.

A Vision and Opportunity for Transformation of On-Demand Air Mobility. Proc., 16th AIAA Aviation Technology, Integration, and Operations Conference, Washington, D.C., American Institute of Aeronautics and Astronautics, Reston, VA, 2016.

Mueller

E. R.

Kopardekar

P. H.

Goodrich

K. H.

Enabling Airspace Integration for High-Density On-Demand Mobility Operations. Proc., 17th AIAA Aviation Technology, Integration, and Operations Conference, Denver, Colorado, American Institute of Aeronautics and Astronautics, Reston, VA, 2017.

Rothfeld

Balać

Antoniou

Potential Urban Air Mobility Travel Time Savings: An Exploratory Analysis of Munich, Paris, and San Francisco. Sustainability, Vol. 13, 2021, pp. 1–20.

10.

Goyal

Reiche

Fernando

Cohen

Advanced Air Mobility: Demand Analysis and Market Potential of the Airport Shuttle and Air Taxi Markets. Sustainability, Vol. 13, 2021, p. 7421.

11.

Yedavalli

Mooberry

2019An Assessment of Public Perception of Urban Air Mobility. https://pdfs.semanticscholar.org/f464/1304ae7082a61d9cd82905917ce694a120c4.pdf?_ga=2.146171191.2046738072.1580045281-1681120550.1580045281. Accessed January 23, 2020.

12.

Moore

M. D.

Misconceptions of Electric Aircraft and Their Emerging Aviation Markets. Proc., 52nd Aerospace Sciences Meeting, National Harbor, Maryland, AIAA SciTech Forum, American Institute of Aeronautics and Astronautics, Reston, VA, 2014.

13.

Johnston

Riedel

Sahdev

Urban Air Mobility Market Study. American Institute of Aeronautics and Astronautics, Reston, VA, 2018.

14.

Lineberger

Hussain

Rutgers

Hanley

The Elevated Future of Mobility. 2019.

15.

Balac

Vetrella

A. R.

Axhausen

K. W.

Towards the Integration of Aerial Transportation in Urban Settings. TRB Annual Meeting Online, Article No. 18-02015. Transportation Research Board, Washington, D.C., 2018.

16.

Fadhil

D. N.

Moeckel

Rothfeld

GIS-Based Infrastructure Requirement Analysis for an Electric Vertical Take-Off and Landing Vehicle-Based Transportation System. Transportation Research Procedia, Vol. 41, 2019, pp. 101–103.

17.

Rothfeld

Balac

Ploetner

K. O.

Antoniou

Initial Analysis of Urban Air Mobility’s Transport Performance in Sioux Falls. Proc., Aviation Technology, Integration, and Operations Conference, Atlanta, Georgia, American Institute of Aeronautics and Astronautics, Reston, VA, 2018.

18.

Rothfeld

Balac

Ploetner

K. O.

Antoniou

Agent-based Simulation of Urban Air Mobility. Proc., Modeling and Simulation Technologies Conference, Atlanta, Georgia, American Institute of Aeronautics and Astronautics, Reston, VA, 2018.

19.

Daskilewicz

German

Warren

Garrow

L. A.

Boddupalli

S.-S.

Douthat

T. H.

Progress in Vertiport Placement and Estimating Aircraft Range Requirements for eVTOL Daily Commuting. Proc., Aviation Technology, Integration, and Operations Conference, Atlanta, Georgia, American Institute of Aeronautics and Astronautics, Reston, VA, 2018.

20.

Bulusu

Onat

E. B.

Yedavalli

Sengupta

Macfarlane

A Traffic Demand Analysis Method for Urban Air Mobility. IEEE Intelligent Transportation Systems Magazine, Vol. 22, No. 9, 2020, pp. 6039–6047.

21.

Gentile

Meschini

Papola

Spillback Congestion in Dynamic Traffic Assignment: A Macroscopic Flow Model with Time-Varying Bottlenecks. Transportation Research Part B: Methodological, Vol. 41, 2007, pp. 1114–1138.

22.

Lim

Hwang

The Selection of Vertiport Location for On-Demand Mobility and its Application to Seoul Metro Area. International Journal of Aeronautical and Space Sciences, Vol. 20, 2019, pp. 260–272.

23.

Syed

Rye

Ade

Trani

Hinze

Swingle

Smith

J. C.

Dollyhigh

Marien

Preliminary Considerations for ODM Air Traffic Management Based on Analysis of Commuter Passenger Demand and Travel Patterns for the Silicon Valley Region of California. Proc., 17th AIAA Aviation Technology, Integration, and Operations Conference, Denver, Colorado, American Institute of Aeronautics and Astronautics, Reston, VA, 2017.

24.

Shamiyeh

Rothfeld

Hornung

A Performance Benchmark of Recent Personal Air Vehicle Concepts. Proc., 31st Congress of the International Council of the Aeronautical Sciences, ICAS, Belo Horizonte, Brazil, 2018.

25.

Robinson

J. N.

Sokollek

M.-D. R.

Justin

C. Y.

Mavris

D. N.

Development of a Methodology for Parametric Analysis of STOL Airpark Geo-Density. Proc., Aviation Technology, Integration, and Operations Conference, Atlanta, Georgia, American Institute of Aeronautics and Astronautics, Reston, VA, 2018.

26.

Zhang

Network Design, Performance Analysis, and Potential Demand Exploration of eVTOL On-Demand Service for Urban Air Mobility. International Conference on Research in Air Transportation, Vol. 10, 2020.

27.

Antcliﬀ

K. R.

Moore

M. D.

Goodrich

K. H.

Silicon Valley as an Early Adopter for On-Demand Civil VTOL Operations. Proc., 16th AIAA Aviation Technology, Integration, and Operations Conference, Washington, D.C., AIAA Aviation Forum, American Institute of Aeronautics and Astronautics, Reston, VA, 2016.

28.

Skabardonis

Varaiya

Petty

K. F.

Measuring Recurrent and Nonrecurrent Traﬃc Congestion. Transportation Research Record: Journal of the Transportation Research Board, 2003. 1856: 118–124.

29.

Yedavalli

Kumar

Waddell

Microsimulation Analysis for Network Traﬃc Assignment (MANTA) at Metropolitan-Scale for Agile Transportation Planning. Transportmetrica A: Transport Science, 2021, pp. 1–22. https://doi.org/10.1080/23249935.2021.1936281.

30.

Boeing

OSMnx: New Methods for Acquiring, Constructing, Analyzing, and Visualizing Complex Street Networks. Computers, Environment and Urban Systems, Vol. 65, 2017, pp. 126–139.

31.

Metropolitan Transportation Commission and Association of Bay Area Governments. Plan Bay Area 2040. Bay Area Metro Center, San Francisco, CA, 2017.

32.

Waddell

UrbanSim Modeling Urban Development for Land Use, Transportation, and Environmental Planning. Journal of the American Planning Association, Vol. 68, No. 3, 2002, pp. 297–314.

33.

Marchetti

Anthropological Invariants in Travel Behavior. Technological Forecasting and Social Change, Vol. 47, 1994, pp. 75–88.

34.

Barbosa

Barthelemy

Ghoshal

James

C. R.

Lenormand

Louail

Menezes

, et al. Human Mobility: Models and Applications. Physics Reports, Vol. 734, 2018, pp. 1–74.

35.

Dohan

Karp

Matejek

K-median Algorithms: Theory in Practice. Working Paper, Princeton, Computer Science, Princeton University, NJ, 2015. https://www.cs.princeton.edu/courses/archive/fall14/cos521/projects/kmedian.pdf

36.

Dasgupta

Frost

Moshkovitz

Rashtchian

Explainable k-Means and k-Medians Clustering. arXiv Preprint arXiv:2002.12538, 2020.

37.

Vascik

P. D.

Hansman

R. J.

Evaluation of Key Operational Constraints Aﬀecting On-Demand Mobility for Aviation in the Los Angeles Basin: Ground Infrastructure, Air Traﬃc Control and Noise. American Institute of Aeronautics and Astronautics, Reston, VA, 2017.

38.

Omohundro

S. M.

Five Balltree Construction Algorithms. Technical Report TR-89-063. ICSI, Berkeley, CA, 1989.

39.

Cohen

M. M.

The Vertiport as an Urban Design Problem. World Aviation Congress & Exposition, 1996. https://doi.org/10.4271/965523.

40.

Federal Aviation Administration. AC 150/5390-2C - Heliport Design. Federal Aviation Administration, Washington, D.C., 2012.

41.

Arhin

Predicting Acceptable Wait Times for Patrons at Transit Bus Stops by Time of Day. Technical Report. San Jose State University, CA, 2018.

42.

Osman

Land Use Regulations and the Dispersion of the IT Industry in the San Francisco Bay Area. Papers in Regional Science, Vol. 99, No. 3, 2020. https://doi.org/10.1111/pirs.12532.

43.

Thipphavong

D. P.

Apaza

Barmore

Battiste

Burian

Dao

Feary

, et al. Urban Air Mobility Airspace Integration Concepts and Considerations. Proc., Aviation Technology, Integration, and Operations Conference, Atlanta, Georgia, American Institute of Aeronautics and Astronautics, Reston, VA, 2018.

44.

Egorov

Kochenderfer

M. J.

Analysis of Fleet Management and Infrastructure Constraints in On-Demand Urban Air Mobility Operations. Proc., AIAA Aviation 2020 Forum (Virtual Event), American Institute of Aeronautics and Astronautics, Reston, VA, 2020.

45.

Kasliwal

Furbush

N. J.

Gawron

J. H.

McBride

J. R.

Wallington

T. J.

De Kleine

R. D.

Kim

H. C.

Keoleian

G. A.

Role of Flying Cars in Sustainable Mobility. Nature Communications, Vol. 10, 2019, pp. 1–9.

46.

Xue

Rios

Silva

Zhu

Ishihara

A. K.

Fe³: An Evaluation Tool for Low-Altitude Air Traﬃc Operations. Proc., Aviation Technology, Integration, and Operations Conference, Atlanta, Georgia, American Institute of Aeronautics and Astronautics, Reston, VA, 2018.

47.

Reiche

Cohen

A. P.

Fernando

An Initial Assessment of the Potential Weather Barriers of Urban Air Mobility. IEEE Transactions on Intelligent Transportation Systems, Vol. 22, 2021, pp. 6018–6027.

48.

Cervero

Going Beyond Travel-Time Savings. Technical Report No. 70206. The World Bank, Washington, D.C., 2011.

49.

Exploring Preferences for Transportation Modes in an Urban Air Mobility Environment: A Munich Case Study. Master’s thesis. Technical University of Munich, 2018.

50.

Tsai

Y.-H.

Quantifying Urban Form: Compactness versus ‘Sprawl’. Urban Studies, Vol. 42, 2005, pp. 141–161.

51.

Kloosterman

R. C.

Musterd

The Polycentric Urban Region: Towards a Research Agenda. Urban Studies, Vol. 4, 2001, pp. 623–633.

52.

Garrow

L. A.

Ilbeigi

Chen

Forecasting Demand for On Demand Mobility. Proc., 17th AIAA Aviation Technology, Integration, and Operations Conference, Colorado, AIAA Aviation Forum, American Institute of Aeronautics and Astronautics, Reston, VA, 2017.

53.

Rothfeld

Antoniou

Exploring Preferences for Transportation Modes in an Urban Air Mobility Environment: Munich Case Study. Transportation Research Record: Journal of the Transportation Research Board, 2019. 2673: 427–442.

Planning Land Use Constrained Networks of Urban Air Mobility Infrastructure in the San Francisco Bay Area

Abstract

Keywords

Vertiport Siting and Network Design

Methodology

Microsimulator

Network

Demand

Network Optimization

Land Use Constraints and Parcel Assignment

Simulation

Modeling Results

Clustering

Simulation

Limitations and Future Work

Conclusion

Footnotes

Acknowledgements

Author Contributions

Declaration of Conflicting Interests

Funding

ORCID iD

References