Land Use Policy 91 (2020) 103963

Contents lists available at ScienceDirect

Land Use Policy

journal homepage: www.elsevier.com/locate/landusepol

Parking futures: Shared automated vehicles and parking demand reduction T

trajectories in Atlanta
Wenwen Zhang , Kaidi Wang
⁎

Department of Urban Affairs and Planning, Virginia Polytechnic Institute and State University, RM 215, 140 Otey Street, Architecture Annex, Blacksburg, VA, USA

ARTICLE INFO ABSTRACT

Keywords: The coming of automated vehicles (AVs) and Mobility-on-Demand (MoD services) is expected to reduce urban
Parking land use parking demand and correspondingly alter the urban parking landscape in a significant way. Multiple modeling
Automated Vehicles (AVs) efforts have already demonstrated that Shared AVs (SAVs) have promising potential to decrease urban parking
Mobility-on-demand (MoD) demand. However, previous studies have only examined SAV parking demand at one point in time, with various
Parking policy
market penetrations. It remains unclear what the demand reduction trajectory will be like during the transition
period when there is a mix of SAVs, Privately-Owned AVs (PAVs), Shared Conventional Vehicles (SCVs), and
Conventional Private Vehicles (CPVs). This study fills this gap by developing an agent-based simulation model to
examine the spatially and temporally explicit parking reduction trends with mixed travel modes from
2020–2040. The results indicate that in the most optimal AV and MoD adoption scenario, the parking demand
will decrease by over 20% after 2030, especially in core urban areas. Meanwhile, the parking demand in re-
sidential zones may double, which could lead to transportation equity concerns. Additionally, parking relocation
may also induce environmental issues by generating a considerable amount of empty Vehicle Miles Traveled
(VMT). To reap the benefits brought by AVs and MoD systems and to mitigate the accompanying social and
environmental issues, our results suggest that proactive policymakers in the next decade will need to modify land
use regulations for both new developments and existing parking infrastructure in commercial and residential
zones, as well as update travel demand management policies.

1. Introduction private personal vehicle ownership through the sharing economy. On-
demand car-sharing services, such as Uber and Lyft, enabled by
In U.S. cities, an average car is parked 95% of the time, resulting in smartphone and GPS technology, have become a popular travel mode
a large parking inventory and numerous parking-related issues (Shoup, choice in some settings (Giovanni et al., 2018). Service data provided
2005). Cities need to provide approximately four parking spaces (or by Transportation Network Companies (TNCs), such as Uber in the U.S.
even eight spaces in some extreme cases) for each private passenger car and Didi in China, suggest that the market share of ride-hailing services
to meet the parking demand (Chester et al., 2010). It is estimated that has grown exponentially in the past few years (Segarra, 2017; Xinhua,
there are about 500 million parking spaces in the U.S., covering a total 2018). MoD services are expected to significantly reduce the public
of 3,590 square miles (Ben-Joseph, 2012). These parking spaces cost parking demand or even make structured parking obsolete by im-
cities millions for construction, operation, and maintenance. Even with proving vehicle utilization rates.
a surplus in parking infrastructure, cruising for parking remains the AVs will also change parking demand, but Shared AVs (SAVs) and
most stressful part of trips and contributes to a series of environmental Privately-owned AVs (PAVs) will play different roles. Several simula-
and economic problems in urban areas (Shoup, 2005). tion endeavors have already suggested that SAVs can dramatically curb
The emergence of disruptive transportation technologies, such as the demand for parking space by reducing vehicle ownership
Mobility-on-Demand (MoD) and Automated Vehicles (AVs) are ex- (International Transport Forum, 2015; Kondor et al., 2018; Zhang and
pected to alleviate urban parking issues. MoD refers to the integration Guhathakurta, 2017; Zhang et al., 2015a). Compared to the existing
of multiple transportation options into one accessible, on-demand shared mobility, SAVs are expected to be more effective in reducing
system that provides door-to-door mobility service. It is an appealing parking demand, as they can be more affordable and efficient under
transportation concept that is envisioned to provide an alternative to centralized operation (Fagnant and Kockelman, 2014). PAVs, on the

⁎
Corresponding author.
E-mail addresses: [email protected] (W. Zhang), [email protected] (K. Wang).

https://doi.org/10.1016/j.landusepol.2019.04.024
Received 24 September 2018; Received in revised form 14 January 2019; Accepted 15 April 2019
0264-8377/ © 2019 Elsevier Ltd. All rights reserved.
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

other hand, may only be able to reduce parking spaces marginally, by a Singapore shows that over 85% of parking spaces may no longer be
mere 10% (Zhang et al., 2018). However, PAVs will alter the spatial needed if all private vehicles are replaced by SAVs (Kondor et al.,
layout of parking spaces in cities, as these vehicles can automatically 2018). The parking reduction rate may increase to 92% if the travelers
cruise to more affordable parking spaces after dropping owners off at are not only willing to share vehicles but also share rides (i.e., multiple
trip destinations. passengers served by one SAV at the same time). Wang et al., 2018a
The majority of studies have examined the impact of SAVs on developed a microscopic simulation to determine the impact of curb-
parking at one point in time (with various market penetration sce- side parking bay design on high-occupancy SAVs, i.e., SAVs with more
narios), although it remains unclear what the demand reduction tra- than four seats. The study shows that the provision of sufficient curb-
jectory will be like during the transition period when there will be a mix side bay areas for SAV loading and unloading can have a positive im-
of PAVs, SAVs, Shared Conventional Vehicles (SCVs), and Private pact on reducing operating costs and average waiting time, thus im-
Conventional Vehicles (PCVs). In other words, there is limited in- proving client experiences.
formation on how quickly the demand for parking space will decrease Compared with SAVs, PAVs may not be able to reduce the parking
during the transition period, how much change in the spatial pattern of demand significantly due to their marginal impact on vehicle utilization
parking should be expected, and how much time is left for planners to rate and vehicle ownership reduction. Zhang et al. (2018) developed a
develop strategies to adjust for possible changes. Such knowledge is greedy algorithm and Mixed Integer Programming (MIP) model to ex-
salient for policymakers to proactively adapt transportation and land plore the impact of PAVs on regional vehicle inventory using data from
use policies. Motivated to narrow the research gaps, this study ex- the Atlanta 10-county metropolitan area. The results show that only
amines spatially explicit parking demand reduction trajectories during 18% of households can reduce vehicle ownership by sharing PAVs
the transition period. among household members. The regional household vehicle ownership
may only be reduced by 10% if households switch from Conventional
2. Existing AV parking models Vehicles (CVs) to PAVs. In other words, PAVs may not be able to reduce
the total parking demand in cities. However, despite PAVs’ limited
This study is an extension of recent endeavors to model the impacts contribution to parking space reduction, this new travel option may
of MoD and AVs on urban parking demand. SAVs can reduce the de- fundamentally change the spatial configuration of parking spaces, as
mand for parking lots by reducing vehicle ownership and improving PAVs can relocate from the trip destination to park somewhere else to
vehicle utilization rate. Multiple studies have already demonstrated save parking costs. Wang et al., 2018b developed a linear programming
that SAVs may significantly reduce the required parking space in the optimization model that optimizes the spatial configuration of parking
city using agent-based simulation models. Zhang et al. (2015) devel- space for PAVs, with the objective of minimizing construction, main-
oped a time-step-based simulation to examine the parking demand of an tenance, and opportunity costs of city-wide parking lots (i.e., the tax
SAV system under various vehicle cruising strategies and fleet size revenue could be generated, if parking spaces were used for commercial
scenarios. The study utilized a hypothetical grid-based network and uses). The model results indicate that the demand for parking in the
national average travel demand patterns. The results suggest that SAVs City of Atlanta may concentrate in Traffic Analysis Zones (TAZs) with
may reduce approximately 90% of the parking space. The parking less intensive economic activities, so that land in business districts can
footprint is negatively correlated with vehicle cruising time and linearly be utilized more efficiently.
and positively associated with the SAV fleet size. However, the SAV However, the future urban mobility will not be dominated by one
system also tends to generate excessive Vehicle Miles Traveled (VMT), a mode of transportation, but rather by a combination of PAVs, SAVs,
measurement widely used in transportation planning for a variety of SCVs, and PCVs, as multiple opinion surveys show a mix of preferences
purposes, especially to assess the amount of travel on the road. In other (Haboucha et al., 2017; Krueger et al., 2016; Lavieri et al., 2017;
words, the SAV system reduces the parking demand at the expense of Megens et al., 2015). We assume that the road will initially be domi-
generating more traffic. nated by PCVs. With the introduction of MoD and AVs, cities may
Simulations using real-world travel demand, network data, and witness an increase in SCVs, SAVs, and even PAVs in the short-term.
various market share assumptions also suggest that SAVs can reduce Eventually, conventional vehicles will be primarily replaced by AVs and
parking demand by over 80%. International Transport Forum (2015) cities will have to regulate the operation of SAVs and PAVs. This study
developed an agent-based simulation to address the interaction be- aims to develop an agent-based model to examine (1) the longitudinal
tween SAVs and travelers with or without high-capacity public trans- changing trajectories of parking demand stemming from vehicle au-
portation in Lisbon, Portugal. In all examined scenarios, the results tonomy and MoD, and (2) possible adaptations to existing land use and
show that SAVs can remove all on-street parking and over 80% of the transportation policies to accommodate these changes and help cities
off-street parking, freeing up parking spaces equivalent to 210 football achieve long-term traffic easement, economic development, and en-
fields in Lisbon. Zhang and Guhathakurta (2017) examined the spatial vironmental preservation goals.
distribution of SAV parking demand in the City of Atlanta using a dis- The rest of the paper is organized as follows: Section Three describes
crete-event, agent-based simulation model. Their results suggest that the methods used to develop market penetration scenarios, design the
parking demand can be reduced by over 90% for households, that are agent-based simulation, and estimate the spatial and temporal dis-
willing to give up private vehicles and use SAVs. However, the spatial tributions of parking demand. Section Four summarizes the model
location of parking demand depends heavily on the policies de- implementation data source and model environment settings. Section
termining the maximum allowed parking cruising time and the cost of Five analyzes and discusses the simulated parking demand and sensi-
parking. For example, SAV parking tends to concentrate in low-income tivity test results, followed by land use and transportation policy in-
neighborhoods when parking becomes expensive, which may lead to dications in Section Six. The last section concludes the primary findings
some transportation equity issues in the era of SAVs. and highlights possible future research directions.
Another recent large-scale parking demand modeling effort in

2
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

3. Methods Table 1
Forecasted Market Penetration of PAVs, SAVs, and CVs by Year.
In this study, we first developed market penetration scenarios based Mode Type Year
on existing AV and MoD forecast studies. We then constructed an agent-
based parking demand simulation model based on the existing SAV 2015 2020 2030 2040
simulation model developed by Zhang and Guhathakurta (2017). The
PAVs 0.0% 4.95% 23.7% 37.35%
model was implemented using data from the Atlanta Metropolitan area. SAVs 0.0% 0.55% 10.1% 37.35%
The detailed methods are described in the following sections. SCVs 0.0% 9.45% 19.9% 12.65%
PCVs 100.0% 85.05% 46.3% 12.65%
3.1. Market penetration development

We first adopted an overall AV market penetration level based on a 5.5%, 33.8%, and 74.7% for the years 2020, 2030, and 2040, respec-
literature review. We then estimated the share of SAVs and SCVs ac- tively.
cording to studies on market penetration levels of shared mobility/
MoD. Eventually, building upon the two streams of literature, we pro- 3.1.2. Market penetration of MoD
posed a market penetration scenario for different travel modes from the A review of the literature in Milakis, Van Arem, and Van Wee
year 2020–2040. (2017) shows that a wealth of studies have examined the secondary
impacts of SAVs based on various market penetration and regulation
3.1.1. Market penetration of AVs intervention assumptions. However, only a few studies to date have
Many efforts have been made toward forecasting the long-term discussed the market penetration of MoD in the future. Deloitte, a
adoption trajectories of AVs. The forecast results vary significantly due consulting firm, predicts that MoD will capture 10%, 30%, and 50% of
to differences in underlying prediction methods, model assumptions, the total market share in cities by the years 2030, 2035, and 2040,
and simplifications. Some studies have relied on projections based on respectively (Deloitte, 2017). The forecast is generated based on
the historical trends associated with other vehicle-related technologies, growing user data from car-sharing programs, the taxi industry, and
such as automatic transmission and hybrid vehicles (Litman, 2014). TNCs. A similar forecast scenario is also reported in a Morgan Stanley
Whereas other studies have modeled market penetration based on blue report (The Economist, 2016).
consumer’s willingness to pay for new technology, the add-on price of Based on the aforementioned study results and the assumption that
AVs, and the regulatory environment (Bansal and Kockelman, 2017). the adoption of AVs and MoD are independent, we estimated the
Some commercial reports forecast adoption trends without detailing the market share of different modes by multiplying the corresponding
projection methodologies (Deloitte, 2017). In short, the forecasted market shares together. For instance, the market share of SAVs in 2040
market penetration of AVs can vary significantly from as low as 20% to is estimated as the multiplication of the market share of AVs and the
over 90% for the year 2040. market share of MoD in 2040. The simulation model adopted the fol-
To the authors’ knowledge, only one study to date has accounted for lowing market penetration levels of PAVs, SAVs, SCVs, and PCVs, as
uncertainties from both the technology supply and demand sides to shown in Table 1.
develop projections. Bansal and Kockelman (2017) developed a simu-
lation-based framework to forecast the long-term adoption of AVs to-
gether with other vehicle technologies using data obtained from a na- 3.2. Urban parking demand simulation model
tional opinion survey. The simulation includes vehicle transaction and
technology adoption modules. When a simulated household decides to This section presents the design of an agent-based model to examine
replace a household vehicle (i.e., modeled via transaction module), the the parking demand with a mix of travel modes in the transportation
household will then choose whether to replace the old conventional system. The model consists of two major parts: 1) travel demand/ac-
vehicle with a level 4 A V or a level 1–3 AV (i.e., modeled via tech- tivity generation, and 2) parking demand agent-based simulation.
nology adoption module). The technology adoption module includes
factors from both the supply side, such as AV prices and development of 3.2.1. Travel demand generation module
regulations, and the demand side, such as willingness to pay and Using a Monte-Carlo simulation method, we randomly assigned
household demographic and socio-economic information. The simula- vehicle trips to different travel modes based on the adopted market
tion was run from 2015 to 2045. For the year 2045, the adoption rate of share of PAVs, SAVs, SCVs, and PCVs. The heterogeneity of traveler
level 4 A Vs (fully automated vehicles) ranged from 25% to 87% across preference for different types of modes was not modeled in this study,
eight scenarios. For our study, we used the most optimistic scenario due to a lack of household- and individual-level data. It was also as-
from the Bansal and Kockelman (2017) simulation to obtain more ob- sumed that SAVs and SCVs will only serve trips within the study area.
servable variations in parking demand across different simulation For trips that both start and end inside of the study area, we first ran-
years. The adopted scenario contains the following assumptions: domly sampled trips to be served by SAVs or SCVs. The rest of the
unselected trips were then randomly and proportionally classified to be
• Consumers are willing to pay an extra 10% of the current vehicle fulfilled by PAVs and PCVs. For the pseudo code of the travel demand
price; generation module as shown below in Algorithm 1, let:
• A 10% decline in technology price (over time?); ti, j be trips with origin i and destination j;
• NHTSA’s current and possible technology adoptions (e.g., adoption ti, j. mode be the travel mode for ti, j (with a default value of None);
of electronic stability control (ESC) from year 2015 and connectivity be the total number of trips;
from year 2020 on all new vehicles) are enacted. A be a set of nodes within the city of Atlanta;
MS{mode} be the assumed market share of different modes.
This scenario suggests that adoption rates of level 4 A Vs will be

3
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

3.2.2. Parking demand agent-based simulation Compared with PAVs and PCVs, the incorporation of SCVs is much
The parking demand agent-based simulation was developed by ex- more straightforward, as they function similarly to SAVs. However,
tending the existing SAV simulation model (Zhang and Guhathakurta, these two fleets will operate under separate central dispatching sys-
2017). The SAV model examines the parking demand for SAV fleets tems, indicating that SCVs will not be used to serve travelers that
only. We enhanced the model by incorporating the simulation of PAVs, choose SAVs, and vice versa. Additionally, the mile-based operation
SCVs, and PCV trips into the model. Before we describe the model ex- costs for SCVs are set to amounts similar to those for Uber and Lyft,
tension methods, we first briefly explain the logic of the original model costs which are significantly higher than for SAVs. It is assumed that the
(see Zhang and Guhathakurta (2017) for model design and im- fleet size of SCVs remains constant in the simulation. In reality, SCV
plementation details). drivers are free to decide when to enter and leave the system.
The SAV simulation model is a discrete event model, including We integrated the simulation of PAVs and PCVs into the current SAV
“call,” “pickup,” “move,” “drop-off,” “find parking,” and “park” events. simulation model by (1) modifying the existing call, find parking, and
A traveler schedules call events to enter the system at the trip departure parking events, and (2) adding a new “leave parking lot” event. We
time. The central dispatching system assigns the closest available SAV revised the call event to be able to handle PAV and PCV users in ad-
to fulfill the travel demand, such as to pick up and drop off the client. dition to SAV and SCV clients. For PAV and PCV users, once they enter
The system identifies the parking destination with the lowest parking the system at the trip departure time, instead of matching them up with
costs including relocation costs, empty travel penalty costs, and parking SAVs or SCVs, the system will directly schedule a find parking event at
ticket costs for idling vehicles, as shown in the equation below. A the trip arrival time and update the vehicle location information cor-
cruising threshold is set for SAVs because the system is not able to respondingly to the destination of the trip. In the find parking event,
determine whether it is more economically efficient to park or to keep similar to SAVs, PAVs will search for a parking space that can minimize
cruising for clients. Past simulation experiments show that the longer the total parking costs, with a constraint that PAVs will have sufficient
the SAVs are allowed to search for parking, the higher the average time to return to pick up the owners, as shown in the equation below.
waiting time for clients, due to the spatial mismatch between where The farthest a PAV may park and still return on time to pick the owner
vehicles park (typically residentially zones) and where the travel de- up is in the TAZ that is half of the activity duration away from the trip’ s
mand is generated (typically commercial districts). destination.
After identifying parking spaces, SAVs will start to navigate along
the shortest path in the network (i.e., minimum time costs) towards the Park TAZ = min (tci, j + i, j + pcj )
j : tti, j t PAVk
assigned parking space. The vehicle may be re-assigned to serve in-
coming clients while relocating for parking. In this case, the parking
space will no longer be used. tPAVk = 0.5 * activity durationk

Park TAZ = min (tci, j + i, j + pcj ) PCVs will directly park at their destination if space remains in the
j : tti, j tcruise
lot. Otherwise, PCVs will park in the nearest TAZ with empty parking
Where, space. If there are multiple TAZs with the same travel distance, the
i is the TAZ index for the owner’s activity location; driver will park in the TAZ with the lowest parking fee.
j is the index for possible parking TAZs; For the parking event, the status of PAVs and PCVs will be set to
tti, j is the travel time between TAZ i and j ; “parking,” and the corresponding parking space will be labeled as
tcruise is the threshold for empty cruising; “occupied.” The system will then schedule a leave parking lot event at
tci, j is the operation cost for relocating between TAZ i and j , esti- the time when PAVs and PCVs no longer need the parking lot. The PAVs
mated by multiplying the mile-based operation costs with distance will automatically leave the parking lots to ensure the owners can be
between TAZ i and j ; picked up at the departure time of the upcoming trip. In other words,
i, j is the penalty for empty relocating between TAZ i and j , esti- the PAVs will leave at the owner’s departure time less the time it takes
mated by multiplying the time-based penalty cost with the travel time to travel from the parking lot to the location of the owner’s latest de-
between TAZ i and j ; parture. The PCVs will free the parking lot at the exact departure time
pcj is the parking cost at TAZ j . In this study, we used a flat parking of the upcoming trip. In the leave parking lot event, the system will
charging scheme, indicating that the vehicle will be charged at the same update the status of PAVs and PCVs to “moving” and label the used
rate, regardless of the length of parking time. parking space as “available” for other vehicles to use.

4
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

Table 2
Input Data Source Summary.
Input Data Data Description Data Boundary Data Source

Synthetic trip profiles Final synthesized weekday trips for years 2015, 2020, 2030, and 2040, 10-County Atlanta Metro Atlanta Regional Commission (ARC)
including trip origin, destination, departure time, travel time, mode, trip ID, Activity-based Model (ABM)
tour ID, household ID, and person ID
Loaded road network Final trip assignment results trips for year 2015, 2020, 2030, and 2040, include 10-County Atlanta Metro ARC ABM
link length, travel speed by time of the day
Downtown parking Public parking in Atlanta Downtown Area, includes number and price of Downtown Atlanta Central Atlanta Progress (CAP)
parking, and parking infrastructure type
Atlanta surface Esri polygon shapefile of parking surface in the City of Atlanta City of Atlanta City of Atlanta
parking
Fulton tax assessor Esri polygon shapefile of 2017 tax assessor parcel, includes appraised parcel Fulton County (includes part Fulton County
parcel value, land use types, square feet, etc. of the City of Atlanta)
Dekalb tax assessor Esri point shapefile of 2017 tax assessor parcel, includes appraised parcel value, Dekalb County (includes part Dekalb County
parcel land use types, and tax year, etc. of the City of Atlanta)

Table 3 assessor parcels as the model inputs. Table 2 provides brief data de-
Number of Synthetic Trips by Trip Type and Simulation Year. scriptions and sources for the input data.
Trip Type Year The trip profiles for year 2015–2040 are obtained from the Atlanta
Regional Commission (ARC)’s latest Activity-Based travel demand
2015 2020 2030 2040 Model (ABM). The trip profiles include trip characteristics, such as trip
origin and destination, departure and arrival time, trip purpose, mode,
Outra-city Trips 839,163 912,774 937,679 1,018,462
(% of total) (37.7%) (37.5%) (35.8%) (35.0%) etc. for each synthesized household and person in the 10-county Atlanta
Intra-city Trips 1,389,409 1,524,492 1,684,818 1,889,296 metropolitan area. Because the parking inventory data, including both
(% of total) (62.3%) (62.5%) (64.2%) (65.0%) number of parking spaces and parking price, are limited to the City of
Total Number of Trips 2,228,572 2,437,266 2,622,497 2,907,758 Atlanta, we only selected synthesized vehicle trips that end within the
that end in Atlanta
Atlanta urban boundary (i.e., trips that need to consume parking spaces
in the City of Atlanta) for simulation purposes. The number of trips
included in the model by simulation year is shown in Table 3. The
3.3. Parking demand and space estimation number of trips increased from 2.2 million in 2015 to approximately
2.9 million in 2040. The percent of intra-city trips, i.e., trips that both
Parking Demand refers to the number of times vehicles park at a start and end in the city boundary, also increases from 62.3%–65%,
certain location, while Parking Space refers to the number of parking indicating that Atlanta expects to have a more compact urban core in
lots needed at the location. The simulation is set to run for 31 simula- the future.
tion days to calculate: The TAZ-level parking capacity and price were imputed using de-
tailed downtown parking inventory data, city-wide parking surface
• the average parking demand by TAZs, travel mode, and time of the data, and tax assessor parcels. According to the latest report by Central
day (PD ); Atlanta Progress (CAP, 2014), there are 93,000 parking spaces in
• the
i, m, t
total parking demand by TAZs and time of the day (PD ); downtown Atlanta, with approximately 300 square feet per parking
• parking
i, t
space requirements by TAZs and mode (PS ); space. The parking capacity for the TAZs outside of downtown areas are
• parking space requirements by TAZs (PS ).
i, m
i then imputed by dividing the TAZ level parking surface square footage
by 300. This imputation approach is also used in an earlier Atlanta
The following equations are used for calculation. The first simula- parking study (Zhang and Guhathakurta, 2017). This method may lead
tion day is excluded from the analysis, as the SAVs are randomly dis- to an underestimation of structured parking, as the parking surface data
tributed to initialize the simulation. do not include building volume. We compared the total parking de-
D mand generated by the synthetic vehicle trips in year 2015 with the
PDi, m, t , d
PDi, m, t = d= 2
TAZ level parking capacity. The synthetic parking demands in all TAZs
D 1 fall below the imputed parking space using the above approach, in-
dicating that there is a limited amount of structured parking outside of
PDi, t = PDi, m, t
m M the downtown area in the City of Atlanta. The parking price in the TAZ
with the highest average land value was set at the highest rate (i.e., $20
PSi, m = max PDi, m, t
0 t 1440 per entrance), whereas the parking price in the urban fringe TAZ where
the average land value is the lowest was set to be free. For the rest of the
PSi = max PDi, t TAZs, the parking price was estimated to be inversely correlated with
0 t 1440
the average land value.
Where, In addition to the above model inputs, the core simulation para-
d denotes the simulation day; meters are shown in Table 4. The fleet sizes of SAVs and SCVs are in-
D is the total number of simulation day; itialized as the corresponding number of clients divided by 30, as-
m denotes the travel mode, such as SAV, SCV, PAV or PCV; suming that each SAV or SCV will serve on average 30 incoming trips.
M is the set with all travel modes. Prior simulation results suggest that SAVs can serve approximately
26–35 trips on a daily basis, without keeping the clients waiting for
4. Data and simulation environment setup more than 10 min (Chen et al., 2016; Fagnant and Kockelman, 2014;
Zhang and Guhathakurta, 2017). In this study, we adopted the average
The proposed simulation model applies synthetic trip profiles, of this range to determine the fleet size of SAVs and SCVs. The intra-
loaded transportation network, local parking inventory, and tax zonal travel time for TAZ i (Timeintra, i ) is a random number that is

5
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

Table 4
Set Up for Key Model Parameters.
Notation Description Parameter Input

NumSAV SAV fleet size Number of SAV clients / 30

NumSCV SCV fleet size Number of SCV clients / 30
Quitthreshold Maximum time before SAV/SCV clients quit 15 minutes (Zhang et al., 2015a)
Timeonboard Time clients used to onboard SAV/SCV 1.5 minutes (Zhang et al., 2015b)
Timedropoff Time clients used to get out of SAV/SCV 1.5 minutes (Zhang et al., 2015a)
Timeintra, i Time vehicles used to travel within the TAZs (i.e., for trips starts and ends in the same TAZ) Areai
rand. uniform (0, )
2 * 30
Cruisingmax Maximum allowed parking cruising threshold for SAVs and SCVs 10 minutes
Costsav SAV operation cost per mile $0.3 per mile
Costscv SCV operation cost per mile $0.6 per mile
Costpav PAV Operation cost per mile $0.5 per mile
Empty travel penalty $0.1 per minute

Table 5 (National Surface Transportation Infrastructure Financing Commission,

Total Parking Space Demand by Simulation Year and Scenarios. 2009).
Scenarios 2015 2020 2030 2040 Furthermore, this model made the following simplifications:

BAU
Transition Scenario
727,863
727,863
818,545
765,087
847,665
665,671
930,357
605,974
• The SAV and SCV systems will not provide ride-sharing services, as
previous studies show that the percentage of ride-sharing does not
SAVs or SCVs N/A 22,116 67,184 121,254
PAVs N/A 41,536 210,341 377,242
have a strong correlation with either the spatial or the total parking
PCVs 727,863 712,909 410,601 128,766 footprint of the systems in a sprawled city like Atlanta (Zhang et al.,
Total Reduction Rate N/A 6.5% 21.5% 34.9% 2015b);
• All travelers will maintain existing travel patterns, in terms of des-
tination choices and trip frequency;
uniformly distributed between 0 and Areai (eai is the area of TAZ i),
2 * 30
assuming the travel speed is 30 miles per hour (mph). The operation
• There is no heterogeneity in the choice of SAVs, SCVs, PAVs, and
PCVs. The differences in the attitudes across population segments
cost for SCVs is set as $0.6 per mile (AAA, 2016), $0.3 per mile for SAVs may potentially change the spatial distribution of parking, but will
(Bridges, 2015; Burns et al., 2013; Zhang and Guhathakurta, 2017), and have a limited impact on the total amount of required parking space.
$0.5 per mile for PAVs (Albright et al., 2016; Barclays, 2016). The
penalty for empty AVs relocating for parking is set as $0.1 per minute The model is implemented using Python 2.7. The model takes

Fig. 1. Spatial Distribution of Parking Demand by Simulation Scenarios.

6
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

20 min, 1 h, and 4 h to complete one simulation-day for years 2020, average, uses 2.6 parking spaces in the city. SAVs and SCVs can reduce
2030, and 2040, respectively, using a computer with 3.1 GHz Intel Core the parking demand by 72–75%, compared to the BAU scenario (i.e. if
i7 CPU and 16 G of memory. the corresponding share of trips is completed by PCVs). The reduction
in parking infrastructure may be as high as 90% in most of the TAZs.
5. Model results and sensitivity tests For some TAZs, especially those with a relatively lower parking price,
the parking space demand may increase due to the AV relocation pro-
In this section, we first discuss the trends in the demand for parking cess. The PAVs may also reduce the parking space demand slightly by
space from 2020–2040. We then describe the spatial distributions of around 5%, as PAVs may keep cruising without parking while waiting
parking space demand and compare the results with a Business as Usual for their owners to complete a brief activity. However, compared with
(BAU) model. Finally, we explore the temporal distributions of parking SAVs and SCVs, PAVs are inefficient in reducing parking demand.
demand by land use types to provide policy implications for future The results suggest that market penetration of car-sharing rather
parking land use planning. than vehicle automation is the key to parking demand reduction. The
parking space reduction is positively and linearly correlated with the
5.1. Total parking demand market share of SAVs and SCVs. PAVs cannot be used as an effective
tool for parking space reduction. The estimated parking space demand,
Table 5 shows the total parking demand by scenarios and year. The however, does not include curbside parking. The synthetic trip profiles
parking demand results for the transition scenarios are further broken are insufficient to determine the current curbside parking demand for
down by travel mode, including SAVs, SCVs, PAVs, and PCVs, to ex- picking up and dropping off travelers. However, it is self-evident that in
amine how parking spaces are consumed by different types of vehicles. the era of car-sharing and AVs, the demand for curbside parking will
The total parking demand in the transition scenario is estimated using increase exponentially, as the majority of the served trips will generate
equation i ( max PDi, t ) , while the mode-specific parking demand is
I at least two curbside parking demand events (i.e., in the loading and
0 t 1440 unloading area). If not managed proactively, curbside parking-induced
estimated as ( max PDi, m, t ) . Therefore, the sum of mode-specific
I
i
0 t 1440
congestion could be catastrophic.
parking space is equal to or larger than the total parking space demand.
The reduction rates are then calculated by comparing the total parking 5.2. Spatial distribution of parking demand
demand from the BAU and the transition scenarios.
Simulation results for transition scenarios from year 2020–2040 In addition to the total parking demand reduction, the spatial dis-
suggest the parking demand may be significantly reduced in the future, tribution of parking space demand may also be fundamentally altered,
as shown in Table 5. The BAU results are estimated by aggregating the as shown in Fig. 1. In the BAU scenarios, the parking demand is con-
parking demand using the ARC synthetic trip profiles (i.e., all the trips centrated in the downtown, midtown, Buckhead, and airport areas. In
are PCV trips). From 2015–2040, the parking space demand increases the transition scenarios, the parking space demands in the above-
from 727,863–903,357 by 24.1% in the BAU scenario. In the transition mentioned hubs are reallocated to TAZs that are four to five miles away,
scenario, the parking space demand first increases from 727,863 to where the combined parking costs are lower. As a result, the parking
765,087 from year 2015–2020, and then decreases dramatically from density in most of the midtown and airport areas falls below 15 spaces
765,087–605,974 by year 2040. The increase in parking demand from per acre, while the parking demand in some residential TAZs, especially
2015 to 2020 may be caused by an increase in the total number of trips in the western and southern parts of Atlanta, increases to above 100
accompanied by limited growth in the market share of SAV and SCV spaces per acre. These neighborhoods are typically disadvantaged
modes. communities, with a higher percentage of minority or low-income
The reduction rate in parking space during the transition period households.
increases from 6.5% in 2020 to 34.9% in 2040. The growth in the de- The parking density in TAZs with different land use types is dis-
mand reduction rate can be primarily attributed to the increase in the played in Fig. 2. For all simulation years, the parking space demand
assumed share of SAVs and SCVs over time. Each SAV or SCV, on density is highest in the Central Business District (CBD). Upon the

Fig. 2. Average Parking Space per Acre by Land Use Types and Simulation Scenarios.
*The classification methods and results of CBD, Commercial-oriented, Mixed-use, and Residential-oriented zones are described by Zhang and Guhathakurta (2017).

7
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

Fig. 3. PAVs Excessive VMT Generation.

introduction of SAVs, SCVs, and PAVs, the parking space demand also have more flexibility in terms of where to park compared with
density in the CBD decreases the most across simulation years. In CBD shared vehicles, as the owners may stay at the work location for the
TAZs, the parking intensity reduction rate tends to increase in the fu- next three or four hours (i.e., providing a larger parking TAZ searching
ture, from 10% in 2020 to 42% in 2030 and 75% in 2040. In Atlanta, up radius/threshold). The PAV empty VMT tends to increase again at noon
to 20% of the downtown space is occupied by parking. The model re- when owners generate work-based trips. VMT due to PAV relocation
sults show that by 2040, 15% of the downtown area may be freed up for decreases after peak evening hours when most of the vehicle owners
redevelopment. Therefore, decision makers may start to plan strategi- return to homes typically located in residential or mixed-use zones,
cally regarding how to re-zone and infill these redundant parking where the parking price is lower. At the penalty of $0.1 per minute, the
spaces for more sustainable development. PAVs tend to generate over 73.9% of the excessive VMT in the city
Meanwhile, the average parking space density in the residential boundary. This may lead to more congestion if PAVs become the pre-
zones may increase from four spaces per acre in 2015 to eight spaces valent travel mode in the future.
per acre in 2040. The spatial distribution of parking indicates that The SAVs and SCVs, on the other hand, only generate around 18%
disadvantaged communities are more likely to witness a spike in the use of the excessive VMT during the parking relocation process. The tem-
of local parking space. The clustering of parking in disadvantaged poral distributions of the SAV and SCV excessive VMT are also different
communities, however, may not necessarily enhance the accessibility of from the PAVs, as shown in Fig. 4. Shared vehicles are more likely to
these neighborhoods. Whether the concentration of parking in low-in- relocate for parking at noon and the early afternoon when the travel
come neighborhoods exaggerates or resolves existing mobility chal- demand is lower compared with peak hours. During peak hours, re-
lenges must be further considered together with the temporal dis- locating vehicles can be easily matched up with incoming clients and,
tribution of parking. therefore, terminate the relocating process. At noon, the shared vehicles
A significant amount of excessive VMT is generated exclusively may automatically relocate to park in TAZs with lower parking prices
during the vehicle parking allocation process, especially for PAVs. The without being assigned for service.
temporal distribution of empty VMT, i.e., AVs moving without pas-
sengers, is shown for PAVs in Fig. 3. The PAVs tend to relocate most 5.3. Temporal distribution of parking demand
significantly during and after peak morning hours. Many PAV owners
may spend the entire morning at work locations in commercial districts, The temporal distributions of parking demand in different types of
where the parking permits tend to be expensive. Additionally, the PAVs TAZs suggest the variations in parking space demand are introduced

Fig. 4. SAVs and SCVs Excessive VMT Generation.

8
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

Fig. 5. Temporal Distribution of Parking Demand by TAZ Land Use Types.

Table 6 Table 7
Sensitivity Test Results by Empty Travel Penalty. Sensitivity Test Results by Parking Price.

Penalty per minute ( ) $0.01 $0.05 $0.1 $0.25 $0.5 Parking price 0.25 * 0.5 * Current 2 * Current
(Base) Current Current Parking Parking
Parking Parking Price Price
PAVs Price Price (Base)
Parking demand/vehicle 4.1 4.2 4.2 4.2 4.3
Excessive VMT (%) 176.3% 104.3% 73.9% 35.1% 19.0% PAVs
SAVs and SCVs Parking demand/ 4.3 4.2 4.2 4.0
Parking demand/vehicle 2.5 2.5 2.6 2.8 3.3 vehicle
Excessive VMT (%) 23.5% 20.1% 18.2% 14.7% 10.6% Excessive VMT (%) 29.7% 56.4% 73.9% 118.5%
SAVs and SCVs
Parking demand/ 2.9 2.8 2.6 2.3
vehicle
Excessive VMT (%) 9.5% 15.7% 18.2% 44.7%
primarily during midday, as shown in Fig. 5. For TAZs in the CBD and
commercial-oriented areas, the majority of parking space resources are
consumed during the daytime and the demand for parking peaks at
noon in the BAU scenarios. Atlanta is a typical monocentric city, where PAV, SAV, and SCV parking spilled over from the CBD and commercial-
the parking lots in the CBD and commercial-oriented areas primarily oriented zones during the day. The reduction in parking demand in
serve businesses and employees living outside of the urban boundary. In mixed-use TAZs can be primarily attributed to the increase in car-
the transition scenarios, the occupancy rate for parking lots in the CBD sharing programs, as the reductions during the night and daytime are
and business hubs diminishes the most during the daytime, especially quite consistent, which differs from the trends found in the CBD and
between 9:30 AM and 3:30 PM. This is because AVs, especially the commercial-oriented zones.
PAVs, tend to relocate to park in nearby mixed-use or residential TAZs The variations in the temporal distribution of parking demand in
where the parking price is lower. residential-oriented zones are different from the rest of the city. Instead
The parking demand in mixed-use TAZs seems to be more stable, of parking demand reduction, the parking demand is likely to increase
compared to CBD and commercial-oriented areas. The parking spaces in dramatically during the daytime. PAVs will consume the majority of the
mixed-use TAZs will serve local residents at night and also attract some parking space during the daytime without benefiting the local residents.

9
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

Furthermore, the SAVs and SCVs tend to park in these zones mostly system users. It is recommended to initialize these adaptations as soon
during the evening and early in the morning. as possible so that long-term parking space reduction and/or re-
development goals can be met.
5.4. Sensitivity tests San Francisco became the first progressive city in the U.S. to remove
the minimum parking requirement from its zoning ordinance in 2018.
In this step, we first ran the model with different empty travel pe- Economic incentives may be necessary to encourage the initial adoption
nalties ( ) to determine how excessive VMT and parking demand may of this type of policy. We anticipate that as AVs and MoD gain more
change correspondingly. We then conducted sensitivity tests to examine traction, a reduction in parking requirements will be more positively
how the parking price will influence model outputs. To reduce simu- perceived and accepted. While removing the minimum parking re-
lation time, all of the sensitivity tests are implemented using trip pro- quirements, authorities may also need to develop a regulatory frame-
files, network data, and assumed market shares for 2020. work and guidance to develop more effective on-site loading and un-
Model results with empty travel penalties ( ) input from $0.01 to loading zones, connectivity to transit stations, and walkability of the
$0.5 per minute are displayed in Table 6. The excessive VMT generated surrounding areas. Such features are important to attract clients using
by PAVs is very sensitive to change in the empty travel penalty. When different transportation modes, especially in commercial districts where
the penalty is set to be low (e.g., $0.01 or 0.05 per minute), the PAVs the demand for shared mobility is high. Adjustment of the building code
will generate over 170% or 104% excessive VMT, respectively, which is for structured parking should be negotiated as well, to support the
greater than the occupied VMT. When the penalty increases to $0.5 per transition of redundant parking decks into more economically viable
minute, the empty VMT declines to 19.2%. The reported excessive VMT uses at a reasonable cost (Nourinejad et al., 2018).
for PAVs only includes the empty VMT generated within the study area, Additionally, AVs and MoD offer a unique opportunity to unbundle
i.e., City of Atlanta. The parking relocation VMT footprint could be parking costs from the development, given they can unload people at
considerably larger at the regional level per PAV owner. The results also activity destinations and park remotely. In the residential-oriented
suggest that parking demand per PAV is not sensitive to the empty TAZs, parking can be decoupled from ownership by rationing and/or
travel penalty fee. The parking demand is only reduced by 10% when charging for permits to park. This policy option will promote high-
the empty travel penalty decreases from $0.5 to $0.01 per mile. density development, reduce vehicle ownership, and provide flexibility
The results for SAVs and SCVs are consistent with prior studies, to manage future on-street parking needs created by AVs and MoD.
indicating that excessive VMT generation is linearly and positively Looking forward, households may be allocated with a baseline number
correlated with the empty travel penalty. Compared with PAVs, SCVs of on-street parking spaces to accommodate their CVs. In the long-run,
and SAVs tend to generate significantly less empty VMT, as the max- these permits can be re-allocated to AV owners and MoD fleet man-
imum time that an SAV or SCV may cruise is 15 min in the model. agement companies. The significant reduction in parking demand at the
Additionally, shared vehicles may be assigned to serve calling clients CBD and employment-oriented TAZs could provide an opportunity to
while relocating to parking. The parking space consumed by each SAV unbundle parking costs from employer packages or shopping fees.
or SCV fleet is quite sensitive to the changes in the empty travel penalty. Employers currently tend to maximize their access to the labor pool by
Unlike PAVs, the longer the SAVs and SCVs cruise for parking, the more providing parking benefits to employees. When the uptake of PAVs
likely they will not consume parking spaces because they are assigned becomes high, employers may only need to offer parking cash-out
to serve a calling client instead. schemes to attract talent. Similarly, retail owners may reconsider the
We also ran the model with different parking price inputs, with provision of free parking or on-site parking as incentives to attract
results shown in Table 7. For both PAVs and shared vehicles, the more consumers (although currently, there is little indication that these po-
expensive the parking, the less parking space will be consumed. For licies work effectively to attract clients [Marsden, 2014]).
PAVs, when the parking price doubles, significantly more vehicles tend
to give up parking and will empty cruise while waiting for their owners, 6.2. Implications for existing parking infrastructures
especially for owner activities with short durations, e.g., shorter than
30 min. SAVs and SCVs also tend to park less when parking becomes 6.2.1. Repurpose existing on-street parking infrastructures
expensive, as they are more likely to cruise in search of incoming clients Such a dramatic increase in the loading and unloading demand
rather than directly park. For both PAVs and shared vehicles, the ex- suggests that there is an urgent need for more flexible and adaptable on-
cessive VMT generation increases when parking becomes more ex- street parking policies in commercial districts. In other words, it is time
pensive, as expected. Shared vehicles remain more competitive by to rethink streets from “parking streets” to “pick-up and drop-off
generating less empty VMT and consuming fewer parking spaces across streets.” Cities with larger-scale uptake of ride-hailing services than
different parking price scenarios. Atlanta, such as San Francisco, New York, and Washington, D.C., have
already witnessed crowded streets due to double-parked Uber or Lyft
6. Land use and transportation policy implications fleets. To address these issues, San Francisco and Washington, D.C. have
actively sought to examine different curbside space reallocation modes
This section discusses implications for land use and transportation in pilot on-street parking conversion projects (e.g., Fehr and Peers,
policies based on the model and sensitivity test results, in particular, 2018; ITF, 2018; Nelson Nygaard, 2014). The recommendations from
possible adaptations to new development regulations, existing parking these pilot cases include (1) reducing on-street parking and flexible
infrastructure, and travel demand management. allocation, (2) dynamic pricing and implementation of loading and
unloading zones using geofencing technology (a technology that uses
6.1. Implications for new development the global positioning system (GPS) to define geographical boundaries
of activities), (3) establishing effective tracking and monitoring of
The model results suggest that there will be a significant reduction overall curbside performance across modes of transportation, (4) close
in parking spaces (around 35% by the year 2040), as well as a spatial collaboration among key local stakeholders for more effective man-
shift in the density of parking demand in the future. Therefore, pol- agement, and (5) prioritizing access for more sustainable road users,
icymakers may need to adjust the existing minimum parking require- such as ride-sharing passengers, transit users, cyclists, and pedestrians.
ment and building codes to avoid an oversupply of parking infra- Meanwhile, the model also suggests that when parking in CBD and
structure in the future. Changes to minimum parking requirements may commercial zones becomes expensive, the PAV, SAV, and SCV parking
take a long time to implement due to objections from key stakeholders, tends to spill over into residential zones, which may cause some equity
such as home buyers, business and retail owners and transportation concerns. Instead of evenly affecting all residential zones in the city,

10
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

AVs and shared fleets are more likely to park in low-income neigh- the parking price, vehicles may prefer continuous empty cruising to
borhoods, where the land price is lower. PAVs tend to occupy these avoid parking and being charged the associated empty VMT penalty
parking spaces during the daytime, while shared fleets may park in fees.
lower-cost spaces primarily at night. When the market penetration of
AVs and MoD is low, such a phenomenon may not induce significant 6.3. City-wide strategic planning for parking infrastructures
issues, and sometimes it is appealing to prioritize parking for MoD in
lower-income areas to encourage the adoption of this travel mode. Due to the “floating” nature of the future parking demand, parking
However, in the long run, lower-income communities may be forced to infrastructure plans should go beyond site plans and be managed at the
bear the opportunity costs of parking infrastructure for the majority of city-wide scale to fit the strategic priority of the city. Moving forward,
the city without adequate compensation. To improve the quality of life cities should outline principles for livable environments in the era of
for these residents, occupied parking spaces could instead be converted shared mobility and automated vehicles to guide long-term parking
into green space, active travel facilities, or flood management infra- infrastructure plans, especially balancing the trade-offs among acces-
structure (i.e., flooding is an issue in several low-income communities sibility, economic development, and social justice. Streets and parking
in Atlanta). Therefore, it will be critical to dynamically charge or issue a infrastructures are increasingly contested public resources. Currently,
limited number of permits for PAV and shared mobility fleet parking, most U.S. cities prioritize automobile use over other travel modes. The
and the corresponding revenue should be intentionally allocated to coming of AVs and MoD services provide cities with a precious op-
improve the built environment in the affected communities. portunity to plan urban spaces strategically - not only for automobiles
In the future, cities will need to standardize the operation of pilot but for heterogeneous usage. Thus, it is imperative for cities to plan out
projects and implement them based on land uses, travel demand pat- the spatial and temporal distributions of parking from a city-wide
tern, and availability of existing transportation infrastructures. perspective, rather than managing individual parking lots and spaces in
Meanwhile, the pilot projects highlight some concerns that need to be a piecemeal fashion.
addressed. First, existing knowledge regarding the use of street curb
space remains poor, due to a lack of common standards for encoding 6.4. Implications for travel demand management
information about curb use, inadequate infrastructure to monitor and
implement curb space regulation strategies, and limited understanding The model results also suggest that policymakers need to consider
of the impacts of ride-hailing services and e-commerce on the use of tools to reduce empty VMT generation. The sensitivity test results
curb space. Second, cities should anticipate and plan for the revenue suggest that empty VMT penalty fees and the spatial distribution of
impacts of shifting on-street parking to passenger loading and un- parking prices may be implemented to curb empty vehicle cruising. To
loading zones and devise new pricing schemes correspondingly, as further reduce city-wide VMT generation, decision makers may allocate
discussed further below. VMT penalty fees, parking revenues, and congestion fees to fund transit
operation and/or capital investment. Chicago and New York have al-
6.2.2. Repurpose existing off-street parking infrastructures ready increased a trip-based tax on Uber and Lyft to fund local transit
Given the substantial anticipated reduction in the demand for on- systems (Lumb, 2018; Small, 2018). The major concern regarding such
site parking infrastructures, planners need to start thinking about a policy is that the allocated funding is not effectively used in the transit
transforming these infrastructures. For instance, the park-and-ride system to attract riders siphoned by Uber and Lyft. Transportation
properties at transit stations may need to be gradually converted into planners and engineers argue that the funding should be allocated
kiss-and-ride spaces to encourage the integration of MoD services and based on transit service performance measures so that travelers may be
transit. For large off-street parking infrastructure in commercial dis- more willing to transfer between ride sourcing and transit services (Yan
tricts, innovative infill or redevelopment policies should be considered et al., 2018). Nonetheless, some transit advocates argue that this type of
to promote more sustainable use of these spaces in the future (e.g., funding mechanism could further disadvantage transit systems when
NACTO, 2017). However, without a practical funding mechanism, these negotiating the annual budget and when competing for funding with
utopian redevelopment scenarios are unlikely to materialize. Some street paving, highway expansion, or bridge maintenance (Goldwyn,
critical questions should be considered, such as who will pay for the 2018).
redevelopment, how to determine if the redevelopment is economically,
environmentally, and socially sustainable, and how to fairly allocate the 7. Conclusions
economic benefits of the redevelopment?
The main contribution of this study is three-fold. First, we devel-
6.2.3. Restructure the current parking pricing scheme oped a novel agent-based simulation model that can be used to examine
Parking price schemes need to switch from fixed structures to more the parking demand of a mixed combination of AV and shared mobility
dynamic and flexible strategies to manage the spatial allocation of modes, such as SAVs, SCVs, PAVs, and PCVs. The model can potentially
parking demand in real-time. Despite some pioneering cities such as be applied to all cities with sufficient input data. Second, the model is
New York and San Francisco, which have already implemented pro- implemented using data from the Atlanta metropolitan area and si-
gressive parking price schemes that vary fees by the time of the day, the mulated to obtain parking forecasts up to the year 2040. The results
conventional parking pricing structure remains static. With the coming show two changing trajectories in parking demand, including a) a re-
of AVs, MoD, and the Vehicle to Infrastructure (V2I) technology, mu- duction in off-street parking demand and a significant increase in
nicipalities will eventually be able to manage parking demand in real curbside loading and unloading demand, and b) spatial shifts in parking
time by altering prices based on the availability of parking spaces and demand from CBD and commercial zones to mixed-use and residential
traffic conditions. However, significant improvements in infrastructure, zones. The results also suggest that cities may witness an increase in
parking revenue collection systems, and real-time dynamic parking empty VMT generation. According to the most optimal AV and MoD
price determination models will be required to support the im- adoption rate, these changes will be quite significant by the year 2030.
plementation of such a pricing policy. The simulated results are not only applicable to the City of Atlanta, but
Moreover, it may not be ideal to associate the environmental per- to many other North American cities of similar size and urban form.
formance of vehicles (e.g. VMT) with parking fees; instead, parking Third, given the changing trajectories in parking demand, this study
should be priced independently. The model results suggest that in the arrives at the following land use and transportation policy indications.
era of AVs and MoD, fleets are expected to generate a significant In the near-term, cities may consider converting existing, static parking
amount of empty VMT. If the empty VMT penalty fees are bundled with infrastructure to flex parking spaces to meet the loading and unloading

11
W. Zhang and K. Wang Land Use Policy 91 (2020) 103963

demand of shared mobility. For instance, cities can designate on-street electric vehicle fleet: implications of vehicle & charging infrastructure decisions.
parking spaces as Uber and Lyft service spaces to avoid double parking, Transp. Res. Part A Policy Pract. 94, 243–254.
Chester, M., Horvath, A., Madanat, S., 2010. Parking infrastructure: energy, emissions,
and can convert park-and-ride spaces at transit stations to kiss-and-ride and automobile life-cycle environmental accounting. Environ. Res. Lett. 5 (3),
spaces to support the integration of shared mobility with the transit 034001.
system. Over time, cities may need to remove minimum parking re- Deloitte, 2017. Gearing for change. Deloitte Development LLC.Gearing for change.
Deloitte Development LLC.
quirements for new development and update parking building codes for Economist, The, 2016. The Driverless, Car-sharing Road Ahead. The Economist.
more economically efficient redevelopment in the future. Similarly, Retrieved from. https://www.economist.com/business/2016/01/09/the-driverless-
decision makers need to devise land redevelopment and infill policies to car-sharing-road-ahead.
Fagnant, D.J., Kockelman, K.M., 2014. The travel and environmental implications of
encourage sustainable conversions of parking spaces in CBD and com- shared autonomous vehicles, using agent-based model scenarios. Transp. Res. Part C
mercial zones, and to introduce dynamic and equitable parking price Emerg. Technol. 40, 1–13.
schemes to compensate the use of parking spaces in residential and Fehr and Peers, 2018. San Francisco Curb Study. Retrieved from. http://www.
fehrandpeers.com/wp-content/uploads/2018/10/SF_Curb_Study_2018-10-19_web-
mixed-use zones. There is also a need to examine the possible erosion of
download.pdf.
existing parking revenue and how to effectively charge for loading and Giovanni, C., Alemi, F., Tiedeman, K., Handy, S., Mokhtarian, P., 2018. The Adoption of
unloading vehicles. Last but not least, the public sector needs to bundle Shared Mobility in California and Its Relationship with Other Components of Travel
travel demand management policies, such as road pricing and conges- Behavior (No. No. NCST-RR-201802). .
Goldwyn, E., 2018. Taxing Uber and Lyft to Fund Transit Isn’t Fair to Transit. Retrieved
tion fees, with land use policies to curb empty VMT generation and to from. https://www.citylab.com/transportation/2018/04/taxing-uber-and-lyft-to-
promote the use of more sustainable travel modes, such as active travel, fund-transit-isnt-fair-to-transit/558812/.
public transit, and high-occupancy vehicle sharing. Haboucha, C.J., Ishaq, R., Shiftan, Y., 2017. User preferences regarding autonomous
vehicles. Transp. Res. Part C Emerg. Technol. 78, 37–49.
Some limitations of this study provide an opportunity for future International Transport Forum, 2015. Urban Mobility System Upgrade: How Shared Self-
research. First, the proposed model is constrained by several model driving Cars Could Change City Traffic.
simplifications and assumptions, especially the uncertainties of tech- ITF, 2018. The Shared-Use City: Managing the Curb. Retrieved from. https://www.itf-
oecd.org/sites/default/files/docs/shared-use-city-managing-curb_3.pdf.
nology adoption and the absence of disaggregated parking inventory Kondor, D., Paolo, S., Kakali, B., Xiaohu, Z., Carlo, R., 2018. Large-scale Estimation of
and parking behavior data. Second, the model does not incorporate Parking Requirements for Autonomous Mobility on Demand Systems. arXiv Preprint
possible changes in latent travel demand induced by AVs, shared mo- arXiv:1808.05935.
Krueger, R., Rashidi, T.H., Rose, J.M., 2016. Preferences for shared autonomous vehicles.
bility, and land use variations in the CBD. Third, this study does not Transp. Res. Part C Emerg. Technol. 69, 343–355.
explore other efficiencies that could be gained via progressive parking Lavieri, P.S., Garikapati, V.M., Bhat, C.R., Pendyala, R.M., Astroza, S., Dias, F.F., 2017.
policies, again due to a lack of travel behavior data. Finally, the model Modeling individual preferences for ownership and sharing of autonomous vehicle
technologies. Transp. Res. Rec. J. Transp. Res. Board (2665), 1–10.
only explores the possible impacts of AVs and shared mobility, while
Litman, T., 2014. Autonomous Vehicle Implementation Predictions. Victoria Transport
the influences of other disruptive innovations, such as vehicle elec- Policy Institute, pp. 28.
trification and e-commerce, are not explored. Lumb, D., 2018. New York Approves Surcharge for Uber and Lyft Rides in Manhattan.
To support dynamic and flexible parking demand management in Retrieved January 14, 2019, from. https://www.engadget.com/2018/04/02/new-
york-surcharge-uber-lyft-manhattan/.
the future, more planning and policy research should be conducted for Marsden, G., 2014. Parking policy. Parking Issues and Policies. Emerald Group Publishing
cities that are dramatically different from Atlanta in terms of urban Limited, pp. 11–32.
form and parking space management policies, such as those in Europe Megens, I.I., Schaefer, W.W., van der Waerden, P.P., Masselink, P.P., 2015. Vehicle
users’preferences concerning automated driving implications for transportation and
and Asia. Additionally, more research should be conducted to evaluate market planning. Explor. Energy Neutral Dev KenW 2 IBrabant Part 4, 99.
the feasibility and effectiveness of pilot parking projects for the new NACTO, 2017. Blueprint for Autonomous Urbanism. Retrieved from. https://nacto.org/
mobility to gain an evolving understanding of curbside management wp-content/uploads/2017/11/BAU_Mod1_raster-sm.pdf.
National Surface Transportation Infrastructure Financing Commission, 2009. Paying Our
strategies, data requirements, and levels of public acceptance. Way: a New Framework for Transportation Finance. Washington, DC.
Nelson\Nygaard, 2014. District Department of Transportation Curbside Management
Acknowledgments Study. Retrieved from. https://comp.ddot.dc.gov/Documents/District
%20Department%20of%20Transportation%20Curbside%20Management%20Study.
pdf.
The authors appreciated collaborating with the Atlanta Regional Nourinejad, M., Bahrami, S., Roorda, M.J., 2018. Designing parking facilities for auton-
Commission, especially Kyeil Kim, Guy Rousseau, and Abby Marinelli, omous vehicles. Transp. Res. Part B Methodol. 109, 110–127.
Segarra, L., 2017. Uber has made an insane number of worldwide trips. Fortune.
who provided cleaned and organized ABM and Four-step model outputs
Shoup, D.C., 2005. The High Cost of Free Parking Vol. 7 Planners Press, American
in the format of comma separated value (.csv) for the years 2015, 2020, Planning Association Washington, DC, USA.
2030, and 2040. Small, A., 2018. How Chicago Should Spend Its Uber Tax. Retrieved January 14, 2019,
from. https://www.citylab.com/transportation/2017/11/how-should-chicago-
spend-its-uber-tax/546233/.
References Wang, B., Ordoñez Medina, S.A., Fourie, P.J., 2018a. Operator and User Perspectives on
Fleet Mix, Parking Strategy and Drop-off Bay Size for Autonomous Transit on
AAA, 2016. Your Driving Cost: How Much Are You Really Paying to Drive? (2016 Demand.
Edition). Retrieved from. http://exchange.aaa.com/wp-content/uploads/2016/04/ Wang, K., Xie, W., Zhang, W., 2018b. Parking space optimization in the era of private
2016-YDC-Brochure.pdf. automated vehicles. Transportation Research Board 95th Annual Meeting.
Albright, J., Bell, A., Schneider, J., Nyce, C., 2016. Marketplace of Change: Automobile Xinhua, 2018. DiDi Completes 7.43b Rides in 2017. ChinaDaily.
Insurance in the Era of Autonomous Vehicles. KPMG.Marketplace of Change: Yan, X., Levine, J., Zhao, X., 2018. Integrating ridesourcing services with public transit:
Automobile Insurance in the Era of Autonomous Vehicles. KPMG. An evaluation of traveler responses combining revealed and stated preference data.
Bansal, P., Kockelman, K.M., 2017. Forecasting Americans’ long-term adoption of con- Transp. Res. Part C Emerg. Technol.
nected and autonomous vehicle technologies. Transp. Res. Part A Policy Pract. 95, Zhang, W., Guhathakurta, S., 2017. Parking spaces in the age of shared autonomous
49–63. vehicles: how much parking will we need and where? Transp. Res. Rec. J. Transp.
Barclays, 2016. Disruptive Mobility: A Scenario for 2040. Barclays Research. Res. Board (2651), 80–91.
Ben-Joseph, E., 2012. Rethinking a Lot: the Design and Culture of Parking. MIT press Zhang, W., Guhathakurta, S., Fang, J., Zhang, G., 2015a. Exploring the impact of shared
Cambridge, MA. autonomous vehicles on urban parking demand: an agent-based simulation approach.
Bridges, R., 2015. Driverless Car Revolution: Buy Mobility, Not Metal. Retrieved From Sustain. Cities Soc. 19, 34–45.
Amazon.cOm. Zhang, W., Guhathakurta, S., Fang, J., Zhang, G., 2015b. The performance and benefits of
Burns, L.D., Jordan, W.C., Scarborough, B.A., 2013. Transforming Personal Mobility. a shared Autonomous vehicles based dynamic ridesharing system: an agent-based
Earth Island Institute. Columbia University. simulation approach. Transportation Research Board 94th Annual Meeting
CAP, 2014. Parking Today: Downtown Atlanta Parking Assessment Existing Conditions. Proceeding.
Retrieved from. http://www.atlantadowntown.com/_files/docs/existing-conditions- Zhang, W., Guhathakurta, S., Khalil, E.B., 2018. The impact of private autonomous ve-
download.pdf. hicles on vehicle ownership and unoccupied VMT generation. Transp. Res. Part C
Chen, T.D., Kockelman, K.M., Hanna, J.P., 2016. Operations of a shared, autonomous, Emerg. Technol. 90, 156–165.

12