Resources, Conservation & Recycling 153 (2020) 104534

Contents lists available at ScienceDirect

Resources, Conservation & Recycling

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

Full length article

Quantifying greenhouse gas emissions reduction from bike share systems: a T

model considering real-world trips and transportation mode choice patterns
Zhaoyu Koua, Xi Wangb, Shun Fung (Anthony) Chiuc,d, Hua Caia,e,*
a
School of Industrial Engineering, Purdue University, West Lafayette, IN, 47907, United States
b
IHS Markit, Southfield, MI, 48076, United States
c
Industrial & Systems Engineering Dept, De La Salle University, Manila, Philippines
d
Institute of Innovation and Circular Economy, Asia University, Taichung City, Taiwan
e
Environmental and Ecological Engineering, Purdue University, West Lafayette, IN, 47907, United States

ARTICLE INFO ABSTRACT

Keywords: The emerging bike share systems provide convenient mobility to short-distance travelers for both leisure and
Bike share commuting purposes. Many cities are rolling out bike share programs. However, few studies have evaluated how
Environmental benefit bike share systems (BSS) are used to quantify their sustainability impacts. This study proposes a Bike Share
Mode replacement Emission Reduction Estimation Model (BS-EREM) to quantify the environmental benefits from bike share trips
Greenhouse gas reduction
and compare the greenhouse gas (GHG) emission reductions from BSS in eight cities in the United States, in-
Shared economy
cluding New York, Chicago, Boston, Philadelphia, Washington D.C., Los Angeles, San Francisco, and Seattle. The
BS-EREM model stochastically estimates the transportation modes substituted by bike share trips, considering
factors such as trip distance, trip purpose, trip start time, the accessibility of public transits, and historical
distributions of transportation mode choices. Based on average life cycle emission factors of different trans-
portation modes, our analysis reveals that the annual GHG emission reductions contributed by the eight BSSs in
year 2016 range from 41 tons of CO2-eq (Seattle) to 5417 tons of CO2-eq (New York City), while the emission
reductions per trip range from 283 to 581 g CO2-eq. The total annual emission reduction is linearly correlated to
the number of trips, bikes, and docks. The bike share stations located in the center of a city contributed to more
total GHG emission reductions due to the high trip volumes, while the stations that are further away have higher
emission reductions on a per trip basis due to longer trips and higher car substitution rate.

1. Introduction focuses on station-based bike share systems.

Evaluating the benefits from bike share systems is important for the
Bike share is becoming an increasingly popular alternative trans- cities to make decisions on supporting the development of bike share
portation mode in many countries, including the United States. The programs. Bike share can potentially bring several social and environ-
year 2017 witnessed 35 million bike share trips in the U.S., a 25% in- mental benefits such as saving transportation time (faster than walking
crease since 2016 (NACTO, 2017). There are mainly two types of bike and even driving in highly congested areas) and expenses, alleviating
share systems (BSS): station-based systems and dockless systems. In a traffic congestion, reducing greenhouse gas (GHG) emissions and air
station-based system, the users check out a bike at a docking station and pollutants, and improving multimodal transport connections (Shaheen
then return it to the same or another station after the trip. In a dockless et al., 2010; Drynda, 2014). Jäppinen et al. (2013) evaluated the spatial
bike share system, because all the necessary electronic devices are in- impacts of a hypothetical BSS on public transit travel time and esti-
corporated into the bikes instead of in the docks, the users can locate mated that bike share can reduce public transit travel time by an
and unlock a bike with a smartphone application and then park and average of 10% in the Greater Helsinki area in Finland. Faghih-Imani
lock the bike without the restriction of docking stations. Currently, et al. (2017) analyzed bike share trips and taxi trips in New York City
station-based bike share systems are still the dominant players, despite and found that, for short trips (less than 3 km), traveling by bike share
the large number of bikes launched by dockless bike share systems. In is either as fast as or faster than traveling by a taxi. Bullock et al. (2017)
2017, 56% of the bikes and 96% of the bike share trips in the U.S. are conducted a survey of bike share users in Dublin, Ireland and showed
from station-based systems (NACTO, 2017). Therefore, this study that the BSS contributed to the urban economy because the journey

⁎
Corresponding author at: 315N. Grant Street, West Lafayette, IN, 47907, United States.
E-mail address: [email protected] (H. Cai).

https://doi.org/10.1016/j.resconrec.2019.104534
Received 27 May 2019; Received in revised form 27 August 2019; Accepted 1 October 2019
Available online 02 November 2019
0921-3449/ © 2019 Elsevier B.V. All rights reserved.
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

time savings and improved connectivity from bike share increased the geospatial distribution of the modal shift due to bike share and con-
productivity of local economic activities. They found that these benefits cluded that higher proportion of bike share trips replaced public transit
are more significant than those of modal shift from cars to bicycles. in urban periphery than in urban areas with high population density.
Using causal inference models, Hamilton and Wichman (2018) proved The analysis from Nair et al. (2013) shows that bike share stations
that the introduction of a bike share system reduced traffic congestion closer to public transit facilities often exhibit higher utilization.
by about 4% in Washington. DC. Fuller et al. (2013) analyzed the data Shaheen et al. (2013) analyzed the modal shift of bike share by con-
from a telephone survey about people’s biking behaviors and concluded ducting user surveys for BSS in four cities: Montreal, Toronto, Twin
that there is a higher likelihood of cycling for people who live in areas Cities, and Washington, D.C. The survey results showed that bike share
where bike share is available, compared to people who have privately users shifted from all transportation modes and the modal shifts vary in
owned bicycles but have no access to bike share programs. Buehler and different cities. About 40% to 50% of the respondents reported that
Hamre carried out a survey of bike share users and local businesses near their use of bike share reduces their bus and rail usage in Montreal,
the bike share stations in Washington DC, U.S. (Buehler and Hamre, Toronto and Washington D.C., while in Twin Cities, 15% of the re-
2015). They found that 73% of the users chose bike share to save travel spondents reported increasing rail usage. In all four cities, bike share
time, while 25% of the users were motivated by cost savings. was found to reduce car and taxi usage. However, such survey results
Among the abovementioned benefits, this study focuses on the en- only provide qualitative information but not quantitative data for the
vironmental benefit, which is one of the major drivers that motivate the changes (e.g., it shows the trend on whether bike sharing increases/
decision makers to introduce BSS into the cities. Understanding the decreases the use of other modes but does not quantify the changes).
modal shift is important to evaluate the net environmental benefits of Similarly, another study also reveals that bike share reduced the usage
BSS. Whether bike share reduces transportation energy use depends on of personal automobiles among about 50% of the bike share users in
the transportation mode replaced by the bike share trips. If bike share five cities: Twin Cities, Salt Lake City, Montreal, Toronto, and Mexico
trips replace personal vehicle or taxi trips, the bike share systems will City (Susan et al., 2014). Two surveys were conducted for the BSS in
reduce transportation fuel use and emissions. On the other hand, Washington, DC, separately targeting the subscription members (users
however, if the bike sharing trips replace walking trips, they may ac- who purchase monthly/annual plans) (LDA Consulting and Capital
tually increase transportation energy use, due to the energy consump- Bikeshare, 2016) or casual users (those who only used the service with
tion required to build the stations, manufacture the bikes, and operate one-day or five-day plans) (Borecki et al., 2012). The survey results
the system (e.g., transporting the bikes between stations to rebalance show that 39% of the members and 53% of the casual users who re-
the bike availability). Therefore, in order to estimate the environmental sponded to the surveys used bike share to replace walking trips for their
benefits of bike share systems, it is critical to consider the transporta- last bike share trips.
tion mode substitution by bike share trips. In existing literatures, there Few studies quantified the environmental benefits of bike share
are limited studies trying to quantify the environmental benefits of BSS, programs using mode substitution information. The research of
in which the mode substitution is either based on simplified assump- Fishman et al. (2014) was one of the studies that first analyzed the bike
tions (Zhang and Mi, 2018; Anderson, 2015) or shifted concepts (e.g., share benefits with consideration of mode substitution. They used the
using the percentage of users information from bike share user surveys bike share user survey data to estimate the mode substitution. In these
as the percentage of travelled miles values) (Fishman et al., 2014). The user surveys carried out by the BSS operators, the BSS users were asked
environmental benefits estimated using these methods may lead to about their alternative travel modes, if not using bike share, to calculate
biased result (more details are provided in the Literature Review part of the aggregated substitution rate (percentage) of different transportation
this paper). Therefore, better methods that consider the heterogeneity modes. For instance, if 40% of the survey participants chose vehicles as
of bike share trips are needed. their alternative mode, it is assumed that 40% of the bike share trips
In this study, we proposed a Bike Share Emission Reduction substituted vehicle trips. Based on the substitution rates, the authors
Estimation Model (BS-EREM), which quantifies the greenhouse gas estimated and compared the car travel distance reduction by BSS in
(GHG) emission reductions from BSS with consideration of the detailed Melbourne, Brisbane, Washington, D.C., Minnesota, and London. The
bike share trip information, such as trip distance, trip purpose, and time substituted car travel at each BSS is calculated as the car substitution
of the day for the trips. Mode substitution of bike share trips is simu- rate multiplied by the total annual trip distance, which is estimated
lated based on historical travel patterns learned from trip data in using the bike share trip duration and an assumed biking speed of
National Household Travel Survey (NHTS). The proposed model is then 12 km/h. However, this approach assumes that 1) the transportation
applied to compare the GHG emission reductions of eight bike share mode substitution is independent of the trip distance and trip purpose,
systems in the United States, including New York, Chicago, Boston, 2) the percentage of trip substitution is equivalent to the percentage of
Philadelphia, Washington D.C., Los Angeles, San Francisco, and Seattle. travel distance substitution, and 3) the trip distance is linearly corre-
The rest of the paper is structured as follows: Section 2 discusses the lated to the trip duration. These assumptions could lead to biased re-
relevant literature, research gaps, and the contributions made by this sults in estimating the environmental benefits of bike share programs.
study; Section 3 provides detailed information about the input data and First, trip distance and purpose do affect people’s transportation mode
methodology of the proposed BS-EREM; In Section 4, the overall and selection. For instance, for long-distance trips, people are more likely to
unit (such as per trip and per bike) GHG emission reduction, as well as prefer car or public transit to walking. Additionally, the transportation
the spatial patterns of different bike share stations’ contributions to mode choice would be different if the trip is for sightseeing instead of
GHG emission reduction are analyzed by applying the proposed model commuting (USDOT, 2009). The user survey data doesn’t include in-
to real-world trip data; a sensitivity analysis is carried out in Section 5 formation for trip distance and purpose. Therefore, assuming that all
to discuss the impacts of key parameters and assumptions of the BS- the bike share trips follow the same mode substitution rates could lead
EREM; lastly, Section 6 summarizes the findings, then discusses the to inaccurate results. Second, because the travel distance from trip to
limitations of this work and future research directions. trip could vary significantly, the percentage of trip substitution (in-
formation asked in the survey) is not equivalent to the percentage of
2. Literature review travel distance substitution (information the calculation is based on).
Third, while the trip distance is linearly related to the trip duration for
In order to estimate the replaced modes for bike share trips, re- some trips (e.g., when the riders go direct from point A to point B, such
searchers have tried to collect modal shift information through user as commuting), the distance of trips with detours or multiple stops (e.g.,
surveys. Martin and Shaheen (2014) collected bike share user survey leisure trips) cannot be accurately estimated using the trip duration
data in Washington, D.C. and Minneapolis. They analyzed the (Kou and Cai, 2019). Therefore, to better understand the transportation

2
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

emission reduction contributed by bike share programs, the analysis distribution of the environmental benefits at the bike share station
needs to consider transportation mode substitution integratedly with level, which provides insights for BSS station siting and planning.
trip distance and trip purpose. In another study which estimated the Results from this study can inform decision makers, city planners, and
environmental benefits of the BSS in Shanghai, China, Zhang and Mi BSS operators to better develop, deploy, and operate BSS programs to
(2018) set a distance threshold of 1 km to divide the bike share trips improve transportation sustainability.
into two groups: trips that are shorter than 1 km, which were assumed
as not replacing car travel and therefore generating zero environmental 3. Data and methods
benefits, and trips that are longer than 1 km, which were assumed as
replacing car trips and providing environmental benefits. The en- 3.1. Input data
vironmental benefits from these trips were calculated by multiplying
the total trip distance with impact factors (e.g., energy saving per km or The proposed BS-EREM model integrates data from multiple sources
emission reduction per km). Although this study considered trip dis- to estimate the environmental benefits of BSS. The input data includes
tance in the analysis, the 1 km distance threshold is set arbitrarily. Trips (1) bike share trip data (Section 3.1.1), (2) National Household Travel
that are less than 1 km could also replace a car trip while trips longer Survey (NHTS) data (Section 3.1.2), which provides historical trans-
than 1 km could also substitute public transit trips or even walking portation mode choice information prior to launching BSS programs,
trips. Anderson (2015) assessed a proposed bike share program in (3) public transit stations/stops near bike share stations (checking
Portland, Maine in the United States, and concluded that the potential whether it is possible for a bike share trip to substitute a public transit
bike share program can improve air quality and rates of physical ac- trip, Section 3.1.3), and (4) GHG emission factors used for the calcu-
tivity for bike share users. The analysis is based on several assumed lation of trip environmental benefits given different mode substitution
parameters such as annual trips, average trip distance, and average by BSS trips (Section 3.1.4). This section explains each type of data
minutes of physical activity per trip. The estimated benefits still need to while the BS-EREM model is presented in Section 3.2.
be further verified with real-world data.
In summary, existing studies quantifying the environmental benefits 3.1.1. Bike share trips in eight U.S. cities
of BSS have two major limitations. First, the calculation of emission Our study analyzed the bike share trip data from programs located
reduction from bike share trips is based on simplified assumptions (for in eight cities: Seattle (Pronto Cycle Share), Los Angeles (Metro Bike
example, “all bike share trips longer than 1 km will replace car trips” Share), Bay Area (Ford Gobike), Philadelphia (Indego Bike Share),
(Zhang and Mi, 2018)). Second, existing studies often replace the Boston (Hubway, now rebranded as Blue Bikes), Washington D.C.
concept of “percent of miles” with “percent of users”, when mode (Capital Bike Share), Chicago (Divvy), and New York City (Citi Bike
substitution data from user surveys are used in the calculation. While Share). To compare all eight BSSs, we used the trip data in year 2016,
user surveys can provide useful information about mode substitution, because Seattle’s Pronto Cycle Share was closed after March 2017. All
existing user surveys only reported the share of respondents claiming eight systems are station-based. These bike share programs have made
that they use bike share to replace a certain mode. However, not all their trip data publically available, which include the timestamps and
users take the same number of bike share trips and not all trips have the locations of the origin and destination of each trip. Because neither trip
same length. It could lead to biased results if we directly applied such distance nor detailed bike trajectory data is provided, we estimated the
“percent of users” information as if they were “percent of miles” data to trip distance between each pair of origin and destination stations using
quantify the environmental benefits. A user who take more trips and Google Maps Distance Matrix API, which could more accurately esti-
longer trips should be given a higher weight. Ideally, the historical bike mate the distance travelled by bike on urban street networks. Many of
share trips made by each survey participant should be used to link the previous studies used trip displacement (e.g., the Great Circle Distance)
detailed trip information with mode substitution. However, because the to measure human movements (Rhee et al., 2011; González et al.,
trip data and the user survey data are collected separately and anon- 2008). However, in an urban transportation system, the movement of
ymously, such linked data is not available at a large scale. Additionally, bikes is restricted by the street networks. The distance estimated by the
because user surveys are self-reported data, the accuracy of the data is API can better reflect the bike share trip along the pathways suitable for
always a concern (Cohen and Shaheen, 2018). Therefore, better mod- biking. Comparing to driving, bike travels are less affected by conges-
eling approaches are needed to evaluate the environmental benefits of tion (Alter, 2008; Lobo, 2013). Therefore, in this study, we assumed
BSS using real-world data and holistically evaluate the transportation that the travel distance and duration between the same pair of start and
mode substitution by bike share trips, considering detailed trip in- end stations stay constant in spite of the trip start time. Round trips,
formation such as trip distance, purpose, and start time. which have the same origin and destination station, are excluded in our
To address the above discussed gaps, this study proposes a Bike analysis due to the difficulty in estimating trip distance. Such round
Share Emission Reduction Estimation Model (BS-EREM) to quantify the trips only make up 2% to 10% of the total trips in the eight systems. For
greenhouse gas emission reductions by bike share trips through prob- the overall emission reduction estimation, we scaled up the results
abilistically simulating bike share trips’ mode substitution based on trip based on the total count of trips, assuming that these round trips share
distance, trip purpose, time of the day for the trip, and historical travel the same pattern as the one-way trips. Because this is an arbitrary as-
patterns (mode choice distributions) before launching bike share pro- sumption, we have also reported the emission reduction values ex-
grams. We applied the model to compare the life cycle greenhouse gas cluding the round trips.
emission reduction from bike share programs in eight U.S. cities and
also evaluated the spatial patterns of reduced GHG emissions at the 3.1.2. National household travel survey (NHTS)
station level in different cities. Compared to the existing literature, the The National Household Travel Survey (NHTS) (USDOT, 2009) is a
unique contributions made by this study are: 1) we proposed a model to nation-wide survey conducted by the U.S. Federal Highway Adminis-
estimate the mode substitution of bike share trips, with the considera- tration, which records the personal and household travel behaviors of
tion of trip distance, trip purpose, trip start time, and the public transit local residents in the United States. Although more recent NHTS data
accessibility near bike share stations; 2) the mode substitution estima- has become available, in this study, we used the 2009 NHTS data to
tion is based on travel survey data and real-world bike share trip data, generate the mode substitution rate distributions, because it could
thus, the proposed method can be generally applied to all cities that better reflect people’s transportation mode choice when bike share
have such data available; and 3) the analysis not only quantifies the service was not available (most of the BSSs in the United States were
overall environmental benefits at the system level, but also evaluated established after the year 2010). There are near 1.05 million trips re-
the environmental benefits for unit distance travelled and the spatial corded in NHTS 2009 data, which contains the detailed trip information

3
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Fig. 1. Distributions of transportation mode choice regards to different trip distance for commuting trips and leisure trips.

and the characteristics of the people who are travelling. We extracted (typically with a population density of more than 1000 persons per
the attributes such as transportation modes, trip purpose, travel time, square mile) (USDOT, 2009) for each of the eight cities. With such in-
and travelled distance (in miles) of those trips taken in urban areas formation, we can develop the distributions of people’s transportation

4
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Table 1
Emission factors used for the GHG emission reduction calculation.
Mode Sub-category Emission factor (g Miles Weighted
CO2-eq/passenger travelled in average of
mile travelled) 2017 NHTS the
(Dave, 2010) (U.S.DOT, emission
2017) factor

Walking Walking 33 33
Bicycling Bicycling 33 33
Vehicle Sedan 382 3,843,053 441
SUV 446 2,278,726
Pickup 619 1,230,229
Public transit Bus Average 326 170,342 299
BART1 136 Average of 39,207
MUNI1 173 three light
Green Line1 224 rail/
subway:
177.7

Fig. 2. Share of transportation mode choice in different time of the day in Los Note: 1 BART: Bay Area Rapid Transit System; MUNI: San Francisco Municipal
Angeles. Railway; Green Line: subway system in Boston; since we do not have the
emission factors for all the light rail/subway systems in the eight cities, we used
the average emission factors of BART, MUNI and Green Line to represent the
mode choice for a specific trip distance, travel time, and purpose. We
unit emission of light rail/subway.
only used NHTS records in the urban area because the bike share sta-
tions analyzed in this study are all located in urban areas as defined in
emission factor calculation). We adopted the emission factors from this
the 2016 Topologically Integrated Geographic Encoding and Referen-
study because this study evaluated the emission factors of different
cing system (TIGER) data from the U.S. Census Bureau (Thompson
transportation modes based on the same analysis scope and assump-
et al., 2016), which is also the data source that NHTS used to define
tions, making the emission factors more comparable to each other than
urban areas (USDOT, 2009). In the case that a bike share system/station
factors drawn from different studies which may be generated based on
is located in rural area, the mode distribution should be generated using
different assumptions. Because the emission factors developed in this
the NHTS trips from rural areas. Fig. 1 shows the mode choice dis-
study include multiple types of public transit (e.g., bus and subway) and
tributions from the NHTS trip data for all time periods of the day with
vehicles (e.g., car, SUV, and pickup), we take the weighted average of
varied trip distance for commuting and leisure trips in each city. These
the emission factors in each transportation mode category to simplify
distributions show that people are more likely to choose walking for
our analysis to include only four transportation modes: walking, bicy-
short-distance trips (e.g., less than 2 miles) than for long-distance trips
cling, vehicle, and public transit (Table 1). The weighting factors are
(e.g., 10 miles). For the same distance range, less people will choose
the corresponding miles travelled using each mode in the NHTS 2017
public transit for leisure trips than for commuting trips. Fig. 2 depicts
data (here we used the 2017 data because it can better reflect the
the share of different transportation modes in different periods of the
distance travelled by different modes in year 2016; NHTS 2017 and
day in Los Angeles. The 24 h of a day are separated into six time in-
2009 are the two most recent NHTS survey datasets available). Because
tervals. We group 21:00 to 5:59 as one period because the number of
it is difficult to find more fine-grained emission factor data for all the
trips during this period is relatively low. For other time of the day, a
analyzed cities, we use the weighted average emission factors described
three-hour interval is used. The mode share varies within a day, espe-
above as the emission factors for the analysis of all eight cities.
cially in the morning and evening. For example, during the period be-
The difference between the life cycle emission factors of the sub-
tween 6 a.m. and 8:59 am, higher proportion of trips are taken by
stituted transportation mode and biking is then used to calculate the
walking and public transit than other periods such as from 9 a.m. to
GHG emission reduction. For example, if a bike share trip substituted a
5:59 pm. These observations demonstrate the need to include trip dis-
vehicle trip, the emission reduction for this trip would be 408 g CO2-eq/
tance, purpose, and time of the day in the analysis of trip mode sub-
mile (biking generates 408 g CO2-eq less than vehicles per passenger
stitution. Therefore, for each city, a total of 12 distributions are de-
mile travelled) multiplied by the trip distance (in miles). How to de-
veloped (2 trip purposes× 6 time intervals).
termine the substituted transportation mode will be discussed in
Section 3.2.
3.1.3. Public transit accessibility
Another factor that could affect people’s mode choice is the acces- 3.2. Bike share emission reduction estimation model (BS-EREM)
sibility of public transit. If there’s no public transit facilities (e.g., bus
stops, subway stations) near the origin and destination stations of a trip, Using the above mentioned data as inputs, we propose a Bike Share
it is less possible that the bike share trip replaced public transit. Using Emission Reduction Estimation Model (BS-EREM) to estimate the GHG
Google Maps Places API, we counted the number of public transit fa- emissions reduction from a bike share trip (Fig. 3). The BS-EREM model
cilities within a 200-m (0.12 miles, about 3–4 minutes’ walk) buffer includes two major components: bike share trip purpose estimation and
zone around each bike share station in each of the eight cities. The bike share trip mode substitution simulation.
impact of choosing 200 m as the public transit accessible zone is tested From the NHTS data, the historical trips are first separated into six
in the sensitivity analysis (Section 5). The details on how we use the time groups as specified in Section 3.1.2 based on the trip start time.
data are described in Section 3.2. Then for each time group, we developed the historical mode choice
distributions for commuting and leisure trips separately. To differ-
3.1.4. GHG emission factors entiate between commuting and leisure trips, we compare the trip
We use life cycle GHG emission factors from Dave (2010) to cal- speed to the average speed of all trips in this city as a way to infer trip
culate the emission reduction associated with each bike share trip (note purpose. Trips whose speeds are lower than the average speed are
that, in the life cycle analysis, emissions from the upstream processes considered as leisure trips (e.g., stop at different locations, take detours
besides the trips, such as road constructions, are also allocated in the to visit different places-of-interest, and not in a hurry to get from the

5
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Fig. 3. Overview of the BS-EREM.

origins to the destinations), while the others are considered as com- MT, respectively (Report, 2017). The changes for 2010 and 2015
muting trips (e.g., directly going from the origins to the destinations). compared to 2005 are only −3% and −2%, respectively.
The impact of using the average speed as the threshold for trip purpose
estimation is tested in the sensitivity analysis (Section 5). Given the
4. Results and discussions
distance of each bike share trip, the estimated trip purpose, time period
of the day, and the corresponding historical distribution of mode choice
To understand the GHG emission reduction contributions of BSS in
regarding the specific trip distance, purpose, and time period, we can
different cities, we analyzed the modeling results from two perspec-
obtain the probability of the bike share trip replacing vehicles, walking,
tives: the total GHG emission reduction from a BSS in each city and the
or public transit to determine the replaced mode in the simulation
unit emission reduction by trip, dock, bike, and mile travelled (Section
process. Using the mode choice distributions from NHTS data to si-
4.1), as well as the spatial distribution of the reduced GHG emissions
mulate the displaced mode by bike share assumes that an average BSS
(Section 4.2).
user is the same as an average person in the city. It is possible that BSS
users are a self-selected group which has different travel patterns (e.g.,
having lower percentage of vehicle trips than an average person). The 4.1. Overall environmental benefits of the BSS
influence of this assumption is tested in the sensitivity analysis (Section
5). If there’s no public transit facility available near the location of the Table 3 listed the basic statistics and modeling results of the bike
bike share trip’s origin or destination, it is unlikely that this bike share share trips for each city. The BSS in New York, Chicago, Boston, and
trip replaces public transit. In this case, the mode share of public transit Washington, DC are the four largest station-based systems in the United
will be set to zero. The pseudo code describing the detailed algorithm to States (NACTO, 2017). In year 2016, these systems generally have more
determine the replaced mode in the simulation is provided in the than 150 stations and generated more than one million bike share trips.
Appendix A. After determining the replaced mode, we can then calcu- These larger systems also have a greater system diameter (indicating
late the GHG emission reduction from the trip using the emission fac- spatial coverage), which is defined as the longest Euclidian distance
tors and trip distance. We used the trips in the entire year of 2016 for between any two of the bike share stations in a BSS. The total GHG
our simulation and compared the results among different cities. emission reduction in 2016 in these four larger systems are much higher
than those in the other four smaller systems (Fig. 4a). New York city has
the largest amount of total GHG emissions reduction: with 10,262,649
3.3. Evaluating the emission reduction from BSS in the context of the bike share trips taken in 2016, New York’s BSS contributed to 5417 tons
transportation sector emissions of GHG emission reduction (in CO2-eq). In contrast, Seattle’s BSS only
reduced 41 tons of GHG emissions. One thing to note is that, we esti-
After obtaining the total GHG emission reduction using BS-EREM, mate the emission reduction of round trips based on the assumption
we compared the results with the total GHG emissions from the trans- that these trips have the same emission reduction per trip. Take New
portation sector in these cities. This reflects the BSS’s relative GHG York as an example, excluding the round trips would change the total
emission reduction contribution to its local transportation systems. The GHG emission reduction to 5305 tons CO2-eq. Although the round trips
data of GHG emission from the transportation sector is collected from only take up a very small portion (2% in New York), this part of esti-
the most recent greenhouse gas inventory report of these cities. The mation could be improved if the detailed trajectory data is available.
cities do not release such data every year but the data from adjacent The emission reduction per trip ranges from 280 to 590 g CO2-eq
years can reflect the general level of transportation sector's GHG (Table 3). The total GHG emission reduction is found to correlate lin-
emissions in these cities (Table 2). The transportation sector’s GHG early to the ridership (total count of trips) in each city with an R
emissions have been staying relatively stable based on historical data squared of 0.998 (Fig. 4b). The linear regression also shows that the
(for instance, the transportation sector’s GHG emissions for City of marginal emission reduction of one additional trip is 533 g CO2-eq.
Chicago in the year 2005, 2010, 2015 are 8.20, 7.95, and 8.05 million Similar linear relationship is also found between the total GHG emission

6
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Table 2
GHG emissions from the transportation sector in the eight evaluated cities.
City GHG emissions from transportation sector (million MT Year of the Data Source
CO2-eq/year) data

Seattle 3.58 2014 2014 Seattle Community Greenhouse Gas Emissions Inventory (E.P and Kevin,
2014)
Los Angeles 33.74 2010 2015 Environmental Report Card for Los Angeles (Gold et al., 2015)
San Francisco 1.98 2016 2016 San Francisco Geographic Greenhouse Gas Emissions Inventory at a Glance
(Program, 2015)
Philadelphia 3.96 2012 Philadelphia Citywide Greenhouse Gas Inventory, 2012 (Spring, 2015)
Boston 1.64 2013 City of Boston Community Greenhouse Gas Inventory 2005-2013 (Boston, 2014)
Washington DC 1.74 2013 District of Columbia Greenhouse Gas Inventory Update 2012-2013 (I. Of, 2012)
Chicago 8.05 2015 City of Chicago Greenhouse Gas Inventory Report, 2015 (Report, 2017)
New York 15.48 2015 Inventory of New York City Greenhouse Gas Emission in 2015 (Pasion et al.,
2017)

Table 3
Basic statistics and analysis results of the bike share systems.
City Seattle Los Angeles Bay Area Philadelphia Boston Washington D.C. Chicago New York

Total trips in 2016 102,606 184,345 193,506 499,306 1,236,199 2,562,718 3,595,383 10,262,649
Total stations 59 64 74 119 327 407 581 687
Total docksa 1,038 1,352 1,357 2,280 5,729 6,720 9,987 20,390
Docks per stationa 17.59 21.12 18.34 19.16 17.52 16.51 17.19 29.68
Percentage of stations with public transit access 57.63% 67.19% 93.24% 100.00% 96.94% 71.01% 98.28% 57.79%
Percentage of trips used for simulationb 91.64% 89.40% 97.58% 91.90% 96.87% 95.36% 95.73% 97.94%
NHTS trips used for simulationc 1,619 49,427 29,284 3,350 3,293 14,199 6,096 40,631
Total number of bikesd 463 763 421 1,023 1,797 4,305 5,746 10,486
System diameter (miles)e 4.71 2.85 2.33 5.06 8.57 14.28 23.29 11.17
Average number of trips per station 1,739 2,880 2,615 4,196 3,780 6,297 6,188 14,938
Average number of trips per dock 99 136 143 219 216 381 360 503
Average number of trips per bike 222 242 460 488 688 595 626 979
Median of trip distance (miles) 1.16 1.2 1.39 1.46 1.43 0.91 1.39 1.31
Average trip distance (miles) 1.27 1.23 1.56 1.7 1.72 1.02 1.71 1.68
Average speed of bike share trips (miles/hour) 6.87 6.24 7.97 7.28 8.21 7.55 7.71 7.92
Percentage of commuting trips 51.92% 48.45% 52.30% 54.37% 52.57% 52.85% 52.88% 46.95%
Total GHG emission reduction (without round trips, ton CO2-eq) 37.21 46.56 67.43 234.83 668.11 1,275.88 2,000.89 5305.13
Total GHG emission reduction (including round trips, ton CO2-eq)f 40.60 52.08 69.10 255.51 689.66 1,338.00 2,090.05 5,416.68
Percentage of GHG emission reduction from commuting trips 57.70% 53.93% 60.89% 60.50% 60.14% 61.49% 60.93% 57.16%
(without round trips)
Average emission reduction (g CO2-eq) per mile travelled 324.57 286.59 298.75 340.89 345.02 329.69 352.66 329.48
Average emission reduction (g CO2-eq) per trip 395.73 282.50 357.12 511.73 557.89 522.10 581.32 527.81
Average emission reduction (ton CO2-eq) per station 0.69 0.81 0.93 2.15 2.11 3.29 3.60 7.88
Average emission reduction (ton CO2-eq) per dock 0.04 0.04 0.05 0.11 0.12 0.20 0.21 0.27
Average GHG emission reduction (ton CO2-eq) per bike 0.09 0.07 0.17 0.26 0.35 0.28 0.37 0.49

Notes:
a. For Seattle, Bay Area, Boston, and Chicago, the station capacity (number of docks of each station) information is extracted from the station information data (year
2016) released by the system operator; while such data is not available for other four cities: for Los Angeles, Philadelphia, and New York, the station capacity
information is obtained from a station status snapshot (real-time station status) in Oct. 15, 2017 from the operators’ website; for Washington, DC, the station capacity
information is extracted from the bike share station information (Aug. 20, 2018) in Open Data of Washington, DC government1 . Although the information for the
latter four cities mentioned is not exact the station information for year 2016, it can reflect the general level of station capacity of these systems.
b. Round trips are excluded in the simulation.
c. Only NHTS survey records in the urban area of the corresponding city were used in our analysis.
d. All bikes that showed up in the trip data in year 2016.
e. The system diameter is the longest Euclidian distance between any two bike share stations of a BSS at the end of year 2016. Data is collected from (Kou and Cai,
2019).
f. Emission reduction for those round trips is calculated by number of trips multiplied by emission reduction per trip of the those one-way trips.

reduction and the total number of bikes (R squared: 0.970, Fig. 4c). 0.1 ton CO2-eq per bike) and per-dock (less than 0.1 ton CO2-eq per
Each bike can contribute to an average of 0.07 to 0.49 tons of GHG dock) emission reduction than those larger systems (more than 0.2 ton
emission reduction depending on the system (Table 3), while the CO2-eq per bike and 0.25 ton CO2-eq per dock). The bikes and docks are
marginal emission reduction of adding one more bike into the system is utilized less efficiently in these smaller systems. Each bike/dock in
0.51 tons CO2-eq in a year. In the same way, we can observe a linear these smaller systems on average served less trips compared with that in
relationship between the total GHG emission and the total number of larger systems, as shown in Table 3. For example, each bike and dock in
docks (R squared: 0.984, Fig. 4d). The marginal emission reduction of Seattle on average served 99 trips and 222 trips in 2016, respectively;
adding one more dock into the system is 0.28 tons CO2-eq in a year. while the trips per bike and trips per dock could be as high as 503 trips
Also note that in the fitted linear models, the marginal emission re-
duction (gradient) is mainly influenced by the four larger systems. For
the smaller systems such as Seattle and Los Angeles which have less 1
Capital Bike Share Locations, Open Data of Washington, DC, http://
bikes and docks, they also exhibit much lower per-bike (less than opendata.dc.gov/datasets/capital-bike-share-locations.

7
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Fig. 4. Total GHG emission reduction in 2016 of the eight systems (a) and its relationship with (b) annual ridership (count of trips), (c) total number of bikes, and (d)
total number of docks.

and 979 trips, respectively in New York. The reasons for this need to be shown in Table 3, the average and median of bike share trip distance
further investigated in future research. Regarding trip purpose, com- are both less than 2 miles for all the cities. However, in larger systems
muting trips take up about half of the total trips in the eight cities, which have a larger spatial coverage of bike share stations, higher
ranging from 47% to 54% (Table 3). However, the GHG emissions re- proportion of emission reduction was from trips within 2–5 miles in-
duction from commuting trips takes up a higher share (54%–61%) than stead of 0–2 miles, compared to those smaller systems. Therefore, the
that from leisure trips, due to combined reasons of travel distance and bike share station network plays an important role in affecting the
mode share distribution. Take Los Angeles as an example, the average users’ travel pattern and the BSS’s ability to reduce GHG emissions.
distance of commuting trips is 1.08 miles, while that of leisure trips is
only 0.90 miles. On the other hand, higher proportion (64.3%) of
4.2. Spatial distributions of GHG emission reduction
commuting bike share trips are car-replacing trips than that of leisure
trips (60.3% car-replacing trips).
To better understand how locations of the bike share stations affect
In order to understand the relative environmental benefits of BSS
the GHG emission reductions, we also analyzed the spatial patterns of
contributed to these cities, we also compared the emission reduction
station level emission reduction (to save the computation time, the
from BSS to the total emission from the transportation sector in each
analysis in this part is based on the trip data in August 2016 instead of
city (Fig. 5a). Generally, the emission reduction from BSS only makes
the entire year’s data). This information could potentially help the de-
up a small part of the total GHG emission from the transportation
cision making in siting future bike share stations. The GHG emission
sector, which ranges from 0.0002% to 0.007% for the four smaller
reduction also exhibits geographic variances within a bike share
systems and 0.026% to 0.077% for the four larger systems. Among
system. We allocated the GHG emission reduction of each trip to its
these cities, Los Angeles has the lowest ratio because it has high
start station (since the origin of a trip generally reflects the demand of
transportation sector emissions but a relatively small BSS.
travel) and plotted the GHG emission reduction by station on the map.
The GHG emission reduction per mile travelled varies among dif-
In the case of Los Angeles and Philadelphia, stations located in the
ferent cities (Fig. 5b): Chicago, Boston, and Philadelphia have the best
center of the city reduced more GHG emissions (Fig. 6(a1) and (b1)),
performance, which reduced 352.7 g, 345.0 g and 340.9 g CO2-eq for
because of the larger ridership (count of trips) at these stations. The
each mile of bike share trip travelled, respectively. From Fig. 5c, we can
GHG emission reduction per trip exhibits the reverse pattern: each trip
observe that in these three cities, higher percentage of bike share trips
reduced less GHG emission if they started from the city center
replaced car trips (74% for Chicago, 75% for Boston, and 75% for
(Fig. 6(a2) and (b2)). This is because trips originated from the city
Philadelphia), which is the main reason that they have better GHG
center are mostly short and more likely to replace walking, while trips
emission reduction per mile travelled. However, the GHG emission re-
made from areas away from city center are relatively longer. Fig. 6(a3)
duction per bike share trip in Chicago is slightly higher because Chi-
and (b3) shows that higher proportion of bike share trips replaced ve-
cago’s larger bike share station network allows the users to travel longer
hicle trips in the stations that are away from the city center. Generally,
distance. Generally, the majority of the GHG emission reduction are
all the eight cities show similar spatial patterns regarding GHG emission
contributed by relatively short trips (less than 5 miles, Fig. 5d). As
reduction (see Fig. B1 in the Appendix A for the figures of the other

8
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Fig. 5. GHG emission reduction in the eight cities: (a) GHG emission reduction from BSS/Total GHG emission from the transportation sector (%), (b) emission
reduction per mile travelled and per trip, (c) percentage of trips replaced by bike share, (d) emission reduction of BSS in different distance range.

cities). These results show that, although the city center has more BSS the result variations using a buffer zone radius ranging from 140 m to
users than areas away from city center, a higher proportion of these 260 m. For the distributions of transportation mode choices, we mod-
users adopt bike share as a substitution of walking. In order to improve ified the probability of the car trips by −30% to +30% of the original
the environmental benefits from the BSS, BSS operators needs to either value, and then evenly distributed the changes to walking and public
attract more vehicle users in the city center to switch to bike share, or transit (for example, if the mode distribution for a trip at given distance
strategically locate and install more stations in the areas that are away is 60% by car, 30% by walking, and 10% by public transit, the −30%
from the center of the city. scenario will decrease the car trip probability to 42% and increase the
probability of walking and public transit to 39% and 19%, respec-
5. Sensitivity analysis tively). These scenarios will represent cases where the bike share users
take less or more vehicle trips than an average person. The focus of
In our model, there are several major factors or assumptions that these different scenarios is on vehicle trips because the vehicle trips
may impact the results: speed threshold to determine the trip purpose, dominate the potential GHG emission reduction (Fig. 5c). For the
buffer zone radius for the public transit accessibility, transportation emission factors, we used the travelled mileage as the weights to cal-
mode distribution from the NHTS data, and the emission factors used in culate the weighted average for vehicle and public transit. In this part,
our analysis. Therefore, we conducted a sensitivity analysis to evaluate instead of using the weighted average, we use the lowest value (sedan –
how sensitive the results are to these factors. For the speed threshold, 382 g CO2-eq/passenger mile travelled (PMT) for vehicle, and BART –
buffer zone radius, and transportation mode distribution, we varied the 136 g CO2-eq/PMT for public transit) and the highest value (pickup –
parameters from 70% to 130% of the values used in the original model, 619 g CO2-eq/PMT for vehicle, and Bus average – 326 g CO2-eq/PMT for
and compared the total emission reduction from the BSSs in Bay Area public transit) to calculate the emission reduction in two extreme cases
and Philadelphia (Fig. 7). For the emission factors, we used the lowest (Table 4). In addition, in order to evaluate the influence of the emis-
and the highest values for vehicle and public transit in Table 1 to sions from stations and rebalance, we add the unit emission of 95 g CO2-
generate the lower and upper bounds for the emission reduction. Ad- eq/PMT (from Luo et al., 2019) for bike share stations (57 CO2-eq/PMT,
ditionally, since the emission factor of bicycling in (Dave, 2010) does including docks) and rebalance (38 CO2-eq/PMT) to the original emis-
not consider the emissions generated from bike share station (including sion factor of 33 g CO2-eq/PMT (from (Dave (2010)) for the emissions
docks) manufacturing and bike rebalance, we also add the unit emis- from bike, road, and human breath, which results in a modified emis-
sions from bike share stations and rebalance from the work of Luo et al. sion factor of 128 g CO2-eq/PMT for bicycling using shared bikes from
(2019) to evaluate the influence from station infrastructure and op- station-based BSS. It is notable that the emission factors from Luo et al.
eration. (2019) reflect the average operational conditions of a station-based
In the original model, we used the average trip speed as the speed system in the U.S.. Because the modified bicycling emission factor is the
threshold to determine whether a trip is for commuting or leisure sum of emission factors from two studies, which may rely on different
purposes. So the sensitivity analysis tested the result variations using a data source and assumptions, we only consider emissions from station
speed threshold that ranges from 70% to 130% of the average speed in and rebalance in the sensitivity analysis. In the sensitivity analysis, we
each city (as listed in Table 3, the average speed varies by city). A buffer use the same emission factors of walking, public transit, and vehicles as
zone radius of 200 m was originally used to extract the public transit stated in Table 1.
accessibility information. Accordingly, the sensitivity analysis tested The sensitivity analysis shows that the total emission reduction is

9
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Fig. 6. Geographic distribution of GHG emission reduction and mode substitution in (a) Los Angeles and (b) Philadelphia: 1. Total emission reduction of each station
in Aug. 2016, 2. Emission reduction per trip, 3. Pie chart of mode substitution of each station.

10
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Fig. 7. Change of total emission reduction by varying parameters in the model: (a) Bay Area, (b) Philadelphia.

the most sensitive to the emission factors and mode choice distribution 6. Conclusions and limitations
change, but stays relatively stable when we change the speed threshold
and buffer zone diameter. Therefore, it is critical to obtain more ac- We proposed a Bike Share Emission Reduction Estimation Model
curate information about BSS user’s transportation choice to refine the (BS-EREM) to evaluate the environmental benefits of bike share systems
estimated environmental benefits of BSS. It is notable that, for longer in eight cities in the United States. The BS-EREM model considers the
trips (e.g., over 5 miles), driving is the dominant mode (e.g., with a trip distance, trip purpose, trip start time, public transit accessibility
probability of 90%). When we evaluate the +30% scenario, the prob- around each bike share station, as well as the historical distributions of
ability of driving would exceed 100%. In these cases, we capped the transportation mode choices in different cities. Although our analysis
probability of vehicle trips to be 100% and all trips would replace ve- only focused on the reduction of GHG emissions, the same approach can
hicle trips). This is why the line for the mode distribution change is not be applied to quantify other environmental impacts (e.g., NOx, PM2.5),
linear in Fig. 7 (the positive side is curved due to the capped 100% using appropriate emission factors as inputs.
probability). When we only consider changing the percentage of vehicle The total annual GHG emission reduction in the eight systems
trips by −30%–0% and fit a linear regression, the linear trend line has a evaluated in this study shown linear relationships with the overall
gradient of 0.6567 in Bay Area (R squared 0.9994) and a gradient of system size and ridership. According to our modeling result, as the
0.6542 in Philadelphia (R squared 0.9996). These results show that, if largest bike share program in the United States, New York’s Citi Bike
the bike share users take 10% less vehicles trips than an average person Share contributed to 5417 tons of GHG emission reduction in year
(whose travel pattern is captured in the NHTS data), the total emission 2016. In contrast, Seattle’s BSS only reduced 41 tons of GHG emissions
reduction of the system will decrease by 6.6% and 6.5% in Bay Area and in 2016. However, the emission reduction contribution from BSS cur-
Philadelphia, respectively. rently is still relatively limited (less than 0.1%) in the context of the
When we change the emission factors in the analysis (Table 4), the total GHG emissions from the entire transportation sector in these ci-
total emission reduction of all the studied cities decreased by 14–16% ties. Expanding the BSS system size (e.g., building more stations and
for the low emission factor case and increased by 41–43% for the high docks, launching more bikes) can help increase the GHG emission re-
emission factor case. This provides a lower and upper bound of the total duction by generating more bike share trips. Another thing to note is
emission reduction given the uncertainty of the emission factors in that, for Los Angeles, even though the overall traffic volume is at the
different cities. It is worth mentioning that it is unlikely that we can top level in the U.S. (Gold et al., 2015), the usage intensity (trips per
expect the high emission factor case to happen, which assumes that all bike, per station, and per dock) of its bike share system is relatively low,
the replaced vehicle trips used pickups. When we use the modified which means that bike share is not as popular there as in New York
emission factor for bicycling (adding emissions of stations and re- City. Improving biking infrastructures such as bike lanes can potentially
balance from Luo et al. (2019)), the total emission reductions are help improve the use of bike share (United Nations Environment
lowered by around 30% (Table 4). Therefore, the emissions from bike Programme (UNEP), 2018).
share stations and rebalance operation also play an important role for In addition, the environmental benefit is also influenced by the
the environmental performance of a BSS. mode substitution, i.e., which transportation mode does the bike share

Table 4
Change of total emission reduction by varying the emission factors for public transit, vehicles and.
Change (%) of total GHG emission reduction Seattle Los Angeles Bay Area Philadelphia Boston Washington DC Chicago New York

Using lower bound of the emission Factors for public transit and vehicles −14.7% −14.7% −15.0% −15.2% −15.7% −15.6% −15.9% −15.7%
Using upper bound of the emission Factors for public transit and vehicles 43.2% 43.3% 42.8% 42.5% 41.8% 41.9% 41.4% 41.7%
Using modified emission factor for bicycling −29.3% −33.1% −31.8% −27.9% −27.5% −28.8% −26.9% −28.8%

11
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

trip replace. In our analysis, the bike share is mainly replacing trips sensitivity analysis has showed that the results are very sensitive to the
made by vehicles and walking, while the distribution of replaced modes transportation mode distribution. Therefore, future research collecting
varies among different cities. The percentage of bike share trips repla- more detailed transportation mode distribution specifically from the
cing car trips is higher in Philadelphia and Chicago, compared to other BSS users can help improve the results accuracy. Additionally, other
cities, leading to higher GHG emission reduction per mile travelled. factors, such as weather and user demographics may also impact the
Such variation is mainly due to the difference of transportation mode mode substitution decisions. For example, we only evaluated whether
choice preferences in these cities and the spatial coverage of the BSS, public transit stations and stops are available near the bike share sta-
which impacts trip distances. tions in the model but did not consider the actual schedule and routes of
Furthermore, the bike share system layout matters. The GHG the public transit networks. It is possible that the users may have to wait
emission reduction is mainly contributed by short bike share trips (less for a long time or have to make too many transfers between two bike
than 5 miles). Bike share stations in the center of a city generally have share stations having access to public transit, making the substitution of
higher total GHG emission reduction but lower unit GHG emission re- public transit less feasible. When surveying BSS users on mode sub-
duction (per trip travelled). The replaced transportation modes also stitutions by bike share trips, it is important to focus the question on a
exhibit spatial variations: a higher proportion of bike share trips re- specific trip, so mode substitution can be linked to trip distance, pur-
placed car trips in areas further away from city center than in the pose, start time, origin, and destination, etc. Such information would
downtown area. Therefore, the unit emission reduction potential (per also benefit the system operators and city planners to optimize the
mile) from BSS could also be influenced by its bike share station net- system and attract more vehicle users to switch to bike share. Second,
work (i.e., how many stations are located in downtown versus areas the emission factor we used in this study may overestimate the en-
away from city center, in the business centers versus leisure areas etc.). vironmental benefits of bike share trips since emissions from other
In order to improve the unit environmental benefit of bike share (per system infrastructure and operations such as bike share station and bike
trip or per mile travelled), the system operators need to attract more rebalancing are not included. Our sensitivity analysis showed that the
vehicle users in the city center to switch to bike share, and also attract emissions from bike share stations and rebalance could decrease the
more users in the areas away from the city center. In order to achieve overall GHG emission reduction by around 30%. Emission factors from
these goals, the cities need to improve biking safety (for instance, a more comprehensive life cycle assessment for different transportation
having more bike lanes) and the convenience of using bike share (such modes, including bike share, tailored to each city would help improve
as simplifying the process of checking out and returning bikes), which the estimation using BS-EREM. Also, the system operators of BSS may
would make bike share a more competitive transportation mode. The apply different rebalancing schedules and use different vehicles (e.g.,
stations that are further away from city center are often distributed less trucks, vans, or electric tricycles). Therefore, to obtain more accurate
densely than in city center, which can be observed in Fig. 6. The dif- estimation, more detailed data on system operations in different cities
ficulty of finding a nearby bike share stations in areas away from city are required. In addition, there could be other factors affecting the
center may discourage users to choose bike share. Installing more sta- emission factors, such as the increased emission from extensive usage of
tions in these regions could help increase bike share usages but will air conditioning in vehicles in certain seasons, different driving beha-
increase the emissions from the station manufacturing. Having a viors, and vehicle ages etc., which are not reflected in our study.
dockless system in areas with less demand can reduce the need of sta- Furthermore, different BSS may use bikes and stations that are manu-
tions and the associated emissions but may increase emissions from factured with different materials or processes and apply different
rebalancing dispersed bikes. Future study considering these factors at maintenance and disposal practices. These factors will also affect the
the station level is needed to further quantify the tradeoffs and assess system’s GHG emissions and require more detailed data to be further
the net benefits of different approaches. In addition, as indicated by Luo addressed. Third, this study only considers the cases that bike sharing
et al. (2019) and our sensitivity analysis results, improving the re- can substitute the entire trip. Because the data only record trip start and
balance efficiency can further improve the sustainability of a BSS. This end stations, we have no information on the actual trip origin and
can be achieved by optimizing the station location and capacity (Park destinations. Bike share could also serve as the “last mile” solution and
and Sohn, 2017; García-Palomares et al., 2012; Romero et al., 2012), contribute to substituting a one-mode trip (e.g., car trip) with a multi-
applying more efficient rebalance vehicle routing algorithms (Chiariotti modal trip (e.g., bike-train-bike mobility chain as mentioned in (United
et al., 2018; Alvarez-Valdes et al., 2016), and using more environ- Nations Environment Programme (UNEP), 2018). Due to the lack of
mental-friendly vehicles (e.g., electric vehicles) instead of internal data to support such analysis, the multi-modal trip substitution is not
combustion engine vehicles for bike rebalancing. considered, which could underestimate the environmental benefits of
In the case of a dockless BSS or hybrid BSS (combining dockless and bike share. Data that tracks trips across different transportation modes
station-based BSS in the same system), the proposed BS-EREM can still (e.g., smart cards that can be used for both public transit and bike
be applied. To reduce the computational intensity due to the dispersed share) and more targeted survey questions could help further analyze
bikes, the service area can be partitioned into small grids. Then the these trips.
center of each grid can be modelled as a pseudo station. All trips
starting (ending) within a grid can be marked as starting (ending) at the Declaration of Competing Interest
corresponding pseudo station. We can also check whether there are
public transit facilities within the grid to determine the public transit The authors have no financial conflict of interest to declare.
accessibility. The grid size should be similar to the spatial coverage of
an existing bike share station. Such map gridding method has also been Acknowledgement
used in existing studies (such as in Zhang and Mi (2018), Pan et al.
(2018)) to model a dockless system. Z.K. thanks the support from the Hugh W. and Edna M. Donnan
Although this study has the merit of proposing a more detailed Fellowship and the Summer Exploratory Research Grant from
model to quantify the environmental benefits of BSS programs, there Environmental and Ecological Engineering at Purdue University. The
are three major limitations that we would like to point out. First, our authors are also immensely grateful to the anonymous reviewers for
results are based on the assumption that the BSS users have the same their comments and suggestions which greatly helped improve the re-
travel pattern as an average person in the urban area of each city. The search.

12
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Appendix A

Detailed procedures of BS-EREM

The BS-EREM involves the following variables

13
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

The algorithm below describes the details of BS-EREM in one round of simulation

14
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Appendix B

Spatial distribution of the GHG emission reduction for all the eight cities

15
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

Fig. B1 Spatial distribution of the emission reduction in different cities (100 stations are randomly selected and visualized in the four larger
systems to avoid overlap in visualization).

References bike-sharing systems. Sensors (Switzerland) 18, 1–22. https://doi.org/10.3390/

s18020512.
Cohen, A., Shaheen, S., 2018. Planning for Shared Mobility. https://doi.org/10.7922/
Alter, L., 2008. Bikes Faster Than Subway or Car, Seven Years Running, TreeHugger. G2NV9GDD.
https://www.treehugger.com/bikes/bikes-faster-than-subway-or-car-seven-years- Dave, S., 2010. Life Cycle Assessment of Transportation Options for Commuters. pp. 1–16.
running.html. (Accessed 18 November 2017). Drynda, P., 2014. Development of bike share systems and their impact on the sustain-
Alvarez-Valdes, R., Belenguer, J.M., Benavent, E., Bermudez, J.D., Muñoz, F., Vercher, E., ability of urban transport. Case study of Opole Bike. Cent. East. Eur. J. Manag. Econ.
Verdejo, F., 2016. Optimizing the level of service quality of a bike-sharing system. 2, 199–215.
Omega (United Kingdom) 62, 163–175. https://doi.org/10.1016/j.omega.2015.09. E.P, Kevin, T., 2014. 2012 Seattle Community Greenhouse Gas Emissions Inventory. .
007. http://www.seattle.gov/Documents/Departments/OSE/2012GHG_inventory_report_
Anderson, N.A., 2015. Portland Bicycle Share Health Impact Assessment. final.pdf.
Borecki, N., Rawls, B., Buck, D., Reyes, P., Al, E., 2012. Virginia Tech Capital Bikeshare Faghih-Imani, A., Anowar, S., Miller, E.J., Eluru, N., 2017. Hail a cab or ride a bike? A
Study. travel time comparison of taxi and bicycle-sharing systems in New York City. Transp.
Boston, G., 2014. GHG Change 2012-2013. Res. A Policy Pract. 101, 11–21. https://doi.org/10.1016/j.tra.2017.05.006.
Buehler, R., Hamre, A., 2015. Business and bikeshare user perceptions of the economic Fishman, E., Washington, S., Haworth, N., 2014. Bike share’s impact on car use: evidence
benefits of capital bikeshare. Transp. Res. Rec. J. Transp. Res. Board. 2520, 100–111. from the United States, Great Britain, and Australia. Transp. Res. D Transp. Environ.
https://doi.org/10.3141/2520-12. 31, 13–20. https://doi.org/10.1016/j.trd.2014.05.013.
Bullock, C., Brereton, F., Bailey, S., 2017. The economic contribution of public bike-share Fuller, D., Gauvin, L., Kestens, Y., Daniel, M., Fournier, M., Morency, P., Drouin, L., 2013.
to the sustainability and efficient functioning of cities. Sustain. Cities Soc. 28, 76–87. Impact evaluation of a public bicycle share program on cycling: a case example of
https://doi.org/10.1016/j.scs.2016.08.024. BIXI in Montreal, Quebec. Am. J. Public Health 103, 85–92. https://doi.org/10.
Chiariotti, F., Pielli, C., Zanella, A., Zorzi, M., 2018. A dynamic approach to rebalancing

16
Z. Kou, et al. Resources, Conservation & Recycling 153 (2020) 104534

2105/AJPH.2012.300917. Systems. pp. 1. http://arxiv.org/abs/1802.04592.

García-Palomares, J.C., Gutiérrez, J., Latorre, M., 2012. Optimizing the location of sta- Park, C., Sohn, S.Y., 2017. An optimization approach for the placement of bicycle-sharing
tions in bike-sharing programs: a GIS approach. Appl. Geogr. 35, 235–246. https:// stations to reduce short car trips: an application to the city of Seoul. Transp. Res. A
doi.org/10.1016/j.apgeog.2012.07.002. 105, 154–166. https://doi.org/10.1016/j.tra.2017.08.019.
Gold, M., Pinceti, S., Federico, F., 2015. 2015 Environmental Report Card for Los Angeles Pasion, C., Oyenuga, C., Gouin, K., LLC, C., 2017. Inventory of New York City Greenhouse
County. pp. 99. www.environment.ucla.edu. Gas Emissions in 2015. pp. 1–56.
González, M.C., Hidalgo, C.A., Barabási, A.-L., 2008. Understanding individual human Program, C., 2015. 2015 San Francisco Geographic Greenhouse Gas.
mobility patterns. Nature. 453, 779–782. https://doi.org/10.1038/nature06958. Report, F., 2017. City of Chicago greenhouse gas inventory report calendar year 2015,
Hamilton, T.L., Wichman, C.J., 2018. Bicycle infrastructure and traffic congestion: evi- Chicago. . https://www.cityofchicago.org/content/dam/city/progs/env/GHG_
dence from DC’s capital bikeshare. J. Environ. Econ. Manage. 87, 72–93. https://doi. Inventory/CityofChicago_2015_GHG_Emissions_Inventory_Report.pdf.
org/10.1016/j.jeem.2017.03.007. Rhee, I., Rhee, I., Shin, M., Shin, M., Hong, S., Lee, K., Hong, S., Kim, S.J., Lee, K., Chong,
I. Of, 2012. District of Columbia Greenhouse Gas Inventory Update 2012-2013. https:// S., Chong, S., 2011. On the levy-walk nature of human mobility: do humans walk like
doee.dc.gov/sites/default/files/dc/sites/ddoe/service_content/attachments/2013_ monkeys? IEEE/ACM Trans. Netw. 19, 630–643. https://doi.org/10.1109/TNET.
Greenhouse_Gas_Inventory_Update_web.pdf. 2011.2120618.
Jäppinen, S., Toivonen, T., Salonen, M., 2013. Modelling the potential effect of shared Romero, J.P., Ibeas, A., Moura, J.L., Benavente, J., Alonso, B., 2012. A simulation-opti-
bicycles on public transport travel times in Greater Helsinki: an open data approach. mization approach to design efficient systems of of bike-sharing. Procedia 54,
Appl. Geogr. 43, 13–24. https://doi.org/10.1016/j.apgeog.2013.05.010. 646–655. https://doi.org/10.1016/j.sbspro.2012.09.782.
Kou, Z., Cai, H., 2019. Understanding bike sharing travel patterns: an analysis of trip data Shaheen, S., Guzman, S., Zhang, H., 2010. Bikesharing in Europe, the Americas, and Asia.
from eight cities. Phys. A Stat. Mech. Appl. 515, 785–797. https://doi.org/10.1016/j. Transp. Res. Rec. J. Transp. Res. Board 2143, 159–167. https://doi.org/10.3141/
physa.2018.09.123. 2143-20.
LDA Consulting, Capital Bikeshare, 2016. Member Survey Report, 2016. Shaheen, S., Martin, E., Cohen, A., 2013. Public bikesharing and modal shift behavior: a
Lobo, A., 2013. Bikes Are Faster Door-to-door Than Cars or Public Transport Within 5- comparative study of early bikesharing systems in North America. Int. J. Transp. 1,
10km, Better by Bicycl. http://www.betterbybicycle.com/2013/12/bikes-are-faster- 35–54. https://doi.org/10.14257/ijt.2013.1.1.03.
door-to-door-than-cars.html. (Accessed 18 November 2017). Spring, P., 2015. Philadelphia Citywide Greenhouse Gas Inventory, 2012 Table of
Luo, H., Kou, Z., Zhao, F., Cai, H., 2019. Comparative life cycle assessment of station- Contents.
based and dock-less bike sharing systems. Resour. Conserv. Recycl. 146, 180–189. Susan, S., Elliot, M., Nelson, C., Adam, C., Mike, P., 2014. Public Bikesharing in North
https://doi.org/10.1016/j.resconrec.2019.03.003. America During a Period of Rapid Expansion : Understanding Business Models,
Martin, E.W., Shaheen, S.A., 2014. Evaluating public transit modal shift dynamics in Industry Trends and User Impacts.
response to bikesharing: a tale of two U.S. Cities. J. Transp. Geogr. 41, 315–324. Thompson, J., Potok, N., Blumerman, L., 2016. TIGER/Line® Shapefiles Technical
https://doi.org/10.1016/j.jtrangeo.2014.06.026. Documentation.
NACTO, 2017. Bike Share in the U. S.: 2017. https://nacto.org/bike-share-statistics- U.S.DOT, 2017. 2017 National Household Travel Survey. http://nhts.ornl.gov.
2017/. United Nations Environment Programme (UNEP), 2018. The Weight of Cities: Resource
Nair, R., Miller-hooks, E., Hampshire, R.C., Bušić, A., Nair, R., Miller-hooks, E., Requirements of Future Urbanization. www.internationalresourcepanel.org.
Hampshire, R.C., Bušić, A., 2013. Large-scale Vehicle Sharing Systems: Analysis of USDOT, 2009. Summary of Travel Trend: 2009 National Household Travel Survey.
Vélib Large-Scale Vehicle Sharing Systems: Analysis. https://doi.org/10.1080/ http://nhts.ornl.gov.
15568318.2012.660115. 8318. Zhang, Y., Mi, Z., 2018. Environmental benefits of bike sharing: a big data-based analysis.
Pan, L., Cai, Q., Fang, Z., Tang, P., Huang, L., 2018. Rebalancing Dockless Bike Sharing Appl. Energy 220, 296–301. https://doi.org/10.1016/j.apenergy.2018.03.101.

17