Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Biking schedules : a new tool for bicycle travel analysis Mohamed, Amr 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2018_may_mohamed_amr.pdf [ 5.85MB ]
Metadata
JSON: 24-1.0365817.json
JSON-LD: 24-1.0365817-ld.json
RDF/XML (Pretty): 24-1.0365817-rdf.xml
RDF/JSON: 24-1.0365817-rdf.json
Turtle: 24-1.0365817-turtle.txt
N-Triples: 24-1.0365817-rdf-ntriples.txt
Original Record: 24-1.0365817-source.json
Full Text
24-1.0365817-fulltext.txt
Citation
24-1.0365817.ris

Full Text

BIKING SCHEDULES: A NEW TOOL FOR BICYCLE TRAVEL ANALYSIS by Amr Mohamed   A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF  THE REQUIREMENTS FOR THE DEGREE OF    MASTER OF APPLIED SCIENCE    in    THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES  (Civil Engineering)     THE UNIVERSITY OF BRITISH COLUMBIA  (Vancouver)     April 2018  © Amr Mohamed, 2018    ii  The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, a thesis/dissertation entitled: BIKING SCHEDULES: A NEW TOOL FOR BICYCLE TRAVEL ANALYSIS submitted by  Amr Mohamed    in partial fulfillment of the requirements for the degree of   Master of Applied Science       in  Civil Engineering  Examining Committee:  Dr. Alex Bigazzi Supervisor Dr. Omar Swei Supervisory Committee Member Supervisory Committee Member Additional Examiner  Additional Supervisory Committee Members:  Supervisory Committee Member Supervisory Committee Member    iii  Abstract With an increasing focus on bicycling as a mode of urban transportation, there is a pressing need for advanced tools for bicycle travel analysis and modeling. The objective of this thesis is to introduce “Biking schedules” to represent archetypal urban cycling dynamics along with its methods of construction and potential applications. Biking schedules are constructed by appending short trip segments, called microtrips, together. Three different methods of constructing biking schedules with both speed and road grade attributes are developed. As an initial proof-of-concept, the methods are applied and compared using a pre-existing demonstration data set of 55 hours of 1-Hz on-road GPS data from three cyclists. Biking schedules are evaluated based on their ability to represent the speed dynamics, power output, and breathing rates of a calibration data set and then validated for different riders. The impact of using coarser 3, 5, and 10 second GPS logging intervals on the accuracy of the schedules is also evaluated. Results indicate that the best biking schedule construction method depends on the volume and resolution of the calibration data set. Overall, biking schedules can successfully represent most of the assessed characteristics of cycling dynamics in the calibration data set within 5%.  As a second step, the biking schedule construction methods are further developed and validated by collecting and applying a large, naturalistic, GPS-based data set of 2314 bicycle trips in Vancouver, Canada. We specifically explored the optimal microtrip definition to be adopted in constructing biking schedules. The choice of the optimal microtrip definition depends on the parameter that biking schedules are originally generated to model. Generally, the 150m microtrips generated the most precise biking schedules.    iv  The collected data are also used to compare the travel characteristics and construct biking schedules for regular and electric bikes. Results show that electric bikes travel 7 km/hr and accelerate 0.17 km/hr/sec faster than regular bicycles. Moreover, the total energy used to move electric bikes is almost twice as much as the energy used by regular bikes. These results have implications in designing bike lanes and safety analysis. Potential applications for biking schedules are also discussed.          v  Lay Summary This research introduces a new tool for bicycle travel analysis. Biking schedules are constructed by appending short trip segments together called microtrips. We looked into three different methods of appending these microtrips using a demonstration data set. Results show that the choice of the best construction method depends on the data resolution and size. A larger data set for more cyclists was then collected in Metro-Vancouver, BC. We used this data set to explore the optimal microtrip definition and to construct biking schedules for comparing between the travel characteristics of regular and electric bikes. Results show that electric bikes generally travel and accelerate faster, and they also use twice as much energy in total. These results help practitioners and designers understand the differences in cycling behavior between electric and regular bike riders, and will have implications in designing bike lanes shared between these two types of vehicles.    vi  Preface This research was conducting by Amr Mohamed under the supervision of Dr. Alex Bigazzi (Assistant Professor at the University of British Columbia). Amr was responsible for conducting literature review, preparing and administering data collection, and data processing and analysis. Several members of the UBC REsearch on ACtive Transportation lab team (REACT) helped in the preparation, recruitment, and administeration phases of the survey. These people are Kathrine Le, Daniel Valenzuela, Omar El Masri, Saki Aono, Xugang Zhong, and Elmira Berjisian. Dr. Alex Bigazzi was continuously providing feedback and guidance throughout the thesis.  The thesis mainly consists of three parts; Proof-of-concept study, Data collection, and Data analysis. The proof-of-concept study was presented in 2017 at the annual Transportation Research Board conference in Washington DC, USA. It was also published in Transportation Research Record journal (Mohamed, A. and A. Bigazzi, “Generation of ‘Biking Schedules’ for Bicycle Travel Analysis.” Transportation Research Record, Journal of Transportation Research Board, Washington DC, USA). The survey conducted in this research was approved by the Behavioral Research Ethics Board (BREB) requirements, under project titled “Incorporating electric bicycles into urban transportation systems in British Columbia”, UBC BREB number: H17-00294.  This research was partially funded by “Social Sciences and Humanities Research Council of Canada (SSHRC).     vii  Table of Contents  Abstract .................................................................................................................................... iii Lay Summary ............................................................................................................................ v Preface...................................................................................................................................... vi Table of Contents .................................................................................................................... vii List of Tables .......................................................................................................................... xii List of Figures ........................................................................................................................ xiii List of Abbreviations .............................................................................................................. xv Acknowledgements .............................................................................................................. xviii Dedication .............................................................................................................................. xix 1. Introduction ........................................................................................................................... 1 2. Literature Review.................................................................................................................. 4 2.1 Current cycling analysis tools ......................................................................................... 4 2.2 Driving schedules ............................................................................................................ 7 2.3 Electric bikes ................................................................................................................. 10 3. Methodology ....................................................................................................................... 13 3.1 Overview of methodology ........................................................................................ 13 3.1.1 Derive biking schedules construction method from driving schedules literature .. 13 3.1.2 Apply to existing bike data in proof-of-concept study ........................................... 13   viii  3.1.3 Collect large naturalistic dataset ............................................................................. 13 3.1.4 Data analysis ........................................................................................................... 14 3.2 Biking schedule construction methodology .................................................................. 14 3.2.1 Construction framework ......................................................................................... 14 3.2.2 Microtrips ............................................................................................................... 15 3.2.3 Assessment criteria ................................................................................................. 15 3.2.4 Testing and selection .............................................................................................. 16 3.2.5 Biking schedule construction.................................................................................. 17 3.3 Proof-of-concept study .................................................................................................. 21 3.4 Data collection............................................................................................................... 24 3.4.1 Introduction ............................................................................................................ 24 3.4.2 Survey design ......................................................................................................... 25 3.4.3 Smartphones vs stand-alone GPS devices .............................................................. 25 3.4.4 Choosing a smartphone application for the survey ................................................ 27 3.4.5 Choosing a heart rate monitor ................................................................................ 29 3.4.6 Preparing a survey plan .......................................................................................... 30 3.4.7 Following up emails ............................................................................................... 31 3.4.8 BREB approval ....................................................................................................... 31 3.4.9 Pilot surveying ........................................................................................................ 32 3.4.10 Recruitment .......................................................................................................... 33   ix  3.4.11 Incentives: ............................................................................................................. 41 3.5 Data processing ............................................................................................................. 43 3.5.1 Preliminary processing ........................................................................................... 43 3.5.2 Speed calculation .................................................................................................... 45 3.5.3 Extracting stop periods ........................................................................................... 46 3.5.4 Data filtering ........................................................................................................... 51 3.5.5 Elevation data ......................................................................................................... 53 3.5.6 Road grade calculation ........................................................................................... 53 3.5.7 Data smoothing ....................................................................................................... 54 3.6 Testing optimal microtrip length ................................................................................... 55 3.7 Comparison between the travel characteristics of electric and regular bikes ............... 57 4. Results ................................................................................................................................. 63 4.1 Proof-of-concept study .................................................................................................. 63 4.1.1 Processing time ....................................................................................................... 63 4.1.2 Schedule progression .............................................................................................. 63 4.1.3 Performance value .................................................................................................. 64 4.1.4 Effect of GPS resolution ......................................................................................... 65 4.1.5 Power and breathing rate results ............................................................................. 66 4.1.6 Transferability to other riders ................................................................................. 67 4.2 GPS survey data overview ............................................................................................ 68   x  4.2.1 General overview .................................................................................................... 68 4.2.2 Data representativeness .......................................................................................... 73 4.3 Optimal microtrips length for biking schedules ............................................................ 77 4.4 Comparison between electric and regular bikes travel characteristics .......................... 82 4.4.1 Target assessment measures ................................................................................... 82 4.4.2 Comparison between biking schedules ................................................................... 85 5. Conclusions ......................................................................................................................... 89 5.1 Research questions/answers .......................................................................................... 89 5.2 Potential biking schedule applications .......................................................................... 93 5.3 Additional findings and unique contributions ............................................................... 94 5.4 Lessons learned in recruiting cyclists for a travel survey ............................................. 96 5.4.1 Attracting participants during field recruitment ..................................................... 96 5.4.2 How to target electric bikes riders .......................................................................... 98 5.5 Limitations and future research ..................................................................................... 98 Bibliography ......................................................................................................................... 101 Appendices ............................................................................................................................ 110 Appendix A: Survey consent form .................................................................................... 110 Appendix B: Survey questionnaire ................................................................................... 112 Appendix C: Smartphone application registration form ................................................... 117 Appendix D: Online uploading registration form ............................................................. 120   xi  Appendix E: Instructions ................................................................................................... 128      xii  List of Tables Table 1: Assessment criteria for biking schedules .................................................................. 16 Table 2: Parameters used in power and breathing rate calculations ....................................... 23 Table 3: Comparison between different smartphone applications .......................................... 28 Table 4: Comparison between different heart rate monitors .................................................. 29 Table 5: Relationship between number of clusters and sum of squares within clusters for the 100m microtrip pool ............................................................................................................... 57 Table 6: PSM output for matching 1....................................................................................... 60 Table 7: PSM output for matching 2....................................................................................... 60 Table 8: PSM output for matching 3....................................................................................... 61 Table 9: PSM output for matching 4....................................................................................... 61 Table 10: Parameters used in calculating power and energy expenditure .............................. 62 Table 11: PVs for the best biking schedule generated from each method .............................. 65 Table 12: Cyclist power output and breathing rate calculated from raw data and best biking schedules generated from each method .................................................................................. 67 Table 13: Individual PVs for the best biking schedules in (%) .............................................. 80 Table 14: Individual average PVs for all biking schedules in (%) ......................................... 81 Table 15: Comparison between the assessment parameters for regular and electric bikes .... 83 Table 16: PV for the best generated biking schedules from each set of trips ......................... 86     xiii  List of Figures Figure 1: Example of driving schedules (Source: Link) ........................................................... 8 Figure 2: Lab dynamometer testing (source: Link) .................................................................. 9 Figure 3: Framework for constructing driving schedules ....................................................... 10 Figure 4: Different electric bike designs ................................................................................. 11 Figure 5: Framework for constructing biking schedules with three different methods .......... 18 Figure 6: Relationship between number of clusters and sum of squared errors within clusters................................................................................................................................................. 20 Figure 7: Rhythm+™ heart rate monitor ................................................................................ 30 Figure 8: Survey flyer ............................................................................................................. 37 Figure 9: Survey flyer with tear-offs ...................................................................................... 38 Figure 10: Survey invitation card ........................................................................................... 39 Figure 11: Locations vistied for field recruitment .................................................................. 40 Figure 12: Hats and socks incentives used in the survey ........................................................ 42 Figure 13: Example of GPS points at stops locations ............................................................. 47 Figure 14: Buffer around the first observation in the potential stop location ......................... 50 Figure 15: Relationship between PV and drive schedule length for each method ................. 64 Figure 16: PV across methods using data resolutions of 1, 3, 5, and 10 seconds ................... 66 Figure 17: Geograhic distribution of participants ................................................................... 69 Figure 18: Trip distribution by mode ...................................................................................... 70 Figure 19: Trip distribution by purpose .................................................................................. 71 Figure 20: Distribution of trip purpose by mode type ............................................................ 72 Figure 21: Gender distribution ................................................................................................ 74   xiv  Figure 22: Income (in $) distribution ...................................................................................... 75 Figure 23: Cumulative age distribution .................................................................................. 76 Figure 24: Relationship between the number and definition of microtrips ............................ 78 Figure 25: PVs generated from each pool ............................................................................... 79 Figure 26 Best biking schedule from the 150m pool .............................................................. 80 Figure 27: Total moved distance of all matches compared with electric bikes (calculated from the best biking schedules) ....................................................................................................... 87 Figure 28: Total energy expenditure of all matches compared with electric bikes (calculated from the best biking schedules) .............................................................................................. 88     xv  List of Abbreviations Kinematic abbreviations  sec Second  min Minute  hr Hour  m Meter  km Kilometer  v Velocity  a Acceleration  G Grade  W Watt  Hz Hertz  L Liter  Biking schedule related abbreviations  BS Biking schedule  PV Performance value  ATS Average trip speed  ARS Average running speed  AAG Average absolute grade  AAC Average absolute acceleration  PTI Percentage time idling  PTA Percentage time accelerating   xvi   PTD Percentage time decelerating  PTC Percentage time cruising  PTPG Percentage time positive grade  PTNG Percentage time negative grade   APW Average positive work per distance  SAGPD Speed acceleration grade probability distribution  STS Stop to stop  Data format abbreviations  tcx Training Center Extensible Markup Language  fit Flexible Image Transport  xlsx Excel Microsoft Office Spreadsheet  gpx Global Positioning System Exchange  Statistics abbreviations  PSM Propensity Score Matching  CI Confidence Interval  sd Standard deviation  SSE Sum of Squared Errors        xvii  Other abbreviations  GPS Global Positioning System  lat Latitude  lon Longitude  HRM Heart Rate Monitor  E-bike Electric bike  API Application Programming Interface  DEM Digital Elevation Model  TMD Theoretical Moved Distance  AMD Actual Moved Distance  UBC University of British Columbia  COPERT Computer Program to calculate Emissions from Road Transport  EMFAC Emission Factor  MOVES Motor Vehicle Emission Simulator  $ LOS Canadian Dollars Level of Service      xviii  Acknowledgements I would like to thank my supervisor Dr. Alex Bigazzi for all his guidance and support to make this research possible. Dr. Alex was a key motivator and he always pushed me forward and helped me out of my frustration. His positivity was flowing all the time and I felt like we are part of a team rather than a supervisor and a student. I would like also to thank all professors who instructed courses to me. Prof. Tarek Sayed who taught me the basics of traffic safety. Dr. Haukaas who changed me into a statistician in one semester. Dr. Alan Russel who taught me basic and advanced construction management. I would like to seize this opportunity to mention my father Prof. Salah who showed me the way and guided me in my youth. He taught me that knowledge is my key for success. My mother who showed me endless love and a never breaking bond no matter the distances. I cannot also forget all 216 participants who participated in the survey, Louise Fogarty who designed the cool hats, and everyone who helped me with the survey.      xix  Dedication I dedicate this research to my parents and siblings. I want to tell them thanks for supporting me until this moment.        1  1. Introduction  Promoting cycling is an increasingly common part of environmental initiatives in the transportation sector to decrease pollution emissions (Intergovernmental Panel on Climate Change & Edenhofer, 2014). As a result of these initiatives, many cities are observing an increase in bicycle lanes either by allocating part of the road or by building off-road paths. These changes motivated plenty of people to commute on bicycles. Nowadays, there are several cycling clubs and associations that organize cycling events and meetups dedicated to cycling enthusiasts. This rise in cycling awareness escalates the need for providing advanced cycling analysis tools to get a better perception of the real-world cycling behavior. Unfortunately research in detailed microsimulation and operational models of on-road urban bicycle dynamics are still in their infancy (Ma & Luo, 2016; Twaddle & Grigoropoulos, 2016; Twaddle, Schendzielorz, & Fakler, 2014). Current research in bicycle analysis is focused on modelling speed and acceleration under different grade settings (Parkin & Rotheram, 2010; Xu, Li, Qu, & Tao, 2015). Some more comprehensive research (Ma & Luo, 2016) developed acceleration models customized for different genders and agility.  There is no doubt that there is a lack of understanding of how cycling behavior changes under different sociodemographic and road conditions. This knowledge gap hinders understanding urban cycling characteristics and in turn how to effectively promote cycling. Knowledge of typical urban cycling dynamics (such as speed, acceleration, etc) would be useful for a number of applications. It has been found that human power requirements of cycling vary greatly with speed, acceleration, road grade, pavement condition, air drag, and rolling resistance (Bigazzi & Figliozzi, 2015; Tengattini, 2017; Wilson, 2004). Human power   2  and energy expenditure, in turn, are linked to physical activity levels, breathing rates and pollution inhalation, and many facets of travel behavior such as speed, route, and mode choices (Bigazzi & Figliozzi, 2015; Bigazzi & Lindsey, Forthcoming; Heinen, van Wee, & Maat, 2010; Willis, Manaugh, & El-Geneidy, 2015).  Up till this day, there is no analysis tool that provides researchers and practitioners with detailed microsimulation for urban cycling that could be customized for different cyclists, road conditions, gender, age, trip purpose, etc. Such analysis tool is already present and implemented in practice for motor vehicles. This tool is called “Driving Schedules” and it is capable of representing real-world driving behavior under different variations (Giakoumis, 2017). This thesis aims at filling the discussed knowledge gap by presenting a new analysis tool for bicycle travel, which we are calling “Biking Schedules”.  The objective of this thesis is to build on the driving schedule literature to introduce the concept of biking schedules along with the methods for their construction and potential applications. The thesis will progressively investigate and answer the following questions i) What are the different potential methods of constructing biking schedules? How do they differ? What is the best method? ii) What is the optimal definition of the biking schedule building unit (known as microtrips)? iii) How could the biking schedules be implemented to compare between the travel characteristics of electric and regular bicycles?  This thesis seeks to answer these questions by first reviewing driving schedule literature (presented in chapter 2) to establish the theoretical foundation of the biking schedules.   3  Chapter 3 discusses the research methodology. It starts with a proof-of-concept study on a small-sized existing data set to explore and evaluate the different methods of constructing biking schedules. Once the theoretical background is established, a high-frequency GPS data set will be used to enhance the biking schedule construction methodology. The GPS data will also be used to generate biking schedules for both electric and regular bicycles. Chapter 3 also includes a detailed description of the GPS data collection method along with its pre-processing methodology. Chapter 4 presents the analysis results. Chapters 5 addresses each of the research questions individually and identifies the most important takeaways from the thesis and learned behavioral practices.      4  2. Literature Review The number of cyclists is steadily increasing in North American cities (CH2M, 2016; Vélo Québec, 2013). In light of this observed interest towards cycling, several design manuals were revisited to accommodate bicycle facilities in road design (AASHTO, 2012; CROW, 2007; NACTO, 2012). Researchers, on the other hand, have been evaluating cycling experiences by different means such as stated preference surveys (Hunt, n.d.; Krizek & Roland, 2005; Sener, Eluru, & Bhat, 2009; Stinson & Bhat, 2003; Tilahun, Levinson, & Krizek, 2007) and revealed preference surveys (Broach, Dill, & Gliebe, 2012; Hood, Sall, & Charlton, 2011; Menghini, Carrasco, Schüssler, & Axhausen, 2010). Other than surveys, the state of research also utilized technology to collect naturalistic cycling data for modeling real-world cycling behavior. These technologies include the use of GPS (Parkin & Rotheram, 2010; Strauss & Miranda-Moreno, 2017), video data (Xu et al., 2015), and bicycle mounted equipment (Dozza & Werneke, 2014). 2.1 Current cycling analysis tools For providing more in-depth understanding of the real-world cycling behavior and how it is influenced by different travel conditions, several researchers contributed to the field by presenting analysis tools for examining microscopic bicycle travel characteristics. This section presents a review on some of the available analysis tools in the literature to address the research gap.  A previous study (Bigazzi & Lindsey, Forthcoming) developed a utility-based behavioral model of bicycle speed choice. The model estimates the speed that maximizes the cycling   5  utility, taking into consideration trade-offs between several factors such as travel time, energy expenditure, and control.  Other research interests included route choice modeling. (Broach et al., 2012) developed a route choice model using GPS data for 164 cyclists in Portland. Participants were provided with a hand-held GPS device that records location every three seconds. The device was also customized so that participants enter the weather and trip purpose before starting the trip. When the device is returned, the data were collected and participants were asked to answer an online questionnaire regarding their trips. The GPS traces were then loaded on a map that includes all network links and nodes. The researchers examined the origin and destination of each trip and the actual route taken by the cyclist. The route choice model was then developed by examining the actual routes in contrast with a set of potential alternatives. Results revealed that route choice is influenced by several factors such as distance, turn frequency, slope, intersection control, and traffic volumes.  A similar study was conducted in San Francisco using GPS data collected from smartphones (Hood et al., 2011). Results show that traffic volumes, number of lanes, crime rates, and nightfall were not considered by the cyclists in their route choice process. This research contradicts with the aforementioned research in whether traffic volumes influences cyclists’ route choice. Some other route choice models used stated preference online survey to decide the most influencing factors in route choice modeling (Sener et al., 2009). The observed increase in bicycle traffic also encouraged some safety engineers to study the interactions between cyclists and other road users, and the effect of those interactions on traffic collisions and road safety in general. (Sayed, Zaki, & Autey, 2013) used automated   6  video data analysis to study the interactions between cyclists and vehicles in downtown Vancouver, British Columbia. The research presents an approach to automatically identify road conflicts and violations. Such analysis is helpful at identifying the riskiest accident-prone locations and also advising some countermeasures to mitigate traffic conflicts. Other studies also include speed and acceleration data analysis. A previous study (Parkin & Rotheram, 2010) supplied 16 cyclists with GPS devices and heart rate monitors. Participants were asked to log 100 min of data during a week. Speed and acceleration data were supplemented with road grade corroborated with map matching to develop linear regression models. Results suggest adopting a design speed of 25 km/hr for road grades between -3% and 3%, whereas a design speed of 30 km/h is more preferred on steeper downgrades. The study also reported that cyclists deliver and average power of 150W on flat grades, which increases to 250W on positive grades.  Another study studied the acceleration behavior from a naturalistic data set collected for eleven cyclists over a course of two data collection periods (Ma & Luo, 2016). The study specifically identifies the different cycling regimes (accelerating, decelerating, and cruising) in bicycle trajectories. It also studies and models acceleration profiles that represent longitudinal bicycle movements without any interaction with any other road users. The study found that gender and agility both play a significant role in influencing acceleration rates. As previously demonstrated, current state-of-research is focused on modelling speed, acceleration, route choice, and road safety. But none of them provided an analysis tool capable of representing archetypal urban cycling behavior under different travel and roadway conditions. Additionally, none of them provided a microsimulation tools which is capable of   7  explaining cycling dynamics such as variations in speed, acceleration, and grade throughout the trip. The existence of such tool will enable studying how measures of interest such as energy expenditure, breathing rate, and heart rate vary along a trip. It can also explain how travel characteristics vary between different riders, weather conditions, and roadway characteristics. The proposed biking schedules are introduced to address all the aforementioned information gaps. Variations in speed, acceleration, and grade are explained by providing an adequate length of speed and grade profiles that encamps all cycling regimes in the actual trip. Any other measure of interests could be simulated microscopically to study their relationships with different travel characteristics.  Biking schedules are derived from driving schedule literature. A tool which was developed to address similar issues for motor-vehicles. The next section provides an overview of driving schedule literature.      2.2 Driving schedules Driving schedules, also referred to as driving cycles, are second-by-second speed profiles designed to represent typical driving patterns (Wang, Huo, He, Yao, & Zhang, 2008). An example of driving schedules is presented in Figure 1. Driving schedules vary across different travel and roadway characteristics such as terrain, traffic, vehicle characteristics, weather conditions, and pavement type, etc. (Giakoumis, 2017).    8   Figure 1: Example of driving schedules (Source: Link)  A main application of driving schedules is in lab dynamometer testing of real-world vehicles as shown in Figure 2. The vehicle is operated on a specific driving schedule that emulates realistic driving conditions. Exhaust from vehicle is then collected and inputted to a gas analyzer system. The role of the gas analyzer is to identify the components of the exhaust and then a computer software is used to correlate each element with the operated driving schedules.    9   Figure 2: Lab dynamometer testing (source: Link)  Similarly, driving schedules could be used to estimate the amount of fuel consumption accompanying a specific driving pattern. Vehicles are operated on chosen driving schedule and the fuel level is monitored. The amount of consumed fuel is then correlated with speed and road grade in order to develop fuel consumption models.  The framework of constructing driving schedules is presented in Figure 3. A large speed-time data set is first collected via GPS navigators or chase-car technique (Bishop, Axon, & McCulloch, 2012; Kamble, Mathew, & Sharma, 2009). A set of predetermined aggregate assessment parameters (such as average speed and acceleration) are calculated for the calibration data set. These parameters describe the characteristics of the data set and the driving schedules are constructed to replicate these parameters. To construct driving schedules, the calibration data set is firstly divided into small snippets called microtrips.   10  Driving schedules are then constructed by appending microtrips together in order to reproduce the assessments parameters calculated for the calibration data set as closely as possible. The precision of the driving schedule is evaluated by a single number indicators called performance value (PV), which represents the differences between the assessment parameters of the calibration data set and the generated driving schedule. If the PV is meeting a certain threshold, the generated driving schedule is accepted, otherwise it is rejected and another driving schedule is generated. We are building on this framework in order to construct biking schedules using data set collected for cyclists.  Figure 3: Framework for constructing driving schedules   2.3 Electric bikes Electric bikes (E-bikes) are motor-assisted bikes that allow riders to travel at higher speeds and climb hills with less effort. The motor uses energy stored in a battery to supply riders with additional propelling force. Electric bikes are not always easy to distinguish from regular bikes. Electric bikes have different styles and designs (NITC, 2014), they can look more like a bicycle (i.e. bicycle style electric bikes), or more like a scooter (i.e. scooter style electric bikes) as shown in Figure 4.    11    Figure 4: Different electric bike designs  There are also different designs for the electric assistance control method. They could have: 1) electric-assisted pedals that provide riders with extra propelling force when they pedals, 2) pedals and throttle: The rider controls the amount of electric-assistance through a throttle, while the pedals are used for propelling using human power. 3) Throttle only: There are no pedals attached. The bike propels solely through the use of a throttle similar to motor-bikes Compared to regular bikes, electric bikes ridership comprises of more people above 60 years old or with physical limitations (MacArthur, Dill, & Person, 2014; Wolf & Seebauer, 2014). Electric bikes are proven to have environmental benefits over motor-vehicles (C. R. Cherry, Weinert, & Xinmiao, 2009). Additionally, electric bikes are more likely to increase the number of cycling trips and travelled distance in contrast with regular bike trips (Fyhri & Fearnley, 2015).    12  In the light of the aforementioned advantages of electric bikes, several countries reported an increase in the usage and ownership of electric bikes (C. Cherry & Cervero, 2007; Jamerson & Ed Benjamin, 2018). This increase in electric bike adoption created a new research topic. However most of the researchers were looking into the impact of the increased electric bike adoption from the environmental, safety, and travel share perspectives (C. Cherry & Cervero, 2007; Pierce, Nash, & Clouter, 2013). Few researchers investigated the differences in travel characteristics between regular and electric bikes. A recent study (Langford, Chen, & Cherry, 2015) reported that the on-road speed of electric bikes is higher than regular bikes, whereas regular bikes travel faster than electric bikes on shared paths. Although this study provides an insight on speed differences, it did not account for the socio-demographic differences between the riders. Additionally, most of the reported values are aggregate measures such as percentiles and averages. These measures are insufficient to examine how different measures such as energy consumptions and heart rate vary along the trip. This emphasis the need for a new microsimulation analysis that can account for different socio-demographic characteristics and provide speed/grade profiles for an archetypal trip. There is also a need to study the differences in travel characteristics between electric bikes and regular bikes after matching the sociodemographic characters that are most likely to affect the trip attributes.         13  3. Methodology 3.1 Overview of methodology 3.1.1 Derive biking schedules construction method from driving schedules literature The first part of the methodology explores the methods of constructing biking schedules. The construction methods were mainly derived from driving schedules literature. We explored three different methods of constructing biking schedules. They all share a similar principle of appending short trip segments until reaching the desired biking schedule length, but they differ in how those segments are appended.  3.1.2 Apply to existing bike data in proof-of-concept study In this part, we seek to validate the established biking schedule construction methods. This was done by applying the methods to a small-sized data set collected in a previous study.  It should be noted that we were not targeting at this step to develop ready-for-practice biking schedules, but rather to explore, test, and compare the construction methods and to evaluate the developed biking schedules.   3.1.3 Collect large naturalistic dataset Once the construction methods are established and evaluated, we seek to collect a larger GPS data set by conducting a survey in Metro-Vancouver, BC, Canada. The methodology of the data collection is presented along with all preparations and administrations. We also present comprehensive description of all post-processing analysis done to the data.   14  3.1.4 Data analysis The data analysis presented in this thesis consists of two parts. First, we attempted to enhance the biking schedule construction methodology. We tested and evaluated different microtrip definitions and their impact on the resulting biking schedule. Second, we utilized all knowledge gained during previous stages to construct biking schedules for each of the regular and electric bikes. The constructed biking schedules will be used to explore the differences in travel characteristics. 3.2 Biking schedule construction methodology 3.2.1 Construction framework  As previously mentioned, the biking schedule construction framework is mainly derived from the driving schedule literature. In general, driving schedule construction methodologies can be classified into microtrip based, segment-based, pattern based, and stochastic modal approaches (Dai, Niemeier, & Eisinger, 2008). All approaches share a general concept of dividing the data set into shorter segments and then append these segments to represent the main data set as closely as possible. In the microtrip based approach, the main data set is firstly divided into short snippets every predefined distance, time, or segment between stops. The segment based approach stratifies the main data set by roadway type and level of service (LOS). The pattern based approach uses statistical methods (such as clustering) to identify different travel patterns in the data set. The stochastic modal approach views trips as a sequence of modal events such acceleration, deceleration, cruising, etc. Since this is the first known attempt to construct biking schedules, the microtrip based approach was utilized because it was easier to implement given the abundant information   15  available in the literature (Hung, Tong, Lee, Ha, & Pao, 2007; Kamble et al., 2009; Seers, Nachin, & Glaus, 2015). Observed travel data are divided into small snippets, called microtrips, which form the elementary building unit of the biking schedules, which is then constructed by appending microtrips together until a desired length is reached.  Departing from driving schedule methods, the microtrips and schedules in this research include synchronous speed and grade data. Road grade affects several aspects of cycling trips such as heart rate, energy expenditure, and breathing rate (Berry, Koves, & Benedetto, 2000; Bigazzi & Figliozzi, 2015). The presence of road grade in the generated biking schedule will extend the applications of biking schedules.   3.2.2 Microtrips Approaches to defining microtrips vary across the literature. One common approach is to define microtrips between two consecutive stops (André, 2004; Berzi, Delogu, & Pierini, 2016), but this method poorly represents travel data with long uninterrupted segments (Giakoumis, 2017). Other approaches involve temporal or spatial segmentations of predetermined sizes. These methods have the advantage of producing microtrips of a consistent desired length regardless of speed dynamics, but with the drawback of requiring a speed continuity criterion to produce realistically smooth speed profiles. Microtrips in this research are delineated at fixed spatial intervals – an approach which was reported to yield the most accurate driving schedules (Lin & Niemeier, 2003).  3.2.3 Assessment criteria The role of the assessment criteria is to ensure that the developed schedules represent the important characteristics of the calibration data set. Thus, assessment criteria should be   16  selected that are relevant to the purposes of the schedule. Target parameter values for the assessment criteria are calculated for the calibration data set firstly, and then schedules are constructed to reproduce those parameters as closely as possible. We propose twelve assessment parameters for biking schedules as listed in Table 1: Assessment criteria for biking schedules. These parameters were adopted from the driving schedule literature to represent speed and acceleration dynamics, with new parameters added for road grade.  Table 1: Assessment criteria for biking schedules  ID Parameter1 Abbreviation Units 1 Average trip speed ATS km/hr 2 Average running speed (𝑣 > 0) ARS km/hr 3 Average absolute grade AAG % 4 Average absolute acceleration AAC km/hr/sec 5 Percentage time idling (𝑣 = 0) PTI % 6 Percentage time accelerating (𝑎 > 0) PTA % 7 Percentage time decelerating (𝑎 < 0) PTD % 8 Percentage time cruising (𝑣 > 1, −0.1 < 𝑎 < 0.1) PTC % 9 Percentage time positive grade (G > 0.5) PTPG % 10 Percentage time negative grade (G < -0.5) PTNG % 11 Average positive work per distance2 (𝑎 > 0) APW m/sec2 12 Speed acceleration grade probability distribution3 SAGPD % 1 𝑣 is speed in km/hr, 𝑎 is acceleration in km/hr/sec, G is grade in % 2 expressed as the “Positive kinetic energy” in (Seers et al., 2015) APW = 1𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒∑ (𝑣𝑖+12 − 𝑣𝑖2) ∀ 𝑣𝑖+1 − 𝑣𝑖𝑛−1𝑖=1  3 percentage time in each cell of a 3-D speed-acceleration-grade distribution matrix with speed intervals of 5 km/hr, acceleration intervals of 0.2 km/hr/sec (Brady & O’Mahony, 2016; Seers et al., 2015), and grade intervals of 1%  3.2.4 Testing and selection The Performance Value (PV) is a single aggregate indicator of the set of assessment criteria, used to evaluate the biking schedules. The PV for each parameter (i) is the absolute percent difference between the schedule and target parameter values   17  𝑷𝑽𝒊 =  |𝒊𝑻𝒂𝒓𝒈𝒆𝒕 − 𝒊𝑺𝒄𝒉𝒆𝒅𝒖𝒍𝒆𝒊𝑻𝒂𝒓𝒈𝒆𝒕| ∙ 𝟏𝟎𝟎%, with the exception of SAGPD for which the PV is root mean square error (RMSE) between the target and schedule time fractions in each cell of the distribution matrix. The total schedule PV is then the weighted average of individual 𝑃𝑉𝑖. Weights are distributed equally among parameters that represent different characteristics, namely, speed, acceleration, grade, and SAGPD. A lower PV means that the schedule is closer to the target parameters and more representative of the calibration data set.   𝑷𝑽𝑻𝒐𝒕𝒂𝒍 = 𝟎. 𝟐𝟓(𝑷𝑽𝑨𝑻𝑺 + 𝑷𝑽𝑨𝑹𝑺 + 𝑷𝑽𝑷𝑻𝑰 + 𝑷𝑽𝑷𝑻𝑪)/𝟒+ 𝟎. 𝟐𝟓(𝑷𝑽𝑨𝑨𝑪 + 𝑷𝑽𝑷𝑻𝑨 + 𝑷𝑽𝑷𝑻𝑫 + 𝑷𝑽𝑨𝑷𝑾)/𝟒+ 𝟎. 𝟐𝟓(𝑷𝑽𝑨𝑨𝑮 + 𝑷𝑽𝑷𝑻𝑷𝑮 + 𝑷𝑽𝑷𝑻𝑵𝑮)/𝟑 + 𝟎. 𝟐𝟓(𝑷𝑽𝑺𝑨𝑮𝑷𝑫)  3.2.5 Biking schedule construction In past research on driving schedules, the appended microtrips have been selected randomly (Amirjamshidi & Roorda, 2013; Hung et al., 2007; Seers et al., 2015), with sophisticated statistical methods such as Markov chain transition matrices (Lin & Niemeier, 2003), or with hybrid approaches (Giakoumis, 2017). In this research, biking schedules are constructed using three different methods, namely, random, best incremental, and single cluster approaches. The details of each approach are described in the following subsections. Figure 5 summarizes the overall framework for constructing biking schedules.    18   Figure 5: Framework for constructing biking schedules with three different methods  For all three methods, the first microtrip in the biking schedule is randomly selected from a subset of microtrips that define a start of a trip. Then, additional microtrips are appended, without repetition, until a schedule duration of 25 minutes is reached, consistent with the 10-30 min driving schedules common in the literature (Hung et al., 2007). Each subsequent microtrip must possess initial speed and grade values that fall within a certain interval with the end of the previous microtrip in order to produce smooth speed and grade profiles. This condition is referred to as continuity criteria in this research. The construction methods differ   19  primarily in how they select the next microtrip from among those meeting the continuity criteria. For each method, several candidate biking schedules are constructed and the best schedule is selected based on lowest PV. The three considered methods are as follows: 3.2.5.1 Random Selection Approach In this approach, microtrips are randomly selected and appended, restricted only by the continuity criteria, until the target duration (25 min) is reached. Consistent with previous studies (Amirjamshidi & Roorda, 2013), new biking schedules are repeatedly constructed until 20 candidate schedules with PV<15% (referred to as PV threshold) have been generated, and then the preferred schedule is selected from those candidates based on lowest PV. Biking schedules with PV exceeding the threshold are discarded. 3.2.5.2 Best Incremental Approach In this approach, also adapted from previous studies (Brady & O’Mahony, 2016; Lin & Niemeier, 2003), microtrips are first clustered by average speed, average acceleration, and average grade using K-means clustering algorithm. The number of clusters is selected based on the sum of squared errors (SSE) within clusters. Figure 6 shows the relationship between the number of clusters and SSE for the small data set used in the proof-of-concept study. The SSE decreases with more clusters (i.e. they become more similar), but with a drawback of having fewer microtrips in each cluster. The optimal number of clusters is likely context-dependent. For the proof-of-concept study, microtrips are grouped into nine clusters.   20   Figure 6: Relationship between number of clusters and sum of squared errors within clusters   After grouping the microtrips into clusters, a transition matrix is generated to represent the probability of transitioning between clusters, based on observed sequences of clusters in the calibration data. A stochastic Markov chain process then generates a sequence of clusters starting from the cluster of the randomly selected initial microtrip (Ashtari, Bibeau, & Shahidinejad, 2014). Then, for each successive cluster: 1. Microtrips in the cluster are filtered to identify candidate microtrips that meet the continuity criteria with the end of the existing (incomplete) schedule,  2. Candidate microtrips are individually appended to the schedule and an interim PV calculated for each,  3. Candidate microtrips are ranked by lowest interim PV, and 4. The best microtrip is appended to the schedule.    21  This process is repeated for each successive cluster in the sequence until the desired schedule length is reached. Consistent with the random selection method, 20 biking schedules are generated using this approach and the best schedule is selected based on lowest PV.  3.2.5.3 Single Cluster Approach This approach is similar to the best incremental approach, but discards the Markov process by combining all microtrips into a single cluster. After selecting the first microtrip randomly, successive microtrips are appended based on fulfilling continuity and lowest interim PV. Because this is no longer a stochastic process, each starting microtrip generates a single deterministic schedule. Hence, the number of unique candidate biking schedules from this method depends on the number of starting microtrips in the pool. As with the other approaches, the best biking schedule is selected from among the candidates based on lowest PV. 3.3 Proof-of-concept study A set of existing on-road cycling GPS data was used to demonstrate and evaluate the proposed biking schedule methods. The data set contains 55 hours of 1 Hz speed and grade data from three cyclists (A, B, and C) in Portland, Oregon (Bigazzi & Figliozzi, 2015). It was noticed that speed and grade values sometimes reach extremely high or low values with a lot of noise. Remedying was carried out by capping grade values at ± 10%, and then smoothing speed and grade data using kernel smoothing algorithm with bandwidths of 3 and 10, respectively (Nouri & Morency, 2017). Different smoothing algorithms including different kernel bandwidths were investigated to reach the best settings that provide realistic speed and grade values while still preserving the physical variations.    22  Microtrips were generated only for cyclist A (for which the most data are available) at fixed spatial interval of 250m as recommended in previous research (Nouri & Morency, 2017). A total of 1,530 microtrips were extracted from the cyclist A data; partial microtrips were discarded.  The continuity criteria considered for this study were determined by two factors. First, we considered the acceptable change in speed and grade between two successive observations. Second, we considered an optimal continuity criteria that leave a considerable amount of microtrips candidates. After several trial and error iterations, we decided to fix the continuity criteria at 2 km/hr speed and 2% grade. Narrower ranges are more likely to filter out more microtrips, which leaves fewer candidates standing up for appending.  Biking schedules were generated for cyclist A and then validated in two ways: by testing application to estimates of cyclist power and breathing rate, and by testing transferability to the other two riders and lower data resolutions.  Power out and breathing rate are calculated using the equations below (Bigazzi & Figliozzi, 2015). ?̇?𝑵 represents the rate of work. It is the summation of four parts: change in kinetic energy, change in potential energy, air resistance, and rolling resistance.  ?̇?𝑴 is the total rate of work, it is calculated as the maximum of 0 and ?̇?𝑵, assuming that negative values represent braking. It should be mentioned that the subjects used in proof-of-concept analysis are the same subjects from which these models were developed (Bigazzi & Figliozzi, 2015). Hence, the masses and resistance values were taken directly from this research and they are shown in Table 2. ?̇?𝑵 =  𝒎𝑻𝟐∆𝒗𝒃𝟐∆𝒕+ 𝒗𝒃𝒎𝑻𝒈𝑮 +𝟏𝟐⁄ 𝛒 𝑪𝑫𝑨𝑭𝒗𝒃𝟑 + 𝒗𝒃𝑪𝑹𝒎𝑻𝒈         23  ?̇?𝑴 = 𝒎𝒂𝒙𝒊𝒎𝒖𝒎 {0, ?̇?𝑵, }   𝐥𝐧(?̇?𝑬)𝒕 =  𝜶 +  𝜷𝑻?̇?𝑴 +  𝜺           Where: ?̇?𝑁 = rate of net work (watt), ?̇?𝑀 = rate of total bicyclist work (watt)  𝑚𝑇 = total mass of the bicycle and rider system (Kg), 𝑣𝑏 = bicycle speed (km/hr), 𝑔 = gravitational acceleration (m/sec2), 𝐺 = road grade (unitless), ρ = air density (kg/m3),  𝐶𝐷 = drag coefficient (unitless), 𝐴𝐹 = frontal area of the bicyclist (m2), 𝐶𝑅 = coefficient of rolling resistance (unitless). ?̇?𝐸 = ventilation rate (liters/min)  α, = intercept value, and 𝛽𝑇 = slope   Table 2: Parameters used in power and breathing rate calculations Parameter Subject A Subject B Subject C Total mass (kg) 105 91 97 𝐶𝑅 0.004 0.004 0.004 𝐶𝐷′  = 0.5ρ𝐶𝐷𝐴𝑓 0.6 0.4 0.4 α 2.185 2.674 2.318 𝛽𝑇 0.00744 0.00417 0.00761    24  Transferability to other riders is tested by generating biking schedules for cyclists B and C based on their aggregate travel characteristics (assessment parameters), but using the microtrips from cyclist A. The rationale for this approach is to test the possibility of developing biking schedules for other riders and conditions knowing only aggregate riding characteristics. To evaluate the effect of GPS data resolution on biking schedule generation, coarser logging intervals of 3, 5, and 10 seconds are simulated from the original data set by deleting observations. Acceleration is recalculated as the difference between consecutive velocity observations, and new biking schedules generated from the revised microtrips and assessment parameters. 3.4 Data collection 3.4.1 Introduction The data set used in the proof-of-concept study was limited to only three cyclists. We aimed at collecting a larger naturalistic data set that comprises of more cyclists (including electric bikers) in more diverse cycling conditions.  The target population in this research was all cyclists over 16 years old who get on their bicycles at least once a week in Metro-Vancouver, BC, Canada. The data collection presented in this thesis was part of a project aimed at investigating the factors that influence the role of electric bicycles in urban transportation systems. Part of this project requires collecting information about cyclists’ perception about electric bikes in British Columbia. Thus, this survey was also aiming at electric bike users in the region.   25  3.4.2 Survey design The data collection method was conducted via an online surveying tool “Fluid surveys” supported by the University of British Columbia (UBC) for easy accessibility and data management. The survey is presented in the Appendix. The survey consists of three main parts. First, as per UBC Behavior Research Ethics Board requirements (BREB), all participants had to provide their consent to participate in the survey. The concept form is presented in Appendix A. The second part aimed at collecting socio-demographic information and cycling preferences and experiences via a questionnaire (Appendix B). The questionnaire was derived from two sources: literature (MacArthur et al., 2014), and a previous survey carried out at UBC (Tengattini, 2017). The third part (Appendices C and D) asked participants to log all their active travel for one week. Participants were asked to record GPS data for their trips in addition to identifying trip purpose and mode. Optionally, participants were also asked to use a heart rate monitor during their travel to track their heart rate. Participants were given two options to provide their GPS data as follows:  1) Use a smartphone application that uses the phone’s built-in GPS. 2) Use any other GPS device or smartphone application they prefer and then manually upload the data on the survey website. 3.4.3 Smartphones vs stand-alone GPS devices Initially, there were two alternatives for collecting GPS data; smartphones and stand-alone GPS devices. The smartphone alternative was preferred over the stand-alone GPS for the following reasons:   26  1) Expenses: Cost is one of the most influencing factors in decision making, especially in academic researches where funding is limited. Choosing stand-alone GPS devices would require purchasing several devices at an average cost of $300 each. On the other side, nowadays, the majority of people own smartphones, which eliminates the cost for purchasing GPS devices.  2) Accessibility: As previously mentioned, the majority of people nowadays have access to smartphones. Whereas, assuming that stand-alone GPS devices were provided, participants would still need to provide their mail address and put on a waiting list till a device becomes available. This would cause several problems as follows  I) Asking for participants’ mail addresses should be avoided whenever possible since this would introduce risk of exposing their location and potentially their identity. II) Creating a dynamic waiting list that automatically sorts participants requires importing data from Fluid Surveys. Building a robust system that automatically updates data requires a long time.  III) There is a risk of late delivery of GPS devices, since mailing GPS devices by express post takes about 2-3 days, not to mention other 2-3 days until the previous GPS user returns the device assuming they are not late.  IV) Putting participants on a waiting list would result in losing some participants. Most people are constrained by a schedule and cannot wait until a device becomes available.   27  3) Forgetting device at home: Generally, participants are more likely to forget a stand-alone GPS rather than their own smartphones. Choosing the smartphone alternative will provide a solution for this issue and so reducing the chances of losing some trips. 3.4.4 Choosing a smartphone application for the survey Adopting a suitable application was a crucial step in the survey. Choosing an inappropriate application for the research could lead into yielding invalid results. We tested different applications and compared them based on different aspects as shown in Table 3.      28  Table 3: Comparison between different smartphone applications Application Strava Ride with GPS Runkeeper Endomondo Runtastic Map my ride Logging interval 1-12 sec 1-18 2-8 2-7 2-10 1-11 Subscription fees Free Free Free Free Free No Premium account (rounded values) $11 monthly or $85 yearly Basic1:   $8 weekly Basic 2:  $5 weekly Premium: $13 monthly $11 monthly,  or $44 yearly $7 monthly or $30 yearly $11 monthly or $58 yearly  $6.5 monthly or $33 yearly Free month premium account No No No Yes NA No Sign in difficulty Easy (Fb/Google) Easy (Fb/Google) Easy (Fb/Google) Easy (Fb/Google) Easy (Fb/Google) Easy (Fb/Google) User Interface Easy Easy Easy Medium Easy (similar with FB) Easy Adding friends Yes Yes Yes Yes Yes Yes Download trips for other users Yes(premium/third party website) Yes NA Yes Yes NA Record missed trips Yes (manual/ GPS/import) Yes (manual/GPS/ import) Yes(manual) Yes (manual/GPS/ import) Yes (manual/ import) Yes (manual/GPS/ import) Pairing with HRM Yes Yes Yes Yes Yes Yes Creating groups Yes No No No No No Joining clubs Free No No No No No Data export formats gpx/crs gpx/tcx/kml gpx/kml gpx/tcx gpx/tcx/kml NA Auto-Pause option Yes (can be turned on/off) Yes(activated by default) Yes (can be turned on/off) Yes (can be turned on/off) Premium accounts only Yes (can be turned on/off) Sharing on social media Yes Yes Yes Yes Yes Yes NA in table means either the information is not available or it was not collected.   29  By comparing these six applications, it was finally decided to choose “Ride with GPS” for several reasons. The application was free to subscribe and easy to use. Plus it can record GPS at 1-second intervals and compatible with several Bluetooth heart rate monitor devices. The application also had another feature which made it the most preferable. The application allows users to add other members to their friend list. Members of the friend list can access and download their friends’ trips. This feature gave us unlimited access to download and save all participants’ trips. Other applications either require users to purchase a premium subscription or do not allow downloading trips for other users. 3.4.5 Choosing a heart rate monitor Choosing a suitable heart rate monitor for the survey was a relatively easy task given the wide variety of options in the market. The heart rate monitor was chosen based on a comparison between six different alternatives as shown in Table 4. Table 4: Comparison between different heart rate monitors Heart rate monitor Wahoo TICKR Polar Loop Polar H7 heart rate sensor Fitbit Flex 2 Mio Link Rhythm+™ Price ($) 50 120 100 100 100 80 Compatibility link  link  IPhone 4S or above Android 4.3 (Jeally bean) + link iphone 4S + Android 4.3 (Jeally bean) + Iphone 5+ Sam S3 + 4.3 (Jelly Bean) Battery life NA link 200 hours 5-10 days 6-8 hours 8 hours Strap Chest wrist Chest wrist wrist Arm  After comparing all proposed heart rate monitors, it was decided to use Rhythm+™ (SCH-RTHM19), shown in Figure 7, for several reasons. The monitor is an arm band which is much more comfortable and easier to wear than chest straps. The monitor was also   30  compatible with a wide range of Android and iPhone devices which ensured that participants will not have problems using it. Additionally, it was capable of recording heart rate data at 1-second interval which is important for future studies combining heart rate data with travel dynamics such as speed, grade, etc.  Figure 7: Rhythm+™ heart rate monitor  3.4.6 Preparing a survey plan After identifying the survey equipment, a clear feasible plan had to be laid out. The plan helps in administering the survey by thinking through each step and how each step feeds in   31  the other. The plan also helps in thinking about every possible scenario that might happen. It is also useful in explaining the survey procedures to other team members.  3.4.7 Following up emails One of the most important tools to keep participants on track and ensure that they do the requested tasks correctly is by continuously following up with them. Sending emails was the most convenient method to follow up with participants because it can be done remotely with minimal costs. We designed a total of 17 emails for every possible situation.   3.4.8 BREB approval Conducting any survey that involves human subjects requires obtaining an approval from the Behavior Research Ethics Board (BREB) in advance. The BREB approval is considered the last step before the survey is ready to be deployed. An application with all research and survey details was submitted to the BREB. The application usually takes about a month before the board reviews the application. The BREB response requested several changes, but their main concern was about the usage of a third-party application for collecting GPS data. Initially, we provided participants with only one option to record their trips, which was through “Ride with GPS”. The research team did not have any administrative role over the application which might expose the identity of participants. The BREB asked for several justifications for the reliance on a third-party application and a clear plan for keeping the identity of participants secure. In response to their comments, we amended the survey so that it allows participants to log their trips using any application or GPS device of their choice. The participants then have to manually upload trips on the survey website. This change gave participants plenty of alternatives if they do not have access to a smartphone or do not want to use “Ride with GPS”. We also asked all participants to change the privacy settings on the   32  application so that their information is kept secure. Other BREB comments were minor, they mainly asked to amend some sentences to align with BREB standards. 3.4.9 Pilot surveying Pilot testing ensures that the survey procedures are smooth and thorough. We asked five volunteers to try the survey and report back immediately if they find any of the steps inconvenient. The subjects were chosen so that they are completely unfamiliar with the survey in order for them to have the first user experience. Pilot surveying was beneficial in drawing attention to some issues that were not considered while laying out the survey plan. Those issues are as follows: 1)  The need for automation: Follow-up emails were originally planned to be sent separately to each participant via Microsoft Outlook. This process was found to be infeasible and very time consuming since these emails contain information that need to be exclusively personalized for each participant such as start and end dates of trip logging. Moreover, we wanted to follow up with participants throughout their recording week by sending daily emails. These emails need to be scheduled for each of the participants according to their start date. The email automation was achieved by two methods as follows: a) Fluid Surveys automation: Fluid survey provides a feature that immediately sends emails to participants if they followed a specific pattern in their answers. Unfortunately, this feature could not schedule emails to be sent at a later time. Thus some follow up emails, especially those tied to their start dates, could not be automated using this feature. Accordingly, this feature could not be used to   33  automate all emails, which brought up the need to rely on another method to automate the remaining emails. b) Integration between Microsoft Excel and Outlook: Microsoft Outlook provides a feature that allows scheduling emails. However, Microsoft Outlook does not integrate with Fluid Surveys in order for the emails to customized and scheduled according to the responses. On the other hand, Fluid Surveys provided an application programming interface (API), which allows all survey responses to be downloaded in xlsx format. The API was then integrated with R analysis software to download responses every day at specific times. The script coded in R also copies the downloaded responses to another excel sheet with pre-designed Macros that schedule and customize all remaining emails automatically be integrating with Microsoft Outlook. The research team was only responsible for activating the Macro only once for each participant.    2) Minor sentences amendments: Some subjects reported that some instructions/questions were not clear or might confuse readers. Those sentences were amended with a clearer and simpler phrases. 3.4.10 Recruitment Survey recruitment was carried out via two methods. The first method is online recruitment via emails and social media. The second method is field recruitment via flyers and invitation cards. The following subsections will discuss each method briefly. 3.4.10.1 Online recruitment The online recruitment was mainly done via invitation emails and advertising through social media as follows.   34  Invitation emails: We contacted several cycling groups and organizations and asked them to invite their members to participate in the survey. Only a few organizations responded by posting our survey link and a brief description on their website or social media. We also keep an email list of all previous participants who would like to receive research results or be contacted for future studies. We sent emails to participants from a previous study to notify them of the study results and invite them to participate in our new survey. This method was found to be very successful. Social media: The social media nowadays comprises people of all ages (Pfeil, Arjan, & Zaphiris, 2009; Sloan, Morgan, Burnap, & Williams, 2015). Most of the meetups and cycling events are organized through social media. We made use of this opportunity by creating a new Facebook page for our research team. We advertised about the survey on the page and invited our friends who are enthusiastic about cycling or who would know someone who might be interested in participating. Although this method might be effective in many different situations, it was inefficient for our case. The new social media page did not get much attention because it was difficult to get a large number of views. It would have been more efficient if the page was already created long time ago and has a big number of followers.  3.4.10.2 Field recruitment Field recruitment was carried out via two methods: 1) Flyers: We designed two versions of flyers (with and without tear-offs) as shown in Figure 8 and Figure 9. The flyers provide a constant but slow recruitment rate. The flyers should be posted in places not only regularly visited by cyclists but also visible.   35  Therefore, we located all bike shops and service locations in Metro-Vancouver. These locations were visited by the research team to ask them to post flyers on the walls and advertising boards. We also asked them to put some invitation cards (shown in Figure 10) on the reception desk in a position visible to customers. We especially targeted electric bike shops in Metro-Vancouver to reach out to electric bike users.  2) On Street recruitment: This is one of the most effective methods of recruitment. The research team visited eight locations in Metro-Vancouver. Figure 11 shows a map of the visited locations. The number between brackets represents the number of times this location had been visited. The choice of the locations is crucial for the success of recruitment. The locations should be strategic and dense with cyclists. After scoping most municipalities in Metro-Vancouver, it was found that most cyclists are concentrated in UBC, Vancouver, and Downtown. Few cyclists were observed in other municipalities such as Surrey, North Vancouver, Burnaby, and Richmond which indicated that recruitment in these locations will not be efficient. It was also found that the majority of cyclists travel from Vancouver or Downtown to these municipalities and then disperse according to their home locations. It was difficult to find a location dense with cyclists in these regions. Therefore, we targeted locations where cyclists are more likely to pass through before dispersing. For example, Canada Line Bikelane is a strategic location where all cyclists have to pass through in order to travel between Vancouver and Richmond. The research team borrowed a truck owned by the University of British Columbia Civil Department to transport shade cover, chairs, trolley, and posters. Once established at the recruitment location, the research team starts talking to cyclists who   36  stop. Cyclists were given a brief talk about the survey and left with an invitation card. This recruitment method allows more direct interaction between potential participants and researchers, opening more channels for discussions to convince cyclists to take part in the survey.     37   Figure 8: Survey flyer   38   Figure 9: Survey flyer with tear-offs   39    Figure 10: Survey invitation card   40    Figure 11: Locations vistied for field recruitment      41  3.4.11 Incentives: Incentives are a symbol of appreciation from researchers to participants. The price of the incentives should not be very expensive so that participants take part in the survey only for receiving incentives rather than their willingness to provide data for scientific research. There were two main types of incentives used in this survey: 1) Amazon electronic gift cards: All participants were entered in draw for 20 amazon gift cards valued at $25 each. The winners were announced at the end of the survey.  2) Hats and Socks: These incentives (shown in Figure 12) were exclusive to participants who recorded their heart rate.   42   Figure 12: Hats and socks incentives used in the survey    43  3.5 Data processing Most of the time, the collected data are not in a proper shape or format for analysis. The collected data in this research mostly came from smartphones built-in GPS and downloaded from “Ride with GPS” in tcx format. This format contains second-by-second raw time stamp, longitude, latitude, attitude, and heart rate (if available). These data had to be processed first in order to get reliable velocity, acceleration, elevation, and grade values.   3.5.1 Preliminary processing 3.5.1.1 Data format conversion The objective of this step is to convert the format of collected data into another format which is compatible with the analysis software R. The collected data in this research came from two sources: 1) Data downloaded from “Ride with GPS” in tcx format. This format can be directly imported in R using the “XML” package, which can read tcx files and convert them into data frames. These data frames had two deficiencies. The first deficiency is that the numerical values (such as longitude, latitude, etc) are read as characters. This has been fixed by applying a function that converts characters into numerical values. The second deficiency is that the columns had complex names. For example, the latitude column was named “value.Position.LatitudeDegrees”. This column has been renamed to “Lat” for simplicity. 2) Data manually uploaded by participants on “Fluid Surveys”. These data were in various formats, namely, gpx, fit, and tcx. The fit and gpx formats cannot be directly imported in R. gpx format can be imported to Microsoft Excel which can then be   44  opened in R. The bigger problem was with the files in fit format. The format is not compatible with either R or Excel. Therefore, these files needed to be converted into a different format first with a different conversion tool. There were several online websites that offer this service for free. However, to keep the data secure and anonymous, it was necessary to avoid them. We decided to use “Garmin Training Center” offline tool which can convert between fit and gpx formats. gpx format can then be opened in Excel as previously discussed.  3.5.1.2 Filtering questionnaire responses The questionnaire responses were downloaded from “Fluid Surveys” in xlsx format. The downloaded files contained several duplicate and empty responses. Since the survey was online, several participants only provided their consent and did not answer any questions which created the empty responses. Whereas duplicate responses were created from participants who logged out of the survey and tried to modify or complete their responses at a later time. The survey was built in a way that enables participants to re-login their responses only if they used the same device and internet browser. The survey relies on cookies saved on the internet browser to recall the participants’ information when they try to re-login. This information was mentioned in the consent form and in almost all of our email reminders. The survey website also provided participants at the end of their response with a link (which can be emailed to them directly if they choose to) that directs them to their response. Some participants ignored or didn’t read this information. If participants used another internet browser or device to try to re-access their responses, they will be logged in as new users and will be asked to fill the whole survey from the beginning. This led into having duplicate responses with the same email address.   45  The questionnaire responses were filtered by deleting the empty responses. Whereas duplicate responses were filtered as follows: If the questionnaire was answered more than once by the same participant (identified by checking entered email addresses), the first response is always considered more accurate. It is presumed that participants would put random answers if they are answering the questions for the second time. There were 18 participants who entered duplicate responses For most cases, participants re-login either to register for the GPS recruitment or to change some responses concerning their recording week such as start date, recording method, or to request a heart rate monitor. For these cases, responses were merged. Questionnaire responses from their first response were merged with latest responses from the GPS recruitment page. 3.5.1.3 Coding To keep the identity of the participants anonymous, it was necessary to give participants codes. The code started with “PRT” (standing for participant) and then a serial number of three numbers with leading zeros representing the order by which they took part in the survey. For instance, the first participant took a code “PRT001” and the fifteenth participant took a code “PRT015”, etc.  3.5.2 Speed calculation Speeds were calculated purely from longitude, latitude, and time stamp. We calculated the moved distance between each two consecutive points by using the “geosphere” package in R (Robert J. Hijmans, 2016). The package contains a function “distm” that calculates the   46  distance (in meters) between two longitudes and latitudes assuming that Earth is a sphere with a radius of 6,378,137m.  The time interval between the observations was not always one second. There were some gaps and duplicate time stamps in data (representing 70% and 0.8%, respectively) due to signal losses, and other factors related to the processing speed of the GPS signal receiver. For duplicate time stamps, we only left the first recorded observation and deleted others with the same time stamp. Time gaps refer to consecutive observations where the time difference between them is more than one second. Speed is calculated as the moved distance between every two consecutive observations (regardless the presence of gaps) divided by the time difference.  3.5.3 Extracting stop periods There were several participants who recorded more than one trip on the same file. These files would give wrong aggregate trip characteristics such as duration and average speed. Therefore, it was necessary to separate these trips. The objective of this step is to create a table that contains all stops longer than five minutes, assuming that this is the longest duration a cyclist would stop during a trip. Other researchers (Strauss & Miranda-Moreno, 2017) assume 90 seconds for this duration. However, we decided to select five minutes instead because we believe that cyclists in Metro-Vancouver can stop for more than 90 seconds at intersections.  The GPS device takes readings even if the user is not moving, and since the GPS is not precise, the readings would move around the same location as shown in Figure 13. Therefore,   47  speeds calculated from raw GPS data were insufficient to extract stop periods. It had to be combined with moved distance in order to determine the motion state of the user.   Figure 13: Example of GPS points at stops locations  There was another issue with using GPS devices which is signal losses. Signal loss can occur when a user is moving on a street with tall trees/constructions or in buildings where concrete walls block the signals. In order to solve this issue, we inserted empty rows where gaps in the data are present. We did not infer any information, we only inserted rows with no information so that the final number of rows in the each trip matches the trip duration in seconds. A new speed column was then created. Speeds in this new column were calculated as previously discussed, and for the time gaps, the inserted rows took a constant speed assuming no acceleration/deceleration was done during time gaps. For example, if the   48  participant moved 100 meters during a gap of 50 seconds, the speed values for all observations within this gap take a value of 100m/50sec = 2m/s. This new speed column was only used for extracting stop periods, it was not used anywhere else in the analysis. It should be emphasized that the purpose of this part is not splitting trips at stops longer than five minutes; we only identified the stopping periods and put them in a table. The methodology of extracting stop periods was derived from the literature (Cich, Knapen, Bellemans, Janssens, & Wets, 2016; Fu, Tian, Xu, & Qiao, 2016). All the speeds mentioned in the next steps refer to speeds from the new speed column. The methodology is applied to all trips individually as follows. 1) First, we identify all potential stop locations in the data set by testing two conditions for every group of 20 consecutive observations. Only one condition needs to be met for these points to be identified as a potential stop location. a. The average speed of the 20 observations is less than 5 km/hr. b. For the second condition we define and compare two distances: Theoretical Moved Distance (TMD) and Actual Moved Distance (AMD). TMD is calculated by multiplying the average speed of the observations (calculated from first condition) by the total time interval (20 seconds). TMD represents the theoretical distance a cyclist should have moved during 20 seconds . While the AMD represents the actual distance between the first and last points in these 20 observations. Conceptually, both distances should be close to each other if the cyclists is moving in one direction. However if the cyclist stopped and the GPS is still recording high speeds due to its inaccuracy, TMD would be significantly bigger than AMD. Therefore a potential stop location is   49  identified if the TMD is more than three times the AMD. We chose the factor of three to be conservative and make sure that the cyclist stopped and was only taking a turn. 2) If a potential stop location was identified, the distances between the first observation of the potential stop location and all successive observations in the trip were calculated. The goal of this step is to identify if the observations are getting farther from the first point or they fall within a vicinity from it. If the GPS are going farther, the distances should be continuously increasing, and if the points are falling within a vicinity from the first point, the distances should be varying within a range. A buffer of 35 meters (about two buildings length) is created around the first point. All observations are counted (if the distances from the first point is less than 35m) until there are 30 consecutive observations lying outside the buffer (see Figure 14). If the number of observations in the buffer exceeded 120 (more than two minutes), this location is identified as a true stop location. The reason for choosing two minutes at this step instead of five minutes is because there is a relationship between the buffer size and the number of points lying in it. Bigger buffers will include more points and might confuse stationary cyclists with cyclists moving at slow speed. On the other hand, smaller buffers will include less points which might not be enough to identity if cyclists came to a stop due to the noise in the data. The combination of 35m with 2 minutes was based on trial and error to get a combination that successfully captures all stops. This combination has been tested on more than 20 random trips from different cyclists and all stops longer than two minutes were identified successfully.   50  We also wanted to include stops occurring at the start and end of the trip, so two minutes seemed ideal.   Figure 14: Buffer around the first observation in the potential stop location  3) If the time difference between two successive stops locations is less than one minute, these two stops are merged together in one stop. 4) All stops are then filtered to exclude all stops that are shorter than five minutes. This step does not apply to stops occurring at the start and end of the trip. 5) To find the start/end of the stop location more precisely, we search within the points identified in the stop location for first observation that possesses the following.  a. Speed less than 5km/hr b. If it is a start of a stop, it has to be followed by two data points with higher speeds. If it is an end of a stop, it has to be preceded with two data points with higher speeds.    51  For example, consider some observations with speeds data as follows “20, 15, 6, 3, 1, 2, 2, and 1”. In this example, we are looking for the start of the stop. The fifth point meets both conditions; it has a speed lower than 5 km/hr and the following two speed points have higher speeds. After extracting a stop duration, the whole process is repeated to find another stop starting from the last observation of the previously identified true stop location. All stops are finally gathered in a table that indicates the trip code, participant number, and stop start/end time stamps. 3.5.4 Data filtering There are four filters applied in this research. 1) Heart rate filter: It was noticed while reviewing some trips that sometimes the heart rate values reach thousands beats per minute. This occurs once and measurements go back to normal. These measurements represent 0.03% of the heart rate data and they were filtered by deleting any heart rate value exceeding 500. Since heart rate was not used in this research, we wanted to keep the raw measured values and remove unrealistic values at the same time. Human heart can reach almost 200 beats per minutes during vigorous exercise (Fred Dyck, 2016). Therefore, we decided to go slightly above 200 to ensure that the data are not altered given that there might be measurement errors. Further filtering may be required in the future if the heart rate data were sought. 2) Low speed filter: Since speed was purely calculated from raw GPS data (longitude and latitude), the calculated speeds would still indicate movement even if the user was not   52  moving (at stationary position). This filter is different from the stop periods filter; it identifies stationary positions regardless of its duration and replaces calculated speed values with zero. The filter works as follows: a. First we identify all observations with speed values less than 5 km/hr. The literature suggest that cyclists come to a stop if the raw GPS speed is below 2km/hr (Langford et al., 2015). However, due to the noise in our data, this value was not high enough to capture all stops.  b. If there are observations less than 10 seconds apart, these observations and all observations in between are put together in a group. This step compensates for the inaccuracy of the GPS device which might result in observing several rapid movements while the user is not actually moving. c. For each group, the TMD and AMD are calculated. If the TMD is three times higher than the AMD, this group is identified as stationary activity and all speeds in this group are replaced with zero. The exact start and end of the stationary activity are identified similarly to the approach discussed in identifying the start and end points of stop periods.  3) High speed filter: This filter deals with all unrealistic high speeds in the data set. Speeds that are 60% higher than both the previous and following observations are deleted. The 60% is based on trial and error iterations to capture all noticeable high values. 4) Stop period filter: The stop period table (extracted from the previous section) is used to split trips recorded on the same file. The stop periods are removed and for a trip to   53  be separated, it has to meet two conditions (Fu et al., 2016). Otherwise it is deleted. These two conditions are as follows: a. Average speed higher than 5 km/hr. b. Total duration longer than 1 minute. 3.5.5 Elevation data The GPS accuracy mainly depends on signal strength between the GPS device and satellites. The altitude recorded by GPS usually suffers from several errors due to exposure to multipath effects, several atmospheric layers, natural factors, obstruction of the satellites by buildings, mountains, trees, etc.(Menard, Miller, Nowak, & Norris, 2011). Therefore, it was necessary to rely on another source to get elevation data.  Extracting elevation data from Digital Elevation Models (DEM) is a very common practice in research (Casello & Usyukov, 2014; Strauss & Miranda-Moreno, 2017). In this research the elevation data were extracted using the Canadian Digital Surface Model (Canadian Digital Surface Model, 2018). The elevation data were extracted using the “extract” function in the “raster” package in R (Robert J. Hijmans, 2017).  3.5.6 Road grade calculation Road grade is calculated as the difference in elevation divided by the travelled distance between every two consecutive points. In cases where there are missing GPS points, the cumulative distance between the last available GPS points are used instead. The aforementioned grade calculation method gave reasonable grade values when the cyclists is moving. However, if the cyclist stopped (i.e. distance moved = zero), the grade returns infinity. To solve this problem, the R script was amended to replace all infinity values   54  with the last calculated grade value assuming that road grade does not change when a cyclist stops. After calculating road grades for the whole data set, the grade values were capped at ± 10% to eliminate all unrealistic steep values (20% of the grade data were removed). These high values arise due to the inaccuracy of the DEMs. Small movements are likely to reflect unrealistic grade values if the elevations extracted from DEMs are inaccurate. 3.5.7 Data smoothing We applied smoothing to speed and road grade values as follows. 1) Speed smoothing/inferring: Speed inferring was carried out for two reasons. Firstly, most of the smoothing algorithms do not work if there are missing data. Secondly, we pursued increasing the completeness of the data set by interpolating speeds if enough data around the missing points are present. We inferred speed data via linear interpolation. For example if we have a set of data points with speeds of “10, 9, NA, 8, NA, NA, 5”. The script will return “10, 9, 8.5, 8, 8, 7, 6, 5”. The interpolation was not applied in cases where there are more than 5 consecutive missing speed data. Thus, big gaps are not filled. After inferring missing speeds, kernel smoothing with bandwidth of 10 is applied. We tried different bandwidths and different smoothing algorithms such as moving average, spline smoothing, local polynomial regression, and Savitzky–Golay. The kernel algorithm outperformed other methods in two way. It conserved zero speeds efficiently and provided realistic speed dynamics. Higher bandwidths would lead into losing some speed details and narrower bandwidth will make speed transitions very sharp. Acceleration is calculated for each observation in the data set as the difference in the smoothed speeds.   55  2) Grade smoothing: grade values were smoothed also by using kernel smoothing algorithm with a bandwidth of 10 after trying different smoothing algorithms and bandwidths. 3.6 Testing optimal microtrip length As previously mentioned, the biking schedule construction methodology was derived from the driving schedule literature. Some of the parameters were directly implemented in the methods without validating its compatibility with bicycle travel analysis. In an effort to enhance the biking schedule construction methods, the collected data set was employed to study the optimal length of microtrips and to study the impact of using different lengths and definitions. In a previous study (Nouri & Morency, 2017), different microtrip definitions and lengths were tested to identify the optimal microtrip definition for driving schedules. A similar approach was adopted in this research but for the purpose of generating biking schedules. The researcher generated several biking schedules using different microtrip definitions. The best definition is identified by looking for the definition that yields the biking schedule with the lowest PV.  Taking into consideration that biking schedules are likely to have different characteristics between different trip purposes, a subset of the collected data was only considered in this research by excluding non-utilitarian trips i.e. exercise trips. Also it was necessary to exclude trips (after filtering and smoothing) with less than 80% speed data since there might be several missing biking activities. Those trips represented 10% of our utilitarian data. In total, there were 1300 biking trips ready to be processed.   56  Seven different microtrip definitions were tested in this research. The filtered trips were processed to generate microtrips every: 100, 150, 200, 250, 300, and 350 meters; in addition to the most common microtrip definition: Stop to Stop (STS). Twenty Biking schedules were generated using the best incremental approach from each definition. The definition that generates the biking schedule with the lowest PV is identified as the best/optimal definition.  The best incremental approach was preferred because it was the most time saving compared with other methods, which is an advantage regarding the enormous amount of GPS data present (as will be discussed in the results section). The methodology of constructing biking schedules for this study (testing optimal microtrip length) deviates from the proof-of-concept study as follows. 1) Number of clusters: The number of clusters was determined based on the percentage difference in sum of squares within clusters as shown in Table 5. The table shows the relationship between the number of clusters and sum of square error (SSE) within clusters for the 100m microtrip pool. As shown in the table, the SSE within clusters decreases as more clusters are added, such that the cluster center becomes closer to the points. The number of clusters is decided when there is less than 10% reduction in the SSE, meaning that adding more clusters does not reflect big improvements in the SSE. This approach is often referred to “elbow method”. For all microtrip pools, we found 15 clusters achieved the required results.     57  Table 5: Relationship between number of clusters and sum of squares within clusters for the 100m microtrip pool Number of  Clusters SSE Difference % Difference 2 2262887 3698073 62.04 3 1206698 1056189 46.7 4 749666 457032 37.9 5 505836 243830 32.5 6 360737 145099 28.7 7 269627 91110 25.3 8 212190 57436 21.3 9 172096 40094 18.9 10 143827 28269 16.4 11 123074 20753 14.4 12 103283 19791 16.1 13 90175 13108 12.7 14 79530 10645 11.8 15 71780 7750 9.74  2) Continuity criteria: We had larger number of microtrips compared with the proof-of-concept study. Therefore, we decided to consider tighter continuity criteria of 1 km/hr speed and 1% grade.  3.7 Comparison between the travel characteristics of electric and regular bikes  We expect cycling behavior to vary among different bike types, trip purposes, and cyclists’ characteristics. In order to conduct fair comparison between regular and electric bike trips, we normalized some factors that might cause the travel characteristics to vary among both types of trips. This was achieved by applying “Propensity Score Matching” (PSM).   58  PSM is a statistical approach which is commonly used in medical disciplines to study the effectiveness of a specific treatment. It is used to find a control group (people who did not take the treatment) that matches the characteristics of the treated patients. A similar approach will be followed in this research. PSM will be applied to find a set of regular bike trips that matches the available electric bike trips. Only trips with more than 80% speed data were considered in this analysis. We considered several characteristics in the matching criteria such as age, gender, home location, level of education, annual gross income, terrain, and trip purpose. The performance of the PSM performance depends on the proximity between the available regular and electric bike trips. Adding several characteristics in the matching criteria makes it harder to find exact matches. We tried different combinations as follows: 1) Age + Sex 2) Age + Sex + Home location 3) Age + Sex + Education level + Income 4) Age + Sex + Education level 5) Age + Sex + Income 6) Age + Sex + Trip Purpose 7) Age + Sex + Terrain 8) Age + Sex + Trip Purpose + Terrain Terrain was defined by the three assessment parameters relevant to road grade (i.e. Average absolute grade (AAG), Percentage time Positive Grade (PTPG), and Percentage Time Negative Grade (PTNG)). Age and sex were included in all combinations because they are expected to have the largest influence on trip characteristics. Only four of the eight   59  considered combinations were successful to find close matches (matches with p value above 0.05). Those four are: Matching 1: Age + Sex Matching 2: Age + Sex + Trip Purpose Matching 3: Age + Sex + Terrain Matching 4: Age + Sex + Trip Purpose + Terrain  The PSM output in R can be summarized in a table and results for the successful combinations are shown in Table 6, Table 7, Table 8, and Table 9. The tables report the number of matched samples and the number of trips in each category. For example, in Table 6, the number of matched regular bike trips were 143 (matching the 143 available electric bike trips). There were 9 trips whose cyclists were aged between 21 and 30 in the electric bike sample and they represent 6.3% of all electric bike trips. PSM can be used to match both discrete and continuous random variables. In case of matching continuous random variables, the table reports the mean and standard deviation instead. PSM also uses chi-square test with confidence interval (CI) 95% to test the representativeness of the matched samples. The results of the chi square test are interpreted by inspecting the P-values. Matching is considered close if the P-value is above 0.05.    60  Table 6: PSM output for matching 1  Electric Regular P-value Number of tips 143 143  Age (%)   1.000 21-30 9 (6.3) 9 (6.3)  31-40 38 (26.6) 38 (26.6)  41-50 75 (52.4) 75 (52.4)  51-60 8 (5.6) 8 (5.6)  60 13 (9.1) 13 (9.1)  Gender = Male (%) 92 (64.3) 92 (64.3) 1.000  Table 7: PSM output for matching 2  Electric Regular P-value Number of tips 143 143  Age (%)   0.996 21-30 9 (6.3) 9 (6.3)  31-40 38 (26.6) 38 (26.6)  41-50 75 (52.4) 77 (53.8)  51-60 8 (5.6) 8 (5.6)  60 13 (9.1) 11 (7.7)  Gender = Male (%) 92 (64.3) 101 (70.6) 0.313 Purpose (%)   0.236 Errand 9 (6.3) 13 (9.1)  Leisure 15 (10.5) 11 (7.7)  Leisure/Other 4 (2.8) 0 (0.0)  Other 22 (15.4) 25 (17.5)  Work 93 (65.0) 94 (65.7)        61  Table 8: PSM output for matching 3  Electric  Regular P-value Number of tips 143 143  Age (%)   0.776 21-30 9 (6.3) 11 (7.7)  31-40 38 (26.6) 36 (25.2)  41-50 75 (52.4) 68 (47.6)  51-60 8 (5.6) 9 (6.3)  60 13 (9.1) 19 (13.3)  Gender = Male (%) 92 (64.3) 81 (56.6) 0.226 AAG 0.02 (0.01) 0.02 (0.01) 0.329 PTPG 0.42 (0.11) 0.41 (0.14) 0.892 PTNG 0.39 (0.10) 0.38 (0.12) 0.653  Table 9: PSM output for matching 4  Electric  Regular P-value Number of tips 143 143  Age (%)   0.176 21-30 9 (6.3) 18 (12.6)  31-40 38 (26.6) 38 (26.6)  41-50 75 (52.4) 64 (44.8)  51-60 8 (5.6) 4 (2.8)  60 13 (9.1) 19 (13.3)  Gender = Male (%) 92 (64.3) 92 (64.3) 1.000 Purpose (%)   0.167 Errand 9 (6.3) 16 (11.2)  Leisure 15 (10.5) 16 (11.2)  Leisure/Other 4 (2.8) 0 (0.0)  Other 22 (15.4) 25 (17.5)  Work 93 (65.0) 86 (60.1)  AAG 0.02 (0.01) 0.02 (0.01) 0.539 PTPG 0.42 (0.11) 0.41 (0.12) 0.54 PTNG 0.39 (0.10) 0.39 (0.11) 0.738    62  As previously mentioned, only four matches were found to be close matches. Therefore, in total, we ended up with five sets of trips; one for electric bikes and four for regular bikes. The comparison will be conducted by comparing each set of regular bike trips with the electric bike trips individually. The comparison was done by two methods as follows: 1) The assessment parameters were calculated for each set of regular bike trips individually and then compared with the electric bike trips. T-test was also conducted to report the significance of the differences. 2) A pool of 150m microtrips was created for each set of trips and twenty biking schedules were then constructed from each pool. The construction method chosen for this practice was the best incremental approach with the same clustering and continuity criteria as presented in testing the optimal microtrip length (see previous section). The best biking schedule from each pool was then used to calculate power and total energy expenditure (Bigazzi & Figliozzi, 2015), which were used as comparison measures. Masses and resistance coefficients were taken from a previous study conducted in Metro-Vancouver as shown in Table 10 (Tengattini, 2017). Table 10: Parameters used in calculating power and energy expenditure Parameter Electric Regular Total mass (kg) 106 90 𝐶𝑑. 𝐴𝑓 (m2) 0.614 0.58 𝐶𝑟 0.0103 0.0079     63  4. Results 4.1 Proof-of-concept study  4.1.1 Processing time Processing time to generate the biking schedules varied between methods but was similar among data resolutions. The random selection method took the longest: up to 3 hours to identify 20 candidate schedules with PV<15%. On average, 70 schedules were created to reach 20 candidate schedules. The processing time for this method would increase with more microtrips, more desired candidate schedules, or a lower PV threshold. In contrast, the best incremental method required the least processing time, 10-20 minutes. The processing time for the single cluster method is mainly determined by the number of microtrips in the trip-starting pool, but, it took the most time to develop one complete schedule compared with the other two methods. Despite the speed difference in schedule construction, the single cluster method is still more time efficient than the random selection since fewer schedules are generated. Failure to find microtrips that met the continuity criteria was an issue for all methods, but predominantly a problem for the best incremental method which is restricted to microtrips within clusters rather than the entire pool. 4.1.2 Schedule progression Figure 15 shows the evolution of PV for each method with biking schedules of increasing length (measured by the number of microtrips). The last data point in each series is the target 25-minute schedule. In general, PV improves (decreases) with increasing schedule length, but not monotonically, and the optimum length depends on the construction method. Longer schedules can in some cases degrade the PV due to the constraints of the continuity criteria   64  when selecting microtrips to append. Substantially shorter schedules generate similar PV for the best incremental and single cluster methods, but the random selection method is less efficient at attaining a low PV. The question of optimal schedule length for biking schedules requires further investigation.  Figure 15: Relationship between PV and drive schedule length for each method  4.1.3 Performance value Table 11 gives the overall and individual-parameter PV for the biking schedules generated from each construction method. Based on the PV, the single cluster method yielded the best biking schedule, followed closely by the best incremental method. The random selection method substantially under-performed the other two methods according to the PV. By   65  inspecting all 60 schedules generation from all methods (20 each), there was no clear pattern of certain parameters having consistently higher PVs than others. Table 11: PVs for the best biking schedule generated from each method  Construction Method     Random Selection Best Incremental Single Cluster Individual-parameter PV x (%) ATS 8.91 1.40 0.39 ARS 8.70 0.89 0.08 PTI 1.94 4.99 3.03 PTC 6.04 1.19 0.86 AAA 7.26 0.42 1.57 PTA 0.96 1.21 1.75 PTD 3.21 2.60 2.66 APW 3.04 1.06 2.26 AAG 2.52 1.03 1.05 PTPG 4.77 1.15 0.06 PTNG 21.4 2.78 1.29 SAGPD 0.15 0.13 0.13 Overall PV (%) 4.93 1.31 1.02  4.1.4 Effect of GPS resolution Figure 16 gives the PV results for all three methods using data resolutions of 1, 3, 5, and 10 seconds (the 1-second results are the same as in Table 11). The ordering among the three methods is consistent across all four data resolutions, with single cluster performing best (lowest PV), followed by best incremental and random selection. The accuracy of the best incremental method degrades at coarser data resolutions. The single cluster method has the most consistently good performance, and no clear relationship with data resolution.   66   Figure 16: PV across methods using data resolutions of 1, 3, 5, and 10 seconds  4.1.5 Power and breathing rate results Table 12 gives the cyclist power output and breathing rate results. Biking schedules generated by all methods and data resolutions provide power output and breathing rate estimates within 10% of the estimates from raw data, and most are within 5%, suggesting that biking schedules can plausibly be used for these applications. Somewhat surprisingly, there is no clear relationship between the accuracy of the power and breathing estimates and the PV.    67  Table 12: Cyclist power output and breathing rate calculated from raw data and best biking schedules generated from each method  Power output (W)  Breathing rate (L/min)  Mean Difference from raw data (%)  Mean Difference from  raw data (%) 1 sec data Raw data 114 -  20.8 - Random selection 114 0.2  20.8 0.2 Best incremental 103 9.8  19.1 8.0 Single cluster 116 1.7  21.1 1.5 3 sec data Raw data 115 -  20.9 - Random selection 113 1.9  20.6 1.1 Best incremental 112 2.7  20.4 1.7 Single cluster 123 7.2  22.3 7.0 5 sec data Raw data 115 -  20.9 - Random selection 120 3.9  21.7 3.4 Best incremental 111 3.6  20.4 3.0 Single cluster 118 2.6  21.5 2.2 10 sec data Raw data 116 -  21.1 - Random selection 108 7.4  19.8 6.2 Best incremental 121 4.1  21.9 3.6 Single cluster 116 <0.1  21.1 <0.1  4.1.6 Transferability to other riders The PV for the biking schedules generated for cyclists B and C (using cyclist A’s microtrips) were 2% and 6%, respectively, by the best incremental method, and 1% and 2%, respectively, by the single cluster method. The biking schedules are less precise, as expected, when constructed from a different cyclist’s GPS data. Still, the biking schedules are able to represent the cycling dynamics reasonably well based on the assessment criteria. The single cluster method was markedly better for this application, likely because it draws from a larger pool of microtrips and does not rely on a transition matrix that was generated from a different cyclist’s data.  Calculated power outputs from these single cluster biking schedules are 11% and 18% higher than from the raw data for cyclists B and C, respectively, and breathing rates are 4% and 8%   68  higher. These differences are likely larger for cyclist C than cyclist B because the dynamics of cyclist C were more distinct from cyclist A. For example, cyclists A, B, and C had mean power outputs of 114, 77, and 57 W, respectively, and mean breathing rates of 21, 20, and 16 L/min. The observation of similar PV but markedly different power/breathing rate accuracy for cyclists B and C supports the idea that the assessment criteria might not well reflect the determinants of cyclist power, and further refinements should be explored. 4.2 GPS survey data overview 4.2.1 General overview  Over a recruitment period of four months in 2017, we recruited a total of 260 participants from all over Metro-Vancouver (see Figure 17). Most of the participants were from Vancouver but we also reached out to Richmond, North Vancouver, Coquitlam, White Rock, Langley, Burnaby, Surrey, and Delta. The geographic distribution of the cyclists aligns with a trip diary report conducted in 2011 in Metro-Vancouver (Regional Trip Diary 2011) that shows that most of the cycling activity is concentrated in Vancouver Lower Mainland with higher intensities at UBC and Kitsilano. This compliance in the geographic distribution suggests that our sample is representative of the actual cycling activity in the region.   69   Figure 17: Geograhic distribution of participants  All of the participants completed the questionnaire, but only 148 participants logged their trips. There were 35 electric bike users in the sample, but only 14 of them logged their trips. It was found that the majority of the participants (131 of the 148) preferred to use smartphones (“Ride with GPS” application) over manually uploading their trips. A total of 2292 trips were collected, which increased to 2314 trips after splitting trips recorded on the same file. There were 934 trips containing heart rate data.  Participants reported a total of 129 missed trips. All missed trips contained information about the mode and purpose. However, other data such as trip total distance, duration, average speed, start point, and end point were not available. Beside missed trips, 62 of the   70  downloaded files were corrupted. The file either had very few GPS points which made it impossible to interpret any useful data, or cannot be opened.   Overall, the collected trips comprised a total moved distance of 14961 km over 875 hours of GPS data after processing. Most of the GPS points were recorded every one second, but due to signal losses and delays, there were some missing GPS points. Before processing the data, the missing GPS points represented 70% of the data. Data processing reduced this percentage to 18% by clipping stop periods. The missing points would typically result in losing the same amount of speed data. However, due to speed inferring, the missing speed data were reduced to 8%. The reason for not having all speed data is because we did not interpolate speed values if there were more than 5 consecutive missing speed values. Figure 18 shows the trip distribution by mode type. The sample consisted of mainly bicycles trips (representing 70% of the data), while running trips were the minority with only 2% of the data. The term “NA” in the figure represents the trips where the mode was not explicitly specified.   Figure 18: Trip distribution by mode 70% 14% 8%2%7%0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%ModeBike walk E-Bike Run NA  71  The distribution of collected trips by purpose (regardless of mode) is shown in Figure 19. Work trips was the dominant type of trips in the data set. They represented 39% of our sample. Errands came in second place, which had a slightly higher percentage than leisure. Exercise trips were the least reported type of trips in the data set.   Figure 19: Trip distribution by purpose  Figure 20 represents the distribution of trip purposes for each mode. Most of the reported bike trips were for commuting between work and home. Whereas, running errands was the second most reported purpose for bike trips. Exercise was the least reported purpose, which implies that most cyclists in our sample use their bikes for utilitarian trips. Exercising might be an implicit purpose shared with all bike trips but it was explicitly mentioned for only 2% of them. Walking on the other hand was mostly for running errands and leisure. Among all other transport modes, those two purposes were the highest for walking trips. For electric bike trips, the purposes distribution was similar to the bike trips. Most of the electric bike 39% 8% 18% 8% 14% 4% 9%0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%PurposeWork School Errand Other Leisure Exercise NA  72  trips were for work commute. None of the reported electric bike trips were for school commute, which implies that most of the electric bike users no longer attend schools. It can also be pointed out from the figure that none of the electric bikes were reported for the purpose of exercise. This implies that cyclists in our sample do not perceive electric bikes as an exercise tool. There were some trips without an explicitly stated mode or purpose. Those trips were referred to with the term “NA” in the figure.  Figure 20: Distribution of trip purpose by mode type 24%11%64%47%3%11%9%45%12%17%7%13%8%3%27%11%14%65%6%2%0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%RunwalkE-BikeBikeWork School Errand Other leisure Exercise  73  4.2.2 Data representativeness We compared age, gender, and income distribution to the trip diary report (Regional Trip Diary 2011) and a previous survey conducted in 2016 (Tengattini, 2017), which sampled cyclists from the same region (Metro-Vancouver), in order to examine the representativeness of our sample. Chi-square tests at 0.05 significance level were conducted between our sample and both references to report the representative of our sample.   Figure 21 shows the gender distribution of participants for the three samples (our sample, 2011 regional report, and the 2016 survey). The distribution, in general, aligns with the previous studies with having more male cyclists than female in the region. The chi-square test between the gender distribution of our sample and the 2011 regional report reported significant differences between the samples. Whereas, the test reported insignificant differences between our sample and the 2016 survey.   74   Figure 21: Gender distribution  Figure 22 shows the income distribution of our sample compared with the other two studies. Our sample mainly aligns with the previous studies. However, it contained participants with income less than $25,000 almost twice as many as the 2011 regional report. This is most likely because our sample had more students since we recruited three times at UBC. The percentage of cyclists with income ranging between $75,000 and $100,000 was noticeably higher for the 2011 regional report compared with our sample and the 2016 survey. However, the number of participants with this income range in our sample was matching with the 2016 study. The chi-square test between the income distribution of our sample and the 2011 regional report reported significant differences between the samples. Whereas, the test reported insignificant differences between our sample and the 2016 survey. 0.00%10.00%20.00%30.00%40.00%50.00%60.00%70.00%80.00%Male FemaleOur sample 2011 Regional report 2016  Survey  75    Figure 22: Income (in $) distribution  Figure 23 shows the cumulative age distribution for the three samples. Our sample did not have any participants under 16 because the BREB states that we cannot recruit anyone under 16 without an official consent from their parents. There were also issues with tracking children and data reliability. In general, the age distribution in our sample follows the other two studies. We couldn’t perform any statistical tests on the age distribution because the age ranges used in our questionnaire is different from the one used in the two other samples. 0.00%5.00%10.00%15.00%20.00%25.00%<25000 25000-50000 50000-75000 75000-100000 100000-150000 >150000Our sample 2011 Regional report 2016  Survey  76   Figure 23: Cumulative age distribution  Overall, the chi-square tests and the plots show that our sample is very close to the 2016 survey. This could return to the fact that the 2016 survey is newer than the 2011 regional report, which explains why the chi-square tests reported significant differences. The characteristics of the cyclists’ community might have changed between the time when the regional report was conducted (2011) and 2017 when we conducted our survey. We also conducted chi-square test between the 2016 survey and the 2011 regional report for the gender and income distributions, and both tests reported significant differences. This result further supports our conclusion that the 2011 is outdated and it should only be used to demonstrate how our sample aligns in general with regional surveys.    0%20%40%60%80%100%0 10 20 30 40 50 60 70 80 90AgeOur Sample 2011 Regional report 2016 Survey  77  4.3 Optimal microtrips length for biking schedules In order to further enhance the biking schedule construction methods, we explored seven different microtrip definitions: 100m, 150m, 200m, 250m, 300m, 350m, and Stop to Stop (STS). The definitions were compared as follows. 1) Number of microtrips generated. In total, after excluding exercise and non-bike trips, 1308 trips stood up in the analysis. The number of microtrips generated from these trips varied significantly among different definitions. The numbers of microtrips generated from each definition is shown in Figure 24 (red color). It can be noticed that the number of microtrips decreases as the microtrip length increases which is reasonable since the microtrip gets longer and less are generated from the data. The STS definition generated the lowest amount of microtrips. The average microtrips length from this definition was 1470 m, which is the longest among all other definitions. It can also be noticed that the number of microtrips generated from the 250m definition in our collected data (34,000) is much higher than the number generated in the proof-of-concept study (1,530). This explains our choice to narrow down the continuity criteria. One of the conditions to append a microtrip in the biking schedule is that it has to meet the continuity criteria. Narrowing the criteria will lower the number of candidate microtrips. However, in our collected sample, the number of microtrips was large enough to narrow down the continuity criteria and still leave considerable amount in the construction process.  2) Processing speed The processing speed is proportionate with microtrip length. The time needed to complete a biking schedule was the longest for the 100m microtrip pool; reaching almost 30 hours. The   78  processing time decreases as the microtrip length increases; reaching about 3 hours for the 350m microtrips, and 20 minutes for the STS microtrips. This happens mainly for two reasons. Firstly, the number of microtrips needed to form a complete biking schedule decreases as the microtrip length increases as shown in Figure 24 (black color). Secondly, as the length increases, fewer microtrips are created in the pool which makes looping through them faster.   Figure 24: Relationship between the number and definition of microtrips   3) Performance value. Figure 25 shows a boxplot of all PVs generated from each pool. As shown in the figure, there is no monotonic behavior in the PV across the pools. The 150m microtrip pool generated the lowest PV among all other definitions. It can also be noticed that the 100m microtrip pool 0102030405060708090100100m 150m 200m 250m 300m 350m STSMicrotrip definitionNumber of microtrips generated (x1000)Number of microtrips needed to complete a biking schedule  79  generated relatively higher PV than the 150m pool. This observation aligns with previous research on driving schedules (Nouri & Morency, 2017). The shortest microtrip length does not necessarily yield the lowest PV. The individual PVs for only the best biking schedule generated from each pool is presented in Table 13 to further support the boxplot. The best biking schedule from the 150m pool is shown in Figure 26.   Figure 25: PVs generated from each pool     80  Table 13: Individual PVs for the best biking schedules in (%) Microtrip Definition 100m 150m 200m 250m 300m 350m STS ATS 5.3 1.5 2.9 3.2 1.2 9.6 2.2 ARS 5.4 1.7 2.7 3.1 1.0 9.3 2.0 PTI 1.8 5.3 7.7 6.1 6.4 5.0 1.9 PTC 1.1 0.7 0.4 2.0 0.7 0.8 3.9 AAA 8.5 2.1 3.5 1.1 5.2 2.8 0.1 PTA 5.0 2.9 0.7 6.4 0.6 10.0 5.2 PTD 3.7 3.6 1.4 1.8 8.4 0.3 10.0 APW 1.6 1.1 3.4 4.4 0.2 0.2 3.6 AAG 1.2 0.6 2.8 1.6 1.4 0.5 7.1 PTPG 1.3 0.2 0.2 0.3 1.0 2.8 0.8 PTNG 0.2 0.4 0.6 1.2 0.0 0.5 6.1 SAGPD 0.103 0.090 0.098 0.090 0.087 0.107 0.082 PV 2.3 1.3 1.8 2.0 1.7 2.7 3.0   Figure 26 Best biking schedule from the 150m pool   81  Table 14 shows the average individual PVs of all twenty biking schedules generated from each pool. The red marker indicates which definition is the best at representing a specific parameter. For example, the 100m microtrips are better at representing PTC and PTA. The 200m definition represented the most parameters. Although this definition was better at representing more parameters and had a better average value for the overall PV compared with the 150m definition, the best biking schedule with the lowest overall PV was still generated from the 150m microtrips.  Table 14: Individual average PVs for all biking schedules in (%) Microtrip Definition Target* 100m 150m 200m 250m 300m 350m STS ATS 17.9 8.4 9.6 8.3 10.2 10.6 13.0 12.1 ARS 18.6 8.2 9.3 8.2 10.0 10.5 12.3 11.9 PTI 3.7 23.1 22.6 17.2 16.7 30.5 19.9 12.2 PTC 19.0 1.5 2.3 2.6 3.9 5.4 2.6 7.6 AAA 0.5 12.3 4.6 5.2 6.1 4.4 3.8 4.2 PTA 36.9 4.3 4.9 5.4 5.5 4.5 6.4 7.4 PTD 36.6 5.3 5.9 3.0 3.1 3.9 4.4 8.3 APW 1.7 14.1 3.9 3.6 5.6 3.8 3.3 5.5 AAG 2.4 1.1 1.0 2.0 3.0 2.2 4.1 4.1 PTPG 44.7 0.9 1.3 0.9 1.9 2.9 2.5 3.9 PTNG 37.9 1.0 0.9 1.5 1.6 1.8 2.1 3.6 SAGPD - 0.1 0.094 0.098 0.107 0.12 0.105 0.102 Overall PV - 5.1 4.2 3.7 4.4 5.2 4.9 5.3 *Units are as indicated in Table 1 It is noticed that the SAGPD values are very low, this could be attributed to two factors. First, this parameter was derived from the speed acceleration probability distribution (SAPD) commonly used in driving schedules and the intervals used for speed, acceleration, and grade in the 3D matrices were directly adopted from these studies (Brady & O’Mahony, 2016; Seers et al., 2015). These intervals are likely to need some adjustments for bike studies. The second   82  factor is using RMSE in the calculation of PVs. RMSE is used to represent the differences between the 3D matrices. This method is different from how PVs are typically calculated (see 3.2.3). Examining the best way to calculate the SAGPD and integrating it with the total PV needs more investigation and was left for future work.   4.4 Comparison between electric and regular bikes travel characteristics For each of the resulting five sets of trips (4 regular and 1 electric), we examined 1) the differences in the target assessment parameters and compared them by conducting t-tests 2) the differences in average power output and total energy expenditure. 4.4.1 Target assessment measures Table 15 summarizes the target assessment measures for each set of trips along with the standard deviation (SD) and t-test results (two tailed test with 95% confidence level). The t-test shows that seven parameters were significantly different between electric bikes and regular bikes (marked in red); all of them are related to speed and acceleration. It is noticed that the differences are either statistically significant or insignificant for the same assessment parameter regardless of the matched criteria. This implies that electric bikes are different at representing certain aspects of the trip no matter the matched sample.     83  Table 15: Comparison between the assessment parameters for regular and electric bikes Parameter* Regular bike trips Electric bike Trips Matching 1 Matching 2 Matching 3 Matching 4 Mean SD** P–value*** Mean SD P-value Mean SD P-value Mean SD P-value Mean SD ATS 17.3 4.35 5.63E-11 18.4 4.86 3.28E-7 17.4 4.97 1.4E-8 17.7 4.74 7.79E-11 24.7 7.23 ARS 18 4.28 4.25E-12 19 4.81 1.88E-8 18 4.85 1.01E-9 18.2 4.74 4.28E-12 25.6 6.98 PTI 3.92 4.52 93.5 2.81 3.23 5.68 3.34 4.54 65.5 2.98 4.26 41.2 3.50 4.05 PTC 19.0 7.91 6.53E-2 18.5 7.64 0.0114 21.8 8.35 2.67E-6 19.6 7.49 9.84E-5 14.7 6.94 AAA 0.458 0.14 4.12E-8 0.481 0.14 2.38E-7 0.42 0.13 2.18E-13 0.459 0.13 2.77E-10 0.622 0.18 PTA 38.9 4.98 0.854 39.8 4.43 2.20 37.8 5.05 0.011 39.2 4.39 5.91E-2 39.6 3.82 PTD 36.9 5.05 6.45E-3 37.7 4.94 0.096 35.5 5.28 2.43E-6 36.8 5.63 2.06E-4 38.9 4.14 APW 1.74 0.55 4.28E-4 1.8 0.58 9.94E-4 1.58 0.50 4.38E-9 1.7 0.50 4.36E-6 2.17 0.65 AAG 2.44 0.58 39.5 2.41 0.64 31.3 2.25 0.67 50.0 2.27 0.70 96.2 2.14 0.89 PTP 44.4 11.04 11.0 45.1 12.43 7.51 43.5 12.31 43.2 41.3 11.04 72.5 40.1 12.00 PTNG 38.9 10.76 45.0 38.3 10.98 44.7 36.8 10.48 76.1 39.1 10.58 43.1 36.9 11.03 *Units are as indicated in Table 1 **SD stands for Standard Deviation ***All P-valued are in (%)  84  The average travel speed (ATS) and average running speed (ARS) for regular bikes in all four matches were similar to each other with an average value of 17.7 km/hr and 18.3 km/hr, respectively. These speeds are within a reasonable range (Bernardi & Rupi, 2015). On the other hand, the ATS and ARS for electric bikes were 24.7 km/hr and 25.6 km/hr, respectively. These values are also within an acceptable range (Tao, Niu, & Chen, 2014). These results indicate that electric bikes travel faster than regular bikes. The difference in average speeds calculated from both ATS and ARS is 7km/hr.  Percentage time idling (PTI) was not statistically different between regular and electric bikes.  This result was expected because all participants were recruited from the same region and they cycled on the same roads which would make them stop at the same places (e.g. intersections) for nearly the same time.  Percentage time cruising (PTC) was lower for electric bikes than regular bikes. This returns to the difference in speeds between the two bike types. As previously discussed, electric bikes travel faster than regular bikes. This will make electric bikes spend more time accelerating to the cruising speed, which makes the cruising activity shorter. The average absolute acceleration (AAA) for regular bikes was similar for the four matches with an average value of 0.45 km/hr/sec. This is also within an acceptable range (Parkin & Rotheram, 2010). AAA for electric bikes was 0.622 km/hr/sec, which is 0.17km/hr/sec higher than regular bikes. This is accounted mostly for the usage of motor in accelerating instead of human power.  The averages for the Percentage time accelerating (PTA) and percentage time decelerating (PTD) were similar among electric bikes and regular bikes. However, the t-test results shows   85  that there are statistically significant differences. This means they came from a different distribution but the means are close. The average positive work (APW) was significantly higher for electric bikes compared with all four regular bike matches. This shows that electric bikers ride more vigorously, which aligns which the AAA results. It also shows that electric bikes use more energy compared with regular bikes. For all assessment parameters associated with terrain (i.e. Average absolute grade (AAG), percentage time positive grade (PTPG), and percentage time negative grade (PTNG)), there were not significant differences as shown in the table. This aligns with our expectations since all participants were recruited from Metro-Vancouver and grade values should be similar in the same region. Additionally, Matchings 3 and 4 were controlled for terrain, which explains the presence of statistical insignificant differences between these matches and the electric bike trips.   4.4.2 Comparison between biking schedules The number of microtrips generated from the four regular bike matches were close to each other since each matching contained the same number of trips and the trips were divided into 150m intervals. There were 249335, 250309, 246093, 225776 microtrips generated from matches 1, 2, 3, and 4, respectively. The number of microtrips generated from electric bike trips was 225544. These results show that the constructed biking schedules are comparable since they had similar numbers of microtrips and were constructed by the same method (Best-incremental method) and continuity criteria (1km/hr speed and 1% grade). It should   86  also be mentioned that applying PSM made the socio-demographic characteristics of the regular bike trips close to the electric bike trips, which further validates the comparison.  Table 16 shows the PV of the best generated biking schedule generated from each set of trips. The overall PVs for all matches were under 5% suggesting that the biking schedules are representative of the trips from which they were yielded. The PTI for matching 3 was noticeably high which led into yielding the highest overall PV compared to other matches.  Table 16: PV for the best generated biking schedules from each set of trips Parameter Matching 1 Matching 2 Matching 3 Matching 4 electric bike ATS 5.23% 5.46% 2.93% 1.22% 5.98% ARS 5.41% 5.43% 3.55% 1.02% 5.81% PTI 4.29% 0.83% 18.59% 6.21% 7.55% PTC 2.09% 2.41% 0.35% 3.41% 0.02% AAA 7.61% 5.17% 0.10% 0.74% 0.17% PTA 0.07% 0.20% 2.65% 2.12% 1.33% PTD 0.11% 1.10% 1.64% 1.28% 0.67% APW 1.07% 5.27% 3.31% 0.75% 0.76% AAG 0.86% 0.07% 2.46% 0.60% 3.21% PTPG 0.53% 0.73% 0.22% 0.77% 0.37% PTNG 0.50% 0.92% 0.34% 3.93% 0.73% SAGPD 0.09% 0.09% 0.09% 0.08% 0.10% Overall PV 1.80% 1.78% 2.34% 1.51% 1.78%  Figure 27 shows the total moved distance calculated from the best generated biking schedules. The distance is calculated by taking the cumulative sum of speeds. The average total moved distance for the regular bikes at the end of the biking schedule was 7.2km. While the total distance for electric bike was 9.7km. These result show that, during 25 minutes, electric bikes can travel 2.5km more than regular bikes.    87   Figure 27: Total moved distance of all matches compared with electric bikes (calculated from the best biking schedules)  The average power output for the best biking schedules generated from matches 1, 2, 3, and 4 were 134W, 107W, 121W, and 113W, respectively. These power values are in line with previous research which suggests that average power output ranges from 50W to 150W (Bigazzi & Figliozzi, 2015; WHITT, 1971). Whereas the average power for the biking schedule generated from the electric bike trips was 256W. It should be mentioned that the   88  stated power values represents the total power used in cycling regardless of the source (i.e. human power or electric motor). The total energy expenditures for the best generated biking schedules are presented in Figure 28. The total energy expenditure at the end of the biking schedule was similar for the four matches (204, 161, 183, and 170 Mega Joules for matching 1, 2, 3, and 4, respectively). Whereas electric bikes used 387 Mega joules at the end of the trip. These results suggest that electric bikes use as much as twice the energy required to ride a regular bike.   Figure 28: Total energy expenditure of all matches compared with electric bikes (calculated from the best biking schedules)    89  5. Conclusions 5.1 Research questions/answers This thesis is conducted to answer three specific research questions (listed below) as mentioned in the introduction chapter.  i) What are the different potential methods of constructing biking schedules? How do they differ? What is the best method? ii) What is the optimal definition of the biking schedule building unit (known as microtrips)? iii) How could the biking schedules be implemented to compare between the travel characteristics of electric and regular bicycles?  We discussed and compared three different methods of constructing biking schedules, namely, best incremental, random, and single cluster. The comparison was conducted by studying the differences in processing times, performance values, and how the methods perform under different data resolutions.  The best incremental approach took the least processing time among all other methods. The resulting biking schedule is most likely to perform well at representing the calibration data set since the PV is continuously monitored in the construction process. Most of the biking schedules generated from this method are likely to pass the PV threshold (i.e. 15%). The best incremental approach also deviates from other methods by following a Markov chain algorithm which creates a sequence of clusters based on the transition matrix. Thus, the created biking schedules are more likely to represent typical real-world cycling more closely   90  since the transition matrix was built based on the sequence of clusters in the calibration data set. The best incremental approach has some drawbacks as well. Data size is crucial in the performance of this method. To consider a microtrip in the construction process, it has to meet two conditions: 1) continuity criteria 2) fall in the same cluster from which microtrips are being chosen. These two conditions add more constraints on the considered microtrips and in some cases, due to the unavailability of multiple alternatives, a microtrip that worsens the PV might be appended.  The single cluster approach follows a similar technique to the best incremental but without clustering microtrips. Therefore, microtrips only have to meet the continuity criteria in order to be considered in the construction process. This method ensures that the best possible biking schedules are generated by appending the fittest microtrips. It also outperformed other methods when different data resolutions were applied. There are several drawbacks for this method. The single cluster method takes the longest time to complete one biking schedule because every time a microtrip is appended, all microtrips that meet the continuity criteria need to be checked. Another disadvantage of using the single cluster is that it does not follow a Markov chain sequence. Consequently, the microtrips sequence in the resulting biking schedule are less likely to follow a similar pattern to the calibration data set. The third method is the random selection method. Microtrips are appended randomly without monitoring the PV. This method is the fastest at generating a complete biking schedule. However, due to the randomness of the process, most of the constructed biking schedules   91  have less PV than the threshold, and thus more biking schedules are needed to be constructed in order to get twenty candidate biking schedules. Results show that, in the random selection method, 70 biking schedules were constructed before reaching twenty that meet the PV threshold, which means that almost 75% of the total number of constructed biking schedules will be discarded. Another drawback of this method is that it gave the worst PVs compared to of methods under all different data resolutions. For all the aforementioned reasons, we recommend adopting the best incremental approach for large data sets with high resolutions. For small-sized data sets or low data resolutions, the single cluster method should be used instead. We also studied the optimal microtrip length by comparing between different spatial lengths, specifically, 100m, 150m, 200m, 250m, 300m, and 350m. We also included the most common microtrip definition, STS. Given that nowadays computer technology is continuously advancing and the processing time is no longer a concern for the majority of researchers and practitioners, the choice of the best microtrip length was mainly determined by the PV. Twenty biking schedules were generated using each microtrip definition and the PVs were compared. Results show that the 150m microtrips gave the lowest PV among all other definitions including the STS – suggesting that 150m is the best microtrip definition to use in constructing biking schedules.  On the other hand, when we studied the PV of individual assessment parameters, it was found that other definitions may be better at describing specific individual parameters. This would affect the choice of the optimal microtrip definition. For example, if the main goal of generating biking schedules is to model acceleration precisely for a certain application, it   92  might be better to use a different definition accordingly. Moreover, the weights used in the calculation on the PV have an influence on the results. If more weight was given to acceleration, different results would be expected.  Therefore, we conclude that, overall, the 150m (which is nearly the average block length in Metro-Vancouver) performs better at generating biking schedules with equal weights between the elements of the assessment parameters (i.e. speed, acceleration, and road grade). However, if a specific parameter was desired to be modeled more precisely, a different definition should be used accordingly.” We also conducted a comparison between regular and electric bikes. The comparison consists of two parts. Firstly, the propensity score matching technique was used to find four sets of regular bike trips that match the electric bike trips. The assessment parameters used in constructing biking schedules were used to compare the travel characteristics of each set of regular bike trips with electric bike trips. Results show that electric bike users travel 7 km/hr faster than regular cyclists. Electric bikers also have higher acceleration rates. Results show that acceleration rates for electric bikers are 0.17 km/hr/sec higher than regular cyclists. Such consideration should be taken into account in designing bike lanes. It is more likely to see electric bikes overtaking regular bikes. Such behavior could have major safety impacts and might lead to several accidents and injuries, especially on narrow bike lanes. These results suggest that more research is required to investigate whether electric bikes should be operated on bike lanes with regular bikes, roadways next to motor-vehicles, or on designated bike lanes.   93  Secondly, we used the 150m microtrips and the best incremental approach to construct biking schedules for regular and electric bikes. We specifically compared total moved distance, average power output, and total energy expenditure. Results show that electric bikes can travel more distance (2.5km) than regular bikes during a course of 25 minutes. Therefore, we can conclude that electric bikes are more useful for travelling longer distances. Results also show that electric bikes use almost twice the energy needed for riding regular bikes. This is accounted for two reasons. First, there are differences in weights and friction coefficients between regular and electric bikes. Second, there are differences in speed and acceleration characteristics between both types of bikes. Electric bike riders travel faster and accelerate more vigorously which increases the energy consumption. These results help electric bike manufacturers understand the real-world cycling behavior of electric bike users. Given that electric bikers are more likely to travel more often and the bigger population are above 60 years old or have medical conditions (MacArthur et al., 2014; Wolf & Seebauer, 2014), they need to be provided with a battery that lasts longer time and supplies sufficient amount of assistance. It should also be mentioned that electric bikes weigh heavier (Tengattini, 2017) and the users are exposed to higher resistance (e.g air, and rolling) emphasizing the need for larger batteries. If the batteries ran out, the users will need to provide twice the energy needed to ride a regular bicycle, which is unreliable considering their age and medical condition.  5.2 Potential biking schedule applications The proposed biking schedules can have a variety of applications. Biking schedules can be used to estimate cyclist power, energy, and breathing rate from aggregate travel data (collected through travel surveys, bike-share systems, smartphone applications, etc.), and   94  thus improve health effects estimates including physical activity and pollution inhalation. Biking schedules could be segmented by rider type (age, experience), equipment (electric bike, cargo bike, road bike), season or weather conditions, or facility type (bike lane, cycle track, multi-use path) to explore and represent systematic, archetypal differences in urban cycling styles/dynamics among population segments, cities, or facilities. In addition to modeling cycling outcomes, biking schedules could potentially be used to represent the typical energy “costs” of network links and thus applied as inputs to route choice models.  Biking schedules can also be used by bicycle designers and manufacturers, particularly of electric bikes and other human/electric hybrid vehicles, similar to the way driving schedules are used in motor vehicle modeling and design. Representative schedules could be implemented in simulation models and laboratory testing to investigate power consumption and battery life, for example. Segmented biking schedules could provide more customized performance information for specific market segments (sport vs. leisure riders, for example), similar to the city/highway fuel economy information supplied to motor vehicle shoppers. Biking schedules could also be used in research and clinical laboratories to investigate human performance under more realistic cycling conditions than traditional tiered-workload exercise tests. 5.3 Additional findings and unique contributions This thesis contributes to the current research literature by presenting a new tool for bicycle travel analysis. Biking schedules generated by all methods and data resolutions provide power output and breathing rate estimates within 10% of the estimates from raw data, and most are within 5%, suggesting that biking schedules can plausibly be used for these applications.    95  The accuracy of the method depends on the proximity of the calibration data to the application conditions. In this study, GPS data from one cyclist was able to generate reasonable biking schedules based on the aggregate travel data of two other cyclists in the same 55-hour data set, albeit with less accuracy. Transferability to more remote conditions (other cities, seasons, etc.) requires further investigation. It is expected that larger calibration data sets encompassing more variability (of riders, terrain, trip purposes, bicycle types, weather conditions, facility types, etc.) would have greater utility for generating realistic biking schedules for other contexts. The transferability of biking schedules, however, depends on the consistency of cycling dynamics across contexts, for which we still have little evidence in the literature.  The ultimate goal of generating biking schedule is to represent real world cycling behavior. All construction methods share a general principle of appending microtrips to replicate the assessment parameters of the calibration data set as closely as possible. The assessment parameters should be chosen to describe the most important aspects of the trips according to the sought application. For example, if the target of developing biking schedules is to determine safe against unsafe riding, the proportion of acceleration and deceleration events should be considered as they likely have influence on safety.  In this study, a set of twelve assessment parameters were chosen to encamps speed, acceleration, and grade characteristics. Power and breathing rate were used to validate the generated schedules. These parameters cannot be used directly to evaluate the biking schedules because assessing only these two values will overlook speed, acceleration, and grade characteristics, which would make the constructed biking schedules unrepresentative of real-world cycling behavior. If desired, the assessment criteria could potentially be refined   96  to more specifically reflect the determinants of power and breathing, or power and breathing estimates could even be used directly as assessment parameters. It should be elaborated that the biking schedule construction methodology is stochastic, meaning that each run will generate a different schedule with different power and breathing rate estimates, which explains the inconsistency of the results in Table 12. Given that there is no clear relationship between PVs and power estimates, biking schedules were evaluated solely on PVs to ensure that the schedules representative of real-world cycling behavior. Other factors of interest (e.g. power and breathing rate) in turn should be measurable from the biking schedules. 5.4 Lessons learned in recruiting cyclists for a travel survey  5.4.1 Attracting participants during field recruitment Field recruitment is one of the most common and effective methods of recruitment. However, it is not necessarily an easy task. Getting the attention of cyclists on bike lanes and getting them to stop is challenging. It is a skill that needs to be developed like any other skill. This was part of the learning process we had to go through in order to make this research successful. Here, we provide some lessons and recommendation we found useful for future researchers performing field recruitment. 1) Signage: This is one of the most important tools during recruitment. Signs tell people who you are and what you are doing from a distance ahead. They simply speak for you. Signs should be clear and big enough to be read from distance. The text should be as brief as possible and to the point. The number of signs also matters significantly, cyclists travelling from different directions should be able to see them.   97  From our experience, cyclists are not willing to pause their travel if they are not certain of who you are and what you are doing. 2) Visibility: It is preferred to have a shade cover, desks, and chairs. Standing in front of them visible for all approaching cyclists encourages cyclists to stop. 3) Team work: Your efforts are magnified significantly with more members on your team. Having more people recruiting gives cyclists the feeling that this is a major event that they do not want to miss.  4) Make eye contact: This is the most important tip to stop cyclists. For safety concerns, you have to be standing far behind the bike path. However, even if the cyclist is not slowing down, he/she can definitely see you. Once they make eye contact with you, you have to promptly act by saying something brief such as “Hi” or “Hello” or “Do you have a minute”. If they are slowing down and you have more time, you could say something longer like “Would you like to learn about our survey”. There is split second before you lose this potential participant forever. We found many cyclists turning back to us when we started speaking. 5) Look professional: Name tags and unified T-shirts with your institution logo make you look professional. People feel safer if you are organized. 6) Bring your incentives into use: Some cyclists are more likely to stop when they hear you have incentives for them and they will be rewarded for their time. In our case, we displayed a specimen of hats and socks. Some participants were excited to try the heart rate monitor more than the actual incentives. They never had a chance to use one of the monitors before. Before going out on the field to recruit, think about what would participants be interested to see. Think as one of them.   98  7) Leave a good impression: Some people are more talkative than others. Some are in a rush because they may be late for work or on their way to pick up their kids. Saving their time leaves a good impression. If they are in a hurry, try to be brief as possible and let them be on their way. Understanding that they have liabilities shows that you care about them and they in return will care about your research. Some cyclists took an invitation card without even fully stop or say a single word. Some other people have the time to talk, give them full description of your survey, why you are doing it and how is it going to benefit the society. Be clear and organized. Let all your team very familiar with the survey and ready for answering all sort of questions.  5.4.2 How to target electric bikes riders The strategy of targeting electric bike riders is no different from regular bikers. However, there is a strategy to increase their participation chance. Complementing their bikes was very effective. Electric bikers appreciate knowing that they are riding something special and more powerful. If cyclists had enough time, we would ask them for the reasons they bought them and if they are having a better cycling experience. Letting them know that you are especially targeting them is helpful.  We also targeted electric bike shops in our recruitment process. We visited all electric bike shops in the region and asked them to post flyers on the walls. We also gave them invitation cards and asked them to let their customers know about the survey.  5.5 Limitations and future research The proposed biking schedules are a promising new tool for bicycle travel analysis, but further work is needed to develop robust construction methods. Much of the proposed   99  approaches were drawn from driving schedule methods, which likely have limited transferability to cycling. The selection of assessment criteria requires further investigation. As this is the first known attempt to develop biking schedules, the 12 assessment measures used in this study should be viewed as preliminary.  For future research, we propose the following 1) Generate biking schedules that represent different travel and socio-demographic characteristics. Biking schedules provide a promising tool that could be used at a wider scale and for different applications, similar to driving schedules. 2) Explore different applications for biking schedules 3) Explore different microtrip definitions 4) Explore the optimal biking schedule length. Results in this research show that the PV, to some extent, relies on the schedule length.  5) Explore alternative weighting schemes to adjust the influence of individual assessment parameters on the overall PV. 6) Explore alternative speed, acceleration, and grade intervals for calculating the SAGPD and different methods of integrating the SAGPD in the PV. 7) Most of the masses and friction coefficients used in the power and breathing rate models were adopted from previous studies. However, considering these values as deterministic is problematic since there is uncertainty associated with them. Future work on performing sensitivity analysis to these models is required for calibration. 8) In this study, the elbow method was used to determine the number of microtrip clusters in the best incremental approach. We propose exploring different methods to select the optimal number of clusters.   100  9) The differences in the assessment parameters between electric and regular bikes were assessed by performing T-test. For future work, it is proposed to explore different approaches such as bootstrapping technique.” 10) Electric bikes have different designs and each design supports the rider with different power proportions. It is proposed for future studies to decompose user-related energy for electric bicycles and compare the results with regular bicycles. 11) Testing the optimal microtrip length in this study was limited to a data set collecting in Metro-Vancouver, Canada. Future studies should verify the results by using a data set collected from a different city. 12) This research relied extensively on GPS data collected from either stand-alone GPS device or smartphones. The usage of GPS introduces sources of error and is a limitation by itself and thus speeds calculated from raw data are inaccurate. For future practices, we recommend exploring different methods to collect more reliable speed. 13) Another limitation of using GPS is its low accuracy at measuring altitude values. In this research, we relied on DEM to get elevation and grade data. However, even after filtering and smoothing, we still observe several grades above 5%. We recommend searching for an alternative approach for extracting elevation data.    101  Bibliography AASHTO. (2012). Guide for the Development of Bicycle Facilities. American Association of State Highway and Transportation Officials, Washington DC. Amirjamshidi, G., & Roorda, M. J. (2013). Development of simulated driving cycles: Case study of the Toronto Waterfront Area. In Annual meeting of Transportation Research part B. Retrieved from http://docs.trb.org/prp/13-2648.pdf André, M. (2004). The ARTEMIS European driving cycles for measuring car pollutant emissions. Science of The Total Environment, 334–335, 73–84. https://doi.org/10.1016/j.scitotenv.2004.04.070 Ashtari, A., Bibeau, E., & Shahidinejad, S. (2014). Using Large Driving Record Samples and a Stochastic Approach for Real-World Driving Cycle Construction: Winnipeg Driving Cycle. Transportation Science, 48(2), 170–183. https://doi.org/10.1287/trsc.1120.0447 Bernardi, S., & Rupi, F. (2015). An Analysis of Bicycle Travel Speed and Disturbances on Off-street and On-street Facilities. Transportation Research Procedia, 5, 82–94. https://doi.org/10.1016/j.trpro.2015.01.004 Berry, M. J., Koves, T. R., & Benedetto, J. J. (2000). The influence of speed, grade and mass during simulated off road bicycling. Applied Ergonomics, 31(5), 531–536. https://doi.org/10.1016/S0003-6870(00)00022-3 Berzi, L., Delogu, M., & Pierini, M. (2016). Development of driving cycles for electric vehicles in the context of the city of Florence. Transportation Research Part D: Transport and Environment, 47, 299–322. https://doi.org/10.1016/j.trd.2016.05.010   102  Bigazzi, A. Y., & Figliozzi, M. A. (2015). Dynamic Ventilation and Power Output of Urban Bicyclists. Transportation Research Record: Journal of the Transportation Research Board, 2520, 52–60. https://doi.org/10.3141/2520-07 Bigazzi, A. Y., & Lindsey, R. (Forthcoming). A utility-based bicycle speed choice model with time and energy factors. Transportation. Bishop, J. D. K., Axon, C. J., & McCulloch, M. D. (2012). A robust, data-driven methodology for real-world driving cycle development. Transportation Research Part D: Transport and Environment, 17(5), 389–397. https://doi.org/10.1016/j.trd.2012.03.003 Brady, J., & O’Mahony, M. (2016). Development of a driving cycle to evaluate the energy economy of electric vehicles in urban areas. Applied Energy, 177, 165–178. https://doi.org/10.1016/j.apenergy.2016.05.094 Broach, J., Dill, J., & Gliebe, J. (2012). Where do cyclists ride? A route choice model developed with revealed preference GPS data. Transportation Research Part A: Policy and Practice, 46(10), 1730–1740. https://doi.org/10.1016/j.tra.2012.07.005 Canadian Digital Surface Model. (2018, March). Retrieved from https://open.canada.ca/data/en/dataset/768570f8-5761-498a-bd6a-315eb6cc023d Casello, J., & Usyukov, V. (2014). Modeling Cyclists’ Route Choice Based on GPS Data. Transportation Research Record: Journal of the Transportation Research Board, 2430, 155–161. https://doi.org/10.3141/2430-16 CH2M. (2016). Transportation Panel Survey, 95.   103  Cherry, C., & Cervero, R. (2007). Use characteristics and mode choice behavior of electric bike users in China. Transport Policy, 14(3), 247–257. https://doi.org/10.1016/j.tranpol.2007.02.005 Cherry, C. R., Weinert, J. X., & Xinmiao, Y. (2009). Comparative environmental impacts of electric bikes in China. Transportation Research Part D: Transport and Environment, 14(5), 281–290. https://doi.org/10.1016/j.trd.2008.11.003 Cich, G., Knapen, L., Bellemans, T., Janssens, D., & Wets, G. (2016). Threshold settings for TRIP/STOP detection in GPS traces. Journal of Ambient Intelligence and Humanized Computing, 7(3), 395–413. https://doi.org/10.1007/s12652-016-0360-9 CROW. (2007). Design manual for bicycle traffic. National Information and Technology Platform for Infrastructure, Traffic, Transport and Public Space. Dai, Z., Niemeier, D., & Eisinger, D. (2008). Driving cycles: a new cycle-building method that better represents real-world emissions. Department of Civil and Environmental Engineering, University of California, Davis. Retrieved from https://www.researchgate.net/profile/Deb_Niemeier/publication/265495453_DRIVING_CYCLES_A_NEW_CYCLE-BUILDING_METHOD_THAT_BETTER_REPRESENTS_REAL-WORLD_EMISSIONS/links/55a13f7508ae1c0e046405e7.pdf Dozza, M., & Werneke, J. (2014). Introducing naturalistic cycling data: What factors influence bicyclists’ safety in the real world? Transportation Research Part F: Traffic Psychology and Behaviour, 24, 83–91. https://doi.org/10.1016/j.trf.2014.04.001   104  Fred Dyck. (2016). Senior Fitness Tuesday: Using your Heart Rate as a Fitness Guide. Retrieved from http://www.ywcasaskatoon.com/senior-fitness-tuesday-using-your-heart-rate-as-a-fitness-guide/ Fu, Z., Tian, Z., Xu, Y., & Qiao, C. (2016). A Two-Step Clustering Approach to Extract Locations from Individual GPS Trajectory Data. ISPRS International Journal of Geo-Information, 5(10), 166. https://doi.org/10.3390/ijgi5100166 Fyhri, A., & Fearnley, N. (2015). Effects of e-bikes on bicycle use and mode share. Transportation Research Part D: Transport and Environment, 36, 45–52. https://doi.org/10.1016/j.trd.2015.02.005 Giakoumis, E. G. (2017). Driving and Engine Cycles. Cham: Springer International Publishing. https://doi.org/10.1007/978-3-319-49034-2 Heinen, E., van Wee, B., & Maat, K. (2010). Commuting by bicycle: An overview of the literature. Transport Reviews, 30(1), 59–96. https://doi.org/10.1080/01441640903187001 Hood, J., Sall, E., & Charlton, B. (2011). A GPS-based bicycle route choice model for San Francisco, California. Transportation Letters, 3(1), 63–75. https://doi.org/10.3328/TL.2011.03.01.63-75 Hung, W. T., Tong, H. Y., Lee, C. P., Ha, K., & Pao, L. Y. (2007). Development of a practical driving cycle construction methodology: A case study in Hong Kong. Transportation Research Part D: Transport and Environment, 12(2), 115–128. https://doi.org/10.1016/j.trd.2007.01.002 Hunt, J. D. (n.d.). Investigation of Cycling Sensitivities.   105  Intergovernmental Panel on Climate Change, & Edenhofer, O. (Eds.). (2014). Climate change 2014: mitigation of climate change: Working Group III contribution to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. New York, NY: Cambridge University Press. Jamerson, F., & Ed Benjamin. (2018). Electric Bikes Worldwide Reports. Retrieved from http://www.ebwr.com/ Kamble, S. H., Mathew, T. V., & Sharma, G. K. (2009). Development of real-world driving cycle: Case study of Pune, India. Transportation Research Part D: Transport and Environment, 14(2), 132–140. https://doi.org/10.1016/j.trd.2008.11.008 Krizek, K. J., & Roland, R. W. (2005). What is at the end of the road? Understanding discontinuities of on-street bicycle lanes in urban settings. Transportation Research Part D: Transport and Environment, 10(1), 55–68. https://doi.org/10.1016/j.trd.2004.09.005 Langford, B. C., Chen, J., & Cherry, C. R. (2015). Risky riding: Naturalistic methods comparing safety behavior from conventional bicycle riders and electric bike riders. Accident Analysis & Prevention, 82, 220–226. https://doi.org/10.1016/j.aap.2015.05.016 Lin, J., & Niemeier, D. A. (2003). Regional driving characteristics, regional driving cycles. Transportation Research Part D: Transport and Environment, 8(5), 361–381. https://doi.org/10.1016/S1361-9209(03)00022-1 Ma, X., & Luo, D. (2016). Modeling cyclist acceleration process for bicycle traffic simulation using naturalistic data. Transportation Research Part F: Traffic Psychology and Behaviour, 40, 130–144. https://doi.org/10.1016/j.trf.2016.04.009   106  MacArthur, J., Dill, J., & Person, M. (2014). Electric Bikes in North America: Results of an Online Survey. Transportation Research Record: Journal of the Transportation Research Board, 2468, 123–130. https://doi.org/10.3141/2468-14 Menard, T., Miller, J., Nowak, M., & Norris, D. (2011). Comparing the GPS capabilities of the Samsung Galaxy S, Motorola Droid X, and the Apple iPhone for vehicle tracking using FreeSim_Mobile. In Intelligent Transportation Systems (ITSC), 2011 14th International IEEE Conference on (pp. 985–990). IEEE. Menghini, G., Carrasco, N., Schüssler, N., & Axhausen, K. W. (2010). Route choice of cyclists in Zurich. Transportation Research Part A: Policy and Practice, 44(9), 754–765. https://doi.org/10.1016/j.tra.2010.07.008 NACTO. (2012). Urban Bikeway Design Guide. National Association of City Transportation Officials, Washington DC. NITC. (2014). Regulations of E-Bikes in North America (No. NITC-RR-564). Nouri, P., & Morency, C. (2017). Evaluating Microtrip Definitions for Developing Driving Cycles. Transportation Research Record: Journal of the Transportation Research Board, 2627, 86–92. https://doi.org/10.3141/2627-10 Parkin, J., & Rotheram, J. (2010). Design speeds and acceleration characteristics of bicycle traffic for use in planning, design and appraisal. Transport Policy, 17(5), 335–341. https://doi.org/10.1016/j.tranpol.2010.03.001 Pfeil, U., Arjan, R., & Zaphiris, P. (2009). Age differences in online social networking – A study of user profiles and the social capital divide among teenagers and older users in MySpace. Computers in Human Behavior, 25(3), 643–654. https://doi.org/10.1016/j.chb.2008.08.015   107  Pierce, J. M. T., Nash, A. B., & Clouter, C. A. (2013). The in-use annual energy and carbon saving by switching from a car to an electric bicycle in an urban UK general medical practice: the implication for NHS commuters. Environment, Development and Sustainability, 15(6), 1645–1651. https://doi.org/10.1007/s10668-013-9454-0 Regional Trip Diary. 2011 Metro Vancouver Regional Trip Diary - Analysis Report. Retrieved from http://www.translink.ca//media/Documents/customer_info/translink_listens/customer_surveys/trip_diaries/2011%20Metro%20Vancouver%20Regional%20Trip%20Diary%20%20Analysis%20Report.pdf Robert J. Hijmans. (2016). geosphere: Spherical Trigonometry. Retrieved from https://CRAN.R-project.org/package=geosphere Robert J. Hijmans. (2017). raster: Geographic Data Analysis and Modeling. Retrieved from https://CRAN.R-project.org/package=raster Sayed, T., Zaki, M. H., & Autey, J. (2013). Automated safety diagnosis of vehicle–bicycle interactions using computer vision analysis. Safety Science, 59, 163–172. https://doi.org/10.1016/j.ssci.2013.05.009 Seers, P., Nachin, G., & Glaus, M. (2015). Development of two driving cycles for utility vehicles. Transportation Research Part D: Transport and Environment, 41, 377–385. https://doi.org/10.1016/j.trd.2015.10.013 Sener, I. N., Eluru, N., & Bhat, C. R. (2009). An analysis of bicycle route choice preferences in Texas, US. Transportation, 36(5), 511–539. https://doi.org/10.1007/s11116-009-9201-4 Sloan, L., Morgan, J., Burnap, P., & Williams, M. (2015). Who Tweets? Deriving the Demographic Characteristics of Age, Occupation and Social Class from Twitter User   108  Meta-Data. PLOS ONE, 10(3), e0115545. https://doi.org/10.1371/journal.pone.0115545 Stinson, M., & Bhat, C. (2003). Commuter bicyclist route choice: Analysis using a stated preference survey. Transportation Research Record: Journal of the Transportation Research Board, (1828), 107–115. Strauss, J., & Miranda-Moreno, L. F. (2017). Speed, travel time and delay for intersections and road segments in the Montreal network using cyclist Smartphone GPS data. Transportation Research Part D: Transport and Environment, 57, 155–171. https://doi.org/10.1016/j.trd.2017.09.001 Tao, S. R., Niu, X. J., & Chen, K. M. (2014). Study on the Bike Path Width Considering the Electric Bike. Applied Mechanics and Materials, 587–589, 1836–1839. https://doi.org/10.4028/www.scientific.net/AMM.587-589.1836 Tengattini, S. (2017). PHYSICAL CHARACTERISATION OF URBAN CYCLISTS FOR ADVANCED BICYCLE TRAVEL MODELS, 138. Tilahun, N. Y., Levinson, D. M., & Krizek, K. J. (2007). Trails, lanes, or traffic: Valuing bicycle facilities with an adaptive stated preference survey. Transportation Research Part A: Policy and Practice, 41(4), 287–301. https://doi.org/10.1016/j.tra.2006.09.007 Twaddle, H., & Grigoropoulos, G. (2016). Modeling the speed, acceleration, and deceleration of bicyclists for microscopic traffic simulation. Transportation Research Record: Journal of the Transportation Research Board, 2587, 8–16. https://doi.org/10.3141/2587-02   109  Twaddle, H., Schendzielorz, T., & Fakler, O. (2014). Bicycles in urban areas. Transportation Research Record: Journal of the Transportation Research Board, 2434, 140–146. https://doi.org/10.3141/2434-17 Vélo Québec. (2013). Le vélo dans l’avenir des villes: Propositions 2014–2021 de Vélo Québec. Wang, Q., Huo, H., He, K., Yao, Z., & Zhang, Q. (2008). Characterization of vehicle driving patterns and development of driving cycles in Chinese cities. Transportation Research Part D: Transport and Environment, 13(5), 289–297. https://doi.org/10.1016/j.trd.2008.03.003 WHITT, F. R. (1971). A Note on the Estimation of the Energy Expenditure of Sporting Cyclists. Ergonomics, 14(3), 419–424. https://doi.org/10.1080/00140137108931261 Willis, D. P., Manaugh, K., & El-Geneidy, A. (2015). Cycling Under Influence: Summarizing the Influence of Perceptions, Attitudes, Habits, and Social Environments on Cycling for Transportation. International Journal of Sustainable Transportation, 9(8), 565–579. https://doi.org/10.1080/15568318.2013.827285 Wilson, D. G. (2004). Bicycling Science (3rd ed.). Cambridge, MA: MIT Press. Wolf, A., & Seebauer, S. (2014). Technology adoption of electric bicycles: A survey among early adopters. Transportation Research Part A: Policy and Practice, 69, 196–211. https://doi.org/10.1016/j.tra.2014.08.007 Xu, C., Li, Q., Qu, Z., & Tao, P. (2015). Modeling of speed distribution for mixed bicycle traffic flow. Advances in Mechanical Engineering, 7(11), 168781401561691. https://doi.org/10.1177/1687814015616918    110  Appendices Appendix A: Survey consent form    111      112  Appendix B: Survey questionnaire    113     114     115     116     117  Appendix C: Smartphone application registration form     118     119       120  Appendix D: Online uploading registration form    121     122     123     124     125     126     127      128  Appendix E: Instructions    129      130      131      132      133   

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0365817/manifest

Comment

Related Items