What Smartphone Bicycle GPS Data Can Tell Us About Current Modeling Efforts Katie Kam, The University of Texas at Austin Qiqian (Angela) Yang, The Fresno Council of Governments Jennifer Duthie, Network Modeling Center, The University of Texas at Austin Presenter: Qiqian (Angela) Yang COLLABORATE. INNOVATE. EDUCATE. Take Away From This Presentation: • GPS app to track cycle routes estimate bike trips for travel demand modeling • A GPS data “cleaning” process a sample set of data • Differences in GPS data and other methods of estimation unique bicycle demand characteristics COLLABORATE. INNOVATE. EDUCATE. Prior Study Examples • GPS-based bicycle route choice model • GPS data analysis for commercial vehicle demand modeling • Location-based social network study Structure of Location-Based Social Network Example of GPS Bike Tracks Example of GPS Truck Trips COLLABORATE. INNOVATE. EDUCATE. Research Purposes • Find out GPS bicycle tracking data V.S. current modeling estimation efforts • Move forward Incorporating GPS bicycle data into the transportation demand modeling process Questions we asked • The role GPS bicycle data has? In the trip generation and trip distribution steps • How to prepare the data? Resulting dataset can be considered a sample dataset suitable for MPO model COLLABORATE. INNOVATE. EDUCATE. Study Area • The metropolitan area of Austin, Texas • Urban area in Austin, Texas • Why Austin? Cycling city, bicycle friendly 4.6 miles per sq. mile Resources Austin Bike lane Map – Downtown Area COLLABORATE. INNOVATE. EDUCATE. Data Collection 1 • CycleTrack app was developed by the San Francisco County Transportation Authority • Collected by volunteers that biked (not randomly selected) in Austin, Texas. 316 volunteers May 1st to October 31st, 2011 1,048,576 continuously collected GPS points CycleTracks Application Screens COLLABORATE. INNOVATE. EDUCATE. Bicyclist Recruiting Process CycleTracks Austin Website Screen Shot COLLABORATE. INNOVATE. EDUCATE. Postcards Distributed in May and September of 2011 to Area Bicyclists COLLABORATE. INNOVATE. EDUCATE. Data Collection 2 • CAMPO – 2006-2007 Household Travel Survey. • CAMPO – 2010 estimation, from 2010 Travel Demand Model. CAMPO 2010 Model - Bike Trips Origin and Destination Data Screen Shot COLLABORATE. INNOVATE. EDUCATE. Methodology - CycleTrack GPS Data Preparation Recruit Bicyclists Attach TAZ and Time Period GPS Kept the Record points to weekday Bike Trips bike trips trips Points to trips Delete Recurring Trips Peak vs Off Peak Reformat GPS Trip Tables Trip tables finalization COLLABORATE. INNOVATE. EDUCATE. Summery Statistic of Trip User Characteristics Gender (n = 302) Age of Participants (n = 304) COLLABORATE. INNOVATE. EDUCATE. Why Data Cleaning ? • Additional cleaning – recurring trips • Keep the O/D pattern consistent with the CAMPO HH Travel Survey method; Before After COLLABORATE. INNOVATE. EDUCATE. Methodology - Data Cleaning Layers 1,048,576 Issues • Kept beginning and ending points. points 650 weekday trips 486 nonrecurring trips • Joined CAMPO 2010 TAZ layer. • Defined trip time periods. • Defined the recurring trips. Four time period? • Finalized the final CycleTracks dataset. No User ID? Repeated trips in the same day? COLLABORATE. INNOVATE. EDUCATE. Data Analysis • Trip Estimations Comparison Compare the CycleTrack trip tables with the CAMPO estimation COLLABORATE. INNOVATE. EDUCATE. Trip Estimation Comparison Origin – Off Peak Hour Origin – Peak Hour Legend Legend HWY HWY polylakes polylakes O_OP_ZeroTrip O_PK_ZeroTrip Differences_GPS_V.S._CAMPO Differences_GPS_V.S._CAMPO diff_op_or diff_pk_or -14 - 0 (53) -19 - 0 (77) 0 (1,750) 1 - 0 (1,771) 1 - 20 (231) 1 - 20 (268) 21 - 40 (93) 21 - 40 (94) 41 - 100 (110) 41 - 100 (44) 101 - 275 (21) 101 - 129 (4) COLLABORATE. INNOVATE. EDUCATE. Trip Estimation Comparison Origin – Off Peak Hour Legend Origin – Peak Hour Legend HighUsageBikeRoute HighUsageBikeRoute LowUsageBikeRoute LowUsageBikeRoute O_OP_NegativeTrip O_PK_NegativeTrip diff_op_or diff_pk_or -14 - -12 (2) -19 - -12 (5) -11 - -5 (3) -11 - -5 (9) -4 - -3 (6) -4 - -3 (10) -2 (13) -2 (11) -1 - 0 (29) -1 - 0 (41) COLLABORATE. INNOVATE. EDUCATE. Trip Estimation Comparison Destination – Off Peak Hour Destination – Peak Hour Legend Legend HWY polylakes D_OP_ZeroTrip Differences_GPS_V.S._CAMPO diff_op_de -10 - 0 (16) 0 (845) 1 - 20 (1,182) 21 - 40 (145) HWY polylakes D_PK_ZeroTrip Differences_GPS_V.S._CAMPO diff_pk_de -22 - 0 (46) 0 (1,040) 1 - 20 (1,081) 21 - 40 (63) 41 - 100 (65) 41 - 100 (24) 101 - 133 (5) 101 - 381 (4) COLLABORATE. INNOVATE. EDUCATE. Trip Estimation Comparison Destination – Peak Hour Destination – Off Peak Hour Legend Legend HighUsageBikeRoute HighUsageBikeRoute LowUsageBikeRoute LowUsageBikeRoute D_OP_NegativeTrip D_PK_NegativeTrip diff_op_de diff_pk_de <-13 (0) -22 - -13 (1) -12 - -5 (1) -12 - -5 (4) -4 - -3 (2) -4 - -3 (2) -2 (3) -2 (9) -1 - 0 (10) -1 - 0 (28) COLLABORATE. INNOVATE. EDUCATE. Comparison of CAMPO and GPS Data Collection CycleTrack Size of valid sample Valid bicycle trips Sample collection Who completed the survey Available data unit Data update frequency Cost Other applications 316 486 Convenience Bicyclists Trips Anytime Many COMPARISON CAMPO CAMPO GPS equipment 1500 153 421 N/A Random Random Householders Householders Days riding bike/per week 10 years Total $1 million, $200/per survey Less (policy analysis) CycleTracks : Non-authorized travel monitoring Transportation infrastructure need analysis Air quality/GHG emission reduction study Transportation and public health modeling, etc. COLLABORATE. INNOVATE. EDUCATE. Future Research • Explore a more refined method in cleaning data (e.g., clustering method). • Consider seasonal factors, and geographic differences. May Jun Oct Jul Sep Aug CycleTracks Recorded Bicycle Trips in Austin per Week COLLABORATE. INNOVATE. EDUCATE. Future Research - Trip Purpose Comparison Account for possible sample bias (smartphone users) Structure approach to get the cycle participators Home Based Non Work Trip Commute CAMPO TDM’s Trip Purpose CycleTrack Participants’ Trip Purpose COLLABORATE. INNOVATE. EDUCATE. Take Away From This Presentation: • CycleTracks or other similar GPS app to track cycle routes may be used to estimate bike trips for travel demand modeling • Acquired GPS data requires a data “cleaning” process that converts the data into a sample set of data • Seeing differences in GPS data results to other methods of estimation (e.g., household surveys) can reveal areas of town with unique bicycle demand characteristics COLLABORATE. INNOVATE. EDUCATE. Thank you! Contact information: Angela Yang: ayang@fresnocog.org Katie Kam: katie.a.larsen@utexas.edu Jennifer Duthie: jduthie@mail.utexas.edu COLLABORATE. INNOVATE. EDUCATE.