Downloaded 31 times
Completing the Cycle: Incorporating CycleTracks into SF-CHAMP
The document discusses the CycleTracks project, which collected GPS data from cyclists in San Francisco to understand their route choices. It summarizes who participated in CycleTracks and the types of data collected. The data was then used to estimate a bike route choice model and update SF-CHAMP, the travel demand model, to better represent bike travel and predict the effects of bike infrastructure projects.
More Related Content
Similar to Completing the Cycle: Incorporating CycleTracks into SF-CHAMP
More from San Francisco County Transportation Authority Technology Data + Analysis
Completing the Cycle: Incorporating CycleTracks into SF-CHAMP
- 1. Completing the Cycle: Incorporating CycleTracks into SF-CHAMP Using technology to understand the needs of cyclists SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY Fall 2012
- 2. Outline 1. Why make CycleTracks? 2. What does CycleTracks do? 3. Who used CycleTracks and why? 4. What data did we get from CycleTracks? 5. What did we do with that data? 6. Evolution and future of CycleTracks
- 3.
- 4. Why CycleTracks? Need toprioritize projects, including bike projects. Estimate a bike choice model that evaluated various bike infrastructure features Needed bike route choice data on a budget.
- 5.
- 6. Enter personal data(optional)
- 7.
- 8.
- 9. That’s it? Bells andwhistles could promote deviation from planned route. Features! Good Data. Flare! Yawn. More users!
- 10. 3. Who usedCycleTracks and Why? - User Recruitment - Participants
- 15.
- 16. SF Participants: Fall2009 to Spring 2010 CycleTracks BATS N-366 N=153 z-stat Age Mean 34 33 1.1 Gender Female 21% 36% -3.5 Cycling Frequency Daily 60% Several Times/Week 34% Several Times/Month 7% Less than once a month 0% N/A
- 17. 4. What datadid we get? - Data Quality - Data Summaries
- 18. Data Quality: somegood, some bad
- 19. Urban Canyon Effect Haight Ashbury vs Downtown
- 20. GPS Signal atBeginning of Trip
- 21. Not on aBike
- 22. Post Processing Warranted 5,178 traces Gaussian 497 users smoothing Activity & mode detection 3,034 bike Map stages h matching 366 users (Schüssler & Axhausen 2009)
- 23. 5. What didwe do with the CycleTracks Data?
- 24. Matched Route Featuresto the Chosen Route…
- 25. …as well asto a set of routes that were not chosen
- 26. What makes uschoose one bike route over another ? Personal Trip Info Features Route Which route Features of was Available Route chosen? Routes Choice Model
- 27. Estimation results Attribute Coef. SE t-stat. p-val. Length (mi) --1.05 0.09 --11.80 0.00 Turns per mile --0.21 0.02 --12.15 0.00 Prop. wrong way --13.30 0.67 --19.87 0.00 Prop. bike paths 1.89 0.31 6.17 0.00 Prop. bike lanes 2.15 0.12 17.69 0.00 Cycling freq. < several per wk. 1.85 0.04 44.94 0.00 Prop. bike routes 0.35 0.11 3.14 0.00 Avg. up-slope (ft/100ft) --0.50 0.08 --6.35 0.00 Female --0.96 0.22 --4.34 0.00 Commute --0.90 0.11 --8.21 0.00 Log(path size) 1.07 0.04 26.38 0.00 2,678 weighted observations, ρ2 = 0.28
- 28. Average Marginal Ratesof Substitution MRS of Length on Street for Value Units Turns 0.10 mi/turn Total Rise 1.12 mi/100ft Length Wrong Way 4.02 None Length on Bike Paths 0.57 None Length on Bike Lanes 0.49 None Length on Bike Routes 0.92 None SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 28
- 29. Updates to SF-CHAMP Synthesized Core, 3 iterations Population Work Location, Land Use Destination Choice, Mode Choice Tour Generation Networks Networks Logsums +Bike Vars! Tour & Trip Mode Choice Bike Route Choice Set Road & Transit Non-Motorized Bike Generation & Assignment/ Skimming (Distances) Logsums Skimming Skimming Initial Road & Transit Assignment/ Skimming Final Bicycle Assignment SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 29
- 30. Bike Accessibility: From4th and King SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 30
- 31. Bike Accessibility: To4th and King SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 31
- 32. Bike Logsums: From4th and King Effect of Bike Plan Build SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 32
- 33. Bike Logsums: To4th and King Effect of Bike Plan Build SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 33
- 34. Preliminary Results: Tour ModeChoice Sensitivity Tour Difference Daily Tours v4.1 Harold v4.3 Fury Bike 300 0.1% 1,300 0.9% Walk 300 0.0% 200 0.0% Transit 200 0.0% -900 -0.1% Auto -1,000 -0.0% -600 -0.0% Total -200 -0.0% 0 0.0% SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 34
- 35. Preliminary Results: Trip ModeChoice Sensitivity Trip Difference Daily Tours v4.1 Harold v4.3 Fury Bike 500 0.1% 3,000 0.8% Walk 1,100 0.0% -500 -0.0% Transit 850 0.0% -600 -0.0% Auto -2,400 -0.0% -1,300 -0.0% Total 0 0.0% 600 0.0% SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 35
- 36. SF-CHAMP Predicted BikeTrips Bikes / hour 0 180 20 360 SF-CHAMP v4.1 “Harold”
- 37. 6. Evolution and Futureof CycleTracks
- 38. All Open Source • GPL3 License • Code on GitHub • Fork us! www.github.com/sfcta
- 39. e.g. AggieTrack http://aggietrack.com
- 40. CycleTracks Works Everywhere… Wealready have the database set up Agencies can download “scrubbed” data Austin,TX Monterey Bay, CA …and more!
- 41.
- 42.
- 43. Bike Accessibility: FromInner Sunset SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 43
- 44. Bike Accessibility: ToInner Sunset SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 44
- 45. Bike Logsums: FromInner Sunset Effect of Bike Plan Build SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 45
- 46. Bike Logsums: ToInner Sunset Effect of Bike Plan Build SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 46
Editor's Notes
- #17 Here is a comparison of the demographics of the participants whose traces survived the data processing to the subpopulation from the Bay Area Travel Survey that reported a cycling trip in San Francisco. As you can see, the CycleTracks sample is over twice the size of BATS, which contained 50,000 households, illustrating why seeking a representative sample to study cycling is not feasible. But, our sample is biased. While the mean age in the two samples are not significantly different, our study does include a lower proportion of women at 21% compared to BATS’ 36%. While we don’t have a population to compare cycling frequency against, we also suspect that our sample is biased toward frequent cyclists. The bias is a limited problem because we were able to account for it with interaction variables in model estimation.
- #23 Schuessler, Nadine and Kay W. Axhausen. “Processing Raw Data from Global Positioning Systems Without Additional Information,” Transportation Research Record : Journal of the Transportation Research Board, No 2105. Washington D.C., 2009, pp. 28-35. http://trb.metapress.com/content/tv306m812140p330/
- #25 SharrowsCongestionNight-timeNo bike lanesCapacity of roadwayBike laneBike pathCrimeWeather
- #26 We use a “Doubly Stochastic Route Search” to find other potential routes in the available choice setBovy, P. & Fiorenzo-Catalano, S. (2007), “Stochastic route choice set generation: behavioral and probabilistic foundations,” Transportmetrica 3, 173-189.
- #28 Here are the coefficients from the path size multinomial logit route choice model. Obviously, cyclists prefer shorter routes, with fewer turns, and don’t go the wrong way down a one way street unneccessarily. The coeficients on the proportions of the different bicycle facility types are measured on the same scale, and so represent the relative preferences for these treatments. Bike lanes are preferred the most, especially by infrequent cyclists, a preliminary indication that installing bicycle lanes may attract new cyclists. Hill climbing is especially disfavored by women and on commute trips. The path size variable corrects for the correlation between alternatives due to route overlap. The coefficient is not significantly different from the theoretically correct value in a model with a scale parameter of one, another indication of the quality with which our choice sets represent the consideration sets. Traffic volume, vehicle speed, number of lanes, crime, and rain had no effect.
- #29 The average cyclist will Avoid a turn if it costs no more than one-tenth of a mileAvoid climbing a hill 100 feet tall as long as the detour is less than roughly one mileAvoid traveling the wrong way down a one-way street unless doing so saves more than four times the distance elsewhereAdd a mile on bike lanes in exchange for only half a mile on ordinary roadsBike paths vs Bike Lanes could be due to limited off-street bike path facilities in SF and other factors that make them less attractive (although Krizek also found similar preferences).Mention the other variables that were significant in estimation: females and work-commuters were more hill-averse
- #31 Talking points/or circle/or flip through
- #33 Yellow doesn’t stand out enough
- #35 Not apples to apples: the bike network coded in Harold was much more aggressive. That said, the results are still interesting:Harold has an auto-stick, but no carrot, so the mode switchers switch to whatever is best for them, which happens to be pretty even across bike, walk and transit.Fury has a very strong bike-carrot, and walk modes benefit mildly from the road diets. Biking draws from auto but also transit because of tour distance.
- #36 Harold again has the auto stick. Walk trips are up because of transit tours being up.Fury has the bike carrot. Walk trips are down (despite the walk tours being up) because of the transit tours being down – many walk trips are part of a transit tour.
- #39 Our code is open source, and there are a number of agencies who have tried their hand at modifying it to their own needs.
- #44 Back pocket all four of these?