Designing a Bicycle and Pedestrian Count Program in Blacksburg, VA Steve Hankey (Virginia Tech) Andrew Mondschein (U or Virginia) Ralph Buehler (Virginia Tech)
Issue/objective Issue No systematic traffic monitoring of bikes/peds Traffic counts important for planning and funding decisions Objective Implement a bike/ped monitoring program: Proof-of-concept Sample ~10% of the network and estimate traffic for all street segments Project Team Steve Hankey (PI), Andrew Mondschein (Co-PI), Ralph Buehler (Co-PI) Graduate students: Tianjun Lu, Kyle Lukacs Community partners: Town of Blacksburg 2
Approach Objective: Systematic Count Program Task 1: Validation of automated counters Task 2: Site selection and data collection Task 3: Estimate AADT for all count sites Task 4: Spatial models to estimate AADT across network Pneumatic tubes (bike; streets) Passive Infrared (ped; sidewalks) Radiobeam (bike & ped; trails) Reference sites (n = 4; full year) Short-duration sites (n = 97; 1-week) Impute missing data (ref. sites) Make scaling factors (ref. sites) & apply (short-duration sites) Tabulate land use variables at count sites Build direct-demand models of traffic 3
Task 1: Counter Validation ~240 hours of validation counts Component of graduate course Develop correction equations MetroCount pneumatic tube counter (n=12) Eco-counter Pyro passive infrared counter (n=10) RadioBeam bicycle-people counter (n=3) 4
Manual Count Manual Count Manual Count Manual Count Task 1: Correction equations Good fit with linear correction (mostly) Correction varies by counter Polynomial fit may be needed at higher volumes More data at high volume sites would be helpful 30 20 10 0 1000 750 500 250 0 Pneumatic Tubes Polynomial R² = 0.90 Linear R² = 0.89 0 10 20 30 Automated Count Infrared Polynomial R² = 0.97 Linear R² = 0.94 0 250 500 750 1000 Automated Count 5 80 60 40 20 0 160 120 80 40 0 Radiobeam: Bike Linear R² = 0.92 0 20 40 60 80 Automated Count Radiobeam: Ped Linear R² = 0.92 0 40 80 120 160 Automated Count
Task 2: Site selection Full year of data Seasonal and daily patterns Variety of location types 1 week of data Good spatial coverage Systematic selection 6
Task 2: Stratified, pseudo-random selection Total: 29% Bike lane: 10% No facility: 19% Total: 51% Bike buildout: 36% Low centrality: 15% Total: 20% Trail transport: 10% Trail neighborhood: 10% 7
Task 2: Final count locations 8
Task 2: Data collection Questions for sampling design How to scale short-duration counts? How long to count at each site? What time of season to count? Case Study: Minneapolis 9
Task 2: Data collection - Scaling Approach 1: Traditional Approach 2: New 10
Mean absolute AADT error Task 2: Data collection Length of count 40% Traditional Old scaling method scaling method New scaling method 30% 20% 10% 0% 0 5 10 15 20 25 Number of short-duration sampling days 11
Task 2: Data collection Season of count 12
Task 2: Summary of sampling campaign 1-week short duration counts (n=97) April October 10% random re-sample of locations Event log and statistical check used to clean data 13
Task 3: Estimate AADT at all count sites Weather and temporal variables Valid data at reference sites Negative binomial regression models Impute missing data and estimate AADT Day-of-Year scaling factors 1-week counts (short-duration sites) AADT (short-duration sites) 14
Task 3: Generate scaling factors 3 Bicycle Day-of-Year Scaling Factors 2 Estimated AADT for reference sites 1 Draper College Giles Huckleberry Bike AADT 21 54 55 179 Ped AADT 98 4,232 289 518 0 Jan 1 Feb 22 April 15 June 6 July 28 Sep 18 Nov 9 Dec 31 3 Pedestrian Day-of-Year Scaling Factors 2 1 0 Jan 1 Feb 22 April 15 June 6 July 28 Sep 18 Nov 9 Dec 31 14
Task 3: Map AADT 16
Task 3: AADT by road type and bicycle facility 250 200 Bicycle AADT * * Bicycle AADT: * indicates statistical significance compared to a road w/o a bike lane (p<0.05). 150 100 50 Pedestrian AADT: Traffic is higher on local roads. Likely due to roads on VT campus classified as local. 0 1000 Road without bike lane Road with bike lane Trail tranport Trail neighborhood Pedestrian AADT 750 500 250 0 17 Local road Major road Trail tranport Trail neighborhood
Task 4: Direct-demand models Goal 1: Determine predictors (land use and infrastructure) of active travel. Goal 2: Estimate spatial patterns of bicycle and pedestrian traffic for exposure assessment. Bike/ped counts Facility-demand models 18
Task 4: Stepwise regression modelbuilding 1. Compile database of candidate predictor variables. 2. Tabulate each variable at many geographic scales (network buffers; 100 3,000 meters). 3. Select predictor variables most correlated with traffic volumes. 4. Keep only variables that improve model. a) Statistical significance (p-value) b) Multi-collinearity (VIF) 5. Reasonable model fit (adj-r 2 ). a) Fully specified models: 0.53-0.64 b) Reduced-form models: 0.42-0.58 19
Task 4: Preliminary model results Total variables (n=133) Variables (n=12) Buffers (n=11); 100 3,000 m Transportation All roads, Off-street trail, On-street facility, Intersections, Centrality Land use Retail, Industry, Open space, Employment density Population House/Pop density, HH income Total selected (n=14); 11% Most commonly selected... Variable Number selected Pop Density 4 Off-street trail 3 Retail 2 Centrality 2 Employment 2 All roads 2 Bike lane, Income, Open space 28% selected at <250 meters 1 20
Task 4: Next steps for direct-demand models Add other variables Update for VT campus Non-residential address, tree cover, bus stops, sidewalks, major/local roads Finalize core models for bike and ped traffic Extrapolate estimates to all street segments 21
Summary and future work Summary of work Successfully implemented systematic count campaign in Blacksburg Developed proof-of-concept for estimating AADT Developed tool for estimating AADT at locations without counts Cost: ~$60,000 (equipment) plus 1.5 year of 20-hour/wk graduate student (~$50,000) Future work Hour-of-day direct-demand models Test method in other (i.e., larger) communities Work with Town of Blacksburg to: Assess buildout of bike/ped network Evaluate investments 22
Thanks and contact info Project Team Steve Hankey (PI), Andrew Mondschein (Co-PI), Ralph Buehler (Co-PI) Graduate students: Tianjun Lu, Kyle Lukacs Community partners: Town of Blacksburg Contact Steve Hankey Urban Affairs and Planning, Virginia Tech hankey@vt.edu 540.231.7508 23