estimating the traffic flow impact of pedestrians with limited data

SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY

Estimating the Traffic Flow Impact of Pedestrians With

Limited Data

Daniel Tischler, Elizabeth Sall,Lisa Zorn & Jennifer Ziebarth

14th TRB Planning Applications ConferenceMay 6th, 2013


2

Making the Case

I have problems.

I’m trying to model San Francisco traffic conditions, but

Pedestrians affect vehicle capacityMy dynamic traffic assignment model is needyData is scarce


3

Network capacity

Road capacity is a key input in traffic assignmentWe assign capacities to roads by facility classification schemes


4

Network capacity

In our DTA model, signal timing becomes the primary determinant of capacity


5

Pedestrians interact with vehicles

Lots of pedestrians crossing the street prevent cars from turning


6

Pedestrian volumes vary

Our meso-level, dynamic assignment model does not simulate pedestrians.It does not know that right turns at “A” are more restricted than at “B”

Intersection A Intersection B


7

Methodologies

Analytical methodsSimulation methodsLocal observations


8

Analytical approaches


9

Analytical approaches


10

Pedestrian volume – vehicle flow relationship

• Assumes 2 sec veh-veh headway, 50% green time

• Doesn’t allow for ped volumes > 5,000

0 500 1000 1500 2000 25000

500

1000

1500

2000

Right Turn Saturation Flow Rate (HCM 2010)

Pedestrians crossing Street (peds / hr)

Sa

tura

tio

n F

low

Ra

te(v

eh

/ h

r)


11

Comparing different methods

HCM2010

Rouphail and Eads, 1997


12

Applying pedestrian friction in the model

Saturation flow rate

Follow-up time

Green time


13

Scale of Application

55%

74%

87%

Existing approach

Under consideration

Area Type Method

Neighborhood TypeMethod

Under consideration

Unique IntersectionMethod

Do nothing approach

Uniform Parameters Method


14

Resources

SFMTA pedestrian count programSFMTA pedestrian modelProject-related pedestrian countsPedestrian-vehicle observations


15

SFMTA pedestrian count program

• 2-hour pedestrian counts at 50 locations (2009/2010)

• Six automatic pedestrian counters (24/7 observations)

Actual pedestrian count binder


16

Automated pedestrian counters

• Time and day profiles

Union Square

Castro

Tenderloin


17

Pedestrian count locations


18

HCM 2010 flow rate adjustment factors


19

SFMTA pedestrian model

• Used count data and log-linear regression model to estimate pedestrian volumes throughout San Francisco

lnYi = β0+ β1 X1i+ β2 X2i+ … + βj Xji

• Combined with auto traffic and collision data to develop pedestrian crossing accident risk factors


20

Pedestrian volume estimates


21

Local validation of pedestrian estimation

8th7t

h6t

h5t

h4t

h3r

d

0

2,000

4,000

6,000

8,000

10,000

12,000

Market

Howard

Ped Model Estimates

PM

Peak

Hour

Ped C

ross

ings

8th

7th

6th

5th

4th

3rd

0

2,000

4,000

6,000

8,000

10,000

12,000

Market

Howard

Ped Counts

PM

Peak

Hour

Ped C

ross

ings


22

Pedestrian-vehicle interaction observations

SFCTA staff observed downtown intersections of varying pedestrian volumesCollected information on quantity of pedestrians and relative restriction of vehicle turning movement capacityThe urban environment is complex:

• Groupings of peds, directions of travel bicycles, unique timing plans, variations in geometry, etc.


23

0 500 1,000 1,500 2,0000%

20%

40%

60%

80%

100%

120%

Pedestrian Impedance of Turning Vehicles

Pedestrian Flow in Crosswalk (peds/hr during walk phase)

Perc

en

t re

du

cti

on

in

Veh

icle

C

ap

acit

y

Observed pedestrian-vehicle interactions

Incremental pedestrians have the greatest impact at low pedestrian volumes


24

HCM2010

Rouphail and Eads, 1997

Observed pedestrian-vehicle interactions


25

Model Tests

Uniform parameters methodArea type methodNeighborhood type method – sat flow rateNeighborhood type method – green time reduction


26

Focus on pedestrian count-rich area

SoMa pedestrian count locations1. 4th / Market2. 6th / Market3. 8th / Market4. 6th /

Mission5. 3rd /

Howard6. 7th / Folsom

1

2

6

4

5

3


27

Hourly pedestrian counts

3,300 3,000 10,000

1,100

600

1,200

Market St

Mission St

Howard St

Folsom St

3rd St

5th St

7th St


28

Network parametersUniform parameter method

All Signals

Sat flow

1,800

Spacing

2.0s


29

Network parametersArea type method

Hi PedsSat flow

1,620

Spacing

2.22s

Sat flow

1,710

Spacing

2.11s

Med Peds


30

Network parametersNeighborhood type method

Sat Flow810

Spacing

4.4s

Sat Flow180

Spacing

20s

Spacing

2.9s

Spacing

2.4s

Sat Flow1,260

Sat Flow1,530

Adjust saturation flow rates


31

Network parametersNeighborhood type method

55% less

green

90% less

green

30% less

green

15% less

green

Reduce green time for turning movements


32

Testing pedestrian friction methods

• Evaluation not yet complete!

• NumbersAt select subarea nodes, turning movement volume validation improves with neighborhood type method

• QueuingNeighborhood method seems to produce more realistic queue formation than area method

Area type method

Neighborhood type method


33

Conclusions

• Pedestrian-vehicle friction is very important at some locations!

• Numerous options to make assignment models sensitive to pedestrians

• The most realistic approaches are very difficult

• We still have work to do:• Run more tests• More local validation• Develop better neighborhood system• Incorporate time of day factors• Develop strategy for scenarios that

change pedestrian volumes


The End.

Questions?

estimating the traffic flow impact of pedestrians with limited data

Documents

san franciscolny

automatic pedestrian

dynamic assignment model

dta model

ped volumes

loglinear regression

green timedoesnt

sec vehveh headway