17tcs walk, don’t run? advancing the state of the practice in pedestrian demand modeling

WALK, DON’T RUN? ADVANCING THE STATE OF THE PRACTICE IN

PEDESTRIAN DEMAND MODELING

Transportation and Communities SummitSeptember 12, 2017

Dr. Kelly J. CliftonPortland State University

2

Dr. Joseph BroachPortland State University

Dr. Patrick SingletonUtah State University

Jaime Orrego OnatePortland State University

Dr. Robert SchneiderUniversity of Wisconsin, Milwaukee

Model of Pedestrian Demand (MoPeD)

3

TAZ = transportation analysis zonePAZ = pedestrian analysis zone

Trip Generation (PAZ)

Trip Distribution or Destination Choice (TAZ)

Mode Choice (TAZ)

Trip AssignmentPedestrian Trips

Walk Mode Split (PAZ)

Destination Choice (PAZ)

I

III

All Trips Pedestrian Trips Vehicular Trips

Opportunities & challengesBehavioral research/data/methods

Adapted from: Wegener and Fürst, 1999

Decision sequencing:activity, mode, destination; mode, destination, activity; destination, activity, mode

Destination choice considerations – choice set generation

Willingness to walk

Path/route choice considerations

Behavioral research

5

Behavioral Research

Built environment– Thresholds & nonlinearities – Mixing– Scale

Lifestyle questions: – Vehicle ownership & residential location– Attitudes, motivations & values

Positive Utility of Travel– What aspects?– Diminishing returns?

Mode feedbacks to trip generation

6

Spatial/Temporal Scale

• Depends on output needed for policy/research

• Capture variations in the pedestrian built & natural environment

• Spatial accuracy• Theory/Behavior

7

Walking Behavior

• Passive data sources– Trip-level information– Multi-day– Multi-modal– Destinations– Routes & speeds

• But also need…– Motivations &

considerations– Barriers– Trips not made

8

Built environment• How & what to

represent?• Indices, proxies• Forecasting

S.R. Gehrke, & K.J. Clifton. (2016). Toward a spatial-temporal measure of land-use mix. Journal of Transport and Land Use, 9(1):171–186S.R. Gehrke, & K.J. Clifton. (2014). Operationalizing land use diversity at varying geographic scales and its connection to mode choice: Evidence from Portland, Oregon. Transportation Research Record: Journal of the Transportation Research Board 2453: 128-136.

9

Networks

• Network representation

• How do we attribute networks?

• Feedbacks of travel costs

• Do we need to assign trips to a network? Broach, J. P. (2016). Travel mode choice framework incorporating realistic bike and walk routes (Order

No. 10061477). Available from Dissertations & Theses @ Portland State University; ProQuest Dissertations & Theses Global.

10

Link to Health Outcomes

• Health impact analysis

• Total time spent walking + speeds

• Physical activity budgets

• Crash risk exposure• Pollutant exposure• Feedback into life

expectancyWoodcock J, Givoni M, Morgan AS. Health Impact Modelling of Active Travel Visions for England and Wales Using an Integrated Transport and Health Impact Modelling Tool (ITHIM). Barengo NC, ed. PLoS ONE. 2013;8(1):e51462

11

ObjectivesUnderstand motives for developing and using pedestrian demand models

Share knowledge & experiences

Discuss key challenges and opportunities in pedestrian modeling

Develop an agenda for improving the state of the research

OutcomesDefine state of the practice

White paper on state of the practice and research needs

TRB workshop

Discuss next steps for MoPed & other efforts

12

Agenda 830-9 Introductions and framing of the workshop (Clifton) 9-930 Why model pedestrian demand? (Singleton) 930-1030 What are the data needs and opportunities? (Schneider) 1030-945 Break 11AM-12 What is the appropriate scale? (Broach and Orrego)12-13 Lunch break13-14 How do we represent networks and attributes? (Broach) 14-1445 How do we forecast inputs? (Clifton) 1445-15 Break 15-1545 How can model outputs link to health impact, safety, and

other modeling tools? (Singleton)1545-16 What are the most important next steps to improve

practice? (Clifton) 13

Pedestrian modeling: State-of-the-practice

“Walk, don’t run?” WorkshopTransportation and Communities Summit

Portland, OR — 12 September 2017

Why model pedestrian demand?

analyze health & safety impacts

utilize new data resources

mode shifts air quality & GhG

plan for pedestrian investments& non-motorized facilities

Early (regional) pedestrian modeling

1988 Metropolitan Service District (Portland, OR) Home-based motorized/non-motorized mode split model

1993 Sacramento Association of Governments (SACOG) Mode choice model w/ separate walking & bicycling modes

Late 1990s Baltimore, Boston, Chicago, Hampton Roads, Los Angeles,

Philadelphia, Portland, Sacramento, San Francisco Bay Area

2005 TRB Special Report 288 54% (35) of large MPOs reported non-motorized modeling

Early pedestrian environment measures

1988 Maryland-National Capital Park and Planning Commission (Washington, DC) Pedestrian and Bicycle Friendliness Index

1990s Making the Land Use, Transportation, Air Quality Connection (LUTRAQ) (Portland, OR) Pedestrian Environment Factor (PEF)

Used in Chicago, Hampton Roads, Miami, Philadelphia, Portland, Sacramento, Salt Lake City, …

2000s Transitioning from subjective indices to objective measures

Pedestrian modeling frameworks

State-of-the-practice reviews Summer 2012 & Summer 2017

Largest 48 US MPOs (> 1,000,000 population)

Reviewed model documentation 2012: 63% model non-motorized travel; 60% in mode choice

2017: 69% model non-motorized travel; 82% in mode choice

Online questionnaires to MPO modelers 2012: 29 responses (60%)

2017: 31 responses (65%)

State-of-the-practice (2012)

104

4

4

4

44

44

0

4

0

4

40

04 0

4

0 4

21

4

00

00

40 4

0

24

302

223

00

0

33 3

0

State-of-the-practice (2017)

204

4

4

4

44

44

0

4

4

4

40

04 0

4

0 4

44

4

40

00

40 4

0

44

302

223

40

0

44 4

0

Barriers & challenges 2012

84% (16) Limited travel survey records

58% (11) Limited built environment data

58% (11) Limited modeling resources

47% (9) Limited decision-maker interest

2017 100% (25) Limited survey records of walking (& bicycling)

81% (21) Few resources for data collection

48% (12) Forecasting future scenarios

40% (10) Insufficient information on the built environment

16% (4) Not a priority among decision-makers

4% (1) Not a priority among the public

Other: Regional scale incompatible with short-distance walking

Pedestrian modeling applications

Project prioritization

Scenario planning

Corridor planning

Traffic safety analysis

Health impact assessment

Infrastructure gap analysis

Currently Future interest

61% (14) 78% (18)

43% (10) 65% (15)

43% (10) 61% (14)

35% (8) 57% (13)

35% (8) 57% (13)

30% (7) 57% (13)

Pedestrian modeling outputs Direct transportation outputs Walk trips generated

Walk trips with origins & destinations

Walk trips with “routes”

Distances walked Pedestrian miles traveled (PMT)

Minutes of walking Physical activity levels (METs)

Classified by… Geographic location

Personal characteristics (socio-demographics)

Questions? In your expert judgement or opinion… What uses motivate pedestrian modeling today?

What potential (existing? new?) uses motivate pedestrian modeling in the near/long-term future?

Patrick A. Singleton [email protected]

Singleton, P. A., Totten, J. C., Orrego-Onate, J. P., Schneider, R. J., & Clifton, K. J. (under review). Making strides: State-of-the-practice of pedestrian forecasting in regional travel models.

Data for Pedestrian Demand Modeling

Robert Schneider, PhDUniversity of Wisconsin-Milwaukee, Department of Urban Planning

Portland State, NITC Workshop – September 2017 1

Data for Pedestrian Demand Modeling

Robert Schneider, PhDUniversity of Wisconsin-Milwaukee, Department of Urban Planning

Portland State, NITC Workshop – September 2017 2

How many?

How safe?

Outline

1) Who is represented in our models?– How do we measure pedestrian activity?– Are there other (better) ways to measure

pedestrian activity?

2) What variables do we use to predict?– What explanatory variables are common?– Are there other (better) ways to represent

behavior? Different needs in the future?

3) How well do our models work?4) Are our models valued in practice? 3

Explanatory Variables

Validation

Dependent Variables

Usefulness

1) Dependent Variables

Pedestrian Data TypesSafety

(Crashes, injuries, behaviors)

UserCharacteristics

(Age, gender)

Infrastructure(Facility coverage

& quality)

PublicOpinion

(Satisfaction, desires)

Exposure/ Volume

(Count, mode share)

Counts Surveys

Household(Phone, mail,

internet)Intercept

Manual(Data collectors, imagery review)

Automated(Continuous

counters, imagery processing)

Pedestrian Data TypesSafety

(Crashes, injuries, behaviors)

UserCharacteristics

(Age, gender)

Infrastructure(Facility coverage

& quality)

PublicOpinion

(Satisfaction, desires)

Exposure/ Volume

(Count, mode share)

Counts SurveysDirect

Demand Models

Household(Phone, mail,

internet)Intercept

Manual(Data collectors, imagery review)

Automated(Continuous

counters, imagery processing)

FlowModels

Pedestrian Counts

Manual Counts

8Automated Counts

Pedestrian Counts

Google Earth—Tele Atlas 2008

Pedestrian Segment/Screenline Counts


Pedestrian Intersection Crossing Counts


Pedestrian Midblock Crossing Counts

Graphic source: Google Earth, 2008.

Counts Direct Demand Models

Travel Surveys

Example: Boulder, CO Region

Surveys Flow Models

Example: Boulder, CO Region

Surveys Flow Models

Source: Clifton, K.J., C.V. Burnier, R.J. Schneider, S. Huang, and M.W. Kang. “Pedestrian Demand Model for Evaluating Pedestrian Risk Exposure,” Prepared by the National Center for Smart Growth Research and Education, University of Maryland for the Maryland SHA, June 2008.

Tour

Graphic source: McGuckin, N. & Y. Nakamoto. Trips, Chains, and Tours—Using an Operational Definition, 2004.Available online: http://onlinepubs.trb.org/onlinepubs/archive/conferences/nhts/McGuckin.pdf

Trips


Non-Home-Based Trips


Stages


Challenge: Secondary walk trip stages

• The travel survey data used to develop flow models does not often capture secondary walk stages

• Even if data were available, could stages be integrated into the structure of flow models?

21

Pedestrian Mode Share on Trips Within 20 Shopping Districts

Intercept Survey NHTS

Overall 65% 71%

Urban core 96% 87%

Suburban Main St. 63% 64%Suburban Thoroughfare 30% 52%Suburban Shop. Center 40% 16%

Urban Core Shopping District (Market Street, San Francisco)Intercept Survey Respondent Pedestrian Path Density

Suburban Main Street Shopping District (Burlingame)Intercept Survey Respondent Pedestrian Path Density

Suburban Main Street Shopping District (Richmond)Intercept Survey Respondent Pedestrian Path Density

Are there other ways to measure pedestrian activity?

26


27Source: USGS, “Earth Explorer,” Available online, https://earthexplorer.usgs.gov/, 2017.

Aerial Imagery or Drone (HAV) Counts


28Source: USGS, “Earth Explorer,” Available online, https://earthexplorer.usgs.gov/, 2017.

Aerial Imagery or Drone (HAV) Counts


29Source: STRAVA Labs. “Strava, 2014 vs. 2015,” Available online, http://labs.strava.com/heatmap/2014-2015.html#6/-120.90000/38.36000/gray/bike, 2016.

GPS Traces: STRAVA running routes

Challenge: Temporal variation

Data Source: Milwaukee County Parks, 2014-2015.

0.0%

0.2%

0.4%

0.6%

0.8%

1.0%

1.2%

1.4%

1.6%

1.8%

2.0%0 6 12 18 24 30 36 42 48 54 60 66 72 78 84 90 96 102

108

114

120

126

132

138

144

150

156

162

Perc

ent o

f Wee

kly

Volu

me

Hour of the Week

Oak Leaf Trail Weekly Volume Pattern (11/4/14 to 11/3/15)

M Tu W Th F Sa Su

Wed., 4-6 pm3.13% of week

2) Explanatory Variables

What explanatory variables are common in our current models?

• Trip distance or time• Local environment factors

– Population and employment density– Proximity to commercial/retail– Proximity to transit– Pedestrian network connectivity

• Socioeconomic factors– HH automobile ownership– HH income

32

These are also mostly

convenience factors

Convenience

Are there other factors that might influence pedestrian behavior?

5) Habit(People who choose a particular mode

regularly are more likely to consider it as an option in the future)

2) Basic Safety & Security(People seek a mode that they perceive to provide a basic

level of safety from traffic collisions and security from crime )

3) Convenience & Cost (People seek a mode that will get them to an activity using an

acceptable amount of time, effort, and money)

4) Enjoyment(People seek a mode that provides personal (e.g., physical, mental, or emotional), social, or environmental benefits)

Pedestrian, Bicycle, Transit, or Automobile?

1) Awareness & Availability(People must be aware of the mode and have it available as

an option to travel to an activity)

Theory of Routine Mode Choice Decisions

Situational Tradeoffs

Soci

oeco

nom

ic F

acto

rs(E

xpla

in d

iffer

ence

s in

how

peo

ple

resp

ond

to e

ach

step

)Schneider, R.J. “Theory of Routine Mode Choice Decisions: An Operational Framework to Increase Sustainable Transportation,” Transport Policy, Volume 25, pp. 128-137, 2013

35Singleton, P.A. and K.J. Clifton. “The theory of travel decision-making: A conceptual framework of active travel behavior,” Presented at the Transportation Research Board Annual Meeting, Washington, DC, January 2015.


• Attractiveness of other modes– Transit service, auto parking pricing & supply, gas

prices, even bicycle infrastructure, AVs

• Safety and security– Pedestrian network attributes (pedestrian

facilities; how hard is it to cross the street?)– Perceived risk of crime

• Social norms; Personal preferences/enjoyment

36


Pedestrian Network Attribute Example

Source: Miranda-Moreno, L.F. and D. Fernandes. “Pedestrian Activity Modeling at Signalized Intersections: Land Use, Urban Form, Weather and Spatio-Temporal Patterns,” Transportation Research Record 2264, pp. 74-82, 2011.

Montreal Signalized Intersection Pedestrian Volume Model

Challenge: Measurement detail…how good is good enough?

Challenge: Geographic scalability

Source: Urbitran Associates. Pedestrian Flow Modeling for Prototypical Maryland Cities, Prepared for Maryland DOT, 2004.

3) Validation

How good (or bad) are our models?


Source: Schneider R.J., L.S. Arnold, and D.R. Ragland. “A Pilot Model for Estimating Pedestrian Intersection Crossing Volumes,” Transportation Research Record 2140, pp. 13-26, 2009.

Central Baltimore Estimated 24-Hour Pedestrian Crossing Volumes



MoPeD, Version 1: Central Baltimore Case Study


Source: National Cooperative Highway Research Program Report 770, “Estimating Bicycling and Walking for Planning and Project Development: A Guidebook.” Authors: Kuzmyak, J.R., J. Walters, M. Bradley, and K.M. Kockelman, Transportation Research Board, Washington, DC, 2014.

Santa Monica Pedestrian Intersection Volume Model

Alameda County Pedestrian Volume Model 2009 Observed Volumes vs. 2008 Pilot Model Predictions

-5,000

0

5,000

10,000

15,000

20,000

25,000

0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000

Wee

kly

Pede

stria

n Vo

lum

e Pr

edic

ted

by P

ilot M

odel

in 2

008

Weekly Pedestrian Volume "Observed" in 2009

Trendline forObserved vs.

Predicted Data

Line Representing Perfect Prediction (Observ. = Pred.)

Other Validation Approaches(validation from within same dataset)

Source: Hankey, S. and G. Lindsey. “Facility-demand models of peak-period pedestrian and bicycle traffic: A comparison of fully-specified and reduced-form models,” Transportation Research Record: Journal of the Transportation Research Board, 2016.

Minneapolis Pedestrian Volume Model Monte Carlo-based random hold-out analysis

Portland Metro Pedestrian Destination Choice Model percent correct & probability correct

Source: Clifton, K.J., P.A. Singleton, C.D. Muhs, and R.J. Schneider. Development of a Pedestrian Demand Estimation Tool, NITC Report, NITC-RR-677, 2015.

Variation in Pedestrian Volumes

• 5 Control Intersections

ID #

2008 Weekly Pedestrian Volume

based on Counts

2009 Weekly Pedestrian Volume

based on CountsAbsolute Difference

(2009 - 2008) Percent Difference1

50 315 310 -5 1.6%2650 15691 16113 422 2.7%9179 8342 7429 -913 12.3%9436 105297 88118 -17179 19.5%

499 5186 3448 -1738 50.4%1) Percent difference is calculated using the smaller number as the base value. If the model value is greater than the actual value, the percent difference is calculated as (2009 - 2008)/2008. If the actual value is greater than the model value, the percent difference is calculated as (2008 - 2009)/2009.

• Time of day, weather, etc. (accounted for)• Measurement error• “Unexplainable” variation

– Individual sickness, people walking for scenery, store sales, etc.– Not feasible to predict in a planning-level model– Require additional data and cost for small benefit

Variation in Pedestrian Volumes

Variation in “Typical” Alameda County Pedestrian Activity Pattern

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%12

AM

4 AM

8 AM

12 P

M4

PM8

PM12

AM

4 AM

8 AM

12 P

M4

PM8

PM12

AM

4 AM

8 AM

12 P

M4

PM8

PM12

AM

4 AM

8 AM

12 P

M4

PM8

PM12

AM

4 AM

8 AM

12 P

M4

PM8

PM12

AM

4 AM

8 AM

12 P

M4

PM8

PM12

AM

4 AM

8 AM

12 P

M4

PM8

PM

95%

Con

fiden

ce In

terv

al P

erce

nt A

bove

& B

elow

Mea

n Ho

urly

Val

ue

M T W Th F Sa SuM T W Th F Sa Su

95% of mid-day counts are between 20% above and 20% below the hourly mean

Challenges: External factors; Special circumstances

Other ideas for validation

• Transferability: Evaluate the performance of models in many different communities– Can they capture different norms, preferences, and

ranges of built and social environments?

• Compare the performance of several different types of models in the same study area

• Have practitioners and advocates carefully review predicted volumes against their local knowledge

4) Usefulness

Challenge: How can we do a better job demonstrating the value of our models?

• How many? – Is there still a need to make the case that there are

high levels of walking activity?• How safe?

– With the concept of Vision Zero, is there still as much of a need to identify high-risk locations?

• What value do our models add to the practice?• Do practitioners understand the value that our

models add?

Source: Hankey, S. and G. Lindsey. “Facility-demand models of peak-period pedestrian and bicycle traffic: A comparison of fully-specified and reduced-form models,” Transportation Research Record: Journal of the Transportation Research Board, 2016.

Minneapolis Intersection Pedestrian Volume Model

Pedestrian Index of the Environment

Source: Singleton, P.A., C. Muhs, R.J. Schneider, and K.J. Clifton. “Representing Pedestrian Activity in Travel Demand Models: A Framework and Proof-of-Concept,” Portland State University, Working Paper, 2015.

Absolute number of crashes suggests safety problem is in

downtown Oakland

Oakland Reported Intersection Pedestrian Crashes (1996-2005)

But the highest risk per pedestrian crossing is along

major arterial roads

Oakland Estimated Intersection Pedestrian Crash Risk (1996-2005)

Model Estimated Walk Commuting

Schneider, R.J., L. Hu, and J. Stefanich. “Development of a Neighborhood Commute Mode Share Model Using Nationally-Available Data,” Transportation, 2017

Change in Walk Commuting

Schneider, R.J., L. Hu, and J. Stefanich. “Development of a Neighborhood Commute Mode Share Model Using Nationally-Available Data,” Transportation, 2017

Mainline Roadway Intersecting Roadway City

Total population within 1/2-mile radius3

Total employment within 1/4-mile radius

Total number of commercial retail properties within 1/4-mile radius

Presence of regional transit station within 1/10 mile (Yes = 1, No = 0)

Estimated Pedestrian Crossings in a Typical Week5,6,7

Dr. Martin Luther King Drive Walnut Street Milwaukee 4924 2390 40 0 8830Dr. Martin Luther King Drive Walnut Street Milwaukee 9848 2390 40 0 13399Dr. Martin Luther King Drive Walnut Street Milwaukee 9848 4780 40 0 18633Dr. Martin Luther King Drive Walnut Street Milwaukee 9848 4780 80 0 22569Notes : 1. This i s a pi lot model based on a s tudy of weekly pedestra in volumes at 50 intersections in Alameda County, CA. The model has a good fi t for the Alameda County s tudy data

(adjusted-R2=0.897). Since the analys is was conducted on 50 intersections in Alameda County, CA, more research i s needed to refine the model equation and determine the appl icabi l i ty of the resul ts for other communities . The model equation i s : Es timated pedestrian intersection cross ings per week = 0.928 * Tota l population within 0.5-mi les of the intersection + 2.19 * Tota l employment within 0.25-mi les of the intersection + 98.4 * Number of commercia l reta i l properties within 0.25-mi les of the intersection + 54,600 * Number of regional trans i t s tations within 0.10-mi les of the intersection - 4910. Deta i l s of the s tudy are provided in two papers : 1) Schneider, R.J., L.S. Arnold, and D.R. Ragland. "Extrapolating Weekly Pedestrian Intersection Cross ing Volumes from 2-Hour Manual Counts ," Transportation Research Record, 2010, and 2) Schneider R.J., L.S. Arnold, and D.R. Ragland. “A Pi lot Model for Estimating Pedestrian Intersection Cross ing Volumes ,” Transportation Research Record, 2010.2. The pedestrian volume estimates produced by the model are intended for planning, priori ti zation, and safety analys is at the community, neighborhood, and corridor levels . Since the model provides rough estimates of pedestrian activi ty, actua l pedestrian counts should be used for s i te-level safety, des ign, and engineering analyses .3. The intersections selected for the s tudy did not include intersections in areas with very low population dens i ties (<50 people per square mi le). Therefore, the model i s not appropriate for intersections below this dens i ty threshold (i .e., the model does not apply i f there are fewer than 64 people within a 1/2-mi le radius ).4. The s tudy of Alameda County, CA found that land use characteris tics are the most important factors for predicting pedestrian activi ty. Roadway des ign factors , such as the presence of s idewalks , median cross ing i s lands , curb radi i , or pedestrian cross ing s ignals may have minor effects on pedestrian volumes , but they are not as s igni ficant for predicting pedestrian activi ty. However, roadway des ign factors are cri tica l for pedestrian safety and comfort. Roadways must be des igned to accommodate pedestrians of a l l abi l i ties , regardless of volume.5. The model output i s an estimate of the number of pedestrian cross ings during a typica l 168-hour week (with an average seasonal volume). Pedestrian cross ings are counted each time a pedestrian crosses any leg of the intersection (e.g., one person i s counted twice i f they cross the east leg and then the south leg of an intersection). Pedestrians do not need to cross completely ins ide the crosswalk; they are counted i f i f they cross within 50 feet of the intersection.6. The model may not perform wel l in locations close to specia l attractors , such as amusement parks , waterfronts , sports arenas , regional recreation areas , and major multi -use tra i l s . Pedestrian volumes in these areas tend to be highly variable, with high volumes during certa in seasons or during nice weather. Bridges and underpasses may a lso channel pedestrian activi ty, so more research may be necessary to adjust volume estimates near these features .

Model OutputIntersection Identification Model Inputs 4

Pedestrian Intersection Crossing Volume ModelPilot Model--January 20091,2

Developed by Robert Schneider, Lindsay Arnold, and David RaglandUniversity of California-Berkeley Traffic Safety Center

Example Pedestrian Volume Model Application:Dr. Martin Luther King Drive & Walnut Street

Concluding Thoughts

• Can we measure pedestrian activity more completely?

• Should we be including other explanatory variables?

• How accurate are our models? Do they scale geographically and transfer across communities?

• Are our models useful to practitioners?

Contact Information

Robert J. Schneider, PhDUniversity of Wisconsin-Milwaukee

Department of Urban [email protected]

Photo by Transportation Research Board

Jaime OrregoJoe BroachPortland State University

TCS 2017 WorkshopSeptember 12, 2017Portland, Oregon

What is the appropriate scale?

What is scale?

Every transportation analysis requires a unit area for the analysis as we need aggregation.

At what level do we sample?

GOAL: Balance resolution with practicality -- desire for resolution capturing individual-level behavior, but also interested low sampling error (people don’t behave the same every time) and data realities

To what level do we summarize?

GOAL: Policy makers need to assess their interventions at a larger scale that assure greater certainty of benefits and fit broader planning goals

Why should we re-consider scale when we model walking?Traditionally transportation planning has addressed problems relating to vehicle demand.

This type of travel is more consistent in terms of analysis and has been largely studied in the history of transportation engineering.

The frameworks are widely used and accepted. Can we just use the same? Why not? What scale do you work at? Traditional 4-step model (Rethinking How We Get

Around Sunnyvale Ria Hutabarat Lo)

https://www.slideshare.net/TransFormCA/rethinking-how-we-get-around-sunnyvale

https://www.slideshare.net/TransFormCA/rethinking-how-we-get-around-sunnyvale

Why it is important to get the scale right?

• Determines the sensitivity to (smaller) scale changes

• Determines limits of the output resolution

Parcel

Fixed Grid

TAZTraffic Analysis Zone

Census Block Group

Census tracts

Regions

Example: Strategic model – household travel survey

Sampling unit to represent population (Census block groups

or tracts)

Input for strategic model

Output in TAZ level

Example: Gerrymandering

What is the appropriate scale?

Behavioral perspective

Many times there are not environmental boundaries for the definition of zones. People don’t see those boundaries. Thus, we have to avoid edge effects.

Our behavioral understanding still is limited. What would you do with finer grained data?

Why kinds of attributes are relevant?

A

BC

D

E

Is the school’s impact on walking behavior the same for A and B? How about B & C? (School by PJ Souders)

How big is (probably) too big?

walk trip distances (mi) for ~7500 walk trips

TAZ

Tract

Block Group

0.6 mi

0.7 mi

1.2 mi

PAZ 0.05 mi

med

ian

zone

siz

e in

Por

tland

obse

rved

wal

k tri

p di

stan

ces

Measurement scale vs. Aggregation scale

Practical implementation for planning

• Scope of the projects

• Sensitivity policy variables that matter

• Compatibility capturing shifts between modes

• Computation challenges

• Forecast-ability (more to come…)

Example:Metro regional travel model (~2010)

Adapting TAZs

Compatibility

Shifted TAZ “centroids” for walking but lost compatibility with auto...

From all good intentions...walk distance was capped at 0.5 mi to improve predictions in larger TAZs. Walk trips out-competed bike and shifted out away from inner Portland!

Planning implementation - MoPeD

Planning implementation - clustering

Planning implementation - sliding scale

To handle intrazonal trips and edge effects and of larger zones…

1) Sample smaller scale locations

2) Aggregate up to compatible/practical scale

Discussion •

Reality transformed into links, nodes, and attributes

From above 600 walk trips, do pedestrians follow networks?

Three components of a travel network.

Block Face Walkway Roadway

…

What were some key issues, challenges, and solutions?

Compatibility or Resolution?

http://rbracket.github.io/GISJammerIdeas/papers_presentations/Walkway_Network_Analysis.pdf

Which side of the street is the middle GPS signal on?What to do when sidewalks come and go?

1 Broach, J., & Dill, J. (2016). Using predicted bicyclist and pedestrian route choice to enhance mode choice models. Transportation Research Record: Journal of the Transportation Research Board, (2564), 52–59.

Scan at sub-block scale. Aggregate to centerline links and intersection nodes. Use crossing logic instead of tracking crossings.

1 Broach, J., & Dill, J. (2016). Using predicted bicyclist and pedestrian route choice to enhance mode choice models. Transportation Research Record: Journal of the Transportation Research Board, (2564), 52–59.

Block Face Walkway Roadway

commercial land-use distance traffic volume

off-street path sidewalks marked crosswalk

park turns ped signal

residential upslope median refuge

downslope substandard street

enclosure (design)

pre-war construction

Attributes tested in pedestrian route and mode choice models. Significant factors (p<0.05) shown in white text, insignificant in gray.

Though project small if measured as zonal change, connectivity value large. Network-based model predicts walking on trip 3.6x more likely with overcrossing, all else equal. Even larger change for round trip.

Why should we care which routes pedestrians are most likely to use?

Many factors from theory (design, land-use) that are difficult to attach to empirical networks

“Micro”-design features may be important -- but are they worth measuring for large-scale models?

Likely a strong dynamic component to pedestrian travel

To date, little traction in regional modeling compared to bicycle network modeling

Network model fit: transit > auto > bike > walk -- we’re pushing it here and may need a hybrid network model

What’s missing?

Thank you!

Joe Broach <[email protected]> web.pdx.edu/~jbroach

mailto:[email protected]

mailto:[email protected]

HOW DO WE FORECAST INPUTS AND DEVELOP SCENARIOS?

Walk, don’t run? Advancing the state of the practice in pedestrian demand modeling

Transportation and Communities SummitSeptember 12, 2017

Background

Models have become much more disaggregate in the representation of travelers and their behaviors.

But, it is difficult to forecast model inputs at the SCALE needed:Land use – population and employmentNetworksWalkability indices/built environment

How can are various policies & improvements to the pedestrian environment represented in these forecast or scenario years. These are new considerations for most demand models. 2

Pedestrian analysis zones

3

264 feet = 80 m ≈ 1 minute walk

Metro: ~2,000 TAZs ~1.5 million PAZs

TAZs PAZs

Home-based work trip productions

Land use: Population and employment

• Most population, employment and other built environment conditions are forecast at the TAZ or aggregate level

• Potential approaches• Allocate TAZ (or other zonal structures) forecasts to pedestrian zones (top-

down)• Parcel (or grid cell-level) land use model that forecasts change incrementally

(bottom-up)• “Paint” scenarios at corridor or area-wide level

• Challenge for regional travel models, integrated models, ABM, as well as pedestrian modeling…

4

Pedestrian Investments & Walkability

• How can we represent spatially-explicit policies intended to improve the pedestrian environment in future scenarios?

• Land use features – density and mix – suffer the problems discussed earlier

• Sidewalks construction, crossing aids, connectivity, etc. pose different challenges.

• Many modeling efforts, including MoPeD, do not directly represent detailed features. Rather they proxy overall walkability using an index.

• How do we related future improvements & policies to these indices?

5

Walkability measures

• Pedestrian Environment Factor (PEF)1

• Walkability Index2

• Walk Opportunities Index3

• Walk Score®4

• Pedestrian Index of the Environment (PIE)5

6

1. Parsons Brinckerhoff Quade and Douglas, Cambridge Systematics, & Calthorpe Associates, 1993. LUTRAQ Volume 4A. http://www.friends.org/resources/reports

2. Frank, Schmid, Sallis, Chapman, & Saelens, 2005. https://doi.org/10.1016/j.amepre.2004.11.0013. Kuzmyak, Baber, & Savory, 2005. https://doi.org/10.3141/1977-19 4. https://www.walkscore.com5. Singleton, Schneider, Muhs, & Clifton, 2014. https://trid.trb.org/view.aspx?id=1289281

What is PIE?

7ULI = Urban Living Infrastructure: pedestrian-friendly shopping and service destinations used in daily life.

People & job density

Transit access

Block size

Sidewalk extent

Comfortable facilities

Urban living infrastructure

The Pedestrian Index of the Environment (PIE)= ∑ (6 dimensions)

weighted by association with walking

Pedestrian Index of the Environment

Portland Region

PAZ Scale

Scored from 20-100

Visualizing PIE

9

100 – Downtown core

80 – Major neighborhood centers

Downtown

Lloyd District

Visualizing PIE

10

60 – Residential inner-city neighborhoods

70 – Suburban downtowns

Laurelhurst

Gresham

Visualizing PIE

11

50 – Suburban shopping malls

40 – Suburban neighborhoods/subdivisions

Visualizing PIE

12

20 – Rural, undeveloped, forested

30 – Isolated business and light industry

N. Marine Drive

Relating to policies

13

People & job density

Block size

Sidewalk extent

Transit access

• Growth in population, housing production and employment

• Need to get at right scale

• New stops/routes• Frequency• Transit network connectivity• New pedestrian connectivity

• No change?• Permeability/Connectivity

Construction/investmentsConnectivity

Other explanatory variables

Increasing recognition that subjective attributes matter in behaviorSynthetic populations offer opportunities to attribute individualHow do we forecast:

– Preferences/Attitudes– Perceptions– Culture– Health outcomes

Weather/ClimateCrime

14

Questions

• Top down vs bottom up?• What does it take to move PIE 10 points? How do we unpack

these indices?• Is moving from a PIE score 90 to 100 the same input as moving

from PIE score 50-60?• Non-linearities? Step functions? Minimum/maximums?• What if there is no change in the attribute (e.g. block size)?• Can we answer the question: what is the impact of X

investment?15

How can model outputs link other tools?

“Walk, don’t run?” WorkshopTransportation and Communities Summit

Portland, OR — 12 September 2017

Why model pedestrian demand?

analyze health & safety impacts

utilize new data resources

mode shifts air quality & GhG

plan for pedestrian investments& non-motorized facilities

Pedestrian modeling outputs Direct transportation outputs Walk trips generated

Walk trips with origins & destinations

Walk trips with “routes”

Distances walked Pedestrian miles traveled (PMT)

Minutes of walking Physical activity levels (METs)

Classified by… Geographic location

Personal characteristics (socio-demographics)

Air quality & emissions models Estimates motor vehicle emissions using outputs like… Vehicle miles traveled

Vehicle speeds

Vehicle fleets

Indirect contributions of pedestrian-enhanced models… Better estimates of auto use

Mode choice model sensitivity to changes in pedestrian environment

Safety analysis Crash rate = Crashes ÷ exposure

Highway Safety Manual Safety performance functions (SPFs)

Estimate expected average crash frequency of a network, facility, or individual site

Crashes = f (exposure, facility characteristics)

Crash modification factors (CMFs) Calculates expected average crash frequency as a result of

geometric or operational modifications to a site that differs from set base conditions

Crashes with treatment = CMFs * Crashes without treatment

Problems for safety analysis Insufficient data for pedestrian safety analysis Crash data

Frequency, severity, injury patterns, contributing factors, types

Exposure data Volume, severity, event information

Traditional data collection problematic Ethical difficulties with experimentation

Pedestrian volume data rare, difficult to collect

http://apps.trb.org/cmsfeed/TRBNetProjectDisplay.asp?ProjectID=4203

http://apps.trb.org/cmsfeed/TRBNetProjectDisplay.asp?ProjectID=4203

Estimating exposure? Models could estimate # walk trips for… An area or neighborhood

A corridor

A facility segment

An intersection

Could be used as measure of exposure for… Pedestrian safety assessment

Crash rates

Safety performance functions

Proulx, Frank. (2016) Using Heterogeneous Demand Data Sources as Exposure in a Bicycle Risk Model (dissertation). Berkeley, CA: University of California, Berkeley.

Health impact assessment Estimating transportation’s health impacts Injury, morbidity (disease), & mortality (death)

Measured in consistent units Dollars (using the value of a statistical life (VSL)) Lives (disability-adjusted life years (DALY))

Two widely-used HIA models Health Economic Assessment Tool (HEAT)

World Health Organization / Europe http://www.heatwalkingcycling.org/index.php

Integrated Transport and Health Impact Modelling Tool (ITHIM) Centre for Diet and Activity Research / James Woodcock http://www.cedar.iph.cam.ac.uk/research/modelling/ithim/

http://www.heatwalkingcycling.org/index.php

http://www.cedar.iph.cam.ac.uk/research/modelling/ithim/

HEAT for walking/bicycling

Kahlmeier et al. (2014) HEAT methodology and user guide

HEAT travel data inputs # people walking (& cycling)

average time spent walking (or cycling) Duration (minutes/day)

Amount (steps/day) & [walking speed]

Distance (miles/day) & [walking speed]

Trips (trips/day) & [walk trip length] & [walking speed]

Notes about travel data inputs Adult population only (20–74 years old)

Average walking levels over year

ITHIM tool Evaluates regional/national… Scenarios

Comparisons

Interventions

Outputs Premature deaths

Disability adjusted life-years (DALYs)

Costs

Health impacts Physical activity

Traffic injuries

Air pollution

ITHIM tool

Maizlish (2016)

ITHIM calibration data

Maizlish (2016)

ITHIM travel data inputs Per capita mean daily travel distance (PMT, VMT) By mode: walk, bicycle, bus, rail, motorcycle, auto driver, auto

passenger, truck

By facility type (for VMT): local, arterial, highway

Per capita mean daily travel time By mode

Distribution of per capita mean daily active travel time (for walking & cycling) By gender (male, female) & age groups

Or ratio of mean vs 15–29 females & standard deviation

Mean walking speed, mean cycling speed, bus occupants

ITHIM travel data inputs Physical activity Distribution of active travel time by gender & age

Walking & cycling speeds

Traffic safety PMT & VMT by mode & facility type

Injuries & fatalities by striking & victim modes

Air quality Emissions by pollutant

Questions? How can travel demand models better link pedestrian outputs

to air quality, safety, health, and other analysis tools?

Does the precision (and the accuracy) of pedestrian outputs match what these tools assume or require?

What other post-model analysis could make use of pedestrian modeling outputs?

Patrick A. Singleton [email protected]