LTC, Jack R. Widmeyer Transportation Research Conference, 11/04/2011, Dohyung Kim

What if Crash Data Does Not Mean for Mapping: Lesson Learned from Crash

Mapping for Riverside County

Do Kim, Ph.D.Assistant ProfessorDepartment of Urban and Regional PlanningCalifornia State Polytechnic University - Pomona

Project Background

• Improvement of bicyclists and pedestrians safety in Riverside County– Finding physical environment factors to bicyclists

and pedestrian crashes– Funded by Leonard Transportation Center

Crash Data

• Crash data is important data for measuring safety on highways, but local governments does not often utilize this data.

• The main reason for the under-usage is the difficulty and inefficiency of the current crash mapping system.

Crash Data Flow

Crash Mapping• Converting test or tabular data to spatial data

that locates crashes on a roadway map

Riverside County Crash Data Analysis• Collected from California Statewide Integrated

Traffic Records System (SWITRS)• 5 year of pedestrian and bicycle crashes (2004

– 2008)• Total 4,769 crashes were reported during the

period (2,230 bicycle and 2,539 pedestrian crashes)

Automatic Mapping Using Geocoding

• ArcGIS Geocoding engine is the most well-known address matching function.

• However, it only matched 1,107 out of 4,769 (23%) crashes after intensive data cleaning and pre-processing.

Main Issue with Geocoding

• Geocoding engine identifies the locations of property addresses and intersections.

• However, the large portion of location information of crash data is certain distance and direction from intersections

W 500

E 300

S 1000

Matching with Customized Application• The application moves crash records from

intersections by given distance and direction.

Crash Record = 500 ft South from University Ave. & 1st St.

500 ft.

Unive

rsity

Ave

.

1 st St.

Results with Customized Application• Matched 2,094 records more (44%)

Manual Matching• Most time consuming and labor intensive works• Need to review the location information of each

individual record one by one using the customized application

• Systemic conflicts + Human errors

Systemic ConflictsHuman Errors

Manual Matching

State road name vs. Local name

Multiple Candidate

Total 1,568

(100%) 629

(40%)159

(10%) 780

(50%)

State Road Names vs. Local Names

• Police officers collect state road numbers, but the street names of roadway network are local names.

State Road vs. Local Name Resolution

• A street alias table can resolve this issue.• 629 records (13%) belong to this category.

Multiple Candidate Issue

• Multiple possibilities of a matching point• ArcGIS Geocoding use zip codes for zonal

details, crash records does not have the codes

Crash Record = ORANGE ST & 10TH ST

Multiple Candidate Resolution

• Screening with city boundaries• 159 (3%) crashes

ORANGE ST & 10TH ST at city of Riverside

Human Errors on Data Collection

• Incomplete information– University Ave & 1st (St) – (W) Palo BLVD & Main St

• Redundant Information– Chicago Ave & 1981 Chicago Ave

• Others– Misspelled street names– Using place names instead of street names (e.g.

Gateway Plaza)– And so on…

Human Error Resolution• Review each individual record one by one and

correct if mistakes are identified• 587 records (12%) matched

Unmatchable Crashes

• Irresolvable humane errors

Crash Record = CYPRESS AVE & PHILBIN AVE

CYPRESS AVE

PHILBIN AVE

Impacts of The Errors

Crash Record = GRAND AVE & 4TH ST

W. G

RAN

D AV

EE. G

RAND AVE

E. 4th STW. 4th ST

• Possibly change the crash hotspots by excluding crashes at particular locations from mapping

Incremental Resolutions

• Reduce human errors by educating police officers and data entry persons

• Construct better quality of roadway network data

• Develop street alias tables• Adopt crash mapping software

MN DOT Case

• Minnesota Crash Mapping Analysis Tool (MnCMAT)– Crash mapping and analysis software covering entire

state

FL DOT Case

• Web-based State Crash Record System– Police officers pinpoint

crash locations on a map that displays an aerial photograph of the area pulled up directly from the sever, much like systems such as Google Maps or Yahoo Maps.

X X