A Comparison of Geocoding Methodologies for Transportation Planning Applications

1
A Comparison of Geocoding Methodologies for
Transportation Planning Applications

• Jennifer Indech Nelson
• Dr. Randall Guensler
• Dr. Hainan Li
• Georgia Institute of Technology

May 9th, 2007
2
Agenda

• Purpose
• Background
• Process
• Acquisition of data
• QAQC
• Final data set
• Analysis
• Positional Accuracy
• Polygon Assignment
• Discussion

• Assess the accuracy of various geocoding methods
  to provide insight on field data collection,
  calibration of travel demand model inputs, and
  automation of travel behavior analysis

3
Geocoding and How It Is Used in Transportation
Planning

• Geocoding - Generation of coordinates within a
  spatial geographic framework, where single points
  serve as proxies for places
• Used to
• Prepare TAZ data from travel diary studies for
  Travel Demand Model development
• Better represent spatial travel patterns
• Verify 4-step model components
• Provide primary input to next generation
  behavior-based micro-simulation Travel Demand
  Models

4
Methods of Obtaining Geocoded Coordinate Data

• GPS field surveys (active)
• Aerial image processing
• Address matching
• Road network address interpolation
• GPS tracking (passive)

Increased automation
5
Geocoding Address Matching Vs. Interpolation
Linear Address interpolation
11 - Check address existence / integrity from
list inc. other attributes
Estimate position from spatial reference (network
link)
Assign coordinates
Address Interpolation
Address Matching
6
GPS and GIS Data Acquisition in Transportation
Commute Atlanta

• Commute Atlanta study
• GPS-instrumented vehicle tracking
• 3 years, second-by-second
• 487 vehicles, 268 households
• 1.8 million trips

7
Data for Comparative Analysis

• Two days of parallel data in March 2004 from 137
  HHs
• Travel diary self-reported locations
• GPS recorded trip files
• Parcel-level geographic reference
• GIS shapefiles generated by MPO and individual
  counties (Fulton and Gwinnett Counties)

8
Example of GPS Trip Ends
All GPS Trip-Ends in 13-County Region during
travel diary survey period
9
Final Data Format

• Each location record has three associated
  coordinates
• GPS trip-end point
• Parcel centroid
• Interpolated location (street network)
• Characteristics
• Unique ID
• Area
• Land use
• TAZ

Centroid
Geocode w/ offset
40
GPS
10
Data Quality Issues GPS/Diaries

• Travel diaries versus GPS trip-ends
• Under-reporting of visited locations in travel
  diaries
• GPS wander
• Dependent on weather, satellite, and hardware
  conditions
• Primarily occurs at lt 5 mph
• Data point is last GPS coordinate at engine-off

11
Data Quality Issues Reference

• GIS parcel boundaries and centroids
• Not all parcels have existent or correct address
  data
• Topology errors may lead to inaccurate centroid
  calculation
• Road network geocoding
• Uses national database generated by NavTeq and
  TeleAtlas, may not have current/correct address
  ranges

12
The Incredible Shrinking Data Set
Metro Atlanta (13 counties )
Two-county subset

• Fulton 195 locations, 119 unique
• Gwinnett 129 locations, 75 unique

13
Analysis Positional Accuracy

• Complete (3-source) data only 324 points (194
  unique)
• 195 Fulton, 129 Gwinnett
• Compare
• GPS trip-end data with parcel centroids
• Interpolated addresses with parcel centroids
• GPS trip-end data with interpolated addresses
• Further comparison according to
• Land use
• Parcel size (e.g. lt 5 acres, gt 5 acres)

14
Positional Accuracy GPS vs Geocode

• GPS significantly more accurate than geocoding
• Combined 273 vs 402
• (Single-family) residential locations more
  accurate than non-residential parcels
• Smaller parcels more likely than larger parcels
  to have better positional accuracy for all
  methods

15
Positional Accuracy Land Use / Size

• GPS to centroid accuracy has some correlation to
  parcel size, but land use and typical parking
  location are probably more important
• Within particular land uses, inverse relationship
  of accuracy to area

16
Results Polygon AssignmentParcel and Blockgroup

• Match rates to potential TDM inputs
• Parcels, Census Blockgroup

17
Results Polygon AssignmentLand Use and TAZ

• Match rates to potential TDM inputs
• Land Use, Traffic analysis zone (TAZ)

18
Polygon Assigment Rate TAZ

• Non-residential locations especially prone to
  mis-assignment

19
Discussion

• Reference Data
• Must be accurate and standardized!
• Positional Accuracy
• Method of creating geocoded data depends on
  degree of accuracy needed
• Most to least accurate (lt10 ft to gt1000 away)
  Address matching, GPS, interpolation
• Off-site parking creates issues for passive
  determination of trip purpose from GPS data

20
Discussion

• Polygon Assignment
• TAZ hit rate lower than expected, particularly
  for non-residential locations
• Degree of zoning homogeneity and size of parcels
  are directly proportional to chance of matching
  correct land use for TDM verification

21
Next Steps

• Assess method of GPS tracking and data gathering
• Quantify error associated trip-ends
• Determine how to evaluate large parcels /
  campuses
• Internal destinations, land uses

22
Any Questions?
Please use the Microphone
22
23
Appendix Sources and Additional Figures

• All figures created by Commute Atlanta
  researchers, except spatial interpolation
  picture (slide 5 from Three Standard Geocoding
  Methods Dramowicz, 2004) and Google Earth
  imagery (slides10 and 21)

• Right GPS position off due to urban canyon (tall
  buildings in Midtown Atlanta)