UrbanSim Model and Data Development

1
UrbanSim Model and Data Development

• John Britting
• Wasatch Front Regional Council

2
Background

• Local application under development for 7 years
• Lawsuit settlement agreement increased our focus
  with respect to application
• Need to determine whether suitable for use
  locally by January 2004 with peer review
• Learned a lot
• modeling system operational
• not suitable for use yet

3
What we want from UrbanSim

• Tool for scenario comparisons
• Interrelationships between infrastructure
  policies land-use
• Can (potentially) inform planning process w.r.t.
  secondary/cumulative impacts

4
Outcome of Peer Review

• Move ahead
• Should be useful in short-term
• Refine to meet WFRCs needs
• Reasonable results for significant policies
• (Eventually) Superior to current process
• Commitment to better planning

• However
• Addl tuning is required for immediate use
• Needs a more timely and improved dataset
• Difficult to interpret the impacts of less
  significant policies
• Not ready for corridor studies

5
WFRC Resolution on UrbanSim

• The Council finds that additional testing of
  UrbanSim is needed, (including) research into
  model refinement, data, policy implications,
  estimation of resources needed, and an outreach
  program to familiarize planning staffs in the
  region on the appropriate and useful applications
  of UrbanSim

6
Good News

• Reasonable response to large-scale policy changes
• The model is operational
• Initial outreach efforts were useful

7
Not Good News

• Little/no sensitivity to less significant
  policies
• Database needs to be improved
• Randomness
• 0 or 100 residential vacancy rates
• Price inflation
• Still discovering bugs

8
Technical Work Plan

• Data Development
• Refine representation of land policies and
  existing land-use
• Model Development
• Re-estimate and validate all models
• Improve logic
• Application Utilities
• Make implementation smoother and more effective
  (and working correctly)
• Overcome randomness
• Summarize and present key indicators quickly
  (utilize SQL and GIS effectively to save time)

9
Need for Data Development
Residential Capacity
Non-Residential Capacity
Dark blue is NYC dense
10
Key Models in UrbanSim

• Land Price Model
• Developer Models
• Residential Location Choice Model
• Employment Location Choice Models

11
Land Price Model Development

• Goals
• Reasonable relationships that will hold over time
• Thorough validation effort
• Account for variation in value by type of use
• Redevelopment analysis
• Appropriate sensitivity to transportation
  accessibility
• Transportation/Land-use interaction

12
Preliminary Results

• Distance from highway
• Residential/commercial (, then -) Industrial
  (-)
• Land price of neighborhood ()
• Access. to employment (Residential )
• Access. to population (Non-residential )
• Consistency between Zoning/Use ()
• Environmental factors (slope, open, roads, etc.)
• Residential (, then -) Non-residential (-)

13
Defining Accessibility

• Regional measure was initially used function of
  logsum and activity at destination
• Local measure was also used in location choice
  models (walking distance)
• We have three urban areas
• Regional, sub-regional and local accessibility
  likely to be important

14
Regional vs. Sub-regional Access

• Regional on left
• Both measures have merit
• Sub-regional shows logical patterns around 3 big
  urban cores

15
Land Price Model Validation

• Initial review suggests patterns are similar
• More to do to validate
• Categorical in application

16
Dealing with Inflation

• Average land price increases substantially over
  33 year simulation (4 per year)
• Accessibility/pop/emp are the culprits
• Average income stays in year 2000 dollars
• Options
• Use a different accessibility measure
• Inflate income or deflate price
• Use categorical price variables

17
Additional Model Development

• Need to use predicted land price in estimation
• Need to make implied behavior more consistent
  with theoretical understanding
• Existing models
• (HH/Dev) higher price is always more attractive
• (HH) closer proximity to highways is preferred
• Tricky to quantify residential character of
  neighborhoods
• A lot of multi-collinearity that clouds
  transparency
• Careful validation is necessary

18
Residential Location

• Segmented by income quartile
• Logical sensitivities to price by income segments
  (non-linear)
• Larger households prefer lower density smaller
  households prefer higher density (non-linear)
• All income groups tend to cluster
• Wrestling with accessibility measures

19
Model Development Challenges

• Not a lot of experience in practice with these
  models
• Many non-linear relationships (e.g. price,
  density, accessibility)
• 0-car vs. 1-car vs. 2-car accessibility
  parameters what is a reasonable relationship?
• Behavior vs. Patterns in data
• KISS

20
Sensitivity to Accessibility

• How much is appropriate?
• Use sensitivity testing to understand model
  response (Right direction? How much change is
  needed to get a response?)
• Quantify average contribution to choice
  probabilities price
• Utilize year-built data and TDM to do historical
  validation (quantify changes in accessibility and
  development from 1992-2002)

21
Randomness
22
Schedule/Milestones

• LRP is due in 28 months
• Expect the work-plan coming out of the peer
  review to be complete in 1 year
• Experimentation is on-going
• Land-use plan refinements will be completed
  within 2 months
• Land price and residential location models will
  be completed within 1 month