UrbanSim at HoustonGalveston Area Council

1
UrbanSim at Houston-Galveston Area Council
Dmitry MessenForecasting Program,
H-GACwww.h-gac.com
2
Modeling at H-GAC

• Socioeconomic Modeling
• Forecasting Group
• UrbanSim
• Transportation Modeling
• TDM Group
• EMME/2 ? Cube Voyager (summer 2006)
• UrbanSim
• 2000-2003 Richard DeBose
• 1st Forecast adopted May 2003
• 2nd Forecast 11/12 2005

3
Principal Features of UrbanSim

• Allocation
• County-level control totals (target levels)
• Spatial Disaggregation
• User-defined regular grid
• Agent-based
• Households
• Workers (employers)
• Real estate developers

4
Base Year

• Spatial System (Regular Grid)
• H-GAC 1,000x1,000 ft 250,000 grid cells
• Grid Attributes
• Housing units
• Value of residential improvements
• Com-ind-gov sq. footage
• Value of com-ind-gov improvement
• Proximity (highways, arterials)
• Physical features ( water, open space, etc.)
• Other (GRID_ID, land use (development) type)
• Cost of Travel data
• Inter-TAZ logsums

5
Base Year (cont.)

• Households
• Table with individual households (household_id)
• Attributes GRID_ID, size ( of people), race,
  age (of hsh head), income, of cars, of
  workers
• Jobs
• Table with individual jobs (job_id)
• Attributes GRID_ID, user-defined sectors
• H-GAC Durables, Non-Durables, Mining, CTPU,
  FIRE-Services, Retail-Wholesale, Gov-Med-Ed

6
Data Sources

• County Appraisal (housing units, non-residential
  sq footage, value of improvements)
• Info-USA (company-level data on location, sector,
  of employees)
• Census (household attributes)
• Household synthesis use marginal probabilities
  from PUMS to disaggregate SF-3 data
• Transportation (logsums, networks)
• Other (Landsat imagery, aerial photo, proprietary
  real estate databases)

7
Control Totals

• UrbanSim only allocates future growth
• Regional (all grid cells combined) levels are
  determined exogenously
• Annual control totals for
• Households (optional segmentation)
• Jobs (optionally,by sector)
• A task in itself
• Buy or develop your own
• H-GAC REMI in 2003, in-house model in 2005

8
UrbanSim Sub-Models (Modules)

• Accessibility
• Household transition
• Employment transition
• Household location choice
• Employment location choice
• Land price
• Developer

9
Annual Sequence of Events

• Accessibilities are announced (computed)
• Households are created/removed
• Jobs are created/removed
• New households select from the stock of available
  housing units
• New jobs select from the stock of available job
  spaces (1 job space X sq ft)
• Market announces new land values (computed)
• Developers build new housing units and
  non-residential sq. footage

10
Accessibility Module

• Uses logsums to compute accessibility to housing,
  employment, different land uses, recent
  developments
• Conveys information about the
• Spatial distribution of housing and jobs
• Disutility (cost) of travel
• Transportation network
• Is computed for every grid cell, for every
  activity
• Exponential distance (cost) decay function

11
Household Transition Module

• Simulates aggregate results of family dynamics
• People are born, marry, divorce, die
• No aging!
• Updates (removes/creates) current households to
  match the Target (Control Total)
• Creates vacant units (when a household is
  removed)
• Puts unplaced (newly created) households
  (GRID_ID0) in a buffer

12
Employment Transition Module

• Simulates aggregate results of employment
  dynamics
• New jobs created, old jobs terminated
• Updates (removes/creates) current jobs to
  matches the Target (Control Total)
• Creates vacant slots (when a job is removed)
• Puts unplaced (newly created) jobs (GRID_ID0)
  in a buffer

13
Relocation Modules

• Household Relocation Module
• Employment Relocation Module
• User-specified of households, jobs wishing to
  change location
• Puts households, jobs in a buffer and releases
  houses units, job slots
• H-GAC not used

14
Household Location Choice Module

• Simulates household decisions on where to reside
• Applies to unplaced households
• Selection from a stock of available housing
• Procedure
• Group unplaced households by attributes
• Sample available locations
• Assign probabilities (from multinomial logit
  model)
• Cost/income ratio, access to employment, housing
  within walking distance
• Use Monte Carlo sampling to pick location and
  assign GRID_ID

15
Employment Location Choice Module

• Simulates employer decisions on where to locate a
  job
• Applies to unplaced jobs
• Selection from a stock of available job slots
• Procedure
• Group unplaced jobs by sector
• Sample available locations
• Assign probabilities (from multinomial logit
  model)
• Land use types, proximity, access to emp, pop
• Use Monte Carlo sampling to pick location and
  assign GRID_ID

16
Land Price Module

• Simulates the determination of land values by the
  market
• Parameters from Hedonic regression
• Site and location attributes

17
Developer Module

• Simulates actions of the real estate developers
• Creates new residential units and non-residential
  sq. footage
• Most important module
• Households/employment locate only in available
  housing/space.
• Development Types (for grid cells)
• min, max (units, sq footage)
• 8 residential, 8 mixed, 3 commercial, 3
  industrial, 1 gov, 1 vacant, 1 undevelopable

18
(No Transcript)
19
Developer Module (cont.)

• Transitions define Real Estate Development
• Transitions
• Within the same dev type
• From one type to another
• Only certain transitions are allowed
  (user-defined)

20
Developer Module (cont.)

• Transition Probability Estimation
• Multinomial logit
• Choice set includes no-build alternatives
• P is a function of grid cells and local
  attributes
• Recent transitions, land value, open space
• Sampling setup
• Artificial observation with 5 alternatives (1
  transition)
• Different grid cells
• Calibration (frequency of transitions)

21
Developer Module (cont.)

• Simulation
• Apply Constrains (exclude certain grid cells)
• Determine Build or NO build
• If Build, then determine quantities
• Distribution for new construction (min, max,
  mean, std)
• Vacancy Adjustment
• Speed up development if vacancy is low, slow down
  if vacancy is high

22
Events

• Deterministic component
• Under construction or planned (known or likely to
  occur)
• Grid-specific
• Development Events
• Employment Events
• Land use Events

23
Running UrbanSim

• UrbanSim Scenario (instructions) file
• Specify the forecast horizon
• Specify years to output
• Reports future grid cell conditions
• Aggregate from grid cells to TAZ
• Households (by attributes)
• Jobs (by sector)
• Give results to transportation modeling group

24
Running UrbanSim (cont.)

• Base year - 2005
• Run 2005-2007 simulation
• Save 2007 conditions (grid cells, jobs,
  households)
• Make 2007 a base year
• Update
• Logsums, Proximities, Travel time
• Base year - 2007
• Run 2007-2010 simulation
• Continue

25
Running UrbanSim (cont.)

• Future travel networks and accessibilities
• Phase 1 2005-2007 (2005 travel data)
• Phase 2 2007-2010 (2007 travel data)
• Phase 3 2010-2013 (2010 travel data)
• Phase 4 2013-2015 (2013 travel data)
• Phase 5 2015-2025 (2015 travel data)
• Phase 6 2025-2050 (2025 travel data)
• 8 separate models (each county)

26
UrbanSim Work Flow Process

• Define basic framework (grid, development types)
  and assemble data (base year and control totals)
• Run the consistency checker utility
• Run the estimation data writer utility
• Prepares data for statistical modeling
• Develop specification and estimate parameters
  (SAS)
• Art/science, no general solution
• Validate parameters
• Recreate past conditions (e.g., 1990), simulate
  development, compare with actual (e.g., 2000)
• Populate the UrbanSim tables (model specification
  and coefficients)
• Run the simulation
• Tabulate and map results

27
Technical Details

• Constantly evolving software
• Stable and nightly releases
• Open source
• Written in Java, will be re-written in Python
• Out next year (?) OPUS
• MySQL server, MySQL table format
• About 60 tables (some are optional) per region
  worth 500 MB (can be smaller)
• Batch files (calls to DOS, MySQL, UrbanSim)

28
H-GAC Plans for 2005-07

• Parcel-level regional database
• Fill the gaps for exempt parcels
• Merge 8 counties into a single model
• Revise development types
• Address sq ft per job and sector-sq ft type
  problems
• Include home-based employment
• Automate grid cells table creation
• Data? Code? UrbanSim
• Re-do statistical modeling
• Thorough validation

29
Wish List

• Link Jobs and Households
• create WORKERS table with job_id and household_id
• Differentiation between renting and owning
• Differentiation between single- and multi-family
  housing units
• Separate Redevelopment Module
• Detailed non-residential space categorization
  (office building, large retail, school, etc)
• Group quarters

30
(No Transcript)
31
Lessons Learned

• Incredible Fun
• If you like GIS stats DBMS complex systems
• Major commitment
• Invest in a parcel-level database
• Dont make it grid-centric
• Design the information system and data collection
  procedures with updates in mind
• Front-end QA/QC
• Documentation!
• Automation
• Recurring tasks