Reimplementing MetroScope: Portland’s Land Use Model Sonny Conder, Portland Metro Ben Stabler, PTV America TRB Planning Applications Conference May 7, 2007

© PTV 2007 Presentation What does Metro use MetroScope for? MetroScope in brief Background of MetroScope Present MetroScope operating requirements Testing MetroScope for use elsewhere MetroScope described in a bit more detail: Short course on model labels MetroScope structure MetroScope details – some examples MetroScope results – some more examples Re-implementation

© PTV 2007 What does Metro use MetroScope for? Transportation Planning > Corridor Studies > Regional Transportation Plan (RTP) > Transit Studies (e.g., Light Rail) Land Use Planning > Jobs and Housing Needs Analysis > Periodic Review > UGB analysis Testing Alternative Land Use & Transportation Policies

© PTV 2007 Metroscope Schematic Job Demand Forecast Travel Times/Access (Travel Demand Model) Metroscope Residential Model HIA Demand Forecast Land Supply / Capacity Data Vacant Land, Refill Supply, UR etc. Metroscope Non- Residential Model HIA location choices Job location choices HIA Location Choices Job Location Choices

© PTV 2007 Background of MetroScope Originally one zone residential model used for housing needs analysis 1996 – 1997: Developed entirely “in house” by Metro technical staff 1998: Multi-zones added 1999: Non-residential model added 2001 – 2002: Partial conversion from entirely spreadsheet to Visual Basic 2004: Complete Visual Basic 2005: Addition of internal travel demand – assignment model (PTV) 2006: Conversion to open source R code (PTV) 2006: Database and programming integration as one model (PTV)

© PTV 2007 Present MetroScope Operating Requirements Model operates in 5 year periods from year 2000 to A 35 year run requires 36 hours. Staff time required per run for model. > Model operation: Less than 1 staff hour for review, initiation and run time monitoring > Run setup: 2 – 6 staff hours for land use schedules by period – UGB expansions, Urban Renewal births/deaths. > Run output evaluation and reporting: 1 – 21 staff days depending on use, mapping and write up needs.

© PTV 2007 Testing MetroScope for use elsewhere Can MetroScope now be set up and be useful to a middle to large sized MPO? How much effort to set it up? How much effort to run and maintain it? Proposal for testing in other MPO’s (Lane County – Mid Willamette ??) MPO supplied – Land use by rzone by zoning category both vacant and refill; employment geocode by ezone by type class; travel network. Other data from published sources or Metro default. Set up accounting system for measured test of time and personnel required for implementing and using MetroScope in 3 – 6 test runs.

© PTV 2007 MetroScope in a bit more detail (Indulging my whimsical side a bit…) For the label thirsty: MetroScope - a 130 year old Walrasian/Paretian, demand and supply, aggregate, static, partial equilibrium model…Did I forget deterministic? What no micro-simulated, discrete, bid rent, neural network, dynamically disequilibrated randomly walking processes? NO…But believe me – humble is hard enough!

© PTV 2007 MetroScope Simulates the Development of Urban Real Estate – Meaning What? Incorporates a theoretical structure to determine the general character and number of equations and what arguments are determined within the model (endogenous) and outside the model (exogenous). Not nearly as rigorous as purely theoretical models so equations must have statistically measured weights (parameter estimation) attached to their arguments. Estimated equations need to be tested against range of variable arguments and compared to literature and common sense. (sensitivity testing). Full system of estimated equations and logical accounting identities need to be “conditioned” to the initial measurements (year 0) of the region. (calibration). Model needs to be repeatedly run in forecast mode with wide range of input assumptions. (destructive testing).

© PTV 2007 Structure Housing Theoretical – MetroScope Gen 3.0 Subject to: Profit (supply) function Subject to: Determined quantities:

© PTV 2007 MetroScope Housing Consumption Function – More than you ever want to know 4 Types of Housing – OSFD,OMFD,RSFD,RMFD in 8 value categories, 425 Rzones, 400 HIAK’s distributed from 72 Ezones. House tenure = f(HIA, Price of good, price of substitute,price of complement, frequency) House type = f(HIA, price of good, price of substitute, price of complement, frequency) House location = f(HIA, travel time, neighborhood, frequency of type and value category, price of good) Choice problem: allocate 1,335,000 households among 391,680,000 possible states.

© PTV 2007 MetroScope Housing Supply Function – shorter because we don’t know much 4 Types of Housing in 425 Rzones in up to 30 zoning density categories by vacant, refill, urban renewal, ugb adds and rural residential land sources. Housing Cost = f(lot size,house size, cap cost per sq. ft. by density, type, location, land cost per sq. ft by type, location, density, SDC’s) Housing production function: CES

© PTV 2007 MetroScope Equilibrium Solution Residential, nonresidential and transportation modules solved for equilibrium condition once every 5 years. Residential solution find Price for 425 zones and 4 housing types that min (S- D)SQR (75 iterations). Nonresidential solution find Price for 72 zones in 6 real estate types that min (S-D)SQR (75 iterations). Travel model change in O-D travel time less than a specified constant.

© PTV 2007 Portland Metro Agency Development Path

© PTV 2007 Forecast Allocation Area Clackamas Co. Clark Co. Washington Co. Multnomah Co. Tailored to allocate growth to UGB, but also anticipates external “leakage” 6.5 counties 2 m pop 425 rzones 72 ezones 2029 taz’s

© PTV 2007 Interface

© PTV 2007 Interface

© PTV 2007 Condo and Single Family Production

© PTV 2007 Annual Land Consumption

© PTV 2007 Land Consumed By Source

© PTV 2007 Residential Development Density by Land Source

© PTV 2007

Re-Implementation The old implementation: > Some calculations in Excel > Res and Non-Res Models in VB6 > Transportation Model in R & EMME/2 > Parcel accounting in Access > Data stored in CSV files > Data transfer via files and The new implementation: > Database > One programming language > Completely automated model runs > Scenario management

© PTV 2007 Tools Metro required open source tools: Database – PostgreSQL 1) Mature open source DB 2) ODBC driver 3) PostGIS Programming Language – R 1) Specifically for data analysis 2) Dialect of S (written in C/C++) 3) Vectorized 4) Statistical 5) Interpreted 6) COM Interface (rcom)

© PTV 2007 Tools Travel Demand Model: Trip Generation, Trip Distribution and Mode Choice in R Assignment and Visualization in PTV Vision - VISUM Communication via the COM interface Matrix I/O via R library visumR - reads/writes compressed binary format

© PTV 2007 Implementation Inputs in CSV format (except networks) DB schema in standard SQL for semi DB independence (no complex data types) R controls the entire process (and is the core interface) DB on Metro’s Postgres server, while R and VISUM run on a Windows desktop Access hook via ODBC to Postgres DB for custom outputs GUI for managing scenarios, inputs, outputs and running the model

© PTV 2007 Implementation Design goal of one day turnaround for model runs Previously 2 week turnaround 50 iterations to balance supply and demand Redesign: 1) ~ 6 hours for 1 run of one time step 2) ~ 48 hours for a 40 year run The RES location choice model consumes ~ 50% of the run time nonResModel(50) resModel(50) nonResModel(50) resModel(50) transModel(3) { x2 Time step n x8 for 40 years

© PTV 2007 Reimplementation Conclusions Integration of all modules enables quick turnaround on model runs The benefits of automated accounting are very important (central open DB that stores everything) Knowledge transfer of database design, programming, and model framework/structure Open source so other agencies can utilize the model

© PTV 2007 Questions Sonny Conder, Metro, Ben Stabler, PTV America,