UrbanSim

Modeling Urban Development

for Land Use, Transportation,
and Environmental Planning
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

Paul Waddell

Metropolitan areas have come under

intense pressure to respond to federal
mandates to link planning of land use,
transportation, and environmental
T he relationships among land use, transportation, and the environ-
ment are at the heart of growth management. The emerging concern
that construction of new suburban highways induces additional
travel, vehicle emissions, and land development, making it implausible to
“build our way out” of congestion, has reshaped the policy context for met-
quality; and from citizen concerns
about managing the side effects of ropolitan transportation planning (Downs, 1992). Recognizing the effects of
growth such as sprawl, congestion, transportation on land use and the environment, the Clean Air Act Amend-
housing affordability, and loss of open ments of 1990 and the 1991 Intermodal Surface Transportation Efficiency
space. The planning models used by Act (ISTEA; 1991) mandated that metropolitan planning organizations
metropolitan planning organizations (MPOs) integrate metropolitan land use and transportation planning. These
were generally not designed to address legislative actions have led to subsequent legal challenges to the traditional
these issues, creating a gap in the abil- approach to transportation planning that ignores these feedback effects
ity of planners to systematically assess (Garret & Wachs, 1996). The passage of the Transportation Equity Act for
them. UrbanSim is a new model sys- the 21st Century (TEA-21; 1998) in 1998, as the successor to ISTEA, soft-
tem that was developed to respond to ened these planning requirements somewhat, but significant pressure re-
these emerging requirements and has
mains to better coordinate metropolitan planning of land use, transporta-
now been applied in three metropoli-
tan areas. This article describes the tion, and the environment.
model system and its application to Requirements for improved and better integrated land use and trans-
Eugene-Springfield, Oregon. portation models have emerged, not only in response to this federal legisla-
tion but also from state growth management programs that promote closer
Waddell is an associate professor in the linkage of land use and transportation planning. Pressure for change has
Department of Urban Design and Planning also come from the community of practicing and academic planners and
and the Daniel J. Evans School of Public Af-
from advocates for the environment and alternative modes of transporta-
fairs at the University of Washington. He is
also director of the Interdisciplinary PhD tion who have become frustrated with the state of the practice, as exemplified
Program in Urban Design and Planning. in the Portland (OR) LUTRAQ project (Blizzard, 1996). In response to the
His current research focuses on land use growing concern regarding the limitations of current land use and trans-
and transportation planning, metropolitan portation models, the Travel Model Improvement Project (TMIP) was
development, and urban simulation.
formed as a collaborative effort by the Federal Highway Administration, the
Journal of the American Planning Association, Federal Transit Administration, and the Environmental Protection Agency.
Vol. 68, No. 3, Summer 2002. © American In 1995, TMIP hosted an international conference on land use modeling
Planning Association, Chicago, IL. to convene practitioners, researchers, and consultants to assess the state of

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 297

 PAUL WADDELL

the practice and to make recommendations for new banSim has since been applied in Honolulu and Salt
model development that address limitations in the cur- Lake City, and other metropolitan areas are beginning
rent practice. Recommendations put forward at this con- to apply it as well. The UrbanSim software is distributed
ference included the following: as Open Source software under the GNU General Public
License (Free Software Foundation, 1991), which allows
• moving fairly quickly toward random utility-based
anyone to use, modify, and redistribute the source code
models;
at no cost. It can be downloaded from the Internet
• using a clear behavioral basis describing the
(UrbanSim Project, 2002).
principal actors and choices involved in urban
The objectives of this article are twofold. The first is
development and transportation;
to describe the UrbanSim model design at a level that
• placing greater emphasis on the use of models for
facilitates an assessment of how it addresses emerging
policy analysis, planning, and sensitivity testing;
requirements for land use and transportation modeling,
• recognizing the varying temporal and geographic
and of how it compares to other existing modeling ap-
scales relevant to different processes in urban
proaches. The second is to describe the application of
development;
the model to the Eugene-Springfield metropolitan area
• moving to disaggregate models and data;
and assess its validity over an historical period. These ob-
• drawing on multiple disciplines;
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

jectives are addressed in turn in the following sections.

• developing modular models;
The article concludes with a description of further re-
• increasing the use of GIS and remote sensing; and
search and development priorities.
• testing the effects of transportation on land use
(Weatherby, 1995).
Federal efforts to improve the state of the practice The Design of UrbanSim
through the TMIP program have focused almost exclu-
sively on long-term investment in a research and devel- Overview and Comparison to Other
opment effort for a new traffic microsimulation model, Operational Models
called TRANSIMS, and have not yet made any invest- The design of UrbanSim differs significantly from
ment in new land use modeling approaches. As a result, several existing operational modeling approaches, in-
the initiative for developing new land use models has cluding the spatial-interaction DRAM/EMPAL models
been taken up at the state and local levels. Leading the developed by Putman (1983); the spatial input-output
way are efforts such as Oregon’s Transportation and TRANUS and MEPLAN models, developed respectively
Land Use Model Integration Project (TLUMIP), Utah’s by de la Barra (1989) and Echenique et al. (1990); the
Quality Growth Enhancement Tools (QGET) and Envi- GIS-based California Urban Futures (CUF, CUF-2)
sion Utah, and the Oahu MPO’s investments in new land Model (Landis, 1994, 1995; Landis & Zhang, 1998a,
use and transportation models. 1998b); the MUSSA model developed by Martínez
Launched in 1996 by the Oregon Department of (1992); and the CATLAS (and later METROSIM and
Transportation, TLUMIP was an ambitious effort to de- NYMTC-LUM) model developed by Anas (1982). These
velop new integrated models to evaluate the interactions models are discussed in detail in several recent reviews
between transportation and land use. The project had (Dowling et al., 2000; Miller et al., 1998; Parsons Brinck-
two components. The first was the statewide implemen- erhoff, 1998; U.S. EPA, 2000), which update a number
tation of an integrated land use and transportation of earlier reviews (Anas, 1987; Harris, 1985; Kain, 1985;
model, called TRANUS™ (de la Barra, 1989). The second Paulley & Webster, 1991; Southworth, 1995; Wegener,
was the development of UrbanSim, a new metropolitan- 1994, 1995). The pitfalls of large-scale urban models
scale land use model for integration with transportation were convincingly articulated almost 3 decades ago (Lee,
models. UrbanSim was designed specifically to address 1973, 1994) and remain significant concerns. The design
the policy analysis requirements of metropolitan growth of UrbanSim has been well informed by these criticisms
management, with particular emphasis on interactions of prior modeling efforts, as well as by advances in the-
between land use and transportation. ory, computation, and econometric methods.
The Oregon TLUMIP effort extended the original To clarify differences between UrbanSim and other
UrbanSim design developed for Honolulu, Hawaii, and operational urban models that have been reviewed else-
implemented a prototype version in the Eugene-Spring- where (Dowling et al., 2000; Miller et al., 1998; U.S. EPA,
field metropolitan area. Testing of the current version of 2000), Table 1 compares the key features of four model
the model in this area, using data from 1980 to 1994, has approaches. A brief description of some terms used in
provided a useful empirical validation of the model. Ur- Table 1 follows.

298 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3

 URBANSIM

TABLE 1. Comparison of operational model characteristics.

Operational urban model

Characteristic DRAM/EMPAL MEPLAN and TRANUS CUF-2 UrbanSim

Model structure Spatial interaction Spatial input-output Discrete choice Discrete choice
Household location choice Modeled Modeled Not modeled Modeled
Household classification Aggregate, Aggregate, user-defined Not Disaggregate, income,
8 categories represented persons, workers, child
Employment location choice Modeled Modeled Not modeled Modeled
Employment classification Aggregate, Aggregate, user-defined Not modeled Disaggregate,
8 categories 10–20 sectors
Real estate development Not modeled Modeled Modeled Modeled
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

Real estate classification 4 land uses Aggregate, user-defined 7 land uses 24 Development types
Real estate measures Acres Acres, units floor space Acres Acres, units floor space
Real estate prices Not modeled Modeled Not modeled Modeled
Geographic basis Census tracts or User-defined zones Grid cells Grid cells
aggregates (2–300)
Temporal basis Quasi-dynamic, Cross-sectional, Annual, Annual, dynamic,
equilibrium (5–10 equilibrium dynamic disequilibrium
year steps)
Interaction with travel models Yes Yes No Yes
Modular model structure Partial No No Yes
Software access Proprietary Proprietary NA Open source

For model structure, spatial interaction refers to mod- For temporal basis, cross-sectional refers to the use of
els that draw on the analogy of the physical relationship one point in time for estimating a model, in contrast to
of gravity. The application to human geography, made a dynamic approach that analyzes changes over multiple
popular by Wilson (1967), recognized the empirical pat- time periods. In the context of economic models, equi-
tern that trips between two locations increase as the ac- librium describes a hypothetical long-term market
tivity (population and employment) in the origin and steady-state condition in which supply, demand, and
destination zones increases and the travel cost decreases. prices are perfectly balanced so that no one can become
A large class of location choice models draws on and ex- better off without someone else becoming worse off. It
tends this metaphor (Fotheringham & O’Kelly, 1989). assumes that all buyers and sellers are operating within
Discrete choice refers to models that draw on discrete competitive markets and have full information about
choice theory and the development of a class of econo- the current and future prices and benefits of consump-
metric models known as random utility maximization tion choices. This condition contrasts with disequilibrium
(RUM). Daniel McFadden (1973, 2000) recently won the where supply and demand are not perfectly balanced, re-
Nobel Prize in economics for his pioneering work in this flecting various limitations in the responsiveness of eco-
area. The approach is suited to modeling choices be- nomic agents. For example, short-term growth in de-
tween alternatives that are mutually exclusive. Many of mand for housing may outpace or lag behind the
the early applications of this class of techniques, includ- growth in housing supply, causing disequilibrium con-
ing multinomial and nested logit models, were focused ditions we commonly refer to as the boom-and-bust
on the transportation mode choice problem. cycle.

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 299

 PAUL WADDELL

In short, the UrbanSim design departs from aggre- tion of its application in the Eugene-Springfield metro-
gate economic and spatial-interaction models that rely politan area.
on cross-sectional equilibrium solutions using large ge-
ogaphic districts, and pursues an approach that is dis- The Database
aggregate and based on predicting changes over small The data integration process for UrbanSim is de-
time steps, as does the CUF-2 model (Landis & Zhang, picted in Figure 1. The input data used to construct the
1998a, 1998b). Unlike the CUF-2 approach, however, the model database, called the data store, include parcel files
UrbanSim design explicitly represents the demand for from tax assessor offices; business establishment files
real estate at each location and the actors and choice from the state unemployment insurance database or
processes that influence patterns of urban development from commercial sources; census data; GIS overlays rep-
and real estate prices. This design approach synthesizes resenting environmental, political, and planning bound-
and extends some of the best features of previous mod- aries; and a location grid. A set of software tools, collec-
eling efforts. It also uses an Open Source approach to tively referred to as the data integration tools, reads these
provide free access to the underlying source code and to input files, diagnoses problems in them such as missing
make the model more open to scrutiny and to further or miscoded data, and applies decision rules to synthe-
extension and adaptation to emerging requirements for size missing or erroneous data and construct the model
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

modeling. data store.

A review of operational models for the Transit Co- The data store represents each household in the
operative Highway Research Program, Project H-12, de- metropolitan area as an individual object, with the pri-
veloped a specification for a proposed “ideal” integrated mary characteristics relevant to modeling location and
land use and transportation model system, and assessed travel behavior: household size, number of workers, pres-
operational models compared to this framework, in- ence of children, age of head, and household income.
cluding DRAM/EMPAL, MEPLAN, TRANUS, NYMTC- The household list is synthesized by integrating census
LUM, MUSSA, and UrbanSim (Miller et al., 1998). The household-level data from the Public Use Microdata
report concluded that UrbanSim came closest to their Sample with Summary Tape File 3A tabulations by cen-
proposed ideal specification. A more recent report by the sus tract and assigning synthesized households to parcel
National Cooperative Highway Research Project exam- data, using a variant of the procedure developed for the
ined operational models and assessed their potential for TRANSIMS model system (Beckman et al., 1996). Em-
use in evaluating the air quality impacts of highway ca- ployment is represented in the data store as individual
pacity expansion (Dowling et al., 2000). The review of records for each job and its employment sector.
land use models included DRAM/EMPAL (ITLUP), ME- The data store represents locations using grid cells
PLAN, NYMTC-LUM, and UrbanSim, and singled out of 150 by 150 meters, which contain just over 5.5 acres
UrbanSim as a foundation for further development. A (the cell size can be modified). This location grid allows
third recent review of models by the Environmental Pro- explicit cross referencing of other spatial features in-
tection Agency (U.S. EPA) inventoried a large number of cluding planning and political boundaries such as city,
land use models and analytical tools, but did not under- county, traffic zones, and urban growth boundaries, as
take any assessment of them (U.S. EPA, 2000). well as environmental features such as wetlands, flood-
The preceding discussion provides a general de- ways, stream buffers, steep slopes, and other environ-
scription of the UrbanSim design, how it addresses mentally sensitive areas.
emerging modeling requirements, and how it compares Figure 2 shows one grid cell superimposed over a
to other modeling approaches. Other references provide digital orthophoto from central Seattle’s Queen Anne
empirical results from the original specification of the neighborhood. Parcel data are collapsed into the cells to
model (Waddell, 2000a), a description of the data devel- generate composite representations of the mix and den-
opment process (Waddell et al., 1998), detailed specifi- sity of real estate at each location, labeled development
cations of the current model implementation (Waddell types. These development types are somewhat analogous
et al., in press), analysis of its relationship to land supply to the development typology developed by Calthorpe
monitoring (Waddell, 2000b), and description of its (1983), in that they represent at a local neighborhood
theoretical foundations (Waddell, 2000c; Waddell & scale the land use mix and density of development. Table
Moore, 2001), its application as a decision support 2 provides the rules for classifying grid cell development
system (Waddell, 2001), and the underlying software in- into types, based on the combination of housing units,
frastructure (Noth et al., 2000). A more detailed de- nonresidential square footage, and the principal land use
scription of the design of the UrbanSim database and of the development. The grid cell shown in Figure 2
model system is provided below, followed by a descrip- would be classified as a development type of R8, or high-

300 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3

 URBANSIM
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

FIGURE 1. UrbanSim data integration process.

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 301

 PAUL WADDELL
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

FIGURE 2. Sample grid cell of 150 by 150 meters, the unit of analysis for
location and development. Light lines are parcel boundaries. (Source data
courtesy of City of Seattle)

density residential, on the basis of containing 98 hous- rived from data about real households, businesses, and
ing units and no nonresidential square footage. parcels, it is a synthetic database that represents only
The data store maintains an explicit accounting of selected characteristics of people, jobs, real estate, and
real estate and occupants, linking individual households locations. Similarly, the models and their estimated pa-
to individual housing units, and individual jobs to job rameters attempt to reflect the patterns of observed be-
spaces that can be either nonresidential square footage or havior of real agents but are simplifications and abstrac-
a residential housing unit to account for home-based tions of real behavior, as are all models.
employment. When jobs or households are predicted to
move, the space they occupy is flagged as becoming va- Model Structure and Processing
cant, and when they are assigned to a particular housing UrbanSim includes model components reflecting
unit or job space, that space is reclassified as occupied. By the key choices of households, businesses, developers,
explicit assignment of housing units and nonresidential and governments (as policy inputs) and their interac-
square footage to grid cells of fixed size, densities and tions in the real estate market. By focusing on the prin-
mixtures of housing units and nonresidential square cipal agents in urban markets and the choices they make
footage of industrial, commercial, or governmental types about location and development, the model deals di-
are inventoried. Land values and residential and nonresi- rectly with behavior that planners, policymakers, and the
dential improvement values are also identified for each public can readily understand and analyze. This behav-
cell in the database. This integrated data store of house- ioral approach provides a theoretical structure more
holds, jobs, land, and real estate is what the model com- transparent than “black-box” models that do not clearly
ponents update over time. Although this data store is de- identify the agents and actions being modeled. The

302 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3

 URBANSIM

TABLE 2. Development type classifications.

Type No. Name Units (low) Units (high) Sq. ft. (low) Sq. ft. (high) Primary use

1 R1 1 1 0 999 Residential
2 R2 2 4 0 999 Residential
3 R3 5 9 0 999 Residential
4 R4 10 14 0 2,499 Residential
5 R5 15 21 0 2,499 Residential
6 R6 22 30 0 2,499 Residential
7 R7 31 75 0 4,999 Residential
8 R8 76 65,000 0 4,999 Residential
9 M1 0 9 1,000 4,999 Mixed R/C
10 M2 10 30 2,500 4,999 Mixed R/C
11 M3 10 30 5,000 24,999 Mixed R/C
12 M4 10 30 25,000 49,999 Mixed R/C
13 M5 10 30 50,000 9,999,999 Mixed R/C
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

14 M6 31 65,000 5,000 24,999 Mixed R/C

15 M7 31 65,000 25,000 49,999 Mixed R/C
16 M8 31 65,000 50,000 9,999,999 Mixed R/C
17 C1 0 9 5,000 24,999 Commercial
18 C2 0 9 25,000 49,999 Commercial
19 C3 0 9 50,000 9,999,999 Commercial
20 I1 0 9 5,000 24,999 Industrial
21 I2 0 9 25,000 49,999 Industrial
22 I3 0 9 50,000 9,999,999 Industrial
23 GV1 0 99,999 0 9,999,999 Government
24 VD2 0 0 0 0 Vacant developable
25 UD3 0 0 0 0 Undevelopable
1Government
2Vacant developable
3Undevelopable

structure allows users to incorporate policies explicitly model is tempered by market information that reflects
and to evaluate their effects. the state of the market as a whole, such as vacancy rates.
UrbanSim is not a single model. It might be better The structure and processing sequence of UrbanSim
described as an urban simulation system, consisting of a are shown in Figure 3. Inputs to the model include the
software architecture for implementing models and a base year data store, control totals derived from regional
family of models implemented and interacting within economic forecasts from an external macroeconomic
this environment. The models that are currently imple- model, travel access indicators derived from external
mented employ a range of techniques and approaches. travel demand models, scenario policy assumptions re-
Some of the models, such as the economic and demo- garding development constraints arising from land use
graphic transition models, are aggregate, nonspatial plans and environmental constraints, as well as any user-
models that deal with the interface to external macro- specified events. The individual model components
economic changes. Other components, such as location predict
choice, are discrete choice models of an agent (a house-
hold, for example) making choices about alternative lo- • the pattern of accessibility by auto ownership level
cations, taking a top-down or bird’s-eye view of the met- (Accessibility Model);
ropolitan area. The developer model, by contrast, takes a • the creation or loss of households and jobs by type
mostly bottom-up view from the vantage point of a de- (Economic and Demographic Transition Models);
veloper or landowner at a single location (grid cell) mak- • the movement of households or jobs within the
ing choices about whether to develop and into what type region (Household and Employment Mobility
of real estate. The bottom-up view in the developer Models);

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 303

 PAUL WADDELL

• the location choices of households and jobs from Broadly speaking, government agencies influence
the available vacant real estate (Household and the land development process through a combination
Employment Location Choice Models); of land use regulations and infrastructure provision.
• the location, type, and quantity of new These are frequently combined into packages that at-
construction and redevelopment by developers tempt to foster a development pattern in ways that pro-
(Real Estate Development Model); and mote planning objectives, for example by pursuing one
• the price of land at each location (Land Price or a combination of the following community visions:
Model).
• containing development within an urban growth
One special component, the Model Coordinator, boundary;
manages the individual model components and handles • focusing development along primary transporta-
the scheduling and implementation of events, such as tion corridors;
reads and writes to the data store. Taken together as a • focusing development within centers connected by
system, these components maintain the data store and multimodal transportation;
simulate its evolution from one year to the next. For sim- • diverting development into new or existing
plicity, the household and employment counterpart satellite communities;
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

models for transition, mobility, and location are repre- • encouraging development in parts of the region
sented jointly in the diagram and described together in with underused infrastructure; and
Figure 4, since they are parallel and almost identical. • promoting development of impoverished areas.
Each of the model components are executed in the se-
quence shown as t.1 . . . t.7, after which the simulation In the UrbanSim context, use of the term scenario dif-
year is incremented and the process repeats for the next fers from a description of a particular vision such as
simulation year. those listed above. An UrbanSim scenario is a collection
of policy assumptions that can be input to the model to
Simulated and User-Specified Events. The model sys-
examine their potential consequences on outcomes such
tem runs on events generated by the model components.
as urban form, land use mix, density, and travel patterns.
A number of choices by households, businesses, and de-
In other words, the system allows interactive testing of
velopers are simulated on an annual basis, and their out-
how different policy strategies fare in achieving a partic-
comes are implemented as scheduled events. Large-scale
ular vision or set of community objectives. It does not
development projects may be scheduled with multiyear
assume that a particular vision can be realized, but facil-
timetables, defined using a template that describes the
itates exploration of the trade-offs that may be involved
characteristics of different types of development events.
in attempting to achieve it, given the range of policies
In addition to model-generated events, the system ac-
available and their costs and consequences. The model
commodates information that planners have about
does not attempt to optimize policy inputs, but is in-
pending development, corporate relocations, or policy
tended to facilitate interactive use to support an itera-
changes. We have developed a capacity to introduce user-
tive, participatory planning process.
specified events such as these into the model, both to
The translation of these scenarios into inputs to Ur-
allow planners to use available information about devel-
banSim involves interpreting policies and creating input
opments that are in the pipeline and also to provide a ca-
files for the model that represent these policy interpre-
pacity for testing the potential effects of a major project
tations. Interpreting the comprehensive land use plan is
on further development and on traffic.
a key part of constructing a policy scenario in UrbanSim.
Scenario Assumptions. UrbanSim allows users to Each land use plan designation (planned land use, or
specify policy inputs and assumptions, generate and PLU) may be described as a set of restrictions on devel-
compare scenarios, compute evaluation measures, and opment options. For example, the plan designation of
query the database of results. The user interface of the “agricultural” may not allow conversion to any devel-
model is focused on the interaction of the user with the oped urban category under restrictive interpretation of
inputs to each scenario. Scenarios consist of a combina- the land use plan, or may allow conversion to rural-
tion of development policies, represented by appropri- density, single-family residential under a less restrictive
ate input data such as comprehensive plans, infrastruc- interpretation. The adopted comprehensive plan guide-
ture plans, urban growth boundaries, and development lines for a local area should spell out the intended inter-
restrictions on environmentally sensitive lands. These pretation of these plan designations, but the user of the
policies are linked to locations at a grid cell, zonal, mu- model may wish to assess the impact of altering these
nicipal, county, or metropolitan scale. constraints as a matter of policy testing.

304 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3

 URBANSIM
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

FIGURE 3. UrbanSim model structure and processing.

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 305

 PAUL WADDELL

Economic and Demographic Transition Models (t.2) has no location. For each such job, a sample of loca-
The Demographic Transition Model simulates births tions with empty sq. ft. or space in housing units for
and deaths in the population of households. Exter- home-based jobs is randomly selected from the set of
nally imposed population control totals determine all possible alternatives.
overall target population values and can be specified The variables used in the Household Location
in more detail by distribution of income groups, age, Model include attributes of the housing in the grid cell
size, and presence or absence of children. This enables (price, density, and age), neighborhood characteris-
the modeling of a shifting population distribution over tics (land use mix, density, average property values,
time. and local accessibility to retail), and regional accessi-
Iterative proportional fitting (Beckman et al., bility to jobs. Variables in the Employment Location
1996) is used to determine how many households of Model include real estate characteristics in the grid
each type are to be created or deleted. Newly created cell (price, type of space, density, and age), neighbor-
households are added to the household list but with- hood characteristics (average land values, land use
out an assignment to a specific housing unit (placed in mix, and employment in each other sector), and re-
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

limbo), to be placed in housing later by the House- gional accessibility to population.

hold Location Choice Model. Households to be de-
leted to meet the control totals are selected at ran- Real Estate Development Model (t.5)
dom, drawn preferentially from households in limbo. The Real Estate Development Model simulates devel-
The Economic Transition Model is responsible for oper choices about what kind of construction to un-
modeling job creation and loss. Employment control dertake and where, including both new development
totals determine employment targets and can be spec- and redevelopment of existing structures. Each year,
ified by distribution of business sector. the model iterates over all grid cells on which devel-
opment is allowed and creates a list of possible tran-
Household and Employment Mobility Models (t.3) sition alternatives (representing different development
The Household Mobility Model simulates households types), including the alternative of not developing.
deciding whether to move. Movement probabilities The probability for each alternative being chosen
are based on historical data. Once a household has is calculated in a multinomial logit model. Variables
chosen to move, it is placed in limbo to indicate it has included in the developer model include characteris-
no current location, and the space it formerly occu- tics of the grid cell (current development, policy con-
pied is made available. straints, and land and improvement values), charac-
The Employment Mobility Model determines teristics of the site location (proximity to highways,
which jobs will move from their current locations dur- arterials, existing development, and recent develop-
ing a particular year using a similar approach to the ment), and regional accessibility to population.
Household Mobility Model.
Land Price Model (t.6)
Household and Employment Location Models (t.4) The Land Price Model simulates land prices of each
The Household Location Choice Model chooses a lo- grid cell as the characteristics of locations change over
cation for each household that has no current loca- time. It is based on urban economic theory, which
tion. For each such household, a sample of locations states that the value of location is capitalized into the
with vacant housing units is randomly selected from price of land. The model is calibrated from historical
the set of all vacant housing. Each alternative in the data using a hedonic regression to include the effect of
sample is evaluated for its desirability to the house- site, neighborhood, accessibility, and policies on land
hold through a multinomial logit model calibrated to prices. It also allows incorporating the effects of short-
observed data. The household is assigned to its most term fluctuations in local and regional vacancy rates
desired location among those available. on overall land prices. Variables used similar are to
The Employment Location Choice Model is re- those in the Real Estate Development Model.
sponsible for determining a location for each job that

FIGURE 4. Descriptions of behavioral models.

306 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3

 URBANSIM

Development regulations may be coded for an en-

tire metropolitan area, for individual counties, cities, or
special overlays such as environmentally sensitive lands
or urban growth boundaries. Overlays such as wetlands,
floodways, steep slopes, or other environmental features
may be used to specify environmental regulations that
impose development constraints. The model interprets
the cumulative impact of the policies by reflecting the
most restrictive policies that apply to a given grid cell.
For example, a general county plan might allow sub-
stantial development for a particular land use plan des-
ignation, but a more restrictive regulation that applies
to wetlands would overrule this for any grid cell that was
in a wetland. Figure 5 shows a portion of the UrbanSim
user interface for specifying development constraints.
The term PLU indicates planned land use, which can rep-
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

resent either the land use plan designation or a zoning

category.
In addition to development constraints, the scenario
inputs include regional control totals from the external
macroeconomic models and assumptions about the
space utilization rates (such as square feet per employee
for different development types). Transportation policy
assumptions are incorporated in the external transpor-
tation model and are embedded in the travel time and
utility outputs from the travel model that UrbanSim
uses to calculate accessibility.
Local and Regional Accessibility. The Accessibility FIGURE 5. UrbanSim user interface for entering
Model (see Figure 3) is responsible for maintaining ac- development constraints.
cessibility values for occupants within each traffic zone,
including access by residents and employees to shopping
and other amenities, to employment, and to the central
business district. The accessibility value for a zone to a transportation system, since running it is relatively cum-
specific type of activity is defined as the sum of the quan- bersome and its outputs generally change more slowly
tity of the activity (jobs, for example) at each possible than other values in the simulation. However, UrbanSim
destination, discounted by a weight between 0 and 1 re- is run annually, updating the accessibility values based
flecting the multimodal travel utility to the destination.1 on the evolving spatial pattern of activities.
Handy (1993) and others have referred to this kind of UrbanSim also incorporates local accessibility mea-
measure as representing “regional accessibility” in that it sures, corresponding to the activities that can be reached
is regional in scope and uses the transportation network by walking,2 using spatial queries of the grid cells in the
on a zone-to-zone basis to represent travel access. It is data store. Achieving this scale of analysis makes Urban-
contrasted with “local accessibility,” which measures ac- Sim the first operational urban model system to support
cess to opportunities within a walkable neighborhood. analysis of location and travel behavior at a level that can
The link between UrbanSim and the Travel Demand effectively represent pedestrian and bicycle scales of
Model System (see Figure 3) is two way, since different travel. Given the ongoing debate over the potential in-
accessibility values from the travel model will influence fluence of neotraditional urban design on travel behav-
the decisions of developers, employers, and residents, ior, this innovation should provide a basis for making
giving rise to different travel demands, which then feed more systematic assessments of the effects of urban de-
back into the travel model. The external travel models sign-scale policies on both location and travel behavior.
provide travel times and utilities to the Accessibility Traditional zone-based travel models are severely lim-
Model. The travel models are typically run to simulate 1 ited by poor performance on intrazonal travel and in-
day every 5 years or when there is a major change to the sufficient representation of nonmotorized travel modes.

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 307

 PAUL WADDELL

By creating a more detailed basis for the land use model,

the main barrier to the improvement of transportation
planning to address nonmotorized modes and the inte-
gration of urban design policies has been effectively
removed.
Data Export. The data export process is responsible
for gathering, aggregating, and exporting data from the
object store to a set of ASCII Output Files for subsequent
analysis and GIS Visualization (see Figure 3). The user
interface allows specification of desired output files and
designation of specific simulation years for which to gen-
erate the outputs. Outputs are created at the grid cell
level and also summarized by traffic zone and for the re-
gion as a whole. The data are written in a standard for-
mat for ease of loading into ArcView, Excel, or other
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

common computer software.

Application to Eugene-Springfield,
Oregon
The UrbanSim model system was fully implemented
first in Eugene-Springfield, Oregon, in TLUMIP, a pro-
ject funded by the Oregon Department of Transporta-
tion. The test site was chosen for a number of reasons, in-
cluding the relatively small size of the metropolitan area
and the availability of data needed for model applica-
tion. The Lane Council of Governments (LCOG) is the
Metropolitan Planning Organization for the area and
provided all the data used in the model development and
application.
The Eugene-Springfield metropolitan area, shown
in Figure 6, lies in Lane County at the south end of Ore-
gon’s Willamette Valley, a fertile agricultural region that
contains most of the economic activity in the state, in- FIGURE 6. Location of Eugene-Springfield in western
cluding the metropolitan areas of Portland and Salem. Oregon. (Source data courtesy of Oregon Geospatial
Eugene-Springfield is a relatively small metropolitan Data Clearinghouse)
area. In 2000, the population of Lane County was
322,959, with 137,893 residing in Eugene and 52,864 in
Springfield (U.S. Census Bureau, 2000).
The base year for the Eugene-Springfield application and Land Price Model (t.4, t.5, and t.6 in Figure 3). Re-
is 1994, since consistent data were available from that sults are summarized for each traffic analysis zone (TAZ)
year for employment, household survey, and parcels. for input to the Travel Demand Model System.
Population data were derived from a 1994 home inter- The classification of households for use in Urban-
view transportation survey and from the 1990 census, Sim is based on the stratification of five household char-
and adjusted to the 1994 base year using the parcel dis- acteristics, shown in Table 3. Employment was classified
tribution of the housing stock in that year. The scope of using 2-digit standard industrial classification codes,
the study area is the extent of the 271 traffic zones in- grouped into sectors that are generally consistent with
cluded in the LCOG transportation model system. The those used in the LCOG transportation models. These
study area was further subdivided into approximately are shown in Table 4.
15,000 grid cells of 150 by 150 meters as the basic unit of Results from this processing of parcel data for the
spatial analysis in the Household and Employment Lo- Eugene-Springfield application of the model are shown
cation Choice Models, Real Estate Development Model, in Figures 7 and 8. The housing distribution in the 1994

308 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3

 URBANSIM

TABLE 3. UrbanSim classifications of household characteristics for Eugene-Springfield.

Annual income Age of head No. persons No. workers No. children

Less than $5,000 15 to 24 1 0 0

$5,000 to $9,999 25 to 34 2 1 1 or more
$10,000 to $14,999 35 to 44 3 2 or more
$15,000 to $24,999 45 to 54 4
$25,000 to $34,999 55 to 64 5 or more
$35,000 to $49,999 65 to 74
$50,000 to $74,999 75 or over
$75,000 to $99,999
$100,000 or more
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

database is shown in Figure 7, with traffic zones and the TABLE 4. Employment sector classifications for
urban growth boundary superimposed. Figure 8 depicts Eugene-Springfield.
the distribution of nonresidential square footage in
1994 and conveys a pattern of decentralized employ- Standard industrial
classification codes Sector description
ment centers beyond the central business district.
After developing the database and estimating the 99– 999 Agriculture
parameters of the equations in UrbanSim using the 2400–2499 Lumber and wood
LIMDEP statistical software (Greene, 1998), the perfor- 2500–2599, 3200–3999 Other durable
mance of UrbanSim in Eugene-Springfield was evalu- 2000–2099 Food products
ated over a historical period. A 1980 database was de- 2100–2399, 2600–3199 Other nondurable
veloped, and the 1994 database that was used to 1500–1799 Construction
calibrate UrbanSim became the observed target for 1000–1499 Mining
comparison of simulation results. UrbanSim was run 4000–4999 Transportation
in annual steps from 1980 to 1994, then compared to 5000–5199 Wholesale trade
the observed 1994 data. Table 5 summarizes the corre- 5200–5999 Retail trade
6000–6999 Fire
lation coefficients between the simulated and the ob-
7000–8199 Services
served 1994 data. 8200–8299 Education
The simulation results are compared to observed 9000–9999 Government
data at three units of geography. The unit used in the
models is the 150-meter-square grid cell, although the
model results are not generally intended for use at this
level of detail. Nevertheless, the model simulation results
correlate well to the observed data after 15 years of sim- TABLE 5. Correlation of simulated to observed 1994
ulation. Aggregation of the results to traffic zones used values in Eugene-Springfield.
in transportation models produced higher correlations,
with all but one category above a 0.9 correlation coeffi- Average over
Characteristic Cell Zone 1-cell radius
cient. Another spatial comparison was made on the grid
cells averaged over the cells within one cell radius, and Employment 0.805 0.865 0.917
these produced correlations at least comparable to the Population 0.811 0.929 0.919
zonal aggregations. Nonresidential sq. ft. 0.799 0.916 0.927
A more stringent benchmark than the preceding Housing units 0.828 0.927 0.918
comparison of simulated to observed 1994 values is the Land value 0.830 0.925 0.908
comparison of the observed changes from 1980 to 1994 to
the simulated changes during this period. Figure 9 depicts
this comparison for households and jobs, using the 271
traffic zones as the basis for comparison. The graph
shows the percentage of traffic zones classified according

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 309

 PAUL WADDELL
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

FIGURE 7. Housing units by grid cell, 1994. (Map courtesy of Lane Council of Governments)

to the difference between the simulated and observed ing. It should be noted that UrbanSim does not contain
changes from 1980 to 1994. For example, the range adjustment factors (sometimes known as K-factors)
“−500 or less” indicates that between 1980 and 1994 the common in the cross-sectional validation of models to
model underpredicted by at least 500 the change in the observed data, so these results are obtained purely from
number of households or jobs. For households, the the underlying behavior of the model and not from ad-
model predicted a change of ±50 (the two middle ranges) justments to correct for errors.
—a total of 57% of the zones. For employment, the model Several factors condition the interpretation of these
predicted a change of 0 to 50—a total of 31% of the zones. results. First, Eugene-Springfield is a fairly small metro-
For households and jobs, the model predicted changes politan region, and the model simulated a period dur-
of ±200 (the four middle ranges)—with respective totals ing which change was modest, with population in the
of 89% and 76% of the zones. These results, while mostly study area growing from 185,000 in 1980 to just over
encouraging regarding the ability of the simulation to 200,000 in 1994, while jobs grew from 75,000 to nearly
reproduce observed changes over time, nevertheless 100,000. Second, policies such as the urban growth
show that changes in some zones were significantly over- boundary were essentially in place at the beginning of
predicted or underpredicted. Employment change was the period, and transportation system changes were rela-
not predicted as well as household change, an outcome tively minor, so changes resulting from policy interven-
that is not entirely surprising given that employment tions were not significant. Third, the difficulties of as-
tends to be more concentrated and volatile than hous- sembling the base year data for model implementation,

310 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3

 URBANSIM
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

FIGURE 8. Nonresidential square footage by grid cell, 1994. (Map courtesy of Lane Council of Governments)

which were substantial even with current data, were While much has been accomplished, many challenges
compounded in the process of assembling a 1980 base remain. Further development priorities include the
year. Archival parcel maps and employment records were following:
difficult to work with, and in the end, considerable error
remained in the historical data, diminishing the capacity • developing software tools for robust data
of the historical validation exercise to inform us about preparation and integration to facilitate applying
how well the model performed. the system to other areas;
The model did not predict isolated events that oc- • developing a version of the system that would be
curred in the region during this period (nor was it de- suitable for classroom use in courses dealing with
signed to). One was a significant downsizing of a Weyer- urban development, transportation, and environ-
houser plant in Springfield and another was the opening mental planning;
of the Gateway Mall on Interstate Highway 5. These • adding environmental components to simulate
kinds of large-scale events will not be accurately mod- land cover change, water demand, and nutrient
eled by any model system and are reminders of the limits emissions (Alberti & Waddell, 2000; Waddell &
of modeling. Alberti, 2000);
• adding an evaluation component that computes
predefined indicators, allows users to construct
Conclusions new ones, and provides useful visualizations of
The UrbanSim project has made significant progress them across multiple scenarios;
toward developing models to support land use and • adding an economic evaluation capacity that
transportation planning and growth management. supports cost-benefit analysis and least-cost

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 311

 PAUL WADDELL
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

FIGURE 9. Difference between observed and simulated changes in households

and employment from 1980 to 1994 by traffic zones in Eugene-Springfield.

planning, incorporating social and environmental Metropolitan land use, transportation, and envi-
externalities and equity considerations; ronmental planning must be more effectively integrated
• developing an interface for access to the system than has traditionally been the case. This integration re-
over the Internet, to support coordination of the quires robust analytical methods and should be open to
model application across local governments within public scrutiny and deliberation in ways that have not
a region and to provide public access to participate been accomplished in the past. Simulation models can
in scenario development and evaluation; and should be part of this deliberative policy process, but
• adding microsimulation of demographic they will have to come out of the “black box” and become
processes (household change); instruments that facilitate discussion between local gov-
• adding a more behavioral real estate development ernments and their constituents. The challenge of bal-
model that represents the roles of landowners, ancing multiple objectives and agendas within urban
lenders, investors, and specialized developers; areas in the U.S. and abroad have grown increasingly in-
• integrating an activity-based travel model; tractable politically, and this work represents a small ef-
• integrating a macroeconomic model that reflects fort to contribute to more deliberative and informed
the potential macro-effects of local choices about metropolitan governance. It is, in closing, only one step
major infrastructure and land policies; forward. What lies ahead is a challenging agenda to re-
• developing more robust methods for calibrating fine the analytical tools for metropolitan and local plan-
and validating the system, incorporating uncer- ning, make them more accessible and robust, and to
tainty about models and data more explicitly; and generate collaboration in the development and use of
• leveraging open source development to enlist planning methods such as these to help communities
collaborators in the development of new tools and that want to grow smarter.
the application of the system in other areas.

312 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3

 URBANSIM

ACKNOWLEDGEMENTS Blizzard, M. (1996). Creating better communities: The LUTRAQ

principles. Tigard, OR: Sensible Transportation Options
The development of the current version of UrbanSim has been
for People (STOP).
funded in part by the National Science Foundation, grants
Calthorpe, P. (1983). The next American metropolis: Ecology, com-
CMS-9818378, EIA-0121326 and EIA-0090832, and by the
munity and the American dream. New York: Princeton Ar-
University of Washington, Puget Sound Regional Synthesis
chitectural Press.
Model project. We also wish to acknowledge the sponsors of
de la Barra, T. (1989). Integrated land use and transport modelling.
earlier development of the system, including Oahu Metropol-
Cambridge, UK: Cambridge University Press.
itan Planning Organization; Oregon Department of Trans-
Dowling, R., Ireson, R., Skabardonis, A., Gillen, D., Stopher, P.,
portation; National Cooperative Highway Research Program
Horowitz, A., et al. (2000). Predicting short-term and long-term
8-32(3), “Integration of Land Use Planning and Multimodal
air quality effects of traffic-flow improvement project (NCHRP
Transportation Planning”; and The Governor’s Office of Plan-
Report No. 25-21). Washington, DC: National Academy
ning and Budget, State of Utah. We particularly want to ac-
Press.
knowledge the assistance of Bud Reiff of the Lane Council of
Downs, A. (1992). Stuck in traffic: Coping with peak-hour traffic con-
Governments.
gestion. Washington, DC: Brookings Institution & Cam-
The development of the current version of UrbanSim was
bridge, MA: Lincoln Institute of Land Policy.
undertaken by an interdisciplinary team at the University of
Downs, A. (1994). New visions for metropolitan America. Wash-
Washington led by Paul Waddell, Associate Professor of Pub-
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

ington, DC: Brookings Institution.

lic Affairs and of Urban Design and Planning, in coordination
Echenique, M. H., Flowerdew, A. D., Hunt, J. D., Mayo, T. R.,
with Alan Borning, Professor of Computer Science and Engi-
Skidmore, I. J., & Simmonds, D. C. (1990). The MEPLAN
neering; Marina Alberti, Associate Professor of Urban Design
models of Bilbao, Leeds and Dortmund. Transport Reviews,
and Planning; and Scott Rutherford, Professor of Civil and En-
10, 309–322.
vironmental Engineering. Michael Noth developed the Ur-
Fotheringham, A. S., & O’Kelly, M. E. (1989). Spatial interaction
banSim software architecture, and Nathan Freier and Michael
models: Formulations and applications. Boston: Kluwer
Becke developed much of the current model implementation
Academic.
and the user interface. Gudmundur Ulfarsson provided sub-
Free Software Foundation. (1991). GNU general public license,
stantial assistance with the calibration of the system.
version 2. Retrieved from <http://www.gnu.org/copyleft/
gpl.html>.
Garrett, M., & Wachs, M. (1996). Transportation planning on trial:
NOTES The Clean Air Act and travel forecasting. Thousand Oaks, CA:
1. The composite utility of travel from zone to zone is based Sage.
on the logsum term in the mode choice component of the Greene, W. (1998). LIMDEP, version 7.0. Plainview, NY: Econo-
Travel Demand Model System, which incorporates times metric Software, Inc.
and costs of all modes from origin to destination. It is Handy, S. (1993). Regional versus local accessibility: Implica-
scaled to a maximum value of 0, and exponentiated to tions for nonwork travel. Transportation Research Record,
achieve a resulting weight between 0 and 1. The accessi- 1400, 58–66.
bility measure will therefore increase as modes are added Harris, B. (1985). Urban simulation models in regional science.
or as they are made faster or less costly. Journal of Regional Science, 25(4), 545–567.
2. Local queries use a distance of 600 meters (approximately Intermodal Surface Transportation Efficiency Act of 1991 (PL
1/3 mile). 102-240), 105 United States Statutes at Large, pp. 1914
(1991, December 18).
Kain, J. F. (1985). Computer simulation models of urban lo-
cation. In E. S. Mills (Ed.), Handbook of regional and urban
REFERENCES
economics, volume II (pp. 847–875). Amsterdam: Elsevier
Alberti, M., & Waddell, P. (2000). An integrated urban devel- Science.
opment and ecological simulation model. Integrated As- Landis, J. (1994). The California urban futures model: A new
sessment, 1(3), 215–227. generation of metropolitan simulation models. Environ-
Anas, A. (1982). Residential location markets and urban transporta- ment and Planning B: Planning and Design, 21, 399–420.
tion: Economic theory, econometrics, and policy analysis with dis- Landis, J. (1995). Imagining land use futures: Applying the
crete choice models. New York: Academic Press. California urban futures model. Journal of the American
Anas, A. (1987). Modeling in urban and regional economics. New Planning Association, 61(4), 438–457.
York: Harwood Academic. Landis, J., & Zhang, M. (1998a). The second generation of the
Beckman, R. J, Baggerly, K. A., & McKay, M. D. (1996). Creating California urban futures model. Part 1: Model logic and
synthetic baseline populations. Transportation Research. theory. Environment and Planning B: Planning and Design, 25,
Part A: Policy and Practice, 30(6), 415–429. 657–666.
Ben-Akiva, M., & Lerman, S. (1987). Discrete choice analysis: The- Landis, J., & Zhang, M. (1998b). The second generation of the
ory and application to travel demand. Cambridge, MA: MIT California urban futures model. Part 2: Specification and
Press. calibration results of the land-use change submodel. En-

APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3 313

 PAUL WADDELL

vironment and Planning B: Planning and Design, 25, 795– tion and housing market components of UrbanSim. En-
824. vironment and Planning B: Planning and Design, 27(2), 247–
Lee, D. (1973). Requiem for large scale urban models. Journal of 263.
the American Institute of Planners, 39(3), 163–178. Waddell, P. (2000b). Monitoring and simulating land capacity
Lee, D. (1994). Retrospective on large-scale urban models. Jour- at the parcel level. In A. Vernez-Moudon & M. Hubner
nal of the American Planning Association, 60(1), 35–40. (Eds.), Monitoring land supply with geographic information sys-
Martinez, F. (1992). The bid-choice land use model: An inte- tems: Theory, practice and parcel-based approaches (pp. 201–
grated economic framework. Environment and Planning A, 217). New York: John Wiley & Sons.
24, 871–885. Waddell, P. (2000c, July). Towards a behavioral integration of land
McFadden, D. (1973). Conditional logit analysis of qualitative use and transportation modeling. Paper presented at the 9th
choice behavior. In P. Zarembka (Ed.), Frontiers in Econo- International Association for Travel Behavior Research
metrics (pp. 105–142). New York: Academic Press. Conference, Queensland, Australia.
Miller, E. J., Kriger, D. S., & Hunt, J. D. (1998). Integrated urban Waddell, P. (2001). Between politics and planning: UrbanSim
models for simulation of transit and land-use policies. Washing- as a decision support system for metropolitan planning.
ton, DC: National Academy of Sciences, Transit Cooper- In R. Brail & R. Klosterman (Eds.), Planning support systems:
ative Research Project. Integrating geographic information systems, models, and visual-
Noth, M., Borning, A., & Waddell, P. (2000). An extensible, ization tools (pp. 201–228). Redlands, CA: Environmental
Downloaded By: [University of Washington Libraries] At: 21:38 14 April 2009

modular architecture for simulating urban development, Systems Research Institute.

transportation, and environmental impacts. Manuscript Waddell, P., & Alberti, M. (2000, September). Integrated simu-
submitted for publication. lation of real estate development and land cover change. Paper
Orfield, M. (1997). Metropolitics: A regional agenda for community presented at the 4th International Conference on Inte-
and stability. Washington, DC: Brookings Institution Press grating Geographic Information Systems (GIS) and En-
& Cambridge, MA: Lincoln Institute of Land Policy. vironmental Modeling, Banff, Canada.
Parsons Brinckerhoff Quade and Douglas, Inc. (1999). Land Waddell, P., & Moore, T. (2001). Forecasting demand for urban
use impacts of transportation: A guidebook (NCHRP Report land. In G. Knaap (Ed.), Land market monitoring for smart
423A). Washington, DC: National Academy Press. urban growth (pp. 185–217). Cambridge, MA: Lincoln In-
Paulley, N. J., & Webster, F. V. (1991). Overview of and inter- stitute of Land Policy.
national study to compare models and evaluate land use Waddell, P., Moore, T., & Edwards, S. (1998, May). Exploiting
transport policies. Transport Reviews, 11, 197–222. parcel-level GIS for land use modeling. In S. Easa & D.
Putman, S.H. (1983). Integrated urban models. London: Pion Ltd. Samdahl (Eds.), Transportation, land use and air quality: Mak-
Southworth, F. (1995). A technical review of urban land use-trans- ing the connection (pp. 632–641). Reston, VA: American So-
portation models as tools for evaluating vehicle reduction strategies ciety of Civil Engineers.
(Report ORNL-6881). Oak Ridge, TN: Oak Ridge Na- Waddell, P., Borning, A., Noth, M., Freier, N., Becke, M., & Ul-
tional Laboratory. farsson, G. (in press). UrbanSim: A simulation system for
Transportation Equity Act for the 21st Century (PL 105-178), land use and transportation. Networks and Spatial Economics.
112 United States Statutes at Large, pp. 107–728 (1998, Weatherby, C. (1995). Summary of workshop observations and
June 9). recommendations. In G. Shunk, P. Bass, C. Weatherby, &
U.S. Environmental Protection Agency. (2000). Projecting land- L. Engleke (Eds.), Land Use Modeling Conference Proceedings,
use change: A summary of models for assessing the effects of com- February 19–21, 1995 (pp. 1–5). Washington, DC: U.S. De-
munity growth and change on land-use patterns (EPA/600/R- partment of Transportation.
00/098). Cincinnati, OH: U.S. Environmental Protection Wegener, M. (1994). Operational urban models: State of the
Agency, Office of Research and Development. art. Journal of the American Planning Association, 60(1), 17–30.
U.S. Census Bureau. (2000). Census 2000 summary tape file 1. Re- Wegener, M. (1995, February). Current and future land use
trieved from <http://factfinder.census.gov>. models. In G. Shunk, P. Bass, C. Weatherby, & L. Engleke
UrbanSim Project. (2002). UrbanSim. Retrieved April 12, 2002 (Eds.), Land Use Modeling Conference Proceedings, February
from <http://www.urbansim.org>. Seattle: University of 19–21, 1995 (pp. 13–40). Washington, DC: U.S. Depart-
Washington. ment of Transportation.
Waddell, P. (2000a). A behavioral simulation model for met- Wilson, A. G. (1967). Statistical theory of spatial trip distribu-
ropolitan policy analysis and planning: Residential loca- tion models. Transportation Research, 1, 253–269.

314 APA Journal ◆ Summer 2002 ◆ Vol. 68, No. 3