UPlan: A Versatile Urban Growth Model
for Transportation Planning
TRB Paper 03-2542
Robert A. Johnston
Dept. of Environmental Science and Policy
University of California
One Shields Ave.
Davis, CA 95616
rajohnston@ucdavis.edu
(530)582-0700
and
David R. Shabazian
Sacramento Area Council of Governments
3000 S. St.
Sacramento, CA 95816
drshabazian@sacog.org
(916) 457-2264
October 22, 2002
Resubmitted to the Transportation Research Board for
Presentation at the Annual Meeting, January 2003
Keywords: GIS, urban modeling, urban planning, impact assessment, induced
development, growth-inducing impacts
Word count: 7,919 plus one table
UPlan: A Versatile Urban Growth Model
for Transportation Planning
ABSTRACT:
We review urban models useful in transportation planning, focusing especially on
ones that are based on geographic information systems (GIS) software. We then describe
UPlan, a simple model written by us in the ArcView GIS. Several different applications
of UPlan are outlined, involving transportation planning and the analysis of the growthinducing effects of new facilitites, to demonstrate its use. Such models are coming into
use for NEPA assessments and for joint land use and transportation planning.
I. INTRODUCTION
In this paper, we describe a Geographic Information System (GIS)-based urban
growth model that runs in the Windows version of ArcView on a personal computer.
The model was designed by us to rely on a minimum amount of data, but allocates urban
growth in several land use types for small (parcel-sized) grid cells. It is a scenario-testing
model that can be applied to any county or metropolitan region and that is transparent to
the user, making it easy to change the assumptions for land use allocation. The model is
rule-based, that is it is not strictly calibrated on historical data and uses no choice or other
statistical models. Projecting the detailed footprint of development with several land use
types allows us to then apply various urban impact models, that forecast soil erosion,
local service costs, and other impacts. We describe several applications of UPlan. Any
state/regional/local transportation agency, citizens group, or county planning department
will be able to utilize this model, using generally available datasets.
II. BACKGROUND
The 1991 Intermodal Surface Transportation Efficiency Act (ISTEA) marked a
major turning point in how transportation modeling is conducted. Prior to the Act,
transportation planning occurred in somewhat of a vacuum. Travel demand models were
run with the same land use inputs for all scenarios, so changes in land uses due to
network improvements were not accounted for. The basic concern of most modeling
efforts was to simply test how network improvements affected congestion and air quality,
overlooking how those improvements influence urban development. Since the enactment
of ISTEA, however, transportation agencies have begun to adopt methods that simulate
changes in land use, as well as changes in travel. These integrated models, or, more
commony, linked transportation and land use models, enable planners to develop a better
understanding of how these urban systems interact.
Urban models are an improvement over travel demand models of the past, both
theoretically and operationally. However, the urban models considered to be the most
comprehensive and behaviorally based (land market bidding models) are also very
tedious to calibrate and operate. They take 2-3 years to develop the data and calibrate the
model and cost $1-2 million. For planning agencies and citizen groups interested only in
the land use results of these models, they will likely find them prohibitively complicated
and expensive. It is in this context that the geographic information system (GIS)-based
land use models are especially useful.
III. TYPES OF URBAN MODELS
As there are many types of urban models, we should start by describing the
various other types, to place GIS models in perspective. The FHWA has recently put up a
Web site, called the Toolbox, that outlines a variety of analysis methods useful in
transportation planning and in evaluating transportation plans and projects
(http://www.fhwa.dot.gov/planning/toolbox/land_develop_forecasting.htm). Their
typology of methods for forecasting land development patterns is:
1. Proximity-Based Forecasting. These are regression models that project development
based on the proximity of past development to transport facilities and other urban
infrastructure.
2. Delphi/Expert Panel. Several case studies of these methods are given. The Delphi
method has also been documented in a TRB report (Land Use Impacts of Transportation
1999).
3. Accessibility-Based Forecasting. Accessibility, derived from a travel model, is used to
forecast development.
4. Simple Land Use Models. These are zone-based models based on a small set of
equations defining relationships with accessibility and past development rates. HLFM II+
is a FHWA-supported model for use by small MPOs.
5. Complex Land Use Models. These can be a land use model that interfaces with an
existing travel model, or an integrated urban model with land development and travel
models together. These models generally use land prices, and sometimes floorspace lease
values, to represent demand for space. They also use accessibility and other factors to
represent site attributes. DRAM/EMPAL has been widely used in the U.S. and does not
use land value or floorspace lease value data and so is the easiest to implement. TRANUS
and MEPLAN have been applied to many regions all over the world and do rely on land
market data. A review of complex land use models can be found at Wegener (1994).
Another way to categorize land use models is to examine those in use in regional
transportation planning agencies. The following table is derived from Miller, Kriger, and
Hunt (1998) and updated to 2001. It shows the combinations of land use models and
travel models in use or in development in the U.S. It is important to note that most MPOs
use the judgement method of land use forecasting and then use this single forecast for all
transportation investment scenarios. This is an inaccurate method, in that improvements
in radial accessibility will generally increase the spread of land development. Significant
additions to road capacity, especially on the edges of congested urban regions, will
increase land development in those areas, according to the official study in the U.S.
(Expanding Metropolitan Highways 1995). If these land use impacts in the outer areas are
not assessed, the NEPA documents will be inaccurate in that the studies will likely bias
the projections of travel and emissions downward for highway improvement plans and
projects. The secondary effects of land development on habitats, water quality,
farmlands, and other systems will also be underprojected.
Note that UPlan is in the Connected Land Use Models category, in its application
in Sacramento with a travel model, which we describe below. Equilibrium Allocation
land use models are a type of model in the Complex Land Development Models category
in the FHWA typology and are used in several regions in the U.S.
We believe that all MPOs should adopt land development models of some sort.
The advantage of taking an overview of these models is so MPOs can see that they can
Travel Model/Land Use Model Integration Matrix
Travel Models
Land Use
Models
Trip-Based Models
Standard
Enhanced
Tour
Complex
Aggregate
Simulation
Activity
Aggregate
Simulation
Stand Alone
Factored
Judgement
New
Hampshire
San Joaquin
Policy+Trends
Allocation
SACMET
Rule-Based
Allocation
Connected
Boise
Fresno
Edmonton
SACMET +
UPLAN
Equilibrium
San Diego
Allocation (e.g.
DRAM)
Puget Sound
San Francisco
Bay Area
Atlanta
Santiago
Portland
(now)
Integrated
Market-Based
Allocation
Aggregate
Economic
(Input/Output)
San Francisco
County
Portland(soon)
Sacramento
(MEPLAN
test)
London
Oregon
Statewide
Disaggregate
Economic
Microsimulation
start with a simple Rule-Based Model, such as UPlan, and then advance to a more
complex model type as they gain expertise and gather more data. This table also shows
that agencies can move to the right, improving their travel models, or they can move
downward, improving their land use modeling, first. As the errors from not forecasting
land development changes can be substantial (Rodier 2001), it seems that MPOs should
advance their land use modeling, at least to the Rule-Based level or the Equilibrium
Allocation level, before improving their travel models to account for trip tours or
household activity allocation.
Once an agency decides which general type of model they wish to develop, they
can make use of another recent review, in which the USEPA suggests criteria that MPOs
can use to select a specific land use model, such as policy relevance, cost, data
requirements, and accuracy (Projecting Land-Use Change 2000).
IV. REVIEW OF GIS-BASED URBAN MODELS
The main applications of GIS have been as “spatial data models” used by
planning and natural resource agencies to organize and display spatial data. Recently,
however, professionals in these fields have begun to recognized the potential for GIS to
also be used as a “spatial process model,” including urban growth models (Heikkila
1998).
Many GIS urban models use suitability criteria, in one form or another, for the
selection of developable sites. Suitability ratings are generally determined by proximity
to infrastructure and services. The higher a suitability rating, the more attractive a site is
for development. The models also have the capability to exclude development from
restricted areas such as public lands, steep slopes, protected habitats, etc. This rating of
the landscape is applied to disaggregated units of land resulting in site-specific
allocations. With the overlay capabilities of GIS, various allocation patterns can be
compared to natural resources data to assess potential environmental impacts, as well as
economic losses resulting from natural disasters such as flooding or wildfires.
The models discussed below are included as examples of the various forms of
GIS-based urban growth models currently available.
A. California Urban Futures Model (CUF)
CUF was one of the first GIS-based urban growth models used to simulate
regional and subregional growth and relies on vector (polygon-based) data for its
analysis. A housing market is modeled where demand is a function of population growth
(calculated exogenously) and land supply is determined by spatial representations of
factors such as general plan land use categories, current land uses, slope, wetlands,
agriculture lands, land development costs, service costs, and jurisdictional boundaries.
These factors are combined to supply the model with the geometry, location, and land
attributes of Developable Land Units (DLU). Each DLU polygon is also evaluated by its
potential profitability, which is determined by subtracting all the development costs from
the sale price of a new home (all costs and prices are exogenous). The model allocates
population to DLUs in four steps:
1. All DLUs are ranked by their potential profitability;
2. DLUs that are unsuitable for development, based on their attributes, are
dropped;
3. Remaining DLUs are sorted from high to low profit potential; and then
5
4. Future population is allocated to DLUs according to density rules set by the
user. DLUs with the highest profitability are “filled” first and then
subsequently lower profitability cells are “filled” until all of the future
population has been allocated (i.e., the market is cleared).
It is assumed that developers are price-takers with respect to the price of new
homes and raw land. It is further assumed that developers want to build on sites that have
the highest profit potential. Once all allocation has occurred, the model then annexes the
new urban lands to incorporated cities or forms new cities from new development that is
noncontiguous to existing incorporated area (Landis 1994; Landis 1995).
The model does have its shortcomings, such as only simulating residential land
consumption. CUF also relies on exogenously specified land prices; therefore,
development spills over jurisdictional boundaries rather than possibly becoming denser in
reaction to high land prices. Nonetheless, CUF has made contributions to the field of
urban modeling. It operates on individual polygons of land rather than zones, making it
fairly disaggregated. It also incorporates the influence of private land developers on land
use changes. Further, it is based in GIS, making it a tool available to many planning
agencies (Landis 1994). Because it requires data for raw land prices, construction costs,
site improvement costs, service costs, development fees, and other development costs, it
is somewhat data-hungry and costly to apply. The model cannot represent infill, because
of the lack of a spatially detailed data layer for existing urban land uses for the base year.
B. Cellular Automaton Model
This urban growth model is based on the cellular automaton construct used to
simulate organic growth and was developed from the same framework Clarke used for a
wildfire model. The model relies on GIS raster (cell-based) data of urban extent, slope,
transportation networks, and protected lands. Cells representing the urban extent for a
particular year are used as the “seeds” of future development. The model then randomly
looks for cells to urbanize. Each cell selected is evaluated in terms of the spatial
properties of surrounding cells, which are determined by:
1. A Diffusion Factor – how development will be dispersed and its movement
outward through the road network.
2. A Neighbor's Coefficient – the level of connectivity to existing urban cells
needed to continue outward expansion or develop infill sites.
3. A Breeding Coefficient – the potential for a cell to become a seed that will
attract new growth.
4. A Spread Coefficient – a control of how much contiguous outward expansion
occurs.
5. A Road Gravity Factor – used to attract development along roads.
6. A Road Weight – a set of weights used to represent distance from a road,
where cells closer to a road have a higher weight.
7. A Slope Resistance Factor – used to discourage development of steep slopes.
The model has been designed to be self-modifying, using another set of
development rules:
1. When the rate of growth exceeds a critical value, the diffusion, breeding and
spread factors are increased.
2. When the rate of growth falls below a critical value, the diffusion and spread
factors are reduced.
6
3. The road gravity factor is increased as roads are added to the network.
4. As the developable land decreases, the slope resistance factor decreases.
5. As development moves up onto steeper slopes, the spread coefficient
increases to encourage development in flatter areas.
If the randomly chosen cell is a suitable site, the model then decides whether or not to
convert that cell to “urban” subject to a set of probabilities (Clarke, et al. 1996a and
1996b).
In order to calibrate Clarke’s model, it is necessary to have several base years of
data so that the factors/coefficients can be adjusted to replicate the growth patterns seen
through time. For example, in the application of the model to the San Francisco Bay
area, Clarke started with a raster image of the urban extent in 1900 as the seed and then
grew the urban area. Rasterized maps of urban extent for 1940, 1954, 1962, 1974, and
1990 were used for model calibration. The development factors were adjusted until a
reasonable fit to the historical data was found. Calibration appears to be quite time
consuming and tedious, as two phases are needed. First, a “visual” phase is conducted to
determine an appropriate range of values for each parameter. Then, numerous runs are
executed until reasonable goodness-of-fit statistics are achieved (Clarke, et al. 1996a).
Future urban development is then projected as a continuation of the past growth patterns
used during the calibration.
Clarke’s model lacks the ability to distinguish activity types as it operates on
simple “urban” and “non-urban” designations. The model also has rather poor resolution
as allocations are assigned to relatively large 300 m cells. It runs in the UNIX
environment and requires a tremendous amount of spatial data. The model also has
neither coherent economic theory nor a behavioral component to help understand its
results. However, the model is probabilistic, it runs quickly, and can be applied to any
region with the necessary data.
C. California Urban Futures Model-2 (CUF-2)
In CUF-2, significant changes were made, resulting in a cell-based model that
uses regression analysis to determine land conversion probabilities. The idea is similar to
Clarke’s model in that Landis uses data for past land use patterns to predict how future
changes will occur. The endogenous population growth model now projects employment
as well as households in ten-year intervals. DLUs take the form of one-hectare (100 m)
cells, rather than polygons, and are constructed using similar land use and geographic
factors as before. CUF-2 differs from Clarke’s model in that historical data are used to
derive probabilities of various land use changes in a cell, estimated on nearby site and
community characteristics such as population and employment growth, proximity to job
centers, proximity to transportation facilities, etc. These probabilities are assigned to
each cell and then the allocation routine uses these probabilities to either develop or
redevelop the cell according to its potential profitability. The land uses can "bid" against
each other in the choice model formulation, but there is no economic basis for the
"bidding."
Like Clarke’s model, CUF-2 has the advantage of being probabilistically based.
However, the land use change probabilities are estimated on the differences observed
between 1985 and 1995 land use data sets. Therefore, the probabilities are estimated on
only two years of data and changes in future development patterns are assumed to be
7
similar to those changes that occurred between 1985 and 1995. The authors state that the
applicability of the results to other time periods is "unclear" (p. 824).
The model has inconsistent signs for many coefficients among the counties,
making the land use change probabilities difficult to interpret. This problem may reflect
the model's lack of a theoretical basis in urban economics. The goodness-of-fit measures
for many counties are poor and highly varying across counties by specific land use
change type. Many are quite poor for changes to specific land use types (pp. 803-821).
The authors overall assessment is that these models are "generally less capable of
predicting which specific sites will shift land use" (p. 824, emphasis theirs).
This project presents useful data and analysis on land use change, but the different
coefficient signs across counties and the low power of the models for each type of
specific land use change tell us that such a model type may not be very useful for
projecting future land uses. In addition, the model did not use local land use plans as an
attractor or constraint and so would probably not be accepted and used by local officials.
D. Other Models
Some GIS models have used accessibility data from a travel model as part of the
GIS development attraction function. The most extensive set of these that we are aware
of is the work by Marshall and Lawe, then at Resource Systems Group, Inc. in Vermont.
They have linked a Lowry-type land allocation model to typical regional travel models in
the Burlington, Vermont region, the seacoast region of Maine/New Hampshire/
Massachusetts, and the Tampa Bay region. Generalized accessibility is used, which is
based on composite impedance for all modes. The other typical data layers include slope,
soils, sewers, existing land uses, and protected lands. Some of the models are menudriven (Marshall and Lawe 1996). No standard software is available to perform this
integration of travel models with the GIS land allocation model.
The new version of the Index model, called Smart Growth Index and supported by
the USEPA is a similar model. It is now going through testing in some regions in the
U.S. It will interface with the Tranplan, Minutp, and Viper travel model systems
software. Initial test versions are available through the Urban and Economic
Development Division at EPA, but future versions will be proprietary and leased from the
developer, Criterion Planners of Portland, Oregon (url: www.crit.com). Cell size ranges
from 2 to 40 ha and any set of land uses can be projected. This model requires input data
for the base year including residences by type and employment by type, by cell or
polygon. The model allocates land uses with decision rules based on the user-set criteria
and weights. This model integrates travel modeling with GIS land use allocation. It is not
clear how this model differs from the work done by consulting firms in the past, on an ad
hoc basis, such as in the models developed by Marshall and Lawe. Smart Growth Index
appears to be difficult to set up and run in conjunction with an agency's regional travel
model.
The available model closest to UPlan seems to be the WhatIf? model (Klosterman
1999). It is a rule-based model, running in ArcView on polygons. The user defines the
rules for attraction for, and constraints to, development. WhatIf? can operate on land use
types as defined by the user. Polygons are 0.25 ha or larger, generally. Redevelopment
is represented, but only crudely. The model produces several reports and has easy-to-use
dialog boxes. The model is proprietary and the consultant's services must be purchased,
to help apply the model.
8
V. THE UPLAN MODEL
A. Model Design Objectives:
Rule-based land use models are a good method for MPOs and counties to start
with, in that most agencies now have a GIS staff. In addition, this class of model uses
datasets that are generally available. We believe that a GIS-based urban model must
project several land use types, in grid cells that roughly match the development parcel
sizes. Grid (raster) data is preferred over vector data also because model runtime is
minimized. At least three residential densities must be represented, in addition to
industrial and two densities of commercial land uses. This helps to identify fiscal, runoff,
water quality, and habitat impacts accurately, in later analyses. The model need not be
strictly calibrated on historical data because its intended use is for long-range scenario
testing. However, it relies on fine-grained grid data that represent existing urban land
uses, local general land use plans, and all other relevant natural and built features that
define the model. It must be deterministic and rule-based, so as to be transparent to users
and to replicate scenarios. The allocation rules must simulate land markets, broadly.
Most importantly, the model must be inexpensive and be applicable to counties,
metropolitan regions, watersheds, and bioregions. If all required digital spatial data are
availble, UPlan can be applied in a few weeks by an ArcView user.
B. The Role of Models:
We view models as tools for consensus building. Models may be used for:
1. Analysis of past and present spatial patterns of phenomena (with maps and descriptive
statistics),
2. Projection of the most likely future patterns of these conditions, and
3. Prescription of desired future conditions and requisite policies (and testing of these
policy sets).
The background studies are more important than generally thought. Analysis and
projection can help to identify common ground among user groups, as they come to
understand and accept current problems and future likely problems. Regarding
prescription, we believe the role of modeling to be one of clarifying value choices,
especially the values that are traded off when one selects any particular policy set. In
order for models to help us to understand these tradeoffs, the models must be complex
enough to represent a great variety of social, economic, and environmental phenomena,
but simple enough for citizens to run scenarios.
So, models allow us to learn together about the urban region and to test
prescriptive concepts. Models can greatly facilitate bargaining, by bringing all interest
groups into the planning process and allowing the quick testing of the ideas of all
participants. Models that run in a few minutes or less on PCs, can be used in meetings,
aiding a detailed discourse.
Models can be useful in all these roles, because they are systematic assemblages
of our assumptions about how the world works. They provide a consistent framework for
our discussions and analyses. Scenario testing models do not provide answers, they
just illustrate various points of view and help rank order scenarios. Since they
employ graphical outputs, models can greatly help get interest groups to meet together
and bargain, because of the evocative methods of analysis and portrayal used.
9
C. Model Structure
UPlan uses any year as its base year and allocates the increment of additional land
consumed in future years. This increment of urban growth is in discrete land use
categories: Industrial, High-Density Commercial, Low-Density Commercial, HighDensity Residential, Medium-Density Residential, and Low-Density Residential. The
user can designate any land use types desired, however. In some applications, we match
the employment categories to those of the MPO, such as Manufacturing, Retail, and
Other. Sometimes, we define a Residential Very-Low category, if there are a lot of verylarge-lot residences.
County or regional land consumption is calculated endogenously, based on userspecified assumptions. UPlan is built with a module that allows the user to input
demographic and land use density factors that are converted to hectares (ha) of land
consumed for each land use. The conversion starts with population projections for
counties or the entire region. The user is then prompted to specify the demographic and
land use characteristics he or she would like to test. To determine ha needed for future
housing, the user specifies persons per household, percent of households in each density
class, and average parcel size for each density class. A similar conversion is used to
determine ha of land consumed for industry and commerce and uses workers per
household, percent of workers in each employment class, and average land area per
worker. These calculations produce a table of land demanded, for each land use type,
from which the model operates its land allocation routine. In general, the input values for
baseline cases are derived from recent experience in each jurisdiction. Overall gross land
consumed per person can be derived from historical data, such as the USDA Natural
Resources Inventory data used in Fulton (2001).
In standard applications, it is assumed that development occurs in areas that are
attractive due to their proximity to existing urban areas and transportation facilities, such
as freeway ramps. It is also assumed that the closer a vacant property is to an attraction,
the more likely it will be developed in the future. Following this assumption, each
development attraction (described below) is surrounded by user-specified buffers. The
user can designate the number and size of the buffers and assign an attractiveness weight
to each buffer. Buffer specifications are applied to each of the attraction grids and then
the grids are overlaid and added together to make a composite Attraction Grid.
There are areas where development cannot occur, which are called exclusions.
Exclusions include features such as lakes and rivers, public open space, existing built-out
urban cells, and other such areas where development cannot occur. There are also other
features that the user can choose to exclude, such as sensitive habitats, 100-year
floodplains, and farmlands. Once the user decides which features are to be excluded, the
model adds together the various exclusion grids to generate a “Mask.” Once the
Attraction Grid and the Mask Grid are generated, the model overlays the two grids and
attraction cells that fall under the mask are converted to “no data” cells, thereby removing
them from possible development allocations. This process creates the Suitability Grid,
which becomes the template for the allocation of projected land consumed in the future.
The Suitability Grid is overlaid with a grid of the 20-year General Plan land use
map for the region enabling the model to further isolate areas which are suitable for each
of the various land use categories that are allocated. The model is then ready to allocate
projected ha of land consumed in the future.
10
The model allocates future development starting with the highest-valued cells. As
the higher-valued cells are consumed, the model looks for incrementally lower-valued
cells until all ha of projected land consumption are allocated. The model does this for
each of the discrete land use categories and the user can decide the order in which land
uses are allocated. By default, the model starts with industry, then proceeds to highdensity commercial, high-density residential, low-density commercial, medium-density
residential, and finally low-density residential. This is roughly the order in which land
uses can outbid each other.
This attraction method does not apply to low-density residential, which is
randomly allocated throughout available rural areas to represent the prevalent
noncontiguous pattern of exurban rural development. Because the allocation is random,
low-density residential does not use the Attraction Grid to find the best locations;
however, the Mask Grid does apply. While all other land uses are allocated in 50 meter
grid cells (≈ 0.25 ha), 200 meter grid cells are used for low-density residential to
represent average parcel size (≈ 4 ha). The user can change the grid cell sizes, if desired.
One can generally calibrate the location of residential development by using the
Census block population data for 1990 and 2000, both in the 2000 blocks. One can
purchase these change data inexpensively from GeoLytics, Inc., and they are especially
useful for determining the location and percentage of growth for Low-Density
Residential. This land use consumes over half of all land, in many jurisdictions.
D. Data Requirements
Most of the data listed below would typically be available in any region where
UPlan is applied.
Attraction Grids: 1. Freeway Ramps, 2. Highways, 3. Major Arterials, 4. Minor
Arterials, 5. Cities, and 6. Passenger Rail Stations. For industrial allocation only, we also
use 7. Airport and 8. Port.
Exclusion Grids: 1. Land Use Plans; 2. Rivers (A user-specified distance buffers
the rivers before they are added to the Mask Grid. This precludes development from
occurring too close to waterways.); 3. Lakes (buffered); 4. Vernal Pools (seasonal
wetlands; buffered); 5. Floodplains; 6. Slope (Steep slopes are used to mask out areas that
are too steep to develop, and moderately steep slopes are used as a discouragement factor
for areas that remain. The discouragement factor works by dividing the sum of the
Attraction Grid weights by values >1, taken from a lookup table.); 7. Public Lands; 8.
Existing Urban (This grid is ofter constructed using satellite data. This layer can be
corrected and updated with parcel data, where such data exist.); 9. Permanent Open
Space; 10. Farmlands (The “exclusive agriculture” designation in a local land use plan.)
VI. EXAMPLE PROJECTS
We review some of our early projects, to show the versatility of UPlan.
A. UPlan by Itself: Espanola, New Mexico
UPlan was used as part of a community outreach effort initiated by the University
of California in the region surrounding the Los Alamos National Laboratory in New
Mexico. The study was focused on the City of Espanola (population 10,000), which is
roughly 30 km north of Santa Fe and 20 km east of Los Alamos. Espanola has begun to
see growth that has spilled over from Los Alamos as that community and the Lab
continue to expand. The Sustainable Communities Consortium at the University of
11
California at Davis was hired to help the region plan for future growth. The Consortium
employed UPlan to assist in forecasting where future growth might occur and what
impacts might result from this urbanization. The study area is about 45 km by 45 km and
was done in 25 m grid cells.
Data for the region were provided by the Earth Data Access Center (EDAC) at the
University of New Mexico at Albuquerque. EDAC possessed all of the topographical
and other base map data needed, but did not have the required land use data, which were
assembled by the Consortium. Staff used aerial photographs and paper maps provided by
the City of Espanola to create coverages of general plan and zoning designations and to
describe existing urban development at the parcel level. Since the staff had only recently
been introduced to the GIS software, the process of digitizing and preparing these
coverages for use in the model required nearly 3 person-months.
While these data were being prepared, other staff members prepared the EDAC
data for use in the model. The EDAC data needed some processing such as reprojecting,
clipping down to the study area, joining and dissolving adjacent coverages, and
reselecting and reclassing attributes so as to be compatible with the model. This
processing took a little over a person-month to complete, again with inexperienced
personnel. The processing of all data was assisted by the authors, who guided
Consortium staff on structuring the data to work with the model. Roughly one week of
our time was needed to guide data preparation.
Once the model was running, the Consortium staff was able to operate it with
little assistance from the authors. The staff became familiar with the operation of the
model and its output within 2-3 days. They were then able to test a wide variety of
planning options for the Espanola region. The group produced several scenarios for two
future years each, testing policies for different densities, degree of infill development,
changes in county land use plans, and allowing development in various adjacent pueblos.
Overall, the implementation of UPlan on the Espanola region took four months.
However, much of this time was spent familiarizing the Consortium staff with the GIS
software. The staff estimates that given their level of knowledge today, they could
implement the model in 1.5 person-months, if no digitizing was needed. Clearly, the ease
of implementing the model is dependent on the availability of data and one’s familiarity
with ArcView (and ArcInfo, to create data layers).
Other examples of using UPlan by itself include a grad student modeling the
effects of a new freeway in the Salt Lake City region. This project was completed in
about two weeks of effort, as most data layers were available. In another study, we have
projected the effects of the proposed Foothill Freeway on urbanization in the San Joaquin
Valley of California. We also evaluated the effects of proposed high-speed rail lines on
development patterns. Further, we have projected the effects of urban growth on
agricultural land availability in the San Francisco Bay area and the Central Valley. A UC
Berkeley research group is examining the effects of urban growth and vineyard
development on oak woodland habitats in Sonoma County, California, using UPlan and a
vineyard expansion model.
B. UPlan with A Travel Model: Sacramento, California
The most recent test application of UPlan is at the Sacramento Area Council of
Governments (SACOG), the MPO for the Sacramento region, where it has been linked to
SACOG’s Minutp-based travel demand model, SACMET. UPlan is being used to test
12
changes in development location due to changes in accessibility. In each analysis zone,
accessibility is determined by a series of calculations in SACMET. First, a representative
income class is chosen. Their trip profiles are used to determine the number of trips that
leave a zone (origins) and the number that come into a zone (destinations). The logsum
of choice coefficient for each mode is then calculated. Each logsum is multiplied by the
number of trips for that mode and then all logsum-trip products for all origin-destination
pairs are summed. The result is a measure of accessibility for both origin and destination
by zone. These accessibilities are fed into UPlan as a measure of attractiveness, where
higher accessibility makes a zone more attractive for development.
In this application of UPlan, accessibility is the only measure of attractiveness
because SACMET captures the conditions of the transportation network that are normally
represented by the various roadway and transit attractiveness grids. For the allocation of
residential development, UPlan uses the origin accessibility to determine if the zone is an
area that people can travel from easily. To test whether a zone is attractive for nonresidential land uses, UPlan uses destination accessibility to measure how easy it is to get
to a business in that zone. When UPlan begins to allocate development, it looks to the
most accessible zones first and then moves on to less accessible zones until is runs out of
development to allocate in that period. As UPlan builds out the most attractive zones,
SACMET calculates more trips in those zones and hence, a rise in congestion. This
reduces the level of accessibility in those zones and causes UPlan to look to other, now
more accessible zones, as suitable development locations. The models are run three
times in sequence to get to the horizon year, 2025.
C. UPlan After An Urban Economic Model
We have also used UPlan as a means of disaggregating the land use projections
made by an integrated urban model, Tranus. In this application, we ran Tranus on our
Sacramento region datasets and performed projections to the year 2020 for four
transportation scenarios (Johnston and de la Barra 2000). We then ran UPlan on each of
the 53 zones in the Tranus model, individually, to disaggregate the land uses spatially.
In an academic comparison, we also ran UPlan by itself on the Sacramento region
and compared these land use maps with those from the exercise where we ran Tranus and
disaggregated its zonal projections with UPlan. The maps compared favorably, that is,
UPlan produced an urbanized layer that overlapped the Tranus-UPlan layer to a fairly
high degree. Using the Lee-Sallee index (Is = (AB)/(AB)), we found over 60%
agreement between Tranus-UPlan and the stand-alone version of UPlan.
D. UPlan Linked to Urban Impact Models
We have developed rudimentary impact models, to demonstrate their use in the
Sacramento region. All of these models were quite easy to implement in ArcView.
1. The Costs of Flooding model uses per-ha real estate assessed values for the six UPlan
land use types and applies these to the ha of new development within the 100-year
floodplain in each scenario being considered. The values are then factored down, since
only buildings are damaged, not land. These values are factored down again, according to
the percentage loss projected in the floodpool. Currently, we assume an average
floodpool depth, but intend to refine this with actual depth data.
2. Costs from Wildfires are projected with a model developed by the California
Department of Forestry and Fire Protection that runs in ArcInfo. A layer for Probability
13
of Fire Occurance was created. Development patterns are evaluated by density and
location. Loss of structures is translated into loss of property value.
3. Local Service Costs are projected, using national data (Frank 1989) for the cost of each
service type, by distance from current services. These data are applied to the ha of new
development of each UPlan land use type, by distance from existing urban land uses. It
would be more accurate to map all actual water and sewer main lines, but we do not have
those data layers and they are not digital in most localities. We attempted to get local data
on service costs for roads, sewer, water, police, and ambulance, but many agencies told
us that costs were the same, regardless of distance from existing facilities. They seemed
to be confusing (their presumably flat) service charges with (by necessity, variable)
actual service costs, and so we used the national data.
4. The Erosion Impact model was based on factors used in the Universal Soil Loss
Equation (USLE). These factors are soil erodibility, slope steepness, and precipitation.
We obtained soil erodibility factor values from the STATSGO GIS database. Slope
steepness was derived from a 30m DEM, then transformed into a slope steepness factor
using a single, continuous function for slope steepness influence on soil loss (Nearing
1997). Annual average precipitation data came from the PRISM GIS database (Daly
1997) in a grid of approximately 1km cells; these data were then resampled to 50m grid
cells using a cubic transformation. All data were converted into 50m grid cells. We then
used ArcView Model Builder to combine these three factors using a multiplicative
arithmetic overlay and used the result as a measure of erosion impact.
5. Habitat Damage. We created a score for wildlife habitat quality using the Ecosystem
Management Decision Support program (EMDS; Reynolds 1999). The analysis looks at
two main issues, wildlife habitat presence and habitat quality, to determine the score for
wildlife habitat quality.
For the purpose of this analysis, we defined wildlife habitat presence as a
combination of aquatic habitat integrity (distance to stream and aquatic diversity areas),
wildlife occurrences (WHR richness and natural diversity database), and terrestrial
habitat integrity (hardwoods, vernal pools, grasslands, wetlands, significant natural areas,
and expert opinion). These data were converted into 50m grid cells and input into an
EMDS knowledge base in this fashion.
EMDS links the GIS program ArcView 3.2 (ESRI 2000) with the knowledge base
Netweaver (Saunders 1989; Saunders 1990). Netweaver is an object-based hierarchical
network, with nodes based on fuzzy logical relationships. EMDS provides Netweaver
with the necessary GIS base data for combining together using fuzzy logic rules to
determine degree of truth for an assertion (Reynolds et al. 1996; Reynolds 1999). These
assertions have “sub-assertions” (dependency networks) that are nested within the
overall network. Eventually, these sub-assertions connect to base-data that address a
single assertion.
Data Needs Identified from Projects Completed
From our experiences to date, we conclude that satellite data are not accurate
enough to be used to represent Existing Urban in the base year. Countywide parcel
datasets are needed for both spatial accuracy and categorical accuracy. Satellite data can
help to identify where buildings and pavement are on large rural parcels, however.
California does not make parcel data available and so researchers and planners are
subject to county-specific data policies. Most counties in California have parcel data, but
14
many counties will not give it to researchers for free. Multispectral airphotos at 1m and
2m resolution are adequate to identify Existing Urban land use types, if interpreted for
this purpose, as raster data. These data are available generally only for urbanized areas,
not rural ones, unless countywide habitat planning is underway. Digital local land use
plans are generally available now, although interpreting a large number of city and
county plans into one coherent set of land use types can be time-consuming, especially if
the documentation is poor.
VII. CONCLUSIONS
We believe UPlan is a reasonable starter model for MPOs, COGs, counties, and
state transportation agencies to use in modeling land use changes. It will project the
footprint of development, which permits the analysis of several kinds of impacts. It will
also allow the agency to iterate the land use model with their travel model, in 5- or 10year steps. A simple rule-based model will not, however, give land market data, such as
floorspace rents. Nor can such models represent redevelopment, except manually. It can
be implemented in a month or two, if most or all data layers are available in GIS. If the
agency wishes to progress on to a more sophisticated model type, the UPlan data and
model will both be useful later for suballocation of growth projections, display of model
outputs, and for other purposes. Current improvements being made are a better user
interface, so all rule changes can be made without recoding, and more flexibility in user
rules. Future improvements include coding in VBA for ArcGIS8.
ACKNOWLEDGEMENTS
We would like to thank USDOT, Caltrans, and USDA for funding the development of
UPlan. Also, we thank the staff at SACOG for their assistance with data in our initial
application.
REFERENCES
Clarke, Keith, Stacy Hoppen, and Leonard J. Gaydos. 1996a. Methods and Techniques
for Rigorous Calibration of a Cellular Automaton Model of Urban Growth. URL:
http://geo/arc/nasa.gov/usgs/clarke/calib.paper.html.
Clarke, Keith, et al. 1996b. An Analysis of Human-Induced Land Transformations in the
San Francisco Bay/Sacramento Area. URL: http://geo/arc/nasa.gov/usgs/
WRR_paper.html.
Clarke, K.C. and S. Hoppen. 1997. A Self-Modifying Cellular Automaton Model of
Historical Urbanization in the San Francisco Bay Area. Environment and Planning: B 24,
pp. 247-261.
Dagang, Deborah A. 1995. Transportation-Related Land Use Strategies to
MinimizeMotor Vehicle Emissions: An Indirect Source Research Study, Final Report.
Prepared for the California Air Resources Board, Contract No. 92-348.
15
Daly, C., G. Taylor, and W. Gibson, 1997, The PRISM Approach to Mapping
Precipitation and Temperature, 10th Conf. on Applied Climatology, Reno,NV, Amer.
Meteor. Soc., 10-12.
Dunphy, Robert T. and Kimberly M. Fisher. 1994. Transportation, Congestion, and
Density:New Insights, Preprint. Transportation Research Board: Washington D.C. Paper
No. 94-0904.
Expanding Metropolitan Highways: Implications for Air Quality and Energy Use.
Transportation Research Board, National Research Council, 1995. Washington, D.C.
Frank, James E. 1989. The Costs of Alternative Development Patterns. Urban Land
Institute, Washington, D.C.
Fulton, William, Rolf Pendall, Mai Nguyen, and Alicia Harrison. 2001. Who Sprawls
Most? How Growth Patterns Differ Across the U.S. Brookings Institution, Center on
Urban & Metropolitan Growth. Washington, D.C. July.
Hanink, Dean M. and Robert G. Cromley. 1998. Land-Use Allocation in the Absence of
Complete Market Values. Journal of Regional Science 38, 3: 465-480.
Harrison, Britton. 1985. Urban Simulation Models in Regional Science. Journal of
Regional Science 25, 4: 545-567.
Heikkila, Eric J. 1998. GIS is Dead; Long Live GIS! Journal of the American Planning
Association 64, 3: 350-360.
Johnston, Robert A. and Tomas de la Barra. 2000. Comprehensive Regional Modeling for
Long-Range Planning: Integrated Urban Models and Geographic Information Systems.
Transp. Res.: 34A, pp. 125-136.
Land Use Impacts of Transportation: A Guidebook. 1999. NCHRS Rept. 423A,
Transportation Research Board, Washington, D.C.
Landis, John D. 1995. Imagining Land Use Futures: Applying the California Urban
Futures Model. Journal of the American Planning Association 61, 4: 438-457.
Landis, J.D. 1994. The California Urban Futures Model: A New Generation of
Metropolitan Simulation Models. Environment and Behavior: B 21, pp. 399-420.
Landis, J. and M. Zhang. 1998. The Second Generation of the California Urban Futures
Model. Part 1: Model Logic and Theory. Environment and Planning: B 25, 5: 657-666.
Part 2: Specification and Calibration Results of the Land-Use Change Submodel.
Environment and Planning: B 25:6, pp, 795-824.
16
Lowry, Ira S. 1964. A Model of Metropolis. RM-4035-RC. Santa Monica: Rand
Corporation.
Marshall, Norman L. and Stephen J.C. Lawe. 1996. Land Use Allocation Modeling in
Uni-Centric and Multi-Centric Regions. Paper delivered at the Transportation Research
Board Annual Meeting, Washington, D.C., January.
Miller, Eric J., David S. Kriger, and J. Douglas Hunt. 1998. Integrated Urban Models for
Simulation of Transit and Land-Use Policies. Final Report. TCRP Project H-12. USDOT.
Washington, D.C. September.
Nearing, M. A. 1997. A single, continuous function for slope steepness influence on soil
loss. Soil Sci. Soc. Am. J. 61: 917-919.
Paterson, J. 1977. Transport and Land-Use Determinants of Urban Structure In D.
Hensher, Urban Transport Economics, Canbridge University Press, New York, 175-198.
Portland Metro. 1994. Region 2040: Decisions for Tomorrow, Metro 2040 Growth
Concept.
Projecting Land-Use Change: A Summary of Models for Assessing the Effects of
Community Growth and Change on Land-Use Patterns. 2000. USEPA Rept. EPA/600/R00/098. September.
Puget Sound Regional Council. 1990 and 1995 update. Vision 2020: Growth and
Transportation Strategy for the Central Puget Sound Region.
Reynolds K. 1999. Netweaver for EMDS user guide (version 1.1); a knowledge base
development system. Gen. tech. Rep. PNW-GTR-471. Portland, OR: U.S. Department of
Agriculture, Forest Service, Pacific Northwest Research Station.
Reynolds K, Cunningham, P, Bednar L (and others). 1996. A design framework for a
knowledge-based information management system for watershed analysis in the Pacific
Northwest U.S. AI Applications. 10: 9-22.
Rodier, Caroline J., Robert A. Johnston, and John E. Abraham. 2001. A Comparison of
Highway and Travel Demand Managment Alternatives Using An Integrated Land Use
and Transportation Model in the Sacramento Region. Submitted to the Transportation
Research Board for presentation at the Annual Meeting, Jan. 2002.
Sacramento County. 1993. Sacramento County General Plan Land Use Element.
Sacramento, CA: Sacramento County.
Saunders, M.C., Rajotte, E.G., Travis, J.W. 1989. The use and significance of expert
systems in small fruit IPM. In: Bostanian,l N.J., Wilson, D.T., Dennehy, T.J., eds.
17
Monitoring and integrated management of arthropod pests of small fruit crops. London:
Intercept.
Saunders, M.C., Coulson, R.N., Folse, J. 1990. Applications of artificial intelligence in
agriculture and natural resource management. In Kent, A., Willeams, J., eds.
Encyclopedia of computer science and technology: volume 25
Simmonds, David. 1995. Available Methods for Land-Use/Transport Interaction
Modelling. David Simmonds Consultancy. March, 1995.
Wegener, Michael. 1994. Operational Urban Models: State of the Art. Journal of the
American Planning Association 60, 1: 17-29.
18