Citation
Exurban residential development and the attraction of natural amenities

Material Information

Title:
Exurban residential development and the attraction of natural amenities an agent-based model
Creator:
Yin, Li
Place of Publication:
Denver, CO
Publisher:
University of Colorado Denver
Publication Date:
Language:
English
Physical Description:
xi, 121 leaves : illustrations ; 28 cm

Subjects

Subjects / Keywords:
Housing development -- West (U.S.) ( lcsh )
Land use -- Planning -- West (U.S.) ( lcsh )
Land settlement patterns -- West (U.S.) ( lcsh )
Housing development ( fast )
Land settlement patterns ( fast )
Land use -- Planning ( fast )
United States, West ( fast )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Bibliography:
Includes bibliographical references (leaves 107-121).
Thesis:
Design and planning
General Note:
College of Architecture and Planning
Statement of Responsibility:
by Li Yin.

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
|Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
57543252 ( OCLC )
ocm57543252
Classification:
LD1190.A735 2004d Y56 ( lcc )

Downloads

This item has the following downloads:


Full Text
EXURBAN RESIDENTIAL DEVELOPMENT AND THE ATTRACTION OF
NATURAL AMENITIES:
AN AGENT-BASED MODEL
by
Li Yin
B. ARCH, Yunnan Polytechnic University, P.R.China, 1992
M.S., Asian Institute of Technology, Thailand, 1998
A thesis submitted to the
University of Colorado at Denver
in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
Design and Planning
2004
j A L j
. Mil


2004 by Li Yin
All rights reserved.


This thesis for the Doctor of Philosophy
degree by
Li Yin
has been approved
by
)_2A2m
Date
Janies O. Huff


Li Yin (Ph.D., Design and Planning)
Exurban Residential Development and the Attraction of Natural Amenities: An
Agent-Based Model
Thesis directed by Assistant Professor Brian H. F. Muller
ABSTRACT
Dispersed residential development has been a conspicuous feature of recent
population migration to the intermountain West. This low-density settlement
pattern has significant implications for land use planning issues such as loss of
habitat and agricultural land, and demands on local government for infrastructure
and services. Many of the models used to explain urban development are not
well-adapted to low-density land markets. Since the 1960s, urban economic theory
has emphasized the journey-to-work as the primary determinant of urban
locational pattern. Recent empirical research on exurban development, however,
focuses on other factors such as household preference for natural amenities and
large lots, new technology, and retirement trends. Agent-based models (ABM)
have the potential to be a powerful tool for simulating dynamics in exurban land
markets. While ABM techniques have recently been applied to a variety of
problems of land use/cover change, there have been relatively few applications of
this approach in exurban areas. This research builds on agent-based urban-rural
fringe land use models (Irwin, 1998), regression-based land conversion policy
models (Muller et al., 2002), and a pilot study of exurban land development (Yin
and Muller, 2002). I investigate patterns of exurban residential development
underway in the American mountain West, and model exurban sprawl as a product
of interactions around amenities, density, and accessibility. Simulation results
IV


suggest the interactive and dynamic exurban development model built in this
research represents the land market system at a reasonably high level of accuracy.
Exurban agent-based models also suggest opportunities for policy applications
that link natural and social models to simulate the effects of alternative planning
regimes. Such agent-based models are important because they enable planners to
consider a broader range of possible cumulative or emergent effects of land use
policies or market trends.
This abstract accurately represents the content of the candidates thesis. I
recommend its publication.
Signed
Brian H.F. Muller
v


ACKNOWLEDGEMENTS
A debt of gratitude is owed to many people who had constructive influences on
my academic development. My deepest gratitude and appreciation go to Professor
Brian H. F. Muller, the principle advisor and mentor, for his support,
encouragement, patience, and guidance that helped shape this dissertation. I am
also very grateful to my committee members, Professor William R. Travis,
Raymond J. McCall, Raymond G. Studer, and James O. Huff, for their time,
dedication, and invaluable comments and suggestions.
Special thanks go to the Director of the Ph.D. Program in Design and Planning,
Professor Willem van Vliet, for giving me the opportunity to study in the States
and his generous support to enable me to concentrate on my research. In addition,
I wish to extend my thankfulness to my fellow Ph.D. students who encouraged
and helped me, such as Sue Wolf and Tamara Laninga.
I would also like to express heartfelt gratitude to my beloved father, mother, and
brother for constant inspiration to my academic accomplishment up to this level,
for their love, encouragement, moral support, and help that made this dissertation
possible. Last but not least, my special thank goes to my husband for his patient
and support.


CONTENTS
Figures........................................................ix
Tables.........................................................xi
CHAPTER
1. INTRODUCTION:
EXURBAN DEVELOPMENT AND THE NEW WEST.................1
Exurban Development in the West..........................1
Planning Implications of Exurban Growth..................6
Objectives of the Study..................................9
Research Organization...................................13
2. LITERATURE REVIEW:
MODELING THE URBAN PERIPHERY................................15
Urban Location Theories.................................16
The Monocentric Model............................16
The Polycentric Model and Other Extensions.......17
Public Goods /Local Amenities Models.............19
Amenities and Theory of the
New West Periphery Development..........................20
Discrete Choice Land Conversion Models..................26
The Agent-based Model and Complexity Theory.............29
Applications of Agent-based Models to
Exurban Development.....................................32
Summary.................................................34
3. RESESARCH METHOD: DESIGN OF AN AGENT-BASED
SIMULATION ON THE URBAN PERIPHERY...........................37
Overall Approach........................................37
Research Questions and Hypotheses.......................38
Data Collection and Processing..........................41
Data Collection..................................42
Data Processing..................................43
Computational Framework.................................45
GIS Techniques and Algorithms....................45


Agent-based Modeling Platforms
49
Model Design............................................53
Overall Design..................................53
Interaction Framework...........................56
Design of Rules.................................59
Case Study/Validation...........................64
Variables...............................................68
4. DISCUSSION AND RESULTS: AGENT-BASED EXURBAN
HOUSEHOLD LOCATION MODEL....................................75
Theoretical Models......................................75
Model 1.........................................77
Model II........................................82
Model III.......................................86
Model Comparison................................91
Case Study/Model Validation.............................92
5. SUMMARY AND CONCLUSIONS.....................................99
Research Findings.......................................99
Planning Implications..................................102
Future Research Directions.............................106
BIBLIOGRAPHY...................................................107
viii



FIGURES
Figure
1-1 Natural Amenity Scale.............................................5
3-1 GIS Raster Overlay...............................................47
3-2 Framework of Agent-agent and Agent-Environment Interaction.......56
3-3 Components of Rule Structure.....................................60
3-4 Rules for Model 1................................................61
3-5 Rules for Model II...............................................62
3-6 Rules for Model III..............................................62
3-7 Pseudo Code of Agent-based Model.................................64
3-8 Rural-Urban Commuting Areas......................................66
3-9 Study Area.......................................................68
3-10 Examples of Viewshed.............................................71
3- 11 Viewshed Illustration............................................72
4- 1 Initial State....................................................76
4-2 Accessibility Model..............................................78
4-3 Model I..........................................................SO
4-4 Model II..........................................................84
IX


4-5 Model III......................................................89
4-6 Modeling Results from Multiple Runs............................95
4-7 Exiting Development (1981-1990 and 1981 -2000).................97
4- 8 Comparison of Neighborhood Density: Existing vs. Predicted.....98
5- 1 Wildfire Hazard in the Study Area.............................104
5-2 Wildfire Hazard Zoning Model..................................105
x


TABLES
Table
1- 1 Population Growth and Land Development in Colorado: 1960-2050.........3
1 -2 Population Growth and Land Development
in Boulder County, Colorado: 1960-2000...............................4
1 -3 Natural Amenities by Region...........................................5
2- 1 Amenities Variables: Literature......................................36
3- 1 Data Types and Sources...............................................43
3-2 Properties of Households.............................................57
3-3 Locational Preferences by Different Types of Households..............61
3- 4 Definitions of Variables.............................................74
4- 1 Some Housing Characteristics of the Study Area.......................93
4-2 Some Household Characteristics of the Study Area.....................94
xi


CHAPTER 1
INTRODUCTION:
EXURBAN DEVELOPMENT AND THE NEW WEST
Exurban Development in the West
Dispersed residential development in mountain valleys and foothills, and outside
city limits, has been a conspicuous feature of recent population migration to the
American Intermountain West. This settlement pattern, loosely described as
exurban, has a variety of planning implications related to loss of habitat and
agricultural land, as well as placing increased demands on local government for
infrastructure and services. However, many of the traditional approaches used by
urban modelers are not well-adapted to this land market environment. Agent-based
models (ABM), on the other hand, have the potential to be a powerful tool for
simulating land development dynamics in exurbia. While ABM techniques have
recently been applied to a variety of problems of land use/cover change, there have
been relatively few applications of this approach in exurban areas. This research
explores the use of a new modeling technique that uses an agent-based approach to
investigate exurban residential development underway in the Intermountain West
by simulating it as an emergent global behavior resulting from household level
evaluation of dynamic and interactive effects of natural amenities, accessibilities,
and neighbor avoidance.
A prominent feature of the contemporary American West is its dramatic transition
in demography and economy from the wild to the new West (Riebsame et al.,
1


1997) and the associated impact on ecosystems. Historically, this region was
characterized by low population and development densities with enormous
undeveloped land (Wilkinson, 1993). During the 1990s, however, the population
growth rate of the Mountain West was 25.4%, making it the fastest growth region
in the country. More than half of the counties grew faster than the national average
(Beyers and Nelson, 2000). The U.S. Census Bureau (2001) reported that
population in the West grew three times as fast as the rest of the U.S. Certain
recent census estimates that the Mountain West will capture as much as 40% of
nations overall population gain in the next quarter century, which would amount
to as many as seven million people. This rapid population growth is fueled by
economic structure changes. The economy of the West is transferring from
traditional resource extraction to a new economy based on high technology, real
estate, and recreation. Ecosystem processes and biodiversity are facing challenges
from the impacts of human-induced activities.
One of the most noteworthy characteristics of the New West is that rapid
population increases are occurring not only in urban but also rural areas. More
than 60% of rural counties in the region are gaining population faster than urban
areas (Theobald, 2000; U.S. Census Bureau, 2001). Many local communities are
confronting rural sprawl as a major issue. In fact, a consistent urban to rural
migration has taken place since the 1970s in the Rocky Mountain region
(Cromartie, 1994). Rapid population growth invigorated land development. Much
agricultural land has been converted to urban uses, especially low-density exurban
residential uses. These exurban growth dynamics in the Mountain West have
become a pressing concern locally and nationally (Benfield, Raimi, and Chen,
1999; Katz and Liu, 2000).
2


In Colorado, for example, the total number of acres used for new development
tripled, while total population only doubled from 1960 to 1990 (Table 1-1). The
most dramatic growth patterns are evident in the exurban zone. More than 65% of
the land developed during this period was used for new homes in exurbia. Within a
30-year period (1960-1990), new homes tripled the amount of low-density exurban
land to 2,269,000 acres, twice the combined land areas for urban and suburban
development. The Center of the American West predictes this trend might persist
for the next 50 years. At University of Colorado, a research team has been
mapping how land uses have changed within Colorado. Table 1-2 shows the
results for Boulder County, which are as striking as at the state and regional level
considering the amount of private land available in the county. Most of the
exurban homes sprung up near the foothills, turning much of the county into a land
of ranchettes and hillside lodges.
able 1-1 Population Growth and Land Development in Colorado: 1960-2050
Population Developed Acres
Urban and Suburban Exurban1 Total
1960 1,753,947 428,000 881,000 1,309,000
1990 3,294,394 1,176,000 2,269,000 3,445,000
2020 5,010,500 2,222,000 3,407,000 5,629,000
2050 6,208,000 2,763,000 4,636,000 7,399,000
Source: Center of the American West
1 The center of the American Wests Western Futures project (2001) defined exurban according to
the density: 1 unit per 10 to 40 acres.
3


Table 1-2 Population Growth and Land Development in Boulder County,
Colorado: 1960-2000
Population Developed Acres
Urban and Suburban Exurban2 Total
1960 74,254 39,632 4,936 44,568
1980 189,625 72,136 9,019 81,155
1990 225,339 85,645 10,960 96,605
2000 291,288 102,487 13,446 115,933
Understanding exurban development patterns requires knowledge of the processes
that push the development. The literature on exurban development suggests that it
is driven by a variety of factors including availability of public land, mountain-
based recreation, infrastructure improvements, and desire for large lot amenities
(Duane, 1999). Some researchers have stated that the rural West is attractive
because of its scenery, wilderness, wildlife, and outdoor recreation opportunities
(Johnson and Rasker, 1995; Beale and Johnson, 1998).
The Mountain West of the United States is historically wild in nature and rich in
natural amenities. The Economic Research Service of the U.S. Department of
Agriculture developed a natural amenity scale (Figure 1-1) at the county level
based on a composite measure of physical characteristics that presumably make a
place attractive to live. These measures include environmental qualities of places,
climate, and typography. Figure 1-1 illustrates that many counties in the Mountain
West have a very high level of natural amenities. The average amenity index in the
region is ranked the second highest in the U.S., following the Pacific region (Table
1-3). Boulder County, having a ranking of six out of seven, is one of the areas with
the second highest amenity index.
2 University of Colorado defined exurban as an area including all the private land outside the city
limit.
4


Figure 1-1: Natural Amenity Scale
H 1 Low Amenities
Source: USDA, Economic Research Service
Table 1-3 Natural Amenities by Region
Natural Amenity Scale across the Region
Region Average Maximum
New England 3.73 5
Middle Atlantic 3.33 4
East North Central 2.61 4
West North Central 2.76 5
South Atlantic 3.64 6
East South Central 3.40 5
West South Central 3.83 6
Mountain 4.82 7
Pacific 5.49 7
Source: USDA, Economic Research Service
5


McGranahan (1999) found that population growth in rural counties in the United
States from 1970 to 1996 was strongly correlated with this natural amenity scale.
In recent years, peoples desire for a high level of natural amenities and
recreational opportunities in the American Mountain West (Power, 1995; Masnick,
2001; Reibsame, 1997; McGranahan, 1999) stimulated a historically
unprecedented human population boom in the region. It was strengthened by
expansion of service, recreation, and information industries. The high quality of
the natural environment is the regions greatest economic asset (Rasker, 1993,
Power, 1995).
In the 1990s, five out of eight Mountain West states were the fastest growth states
in the country, Colorado being one of them. Counties with major ski resorts grew
at an annual rate of 8.5% from 1960 to 1990. In the early 1990s, population growth
rates in nearly one fifth of the counties exceeded 5% (Theobald, 2000), and ten out
of the fifty fastest growing counties in the U.S. were in the mountain areas
(Riebsame et al., 1996). Rapid population growth and amenity migration to
Colorado is marked by the conversion of land use from agriculture to low-density
residential development. From 1992 to 1997, some 271,815.92 acres were
converted (Oberman, Carlson, and Batchelder, 2000). This conversion to low-
density exurban development continues, suggesting a long-term alteration of the
natural heritage of this region.
Planning Implications of Exurban Growth
Exurban development, a fast growing component on the landscape of the
American Mountain West, has far-reaching implications for the existing land use
systems, local communities and ecological biodiversity. Under current policies
and market trends, new residential development is likely to remain at current or
6


even decreasing densities in the Mountain West. This suggests that the overall
patterns of population dispersion in the region are likely to continue if not
intensify.
While natural amenities attract people to exurban areas in the West, the resulting
land use changes challenge the effort to sustain local communities and ecosystem
health, and threaten agricultural and wildlife habitat. A host of natural and human
environmental activities have many profound impacts upon the ecosystem, but the
human induced land transformation of the current era is the most striking.
Hundreds and thousands of acres of land are converted each year from natural to
human settlements and workplace uses. Studies have shown that exurban
development affects ecosystem processes and biodiversity (Theobald, 2000).
Colorado has seen significant threats to habitat due to development pressures.
Many ranchettes, sprawling subdivisions, and recreational activities are replacing
natural habits and agricultural lands, displacing wildlife (Miller et al., 1998),
reducing the population of some species, harming environmentally sensitive areas,
and changing disturbance regimes, such as wildfire.
The flow of population migration to natural amenity-rich areas in the United States
has also created a difficult set of regulatory challenges for hazard and land use
managers in rural and local governments. For example, wildfire mitigation is
evolving into a significant public responsibility for rural and urban edge county
governments in the Western United States. The traditional public safety
responsibility of county governments puts them at the intersection of two social
and ecological trends, each with the potential for enormous public risk. On the
one hand, rapid residential development in fire-prone areas of many counties is
increasing the overall vulnerability of county residents to injury or loss of property
from wildfire. On the other hand, long-term ecological processes occurring in the
7


West, notably pest infestation and effects of fire suppression, may increase the
probability of serious fire events.
County officials describe the threat of wildfire as a kind of Sword-of-Damocles
hanging over their neighbors, careers and the fiscal viability of county
government. In addition, new frameworks are emerging for discussions between
rural counties and federal agencies, specifically FEMA and the federal land
managers in the U.S. Forest Service of the Bureau of Land Management. In short,
there is both a sense of threat in county governments and considerable concern
among wildfire planners.
These challenges are in large part a organizational problem due to high migration
rates to rural counties with historically weak land management organizations. They
also have another dimension, however, intrinsic to the character of amenity-based
real estate development. Land use policy tools can be used to direct residential
development away from conservation and hazardous areas, but such areas often
have the the most desirable building site in the context of amenity market and
attract the most demand. Exurban amenity migration is linked to localized
configurations of natural features such as views, trees and beach quality. These
local configurations tend to create spatially-complex patterns of market demand,
and over large areas, generate fragmented landscapes with highly interactive
natural and social processes. Such landscapes pose major technical challenges to
land use managers attempting to identify appropriate regulations related to hazard
and farmland and habitat conservation. Finally, natural amenities have a dynamic
character; too much nearby development tends to degrade them. Residents
frequently act to protect the quality of local amenities against increased density
through zoning and ownership patterns; both can create rigidities that interfere
with sensible hazard and conservation planning.
8


There are increasing concerns at the local and regional level about urban sprawl
spilling into exurban areas. Hundreds of cities and counties across the
Intermountain West are developing strategies to mitigate exurban development
impacts on ecosystems. It is possible that more stringent land use regulations may
emerge from this environment, but a clear trend in this direction has not been
documented. Certainly, many counties are considering more rigorous
interventions; discussions are occurring along these lines even in small, rural
counties where property rights traditions are strong.
Planners are challenged to both accommodate growth and conserve our natural
resources, and to minimize the effects on the environment as people move beyond
suburbs into the exurbs, and consume agricultural and wildlife habitat. A foremost
research need is to develop ways of understanding exurban development patterns,
to evaluate alternative planning policies that will guide development, and to
examine the consequences of development patterns for natural and social systems.
Objectives of the Study
Nelson and Dueker (1990) and Nelson (1990, 1991) defined the term exurban
based on county boundaries. Their exurban counties include some low-density,
metropolitan counties and all of the nonmetropolitan counties. While there are
many different definitions, exurban in this dissertation includes all private land
that is outside city limits in both nonmetropolitan and metropolitan counties.
As large numbers of new housing constructions are taking place in the exurban
landscape across the American Mountain West each year, forests, agricultural land
9


and wildlife habitat increasingly become threatened by urban sprawl. The rural
living style preferences of seemingly small and innocuous households may be
producing a cumulative effect that is hard to reverse. We need to protect forest and
habitat lose and fragmentation from exurban development. In this research, I
explore the application of an agent-based approach to understand exurban
residential location and trajectories of the cumulative and dispersed exurban
development in the Mountain West. An agent-based model is constructed to detect
the dynamic exurban sprawl influenced by household interaction with respect to
effects of amenities, density (neighbor avoidance and large lot development), and
accessibility.
Computer models for urban growth and policy development have a long history.
Yet, many of the traditional approaches used by urban modelers are not well
tailored to exurban land market environments. Since the 1960s, urban economic
theory has emphasized the joumey-to-work (employment center) as the primary
determinant of urban locational pattern. Residential location behavior is assumed
to be based on the trade off between transportation costs to urban centers and land
rent. However, recent literature suggests that exurban development is driven by a
range of factors including availability of public land, recreation, recent
infrastructure improvements, and desire for large lot amenities (Duane, 1999).
The drivers of low-density exurban development suggested in this literature are not
integrated in the conventional models.
Agent-based models (ABM) have a high potential to reach beyond the
conventional models by simulating interactions between households and between
households and natural amenities, and the dynamics of exurban land development
decisions and land use policy effects. A richer, more intuitive understanding of
dispersed exurban development may be reached through integrating exurban
10


households land development behavior related to amenities, density, and
accessibility. Instead of averaging social actors and their behaviors, ABMs can be
constructed to represent them with various characteristics to reflect the diversity of
traits found in social systems, such as different types of exurban households with
different locational preferences. It helps to differentiate entities and processes at
play in the exurban system while allowing the direct modeling of interactions
between them. That is, ABMs can be used to simulate the process of an exurban
resident selecting a site for development in relation to his individual wants (e.g.
preference for being close to a stream) and his interaction with other residents in
the neighborhood. A process in an ABM is directly analogous to the one in the real
world. In addition, the ABM approach is a powerful way to build a model that
more easily relates to policy decisions and integrates policy effects.
Furthermore, the agent-based approach provides a means to assess temporal and
autonomous exurban residential development decision making at the household
level and link these decisions to aggregate land use changes. Simulations can
better capture timely changing activities at the micro-level, which is important
when considering land use development because development always concerns the
growth of new activities and adaptation to these changes. In the context of exurban
development, households find out about the densities, amenities, and
accessibilities, and make land use decisions based on them. Some households
prefer to be close to other households, while others have a strong desire for more
space, natural amenities, and a sense of isolation (Davis et al., 1994; Nelson, 1992;
Riebsame et al., 1996). They respond differently to the land development activities
around them.
ABM techniques have recently been applied to a variety of problems of land
use/cover change (Parker et al., 2003); however, there have been relatively few
11


applications of this approach in exurban areas, much less applications validated
using actual data. One exception is Irwin (1998), who demonstrated that the agent-
based approach is well suited to explain land use conversion as a result of land use
externalities at the urban-rural fringe. This research builds on Irwin (1998), as
well as on regression-based land conversion policy models (Muller et al., 2002),
and a pilot study on exurban land development (Yin and Muller, 2002) in design of
an agent-based simulation of exurban land development in the Western United
States.
In this research, I build a model using the agent-based approach to explore how
exurban residential development patterns may result from interactions between
decentralized individual behaviors. In addition, I compare the static ABM
approach emphasizing accessibility factors to various ABM models emphasizing
amenities and neighbor avoidance.
I begin by using data on actual development patterns in a 16-square-mile area at
the edge of Lyons, Colorado, to look at the relationship among preferences for
amenity, density, and accessibility as a pilot study of exurban development. Lyons
is a town of about 1500 people located in a mountain foothills area in the
northwestern part of Boulder County. From this, I build three theoretical models
and validate them in the study area, northwestern Boulder County, Colorado. I
evaluate exurbanization as an interactive and dynamic process at a household level
of amenity choice, neighbor avoidance, and accessibility preference.
Exurban agent-based models also suggest opportunities for policy applications that
link natural system and social models to simulate the effects of alternative
planning regimes. This research evaluates alternative wildfire planning regimes
from the perspective of land use policy and site design. I report on applications of
12


this agent-based household location choice model to planning around wildland fire
in the study area.
Such agent-based models are important because they study micro-level
information and link it to macro level patterns. They can both inform the planning
theorist and improve planning practice in exurban areas by providing rich
understanding of households locational behaviors, which standard quantities
models do not. And they enable planners to consider a broader range of possible
cumulative or emergent effects of land use policies or market trends.
My work is guided by the following questions. Why is the exurban development
so dispersed? How does exurban dispersion emerge from the interactive effects of
amenities, settlement density and accessibility? More specifically, what are the
effects of amenities on exurban residential location? The flexibility of ABMs
allowed me to build models at a level of detail and complexity appropriate to the
questions I ask. My goal is not to create models that reproduce the specific details
of particular historic exurban development pattern, but to explore the use of a
novel modeling method the agent based approach to build models that help us
understand the processes that lead to the patterns and dynamics behind it.
Research Organization
The introduction includes a discussion of non-urban growth trends in the
intermountain West and an outline of the research. In the next chapter literature
review, I place agent-based models in the context of the history of urban growth
modeling and assess recent advances in agent-based models and applications of
ABMs in the urban systems. The methods chapter covers the major contribution
of this project, the research questions and hypotheses and the refinements of
13


methods for application of agent-based models to problems of peripheral urban
development, within a Geographic Information System (GIS) environment. The
empirical results chapter includes a review of the results of the application of the
model as well as a discussion of the background and case study. The last chapter -
summary and conclusions provides and interpretation of the project findings and
discusses the policy implications of the model.
14


CHAPTER 2
LITERATURE REVIEW:
MODELING THE URBAN PERIPHERY
Exurban residential development is a widespread phenomenon and is becoming an
increasingly important issue in many counties across the U.S. This chapter starts
with a review of urban location theories, followed by a description of amenities
and theory of the new West peripheral development, and finally discusses urban
modeling techniques and places an agent-based approach in the context of the
history of urban location modeling and recent studies of exurban development. It
attempts to build a theoretical basis for studying exurban household location
behavior with respect to amenities, density, and accessibility using an agent-based
model.
Exurban development pattern at the aggregate level emerges from many
households locational decisions and behaviors. Thus, modeling exurban
residential development requires an understanding of individual households land
use decision-making and the motivations behind it. Urban location theories,
discrete choice land conversion models, and agent-based theories of urban and
regional spatial structure offer approaches that begin by studying the individual
behaviors and locational decisions and end by deriving land use patterns from
them.
15


Urban Location Theories
There is a rich history of urban location theories. Two classes of theories
particularly focus on household location choices: Monocentric models (Alonso,
1964; Muth, 1969; Mills, 1972) and their extensions, and public goods/local
amenities models derived from the Tiebout hypothesis (Tiebout, 1956).
The Monocentric Model
The Monocentric model forms the basis of modem urban economic theories of
urban spatial structure. The two most popular and influential theoretical schemata
used to explain urban land use patterns and how people individually and
collectively compete over urban land uses are bid rent theory and ecological
processes as described by the Chicago School. The bid rent function theory
developed by a north German estate owner, J. H. von Thunen in 1826 originated in
the agricultural land rent theory. It was revised and applied in the urban areas by
William Alonso in 1964, and supplemented by Burgesss ecological rings.
Theorists in the early bid-rent and social ecology schools described cities as
monocentric (Alonso, 1964; Muth, 1969; Mills, 1972). The idealized traditional
urban spatial structure pattern is characterized by a central business district (CBD),
where nearly all employment is concentrated, and a circular residential area from
inside out. Assuming that transportation costs in all directions on the presumably
uniform and featureless plain are equal, each household makes optimal locational
decisions based on a unique equilibrium determined by a positive commuting cost
to the CBD gradient and a negative land rent gradient of the house location. In
addition, all households are assumed to be alike with the same income, family
structure, and preferences. Distance to the CBD is the variable to determine
housing locations and the motivation behind households location decisions in
16


monocentric models. The shorter the distance to CBDs, the higher the housing rent
is. Other factors were added to the static monocentric models later, such as income
(Beckman, 1969; Solow, 1973; Wheaton, 1977; Fujita, 1987), time costs of
commuting (Beckmann, 1974; Henderson, 1977), and family structures
(Beckmann, 1973; Muth, 1969; Fujita, 1987).
Unfortunately, monocentric models are too simplistic to adequately represent more
complex urban land use patterns. Many extensions to this basic model were
therefore proposed including polycentric theory of urban structure and dynamic
models of urban growth (Brueckner, 1981; Fujita, 1976).
The Polycentric Model and Other Extensions
Some theoretical and empirical urban economics studies show that urban structure
changed dramatically over the last half century as metropolitan areas grew and
decentralized. Economic activities and employment opportunities cluster in several
interacting subcenters in metropolitan areas to rival traditional CBDs (Giuliano
and Small, 1991).
Polycentric theory proponents ague that polycentricism characterizes most of the
cities in the world today, disputing the central idea of monocentric models, which
assumes that urban employment is concentrated in the traditional CBDs. Instead of
having a single center to which people will travel to work and shop, polycentric
models include both CBD and subcenters, and consider households commuting
costs to these employment centers as the fundamental factor that shapes the urban
structure (Gordon, 1988). This change from considering distance to a single CBD
to multiple employment centers introduced more realism into polycentric models.
17


Both monocentric and polycentric models are limited in their ability to
approximate and analyze observed land use patterns because of their restrictive
assumptions. To better capture land use development pattern and factors that affect
households locational decision making, we need more general models. Many
studies have been dedicated to relaxing the assumptions of the basic model.
Beckmann (1976) and Borukhov and Hochamn (1977) presented models with no
predetermined, exogenously-given centers. Other important studies include those
that considered the presence of externalities like traffic congestion (Mills and
DeFerranti, 1971, Mills, 1972), technological change (Amott, et al., 1986), and
neighborhood amenities (Brueckner, et ah, 1999).
Irrespective of these efforts, the essential tenets of the monocentric model and its
extensions remained intact. The fundamental limitations of the theory for
explaining dispersed residential development in exurban also remains. First of all,
these models are still operating on a presumably uniform, static and orderly
landscape, which is a crude representation of reality and inadequate to explain
complex spatial phenomena related to urban growth, particularly, the dispersed
and low-density residential development dotting the exurban landscape. The low-
density exurban development in effect is highly complex. Land use changes under
the influence of many macro and micro factors, acting and interacting within
varying time frames.
Secondly and more importantly, the soul of all monocentric-based models is
distance to work or costs to jobs. During the mid 19th century, urban areas were
spatially dense because of high transportation cost of goods and people. Later, as
modem transportation modes developed, people were able to live farther and
farther away from work. The potential areas in which new homes may find a
market have expanded dramatically to suburbia first, and then to exurbia. Is
18


distance to work still the primary factor in deciding where to locate in exurbia as it
is in urban areas?
Public Goods/Local Amenities Models
In the monocentric model and its extensions, residential location behavior is
primarily based on the trade off between distance to jobs and the land rent.
However, recent empirical literature suggests that exurban households tend to have
more non-commuting workers and social values motivating exurban location are
closely linked to natural features and systems (see the next section). Rudzitis
(1999) found that only 30 percent of the respondents cited job related reasons for
migrating to the rural West. Patel (1980) showed that the three most important
characteristics of neighborhood for the exurban residents living in Kentucky
Bluegrass are quiet surroundings, safety, and open space. Monocentric-based
models leave out local amenities associated with specific locations, i.e.
environmental qualities, characteristics of neighbors, and some other features of
the location. However, these happen to be the factors found important to
residential location behavior in exurbia.
The second class of urban location theory stresses the role of public goods in
shaping household locational choices. Tiebout (1956) argued that people vote
with their feet and choose to live in places with the public goods that best suit
their preferences. There is a form of intergovernmental competition among the
jurisdictions that guide the provision of desired public goods to attract people.
The Tiebout model has important implications for modeling how attractive a place
is for households. It posited that households choose their residences based on
location specific public goods. Higher quality environment, public schools, less
congestion, and fewer taxes draw some of the city residents migrate to the suburbs,
19


and farther into exurbia. This theory was applied to issues of central city decline
and suburbanization by a number of studies (Mieszkowski and Mills, 1993; Voith,
1996; Garasky and Haurin, 1997).
Amenities and
Theory of the New West Periphery Development
The Mountain West used to be well known and characterized by the word wild
because of low human population densites and vast amount of undeveloped land
(Wilkinson, 1993; Power, 1998). However, from 1990 to 2000, population grew
very rapidly. More than 60% of the rural counties in this region are gaining
population faster than urban areas (Theobald, 2000; U.S. Census Bureau, 2001).
Fast population growth and in-migration are marked by the conversion of land use
from agriculture or wildness to low-density residential development. How does
one explain this unprecedented exurban growth?
Urban economic theory has emphasized the journey-to-work (employment center)
as the primary determinant of urban locational pattern since the 1960s. People are
commuting to CBDs where employment is concentrated because of the superior
infrastructure provision and convenient access to markets and suppliers. On the
other hand, the periphery is left from the development due to the limitation
imposed by space and time. Nevertheless, some researchers argued that the current
spatial structure of the city is quite different from when these traditional theories
were formulated. Urban structure is decentralizing. Changing social factors and
technological advances are producing an impact on the core-periphery relationship
and our understanding of conventional residential location theories. Households
are settling farther from the urban core than before. What are the main driving
20


forces behind the recent peripheral development in the small-towns and rural
mountains in the West?
Recent empirical research on exurban development focuses on factors including
households preferences on amenities and large lots, decentralization of the
employment, technological influences, retirement trends, and flexible working
time and places.
One strand of the literature assert that exurbia is attractive because of its natural
amenities, including scenery, wilderness, and wildlife (Johnson and Rasker, 1995;
Beale and Johnson, 1998). People want to be close to places that are good for
fishing, hunting, skiing, and hiking. Studies have shown that population growth in
rural counties in the U.S. was strongly correlated with natural amenities during the
period from 1970 to 1996 (McGranahan, 1999). Natural amenity was found to be a
more important reason for relocation than job opportunity or cost of living
(Johnson and Rasker, 1995; Rudzitis, 1999). Proximity to wilderness (Rudzitis and
Johansen, 1991) and presence of wildlife (Ingram and Lewandrowski, 1999) were
cited to be important factors as well by residents. Public lands are also stated as a
magnet to amenity-seeking migrants (Rudzities, 1993, 1996, 1999; Rudzities and
Johansen, 1989, 1991). In western Colorado, around 60% of the land is public,
managed by the USFS and the BLM. Adjacency to public land is a popular selling
feature of mountain real estate. Counties in the West that contain federally
designated wilderness areas grew two to three times faster than all other counties
in the nation from the 1970s to the 1990s.
Some studies emphasize developed amenities including ski resorts and other
mountain recreational sites (Duane, 1999; Ringholz, 1992). Resort areas have been
the focus of mountain development since the 1960s. Development used to be
21


clustered around the ski resorts. In recent years, however, it is also found to be
around other mountain-based recreation centers.
Another argument focuses on the attractions of space. A number of studies indicate
that the location preferences of these households are strongly shaped by a desire
for more space and a sense of isolation (Davis et al., 1994; Nelson, 1992;
Riebsame et al., 1996). The latent desire of Americans for the rural life-style is a
very important driving force of exurban development. These preferences for a
rural life style are often defined in terms of local or on-lot amenities including low
levels of noise and pollution, and availability of natural features such as vegetation
and trees (Lessinger, 1986; Yamada, 1972).
Other analyses show that the decentralization of employment and economic
structure change are significant factors giving rise to exurbanization. The high
quality environment in exurbia is a great economic asset (Rasker, 1993, Power
1995). There is a shift to value-added manufacturing and service economy that
makes proximity to markets and supplies less important. As some commercial
development moved to the periphery, many exurban locations became within
commuting distance of new employment opportunities (Garreau, 1991; Cervero,
1986, 1991, 1993). This opened up the exurban landscape to settlement. This
settlement was further fueled by more flexible working time and locations (Dowall
and Salkin, 1986).
A fourth group of researchers emphasize the effects of changing infrastructure
technology on exurban migration. New technology is used to improve
infrastructure provision in the periphery to overcome the constraint imposed by
space and time. The focus in this literature has shifted among various technologies
including telecommunications and computers, the interstate highway system, all-
22


weather roads, expanding commercial airline service, satellite dishes, the
availability of modem septic systems, and improved water wells (Nelson, 1992;
Riebsame et al., 1996; Allan, 1986, Levitt, 2002). The advance in infrastructure
technology strongly influences the accessibility to exurban mountainous regions
and leads to increased decentralization of commercial and residential development
by providing virtually worldwide access to information regardless of distance.
That makes living and working in rural areas easier and makes it possible for
people to take pleasure in rural amenities and escape from urban externalities
while enjoying modem conveniences. A fifth group of researchers argue that
changing patterns of retirement since the 1950s have created a new urban-to-rural
migration (Cuba, 1989; McHugh, 1990).
Finally, several studies state that certain numbers of exurbanites were found to be
non-commuting workers (Nelson and Sanchez, 1999). Nelson and Sanchez (1999)
pointed out that exurban households tend to have more non-commuting workers.
More exurbanites worked at home than suburbanites. They do not have to go to
work every day. Job decentralization and the recent technology improvements
have widened the field of employment opportunity. Some exurbanites can work at
home through telecommunications and computers. Some have flexible working
schedules or have a metro fringe work places with less commuting time (Dueker et
al., 1983; Gordan, Kumar, and Richardson, 1989). Others are retirees migrating
from other places (Cuba, 1989; McHugh, 1990) or amenity-seekers to whom
natural amenity was a more important factor for location than job opportunity or
cost of living (Johnson and Rasker, 1995; Rudzitis, 1999). Therefore, the potential
areas in which new homes may find a market have expanded dramatically to
exurbia.
23


In sum, the literature on exurban development puts forward that it is driven by a
mixture of factors such as desire for a lower density and more rural living
environment, availability of public land and recreation, decentralization of
employment, and recent infrastructure improvements (Duane, 1999; Fuguitt and
Zuiches, 1975; Williams and McMillan, 1980; Williams and Jobes, 1990; Jobes,
1988, 1995, 2000; Davis et ah, 1994; Nelson, 1992; and McHugh, 1990).
Increasing numbers of households have been moving to exurban areas for non-job
related purpose. Exurbanization is suggested to be a result of how an individual
household evaluates location specific natural amenity and quality of life. People
are attracted to exurbia for its scenery, open space and what is perceived as high
quality of life (e.g. low crime and traffic).
Another strand of literature that lay emphasis on location specific amenities is
hedonic price theory of housing markets (Lancaster, 1966; Griliches, 1961, 1967).
This body of work focuses on predicting housing prices through valuation of the
attributes in that property by consumers. Rosen (1974) provided the theoretical
underpinnings for justifying the relation between market prices and the
characteristics of housing. The three sets of independent variables that have been
utilized to measure these attributes are: 1) property characteristics; 2)
neighborhood attributes such as air quality, water quality, undesirable land uses
and proximity to amenities and shopping centers (Ridker and Henning, 1967; Epp
and AL-Ani, 1979; Michaels and Smith, 1990; Spahr and Sunderman, 1999); and
3) economic factors.
A broad amount of literature exists on estimating preferences for neighborhood
attributes using the hedonic model and McFaddens (1974) random utility model.
Some of them concentrate specifically on the economic values of scenery and
natural amenities in urban housing price, including open space, watersheds,
24


wetlands, lakes, scenic views, green vegetation, trees, and ecological diversity
(Shultz, 2001; Vaughn, 1981; Acharya and Lewis, 2001; Doss and Taff, 1996;
Mahan et al., 2000; Benson et al., 1998; Sengupta, 2003; Colby and Wishart,
2002; Chattopadhyay, 2000; Geoghegan et al., 1997).
The Hedonic approach provides tools to analyze market values of site amenities
and sheds light on the study of forces behind land use conversion in exurbia. It
links consumer preference for environmental attributes to housing price by
monetizing a households evaluation of amenities into the total value of the
property. Research has been done showing that amenities around a location do
matter when people are selecting houses.
Hedonic models have the advantages of incorporating price, but there are
drawbacks. For one thing, this method is relatively complex to implement. In
addition, it requires a high degree of statistical expertise to interpret the results and
to cope with two intrinsic econometric problems in the estimation of hedonic price
functions collinearity and spatial autocorrelation because of lack of stochastic
independence between observations. (Dubin, 1988, 1992). Thirdly, it is not closely
attached to policy. Finally, it is mainly used to predict housing price, not location
choices as I need for this study.
Exurban land development is under the influence of many macro and micro factors
and characterized by multiple interactions within varying time frames the
process of development itself has an impact on the environment which will affect
future development in turn with respect to different preferences of different
households for various environmental attributes. While there is now a large
amount of literature describing the attractiveness of forest landscapes and natural
25


systems for residential location, little work has been done to model these factors in
a dynamic and interactive framework.
Using the agent-based approach, I attempt to extend the monocentric model in this
research by integrating these strands of literature and incorporating a location-
dependent environmental quality aspect into location models. I examine the
locational consequences of households choice as influenced by a complicated
interplay among four factors: households characteristics, natural amenities,
density, and transportation costs. In the rest of this chapter, I will review some
modeling techniques that show potentials of simulating the effect of these four
factors and the exurban land development decisions.
Discrete Choice Land Conversion Models
Monocentric models assume a circular city with linear transport costs, identical
households with identical and fixed lot sizes, and homogenous land with
predetermined location and the same qualities. Residential choices of the
households in these models are examined primarily based on the trade-off between
residential location and accessibility to urban centers. Land, however, is not
homogenous. Residential parcels are different in their size, accessibility, and site
attributes (e.g. slope).
Discrete choice land models are built on McFadden (1978)s logit method derived
from random utility theory to model individual development decision-making on
lots with different attributes. Because the non-linear logit method models
individual choices more appropriately, these models enhance the simulation of the
formation of urban land use pattern as combinations of observed individual
26


household or landowners behaviors and choices based on the attributes of the
land.
In the 1990s, John Landis and his colleagues at the University of California,
Berkeley, developed the California Urban Futures Model (CUF), one of the first
GIS-based urban models. It combines discrete choice theory and GIS softwares
powerful analyzing ability on longitudinal and micro-scale data. CUF models are
disaggregate. They allocate growth to the level of developable land units (DLUs),
which are potentially developable sites of approximately one-hectare (100m x
100m) grid cells in the second generation CUF models. Development decision on
each developable land unit is based on equations in multinomial logit procedure
and a great deal of fine-scale information describing land development potential
for the DLUs stored in a spatial database (Landis, 1993, 1994, 1995, 1998).
Unlike monocentric models, which assume rational man selecting sites based on
full information on a uniform, static and orderly landscape, the CUF model
supposes that individuals are boundedly rational with limited information and
operate in a heterogeneous environment.
Many urban growth models have been developed based on the CUF model since
the late 1990s. The Alternative Growth Futures (AGF) model is one of them. The
model in effect mimics the calculations of an individual who is surveying and
comparing raw land sites within a market area. Logit regression is used to
evaluate influences on land conversion between two historical points. These
influences can be interpreted as variables that jointly define the development
profitability of alternative sites (Landis 1994, 1995, 1998; Bradshaw and Muller
1998). Variables include network accessibility, land regulation, jurisdiction type,
urban proximity, neighborhood and site attributes. A group at the University of
Colorado has continued to refine the Alternative Growth Futures (AGF) method
27


through case studies in selected California and Colorado communities. (Muller and
Yin, 2001; Muller, Bertron and Yin, 2002; Muller, Puccio, Baker and Yin, 2002).
As a response to Lees criticism of large-scale models (Lee, 1973; 1994), CUFM
shows clearly the interactions between the different local, county, and regional
development policies and land use change, and how current technology
development on computers, software, and digital data can be combined with the
urban development theory to improve the planning process at local and regional
level (Wegener, 1994, 1995; Klosterman, 1994; Harris, 1994). Discrete choice
land conversion models utilize GIS and have the advantages of being
disaggregated. They are, however, static models. Moreover, different motives and
characteristics of the agents are not taken into account.
Discrete choice land models are built on longitudinal data which usually focus on
two historical points separated by five or ten years with no information on the
intermediate period, while it is well-recognized that urban development is a
cumulative and path-dependent process (Landis and Zhang, 1997; Krugman,
1991). Development on each piece of land depends on its own state and history as
well as the state and history of its neighbors. Therefore, development dynamics are
poorly represented in these models. Discrete choice land conversion models
consider the information on the characteristics of the individual site, not the
characteristics or motivations of individual landowners and developers, and yet the
timing and nature of actual land use change that reflect the economic
characteristics and personal motivations of real people in addition to the locational
characteristics of the sites.
28


The Agent-based Model and Complexity Theory
With the high potential of remedying some of the shortcomings of earlier urban
models and the ability to model interactions between households and between
household and natural amenities, the agent-based approach shows promise of
being a powerful tool for simulating the dynamics of exurban land development
decisions and policy effects.
The agent-based approach is capable of simulating temporal, decentralized, and
autonomous household level decision making and human-environment interactions
iteratively, and how aggregate level land use changes emerge from these local
decisions. Simulation has the advantage of capturing timely changing activities. It
is especially important when studying land use development because development
always concerns the growth of new activities and adaptation of the population to
these changes. In the context of exurban development, households are surveying
about the congestion and amenities and trying to avoid congestion.
The Agent-based model has its origin in computational artificial intelligence
systems, in which the complexity of the system grows out of interactions among
system components. The concepts are from the theory of complex system, which is
about the understanding of how collective properties of a system emerge from the
properties of its components, that is, how collectives behaviors of a system arises
from the detailed, microscopic behavior and relationships of its components. There
are typically many components in a complex system. This theory attempts to
access the holism and synergy from interactions of system components, which are
closely connected but behaving differently. The connection (interdependency) and
the distinction (diversity) are two dimensions characterizing the system. When the
interdependency and diversity of the components change, the system behavior
29


changes as well. The key concepts in complexity theory are relationships among
components and between component and surrounding environment, the learning
and memorizing ability of the components and the system, and the emergent
qualities of the system (Manson, 2001). These concepts and theory have been
developed for the most part in the physical and biological sciences. In recent years,
the relevance of these was explored for the social sciences. The University of
Chicago has participated in this intellectual movement through the development of
the agent-based simulation platform RePast, which is used in this research.
Complex system concepts provide a new approach to the study of urban spatial
pattern. One example of a complex system could be a land use system formed out
of locational behaviors of people. The interactions and interdependencies between
the components (e.g. households and developers), which is the heart of the system,
create the collective properties associated with the land use system as a whole.
Complexity theory often focuses on Complex Adaptive System (CAS) in which
components interact while adapting to their environment. CAS uses computer
simulations extensively as a research tool. Cellular automata (CA) and agent-based
approaches are two examples. These have now developed an extensive literature;
each has benefits and disadvantages.
A cellular automaton system consists of four basic elements: two-dimensional
gridded space, a set of transition rules defined over that space, neighbors, and state
of each grid cell. In the CA model, the behavior of a system emerges from
interactions between cells and their immediate neighborhood. The state of each
cell is determined by transitional rules derived from the states of neighboring cells
in previous time periods. Building on the groundbreaking work of Von Neumann
and Ulam (1962), researchers began in the late 1980s to use cellular automata to
30


explore the dynamics of urban systems (Couclelis, 1989; Batty, 1998; Batty and
Xie, 1994, 1997; Clark et al., 1997; O'Sullivan, 2000; O'Sullivan and Torrens,
2000; Ren and White, 1995; White and Engelen 1997; Webster et al., 1998; Wu,
1998; Webster and Wu, 1998). It is more flexible for trying out new land use rules
or constraints on urban development comparing with regression-based models
because the regression does not incorporate such constraints. CA models have
been tested in a variety of environments.
Human behavior is generally missing in CA models, however; the agent-based
approach supports the study of system evolution through accumulations of
individual interactions both among humans and between humans and their
environment (Franklin and Graesser, 1997). An agent-based model usually
consists of some autonomous and heteronymous agents and a framework to
simulate their decisions and interactions, and their adaptations to environment to
produce emergent system behavior. Emergent behavior occurs because the
systems behavior is more complicated than the simple sum of its components
behaviors (Holland, 1998).
Agents have certain attributes and behavior rules, which can be coded into the
simulation model. Some agents are able to evolve over time through learning and
altering their attributes and behavior, e.g. by genetic algorithms. They remember
and learn information from their own and other agents past decisions. They
interact with each other and with their biophysical environment across time and
space (Franklin and Graesser, 1997).
Unlike monocentric models that simplify the real world to serve the model
building, and discrete choice land use models which have a very complicated
equilibrium structure, CA and agent-based simulations inherently produce
31


complex patterns based on behavioral rules. Wolfram (2002) showed examples of
how remarkably simple rules give rise to behavior of great complexity. Wolframs
work in early 1980s also provided strong evidence that complex behavior which is
very much like what we observed in reality could emerge from simple CA rules.
Agent-based perspective distills the behaviors of agents in their daily decision
making into rules. It provides a way to represent spatial actors having complex
behaviors and capture directly the interactive properties of human and other
systems, and the complex system behavior emerges from the interaction. It gives
significance to human agents and their land use behaviors and relates them to the
rise of urban spatial pattern. Agent-based models can operate interactively at
different spatial and temporal scales, linking local interactions to aggregate level
of land use pattern and vice versa (Moran, Ostrom and Randolph, 1998). They
provide a very useful approach to allow us explore relationships between micro-
level individual behaviors and emergent macro-level phenomena.
Applications of Agent-based Models
to Exurban Development
Agent-based models have been used extensively in evaluation of land cover
change (Parker et al., 2003). Balmann (1997, 2001), Berger (2001), and Lim et al.
(2001) apply a multi-agent approach to studies of farmer behaviors. Rajan and
Shibasaki (2000) employ an agent-based model to examine the land use and land
cover change at the national level in Thailand. Conte and Gilbert (1995), Drogoul
and Ferber (1995), Findler and Malyankar (1995) demonstrate the potential of the
agent-based approach to carry out social experiments under laboratory conditions.
Gimblett (2002) focuses on the use of intelligent agents to explore human-wildlife-
landscape interactions. Researchers have also coupled agent-based and cellular
32


automata models to explore interactions between actors and a landscape (Manson,
2000, Torrens, 2001, Rand et al., 2002).
ABMs also have been applied to a variety of urban issues including pedestrian
movement (Batty et al., 1998; Schelhom et al., 1999), traffic (Nagel et al., 1998),
urban sprawl, residential choice in urban areas (Schelling, 1971; Portugali, 2000),
evolution of settlement and urban transition (Sanders et al., 1997; Bura et al.,
1996) and land use changes (Irwin, 1998; Irwin and Bockstael, 2002; Torrens,
2001).
The Agent-based approach shows the potential to greatly enhance our ability to
model urban systems and to be a powerful tool for simulating the dynamics of
exurban land development decisions. A primary benefit of agent-based models is
that they are dynamic (Torrens, 2001). Exurbanites are more sensitive to space and
natural amenities than their suburban and urban counterparts. But natural
amenities have a dynamic character: they are degraded by too much nearby
development. Therefore, exurban development on each piece of land greatly
depends on its own state and history as well as the state and history of its
neighbors. By including interactions across space and time among agents and
between agents and their environment, the agent-based approach can represent
behavior of homeowners at a relatively high level of complexity. Homeowners
location choices are imitated through translating how their decisions are
continuously affected by what happens around them into the rules, including the
decisions of their neighbors and the change of the environmental characteristics.
These dynamic and interaction rules at the micro level form a fundamental
variation from the static models.
33


A second important benefit of agent-based models is their capacity for capturing
heterogeneity among agents as well as environmental conditions. Agent-based
approach offers advantages when independent components in a system must inter-
operate in a heterogeneous environment. By including interactions across space
and time among agents and between agents and their environment, the agent-based
approach can represent behavior of homeowners at a relatively high level of
complexity. A third important benefit of agent-based models is their capacity to
capture heterogeneity among agents as well as environmental conditions. Different
types of households with varied preferences in exurban areas will act according to
dynamic rule sets, within a varied and dynamic biophysical environment. By using
decision rules other than imposing equilibrium conditions as a lot of mathematics-
based urban models do, agent-based models potentially offer a higher degree of
flexibility for accounting for heterogeneity and interdependency among agents and
their environment.
Although the agent-based approach has recently been applied to a variety of
problems of land use/land cover change and urban system, there have been
relatively few applications of this approach to issues of exurban land development
based on real data. There have been few applications of agent-based models in GIS
environments, either.
Summary
The review of urban location theories, amenities and theory of the new West
peripheral development, and some urban modeling approaches suggests the use of
the agent-based approach to model the dynamic exurban household location
behaviors with respect to amenities, neighbor avoidance, and accessibility.
34


The central problem in this research concerns the construction of an agent-based
model to simulate the exurban residential development as a result of the dynamic
relationship between amenities, density, and accessibilities. Each of these factors
individually has been discussed widely in the literature; however, they have not
been evaluated in an interactive and iterative simulation. Amenity is a key concept
in this research. Darling (1973) and Brown and Pollakowski (1977) found that
distance from lakes was a significant determinant of property values. Distance
from the greenbelt or open spaces was tested to be negatively correlated with
housing prices by some studies (Correll, Lillydahl, and Singell, 1978; Kitchen and
Hendon, 1967; Hammer et al., 1974; Peiser and Schwann, 1993). A number of
studies have found that scenic view adds significantly to the value of residential
real estate (Do and Sirmans, 1994; Rodriguez and Sirmans, 1994; Cassel and
Mendelsohn, 1985; Gillard, 1981; Rodriguez and Sirmans, 1994; Benson et al.,
1998). Chattopadhyay (2000) found trees are important to housing prices. The
amenities variables in the literature were summarized below in Table 2-1. Types of
amenities in this research will include forestland, water bodies, riparian, trees, and
views.
The density level people can endure is different for different types of households,
landscape type, and development pattern. Commuters would endure higher levels
of congestions than second-home owners. Some topology and landscape type can
make your neighbors hidden behind, for example, trees; some make them
conspicuous. Organization of open spaces and distance to the nearest structures
will make a difference when households are evaluating density levels.
35


Table 2-1 Amenities Variables: Literature
Variables Literature
Distance to water bodies (Lakes and Reservoirs) Brown and Pollakowski, 1977; Lansford and Jones, 1995, Milon, Gressel, and Mulkey, 1984; Darling, 1973; Doss and Taff, 1996; Sengupta and Osgood, 2003
Scenic View (Mountain view, Lake view, etc.) Do and Sirmans, 1994; Rodriguez and Sirmans, 1994; Cassel and Mendelsohn, 1985; Gillard, 1981; Plattner and Campbell, 1978; Morton, 1977; Benson, et al., 1997; Benson et al., 1998; Garrod and Willis, 1992a, b, c; Do and Grudnitski, 1995; Rodriguez and Sirmans, 1994; Geoghegan et al., 1997;
Distance to greenery (green belt, forest, parks, etc.) Correll, Lillydahl, and Singell, 1978; Kitchen and Hendon, 1967; Hammer et al., 1974; Peiser and Schwann, 1993; Vaughn, 1981; Palmquist, 1992; Geoghegan et al., 1997; Shultz and King, 2001; Sengupta and Osgood, 2003; Gupta, 2003
36


CHAPTER 3
RESESARCH METHOD: DESIGN OF AN AGENT-BASED
SIMULATION ON THE URBAN PERIPHERY
This chapter describes a method for application of agent-based models to exurban
development. I discuss the research strategy adopted and steps followed to achieve
the objectives of the study. The chapter is divided into six parts: overall approach,
research questions and hypotheses, data collection and processing, computational
framework, model design, and variables.
Overall Approach
I start with an exploration of the relationship among preferences for amenity,
density, and accessibility through a pilot study of exurban development in a 16
square miles area at the edge of Lyons, Colorado. This section is followed by
discussion of theoretical models built on an abstract, artificial constructed square
grid. In order to achieve an empirically validated understanding of how land use
decisions of individual exurban households with different preference and under
different policy regimes will affect exurban residential development pattern, I
conduct a case study in the northwestern Boulder County, Colorado. Research
approaches include: 1) collection and processing of necessary datasets; 2)
quantitative spatial evaluation of the past land development in the area; 3) study of
some modeling results from Alternative Growth Futures project applied in the area
by a research team at University of Colorado (Muller, Bertron and Yin, 2002;
Muller, Puccio, Baker and Yin, 2002); and 4) construction of simulations using
37


ArcGIS and RePast. I examine how land use decisions made at the household level
affect outcomes at a higher/aggregate level in the urban system. Specifically, I
build three models. The first emphasizes amenity and accessibility factors; the
second stresses neighbor avoidance and accessibility; and the third focuses on the
interactive effects of amenities, settlement density, and accessibility.
Geographic Information System (GIS) plays a role as a tool for data compiling,
processing, and spatial database building. It also serves as a modeling tool for the
first stage agent-based model. The multi-agent based modeling tool RePast -
simulates the temporal and spatial land conversion from one state to another
according to a set of predefined transitional rules based on evaluation of past
patterns of exurban land use decision-making.
Research Questions and Hypotheses
Why is exurban development so dispersed? I focus on three types of variables
described in the literature as generating exurban dispersion: site amenity
attractiveness, development density (neighbor avoidance), and accessibility. I
apply an agent-based approach to model population dispersion resulting from
household level locational choices and the interactive effects of amenities,
settlement density, and accessibility to urban services and transportation network
over time.
This research seeks to understand and explain the spatial pattern of exurban
development in the American Mountain West by integrating understanding gained
from historical-empirical narratives, monocentric-based urban location models,
public goods/local amenities models derived from the Tiebout hypothesis,
amenities and theory of the new West periphery development, hedonic price
38


theory, complex system theory, and geographic information system. Because of
advances in technology such as GIS-based analysis ability on very large databases
and availability of enhanced fine-grained GIS data with detailed information on
land use patterns, we now have an unparallel opportunity to expand our
understanding of the exurban development dynamics. Furthermore, the agent-
based approach put the phenomenon of dispersion in a dynamic context and an
interactive behavior framework, which enables us to understand additional
dimensions of exurban dispersion. I employ recently developed techniques in
complexity theory and agent-based simulation in this research to model exurban
sprawl as a product of interaction with respect to amenities, density, and
accessibilities.
The primary contribution of this research is methodological. I explore a new
potentially powerful way of modeling the application of agent-based approach -
for exurban development. The theoretical contribution is mainly in the area of
urban location theory. It extends monocentric model by including dynamic effects
of location specific amenities and density. I simulate the emergence of spatial
network and dynamic filling up process as representations of exurban
development. How is exurban dispersion emerged from the interactive effects of
amenities, settlement density and accessibility? I take a building blocks approach
to tackle this question by beginning with a simpler question leading to the
interaction one.
Accessibility is often considered the primary determinant of urban locational
pattern. In this research, I begin with adding effects of amenities to accessibility,
following by replacing amenities with density to examine the influences of density
and accessibility, and finally study the interactive effects of all three sets of
variables. At first, I examine the effects of amenities and density one at a time,
39


then a complex model with interactive effects. My three primary research
questions are as follows: 1) What are the effects of amenities and accessibility on
household location in exurban area? More specifically, what are the effects of
multi-agent bidding between amenity-seeking second-home owners and
commuters who are competing for exurban locations? 2) What are the effects of
density and accessibility on exurban household location choices? In particular,
what are the effects of multi-agent bidding between large-lot second-home owners
and commuters who are competing for alternative exurban lot sizes? 3) What are
the dynamic and interactive effects of amenities, density, and accessibility on
exurban location? Specifically, how do development densities in one period
influence location and density in a later period in an amenity-based neighborhood?
Recent empirical literature suggests that exurban households tend to have more
non-commuting workers than their suburban and urban counterparts and social
values motivating exurban location are closely linked to natural features and
systems. Amenities and density variables were found to be more important factors
for exurban location than jobs. Because of some exurbanites preferences for
larger space or low density, relocation is expected after reaching certain levels of
density, which push further exurban sprawl. The following hypotheses are
formulated:
1. Exurban form is a result of household evaluation of the effects of
amenities, accessibilities, and neighbor avoidance.
la.Commuter model tends to concentrate development too tightly around
rural places and transportation networks in comparison with the actual
development pattern.
40


lb. Amenities variables tend to focus development too tightly around
natural amenities in comparison with the actual development pattern.
lc. Exurban development is best explained by mixed preferences of
amenities, accessibilities, and neighbor avoidance.
2. Exurban development is a dynamic system defined by threshold effects as
development reaches certain density.
2a. There are development phasing effects at which exurban
development shifts from a land market dominated by second-home
owners to commuters.
2b. The switch between second-home owners and commuters is sudden
because second-home owners reach density threshold.
This model will shed light on the determinants and constitution of the new West
exurban development and draw on the notion that urban spatial structure is
determined by interdependencies among spatially distributed agents/households
and the biophysical aspects of the landscape environment. This research is
designed to be helpful in evaluation of timing and possible sequence of future
development, and in guiding policy development related to urban land use.
Data Collection and Processing
I use the power of GIS and information technology to store, process, and manage
data that are geographically located, build up a spatial database, and develop a
raster GIS for the simulation. Development of a GIS includes the following steps:
41


survey of data availability; compilation of data from local governments and other
sources; reconciliation of attribute inconsistencies; re-projection to a consistent
geography; and overlay and processing to create necessary new data sets for the
project.
Data Collection
I collected data from municipalities and county in Boulder, Colorado, and other
sources and produced spatial metrics. In order for a GIS to be useful, an accurate,
complete, and meaningful database is the foundation. This requires survey of data
availability and assurance of data precision so that complete and accurate datasets
are assembled for the information system. I also performed a qualitative spatial
evaluation of some data such as longitudinal U.S. Geological Survey (USGS) land
Use/Land Cover data and compared it with other datasets available.
After data survey, raw data were derived from a variety of sources. These are
outlined in Table 3-1.1 collected infrastructure data from U.S. Census TIGER that
include local streets and highways. Parcel data was from the County Assessors
Office. With information on structure and land valuation attributes for each parcel,
it provides the finest resolution land use data available and offers way to better
measure land use changes in exurbia. The map of land ownership came from
Bureau of Land Management. These datasets came in different formats, including
shapefiles, EOOs (Arc/Info export files), and coverages. USGS has digital elevation
data at 1:24K scale available in Spatial Data Transfer Standard (SDTS) format for
download on their website. Census 1990 and 2000 were collected at tract and
block level that contain fine-grained, detailed spatial housing data. Census
geography follows a hierarchical structure. Census block is the smallest
geographic unit, followed by block groups and tracts. Block groups consist of sets
42


of contiguous blocks and usually contain certain income data categories.
Additional data was derived from a related research project, Alternative Growth
Futures Project: Colorado Settlement Pattern and Wildfire Risk.
Table 3-1: Data Types and Sources
Type Source
Parcels County Assessors Offices and County GIS Offices
Streets Census TIGER 1990 data
Highways Census TIGER 1990 data
Municipality Boundaries Census TIGER 1990 data
Census Tract and Blocks Census 1990 data
Digital Elevation Model USGS
Water bodies USGS
Stream TIGER (U.S. Census) and USGS
Boulder County Open Space Boulder County GIS
Land Ownership BLM
Data Processing
Significant processing on raw data was necessary to 1) check and review data
accuracy; 2) convert raw data into a useable form for the model and build the
database; 3) recheck the data for inconsistencies. The accuracy check involves a
review and crosschecking much of the information obtained from a variety of
sources. Digital database is divided into two types: vector and raster. Most of the
raw data are vectors. In order to carry out many grid-based operations to build the
model, I converted all the vector features to raster grids.
One issue in data processing concerns making data readable by GIS. For example,
Digital Elevation Model (DEM) data files were used to generate elevation maps. I
got them from USGS website which made available free of charge in SDTS format
43


at a scale of 1:24k. Neither ArcMap nor Arcview can read SDTS files directly. In
order to be incorporated into the regional GIS, they were converted into grids, and
elevation was calculated in Arclnfo. This same operation was performed on more
than 20 tiles for the county. Measurement intervals were then reclassified and the
tiles were merged into a single elevation map for inclusion in the model.
Unifying the representation of features is another important issue when processing
data. GIS provides tools that can integrate all the different data sets within a
common reference framework defined by the geographic projection system.
Geographic projections are mathematical techniques used to convert features from
a spherical surface, earth, onto a flat surface. The datasets I collected came with
different projection information. In order to line them up properly and present
them in the most geographically effective manner, I performed re-projection on all
the raw data. All the data layers were converted to the UTM, zone 13.
After checking and assembling data and defining the right projection, data are
ready to be accurately compared and converted to raster representations. The grids
I converted and created have a cell size of 1-hectare or 2.5 acres. Unification
issues for raster data representation relate to map extents and cell size. Most of the
grid algorithms treat the grids as matrices of data or arrays of numbers. All grids
must have identical extents, cells size, and projection.
I automated this conversion process using GIS software Arc/Info, Arc Macro
Language (AML), and Unix C shell scripts and produced grid layers describing the
biophysical and social characteristics of the environment with which households
interact. Arclnfo was used to calculate the proximity measures as well. The
distance from each cell to highways, highway ramps, local roads, and to the
44


nearest developed area was calculated. Most of the processes were done using
AML to automate and speed up the process.
Computational Framework
GIS Techniques and Algorithms
A variety of software is used to process data and build the model. GIS has an
inherent ability to capture, combine, retrieve, analyze, and display multiple data
layers spatially referenced to earth. During the early phases of the research, I build
an agent-based model using Arc/Info Grid modeling platform and Map Algebra
based on a model of exurban growth pattern: Alternative Growth Futures (AGF)
(Muller, Bertron and Yin, 2002; Muller, Puccio, Baker and Yin, 2002).
GIS data sets are digital equivalents of paper maps. When building a geographic
information system, real world geographic data need to be translated into
representations that can be stored and manipulated in a computer. Two data
models that currently dominate commercial GIS software can satisfy this purpose:
the vector and raster models. The vector data model represents spatial data as
geometric objects: points, lines, and polygons; raster data model represents them
as images files composed of grid cells or pixels. Each pixel is a single picture
element and acts as the basic unit where information is explicitly recorded. Each
pixel or cell is assigned only one value that represent a geographic phenomenon,
e.g. if it is within certain municipal boundary. The matrix of cells is called a grid,
and are also represented as arrays of numbers. The resolution of the grid depends
on its cell size. The smaller the cell size, the higher the resolution. The strength of
the raster data model is its simplicity and ease of use in mathematical
45


computations. Most GIS operations can be performed on both data models and
conversion from one to the other is easy.
A GIS spatial database includes layers of spatial features such as roads, streams,
and administrative boundaries. Any geographic object on a digital map has
detailed information (attributes) stored in data tables linked to the digital spatial
database. GIS software has tools for organizing information on the spatially
defined features and integrating all the different data sets so that data for the same
objects on different data layers can be related and combined for analysis and
mapping purposes. GIS also allows for the creation new data layers using the
existing ones.
Arc/Info has the most comprehensive functionality among the ESRIs ArcGIS
software family. It includes all the functionality of ArcView and ArcEditor. It is a
complete GIS data creation, update, query, mapping, and analysis system and it is
de facto standard for GIS. The raster-based geographic information system,
Arc/Info GRID, provides a powerful framework and tools for model development.
All data in the Arc/Info Grid module are referenced to a fixed location, that is, a
cell with a size defined by users and position defined in terms of x and y
coordinates. The grid module was used to store and construct a raster-based spatial
database. With these tools, I performed data conversion, generalization,
aggregation, overlays, buffer creation, statistical calculations, and much more.
There are many factors that drive people to make decisions with respect to
building a house on a particular site, for example, environmental and ecological
constraints, existing land development, accessibility, and policy constraints.
However, display of all factors together on a single map is problematic because of
46


too much information. Raster based map overlays help ensure that all the
information about the same place can be consolidated.
A raster database can contain many congruent raster layers with an identical
geographic extent and the same number of rows and columns. Each layer shows
one theme and contains a set of data describing a single characteristic for each cell.
Each cell is coded with integer values such as distance to the nearest local road,
the year a house was built in the cell, and whether the cell is located on a 200-
meter public land buffer. Thus, a household can move anywhere on a grid and is
associated with information such as distance to the nearest local road, whether the
cell is built out, and whether the cell is coded as public land buffer. These are
some of the questions a household will ask when looking for a site for
development. Consolidations of layers combines information about a single cell
relevant to land use development decisions (Figure3-1).
Figure 3-1 GIS Raster Overlay
Parcel
Road
Public Land
Buffer
Stream
Buffer
47


Raster overlays are faster and easier to work with mathematically than vector
overlays. Grids can be created from arc or polygon overages, shapefiles, or by
application of statistical or logical functions to other grids. Arc/Info Grids provide
a set of very powerful tools for the geographic data analysis. However, creating
and manipulating grids could be a daunting task.
Map Algebra is a data manipulation language designed specifically for geographic
cell-based systems. Map algebra provides a high-level language interface to
describe and manipulate data. Arc/Info Grid uses map algebra to increase its
computational capacities and enhance its mathematical and spatial functions. Grids
allow combinations of various functions, operators, grids, numbers, and scalars.
Models can be computed on a cell-by cell basis on columns and rows in a
clockwork manner.
A docell block was used to control per-cell processing on a cell-by-cell basis
within the analysis window. A sequence of operations were performed at a cell
before moving to the next. The docell block also allows the use of IF and WHILE
statements to support models that require iterative processes at the individual cell
level. Therefore, it allows researchers to examine and analyze the features for each
cell. The following example shows that all the cells that are not on the open spaces
were examined to see if they were developed before 1990. If not, they will be
considered for development under the condition that they are located in the stream
buffer.
docell
if (K:/forest/lyons/grid/openspace eq 0)
{
if (K:/forest/lyons/grid/dev90 o 1 and
48


K:/forest/lyons/grid/streambuf = 1)
{
K:/forest/lyons/grid/devcom = 1
count += 1
}
}
end
After considering a number of options, the Arc/Info Grid modeling platform, Map
Algebra, and ARC Macro Language (AML) were chosen in my first stage model.
The advantage of using this model is that it does not require data transformation,
input and output. However, as information is needed for each cell and the number
of cells used in the model increases, grid data sets tend to take up too much file
space and slow down the calculation speed. Moreover, modeling results are not
directly viewable. Data need to be joined and mapped in Arcview or ArcMap.
Agent-based Modeling Platforms
Arc/Info Grid execution times become unacceptably long as models become more
complex. Other computer programming languages that implement multi-
dimensional arrays were considered to reproduce the spatial analysis performed in
GRID, for example, Java.
In order to better implement the agent-based model, I reconstruct the model code
on a more robust computational platform. Modeling platforms developed
especially for agent-based models include Swarm, created by Nelson Minar and
Chris Langton at Santa Fe Institute; Kenge developed by Paul Box and colleagues
at Utah State University; Ascape by Joshua Epstein and Robert Axelrod at the
49


Brookings Institution; Starlogo by Mitchell Resnick at the MIT Media Lab; and
RePast by the Social Science Research Computing at University of Chicago. I
reviewed several such platforms and scripting languages and decide to move much
of the existing code to a more flexible platform based on Java, that is, RePast.
RePast is a Java language based software framework for creating agent-based
simulations. It is developed by the Social Science Research Computing at
University of Chicago based on Swarm, a software package for multi-agent
simulation of complex systems used mostly for physical and biological sciences.
RePast is an acronym for Iterative Porous Agent Simulation Toolkit. It provides a
class library to help collect, create, run, and display data for an agent-based
simulation. In addition, it has ready-to use mechanism for taking snapshots of
running simulation, or even creating QuickTime movies of simulations.
I downloaded RePast system version 2.0.1 from the website
www.repast.sourceforge.net and installed it together with Java 2.1. To develop
models using RePast, I acquired a copy of Java Software Development Kit, which
includes a java runtime required for running the RePast demonstration simulation.
In RePast, an agent-based simulation proceeds in two stages. The first stage sets up
the simulation; the second stage runs it. Simulation is divided into ticks or time
steps. The programmer needs to describe what happens during the setup and every
tick and place the information in a class called SimpleModel. There are typically at
least two classes in RePast, agent and model classes. The agent class describes the
behavior of the agent (household in this research) and the model class coordinates
the simulation setup and running of the model. Agents are implemented according
to the properties of exurban households described latter in this chapter. It is
50


facilitated by the used of Java interface, the Drawable interface. I placed agents in
the RePast space by implementing the methods like "draw".
SimpleModel is served as the basis of my model class and extended to suit my
specific simulation purposes. Setup(), buildModel(), and step() are the three
methods I needed to create the simplest simulation satisfying my purposes. I used
them as follow.
import uchicago.src.sim.engine.SimpleModel;
public class LifeModel extends SimpleModel {
private ArrayList birthList = new ArrayListQ;
private Space space;
private int width;
private int height;
private Display Surface dsurf;
private Displayable display;
public void setup() {
super.setup();
width 150;
height 150;
dsurf = new Display Surface(this, "Northwestern Boulder Display");
}
public void buildModel() {
space = new LifeSpace
("E:/agent/lyons/model/lyons4/ascii/streambufasc,
this, "E:/agent/lyons/model/lyons4/ascii/ybtpcl90.asc,
display = space. getDisplayQ;
((Object2DDisplay) display).setObjectList(agentList);
}
51


public void stepQ {
int size = agentList.sizeQ;
space, step (agentList);
}
}
Setup() is used to do housecleaning. It is called when the simulation is first started
and whenever the setup button is clicked on the interface. BuildModel() is used to
create the objects that my simulation needed to use including agents and space in
where they operate. Agentlist is an array to keep all the agents. LifeSpace is the
raster space which agents inhabit. Now that the setup stage is finished,
SimpleModel calls step() method for each tick of the scheduler. Step() tells the
simulation what need to be done for each tick or time step. In the step() method, I
iterated through all the agents (households) and also called relevant method to
execute their behavior on each one.
Some action occurs at each tick according to the results of previous actions, the
current values of all the agents variables and space in which they operate. The
history of the simulation is the history of the states of all the ticks. Any changes to
the states occurred are recorded and used as input for the next tick.
For example, if there is a house built in a neighborhood, this information will be
recorded and reflected immediately in the next tick so that, for example,
households who are sensitive to the high density around them would have an
opportunity to reconsider their development decisions. It is one of the advantages
of using RePast to build agent-based exurban household location model. Neither
discrete choice land use model nor agent-based model we developed in the first
stage using Arc/Info grid can do this efficiently.
52


RePast also provides a graphical user interface (gui) to allow an interactive
simulation. User can start, stop, pause, setup, view settings, and exit a simulation
through the toolbar. In addition, it has a Step button that would allow users to
simulate by repeating single iterations. In the settings window, there are Custom
Actions and Repast Actions tab. I used the first two buttons in the RePast Actions
tab to make movies and take snapshots. When the setup button is clicked, the code
in the setup() method is executed while if step button is clicked, the code in
buildModel() and step() is executed.
RePast version 2.0 introduced interoperability with geographical information
systems. The space on which agents operate can be imported from GIS. ArcGIS
raster files were exported as raster ascii files. RePast can read them in as
RasterSpaces, a class defined in RePast.
Model Design
Overall Design
Many urban growth models are static in nature. They usually refer to a single or
two points in time and build on longitudinal data focusing on these historical
points separated by five or ten years with no information on the intermediate
period. Such models ignore the transformation process from one time point to
another. However, urban development is a cumulative and path-dependant process.
It is important to understand and include inherently dynamic nature of growth in
urban models. Development on each piece of land depends on its own state and
history as well as the state and history of its neighbors. Because development
dynamics are poorly represented in these models, they are inefficient to investigate
53


complex dynamic relationships between components of the spatio-temporal urban
development.
Some new types of models developed recently based on complexity theory on the
other hand can include the changes and interactions of model components over
time. These dynamic simulation models can be used to represent behaviors of
homeowners and the evolution of every individual parcel at a relatively high level
of complexity by using a process-based approach. They attempt to capture the
myriad component interactions that comprise the development of an urban system.
Homeowners keep discovering the continual changes in their neighborhood within
such models. The interpretation of urban development is evolving from a static,
fixed model of concrete relations to a more dynamic model of fluid and complex
systems. These models have a better chance of demonstrating the possible results
of policy changes. Such models have recently been applied to a variety of
problems of land use/land cover change such as what was discussed by Parker et
al. (2003), residential segregation (Schelling, 1972), and to simulate urban systems
(Torrens 2001). In this research, I use agent-based approach to simulate land use
dynamics in exurbia. Theoretical models are built in three phases:
Model I: Assess the static effects of amenities and accessibility on
location.
Model II: Evaluate the dynamic effects of lot size preference on long time
density change. Second-home owners were assumed to be space
sensitive. They like large lot while commuters are space-neural.
This phase focuses on the influences of density and
accessibility.
Model III: Evaluate the effects of amenity and lot size preference on long
time location and density change. In this phase, I build a
54


dynamic model to incorporate interactive effects of amenities,
settlement density and accessibility.
The theoretical models are built on an abstract grid. An ASCII file is created and
imported to Arcview to create a grid of 150 x 150 cells, with a resolution of 100 x
100 meters per grid cell. In order to run the above three theoretical models, road
network, two small towns assumed to be within 8 cells, public owned lands, a lake,
and some streams are drawn randomly and added to the grid. This grid sets up the
basis for creating the following variables: distance to road, distance to city,
distance to public land, distance to lake, and distance to stream.
Model verification and validation are essential parts of model development
process. After the initial model development, I engaged in a series of activities
such as debugging, verification, model review, and validation before further model
development. Debugging involves the use of various techniques to determine the
cause of a bug and fix it. I validate the theoretical models by evaluating how they
work empirically in the study area to assure that they represent the real system to a
sufficient level of accuracy. I collect performance measures of the system for
defined periods of time, run the model over 100 times with the given input data,
and compare model outcomes to the real-world observations over the given period
of time.
A series of tasks were carried out to verify and validate my model to ensure that
they are sufficiently accurate with reference to the purpose of this research;
however, it should be noted that no model is fully verified and validated, and no
model is 100% accurate. Any model is only a representation of a system.
55


Interaction Framework
The five building blocks of my agent-based model are agents, states,
neighborhood, environment, and rules. I designed a conceptual agent-agent and
agent-environment interaction framework using these five components and
mapped it onto an computer model to simulate agents decisions concerning the
state conversion of a piece of land following certain rules (figure 3-2).
Figure 3-2. Framework of Agent-agent and Agent-Environment Interaction
56


In my model, agents represent households who are searching for exurban
residential locations based on certain rules. Agents have heterogeneous attributes
and behaviors. In this research, I consider only two types of households:
commuters, and second-home owners/telecommuters. Commuters decisions are
based on accessibility factors, site characteristics and urban proximity. Households
from different groups interact with other households and respond to natural
amenities and other factors. Household behavior is also a reaction to the macro
level land use policies. By evaluating these properties of households, I can portray
their behaviors at a relatively high level of complexity and realism (see Table 3-2).
Table 3-2. Properties of Households
Objective-orientation Search for an exurban residential location.
Heterogeneity Two types of households (commuters and second- home owners) behaving differently. Varied biophysical environment.
Interaction Household-household interactions.
Respond to the characteristics of the environment.
Respond to the macro level land use policy (i.e. wildfire mitigation policies).
Agents are typically individual programming objects that respond to a variety of
social and environmental information and filter this information through a set of
rules that govern decisions about household location and mobility across a
landscape. Agents keep searching for exurban residential locations which satisfy
their requirements. Each grid cell has one of the two states: developable and
undevelopable. Undevelopable cells include those on roads, water bodies, and
public owned land, and those occupied by other agents. Agents are allowed to
locate only on developable cells in the simulation.
57


The decisions of agents in a neighborhood influence and are influenced by other
agents in that neighborhood because the presence of one household affects
residential location decisions of another. I assumed that some households prefer to
be close to other households for a number of reasons. For example, commuters
have a preference for neighbors because of the presence of schools and services
after a certain level of clustering. On the contrary, some studies indicate that the
location preferences of some households, like second-home owners, are strongly
shaped by a desire for more space, natural amenities, and a sense of isolation
(Davis et al., 1994; Nelson, 1992; Riebsame et ah, 1996). They are more sensitive
to levels of noise, scenic views, and congestion. Therefore, they are repelling
rather than being attracted by other households, which create pressures for low
density development. These differences between households and their decision-
making are reflected in the decision rules of the agents. Two major ways of
defining neighborhood in the literature are the Von Neumann and Moore
neighborhood (Epstein and Axtell, 1996). In my model, neighborhood is an area of
13 by 13 cells, that is, an area of 169 hectares.
The environment is a two-dimensional array of regular spaces represented as a
mosaic of grid cells. It is the virtual space agents live in and interact with based on
decision rules. It is the combination of the developable land layer and cell
characteristics layer. Cell characteristics include: 1) site amenities; 2) Measures of
contiguity and fragmentation neighborhood effects; and 3) accessibility factors.
The advantage of this modeling framework is that it provides a comprehensive
portrait of actual land use change under market and policy conditions present
during the study period. Thus, the allocation and visualization of exurban growth
is based on current land market patterns. Moreover, this approach relies heavily
on the evaluation of decision-making by individual landowners. It describes land
58


markets in terms of the preference of households and characteristics of specific site
that make them attractive to different households. In this respect, the model can be
explained relatively easily to a layperson.
I develop models in both Arc/info Grid platform and RePast using agents who
have exurban locational preferences based on this framework. They exist on a
heterogeneous landscape which is defined using data stored in the geographic
information system described early in this chapter. It is a two dimensional irregular
square lattice.
Design of Rules
There are two main objects encoded in the model: cells and agents. Agents or
households select exurban sites or cells following a variety of land conversion
rules. The rule structure consists of two main components. They are the different
locational preferences of second-home owners and commuters (figure 3-3). Rule
sets vary among different types (groups) of households to guide their behaviors
according to preferences. Preferences are determined by the household types and
the three types of independent variables. Sets of preference functions were
constructed with respect to the independent variables, including preference for
natural amenities, neighborhood, and accessibility to employment centers,
services, and infrastructure. While some of the independent variables were
introduced into the model to determine the non-homogeneous nature of the
physical space with which agents interact and where the land use dynamics unfold,
neighborhood effects variables and associated household preferences are very
important parts in the model because they reflect the dynamic impact of land uses
in the immediate surrounding of a location.
59


Figure 3-3 Components of Rule Structure
At the first stage, I build a very simple model to explore rules in a limited
geographic test area. I design a number of rules based on some important
characteristics of the sites which most likely attract households according to the
literature review and results of an earlier model: Alternative Growth Futures
(Muller, Bertron and Yin, 2002; Muller, Puccio, Baker and Yin, 2002). Rules are
coded so agents are able to navigate through geographical environments and make
decisions on which piece of land they want to develop.
Each cell on the grid is given two summary preference scores: one for each agent
type. The preference score is determined by the presence of their neighbors, level
of services and the biophysical characteristics of the environment. I use various
rating, weighting, ranking, and map overlays techniques to create scores related to
probabilities of urban transformation. Agents are programmed to look at each cell,
add up the weighs, compare scores, randomly select cells among top-ranked cells,
and record the development for each year. The score is continuously changed by
agents decisions every time step. Through the iterative application of the rules on
the households for twenty consecutive years (1980 to 1999), I generated the
development pattern for 1999.
In the second stage, I build some theoretical models on a more flexible platform
(RePast) using Java and further explore household land conversion rules according
60


to some locational preferences (Table 3-3). In this phase, models are built on
abstract grid spaces. The first model is designed to detect the static effects of
amenities. There are two types of agents or households: commuters who value
short commuting distance most, and second-home owners or amenity-seekers
whose locational choices are strongly influenced by site amenities, for example,
presence of open space or stream, or what the scenic view is from the site (Figure
3-4). The second model simulates dynamic effects of density with respect to
location. Commuters were assumed to be density-neutral, but second-home owners
are very sensitive to space and skip over properties to obtain bigger lot further out
(Figure 3-5).
Table 3-3 Locational Preferences by Different Types of Households
Locational Preference Priorities Second-home Owners Commuters
First Proximity to Public land or Lakes or Streams Proximity to Jobs or Highways
Second Quite environment or Large Lot. Proximity to Roads or Shopping High Level of Development in the Neighborhood
Third Proximity to Open Space or Lakes or Streams
Figure 3-4 Rules for Model I__________________________________________________
Commuters:
1. Determine developable sites.
2. Determine accessibility to roads and towns.
3. Determine accessibility to natural amenities.
3. Find location to satisfy: a) distance to roads and towns is minimal; b) natural
amenities score is highest.______________________________________________
Second-home Owners:
1. Determine developable sites.
2. Determine accessibility to natural amenities.
3. Determine accessibility to roads and towns.
4. Find location to satisfy: a) natural amenities score is highest; b) distance to
roads and towns is minimal.______________________________________________
Bidding: Two types of households bid with their preference score.___________
61


Figure 3-5 Rules for Model II______________________________________________
Commuters:
1. Determine developable sites.
2. Determine accessibility to roads and towns.
3. Find location where distance to roads and towns is minimal.___________
Second-home Owners:
1. Determine developable sites.
2. Determine locations of other households.
3. Determine accessibility to roads and towns.
4. Find location a) far away from other households; b) distance to roads and
towns is minimal._______________________________________________________
Bidding: Two types of households bid with their preference score._________
The third model better represents the complex spatial location behaviors of
households in exurbia. Second-home owners have a higher level of preference for
space in the amenity-rich areas, and it dynamically affects their location and
relocation (Figure 3-6). After reaching certain level of density, households begin to
respond by either not to move in or move out.
Figure 3-6 Rules for Model III______________________________________________
Commuters:
1. Determine developable sites.
2. Determine accessibility to roads and towns.
3. Determine level of development in the neighborhood.
3. Determine accessibility to natural amenities.
3. Find location to satisfy: a) distance to roads and towns is minimal; b) Level
of neighborhood development is high; c)natural amenities score is high.
Second-home Owners:
1. Determine developable sites.
2. Determine accessibility to natural amenities.
3. Determine locations of other households.
4. Determine accessibility to roads and towns.
5. Find location which balance the preference for amenities and space and
accessibility to road or town.__________________________________________
Bidding: Two types of households s bid with their preference score._______
62


Agents visit each cell on the grid and find out if it is developable, and then look at
other attributes of the cell. Households do not make any decision until they finish
surveying all the cells to see the potential of each for development. Adjacency to
natural amenity such as public land and streams, and accessibility to road are
desirable site features. Next, agents compare and rank relative probabilities of
urban conversion for each cell. Each cell has two sets of development
probabilities: one for second-home owners and the other for commuters. Because
households are boundedly-rational and have limited information, a random
mechanism is included in the model. Second-home owners and commuters bid for
cell locations. The cells ranked in the top 100 highest probability are selected
randomly for development. It may be a second-home owner or a commuter
depending on the set from which the highest probability is chosen.
Development information is immediately recorded after the event and affects the
decision of other households for the next iteration. Each time step or iteration in
the RePast model is not one year or one month. It is the time period it takes for the
next development activity taking place. This is one of advantages of using agent-
based model because the desirable or repellent land uses are of great significance
for the quality of a location and for its appeal to particular types of household. For
each time step, agents assess the quality of its neighborhood, a 13x13 cells area
around each particular cell and the distance from each cell to its nearest neighbor.
New activities and land uses occurring in a neighborhood over time change its
attractiveness level for households (Figure 3-7). Only one household moved in
each time step in RePast. The total number of residents moved into the landscape
is determined exogenously by the results of demographic analysis and analysis of
the assessors parcel data. Finally, I test and validate theoretical models
empirically in the study area.
63


Figure 3-7 Pseudo Code of Agent-based Model
Case Study/Validation
Theoretically, agent-based models have a very high potential to be a powerful tool
for simulating the dynamics of exurban land development decisions. They in
theory can provide a qualitative description of land use development. However, to
date, these models rely on stylized heuristic decision rules not derived from
empirical investigation. Rather than building a model on an abstract, artificial
constructed square grid with fabricated rules as most of agent-based urban
development models do, this research begins with building simple empirical
models in a small area around Lyons, Colorado to explore rules, followed by some
theoretical models, and finally tests and validates the theoretical model in a study
area in northwestern Boulder County, Colorado.
64


In order to explore rules, I build two agent-based simulations at high geographical
resolutions using GIS data layers in a 4 by 4 mile area west of Lyons, Colorado.
Starting with a small area not only makes it easier to study the behavior of the
model and play with rules, but also allows me to better understand which
phenomena are idiosyncratic to the study area.
Lyons is a town of about 1585 people (U.S. Census 2000) located 18 miles
northwest of the city of Boulder, Colorado and 20 miles east of Rocky Mountain
National Park. It sits in a mountain foothills area surrounded by hills of ponderosa
pine and red sandstone with mild climate and lots of sunshine. It is well known as
the Double Gateway to the Rockies because of the two different roads leading to
Estes Park; Colorado highway 7 winds up to Estes Park from the south and U.S.
highway 36 goes directly north to the park. Lyons is now feeling the impacts of
growth from the Denver metropolitan area. Development has sprawled into Lyons
surrounding areas.
Lyons area is an ideal study subject to begin this research because of its unique
location and the picturesque landscape of rocky red hills, the rushing St. Vrain
River, agricultural lands, and mountain vistas. It is a beautiful little mountain
community at the edge of a metropolitan region, the mouth of the St. Vrain River,
and the convergence of Colorado Highways 36 and 7. It has a strong lure for
people looking for places with advantages of both being close to work and to
natural amenities. Most of the residents in and around Lyons are commuting to
work in Boulder, Longmont and other metropolitan employment centers (Figure 3-
8).
65


Figure 3-8: Rural-Urban Commuting Areas
Rural-Urban
Commuting Areas
Metropolitan area core
i Metropolitan area
high commuting
Metropolitan area
low commuting
Large town core
Large town
high commuting
Large town
low commuting
Small town core
Small town
high commuting
Small town
low commuting
Rural areas
Arapahoe ?*
Park
Source: USDA, Economic Research Service
I evaluate the land use development pattern from 1950 to 2000 in this area. Parcel
data is used to learn how land use changes each year. Information on year of
building each structure is the study focus in parcel data. Other maps such as land
ownership, local streets, highways, and streams are useful as well to understand
how land use change is influenced by the presences of public land, streams, and
accessibility to roads.
Models are built in two stages using Arc/Info Grid. First, I build an accessibility
model emphasizing commuting and access to services, and then a model including
other variables derived from amenities and the theory of the new West periphery
development. In the accessibility model, the primary variables are travel distances
to transportation network and urban areas. In the second model, primary variables
66


relate to neighbor avoidance and amenities, as well as travel distance. Both
household agent types appear in the second model. In addition, I run the simulation
model iteratively (year-by-year) in the second model in order to capture dynamic
effects. The number of cells developed at each iteration is derived through the
study on development pattern mentioned above. It equals to one tenth of the total
change number from 1981 to 2000. Through the iterative application of the rules
on the households for ten consecutive years (1981 to 2000), I generate a
development pattern for 2000.
The theoretical models are tested and validated in the study area at the final stage.
The study area is the mountainous part of Boulder County (Figure 3-9). It is next
to and west of the Lyons area. As in Lyons, most of the residents in the study area
are commuting to work in Boulder, Longmont and other metropolitan employment
centers (Figure 3-8). The study area is situated in several valleys in the foothills of
the Rocky Mountains. It reaches the city limit of Boulder and Lyons in the east,
Larimer County boundary in the north, and Nederland in the south. Land available
to development in the study area is sharply limited by land ownership and
topographical constraints. Much of the area is public-owned including large blocks
of public forest land and Boulder County open space. It is an area with very high
natural amenities because of picturesque landscape of Rocky Mountains,
numerous of stream and lakes, and wildlife. It has convenient access to National
Forest, Rocky Mountain National Park, and metro job centers. Housing market
pressures are expected to intensify as a result of continuing population spillovers
from city Boulder and Longmont, continued retirement and amenities-seeking, and
growth in employment opportunities in the metro area.
67


Figure 3-9 Study Area
The third theoretical model is validated in the study area from 1981 to 2000. The
number of household increase is obtained from assessors data. I collect data
necessary for creating amenities, neighborhood effect, and accessibility variables
in this area for the same period. After getting information for the study area, I run
simulations over 100 times using the input data over the given period of time in
order to generate or average a pattern to compare with the real world development.
Variables
The dependent variable for the first research question is measured on land
conversion from rural to urban land uses. This is determined based on the year
built attribute in the assessors database. In other words, conversion occurs
68


during the year in which the assessor indicates a structure is built on a parcel.
This information is carried into the grid cells when converting vector data into
raster data. For the second research question, it is development density. This is
measured by the total number of developed cells in a 169-hectare window (as
divided by the total number of the cells in the window). The dependent variable for
the third research question is measured by the conversion from one household type
to another.
Independent variables (Table 3-4) were selected with reference to literature
review, and current growth modeling practice as represented by projects such as
the California Urban Futures Model developed at the University of Califomia-
Berkeley and the SLEUTH model developed at the University of Califomia-Santa
Barbara. Three primary types of independent variables are included in the model:
1. Site amenities These variables are designed to capture the
attractiveness of the site itself to households. Variables include
presence of 200-meter stream buffer and trees, distance to federal and
state land, distance to county open space, and viewshed. All of the
variables were found to be significantly related to housing choices by
previous studies.
Viewshed is a recently coined term used to indicate the entire area an
individual can see from a given point. It is characterized by visibility
between locations. Viewsheds was generated by GIS based on
topographic analysis and tested by field visits. I conduct a viewshed
analysis based on the above information. The aim is to allow the
visualization of what and how far a person might see within the
abstract lattice when standing at the center of an area (figure 3-10, 3-
69


11). Calculations are based on assumed eyes or cameras at 1 meter
high from the surface. Because of the tremendous calculation tasks
and CPU time involved in the analysis, I calculate only the viewshed
for each 9 x 100 x 100 meters (9 cells) area. This process begins with
placing observers in the center of each 9-cells. Visual units include
lake, public lands, mountain peaks, streams, and general areas. If the
lake or peaks of mountains are visible from the site, the view quality
of the site is considered better than other sites where none of natural
amenities are viewable. The result is a grid with a score of view
quality of each cell.
2. neighborhood effects After testing several measures of contiguity and
fragmentation, two measures were introduced into the model: distance
from each cell to the nearest urban development and the number of
developed cells in a neighborhood. These variables provide an
indication of the significance for development, based on proximity to
the neighbors and development densities.
70


| ma Visible bd Not Visible M Contour Lines Observers||
71


Figure 3-11 Viewshed Illustration
Source: Institute of Water Research, Michigan State University
3. Accessibility factors These measure distance from each cell to nearby
highways, roads and streets, and highway ramps. They provide a crude
indicator of the relative costs of extending roads or streets to service a
site, as well as travel times for commuting or shopping trips.
Independent variables can also be grouped into static and dynamic factors
determined by whether they are changing in the simulation. Static variables are
constant throughout the simulation. For example, some global configuration of the
grid space like where the streams are and where the lakes are. Dynamic variables
change at every time step of the simulation, which reflect the consequences of land
development, and cell state transitions for successive states. An example is
distance to nearest neighbor.
1. Dynamic variables Neighborhood effects.
2. Static variables Site amenities and Accessibility Factors.
72


The unit of analysis is a developable cell defined by site characteristics like
whether it is on the right of ways, waterways, and whether it is occupied by other
households). The model relates land use change to site amenities and other
attributes of the cell, including proximity to regional freeways, boundaries of
developed areas, and whether the site is within a 200-meter stream or water body
buffer.
73


Table 3-4 Definitions of Variables
Variables Definitions
Dependent Variable
Change Binary variable. 1: land conversion from rural to exurban; 0: no change
Amenity Variables
Distance to Water Bodies Continuous variable: Euclidean distance to lakes and reservoirs.
Distance to Streams Continuous variable: Euclidean distance to streams.
Viewshed Binary variable. 1: cells an individual can see from a site. 0: cells an individual cannot see from a site.
Distance to Public Land Continuous variable: Euclidean distance to federal or state owned land
Distance to Open Space Continuous variable: Euclidean distance to county open space
Neighborhood Variables
Density of Neighboring Areas Developed Continuous variable: Number of developed cells divided by total number of cells in a 169 hectare window
Distance to Developed Neighbor Continuous variable: Euclidean distance to the nearest developed neighbor
Accessibility Variables
Distance to Roads Continuous variable: Euclidean distance to the nearest local road
Distance to Highway Continuous variable: Euclidean distance to the nearest highway
Distance to City Continuous variable: Euclidean distance to the nearest city center/subcenters
74


CHAPTER 4
DISCUSSION AND RESULTS: AGENT-BASED
EXURBAN HOUSEHOLD LOCATION MODEL
This chapter is divided into two sections. Part one discusses the three theoretical
models in detail and presents results of the simulations. These models are designed
in a way that disaggregated time, space, and agents are tackled one at a time and
then integrated into a dynamic and interactive framework using a disequilibrium
approach. In the second part, the third theoretical model is validated with a case
study in northwestern Boulder County.
Theoretical Models
Theoretical models are built on an abstract grid with 150 x 150 cells, with a
resolution of 100 x 100 meters per grid cell as shown in figure 4-1. This grid is
considered the initial state of the development or the development at time step 0. It
sets up the basis for simulating each of the three theoretical models.
75


Figure 4-1 Initial State
Initial
State
Legend
La ke
Publand Land
I I Stream
Road
| | Town
Model I, II, and III each consist of two types of households, commuters and
second-home owners, which encapsulate the behaviors of the diverse individuals
that make up the system. Households preferences and behaviors are adjusted in
each of the models in accordance with the purpose of the model. Commuters and
second-home owners enter the environment (the abstract lattice) and interact with
it. One of the households takes up one site or cell in each time step (iteration)
depending on their preference bidding. I run simulations, collect, and record the
development pattern for each 30 time steps, that is, a snapshot is taken after every
30 households choosing their sites on the abstract grid. The results are presented in
this chapter at the 60th, 180th, 360th, 600th, and 780th time step respectively with
some variations for each model.
76


Model I
Model I attempts to assess the static effects of natural amenities and accessibility
on household exurban locational choice. The two types of households in this
model, commuters and second-home owners, have corresponding preferences for
accessibility and amenities. I expect that a model based on a mix of preferences for
amenities and accessibility provides a more accurate predictor of development
than a accessibility model.
Model I is built in two stages. The first stage contains homogeneous residents, and
the second stage or the full model has both of the two household types. Results are
presented as two sets of snapshots from these two stages. With the intention of
looking at dissimilar effects of amenities and accessibilities, I set preference for
amenities to null for all households in the first stage and display the results from
the time step 60 and 600 (Figure 4-2). The result shows that build-out first occurs
in the areas around the rural places/the employment centers (Step 60) and then
extends to the part along the transportation corridors (Step 600).
Accessibility factors pull all the development towards the places with a high level
of accessibility. This pattern is persistent from time step 60 to 600. It illustrates
that if all the exurban households have the same simple preferences for the
proximity to job centers and road network, at the aggregate level, exurban
development patterns would be similar to a monocentric pattern. Next stage shows
what the exurban development patterns would look like if we add amenities
variables into the simulation.
77


New Development
Step 60
Step 600


In the second stage (Figure 4-3), commuters and second-home owners bid for land
and develop one site/cell in each iteration. Commuters give preference to sites that
are close to job centers and transportation network. Sites are favorable if they are
close to natural amenities as well. On the other hand, second-home owners desire
sites that are close to public land, water bodies, streams, or sites with great views;
less attention is given to accessibility.
Figure 4-3 demonstrates that because second-home owners chose to develop
places in close proximity to natural amenities, clusters emerged not only along
transportation corridors and around job centers, but also in the areas with rich
natural amenities (i.e. lakes and public land) or with easy access to both road and
natural amenities. From time step 60 to 600, the pattern is consistent. This model
shows how diverse land use decisions made by different types of households at the
micro level affect outcomes at higher level in the exurban development. The mix
of preferences for amenities and accessibility makes Figure 4-3 look more similar
to the exurban development usually seen in the real world than the accessibility
model (Figure 4-2). The inclusion of the amenity variables clearly makes the
model more explanatory.
Nevertheless, the dynamics of exurbanization is still not adequately represented in
this model. Although model I includes heterogeneous agents and bio-physically
varied environment, agents/households only respond to the bio-physical
environment around them, not the characteristics of their neighbors or level of
development in the neighborhood. It is important to include these factors in the
exurban simulation because development always concerns the growth of new
activities close by and adaptation to these changes. Household-household
interaction is missing in model I.
79


New Development
nnj
City
Public Land
d
Lake
Stream
Road
Step 60
Step 180
80


New Development
nu
City
Public Land
I____I
Lake
Stream
Road
Step 360
Step 600
81


Model II
Model II evaluates the dynamic effects of lot size preference on long time density
change. I assume that second-home owners are very space sensitive, that is, they
do not like to live close to each other. They prefer large lots while commuters are
space-neural. This phase focuses on the influences of density and accessibility.
The results of the simulation are presented in Figure 4-4. It displays development
from time step 60 to 780 in five snapshots. Commuters chose areas near
cities/rural places and roads as what they did in Model I. Second-home owners in
this model favor spacious areas with some consideration on accessibility.
At the early stage when there are a large number of empty places, various sites or
cells can satisfy second-home owners to a great extent. Therefore, they offer
higher bid and win their bids more often. At time step 60, most of the
developments are scattered as a result of second-home owners bid triumph. Time
step 180 shows the development in two extremes: clustering on cities and roads,
and dispersion with some tendency to be close to roads. We also notice that there
are more cells developed by commuters than by second-home owners from step 60
to 180.
At time step 360, scattered development is pushed into the areas away from roads
because of lack of space. When development gets more and more densified and
accessibility clusters are stretched out, fewer and fewer cells suit second-home
owners needs while commuters can still find many sites that highly satisfy their
needs. From time step 360 to 600 to 780, large clusters emerged around areas with
good accessibility. However, very few (4) cells are developed by second-home
owners in the empty spaces between previous developments. In some areas, no
second-home owners moved in after step 360.
82


Since every second-home owner attempts to avoid other development and skips
over properties to obtain bigger lots further out, they all together create pressures
for low density development and make the dispersion pattern persistent at the
aggregate level. However, second-home owners densification process slows down
over time and will eventually stop after reaching certain density levels. This is
because these households are not satisfied with the empty cells available for
development after a certain time and therefore, are not willing to bid high enough
to get them developed. Timing plays an important role in this model. Figure 4-4
illustrates how a small behavior preference for neighbor avoidance and larger lots
over time at household level leads to significant and disproportionate reduction in
average density of development.
83


84


New Development
Ci£_
Public Land
Lake
Stream
Road
Step 360
Step 600


Model ID
Model I and II examine different influences of accessibility, natural amenity, and
neighborhood density/neighborhood avoidance on exurban household locational
behaviors. Model in assesses the effects of amenity and lot size preference on long
time location and density change. In this phase, I build a dynamic model to
incorporate interactive effects of amenities, settlement density and accessibility.
Commuters tend to favor sites that are close to job centers and road network most,
but they also consider factors like proximity to natural amenities, space, and level
of development in the neighborhood due to the potential development of schools
and services after a certain level of clustering is reached. Location preferences of
second-home owners are strongly shaped by a desire for natural amenities, such as
86


public land, water bodies, streams, or great views. They also would prefer to avoid
each other and be somewhat attracted to road access.
Figure 4-5 reports the simulation results from time step 60 to 780. It shows that
development first takes place in areas rich in natural amenities, or with easy access
to roads and cities, or both. After some time, a degree of clustering emerges
around these areas (step 180). These are the outcomes of household level decision-
making: commuters prefer cells with good accessibility while second-home
owners prefer natural amenities.
Noticeably, from time step 360 to 600, there is very little change in the upper part
of the abstract grid. From step 600 to 780, there is no difference in the same area,
that is, no cell gets developed. It suggests that after reaching a certain density
level, second-home owners stop moving in because of amenity and space
concerns. They need cells that are not only close to natural amenities but also
distant from their neighbors. After time step 600, it is hard for them to find cells
that satisfy their preferences.
Model III simulates what happens if every second-home owner has a preference
for both natural amenities and space (neighborhood avoidance and larger lots).
Initially, second-home owners bid high for cells because of voluminous spaces
available in amenity rich areas. As more and more cells are developed, the areas
with high amenities densify to a level so that they become less and less valuable
for second-home owners. Natural amenities have a dynamic character: they
become degraded with increased development. When a certain development
density is reached, the densification process stops because second-home owners
want to protect the quality of local amenities against increased density. They
choose to either buy large lots to retain lower density or relocate.
87


Model III is constructed to represent varied types of exurban households with
different locational preferences to reflect the diversity of characteristics found in
social systems. It simulates the process of how an exurban resident selects a site
for development in relation to his individual desire (i.e. preference for being close
to public lands) and his interaction with other residents in the neighborhood, a
process that is directly analogous to the one in the real world. It integrates a
dynamic and interactive framework to explore effects of amenities, density, and
accessibility on exurban location and how exurban spatial structure is determined
by interdependencies among spatially distributed agents and the biophysical
aspects of the physical environment.
88


89


Full Text

PAGE 1

EXURBAN RESIDENTIAL DEVELOPMENT AND THE ATTRACTION OF NATURAL AMENITIES: AN AGENT-BASED MODEL by Li Yin B. ARCH, Yunnan Polytechnic University P.R.China, 1992 M.S. Asian Institute ofTechnology, Thailand, 1998 A thesis submitted to the University of Colorado at Denver in partial fulfillment of the requirements for the degree of Doctor of Philosophy Design and Planning 2004

PAGE 2

by Li Yin All rights reserved.

PAGE 3

This thesis for the Doctor of Philosophy degree by Li Yin has been approved by -----James 0. Huff

PAGE 4

1-" .. -, 1--I Li Yin (Ph.D., Design and Planning) E x urban Residential De v elopm e nt and the Attraction o f Natural Amenities : An Agent-Based Model Thesis directed by Assistant Professor Brian H. F. Muller ABSTRACT Dispersed residential development has been a conspicuous feature of recent population migration to the intermountain West. This low-density settlement pattern has significant implications for land use planning issues such as loss of habitat and agricultural land, and demands on local government for infrastructure and services Many of the models used to explain urban development are not well-adapted to low-density land markets. Since the 1960s, urban economic theory has emphasized the journey-to-work as the primary determinant of urban locational pattern. Recent empirical research on exurban development however, focuses on other factors such as household preference for natural amenities and large lots new technology and retirement trends Agent-based models (ABM) ha v e the potential to be a powerful tool for simulating dynamics in exurban land markets. While ABM techniques have recently been applied to a variety of problems of land use / cover change there have been relatively few applications of this approach in exurban areas This research builds on agent-based urban-rural fringe land use models (Irwin 1998) regression-based land conversion policy models (Muller et al., 2002) and a pilot study of e x urban land de v elopment (Yin and Muller, 2002). I investigate patterns of exurban residential development underway in the American mountain West and model exurban sprawl as a product of interactions around amenities density and accessibility. Simulation results IV i I li

PAGE 5

suggest the interactive and dynamic exurban development model built in this research represents the land market system at a reasonably high level of accuracy. Exurban agent-based models also suggest opportunities for policy applications that link natural and social models to simulate the effects of alternative planning regimes. Such agent-based models are important because they enable planners to consider a broader range of possible cumulative or emergent effects of land use policies or market trends. This abstract accurately represents the content of the candidate's thesis. I recommend its publication. Signed v

PAGE 6

' I ACKNOWLEDGEMENTS A debt of gratitude is owed to many people who had constructive influences on my academic development. My deepest gratitude and appreciation go to Professor Brian H. F. Muller, the principle advisor and mentor, for his support, encouragement, patience, and guidance that helped shape this dissertation. I am also very grateful to my committee members, Professor William R. Travis, Raymond J. McCall, Raymond G. Studer, and James 0 Huff, for their time, dedication, and invaluable comments and suggestions. Special thanks go to the Director of the Ph.D. Program in Design and Planning, Professor Willem van Vliet, for giving me the opportunity to study in the States and his generous support to enable me to concentrate on my research. In addition, I wish to extend my thankfulness to my fellow Ph.D. students who encouraged and helped me, such as Sue Wolf and Tamara Laninga I would also like to express heartfelt gratitude to my beloved father, mother, and brother for constant inspiration to my academic accomplishment up to this level, for their love, encouragement, moral support, and help that made this dissertation possible. Last but not least, my special thank goes to my husband for his patient and support.

PAGE 7

I' I i'' -i ,, 'i : CONTENTS Figures .. .. .. ................. ............................................................. ix Tables ........................................................................................ xi CHAPTER 1. INTRODUCTION: EXURB AN DEVELOPMENT AND THE NEW WEST ...................... .1 Ex urban Development in the West. ............... .. ...... .................. 1 Planning Implications ofExurban Growth .... ............................. 6 Objectives of the Study ................. .................................... ... 9 Research Organization ....................................................... 13 2. LITERATURE REVIEW: MODELING THE URBAN PERIPHERY ........................................ 15 Urban Location Theories ..................................................... 16 The Monocentric Model ..................... ..................... 16 The Polycentric Model and Other Extensions ................. 17 Public Goods / Local Amenities Models .... ................. .. 19 Amenities and Theory of the New West Periphery Development. ........... .............. .............. 20 Discrete Choice Land Conversion Models ................................ 26 The Agent-based Model and Complexity Theory ......................... 29 Applications of Agent-based Models to Exurban Development ....................................................... 32 Summary .............. .. ....................... ................................ 34 3. RESESARCH METHOD: DESIGN OF AN AGENT-BASED SIMULATION ON THE URBAN PERIPHERY .. ............................ 37 Overall Approach ............................................................. 37 Research Questions and Hypotheses ............. .................... ...... 38 Data Collection and Processing ............................................ 41 Data Collection .................................................... .42 Data Processing ................................................... .43 Computational Framework ................................................. .45 GIS Techniques and Algorithms ............................... .45 Vll

PAGE 8

' Agent-based Modeling Platforms ............................... 49 Model Design .. ........... ................................................... 53 -Ov era ll Design .. ................................................... 53 Interaction Framework ..... ................... ...... ............ 56 Design of Rules .............................................. .... 59 Case StudyNalidation ........................................... 64 Variables ...... . ................................. ......... ...................... 68 4 DISCUSSION AND RESULTS: AGENT-BAS E D EXURBAN HOUSEHOLD LOCATION MODEL. .............. .... ...... ........... .. .... 75 Theoretical Models .......................................... ... ... .. ..... ... 7 5 Model I. .. ................... ............................ ........ 77 Model II .... ........................................... .. ........... 82 Model ill ....... .. .... .............................................. 86 Model Comparison .. ................ ............................. 91 Case Study/Model Validation ... ................................... .... ... 92 5 SUMMARY AND CONCLUSIONS .. ......... ....................... ...... ... 99 Research Findings . .. .. .. ..... ... ........................................ .... 99 Planning Implications .................... ....... ....... ...................... .1 02 Future Research Directions ................................................. 1 06 BIBLIOGRAPHY ........................................................ .. .. ... ....... 1 07 Vlll

PAGE 9

FIGURES Figure 1-1 Natural Amenity Scale .............. .................................... .......... 5 3-1 GIS Raster Overlay . ....... ..................... ... .... ............ ........... .... .47 3-2 Framework of Agent-agent and Agent-Environment Interaction ............ 56 3-3 Components of Rule Structure ........ . . . .......................... . ......... 60 3-4 Rules for Model 1.. ................................................................. 61 3-5 Rules for Model II .......................... ................................ ....... 62 3-6 Rules for Model ill .. ......... .................. ........ ... ..... ... ................. 62 3-7 Pseudo Code of Agent-based Model. .......... ... . ................... .... ...... 64 3-8 Rural-Urban Commuting Areas ................................................. . 66 3-9 Study Area .......................................... ................................... 68 3-10 Examples ofViewshed ...... ............... ........................................ 71 3-11 Viewshed Illustration ... . ............. ............ .................... .... ... 72 4-1 Initial State .......... .. .. .. .................................. ....................... 76 4-2 Accessibility Model. ....................................................... ...... .. 78 4-3 Model 1. .................... . .................................................. .. .. .. 80 4-4 Model II .. ................ ... .... ......... . .. .. ... ......... .. ..................... 84 IX

PAGE 10

4-5 Model III ............ .. ............................ ...... ............................. 89 4-6 Modeling Results from Multiple Runs ..... ..... .. ......... ... ....... . ........ 95 4-7 Exiting Development (1981-1990 and 1981-2000) ....................... .... 97 4-8 Comparison ofNeighborhood Density : Existing vs Predicted ..... ...... ... 98 5-1 Wildfire Hazard in the Study Area ......... .. ........ ................. .. ...... 1 04 5-2 Wildfire Hazard Zoning Model.. ......................................... ....... 1 05 X

PAGE 11

TABLES Table 1-1 Population Growth and Land Development in Colorado: 1960-2050 ..... ... 3 1-2 Population Growth and Land Development in Boulder County, Colorado: 1960-2000 ................. ............. ......... .4 1-3 Natural Amenities by Region ................... ....................... ........... 5 2-1 Amenities Variables: Literature ...................... ... .. .................... 36 3-1 Data Types and Sources .. .. .. ................................. ....... ......... .43 3-2 Properties of Households .... ....................... . ........................... 57 3-3 Locational Preferences by Different Types of Households ... ... ............ 61 3-4 Definitions of Variables .................. . ... .. ...................... ....... . 74 4-1 Some Housing Characteristics of the Study Area ................. .......... 93 4-2 Some Household Characteristics of the Study Area ................ ..... ... 94 XI

PAGE 12

CHAPTER 1 INTRODUCTION: EXURBAN DEVELOPMENT AND THE NEW WEST Exurban Development in the West Dispersed residential development in mountain valleys and foothills, and outside city limits, has been a conspicuous feature of recent population migration to the American Intermountain West. This settlement pattern, loosely described as "exurban", has a variety of planning implications related to loss of habitat and agricultural land, as well as placing increased demands on local government for infrastructure and services. However, many of the traditional approaches used by urban modelers are not well-adapted to this land market environment. Agent-based models (ABM), on the other hand have the potential to be a powerful tool for simulating land development dynamics in exurbia. While ABM techniques have recently been applied to a variety of problems of land use / cover change, there have been relatively few applications of this approach in exurban areas This research explores the use of a new modeling technique that uses an agent-based approach to investigate exurb an residential development underway in the Intermountain West by simulating it as an emergent global behavior resulting from household level evaluation of dynamic and interactive effects of natural amenities accessibilities, and neighbor avoidance. A prominent feature of the contemporary American West is its dramatic transition in demography and economy from the "wild" to the "new" West (Riebsame et al., 1

PAGE 13

1997) and the associated impact on ecosystems Historically this regwn was characterized by low population and development densities with enormous undeveloped land (Wilkinson 1993). During the 1990s however the population growth rate of the Mountain West was 25.4% making it the fastest growth region in the country. More than half of the counties grew faster than the national average (Beyers and Nelson 2000) The U.S. Census Bureau (2001) reported that population in the West grew three times as fast as the rest of the U.S Certain recent census estimates that the Mountain West will capture as much as 40% of nation's overall population gain in the next quarter century, which would amount to as many as seven million people. This rapid population growth is fueled by economic structure changes. The economy of the West is transferring from traditional resource extraction to a new economy based on high technology real estate, and recreation Ecosystem processes and biodiversity are facing challenges from the impacts ofhurnan-induced activities. One of the most noteworthy characteristics of the New West is that rapid population increases are occurring not only in urban but also rural areas. More than 60% of rural counties in the region are gaining population faster than urban areas (Theobald 2000; U.S. Census Bureau, 2001). Many local communities are confronting rural sprawl as a major issue. In fact a consistent urban to rural migration has taken place since the 1970s in the Rocky Mountain region (Cromartie, 1994) Rapid population growth invigorated land development. Much agricultural land has been converted to urban uses, especially low-density exurban residential uses These exurb an growth dynamics in the Mountain West have become a pressing concern locally and nationally (Benfield, Raimi and Chen, 1999; Katz and Liu, 2000). 2

PAGE 14

In Colorado, for example the total number of acres used for new development tripled while total population only doubled from 1960 to 1990 (Table 1-1 ). The most dramatic growth patterns are evident in the exurban zone. More than 65% of the land developed during this period was used for new homes in exurbia Within a 30-year period (1960-1990), new homes tripled the amount of low-density ex urban land to 2 269,000 acres twice the combined land areas for urban and suburban development. The Center of the American West predictes this trend might persist for the next 50 years At University of Colorado, a research team has been mapping how land uses have changed within Colorado Table 1-2 shows the results for Boulder County which are as striking as at the state and regional level considering the amount of private land available in the county. Most of the e x urban homes sprung up near the foothills, turning much of the county into a land o f ranchettes and hillside lodges. Ta bl e 1-1 Population Gro wth a n d L and D evel opm e n t in C olorado: 196 0 -205 0 Population D eve lop e d A cres Ur b an and S ub urban Exurban 1 Total 19 60 1,753,947 428 ,000 881,000 1 309 000 19 90 3,294,394 1 176 000 2,269,000 3,445 ,000 2 020 5,010,500 2 222,000 3,407,000 5,629 ,000 2 050 6,208,000 2 ,763,000 4,636,000 7 399 ,000 Source : Center of the American West 1 T h e ce nter of th e Am erica n West s We s t e rn Futur e s proj e ct (2001) d e fin e d ex urb a n a cc ordin g t o th e d e n s ity: 1 unit p e r 10 to 40 acre s 3

PAGE 15

Table 1-2 Population Growth and Land Development in Boulder County, Colorado: 1960-2000 Population f----------=D:....:e:....:v-=e.::..;lo:_,.p:....:te-=d:....:A=.:...cr:....:e-=s----....--.---------l Urban and Suburban ExurbanL Total 1960 74,254 39,632 4,936 44,568 1980 189,625 72,136 9,019 81,155 1990 225,339 85,645 10,960 96,605 2000 291,288 102,487 13,446 115,933 Understanding exurban development patterns requires knowledge of the processes that push the development. The literature on exurban development suggests that it is driven by a variety of factors including availability of public land, mountain based recreation, infrastructure improvements, and desire for large lot amenities (Duane, 1999). Some researchers have stated that the rural West is attractive because of its scenery, wilderness, wildlife, and outdoor recreation opportunities (Johnson and Rasker, 1995; Beale and Johnson, 1998). The Mountain West of the United States is historically wild in nature and rich in natural amenities. The Economic Research Service of the U.S. Department of Agriculture developed a natural amenity scale (Figure 1-1) at the county level based on a composite measure of physical characteristics that presumably make a place attractive to live. These measures include environmental qualities of places, climate, and typography. Figure 1-1 illustrates that many counties in the Mountain West have a very high level of natural amenities. The average amenity index in the region is ranked the second highest in the U.S., following the Pacific region (Table 1-3). Boulder County, having a ranking of six out of seven, is one of the areas with the second highest amenity index. 2 University of Colorado defmed exurban as an area including all the private land outside the city limit. 4

PAGE 16

Amenity Level 7 Amenities 6 l Figure 1-1: Natural Amenity Scale Source: USDA, Economic Research Service T bl 1 3 N t I Am r b R a e -aura em 1es 'Y eg10n Natural Amenity Scale across the Region Region Average Maximum New England 3.73 5 Middle Atlantic 3.33 4 East North Central 2.61 4 West North Central 2.76 5 South Atlantic 3.64 6 East South Central 3.40 5 West South Central 3.83 6 Mountain 4 .82 7 Pacific 5.49 7 Source: USDA, Economic Res earch Service 5

PAGE 17

McGranahan (1999) found that population growth in rural counties in the United States from 1970 to 1996 was strongly correlated with this natural amenity scale. In recent years people's desire for a high level of natural amenities and recreational opportunities in the American Mountain West (Power 1995 ; Masnick 2001; Reibsame, 1997; McGranahan, 1999) stimulated a historically unprecedented human population boom in the region It was strengthened by e x pansion of service recreation and information industries. The high quality of the natural en v ironment is the region's greatest economic asset (Rasker, 1993 Power, 1995). In the 1990s five out of eight Mountain West states were the fastest growth states in the country Colorado being one of them. Counties w ith major ski resorts grew at an annual rate of8. 5 % from 1960 to 1990. In the early 1990s population growth rates in nearly one fifth of the counties exceeded 5 % (Theobald 2000) and ten out of the fifty fastest growing counties in the U.S were in the mountain areas (Riebsame et al. 1996). Rapid population growth and amenity migration to Colorado is marked by the conversion of land use from agriculture to low-density residential development. From 1992 to 1997, some 271, 815 92 acres were converted (Oberman Carlson and Batchelder 2000). This conversion to low density exurban development continues suggesting a long-term alteration of the natural heritage ofthis region. Planning Implications of Exurb an Growth Exurban development a fast growing component on the landscape of the American Mountain West has far-reaching implications for the existing land use systems local communities and ecological biodiversity. Under current policies and market trends new residential development is likely to remain at current or 6 : .'' ;'

PAGE 18

even decreasing densities in the Mountain West. This suggests that the overall patterns of population dispersion in the region are likely to continue if not intensify. While natural amenities attract people to exurban areas in the West the resulting land use changes challenge the effort to sustain local communities and ecosystem health and threaten agricultural and wildlife habitat. A host of natural and human environmental activities have many profound impacts upon the ecosystem, but the human induced land transformation of the current era is the most striking. Hundreds and thousands of acres of land are converted each year from natural to human settlements and workplace uses. Studies have shown that exurban development affects ecosystem processes and biodiversity (Theobald 2000). Colorado has seen significant threats to habitat due to development pressures. Many ranchettes sprawling subdivisions and recreational activities are replacing natural habits and agricultural lands displacing wildlife (Miller et al., 1998) reducing the population of some species harming environmentally sensitive areas and changing disturbance regimes such as wildfire The flow of population migration to natural amenity-rich areas in the United States has also created a difficult set of regulatory challenges for hazard and land use managers in rural and local governments. For example, wildfire mitigation is e v olving into a significant public responsibility for rural and urban edge county governments in the Western United States. The traditional public safety responsibility of county governments puts them at the intersection of two social and ecological trends, each with the potential for enormous public risk. On the one hand, rapid residential development in fire-prone areas of many counties is increasing the o v erall vulnerability of county residents to injury or loss of property from wildfire. On the other hand long-term ecological processes occurring in the 7

PAGE 19

West, notably pest infestation and effects of fire suppression may increase the probability of serious fire events. County officials describe the threat of wildfire as a kind of "Sword-of-Damocles" hanging over their neighbors, careers and the fiscal viability of county government. In addition, new frameworks are emerging for discussions between rural counties and federal agencies, specifically FEMA and the federal land managers in the U S Forest Service of the Bureau of Land Management. In short there is both a sense of threat in county governments and considerable concern among wildfire planners These challenges are in large part a organizational problem due to high migration rates to rural counties with historically weak land management organizations. They also have another dimension, however, intrinsic to the character of amenity-based real estate development. Land use policy tools can be used to direct residential development away from conservation and hazardous areas but such areas often have the the most desirable building site in the context of amenity market and attract the most demand. Exurban amenity migration is linked to localized configurations of natural features such as views, trees and beach quality. These local configurations tend to create spatially-complex patterns of market demand, and over large areas generate fragmented landscapes with highly interactive natural and social processes. Such landscapes pose major technical challenges to land use managers attempting to identify appropriate regulations related to hazard and farmland and habitat conservation Finally natural amenities have a dynamic character; too much nearby development tends to degrade them Residents frequently act to protect the quality of local amenities against increased density through zoning and ownership patterns; both can create rigidities that interfere with sensible hazard and conservation planning. 8

PAGE 20

There are increasing concerns at the local and regional level about urban sprawl spilling into exurban areas. Hundreds of cities and counties across the Intermountain West are developing strategies to mitigate exurban development impacts on ecosystems It is possible that more stringent land use regulations may emerge from this environment but a clear trend in this direction has not been documented. Certainly, many counties are considering more rigorous interventions; discussions are occurring along these lines even in small, rural counties where property rights traditions are strong Planners are challenged to both accommodate growth and conserve our natural resources, and to minimize the effects on the environment as people move beyond suburbs into the exurbs, and consume agricultural and wildlife habitat. A foremost research need is to develop ways of understanding exurb an development patterns to evaluate alternative planning policies that will guide development, and to examine the consequences of development patterns for natural and social systems Objectives ofthe Study Nelson and Dueker (1990) and Nelson (1990 1991) defined the term exurban based on county boundaries Their exurban counties include some low-density, metropolitan counties and all of the nonmetropolitan counties. While there are many different definitions, exurban in this dissertation includes all private land that is outside city limits in both nonmetropolitan and metropolitan counties. As large numbers of new housing constructions are taking place in the exurban landscape across the American Mountain West each year forests agricultural land 9

PAGE 21

and wildlife habitat increasingly become threatened by urban sprawl. The rural living style preferences of seemingly small and innocuous households may be producing a cumulative effect that is hard to reverse. We need to protect forest and habitat lose and fragmentation from exurban development. In this research, I explore the application of an agent-based approach to understand exurban residential location and trajectories of the cumulative and dispersed exurban development in the Mountain West. An agent-based model is constructed to detect the dynamic exurban sprawl influenced by household interaction with respect to effects of amenities, density (neighbor avoidance and large lot development), and accessibility. Computer models for urban growth and policy development have a long history. Yet, many of the traditional approaches used by urban modelers are not well tailored to exurban land market environments. Since the 1960s, urban economic theory has emphasized the journey-to-work (employment center) as the primary determinant of urban locational pattern. Residential location behavior is assumed to be based on the trade off between transportation costs to urban centers and land rent. However, recent literature suggests that exurban development is driven by a range of factors including availability of public land, recreation, recent infrastructure improvements, and desire for large lot amenities (Duane, 1999) The drivers of low-density exurban development suggested in this literature are not integrated in the conventional models. Agent-based models (ABM) have a high potential to reach beyond the conventional models by simulating interactions between households and between households and natural amenities, and the dynamics of exurban land development decisions and land use policy effects. A richer, more intuitive understanding of dispersed exurban development may be reached through integrating exurban 10

PAGE 22

households' land development behavior related to amenities density and accessibility. Instead of averaging social actors and their behaviors ABMs can be constructed to represent them with various characteristics to reflect the diversity of traits found in social systems such as different types of exurban households with different locational preferences It helps to differentiate entities and processes at play in the exurban system while allowing the direct modeling of interactions between them. That is, ABMs can be used to simulate the process of an exurban resident selecting a site for development in relation to his individual wants (e.g. preference for being close to a stream) and his interaction with other residents in the neighborhood. A process in an ABM is directly analogous to the one in the real world In addition, the ABM approach is a powerful way to build a model that more easily relates to policy decisions and integrates policy effects Furthermore, the agent-based approach provides a means to assess temporal and autonomous exurban residential development decision making at the household level and link these decisions to aggregate land use changes. Simulations can better capture timely changing activities at the micro-level which is important when considering land use development because development always concerns the growth of new activities and adaptation to these changes. In the context of exurban development, households find out about the densities, amenities, and accessibilities and make land use decisions based on them Some households prefer to be close to other households while others have a strong desire for more space natural amenities, and a sense of isolation (Davis et al., 1994; Nelson, 1992; Riebsame et al., 1996) They respond differently to the land development activities around them. ABM techniques have recently been applied to a variety of problems of land use / cover change (Parker et al., 2003) ; however there have been relatively few 11

PAGE 23

--.----applications of this approach in exurban areas, much less applications validated using actual data. One exception is Irwin (1998), who demonstrated that the agent based approach is well suited to explain land use conversion as a result of land use externalities at the urban-rural fringe. This research builds on Irwin (1998), as well as on regression-based land conversion policy models (Muller et al., 2002), and a pilot study on exurban land development (Yin and Muller, 2002) in design of an agent-based simulation of exurban land development in the Western United States. In this research, I build a model using the agent-based approach to explore how exurban residential development patterns may result from interactions between decentralized individual behaviors In addition, I compare the static ABM approach emphasizing accessibility factors to various ABM models emphasizing amenities and neighbor avoidance I begin by using data on actual development patterns in a 16-square-mile area at the edge of Lyons Colorado, to look at the relationship among preferences for amenity density and accessibility as a pilot study of exurban development. Lyons is a town of about 1500 people located in a mountain foothills area in the northwestern part of Boulder County. From this, I build three theoretical models and validate them in the study area, northwestern Boulder County Colorado. I evaluate exurbanization as an interactive and dynamic process at a household level of amenity choice neighbor avoidance, and accessibility preference Exurban agent-based models also suggest opportunities for policy applications that link natural system and social models to simulate the effects of alternative planning regimes. This research evaluates alternative wildfire planning regimes from the perspective of land use policy and site design I report on applications of 12

PAGE 24

this agent-based household location choice model to planning around wildland fire in the study area. Such agent-based models are important because they study micro-level information and link it to macro level patterns. They can both inform the planning theorist and improve planning practice in exurban areas by providing rich understanding of households' locational behaviors, which standard quantities models do not. And they enable planners to consider a broader range of possible cumulative or emergent effects of land use policies or market trends. My work is guided by the following questions Why is the exurban development so dispersed? How does exurban dispersion emerge from the interactive effects of amenities, settlement density and accessibility? More specifically, what are the effects of amenities on exurban residential location? The flexibility of ABMs allowed me to build models at a level of detail and complexity appropriate to the questions I ask. My goal is not to create models that reproduce the specific details of particular historic exurban development pattern, but to explore the use of a novel modeling method the agent based approach to build models that help us understand the processes that lead to the patterns and dynamics behind it. Research Organization The introduction includes a discussion of non-urban growth trends in the intermountain West and an outline of the research. In the next chapter literature review, I place agent-based models in the context of the history of urban growth modeling and assess recent advances in agent-based models and applications of ABMs in the urban systems. The methods chapter covers the major contribution of this project the research questions and hypotheses and the refinements of 13

PAGE 25

methods for application of agent-based models to problems of peripheral urban development, within a Geographic Information System (GIS) environment. The empirical results chapter includes a review of the results of the application of the model as well as a discussion of the background and case study. The last chapter summary and conclusions provides and interpretation of the project findings and discusses the policy implications of the model. 14

PAGE 26

CHAPTER2 LITERATURE REVIEW: MODELING THE URBAN PERIPHERY Exurban residential development is a widespread phenomenon and is becoming an increasingly important issue in many counties across the U.S. This chapter starts with a review of urban location theories, followed by a description of amenities and theory of the new West peripheral development, and finally discusses urban modeling techniques and places an agent-based approach in the context of the history of urban location modeling and recent studies of exurban development. It attempts to build a theoretical basis for studying exurban household location behavior with respect to amenities density, and accessibility using an agent-based model. Exurban development pattern at the aggregate level emerges from many households' locational decisions and behaviors Thus, modeling exurban residential development requires an understanding of individual household's land use decision-making and the motivations behind it. Urban location theories discrete choice land conversion models, and agent-based theories of urban and regional spatial structure offer approaches that begin by studying the individual behaviors and locational decisions and end by deriving land use patterns from them. 15

PAGE 27

Urban Location Theories There is a rich history of urban location theories. Two classes of theories particularly focus on household location choices: Monocentric models (Alonso, 1964 ; Muth 1969; Mills, 1972) and their extensions, and public goods / local amenities models derived from the Tiebout hypothesis (Tiebout 1956) The Monocentric Model The Monocentric model forms the basis of modem urban economic theories of urban spatial structure. The two most popular and influential theoretical schemata used to explain urban land use patterns and how people individually and collectively compete over urban land uses are bid rent theory and ecological processes as described by the Chicago School. The bid rent function theory developed by a north German estate owner J. H. von Thunen in 1826 originated in the agricultural land rent theory. It was revised and applied in the urban areas by William Alonso in 1964 and supplemented by Burgess's ecological rings Theorists in the early bid-rent and social ecology schools described cities as monocentric (Alonso 1964; Muth, 1969; Mills, 1972). The idealized traditional urban spatial structure pattern is characterized by a central business district (CBD), where nearly all employment is concentrated and a circular residential area from inside out. Assuming that transportation costs in all directions on the presumably uniform and featureless plain are equal each household makes optimal locational decisions based on a unique equilibrium determined by a positive commuting cost to the CBD gradient and a negative land rent gradient of the house location. In addition all households are assumed to be alike with the same income, family structure, and preferences. Distance to the CBD is the variable to determine housing locations and the motivation behind households' location decisions in 16 ' I

PAGE 28

monocentric models. The shorter the distance to CBDs, the higher the housing rent is. Other factors were added to the static monocentric models later, such as income (Beckman, 1969; Solow, 1973; Wheaton 1977; Fujita, 1987), time costs of commuting (Beckmann, 1974; Henderson, 1977), and family structures (Beckmann, 1973; Muth, 1969; Fujita, 1987). Unfortunately, monocentric models are too simplistic to adequately represent more complex urban land use patterns Many extensions to this basic model were therefore proposed including polycentric theory of urban structure and dynamic models ofurban growth (Brueckner 1981; Fujita, 1976) The Polycentric Model and Other Extensions Some theoretical and empirical urban economics studies show that urban structure changed dramatically over the last half century as metropolitan areas grew and decentralized. Economic activities and employment opportunities cluster in several interacting subcenters in metropolitan areas to rival traditional CBDs (Giuliano and Small, 1991). Polycentric theory proponents ague that polycentricism characterizes most of the cities in the world today, disputing the central idea of monocentric models, which assumes that urban employment is concentrated in the traditional CBDs. Instead of having a single center to which people will travel to work and shop, polycentric models include both CBD and subcenters, and consider households' commuting costs to these employment centers as the fundamental factor that shapes the urban structure (Gordon 1988). This change from considering distance to a single CBD to multiple employment centers introduced more realism into polycentric models 17

PAGE 29

Both monocentric and polycentric models are limited in their ability to approximate and analyze observed land use patterns because of their restrictive assumptions. To better capture land use development pattern and factors that affect households locational decision making, we need more general models Many studies have been dedicated to relaxing the assumptions of the basic model. Beckmann (1976) and Borukhov and Hocharnn (1977) presented models with no predetermined exogenously-given centers. Other important studies include those that considered the presence of externalities like traffic congestion (Mills and DeFerranti 1971, Mills, 1972) technological change (Amott et al. 1986) and neighborhood amenities (Brueckner, et al., 1999). Irrespective of these efforts the essential tenets of the monocentric model and its extensions remained intact. The fundamental limitations of the theory for explaining dispersed residential development in exurban also remains. First of all, these models are still operating on a presumably uniform, static and orderly landscape, which is a crude representation of reality and inadequate to explain complex spatial phenomena related to urban growth particularly the dispersed and low-density residential development dotting the exurban landscape. The low density exurban development in effect is highly complex. Land use changes under the influence of many macro and micro factors acting and interacting within v arying time frames. Secondly and more importantly the soul of all monocentric-based models is distance to work or costs to jobs. During the mid 19th century urban areas were spatially dense because of high transportation cost of goods and people. Later as modem transportation modes developed, people were able to live farther and farther away from work. The potential areas in which new homes may find a market have expanded dramatically to suburbia first, and then to exurbia. Is 18

PAGE 30

distance to work still the primary factor in deciding where to locate in exurbia as it is in urban areas? Public Goods/Local Amenities Models In the monocentric model and its extensions, residential location behavior is primarily based on the trade off between distance to jobs and the land rent. However, recent empirical literature suggests that exurban households tend to have more non-commuting workers and social values motivating exurban location are closely linked to natural features and systems (see the next section). Rudzitis (1999) found that only 30 percent of the respondents cited job related reasons for migrating to the rural West. Patel (1980) showed that the three most important characteristics of neighborhood for the exurban residents living in Kentucky Bluegrass are quiet surroundings, safety, and open space. Monocentric-based models leave out local amenities associated with specific locations, i.e. environmental qualities, characteristics of neighbors, and some other features of the location. However, these happen to be the factors found important to residential location behavior in exurbia. The second class of urban location theory stresses the role of public goods in shaping household locational choices. Tiebout (1956) argued that people "vote with their feet" and choose to live in places with the public goods that best suit their preferences. There is a form of intergovernmental competition among the jurisdictions that guide the provision of desired public goods to attract people The Tiebout model has important implications for modeling how attractive a place is for households. It posited that households choose their residences based on location specific public goods. Higher quality environment, public schools, less congestion, and fewer taxes draw some of the city residents migrate to the suburbs 19

PAGE 31

and farther into exurbia. This theory was applied to issues of central city decline and suburbanization by a number of studies (Mieszkowski and Mills, 1993; Voith 1996; Garasky and Haurin, 1997). Amenities and Theory of the New West Periphery Development The Mountain West used to be well known and characterized by the word "wild" because of low human population densites and vast amount of undeveloped land (Wilkinson, 1993; Power, 1998) However, from 1990 to 2000, population grew very rapidly. More than 60% of the rural counties in this region are gaining population faster than urban areas (Theobald, 2000; U .S. Census Bureau, 2001). Fast population growth and in-migration are marked by the conversion of land use from agriculture or wildness to low-density residential development. How does one explain this unprecedented exurban growth ? Urban economic theory has emphasized the journey-to-work (employment center) as the primary determinant of urban locational pattern since the 1960s. People are commuting to CBDs where employment is concentrated because of the superior infrastructure provision and convenient access to markets and suppliers On the other hand, the periphery is left from the development due to the limitation imposed by space and time Nevertheless, some researchers argued that the current spatial structure of the city is quite different from when these traditional theories were formulated. Urban structure is decentralizing Changing social factors and technological advances are producing an impact on the core-periphery relationship and our understanding of conventional residential location theories. Households are settling farther from the urban core than before. What are the main driving 20

PAGE 32

forces behind the recent peripheral development m the small-towns and rural mountains in the West ? Recent empirical research on exurban development focuses on factors including households preferences on amenities and large lots, decentralization of the employment, technological influences, retirement trends and flexible working time and places One strand of the literature assert that exurbia is attractive because of its natural amenities including scenery wilderness, and wildlife (Johnson and Rasker, 1995; Beale and Johnson 1998). People want to be close to places that are good for fishing, hunting skiing and hiking. Studies have shown that population growth in rural counties in the U.S. was strongly correlated with natural amenities during the period from 1970 to 1996 (McGranahan 1999). Natural amenity was found to be a more important reason for relocation than job opportunity or cost of living (Johnson and Rasker, 1995; Rudzitis 1999). Proximity to wilderness (Rudzitis and Johansen, 1991) and presence of wildlife (Ingram and Lewandrowski 1999) were cited to be important factors as well by residents Public lands are also stated as a magnet to amenity-seeking migrants (Rudzities 1993 1996, 1999; Rudzities and Johansen, 1989, 1991). In western Colorado, around 60% of the land is public managed by the USFS and the BLM. Adjacency to public land is a popular selling feature of mountain real estate. Counties in the West that contain federally designated wilderness areas grew two to three times faster than all other counties in the nation from the 1970s to the 1990s Some studies emphasize developed amenities including ski resorts and other mountain recreational sites (Duane 1999; Ringhol z 1992) Resort areas have been the focus of mountain development since the 1960s Development used to be 21

PAGE 33

clustered around the ski resorts. In recent years, however, it is also found to be around other mountain-based recreation centers. Another argument focuses on the attractions of space. A number of studies indicate that the location preferences of these households are strongly shaped by a desire for more space and a sense of isolation (Davis et al., 1994; Nelson, 1992; Riebsame et al., 1996). The latent desire of Americans for the rural life-style is a very important driving force of exurban development. These preferences for a rural life style are often defined in terms of local or on-lot amenities including low levels of noise and pollution, and availability of natural features such as vegetation and trees (Lessinger, 1986; Yamada, 1972). Other analyses show that the decentralization of employment and economic structure change are significant factors giving rise to exurbanization The high quality environment in exurbia is a great economic asset (Rasker, 1993, Power 1995). There is a shift to value-added manufacturing and service economy that makes proximity to markets and supplies less important. As some commercial development moved to the periphery, many exurban locations became within commuting distance of new employment opportunities (Garreau, 1991; Cervero, 1986, 1991, 1993). This opened up the exurban landscape to settlement. This settlement was further fueled by more flexible working time and locations (Dowall and Salkin, 1986). A fourth group of researchers emphasize the effects of changing infrastructure technology on exurban migration. New technology is used to improve infrastructure provision in the periphery to overcome the constraint imposed by space and time. The focus in this literature has shifted among various technologies including telecommunications and computers, the interstate highway system, all-22

PAGE 34

weather roads, expanding commercial airline service, satellite dishes, the availability of modem septic systems, and improved water wells (Nelson, 1992 ; Riebsame et al., 1996; Allan 1986, Levitt, 2002). The advance in infrastructure technology strongly influences the accessibility to exurban mountainous regions and leads to increased decentralization of commercial and residential development by providing virtually worldwide access to information regardless of distance. That makes living and working in rural areas easier and makes it possible for people to take pleasure in rural amenities and escape from urban externalities while enjoying modem conveniences. A fifth group of researchers argue that changing patterns of retirement since the 1950s have created a new urban-to-rural migration (Cuba 1989; McHugh, 1990). Finally, several studies state that certain numbers of exurbanites were found to be non-commuting workers (Nelson and Sanchez, 1999) Nelson and Sanchez (1999) pointed out that exurban households tend to have more non-commuting workers. More exurbanites worked at home than suburbanites. They do not have to go to work every day Job decentralization and the recent technology improvements have widened the field of employment opportunity. Some exurbanites can work at home through telecommunications and computers. Some have flexible working schedules or have a metro fringe work places with less commuting time (Dueker et al., 1983; Gordan Kumar, and Richardson, 1989) Others are retirees migrating from other places (Cuba 1989; McHugh, 1990) or amenity-seekers to whom natural amenity was a more important factor for location than job opportunity or cost of living (Johnson and Rasker, 1995; Rudzitis 1999). Therefore the potential areas in which new homes may find a market have expanded dramatically to exurbia. 23

PAGE 35

In sum the literature on exurban development puts forward that it is driven by a mixture of factors such as desire for a lower density and more rural living environment, availability of public land and recreation, decentralization of employment and recent infrastructure improvements (Duane 1999; Fuguitt and Zuiches, 1975 ; Williams and McMillan, 1980; Williams and Jobes 1990; Jobes 1988, 1995 2000; Davis et al., 1994; Nelson, 1992 ; and McHugh, 1990) Increasing numbers of households have been moving to exurban areas for non-job related purpose. Exurbanization is suggested to be a result of how an individual household evaluates location specific natural amenity and quality of life. People are attracted to exurbia for its scenery, open space and what is perceived as high quality of life (e .g. low crime and traffic). Another strand of literature that lay emphasis on location specific amenities is hedonic price theory of housing markets (Lancaster 1966; Griliches, 1961, 1967) This body of work focuses on predicting housing prices through valuation of the attributes in that property by consumers. Rosen (1974) provided the theoretical underpinnings for justifying the relation between market prices and the characteristics of housing. The three sets of independent variables that have been utilized to measure these attributes are: 1) property characteristics; 2) neighborhood attributes such as air quality, water quality, undesirable land uses and proximity to amenities and shopping centers (Ridker and Henning, 1967 ; Epp a nd AL-Ani 1979; Michaels and Smith 1990; Spahr and Sunderman 1999); and 3) economic factors. A broad amount of literature e x ists on estimating preferences for neighborhood attributes using the hedonic model and McFadden s (1974) random utility model. Some of them concentrate specifically on the economic values of scenery and natural amenities in urban housing price, including open space watersheds 24

PAGE 36

-wetlands, lakes, scemc views green vegetation, trees, and ecological diversity (Shultz, 2001; Vaughn, 1981; Acharya and Lewis, 2001; Doss and Taff, 1996; Mahan et al. 2000 ; Benson et al. 1998 ; Sengupta, 2003; Colby and Wishart 2002; Chattopadhyay, 2000; Geoghegan et al. 1997). The Hedonic approach provides tools to analyze market values of site amenities and sheds light on the study of forces behind land use conversion in exurbia. It links consumer preference for environmental attributes to housing price by monetizing a household's evaluation of amenities into the total value of the property. Research has been done showing that amenities around a location do matter when people are selecting houses Hedonic models have the advantages of incorporating pnce, but there are drawbacks For one thing, this method is relatively complex to implement. In addition, it requires a high degree of statistical expertise to interpret the results and to cope with two intrinsic econometric problems in the estimation of hedonic price functions collinearity and spatial autocorrelation because of lack of stochastic independence between observations. (Dubin 1988 1992). Thirdly it is not closely attached to policy. Finally, it is mainly used to predict housing price not location choices as I need for this study Ex urban land development is under the influence of many macro and micro factors and characterized by multiple interactions within varying time frames the process of development itself has an impact on the environment which will affect future development in turn with respect to different preferences of different households for various environmental attributes While there is now a large amount of literature describing the attractiveness of forest landscapes and natural 25

PAGE 37

systems for residential location, little work has been done to model these factors in a dynamic and interactive framework. Using the agent-based approach, I attempt to extend the monocentric model in this research by integrating these strands of literature and incorporating a location dependent environmental quality aspect into location models. I examine the locational consequences of households' choice as influenced by a complicated interplay among four factors: households' characteristics, natural amenities density, and transportation costs. In the rest of this chapter, I will review some modeling techniques that show potentials of simulating the effect of these four factors and the exurban land development decisions Discrete Choice Land Conversion Models Monocentric models assume a circular city with linear transport costs, identical households with identical and fixed lot sizes, and homogenous land with predetermined location and the same qualities Residential choices of the households in these models are examined primarily based on the trade-off between residential location and accessibility to urban centers. Land, however, is not homogenous. Residential parcels are different in their size, accessibility, and site attributes (e.g. slope). Discrete choice land models are built on McFadden (1978)'s legit method derived from random utility theory to model individual development decision-making on lots with different attributes. Because the non-linear legit method models individual choices more appropriately, these models enhance the simulation of the formation of urban land use pattern as combinations of observed individual 26

PAGE 38

household or landowners' behaviors and choices based on the attributes of the land. In the 1990s, John Landis and his colleagues at the University of California, Berkeley, developed the California Urban Futures Model (CUF), one of the first GIS-based urban models It combines discrete choice theory and GIS software's powerful analyzing ability on longitudinal and micro-scale data. CUF models are disaggregate. They allocate growth to the level of developable land units (DLUs), which are potentially developable sites of approximately one-hectare (1OOm x 1OOm) grid cells in the second generation CUF models. Development decision on each developable land unit is based on equations in multinomial logit procedure and a great deal of fine-scale information describing land development potential for the DLUs stored in a spatial database (Landis, 1993, 1994, 1995, 1998). Unlike monocentric models, which assume rational man selecting sites based on full information on a uniform, static and orderly landscape, the CUF model supposes that individuals are boundedly rational with limited information and operate in a heterogeneous environment. Many urban growth models have been developed based on the CUF model since the late 1990s. The Alternative Growth Futures (AGF) model is one of them. The model in effect mimics the calculations of an individual who is surveying and comparing raw land sites within a market area. Logit regression is used to evaluate influences on land conversion between two historical points. These influences can be interpreted as variables that jointly define the development profitability of alternative sites (Landis 1994, 1995, 1998; Bradshaw and Muller 1998). Variables include network accessibility, land regulation, jurisdiction type, urban proximity, neighborhood and site attributes A group at the University of Colorado has continued to refine the Alternative Growth Futures (AGF) method 27

PAGE 39

through case studies in selected California and Colorado communities. (Muller and Yin, 2001; Muller, Bertron and Yin, 2002; Muller, Puccio, Baker and Yin, 2002). As a response to Lee's criticism of large-scale models (Lee, 1973; 1994), CUFM shows clearly the interactions between the different local, county, and regional development policies and land use change, and how current technology development on computers, software, and digital data can be combined with the urban development theory to improve the planning process at local and regional level (Wegener, 1994, 1995; Klosterman, 1994; Harris, 1994). Discrete choice land conversion models utilize GIS and have the advantages of being disaggregated. They are, however static models. Moreover, different motives and characteristics of the agents are not taken into account. Discrete choice land models are built on longitudinal data which usually focus on two historical points separated by five or ten years with no information on the intermediate period while it is well-recognized that urban development is a cumulative and path-dependent process (Landis and Zhang, 1997; Krugman, 1991). Development on each piece of land depends on its own state and history as well as the state and history of its neighbors Therefore, development dynamics are poorly represented in these models. Discrete choice land conversion models consider the information on the characteristics of the individual site, not the characteristics or motivations of individual landowners and developers, and yet the timing and nature of actual land use change that reflect the economic characteristics and personal motivations of real people in addition to the locational characteristics ofthe sites. 28

PAGE 40

--The Agent-based Model and Complexity Theory With the high potential of remedying some of the shortcomings of earlier urban models and the ability to model interactions between households and between household and natural amenities, the agent-based approach shows promise of being a powerful tool for simulating the dynamics of exurban land development decisions and policy effects. The agent-based approach is capable of simulating temporal, decentralized, and autonomous household level decision making and human-environment interactions iteratively, and how aggregate level land use changes emerge from these local decisions Simulation has the advantage of capturing timely changing activities. It is especially important when studying land use development because development always concerns the growth of new activities and adaptation of the population to these changes In the context of exurban development, households are surveying about the congestion and amenities and trying to avoid congestion. The Agent-based model has its ongm m computational artificial intelligence systems, in which the complexity of the system grows out of interactions among system components The concepts are from the theory of complex system, which is about the understanding of how collective properties of a system emerge from the properties of its components, that is how collectives behaviors of a system arises from the detailed microscopic behavior and relationships of its components. There are typically many components in a complex system. This theory attempts to access the holism and synergy from interactions of system components, which are closely connected but behaving differently. The connection (interdependency) and the distinction (diversity) are two dimensions characterizing the system When the interdependency and diversity of the components change, the system behavior 29

PAGE 41

changes as well. The key concepts in complexity theory are relationships among components and between component and surrounding environment, the learning and memorizing ability of the components and the system, and the emergent qualities of the system (Manson, 2001). These concepts and theory have been developed for the most part in the physical and biological sciences. In recent years, the relevance of these was explored for the social sciences. The University of Chicago has participated in this intellectual movement through the development of the agent-based simulation platform RePast, which is used in this research. Complex system concepts provide a new approach to the study of urban spatial pattern. One example of a complex system could be a land use system formed out of locational behaviors of people. The interactions and interdependencies between the components (e.g. households and developers), which is the heart of the system, create the collective properties associated with the land use system as a whole. Complexity theory often focuses on Complex Adaptive System (CAS) in which components interact while adapting to their environment. CAS uses computer simulations extensively as a research tool. Cellular automata (CA) and agent-based approaches are two examples. These have now developed an extensive literature; each has benefits and disadvantages. A cellular automaton system consists of four basic elements: two-dimensional gridded space, a set of transition rules defined over that space, neighbors, and state of each grid cell. In the CA model, the behavior of a system emerges from interactions between cells and their immediate neighborhood. The state of each cell is determined by transitional rules derived from the states of neighboring cells in previous time periods. Building on the groundbreaking work of Von Neumann and Ulam (1962), researchers began in the late 1980s to use cellular automata to 30

PAGE 42

explore the dynamics of urban systems (Couclelis 1989; Batty 1998; Batty and Xie 1994, 1997; Clark et al., 1997 ; O'Sullivan, 2000 ; O'Sullivan and Torrens 2000; Ren and White 1995; White and Engelen 1997 ; Webster et al., 1998; Wu 1998; Webster and Wu, 1998). It is more flexible for trying out new land use rules or constraints on urban development comparing with regression-based models because the regression does not incorporate such constraints. CA models have been tested in a variety of environments. Human behavior is generally missing in CA models, however; the agent-based approach supports the study of system evolution through accumulations of individual interactions both among humans and between humans and their environment (Franklin and Graesser, 1997). An agent-based model usually consists of some autonomous and heteronymous agents and a framework to simulate their decisions and interactions, and their adaptations to environment to produce emergent system behavior. Emergent behavior occurs because the system's behavior is more complicated than the simple sum of its components' behaviors (Holland, 1998) Agents have certain attributes and behavior rules, which can be coded into the simulation model. Some agents are able to evolve over time through learning and altering their attributes and behavior e g by genetic algorithms. They remember and learn information from their own and other agents' past decisions They interact with each other and with their biophysical environment across time and space (Franklin and Graesser, 1997) Unlike monocentric models that simplify the real world to serve the model building, and discrete choice land use models which have a very complicated equilibrium structure, CA and agent-based simulations inherently produce 31

PAGE 43

complex patterns based on behavioral rules Wolfram (2002) showed examples of how remarkably simple rules give rise to behavior of great complexity. Wolfram's work in early 1980s also provided strong evidence that complex behavior which is very much like what we observed in reality could emerge from simple CA rules Agent-based perspective distills the behaviors of agents in their daily decision making into rules. It provides a way to represent spatial actors having complex behaviors and capture directly the interactive properties of human and other systems, and the complex system behavior emerges from the interaction. It gives significance to human agents and their land use behaviors and relates them to the rise of urban spatial pattern. Agent-based models can operate interactively at different spatial and temporal scales, linking local interactions to aggregate level of land use pattern and vice versa (Moran, Ostrom and Randolph, 1998). They provide a very useful approach to allow us explore relationships between micro level individual behaviors and emergent macro-level phenomena. Applications of Agent-based Models to Exurban Development Agent-based models have been used extensively in evaluation of land cover change (Parker et al., 2003). Balmann (1997, 2001), Berger (2001) and Lim et al. (2001) apply a multi-agent approach to studies of farmer behaviors. Rajan and Shibasaki (2000) employ an agent-based model to examine the land use and land cover change at the national level in Thailand. Conte and Gilbert (1995), Drogoul and Ferber (1995), Findler and Malyankar (1995) demonstrate the potential of the agent-based approach to carry out social experiments under laboratory conditions. Gimblett (2002) focuses on the use of intelligent agents to explore human-wildlife landscape interactions. Researchers have also coupled agent-based and cellular 32

PAGE 44

automata models to explore interactions between actors and a landscape (Manson 2000, Torrens 2001 Rand et al. 2002) ABMs also have been applied to a variety of urban issues including pedestrian movement (Batty et al., 1998 ; Schelhom et al., 1999) traffic (Nagel et al. 1998) urban sprawl, residential choice in urban areas (Schelling, 1971; Portugali, 2000) evolution of settlement and urban transition (Sanders et al. 1997 ; Bura et al. 1996) and land use changes (Irwin 1998 ; Irwin and Bockstael 2002 ; Torrens 2001). The Agent-based approach shows the potential to greatly enhance our ability to model urban systems and to be a powerful tool for simulating the dynamics of exurban land development decisions. A primary benefit of agent-based models is that they are dynamic (Torrens, 2001). Exurbanites are more sensitive to space and natural amenities than their suburban and urban counterparts But natural amenities have a dynamic character : they are degraded by too much nearby development. Therefore exurban development on each piece of land greatly depends on its own state and history as well as the state and history of its neighbors By including interactions across space and time among agents and between agents and their environment, the agent-based approach can represent behavior of homeowners at a relatively high level of complexity Homeowners location choices are imitated through translating how their decisions are continuously affected by what happens around them into the rules including the decisions of their neighbors and the change of the environmental characteristics These dynamic and interaction rules at the micro level form a fundamental vari a tion from the static models. 33

PAGE 45

-A second important benefit of agent-based models is their capacity for capturing heterogeneity among agents as well as environmental conditions. Agent-based approach offers advantages when independent components in a system must inter operate in a heterogeneous environment. By including interactions across space and time among agents and between agents and their environment, the agent-based approach can represent behavior of homeowners at a relatively high level of complexity. A third important benefit of agent-based models is their capacity to capture heterogeneity among agents as well as environmental conditions Different types of households with varied preferences in exurban areas will act according to dynamic rule sets, within a varied and dynamic biophysical environment. By using decision rules other than imposing equilibrium conditions as a lot of mathematics based urban models do, agent-based models potentially offer a higher degree of flexibility for accounting for heterogeneity and interdependency among agents and their environment. Although the agent-based approach has recently been applied to a variety of problems of land use/land cover change and urban system, there have been relatively few applications of this approach to issues of exurban land development based on real data. There have been few applications of agent-based models in GIS environments, either. Summary The review of urban location theories, amenities and theory of the new West peripheral development, and some urban modeling approaches suggests the use of the agent-based approach to model the dynamic exurban household location behaviors with respect to amenities, neighbor avoidance, and accessibility. 34

PAGE 46

The central problem in this research concerns the construction of an agent-based model to simulate the ex urban residential development as a result of the dynamic relationship between amenities density and accessibilities. Each of these factors individually has been discussed widely in the literature ; however they have not been evaluated in an interactive and iterative simulation. Amenity is a key concept in this research. Darling (1973) and Brown and Pollakowski (1977) found that distance from lakes was a significant determinant of property values. Distance from the greenbelt or open spaces was tested to be negatively correlated with housing prices by some studies (Correll, Lillydahl and Singell 1978 ; Kitchen and Hendon 1967 ; Hammer et al. 1974; Peiser and Schwann 1993) A number of studies have found that scenic view adds significantly to the v alue of residential real estate (Do and Sirmans 1994; Rodriguez and Sirmans 1994 ; Cassel and Mendelsohn 1985 ; Gillard 1981 ; Rodriguez and Sirmans, 1994 ; Benson et al., 1998) Chattopadhyay (2000) found trees are important to housing prices. The amenities variables in the literature were summari z ed below in Table 2-1. Types of amenities in this research will include forestland water bodies riparian trees and V Iews. The density level people can endure is different for di f ferent types of households, landscape type and de v elopment pattern. Commuters would endure higher levels of congestions than second-home owners. Some topology and landscape type can make your neighbors hidden behind for example trees; some make them conspicuous. Organi z ation of open spaces and distance to the nearest structures will make a difference when households are evaluating density levels 35

PAGE 47

Table 2-1 Amenities Variables: Literature Variables Distance to water bodies (Lakes and Reservoirs) Scenic View (Mountain view, Lake view, etc.) Distance to greenery (green belt, forest, parks, etc.) Literature Brown and Pollakowski 1977; Lansford and Jones, 1995, Milon, Gressel, and Mulkey, 1984; Darling, 1973; Doss and Taff, 1996; Sengupta and Osgood 2003 Do and Sirmans, 1994; Rodriguez and Sirmans, 1994; Cassel and Mendelsohn, 1985; Gillard, 1981; Plattner and Campbell, 1978; Morton, 1977; Benson, et al., 1997; Benson et al., 1998; Garrod and Willis, 1992a, b, c; Do and Grudnitski, 1995; Rodriguez and Sirmans, 1994; Geoghegan et al., 199_7--"-;------1 Correll, Lillydahl, and Singell, 1978; Kitchen and Hendon, 1967; Hammer et al., 1974; Peiser and Schwann, 1993; Vaughn, 1981; Palmquist, 1992; Geoghegan et al., 1997; Shultz and King, 2001; Sengupta and Osgood 2003; Gupta,_2_00_3 ____ _, 36

PAGE 48

CHAPTER3 RESESARCH METHOD : DESIGN OF AN AGENT-BASED SIMULATION ON THE URBAN PERIPHERY This chapter describes a method for application of agent-based models to exurban development. I discuss the research strategy adopted and steps followed to achieve the objectives of the study. The chapter is divided into six parts: overall approach, research questions and hypotheses, data collection and processing, computational framework, model design, and variables. Overall Approach I start with an exploration of the relationship among preferences for amenity, density, and accessibility through a pilot study of exurban development in a 16 square miles area at the edge of Lyons, Colorado. This section is followed by discussion of theoretical models built on an abstract, artificial constructed square grid. In order to achieve an empirically validated understanding of how land use decisions of individual exurban households with different preference and under different policy regimes will affect exurban residential development pattern, I conduct a case study in the northwestern Boulder County, Colorado. Research approaches include: 1) collection and processing of necessary datasets; 2) quantitative spatial evaluation of the past land development in the area; 3) study of some modeling results from Alternative Growth Futures project applied in the area by a research team at University of Colorado (Muller, Bertron and Yin, 2002; Muller, Puccio, Baker and Yin, 2002); and 4) construction of simulations using 37

PAGE 49

ArcGIS and RePast. I examine how land use decisions made at the household level affect outcomes at a higher / aggregate level in the urban system. Specifically I build three models. The first emphasizes amenity and accessibility factors; the second stresses neighbor avoidance and accessibility ; and the third focuses on the interactive effects of amenities, settlement density, and accessibility. Geographic Information System (GIS) plays a role as a tool for data compiling, processing, and spatial database building. It also serves as a modeling tool for the first stage agent-based model. The multi-agent based modeling tool RePast simulates the temporal and spatial land conversion from one state to another according to a set of predefined transitional rules based on evaluation of past patterns of exurban land use decision-making. Research Questions and Hypotheses Why is exurban development so dispersed? I focus on three types of variables described in the literature as generating exurban dispersion: site amenity attractiveness, development density (neighbor avoidance), and accessibility. I apply an agent-based approach to model population dispersion resulting from household level locational choices and the interactive effects of amenities, settlement density, and accessibility to urban services and transportation network over time. This research seeks to understand and explain the spatial pattern of exurban development in the American Mountain West by integrating understanding gained from historical-empirical narratives, monocentric-based urban location models public goods / local amenities models derived from the Tiebout hypothesis amenities and theory of the new West periphery development, hedonic price 38

PAGE 50

theory complex system theory, and geographic information system Because of advances in technology such as GIS-based analysis ability on very large databases and availability of enhanced fine-grained GIS data with detailed information on land use patterns we now have an unparallel opportunity to expand our understanding of the exurban development dynamics Furthermore, the agent based approach put the phenomenon of dispersion in a dynamic context and an interactive behavior framework which enables us to understand additional dimensions of exurban dispersion. I employ recently developed techniques in complexity theory and agent-based simulation in this research to model exurban sprawl as a product of interaction with respect to amenities density and accessibilities. The primary contribution of this research is methodological. I e x plore a new potentially powerful way of modeling -the application of agent-based approach for exurban development. The theoretical contribution is mainly in the area of urban location theory. It extends monocentric model by including dynamic effects of location specific amenities and density. I simulate the emergence of spatial network and dynamic filling up process as representations of exurban de v elopment. How is exurban dispersion emerged from the interactive effects of amenities settlement density and accessibility? I take a building blocks approach to tackle this question by beginning with a simpler question leading to the interaction one. Accessibility is often considered the pnmary determinant of urban locational pattern In this research I begin with adding effects of amenities to accessibility following by replacing amenities with density to examine the influences of density and accessibility, and finally study the interactive effects of all three sets of v ariables. At first I examine the effects of amenities and densit y one at a time 39

PAGE 51

then a complex model with interactive effects. My three pnmary research questions are as follows: 1) What are the effects of amenities and accessibility on household location in exurban area? More specifically, what are the effects of multi-agent bidding between amenity-seeking second-home owners and commuters who are competing for exurban locations? 2) What are the effects of density and accessibility on exurban household location choices? In particular, what are the effects of multi-agent bidding between large-lot second-home owners and commuters who are competing for alternative exurban lot sizes? 3) What are the dynamic and interactive effects of amenities, density, and accessibility on exurban location? Specifically, how do development densities in one period influence location and density in a later period in an amenity-based neighborhood? Recent empirical literature suggests that exurban households tend to have more non-commuting workers than their suburban and urban counterparts and social values motivating exurban location are closely linked to natural features and systems. Amenities and density variables were found to be more important factors for exurban location than jobs. Because of some exurbanites' preferences for larger space or low density, relocation is expected after reaching certain levels of density, which push further exurban sprawl. The following hypotheses are formulated: 1. Exurban form is a result of household evaluation of the effects of amenities accessibilities, and neighbor avoidance. la.Commuter model tends to concentrate development too tightly around rural places and transportation networks in comparison with the actual development pattern. 40

PAGE 52

1 b. Amenities variables tend to focus development too tightly around natural amenities in comparison with the actual development pattern lc. Exurban development is best explained by mixed preferences of amenities, accessibilities, and neighbor avoidance. 2 Exurban development is a dynamic system defined by threshold effects as development reaches certain density 2a. There are development phasing effects at which exurban development shifts from a land market dominated by second-home owners to commuters. 2b. The switch between second-home owners and commuters is sudden because second-home owners reach density threshold. This model will shed light on the determinants and constitution of the new West exurban development and draw on the notion that urban spatial structure is determined by interdependencies among spatially distributed agents/households and the biophysical aspects of the landscape environment. This research is designed to be helpful in evaluation of timing and possible sequence of future development and in guiding policy development related to urban land use Data Collection and Processing I use the power of GIS and information technology to store process and manage data that are geographically located build up a spatial database and develop a raster GIS for the simulation. Development of a GIS includes the following steps: 41

PAGE 53

survey of data availability ; compilation of data from local governments and other sources; reconciliation of attribute inconsistencies ; re-projection to a consistent geography ; and overlay and processing to create necessary new data sets for the project. Data Collection I collected data from municipalities and county in Boulder Colorado, and other sources and produced spatial metrics. In order for a GIS to be useful an accurate complete, and meaningful database is the foundation. This requires survey of data availability and assurance of data precision so that complete and accurate datasets are assembled for the information system I also performed a qualitative spatial evaluation of some data such as longitudinal U.S Geological Survey (USGS) land Use / Land Cover data and compared it with other datasets available. After data survey raw data were derived from a variety of sources These are outlined in Table 3-1. I collected infrastructure data from U.S. Census TIGER that include local streets and highways Parcel data was from the County Assessor s Office With information on structure and land valuation attributes for each parcel it provides the finest resolution land use data available and offers way to better measure land use changes in exurbia. The map of land ownership came from Bureau of Land Management. These datasets came in different formats including shapefiles EOOs (Arc/Info export files) and coverages. USGS has digital elevation data at 1 : 24K scale available in Spatial Data Transfer Standard (SDTS) format for download on their website. Census 1990 and 2000 were collected at tract and block level that contain fine-grained, detailed spatial housing data. Census geography follows a hierarchical structure. Census block is the smallest geographic unit followed by block gr oups and tracts Block groups consist of sets 42

PAGE 54

of contiguous blocks and usually contain certain mcome data categories. Additional data was derived from a related research project, Alternative Growth Futures Project : Colorado Settlement Pattern and Wildfire Ri sk. T bl 3 1 D dS T a e -. ata ypes an ources Type Source Parcels County Assessors' Offices and County GIS Offices Streets Census TIGER 1990 data Highways Census TIGER 1990 data Municipality Boundaries Census TIGER 1990 data Census Tract and Blocks Census 1990 data Digital Elevation Model USGS Water bodies USGS Stream TIGER (U.S. Census) and USGS -Boulder County Open Space Boulder County GIS Land Ownership BLM Data Processing Significant processing on raw data was necessary to 1) check and review data accuracy; 2) convert raw data into a useable form for the model and build the database; 3) recheck the data for inconsistencies. The accuracy check involves a review and crosschecking much of the information obtained from a variety of sources. Digital database is divided into two types: vector and raster. Most of the raw data are vectors. In order to carry out many grid-based operations to build the model, I converted all the vector features to raster grids. One issue in data processing concerns making data readable by GIS. For example, Di gital Elevation Model (DEM) data files were used to generate elevation maps I got them from USGS website which made available free of charge in SDTS format 43

PAGE 55

at a scale of 1 :24k. Neither ArcMap nor Arcview can read SDTS files directly. In order to be incorporated into the regional GIS, they were converted into grids, and elevation was calculated in Arclnfo. This same operation was performed on more than 20 tiles for the county. Measurement intervals were then reclassified and the tiles were merged into a single elevation map for inclusion in the model. Unifying the representation of features is another important issue when processing data. GIS provides tools that can integrate all the different data sets within a common reference framework defmed by the geographic projection system. Geographic projections are mathematical techniques used to convert features from a spherical surface, earth onto a flat surface The datasets I collected came with different projection information. In order to line them up properly and present them in the most geographically effective manner, I performed re-projection on all the raw data. All the data layers were converted to the UTM, zone 13. After checking and assembling data and defining the right projection, data are ready to be accurately compared and converted to raster representations. The grids I converted and created have a cell size of 1-hectare or 2.5 acres. Unification issues for raster data representation relate to map extents and cell size. Most of the grid algorithms treat the grids as matrices of data or arrays of numbers All grids must have identical extents, cells size, and projection. I automated this conversiOn process usmg GIS software Arc / Info Arc Macro Language (AML) and Unix C shell scripts and produced grid layers describing the biophysical and social characteristics of the environment with which households interact. Arclnfo was used to calculate the proximity measures as well. Th e distance from each cell to highways, highway ramps local roads, and to the 44

PAGE 56

nearest developed area was calculated. Most of the processes were done using AML to automate and speed up the process Computational Framework GIS Techniques and Algorithms A variety of software is used to process data and build the model. GIS has an inherent ability to capture, combine, retrieve, analyze and display multiple data layers spatially referenced to earth During the early phases of the research, I build an agent-based model using Arc / Info Grid modeling platform and Map Algebra based on a model of exurban growth pattern: Alternative Growth Futures (AGF) (Muller, Bertron and Yin, 2002; Muller, Puccio, Baker and Yin, 2002). GIS data sets are digital equivalents of paper maps. When building a geographic information system, real world geographic data need to be translated into representations that can be stored and manipulated in a computer. Two data models that currently dominate commercial GIS software can satisfy this purpose : the vector and raster models The vector data model represents spatial data as geometric objects : points lines and polygons ; raster data model represents them as images files composed of grid cells or pixels Each pixel is a single picture element and acts as the basic unit where information is explicitly recorded. Each pixel or cell is assigned only one value that represent a geographic phenomenon, e.g. if it is within certain municipal boundary. The matrix of cells is called a grid and are also represented as arrays of numbers The resolution of the grid depends on its cell size. The smaller the cell size the higher the resolution. The strength of the raster data model is its simplicity and ease of use in mathematical 45

PAGE 57

computations. Most GIS operations can be performed on both data models and conversion from one to the other is easy A GIS spatial database includes layers of spatial features such as roads, streams, and administrative boundaries Any geographic object on a digital map has detailed information (attributes) stored in data tables linked to the digital spatial database. GIS software has tools for organizing information on the spatially defined features and integrating all the different data sets so that data for the same objects on different data layers can be related and combined for analysis and mappmg purposes GIS also allows for the creation new data layers using the existing ones Arc / Info has the most comprehensive functionality among the ESRl's ArcGIS software family It includes all the functionality of ArcView and ArcEditor. It is a complete GIS data creation, update query mapping, and analysis system and it is de facto standard for GIS. The raster-based geographic information system, Arc / Info GRID, provides a powerful framework and tools for model development. All data in the Arc/Info Grid module are referenced to a fixed location that is, a cell with a size defined by users and position defined in terms of x and y coordinates. The grid module was used to store and construct a raster-based spatial database. With these tools, I performed data conversion generalization aggregation overlays, buffer creation, statistical calculations, and much more. There are many factors that drive people to make decisions with respect to building a house on a particular site for example, environmental and ecological constraints existing land development, accessibility and policy constraints. However, display of all factors together on a single map is problematic because of 46

PAGE 58

too much information. Raster based map overlays help ensure that all the information about the same place can be consolidated. A raster database can contain many congruent raster layers with an identical geographic extent and the same number of rows and columns Each layer shows one theme and contains a set of data describing a single characteristic for each cell. Each cell is coded with integer values such as distance to the nearest local road the year a house was built in the cell, and whether the cell is located on a 200meter public land buffer. Thus, a household can move anywhere on a grid and is associated with information such as distance to the nearest local road, whether the cell is built out, and whether the cell is coded as public land buffer. These are some of the questions a household will ask when looking for a site for development. Consolidations of layers combines information about a single cell relevant to land use development decisions (Figure3-l ) Figure 3-1 GIS Raster Overlay 47 Parcel Road Public Land Buffer Stream Buffer

PAGE 59

Raster overlays are faster and easier to work with mathematically than vector overlays. Grids can be created from arc or polygon overages shapefiles, or by application of statistical or logical functions to other grids. Arc/Info Grids provide a set of very powerful tools for the geographic data analysis. However, creating and manipulating grids could be a daunting task. Map Algebra is a data manipulation language designed specifically for geographic cell-based systems. Map algebra provides a high-level language interface to describe and manipulate data. Arc / Info Grid uses map algebra to increase its computational capacities and enhance its mathematical and spatial functions. Grids allow combinations of various functions operators grids, numbers, and scalars. Models can be computed on a cell-by cell basis on columns and rows in a clockwork manner. A docell block was used to control per-cell processing on a cell-by-cell basis within the analysis window A sequence of operations were performed at a cell before moving to the next. The docell block also allows the use of IF and WHIT.,E statements to support models that require iterative processes at the individual cell level. Therefore, it allows researchers to examine and analyze the features for each cell. The following example shows that all the cells that are not on the open spaces were examined to see if they were developed before 1990. If not, they will be considered for development under the condition that they are located in the stream buffer docell if (K:/forest/lyons / grid/openspace eq 0) { if (K :/ forest/lyons / grid/dev90 <> 1 and 48

PAGE 60

} end K :/ forest/lyons / grid/streambuf = 1) { } K:/ forest/lyons / grid/devcom = 1 count+= 1 After considering a number of options the Arc/Info Grid modeling platform Map Algebra, and ARC Macro Language (AML) were chosen in my first stage model. The advantage of using this model is that it does not require data transformation input and output. However as information is needed for each cell and the number of cells used in the model increases grid data sets tend to take up too much file space and slow down the calculation speed Moreover modeling results are not directly viewable Data need to be joined and mapped in Arcview or ArcMap. Agent-based Modeling Platforms Arc / Info Grid execution times become unacceptably long as models become more complex Other computer programming languages that implement multi dimensional arrays were considered to reproduce the spatial analysis performed in GRID for example Java. In order to better implement the agent-based model I reconstruct the model code on a more robust computational platform Modeling platforms developed especially for agent-based models include Swarm created by Nelson Minar and Chris Langton at Santa Fe Institute ; Kenge developed by Paul Box and colleagues at Utah State University; Ascape by Joshua Epstein and Robert Axelrod at the 49

PAGE 61

Brookings Institution ; Starlogo by Mitchell Resnick a t the MIT Media Lab; and RePast by the Social Science Research Computing at University of Chicago I reviewed several such platforms and scripting languages and decide to move much of the existing code to a more flexible platform based on Java that is, RePast. RePast is a Java language based software framework for creating agent-based simulations. It is developed by the Social Science Research Computing at University of Chicago based on Swarm, a software package for multi-agent simulation of complex systems used mostly for physical and biological sciences RePast is an acronym for Iterative Porous Agent Simulation Toolkit. It provides a class library to help collect create run and display data for an agent-based simulation. In addition, it has ready-to use mechanism for taking snapshots of running simulation or even creating QuickTime movies of simulations I downloaded RePast system versiOn 2 0 1 from the website www.repast.sourceforge.net and installed it together with Java 2 .1. To develop models using RePast I acquired a copy of Java Software Development Kit which includes a java runtime required for running the RePast demonstration simulation In RePast, an agent-based simulation proceeds in two stages The first stage sets up the simulation ; the second stage runs it. Simulation is divided into ticks" or time steps. The programmer needs to describe what happens during the setup and every tick and place the information in a class called SimpleModel. There are typically at least two classes in RePast, agent and model classes The agent class describes the behavior of the agent (household in this research) and the model class coordinates the simulation setup and running of the model. Agents are implemented according to the properties of exurban households described latter in this chapter. It is 50

PAGE 62

facilitated by the used of Java interface, the Drawable interface. I placed agents in the RePast space by implementing the methods like "draw". SimpleModel is served as the basis of my model class and extended to suit my specific simulation purposes. Setup(), buildModel(), and step() are the three methods I needed to create the simplest simulation satisfying my purposes. I used them as follow. import uchicago. src.sim. engine. Simp/eM ode!; public class LifeModel extends SimpleModel { private ArrayList birthList =new ArrayListO; private Space space; private int width; private int height; private DisplaySurface dsurf; private Displayable display; public void setupO { super.setupO; width= 150; height= 150; dsurf = new DisplaySurface(this, "Northwestern Boulder Display''); } public void buildModelO { space = new LifeSpace ("E:I agent l lyonslmodel/lyons4/ ascii / s treambuf asc ", this, "E:/ agentllyons l model/lyons4/ ascii /y btpcl90 asc .. .); display= space.getDisplayO; ((Object2DDisplay) display) setObjectList(agentList) ; } 51

PAGE 63

public void stepO { } } int si z e = agentList.sizeQ ; space.ste p(agentList); Setup() is used to do housecleaning It is called when the simulation is first started and whenever the setup button is clicked on the interface BuildModel() is used to create the objects that my simulation needed to use including agents and space in where they operate. Agentlist is an array to keep all the agents LifeSpace is the raster space which agents inhabit. Now that the setup stage is finished SimpleModel calls step() method for each tick of the scheduler Step() tells the simulation what need to be done for each tick or time step. In the step() method I iterated through all the agents (households) and also called relevant method to execute their beha v ior on each one Some action occurs at each tick according to the results of pre v ious actions the current v alues of all the agents variables and space in which they operate. The history of the simulation is the history of the states of all the ticks Any changes to the states occurred are recorded and used as input for the next tick For example if there is a house built in a neighborhood this information will be recorded and reflected immediately in the next tick so that for example households who are sensitive to the high density around them would have an opportunity to reconsider their development decisions It is one of the advantages of using RePast to build agent-based exurban household location model. Neither discrete choice land use model nor agent-based model we developed in the first sta g e using Arc/Info grid can do this efficiently. 52

PAGE 64

RePast also pro v ides a graphical user interface (gui) to allow an interactive simulation User can start stop pause setup view settings, and exit a simulation through the toolbar In addition it has a Step" button that would allow users to simulate by repeating single iterations In the settin g s window there are Custom Actions and Repast Actions tab I used the first two buttons in the RePast Actions tab to make movies and take snapshots. When the setup button is clicked the code in the setup() method is e x ecuted while if step button is clicked the code in buildModel() and step() is executed. RePast versiOn 2.0 introduced interoperability with geographical information systems The space on which agents operate can be imported from GIS. ArcGIS raster files were exported as raster ascii files RePast can read them in as RasterSpaces a class defined in RePast. Model Design Overall Design Many urban growth models are static in nature They usually refer to a single or two points in time and build on longitudinal data focusing on these historical points separated by five or ten years with no information on the intermediate period Such models ignore the transformation process from one time point to another. However, urban development is a cumulati v e and path-dependant process. It is important to understand and include inherently dynamic nature of growth in urban models. Development on each piece of land depends on its own state and history as well as the state and history of its neighbors Because development dynamics are poorly represented in these models they are inefficient to investigate 53

PAGE 65

complex dynamic relationships between components of the spatio-temporal urban development. Some new types of models developed recently based on complexity theory on the other hand can include the changes and interactions of model components over time. These dynamic simulation models can be used to represent behaviors of homeowners and the evolution of every individual parcel at a relatively high level of complexity by using a process-based approach They attempt to capture the myriad component interactions that comprise the development of an urban system. Homeowners keep discovering the continual changes in their neighborhood within such models. The interpretation of urban development is evolving from a static, fixed model of concrete relations to a more dynamic model of fluid and complex systems. These models have a better chance of demonstrating the possible results of policy changes. Such models have recently been applied to a variety of problems of land use / land cover change such as what was discussed by Parker et al. (2003), residential segregation (Schelling, 1972), and to simulate urban systems (Torrens 2001). In this research, I use agent-based approach to simulate land use dynamics in exurbia. Theoretical models are built in three phases: Model I: Assess the static effects of amenities and accessibility on location. Model II: Evaluate the dynamic effects of lot size preference on long time density change. Second-home owners were assumed to be space sensitive. They like large lot while commuters are space-neural. This phase focuses on the influences of density and accessibility Model III: Evaluate the effects of amenity and lot size preference on long time location and density change In this phase, I build a 54

PAGE 66

dynamic model to incorporate interactive effects of amenities, settlement density and accessibility. The theoretical models are built on an abstract grid. An ASCII file is created and imported to Arcview to create a grid of 150 x 150 cells, with a resolution of 100 x 100 meters per grid cell In order to run the above three theoretical models road network, two small towns assumed to be within 8 cells, public owned lands, a lake, and some streams are drawn randomly and added to the grid This grid sets up the basis for creating the following variables: distance to road, distance to city, distance to public land, distance to lake, and distance to stream. Model verification and validation are essential parts of model development process. Mter the initial model development, I engaged in a series of activities such as debugging, verification, model review, and validation before further model development. Debugging involves the use of various techniques to determine the cause of a bug and fix it. I validate the theoretical models by evaluating how they work empirically in the study area to assure that they represent the real system to a sufficient level of accuracy. I collect performance measures of the system for defined periods of time, run the model over 100 times with the given input data, and compare model outcomes to the real-world observations over the given period of time. A series of tasks were carried out to verify and validate my model to ensure that they are sufficiently accurate with reference to the purpose of this research; however, it should be noted that no model is fully verified and validated, and no model is 1 00% accurate. Any model is only a representation of a system 55

PAGE 67

Interaction Framework The five building blocks of my agent-based model are agents, states neighborhood environment and rules. I designed a conceptual agent-agent and agent-environment interaction framework using these five components and mapped it onto an computer model to simulate agents' decisions concerning the state conversion of a piece of land following certain rules (figure 3-2) Figure 3-2. Framework of Agent-agent and Agent-Environment Interaction Land Use Policies I ____ _J __ Neighborhood rn ... ,.. ____ ........ rn I Households I + 56

PAGE 68

In my model, agents represent households who are searching for exurban residential locations based on certain rules. Agents have heterogeneous attributes and behaviors. In this research, I consider only two types of households: commuters, and second-home owners / telecommuters. Commuters' decisions are based on accessibility factors, site characteristics and urban proximity. Households from different groups interact with other households and respond to natural amenities and other factors. Household behavior is also a reaction to the macro level land use policies. By evaluating these properties of households, I can portray their behaviors at a relatively high level of complexity and realism (see Table 3-2) T bl 3 2 P f h ld fH a e -. roper 1es o ouse o s Objective-orientation Search for an exurban residential location. Heterogeneity Two types ofhouseholds (commuters and secondhome owners) behaving differently Varied biophysical environment. Household-household interactions. Interaction Respond to the characteristics of the environment. Respond to the macro levelland use policy (i.e. wildfire mitigation policies). Agents are typically individual programming objects that respond to a variety of social and environmental information and filter this information through a set of rules that govern decisions about household location and mobility across a landscape. Agents keep searching for exurban residential locations which satisfy their requirements. Each grid cell has one of the two states: developable and undevelopable. Undevelopable cells include those on roads water bodies, and public owned land, and those occupied by other agents. Agents are allowed to locate only on developable cells in the simulation 57

PAGE 69

-The decisions of agents in a neighborhood influence and are influenced by other agents in that neighborhood because the presence of one household affects residential location decisions of another. I assumed that some households prefer to be close to other households for a number of reasons. For example, commuters have a preference for neighbors because of the presence of schools and services after a certain level of clustering. On the contrary, some studies indicate that the location preferences of some households, like second-home owners, are strongly shaped by a desire for more space, natural amenities, and a sense of isolation (Davis et al., 1994; Nelson, 1992; Riebsame et al., 1996). They are more sensitive to levels of noise, scenic views, and congestion. Therefore, they are repelling rather than being attracted by other households, which create pressures for low density development. These differences between households and their decision making are reflected in the decision rules of the agents. Two major ways of defining neighborhood in the literature are the Von Neumann and Moore neighborhood (Epstein and Axtell, 1996) In my model, neighborhood is an area of 13 by 13 cells, that is, an area of 169 hectares. The environment is a two-dimensional array of regular spaces represented as a mosaic of grid cells. It is the virtual space agents live in and interact with based on decision rules. It is the combination of the developable land layer and cell characteristics layer. Cell characteristics include: 1) site amenities; 2) Measures of contiguity and fragmentationneighborhood effects; and 3) accessibility factors. The advantage of this modeling framework is that it provides a comprehensive portrait of actual land use change under market and policy conditions present during the study period. Thus, the allocation and visualization of exurban growth is based on current land market patterns. Moreover, this approach relies heavily on the evaluation of decision-making by individual landowners. It describes land 58

PAGE 70

markets in terms of the preference of households and characteristics of specific site that make them attractive to different households. In this respect, the model can be explained relatively easily to a layperson I develop models in both Arc / info Grid platform and RePast using agents who have exurban locational preferences based on this framework. They exist on a heterogeneous landscape which is defined using data stored in the geographic information system described early in this chapter. It is a two dimensional irregular square lattice. Design ofRules There are two main objects encoded in the model : cells and agents Agents or households select exurban sites or cells following a variety of land conversion rules The rule structure consists of two main components. They are the different locational preferences of second-home owners and commuters (figure 3-3). Rule sets vary among different types (groups) of households to guide their behaviors according to preferences Preferences are determined by the household types and the three types of independent variables Sets of preference functions were constructed with respect to the independent variables, including preference for natural amenities, neighborhood, and accessibility to employment centers, services, and infrastructure While some of the independent variables were introduced into the model to determine the non-homogeneous nature of the physical space with which agents interact and where the land use dynamics unfold neighborhood effects variables and associated household preferences are very important parts in the model because they reflect the dynamic impact of land uses in the immediate surrounding of a location. 59

PAGE 71

-F i g ur e 3-3 Component s of Rul e S tructur e Preference S econd hom e O w n e r s Amen i ty Seeking, l ow 1.,. ..... __ -+----+--..r .... d ensity, etc. ...,. .,... C ommut ers Comm u ting D i s t ance, etc. At the first stage, I build a very simple model to explore rules in a limited geographic test area. I design a number of rules based on some important characteristics of the sites which most likely attract households according to the literature review and results of an earlier model: Alternative Growth Futures (Muller Bertron and Yin 2002 ; Muller Puccio Baker and Yin, 2002). Rules are coded so agents are able to navigate through geographical environments and make decisions on which piece of land they want to develop Each cell on the grid is given two summary preference scores: one for each agent type The preference score is determined by the presence of their neighbors level of services and the biophysical characteristics of the environment. I use various rating weighting ranking, and map overlays techniques to create scores related to probabilities of urban transformation. Agents are programmed to look at each cell, add up the weighs compare scores randomly select cells among top-ranked cells and record the development for each year. The score is continuously changed by agents' decisions every time step Through the iterative application of the rules on the households for twenty consecutive years (1980 to 1999) I generated the development pattern for 1999 In the second stage I build some theoretical models on a more flexible platform (RePast) using Java and further explore household land conversion rules according 60

PAGE 72

-to some locational preferences (Table 3-3). In this phase, models are built on abstract grid spaces. The first model is designed to detect the static effects of amenities There are two types of agents or households: commuters who value short commuting distance most, and second-home owners or amenity-seekers whose locational choices are strongly influenced by site amenities, for example, presence of open space or stream, or what the scenic view is from the site (Figure 3-4) The second model simulates dynamic effects of density with respect to location. Commuters were assumed to be density-neutral, but second-home owners are very sensitive to space and skip over properties to obtain bigger lot further out (Figure 3-5). Table 3-3 Locational Preferences by Different Types of Households Locational Preference Second-home Owners Commuters Priorities First Proximity to Public land Proximity to Jobs or Second Third or Lakes or Streams Quite environment Large Lot. Proximity to Roads Shopping Fi2ure 3-4 Rules for Model I Commuters: 1 Determine developable sites. 2 Determine accessibility to roads and towns. 3. Determine accessibility to natural amenities. Highways or High Level of Development in the Neighborhood or Proximity to Open Space or Lakes or Streams 3 Find location to satisfy: a) distance to roads and towns is minimal; b) natural amenities score is highest. Second-home Owners: 1. Determine developable sites. 2 Determine accessibility to natural amenities. 3. Determine accessibility to roads and towns. 4. Find location to satisfy: a) natural amenities score is highest; b) distance to roads and towns is minimal. Bidding: Two types of households bid with their preference score. 61

PAGE 73

Figure 3-5 Rules for Model II Commuters : 1 Determine developable sites 2. Determine accessibility to roads and towns. 3 Find location where distance to roads and towns is minimal. Second-home Owners: 1 Determine developable sites. 2. Determine locations of other households 3. Determine accessibility to roads and towns 4 Find location a) far away from other households ; b) distance to roads and towns is minimal. Bidding: Two types of households bid with their preference score. The third model better represents the complex spatial location behaviors of households in exurbia. Second-home owners have a higher level of preference for space in the amenity-rich areas and it dynamically affects their location and relocation (Figure 3-6). After reaching certain le vel of density, households begin to respond by either not to move in or move out. Figure 3-6 Rules for Model III Commuters : 1 Determine developable sites 2 Determine accessibility to roads and towns. 3 Determine level of development in the neighborhood. 3 Determine accessibility to natural amenities 3 Find location to satisfy : a) distance to roads and towns is minimal; b) Level of neighborhood development is high; c )natural amenities score is high Second-home Owners: 1 Determine de v elopable sites 2 Determine accessibility to natural amenities. 3 Determine locations of other households. 4 Determine accessibility to roads and towns 5 Find location which balance the preference for amenities and space and accessibility to road or town. Bidding : Two types of households s bid with their preference score 62

PAGE 74

-Agents visit each cell on the grid and find out if it is developable and then look at other attributes of the cell. Households do not make any decision until they finish surveying all the cells to see the potential of each for development. Adjacency to natural amenity such as public land and streams, and accessibility to road are desirable site features. Next agents compare and rank relative probabilities of urban conversion for each cell. Each cell has two sets of development probabilities: one for second-home owners and the other for commuters. Because households are boundedly-rational and have limited information, a random mechanism is included in the model. Second-home owners and commuters bid for cell locations The cells ranked in the top 100 highest probability are selected randomly for development. It may be a second-home owner or a commuter depending on the set from which the highest probability is chosen. Development information is immediately recorded after the event and affects the decision of other households for the next iteration. Each time step or iteration in the RePast model is not one year or one month. It is the time period it takes for the next development activity taking place This is one of advantages of using agent based model because the desirable or repellent land uses are of great significance for the quality of a location and for its appeal to particular types of household For each time step, agents assess the quality of its neighborhood a 13x13 cells area around each particular cell and the distance from each cell to its nearest neighbor. New activities and land uses occurring in a neighborhood over time change its attractiveness level for households (Figure 3-7). Only one household moved in each time step in RePast. The total number of residents moved into the landscape is determined exogenously by the results of demographic analysis and analysis of the assessor's parcel data Finally I test and validate theoretical models empirically in the study area 63

PAGE 75

Figure 3-7 Pseudo Code of Agent-based Model Agent and Environment Initialization Begin Time Step 1 to N .. Visit Site N/Cell N Visit and Get t Info. for t h.e r-----G-a_t_h_e_r_I_n-fo_r_m_..a_t-io_n __ o_f_S_it_e_N __ _, Next Cell Market Interactions Randomly Develop a Cell ranked Top 100 Bio-physical Environment Interaction with Other Households ... Bidding and Agent .. Chose Actions 1 Update Site Information Begin the Ne t Time Step Case StudyN ali dation Theoretically, agent-based models have a very high potential to be a powerful tool for simulating the dynamics of exurban land development decisions. They in theory can provide a qualitative description of land use development. However, to date these models rely on stylized heuristic decision rules not derived from empirical investigation Rather than building a model on an abstract artificial constructed square grid with fabricated rules as most of agent-based urban development models do this research begins with building simple empirical models in a small area around Lyons, Colorado to explore rules, followed by some theoretical models and finally tests and validates the theoretical model in a study area in northwestern Boulder County Colorado. 64

PAGE 76

-In order to explore rules, I build two agent-based simulations at high geographical resolutions using GIS data layers in a 4 by 4 mile area west of Lyons Colorado Starting with a small area not only makes it easier to study the behavior of the model and play with rules, but also allows me to better understand which phenomena are idiosyncratic to the study area. Lyons is a town of about 1585 people (U.S Census 2000) located 18 miles northwest of the city of Boulder Colorado and 20 miles east of Rocky Mountain National Park. It sits in a mountain foothills area surrounded by hills of ponderosa pine and red sandstone with mild climate and lots of sunshine It is well known as the Double Gateway to the Rockies because of the two different roads leading to Estes Park; Colorado highway 7 winds up to Estes Park from the south and U.S. highway 36 goes directly north to the park. Lyons is now feeling the impacts of growth from the Denver metropolitan area. Development has sprawled into Lyons' surrounding areas. Lyons area is an ideal study subject to begin this research because of its unique location and the picturesque landscape of rocky red hills the rushing St. Vrain River, agricultural lands and mountain vistas It is a beautiful little mountain community at the edge of a metropolitan region the mouth of the St. Vrain River, and the convergence of Colorado Highways 36 and 7 It has a strong lure for people looking for places with advantages of both being close to work and to natural amenities Most of the residents in and around Lyons are commuting to work in Boulder Longmont and other metropolitan employment centers (Figure 38) 65

PAGE 77

RuraJ..!I:Jrban Comm u t i ng A r a s II Me! r opo l an area earn Metropo l an area h h comm u g Metropo l flan area low commuling II large town core II large lown lhtJth comm u g large tm low commuUng Sma towncore Sma town hl,gh commuUng Small town low commuling Rural areas Areas Source: USDA Economic Research Service I evaluate the land use development pattern from 1950 to 2000 in this area. Parcel data is used to learn how land use changes each year. Information on year of building each structure is the study focus in parcel data Other maps such as land ownership local streets highways and streams are useful as well to understand how land use change is influenced by the presences of public land streams and acc e ssibility to roads Models are built in two stages using Arc / Info Grid First I build an accessibility model emphasizing commuting and access to services and then a model including other variables derived from amenities and the theory of the new West periphery de v elopment. In the accessibilit y model the primary variables are travel distances to transportation network and urban areas In the second model primary variables 66

PAGE 78

relate to neighbor avoidance and amenities, as well as travel distance. Both household agent types appear in the second model. In addition, I run the simulation model iteratively (year-by-year) in the second model in order to capture dynamic effects. The number of cells developed at each iteration is derived through the study on development pattern mentioned above. It equals to one tenth of the total change number from 1981 to 2000. Through the iterative application of the rules on the households for ten consecutive years (1981 to 2000), I generate a development pattern for 2000. The theoretical models are tested and validated in the study area at the final stage The study area is the mountainous part of Boulder County (Figure 3-9). It is next to and west of the Lyons area. As in Lyons, most of the residents in the study area are commuting to work in Boulder, Longmont and other metropolitan employment centers (Figure 3-8) The study area is situated in several valleys in the foothills of the Rocky Mountains. It reaches the city limit of Boulder and Lyons in the east Larimer County boundary in the north, and Nederland in the south. Land available to development in the study area is sharply limited by land ownership and topographical constraints. Much of the area is public-owned including large blocks of public forest land and Boulder County open space. It is an area with very high natural amenities because of picturesque landscape of Rocky Mountains, numerous of stream and lakes, and wildlife It has convenient access to National Forest, Rocky Mountain National Park, and metro job centers. Housing market pressures are expected to intensify as a result of continuing population spillovers from city Boulder and Longmont, continued retirement and amenities-seeking, and growth in employment opportunities in the metro area. 67

PAGE 79

Figure 3-9 Study Area The third theoretical model is validated in the study area from 1981 to 2000. The number of household increase is obtained from assessors' data. I collect data necessary for creating amenities, neighborhood effect, and accessibility variables in this area for the same period. After getting information for the study area, I run simulations over 100 times using the input data over the given period of time in order to generate or average a pattern to compare with the real world development. Variables The dependent variable for the first research question IS measured on land conversion from rural to urban land uses. This is determined based on the "year built" attribute in the assessor's database. In other words, conversion occurs 68

PAGE 80

during the year in which the assessor indicates a structure is built on a parcel. This information is carried into the grid cells when converting vector data into raster data. For the second research question, it is development density. This is measured by the total number of developed cells in a 169-hectare window (as divided by the total number of the cells in the window). The dependent variable for the third research question is measured by the conversion from one household type to another Independent variables {Table 3-4) were selected with reference to literature review, and current growth modeling practice as represented by projects such as the California Urban Futures Model developed at the University of California Berkeley and the SLEUTH model developed at the University of California-Santa Barbara Three primary types of independent variables are included in the model: 1. Site amenities -These variables are designed to capture the attractiveness of the site itself to households. Variables include presence of 200-meter stream buffer and trees, distance to federal and state land, distance to county open space, and viewshed. All of the variables were found to be significantly related to housing choices by previous studies. Viewshed is a recently coined term used to indicate the entire area an individual can see from a given point. It is characterized by visibility between locations. Viewsheds was generated by GIS based on topographic analysis and tested by field visits. I conduct a viewshed analysis based on the above information. The aim is to allow the visualization of what and how far a person might see within the abstract lattice when standing at the center of an area (figure 3-10, 369

PAGE 81

11 ). Calculations are based on assumed eyes or cameras at 1 meter high from the surface. Because of the tremendous calculation tasks and CPU time involved in the analysis, I calculate only the viewshed for each 9 x 100 x 100 meters (9 cells) area This process begins with placing observers in the center of each 9-cells. Visual units include lake, public lands, mountain peaks, streams, and general areas. If the lake or peaks of mountains are visible from the site, the view quality of the site is considered better than other sites where none of natural amenities are viewable. The result is a grid with a score of view quality of each cell. 2. neighborhood effects-After testing several measures of contiguity and fragmentation, two measures were introduced into the model: distance from each cell to the nearest urban development and the number of developed cells in a neighborhood. These variables provide an indication of the significance for development, based on proximity to the neighbors and development densities. 70

PAGE 82

71

PAGE 83

Figure 3-11 Viewshed Illustration Visible Source: Institute of Water Research Michigan State University 3. Accessibility factors These measure distance from each cell to nearby highways, roads and streets, and highway ramps They provide a crude indicator of the relative costs of extending roads or streets to service a site, as well as travel times for commuting or shopping trips Independent variables can also be grouped into static and dynamic factors determined by whether they are changing in the simulation Static variables are constant throughout the simulation. For example, some global configuration of the grid space like where the streams are and where the lakes are. Dynamic variables change at every time step of the simulation, which reflect the consequences of land development and cell state transitions for successive states. An example is distance to nearest neighbor. 1. Dynamic variables -Neighborhood effects. 2. Static variabl e s Site amenities and Accessibility Factors 72

PAGE 84

The unit of analysis is a developable cell defined by site characteristics like whether it is on the right of ways, waterways and whether it is occupied by other households). The model relates land use change to site amenities and other attributes of the cell including proximity to regional freeways boundaries of developed areas and whether the site is within a 200-meter stream or water body buffer. 73

PAGE 85

Table 3-4 Definitions of Variables Variables Definitions Dependent Variable Change Binary variable. 1: land conversion from rural to exurban; 0: no change Amenity Variables Distance to Water Bodies Continuous variable: Euclidean distance to lakes and reservoirs. Distance to Streams Continuous variable: Euclidean distance to streams. Viewshed Binary variable. 1: cells an individual can see from a site. 0: cells an individual cannot see from a site. Distance to Public Land Continuous variable: Euclidean distance to federal or state owned land Distance to Open Space Continuous variable: Euclidean distance to county open space Neighborhood Variables Density ofNeighboring Continuous variable: Areas Developed Number of developed cells divided by total number of cells in a 169 hectare window Distance to Developed Continuous variable: Neighbor Euclidean distance to the nearest developed neighbor Accessibility Variables Distance to Roads Continuous variable: Euclidean distance to the nearest local road Distance to Highway Continuous variable: Euclidean distance to the nearest highway Distance to City Continuous variable: Euclidean distance to the nearest city center/subcenters 74

PAGE 86

CHAPTER4 DISCUSSION AND RESULTS: AGENT-BASED EXURBAN HOUSEHOLD LOCATION MODEL This chapter is divided into two sections. Part one discusses the three theoretical models in detail and presents results of the simulations. These models are designed in a way that disaggregated time, space, and agents are tackled one at a time and then integrated into a dynamic and interactive framework using a disequilibrium approach. In the second part, the third theoretical model is validated with a case study in northwestern Boulder County. Theoretical Models Theoretical models are built on an abstract grid with 150 x 150 cells, with a resolution of 100 x 100 meters per grid cell as shown in figure 4-1. This grid is considered the initial state of the development or the development at time step 0. It sets up the basis for simulating each of the three theoretical models. 75

PAGE 87

F igur e 4 1 Ini tial State Initial State Model I, II, and III each consist of two types of households, commuters and second-home owners, which encapsulate the behaviors of the diverse individuals that make up the system. Households' preferences and behaviors are adjusted in each of the models in accordance with the purpose of the model. Commuters and second-home owners enter the environment (the abstract lattice) and interact with it. One of the households takes up one site or cell in each time step (iteration) depending on their preference bidding. I run simulations, collect, and record the development pattern for each 30 time steps, that is, a snapshot is taken after every 30 households choosing their sites on the abstract grid. The results are presented in this chapter at the 60th, 180th, 360th, 600th and 780th time step respectively with some variations for each model. 76

PAGE 88

Model I Model I attempts to assess the static effects of natural amenities and accessibility on household exurban locational choice. The two types of households in this model, commuters and second-home owners, have corresponding preferences for accessibility and amenities. I expect that a model based on a mix of preferences for amenities and accessibility provides a more accurate predictor of development than a accessibility model. Model I is built in two stages. The first stage contains homogeneous residents, and the second stage or the full model has both of the two household types. Results are presented as two sets of snapshots from these two stages. With the intention of looking at dissimilar effects of amenities and accessibilities, I set preference for amenities to null for all households in the first stage and display the results from the time step 60 and 600 (Figure 4-2). The result shows that build-out first occurs in the areas around the rural places/the employment centers (Step 60) and then extends to the part along the transportation corridors (Step 600). Accessibility factors pull all the development towards the places with a high level of accessibility. This pattern is persistent from time step 60 to 600 It illustrates that if all the exurban households have the same simple preferences for the proximity to job centers and road network, at the aggregate level, exurban development patterns would be similar to a monocentric pattern Next stage shows what the exurban development patterns would look like if we add amenities variables into the simulation. 77

PAGE 89

New Development I I Stream -Road 78

PAGE 90

In the second stage (Figure 4-3), commuters and second-home owners bid for land and develop one site / cell in each iteration. Commuters give preference to sites that are close to job centers and transportation network. Sites are favorable if they are close to natural amenities as well. On the other hand, second-home owners desire sites that are close to public land water bodies, streams, or sites with great views; less attention is given to accessibility Figure 4-3 demonstrates that because second-home owners chose to develop places in close proximity to natural amenities, clusters emerged not only along transportation corridors and around job centers but also in the areas with rich natural amenities (i.e. lakes and public land) or with easy access to both road and natural amenities From time step 60 to 600, the pattern is consistent. This model shows how diverse land use decisions made by different types of households at the micro level affect outcomes at higher level in the exurban development. The mix of preferences for amenities and accessibility makes Figure 4-3 look more similar to the exurban development usually seen in the real world than the accessibility model (Figure 4-2). The inclusion of the amenity variables clearly makes the model more explanatory Nevertheless, the dynamics of exurbanization is still not adequately represented in this model. Although model I includes heterogeneous agents and bio-physically varied environment agents/households only respond to the bio-physical environment around them, not the characteristics of their neighbors or level of development in the neighborhood It is important to include these factors in the exurban simulation because development always concerns the growth of new activities close by and adaptation to these changes. Household-household interaction is missing in model I. 79

PAGE 91

4-3 Model I -New Development I I Public Land -Lake I Stream -Road Step 60 Step 180 80

PAGE 92

New Development I I Stream -Road 81

PAGE 93

Model II Model II evaluates the dynamic effects of lot size preference on long time density change. I assume that second-home owners are very space sensitive that is they do not like to live close to each other. They prefer large lots while commuters are space-neural. This phase focuses on the influences of density and accessibility. The results of the simulation are presented in Figure 4-4. It displays development from time step 60 to 780 in five snapshots Commuters chose areas near cities / rural places and roads as what they did in Model I. Second-home owners in this model favor spacious areas with some consideration on accessibility. At the early stage when there are a large number of empty places v arious sites or cells can satisfy second-home owners to a great extent. Therefore, they offer higher bid and win their bids more often At time step 60, most of the developments are scattered as a result of second-home owners' bid triumph. Time step 180 shows the development in two extremes: clustering on cities and roads and dispersion with some tendency to be close to roads. We also notice that there are more cells developed by commuters than by second-home owners from step 60 to 180 At time step 360, scattered development is pushed into the areas away from roads because of lack of space When development gets more and more densified and accessibility clusters are stretched out fewer and fewer cells suit second-home owners' needs while commuters can still find many sites that highly satisfy their needs. From time step 360 to 600 to 780 large clusters emerged around areas with good accessibility. However, very few (4) cells are developed by second-home owners in the empty spaces between previous developments In some areas no second-home owners moved in after step 360 82

PAGE 94

Since every second-home owner attempts to avoid other development and skips over properties to obtain bigger lots further out, they all together create pressures for low density development and make the dispersion pattern persistent at the aggregate level. However, second-home owners' densification process slows down over time and will eventually stop after reaching certain density levels This is because these households are not satisfied with the empty cells available for development after a certain time and therefore, are not willing to bid high enough to get them developed. Timing plays an important role in this model. Figure 4-4 illustrates how a small behavior preference for neighbor avoidance and larger lots over time at household level leads to significant and disproportionate reduction in average density of development. 83

PAGE 95

4-4 Model II New Development I I Stream -Road 84

PAGE 96

New Development I I t==J Stream -Road 85

PAGE 97

-New Development I I Public Land -Lake I Stream -Road Step 780 Model ill Model I and II examine different influences of accessibility, natural amenity, and neighborhood density/neighborhood avoidance on exurban household locational behaviors. Model ill assesses the effects of amenity and lot size preference on long time location and density change. In this phase, I build a dynamic model to incorporate interactive effects of amenities, settlement density and accessibility. Commuters tend to favor sites that are close to job centers and road network most, but they also consider factors like proximity to natural amenities, space, and level of development in the neighborhood due to the potential development of schools and services after a certain lev el of clustering is reached. Location preferences of second-home owners are strongly shaped by a desire for natural amenities, such as 86

PAGE 98

-public land, water bodies, streams or great views They also would prefer to avoid each other and be somewhat attracted to road access. Figure 4-5 reports the simulation results from time step 60 to 780. It shows that development first takes place in areas rich in natural amenities, or with easy access to roads and cities or both. After some time, a degree of clustering emerges around these areas (step 180). These are the outcomes ofhousehold level decision making: commuters prefer cells with good accessibility while second-home owners prefer natural amenities Noticeably, from time step 360 to 600, there is very little change in the upper part of the abstract grid. From step 600 to 780, there is no difference in the same area, that is, no cell gets developed. It suggests that after reaching a certain density level second-home owners stop moving in because of amenity and space concerns. They need cells that are not only close to natural amenities but also distant from their neighbors. After time step 600, it is hard for them to find cells that satisfy their preferences Model ill simulates what happens if every second-home owner has a preference for both natural amenities and space (neighborhood avoidance and larger lots). Initially second-home owners bid high for cells because of voluminous spaces available in amenity rich areas As more and more cells are developed, the areas with high amenities densify to a level so that they become less and less valuable for second-home owners. Natural amenities have a dynamic character: they become degraded with increased development. When a certain development density is reached, the densification process stops because second-home owners want to protect the quality of local amenities against increased density. They choose to either buy large lots to retain lower density or relocate. 87

PAGE 99

Model ill is constructed to represent varied types of exurban households with different locational preferences to reflect the diversity of characteristics found in social systems. It simulates the process of how an exurban resident selects a site for development in relation to his individual desire (i.e. preference for being close to public lands) and his interaction with other residents in the neighborhood, a process that is directly analogous to the one in the real world. It integrates a dynamic and interactive framework to explore effects of amenities, density, and accessibility on exurban location and how exurban spatial structure is determined by interdependencies among spatially distributed agents and the biophysical aspects of the physical environment. 88

PAGE 100

re 4-5 Model III New Development I I Stream -Road 89

PAGE 101

-New Development I I Public Land -Lake I Stream -Road Step 360 Step 600 90

PAGE 102

New Development I I Stream -Road Model Comparison Figure 4-2 to Figure 4-5 suggest that the accessibility model (Stage one of the Model I) overemphasizes clustering factors around areas with urban services and along transportation corridors. Model I exaggerates the build out around amenity rich areas. Model II overemphasizes dispersal effects In other words, accessibility variables cluster development on job centers and along highways and roads; amenity variables pull development into natural amenity rich areas; and neighborhood effect variables have both pull and push power. They draw commuters together while second-home owners spread out into underdeveloped areas and become seeds for later development. 91

PAGE 103

Model III (Figure 4-5) is a more robust simulation because it includes households' connection and adaptation to changes in their environment, both bio-physically and socially. It includes complex household decision making behaviors and how that influences exurban development. Households have the ability to receive and interpret GIS data in light of an overall plan to decide where to develop. It integrates a searching process for exurban household that has direct analogy to the real world. Model III tends to scatter development more widely with respect to amenities. However, the three theoretical models and results presented here are not based on actual data. They serve as tools to explore factors driving exurban development. In the next section, the model is validated by a case study in northwestern Boulder County, Colorado. Case Study/Model Validation Model validation ensures that simulations are sufficiently accurate with reference to the purpose of this research Planning demographic and economic models typically rely upon standard statistical forecasting techniques such as linear, multivariate, and logistic regressions. A range of techniques were developed to verify and validate these models. Validation methods for the agent-based model, however, was not fully established. The literature suggests that a mix of "eyeball" similarity comparison (Robinson, 1997) to the real-world settlement patterns and some aggregate statistics (Rand, et al., 2003; Parker et al., 2003; Sargent 1988) may be the best way to perform agent-based model validation. I build three theoretical models on an abstract grid and validate by evaluating how it works empirically in the study area to assure that it represents the real system to 92

PAGE 104

-a sufficient level of accuracy My approach of model validation includes: 1) collect performance measures of the system in the study area northwestern Boulder County for the time period of 1980-2000; 2) average the model results of multiple simulation runs (over 100 runs) with the given input data; and 3) compare the model outcomes to real-world observations over the given time period both visually and statistically using neighborhood density. Through multiple runs I assess the sensitivity of exurban landscape change. The case study was conducted in northwestern Boulder County which consists of three census block groups. Results of a study on the housing characteristics of the study area show that a high percentage of housing units is used for seasonal or recreation purpose especially in two of the three block groups. U.S. Census data from 1990 and 2000 report that more than 60 % of housing units are for seasonal or recreation use in two block groups and more than 15% in the third group (Table 4-1 ). Table 4-1 Some Housing Characteristics of the Study Area Census Block Occupied Housing Units For Tract Group Housing Seasonal or Recreational Units use 1990 2000 1990 2000 13602 1 164 297 657 519 13602 2 127 166 215 329 13701 4 693 596 259 122 Source: U.S. Census Bureau Percentage of Seasonal or Recreational use(% ) 1990 2000 80.02 63.60 62.87 66.46 27.21 16. 99 Furthermore, in these two block groups, more than 10% of householders are 65 years or older, which may include retirees (Table 4-2) Table 4-1 and 4-2 suggest that there are some households in the study area with non-commuters. This makes the study area ideal for validating the theoretical model presented in the previous 93

PAGE 105

section m which the two households types are commuters and second-home owners Table 4-2 Some Household Characteristics of the Study Area Census Block Age (percentage) Households by Type Persons Tract Group (percentage) per 13602 1 13602 2 13701 4 < 18 18-65 > 65 Married NonHouseholder Household -couple family 65 years and over 22.25 60.05 17.70 59.76 33.54 10. 37 2.27 20 .65 66.30 13.05 55.91 37.01 10.24 2.17 26.11 70.06 3.23 49.35 43.29 1.73 2.37 Source: U.S. Census Bureau Agents in the validation model make decisions in ways that approximate real world household decision-making They visit the site and obtain information on the neighborhood and bio-physical environment. Through the iterative application of the rules on the households for twenty consecutive years (1981 to 2000) over 1 00 runs, I average and generate development patterns for 1990 and 2000 respectively. Figure 4-6 shows the results from three of the 100 runs. The maps on the left are the simulation predicted development from 1981 to 1990; and on the right they are predicted development from 1981 to 2000 The land use pattern emerging on the grid is the result of interaction between two types of agents (commuters and second-home owners) across space and time In the upper middle part of the study area development clings to highway 7 This area has high levels of previous development and is close to Rocky Mountain National Park. There are also a few dots / developed cells by second-home owners on the right side Land in the bottom right hand comer has attracted a lot of development due to its proximity to the City of Boulder. 94 I li

PAGE 106

Results from Mu 1981-1990 95

PAGE 107

-The maps of actual development between 1981 and 1990, and between 1981 and 2000 are shown in Figure 4-7. The visual comparison of snapshots from the simulation and the existing development suggests that the model generates clusters and patterns of dispersion that to some extent align with data from the real world This validation suggests that the model has captured some aspects of the exurban development process. Timing space, and amenities play important roles in shaping the exurban development pattern. After the initial visual comparison of snapshots from the simulation and the existing development I export the model results from RePast as an ASCII file and import it to Arcview to create a grid showing predicted development in 2000. Arc/Info Grid and Map Algebra are used to create neighborhood density for each cell. It is the total number of developed cells in a 13 x 13 window / neighborhood around each cell. Since the purpose of this study is not to predict precisely in which cell development would occur, neighborhood density is a good indicator to be used for validating how accurate my agent-based model is for exploring exurban residential development patterns. Figure 4-8 presents the existing neighborhood density and the predicted density for the year 2000. Comparison of these two maps suggests that my model reasonably represents the real system. 96

PAGE 108

Existing development: 19811990 . l .tD .. :T J 10 r-' !'. .. __ / "'-__,-/ -_,/ (. :. .. "? .. .. j':. 1\1 Highway 1 Local Road D Study Area 72 D Previous Development \"" E:\.isting Development: 198(r: fJ ., J . .t .. Existing development: 1981-2000 .,.t-3 .. l .. 10 .. r-' . .... r. / "?' r:. .. tl.. ;: .. 1\1 Highway 1 Local Road ; CJ Study Area 72 D Previous D evelopment \...,.. Existing Development: 198 1,..: foQOO \l..S 97

PAGE 109

Existing Neighborhood Density: 2000 Legend Predicted Neighborhood Density: 2000 Legend City j\/Highway Predicted Neighborhood Density: 2000 D -1 0 Std. Dev. DMean D 0-1 Std. Dev. CJ 1-2 Std. Dev. 2-3 Std. Dev. > 3 Std. Dev. 98

PAGE 110

-CHAPTERS SUMMARY AND CONCLUSIONS In chapter 4, I present the modeling results from the three theoretical models and the validation model. This chapter contains a brief summary of the research findings, the planning implications, and future research directions. Research Findings In this section, I summarize the research findings of this dissertation. The accessibility model (the first part of theoretical model I) illustrates how clusters emerged around the employment centers and along the transportation corridors because of household level of preference on accessibility. It tends to concentrate development too tightly around rural places and transportation networks in comparison with the actual development pattern. This model is an extreme; however, it demonstrates that if all exurban households had only preferences for proximity to road network and job centers, at the aggregate level, exurban development pattern would have been similar to a monocentric pattern. In the full theoretical model I, much of the growth goes into either amenity rich areas or areas with easy access to highways and jobs, reflecting the market attractiveness of factors such as the presence of public lands, water bodies, or streams to second-home owners, and proximity to existing urban services and accessibility to major highways to commuters. The mix of preferences for amenities and accessibility makes the model a more accurate predictor of exurban 99

PAGE 111

development than models based exclusively on accessibility. However, amenity variables tend to focus development too tightly around natural amenities in comparison with the actual development pattern Theoretical model II illustrates two extremes of development patterns : cluster and dispersion resulting from different household locational preferences Commuter preferences (proximity to work and transportation networks) fuel higher-density development in lands surrounding cities and highways; behavior preferences of second-home owners for large lots and neighbor avoidance push development into wilderness areas which may become seeds for later development. It indicates that patterns of exurban growth are defined by factors such as spacing of lots and distance from infrastructure. Households interact with each other in exurban locational decision making. Second-home owners skip over properties close in to obtain bigger lots further out. This creates pressure for low density development and a persistent dispersion pattern and significant and disproportionate reduction in average density of development at the aggregate level. Theoretical model II demonstrates that there are development phasing effects at which exurban development shifts from a land market dominated by second-home owners to commuters Yet the switch between second-home owners and commuters land markets occurs only gradually after second-home owners reach the density threshold Model I introduces accessibility variables followed by amenity variables Model II evaluates the dynamic effects of lot size preference on long term density change. However, accessibility variables overemphasize clustering factors around areas with urban services and highway corridors; amenity variables exaggerate build-out around amenity rich areas; and neighborhood effect variables overstress dispersal effects. 100

PAGE 112

-Theoretical model Ill is a more robust simulation incorporating varied types of exurban households with different locational preferences to reflect the diversity of characteristics found in social systems, and households' connection and adaptation to changes in their neighborhoods and their responses to the bio-physical environment. Households have the capacity to receive and interpret GIS data based on the presence of natural amenities, neighborhood density, and urban and transportation proximities. This model simulates the process in which an exurban resident selects a site for development in a way that is directly analogous to the real world. Model ill tends to scatter development more widely with respect to amenities. Development also builds out in other areas that are attractive because of proximity to roads and cities, and the presence of existing development. Theoretical model Ill shows that exurban development is best explained by the mixed preferences of amenities, accessibilities and neighborhood avoidance In the model validation for the study area we see large amount of land development in the bottom right comer because of its accessibility to the City of Boulder and some scattered development by second-home owners. The general pattern shows that an agent-based model that encapsulates the behaviors of the diverse individuals making up the system can generate aggregate patterns that match the real world empirical data It suggests this model represents the real land market system at a reasonably high level of accuracy and captures some aspects of the exurban development process The agent-based approach allows resesarchers to look at independent and interactive effects of natural amenities, accessibilities and neighborhood avoidance, and phasing effects of exurban development. The interactive and dynamic agent-based exurban development model built in this research permits the incorporation of locational behavior understanding into urban location theories It 101

PAGE 113

-extends existing growth research which has not fully incorporated a fundamental insight regarding how exurban locational behaviors are sensitive to fine-grained variations in their social and biophysical environment. It models how exurban dispersion emerges from the interactive effects of amenities settlement density and accessibility This type of modeling can not only inform planning theorists but also improve planning practice by providing an understanding of households locational behaviors that standard quantitative models do not, and they enable planners to consider a broader range of possible cumulative or emergent effects of land use policies or market trends. Finally this model sheds light on the determinants and constitution of the new American West ex urban development, and draws on the notion that urban spatial structure is determined by interdependencies among spatially distributed heterogenous agents and the biophysical aspects of the landscape environment. Planning Implications The results of the agent-based exurban model suggest that without the adoption of a growth management strategy future development in the mountain valleys and foothills of the American West is likely to be highly fragmented. Ex urban d e velopment patterns have been extremely scattered and it is anticipated that these patterns will continue unless measures are taken to guide development. Understanding patterns of e x urban development the dri v ing forces behind them and consequences of this rapid exurban growth are critical to planners and land managers. Some research suggests that people seem to get increasingly dissatisfied by the stresses and cost of city life and more interested in the amenities in the rural area Exurban areas are further invigorated by modem information and 102

PAGE 114

communication technology. While it tends to attact a further flow of people to the high amenity rural West, we can expect questions and issues resulting from exurban development to be heightened in the near future. Simulations developed in this research support the incorporation of our behavioral understanding of exurban households into urban growth models for policy makers. These simulations are not only mechanically simple to understand because of the use of behavior rules of households, but also easy to include and evaluate government policies, restrictions, incentives, and disincentives that might be used to affect households' decision-making, such as zoning and rural cluster boundaries. It might be used as a tool to influence how policy makers guide the development patterns. In this final section, I present an example of using an agent-based model to control development in a high wildfire hazard zone in the study area. The flow of migration to natural amenity-rich areas in the United States has created a difficult set of regulatory challenges for hazard and land use managers in rural local governments. Immigrants have a variety of impacts on the forests as a result of different values and ways of interacting with the environment. With high rates of exurban development over the past three decades, this landscape shows signs of increasing wildfire risk and the loss of property and life. In this simulation, both commuters and second-home owners are restricted from building in high wildfire hazard zones (Figure 5-l). Every household responds to this policy and takes it into consideration when selecting a cell to develop. This assumes that choice of location is market-driven, but wildfire hazard zoning is imposed to control development in high fire zones. In an area such as western Boulder County where there are many possible future development sites in high 103

PAGE 115

wildfire hazard areas it is important to study the household's perceptions of wildfire risk. In the wildfire hazard zoning model, the maps describe roughly the same locational pattern as in the model I present in the last chapter except that development was pushed into areas with low wildfire risk (Figure 5-2) Legend D City NHighway Wildfire Hazard Low Medium Medium D Medium CJHigh 2 :\Jil es 104

PAGE 116

5-2 Wildfire Hazard Zonin 105

PAGE 117

This example has a relatively low level of regulatory intervention. However, it shows some planning applications of the agent-based models built in this research It will be very helpful in evaluating the possible timing and sequence of future development, guiding policy development, and channeling development as wanted at this critical time in the evolution of land use in the rural West. Future Research Directions In this research, I include only two types of households in my simulation of exurban development, representing households with preferences for commuting, and natural amenities and lot size respectively. However, some research shows that there are other types of preferences/households representing exurbanites (Fernandez, et al., 2003; Nelson and Sanchez, 1997; Varady, 1980), such as a preference for living close to friends and families, for the availability of good schools, and for the overall appearance of an area. A study on demographic data (i.e family structure, age, income, and correlations between block level demographic changes and exurban residential development) will make the model more realistic. These factors can be included in the simulation to make it more accurate for the future work. 106

PAGE 118

BIBILOGRAPHY Acharya, G., and Lewis, B. L. 2001. Valuing Open Space and Landuse Patterns in Urban Watersheds. Journal of Real Estate Finance and Economics 22 : 221-237 Allan, N J. R. 1986. Accessibility and Altitudinal Zonation Models of Mountain Mountain Research and Development 6 (3): 185-194. Alonso W 1964. Location and Land Use: Toward a General Theory of Land Rent Cambridge, MA: Harvard University Press Amott, R., Pines, D., and Sadka, E 1986. The Effects of an Equiproportional Transport Improvement in a Fully-Closed Monocentric City. Regional Science and Urban Economics 16: 387-406. Balmann, A. 1997. Farm-Based Modeling of Regional Structural Change European Review of Agricultural Economics 25 (1) : 85-108 Balmann, A. 2001. Modeling Land Use with Multi-Agent Systems: Perspectives for the Analysis of Agricultural Policies. In Microbehavior and Macroresults: Proceedings of the Tenth Biennial Conference of the International Institute of Fishery Economics and Trade (IIFET), Compiled by R. S. Johnston and A. L. Shriver. Corvallis, OR : ITFET Batty, M 1998. Urban Evolution on the Desktop : Simulation with the Use of Extended Cellular Automata. Environment and Planning A 30: 1943-1967. Batty, M., Jiang, B. and Thurstain-Goodwin, M 1998. Local Movement : Agent Based Models of Pedestrian Flows. CASA Working Paper 4. London : The Center for Advanced Spatial Analysis. Batty, M. and Xie, Y. 1994. From Cells to Cities. Environment and Planning B 21: 31-48. Batty, M. and Xie Y. 1997. Possible Urban Automata Environment and Planning B 24: 175-192. 107

PAGE 119

Beale, C. L., and Johnson, K. M. 1998. The Identification of Recreation Counties in Non-Metropolitan Areas of the USA Population Research and Polic y Review 17: 37 -53. Beckmann, M. J. 1969. On the Distribution of Urban Rent and Residential Density Journal of Economic Theory 1: 60-67. Beckmann, M. J 1973. Spatial Equilibrium in the Housing Market. Journal of Urban Economics 1: 99-1 07 Beckmann, M. J. 1976. Spatial Equilibrium in the Dispersed City. In Mathematical Land Use Theory, edited by Y. Y. Papageorgiou. Lexington, Mass: Lexington Books 117-125 Benfield, E. K., Raimi M. D., and Chen, D. D. 1999. Once There Were Greenfields : How Urban Sprawl is Undermining America s Environment, Economy and Social Fabric. New York: National Resources Defense Council. Berger, T 2001. Agent-Based Spatial Models Applied to Agriculture: A Simulation Tool for Technology Diffusion, Resource Use Changes and Policy Analysis Agricultural Economics 25 (2/ 3): 245-260 Beson, E. D. Hansen J. L., Arthur, J., Schwartz, L., and Smersh, G. T. 1998. Pricing Residential Amenities: The Value of a View Journal of Real Estate Finance and Economics 16 : 55-73 Beyers, W. B. and Nelson, P B 2000. Contemporary Development Forces in the Nonrnetropolitan West: New Insights From Rapidly Growing Communi Ties. Journal of Rural Studies 16: 459-474. Borukhov, E., Hochman 0. 1977. Optimum and Market Equilibrium in a Model of a City Without a Predetermined Center. Environment and Planning A 9: 849 856. Bradshaw, Ted and Brian Muller. 1998 Impacts of Rapid Urban Growth on Farmland Conservation: Application of New Regional Land Use Policy Models and Geographic Information Systems. Rural Sociology 63 (1) : 1-25 108

PAGE 120

Brown, G. M., Jr., and Pollakowski, H.O. 1977. Economic Value of Shoreline. Review of Economics and statistics 59 (August): 272-278. Brueckner, J. K. 1981. A Dynamic Model of Housing Projection. Journal of Urban Economics 10: 1-14. Brueckner, J. K., Thisse, J. F., and Zenou, Y. 1999. Why is Central Paris Rich and Downtown Detroit Poor? An Amenity-Based Theory. European Economic Review 43: 91-107. Cassel, E. and Mendelsohn, R. 1985. The Choice ofFunctional Forms for Hedonic Price Equations: Comments. Journal of Urban Economics 18: 135-142. Cervero, R. 1986. Suburban Gridlock. New Brunswick, New Jersey: Center for Urban Policy Research Press. Cervero, R. 1991. Land Use and Travel at Suburban Activity Centers. Transportation Quarterly 45: 479-491. Cervero, R. 1993. Ridership Impacts of Transit-Focused Development in California. Monograph 45, Berkeley, California: Institute of Urban and Regional Development, University of California. Chattopadhyay, S. 2000. The Effectiveness ofMcfaddens's Nested Logit Model in Valuing Amenity Improvement. Regional Science and Urban Economics 30: 23-43. Clarke, K.C., Hoppen, S. and Gaydos, L. 1997. A Self-Modifying Cellular Automaton Model of Historical Urbanization in the San Francisco Bay Area. Environment and Planning B 24: 247-261. Colby, B., Wishart, S. 2002. Riparian Areas Generate Property Value Premium for Landowners, Working paper, University of Arizona College of Agriculture and Life Sciences, Department of Agricultural and Resource Economics. Conte, R. and Gilbert, G. N., ed. 1995. Artificial Societies: The Computer Simulation of Social Life. London: UCL. 109

PAGE 121

-Correll M. R. Lillydahl, J H., and Singell, L. D. 1987. The Effects of Greenbelts on Residential Property Values: Some Findings on the Political Economy of Open Space Land Economics 54 (May): 207-217. Couclelis H. 1989. Macrostructure and Microbehavior in a Metropol i tan Area Environm e nt and Planning B 16 : 141-54 Cromartie J 1994. Recent Demographi c and Economic Chang e s in the W e st Statement before the House Committee on Natural Resources Hearing on The Changing Needs of the West. Economic Research Service U.S. Department of Agriculture Cuba, L. 1989. From Visitor to Resident: Retiring in Vacationland. Generations 13: 63-67 Darling, A. H. 1973 Measuring Benefits Generated by Urban Water Parks. Land Economics 49 (February): 22-34 Davis, J S Nelson, A. C. and Dueker K. J 1994. The New 'Burbs: the Exurbs and Their Implications for Planning Policy Journal of American Planning Association 60: 45-59. Do, A.Q and Sirmans C.F. 1994. Residential Property Tax Capitalization : Discount Rate Evidence from California. National Tax Journal 47 (June): 341 348 Doss C. and Taff, S. 1996. The Influence of Wetland Type and Wetland Proximity on Residential Property Values. Journal of A gri c ultural and R e source Economics 21 (1): 120-129 Dowall D E. and Salkin, M 1986. Office Automation and the Implications for Office D e ve lopment. Working Paper No. 447 University of California Berkeley Institute of Urban and Regional Development. Drogoul, A. and Ferber, J. 1995. Multi-Agent Simulation as a Tool for Studying Emergent Processes. In Simulating Soci e ti e s : The comput e r s imulation of social phenom e na edited by N. Gilbert and J Doran. London: UCL Press 127-142. Duane T.P. 1999. Shaping the Si e rra: Nature Culture, and Confli c t in the Changing W e st. Berkeley Calif: University of California Press 110

PAGE 122

-Dubin, R. A. 1988. Estimation of Regression Coefficients in the Presence of Spatially Autocorrelated Error Terms. Revi ew of Economics and Statistics 70(3): 466-74. Dubin R. A. 1992 Spatial Autocorrelation and Neighborhood Quality. Regional Science and Urban Economics 22(3): 433-52. Dueker, K. J., Strathman, J. G., Levin, I. P., and Phipps A. G. 1983. Rural Residential Development within Metropolitan Area Computers Environment and Urban Systems 8(2): 121-29. Epp, D. J. and A1-Ani, K. S 1979. The Effect of Water Quality on Rural Non-farm Residential Property Values. American Journal of Agricultural Economics 61: 529-534. Epstein, J. M. and Axtell, R. 1996. Growing Artificial Societies; Social Science from the Bottom Up. Cambridge, Massachusetts: MIT Press. Findler, N.V. and Malyankar, R.M. 1995. Emergent Behavior in Societies of Heterogeneous, Interacting Agents; Alliances and Norms. In Artificial Societies: The computer simulation of social life, edited by Gilbert, N. and Conte, R London: UCL Press. 212-236. Franklin, S. and Graesser, A. 1997. Is It an Agent, or Just a Program?: A Taxonomy for Autonomous Agents. Intelligent Agents III Berlin : Springer Verlag, 21-35. Fuguitt, G V and Zuiches, J. J. 1975. Residential Preferences and Population Distribution Demography 12: 491-504. Fujita, M 1976. Toward a Dynamic Theory of Urban Land Use. Papers of R egiona l Science Association 37: 133-165. Fujita, M. 1987 Residential Land Use Theory. Cambridge, MA: Cambridge University Press. Garasky, S. and Haurin, D. R. 1997. Tiebout Revisited: Redrawing Jurisdictional Boundaries Journal of Urban Economics 42 (3): 366-376. Garreau, J. 1991. Edge City: Life on the New Frontier, New York: Doubleday. 111

PAGE 123

Geoghegan, J., Wainger, L., and Bockstael, N.E. 1997. Spatial Landscape Indices in a Hedonic Framework: An Ecological Economics Analysis Using GIS Ecological Economics 23 (3) : 251-264. Gillard, Q 1981. The Effect of Environmental Amenities on House Values: The Example of a View Lot. Professional Geographer 33 (May) : 216-220. Gimblett, H. R., ed. 2002. Integrating Geographic Information Systems and Agent Based Modeling Techniques for Simulating Social and Ecological Processes. Oxford UK: Oxford University Press Giuliano G. and Small, K. 1991. Subcenters in the Los Angeles Region. Regional Science and Urban Economics 21: 163-182 Gordon P., Kumar A. and Richardson H. W. 1988 Beyond the Journey to Work. Transportation Research A 22A (6) : 419-426. Gordan, P., Kumar A. and Richardson, H. W. 1989. The Influence of Metropolitan Structure on Commuting Time. Journal of Urban Economics 26: 138-51. Griliches, Z. 1961. Hedonic Price Indexes for Automobiles: An Econometric Analysis Of Quality Change. The Price Statistics of the Federal Government 73: 137-196. Griliches, Z 1967 Hedonic Price Indexes Revisited: Some Notes on the State of the Art. In Proceedings of the Business and Economic Statistics Section of the American Statistical Association 324-332; reprinted in slightly altered form in Price Indexes and Quality Change, edited by Z. Griliches. Cambridge, MA: Harvard University Press, 1971. Hammer, T R., Coughlin, R. E., Hom IV, E. T 1974. The Effect of a Large Urban Park on Real Estate Values American Institute of Planners Journal 40 ( 4): 274--277. Harris B. 1994 The Real Issues Concerning Lee's "Requiem." Journal of the American Planning Association 60 : 31-34. Hatz B., and A. Liu. 2000. Moving Beyond Sprawl: Toward a Broader Metropolitan Agenda. Brookings Review 18 (2) : 31-34. 112

PAGE 124

-Henderson, J.V. 1977. E c onomi c The o ry and the Citi e s, New York: Academic Press. Holland J. H. 1998. Emergen ce: From Chao s to Ord e r New York: Helix Books. Ingram, K and Lewandrowski J. 1999. Wildlife Conservation and Economic Development in the West. Rural D e velopm e nt P e r s p e ctiv e s 14 (2): 44-51. Irwin E. 1998. Do Spatial Externalities Matter? The Role of Endogenous Interactions in the Evolution of Land Use Patt ern. Unpublished doctoral dissertation University of Maryland Maryland. Irwin E., and Bockstael, N. 2002. Interacting Agents Spatial Externalities, and the Evolution of Residential Land Use Patterns. Journal of Economic Geograph y 2 (1): 31-54. Jobes P C. 1988. Nominalism Realism and Planning in a Changing Community. International Journal of Env ironm e ntal Studi e s 31: 279-290. Jobes P. C. 1995. Migration in the West: a Gallatin Valley Montana, Case Study. The Western Planner 16 (3) : 10-13 Jobes P. C 2000. Moving Near e r to H e aven : The Illusions and Disillusions of Migrants to Scenic Rural Places. Westport, CT: Praeger Johnson J. D and Rasker R. 1995. The Role of Economic and Quality of Life Values in Rural Business Location. Journal of Rural Studi e s 11: 405-416 Kitchen J W and Hendon W S. 1967 Land Values Adjacent to an Urban Neighborhood Park. Land Economics 43: 357-360 Klosterman, R E 1994. An Introduction to the Literature on Large-Scale Urban Models. Journal of the American Planning Association 60 : 41-44 Krugman, P 1991 G e ograph y and Trad e Leuven University: MIT Press Krugman, P 1996 The Self-Organi z ing Economy. Malden, MA: Blackwell Lancaster K.J. 1966. A New Approach to Consumer Theory. Journal of Politic a l Econom y 74 : 132-57. 113

PAGE 125

-Landis, J D. 1993. CUF Model Simulation Results: Alternative Futures for the Greater Bay Area Region. Working Paper 592 Berkeley, CA: Institute of Urban and Regional Development University of California at Berkeley. Landis, J. 1994 The California Urban Futures Model : A New Generation of Metropolitan Simulation Models. Environment and Planning B 21: 399420. Landis, J. 1995 Imagining Land Use Futures: Applying the California Urban Futures Model. Journal of the American Planning Association 61 (4): 438457. Landis, J.D. 1998 The Second Generation ofthe California Urban Futures Model. Part 2: Specification and Calibration Results of the Land-use Change Submodel. Environment and Planning B 25: 795824. Landis, J. and Zhang, M 1997. Modeling Urban Land Use Change: The Next Generation of the California Urban Futures Model Retrieved April 10, 2002, from http:/ / www .ncgia. ucsb. edu / conf/landuse97 Lee, D.B. 1973 Requiem for Large Scale Models. Journal of the American Institute for Planners 39: 163-78. Lee, D.B. 1994. Retrospective on Large-scale urban models Journal of the American Planning Association 60: 35-41. Lessinger, J 1986. Regions of Opportunity. New York: Times Books. Levitt J (ed.) 2002. Conservation in the Internet Age, Washington D.C.: Is-land Press Lim, K., Deadman, P., Moran, E., Brondizio E., McCracken S 2001. Agent based Simulations of Household Decision Making and Land Use Change Near Altamira, Brazil. In Integrating GIS and Agent Based Modeling Techniqu es for Understanding Social and Ecological Processes edited by R. Gimblett. Oxford, U.K., and Santa Fe, N.Mex.: Oxford Press and the Santa Fe Institute Mahan, B., S Polasky, and R.M. Adams. 2000. Valuing Urban Wetlands: A Property Price Approach. Land Economics 76: 100-113. 114

PAGE 126

Manson, S.M. 2000. Agent-based Dynamic Spatial Simulation of Land-use / Cover Change in the Yucatim Peninsula, Mexico. Paper presented in the Fourth International Conference on Integrating GIS and Environmental Modeling (GIS / EM4), Banff, Canada. Manson, S.M 2001. Simplifying Complexity: A Review of Complexity Theory. Geoforum 32 : 405-414. Masnick, George. 2001. America's Shifting Population: Understanding Migration Patterns in the West. The Rocky Mountain West's Changing Landscape 2 (Winter / Spring): 8-15. McFadden, D. 1974. Conditional Logit Analysis of Qualitative Choice Behavior. In Frontiers in Economentrice, edited by P. Zarembka New York : Academic Press 105-142. McFadden, D. 1978. Modeling the Choice of Residential Location. In Spatial Interaction Theory and Planning Models, edited by A. Karlqvist, L. Lundqvist, F Snickars and J.W. Weibull. Amsterdam: North-Holland: 7596. McGranahan, D.A. 1999. Natural Amenities Drive Rural Population Change Agricultural Economic Report No. 781, Washington, D.C.: U.S. Department of Agriculture, Economic Research Service. McHugh, K. E 1990 Seasonal Migration as a Substitute for, or Precursor to, Permanent Migration. Research on Aging 12: 229-245. Michaels R. G and V. K. Smith. 1990 Market Segmentation And Valuing Amenities With Hedonic Models : The Case Of Hazardous Waste Sites. Journal of Urban Economics 28 (2): 223-242. Mieszkowski, P and Mills, E. S 1993. The Causes of Metropolitan Suburbanization. Journal of Economic Perspectives 7 (3) : 135-147. Miller, S.G., Knight R L., and Miller C.K 1998. Influence of Recreational Trails on Breeding Bird Communities. Ecological Applications 8: 162 169. Mills, E.S. and D.M.deFerranti. 1971. Market Choices and Optimum City Size. American Economic Review 61(2): 340-345. 115

PAGE 127

Mills, E S. 1972. Studies in the Structure of the Urban Econom y. Baltimore: The Johns Hopkins Press Moran, E E Ostrom, and J. C Randolph 1998. A Multilev e l Approach to Studying Global Environmental Change in For e st Ecos y stems Bloomington: Center for the Study of Institutions, Population, and Environmental Change and Workshop in Political Theory and Policy Analysis, Indiana University Muller, B., Bertron C., and Yin, L. 2002. Alternatives for Futur e Growth in the Tri-River Region. Washington, D C.: American Farmland Trust. Muller, B ., Puccio K., Baker, H and Yin, L. 2002. Custer County Colorado: Alternative Growth Futures Tucson, Arizona: Sonoran Institute. Muller, B and Yin L. 2001. Salinas-Pajaro Alternative Gro wth Futures Project : Analysis of Growth Patterns and Alternatives Working paper University of Colorado at Denver Muth R. F 1969 Cities and Housing. Chicago: University of Chicago Press Nagel K., Beckman, R. J and Barrett C. L. 1998 TRANSIMS for Transportation Planning. In Proceedings of the International Confer e nce on Complex S y stems edited by Y. Bar-Yam. Nashua, NH: New England Complex Systems Institute. Nelson, A. and T. Sanchez. 1999. Debunking the Exurban Myth: A Comparison of Suburban Households Housing Polic y Debat e 1 0(3 ) : 689-709. Nelson A. C 1990. Regional Patterns of Exurban Industrialization Economic Development Quarterly 4 (4): 320-33 Nelson, A.C. 1991: Toward a Definition of Exurbia. Journal of Planning Literature 6 (4): 350-68. Nelson A.C 1992. Characterizing Exurbia Journal of Planning Literature 6 (4) : 350-368. Nelson A C. and Dueker, K.J. 1990. Socioeconomic Characteristics and Location Behavior of Exurban Households with Planning Policy Implications Journal of Planning Education and R e s e arch 9(2) : 91-100 116

PAGE 128

Oberman, W., Carlson, D., and Batchelder, J., Ed. 2000. Tracking Agricultural Land Conversion in Colorado: An Interagency Summary by the Colorado Department of Agriculture, Natural Resources Conservation Service, and Colorado Agricultural Statistics Service. September. O'Sullivan, D. 2000. Graph-based Cellular Automata Models of Urban Spatial Processes Unpublished doctoral dissertation University College London, London. O'Sullivan, D. and Torrens, P. 2000. Cellular Models of Urban Systems. In Theoretical and Practical Issues on Cellular Automata, edited by S Bandini and T. Worsch. London: Springer-Verlag, Parker, D. C., Evans, T. P., and Meretsky, V 2001. Measuring Emergent Properties of Agent-Based Land-cover I Land-use Models Using Spatial Metrics. Paper presented at the 7th annual conference of the Society for Computational Economics, Yale University June 28-29th. Parker, D. C., Manson, S. M., Janssen, M. A., Hoffinann, M. J., and Deadman P 2003. Multi-Agent Systems for the Simulation of Land-Use and Land Cover Change: A Review. Annals of the Association of American Geographers 93 (2). Peiser, R.B., Schwann, G. M 1993 The Private Value of Public Open Space within Subdivisions Journal of Architectural and Planning Research 10 (2): 91104. Portugali, J. 2000 Self-Organization and the City. Berlin : SpringerVerlag. Power, T. M., Ed. 1995. Economic Well-Being and Environmental Protection in the Pacific Northwest Massachusetts, MT: University of Montana Missoula, Department of Economics Power, T. M. 1998. Lost Landscapes and Failed Economies: The Search for a Value of Place. Washington, DC: Island Press. Rajan, K. S. and Shibasaki, R 2000. Land Use/ Cover Changes and Water Resources Experiences from AGENT-LUC Model. Tokyo, Japan: International Center for Disaster Mitigation Engineering (INCEDE), Institute of Industrial Science, University of Tokyo Publication 19. 117

PAGE 129

Rand, W., Brown, D. G., Page, S. E., Riolo R., Fernandez, L. E., and Zellner, M 2003. Statistical Validation of Spatial Patterns in Agent-Based Models Paper presented at Agent-based Simulation 4, Montnellier, France, April 28. Rand, W., Zellner, M., Page, S.E., Riolo, R., Brown D G., Fernandez, L.E 2002 The Complex Interaction of Agents and Environments: An Example in Urban Sprawl. In Proceedings of Agent 2002 Conference on Social Agents: Ecology E x change and Evolution. Chicago, IL: Argonne National Laboratory, University of Chicago. Rasker, R. 1993 Rural Development, Conservation and Public Policy in the Greater Yellowstone Ecosystem. Society and Natural Resources 6: 109 126. Ren, J and White, R. 1995 The Simulation ofUrban System Dynamics in Atlantic Canada, 1951-1991. Canadian Geographer-Geographe Canadien 39 (3) : 252-262 Ridker, R. and Henning, J. 1967 The Determinants of Residential Property Values with Special Reference to Air Pollution. Review of Economics and Statistics 49 (May): 246-257 Riebsame, W. E., Gosnell H., and Theobald D. M 1996 Land Use and Landscape Change in the Colorado Mountains I: Theory, Scale, and Pattern Mountain Research and Development 16 (4): 395-405. Riebsame, W. E. Robb, J., Boulder Center of the American West University of Colorado and Gosnell, H., Ed. 1997. Atlas of the New West: Portrait of a Changing Region. New York : W W Norton & Company Ringholz, R. 1992. Small Town Blues: Voices from the Changing West. Salt Lake City : Perigrine Smith Books Robinson, S. 1997. Simulation Model Verification and Validation: Increasing the Users' Confidence In Proceedings of the 199 7 Winter Simulation Conference, edited by S Andradottir, K. J. Healy, D H. Withers, and B. L. Nelson. Washington D.C. : IEEE Computer Society Press 118

PAGE 130

Rodriguez M and Sirmans C. F 1994. Quantifying the Value of a View in Single-Family Housing Markets The Appraisal Journal62: 600-603 Rosen S. 1974. Hedonic Prices and Implicit Markets : Product Differentiation in Pure Competition. Journal of Political Econom y 82 (Jan./Feb): 34-55 Rudzitis G. 1993 Non-metropolitan Geography: Migration, Sense of Place, and the American West. Urban G e ograph y 14 : 574-585 Rudzitis G. 1996. Wilderness and the Changing Am e rican W e st New York: John Wiley & Sons Rudzitis, G. 1999. Amenities Increasingly Draw People to the Rural West. Rural D e velopm e nt Perspectiv e s 14 (2): 9-13. Rudzitis G. and Johansen H E. 1989 Migration into Western Wilderness Counties : Causes and Consequences. Western Wildlands (Spring): 19-23. Rudzitis, G. and Johansen, H. E 1991. How Important is Wilderness? Results from a United States Survey. Environmental Management 15: 227-231. Sargent, R. G 1988 A Tutorial on Validation and Verification of Simulation Models. In Proceedings of 1988 Winter Simulation Confe r e nce, edited by M. Abrams. New York NY: ACM Press Schelhorn T. O'Sullivan, D., Haklay, M. and Thurstain-Goodwin M. 1999 STREETS : An Agent-Bas e d Pedestrian Model University College London London: Centre for Advanced Spatial Analysis Schelling, T 1971. Dynamic Models of Segregation. Journal of Mathematical Sociology 1: 143-186. Schelling T. C. 1972. A Process of Residential Segregation: Neighborhood Tipping In Racial Discrimination in Economic life, edited by A Pascal. Lexington MA: Lexington Books Shultz, S.D. and D .A. King 2001. The Use of Census Data for Hedonic Price Estimates of Open-Space Amenities and Land Use. Journal of Real Estat e Financ e and Economi c s 22 (2/ 3): 239-252. 119

PAGE 131

-Sengupta, S. and D. E.Osgood. 2003. The Value of Remoteness: A Hedonic Estimation ofRanchette Prices. Ecological Economics 44: 91-103. Solow, R. H. 1973. On Equilibrium Models of Urban Locations. In Essays in Modern Economics, edited by J M. Parkin. London: Longman. Spahr, R. W. and Sunderman, M. A 1999. Valuation of Property Surrounding a Resort Community. Journal of Real Estate Research 17 (112): 227-43. Theobald, D. M. 2000. Fragmentation by Inholdings and Exurban Development. In Forest Fragmentation in the Central Rocky Mountains, edited by R. L. Knight, F. W. Smith, S. W. Buskirk, W. H. Romme, and W. L. Baker. Boulder: University Press of Colorado: 155 174. Tiebout, C. 1956. A Pure Theory of Local Expenditures. Journal of Political Economy 64: 416-424. Torrens, P.M. 2001. Can Geocomputation Save Urban Simulation? Throw Some Agents into the Mixture, Simmer and Wait .... Working paper 32. London: University College. Ulam, S. 1962. On Some Mathematical Properties Connected with Patterns of Growth of Figures. Proceedings of Symposia on Applied Mathematics 14: 215-224. United States Census Bureau. 2001. Census 2001, Retrieved November 15, 2003, from http://www.census.gov. Vaughn, R.J. 1981. The Value of Urban Open Space. Research zn Urban Economics 1: 103-130. Voith, R. 1996. Central City Decline: Regional or Neighborhood Solutions? Business Review 3. Webster, C.J. and Wu, F 1998. Simulation of Land Development through the Integration of Cellular Automata and Multicriteria Evaluation. Environment and Planning B 25: 103-126. Wegener, M. 1994. Operational Urban Models: State of the Art. Journal of the American Planning Association 60 (1): 17-29. 120

PAGE 132

Wegener, M 1995. Current and Future Land Use Models. Proceedings of the Travel Model hnprovement Program, Land Use Modeling Conference, February 19-21, Washington D.C.: U.S. Departments of Transportation and Energy and the Environmental Protection Agency. Wheaton, W C. 1977. Income and Urban Residence: An Analysis of Consumer Demand for Location. American Economic Review 67 : 620-631. White, R. and Engelen, G 1997. Cellular Automata as the Basis of Integrated Dynamic Regional Modeling. Environment and Planning B: Planning and Design 24 : 235 246. Wilkinson, C. F 1993 Crossing the Next Meridian: Land Water, and the Future of the West. Washington D C.: Island Press. Williams, A. S and Jobes, P. C. 1990. Economic and Quality-of-Life Considerations in Urban-rural Migration. Journal of Rural Studies 6: 187194. Williams, J. D. and McMillan, D. B. 1980. Migration Decision Making among Nonmetropolitan Bound Migrants. In New Directions in Urban-Rural Migration: The Population Turnaround in Rural America edited by J. M. Wardwell and D. L. Brown. New York : Academic Press: 189-212 Wolfram, S. 2002. A New Kind of Science. Champaign, IL: Wolfram Media. Wu, F. 1998 An Experiment on the Generic Polycentricity of Urban Growth in a Cellular Automatic City. Environment and Planning B 25 : 731-752. Yamada, H. 1972. On the Theory of Residential Location: Accessibility, Space, Leisure and Environmental Quality. Papers of the Regional Science Association 29: 125-35 Yin, L. and Muller, B. 2002. Exurban Residential Development in Colorado: An Agent Based Policy Model. Paper presented at the Annual Conference of the American Collegiate Schools ofPlanning, Baltimore, MD November 23. 121