DataPDF Available

Exurbia from the bottom-up: Confronting empirical challenges to characterizing a complex system

Authors:
Exurbia from the bottom-up: Confronting empirical challenges
to characterizing a complex system
Daniel G. Brown
a,b,*
, Derek T. Robinson
a,b
,LiAn
c
, Joan I. Nassauer
b
, Moira Zellner
a,d
,
William Rand
a,e
, Rick Riolo
a
, Scott E. Page
a,f
, Bobbi Low
b
, Zhifang Wang
b
a
Center for the Study of Complex Systems, University of Michigan, United States
b
School of Natural Resources and Environment, University of Michigan, United States
c
Department of Geography, San Deigo State University, United States
d
College of Urban Planning and Public Affairs, University of Illinois at Chicago, United States
e
Northwestern Institute on Complex Systems, Northwestern University, United States
f
Departments of Political Science and Economics, University of Michigan, United States
Received 21 February 2006; received in revised form 4 December 2006
Abstract
We describe empirical results from a multi-disciplinary project that support modeling complex processes of land-use and land-cover
change in exurban parts of Southeastern Michigan. Based on two different conceptual models, one describing the evolution of urban
form as a consequence of residential preferences and the other describing land-cover changes in an exurban township as a consequence
of residential preferences, local policies, and a diversity of development types, we describe a variety of empirical data collected to support
the mechanisms that we encoded in computational agent-based models. We used multiple methods, including social surveys, remote sens-
ing, and statistical analysis of spatial data, to collect data that could be used to validate the structure of our models, calibrate their spe-
cific parameters, and evaluate their output. The data were used to investigate this system in the context of several themes from complexity
science, including have (a) macro-level patterns; (b) autonomous decision making entities (i.e., agents); (c) heterogeneity among those
entities; (d) social and spatial interactions that operate across multiple scales and (e) nonlinear feedback mechanisms. The results point
to the importance of collecting data on agents and their interactions when producing agent-based models, the general validity of our
conceptual models, and some changes that we needed to make to these models following data analysis. The calibrated models have been
and are being used to evaluate landscape dynamics and the effects of various policy interventions on urban land-cover patterns.
Ó2007 Elsevier Ltd. All rights reserved.
Keywords: Urban sprawl; Land-cover change; Land-use change; Spatial modeling; Ecological effects
1. Introduction
One of the most dramatic changes on the landscape of
the Eastern United States in the last 50 years has been
the fivefold increase in the area of land settled at exurban
densities (Brown et al., 2005a). This dispersed pattern of
land development at and outside the fringe of urban areas
has a range of effects on the functioning of ecological sys-
tems, through alterations in land-cover types and patterns
(Brown et al., 2000), surface hydrology (e.g., Groffman
et al., 2003), terrestrial habitat quality (Hansen et al.,
2005), and material and energy flows such as carbon
sequestration (Pickett et al., 2001). Understanding these
land-use and land-cover patterns and the processes that
give rise to them is important for managing landscapes to
minimize negative ecological effects and enhance positive
ones in the future. Importantly, because human–environ-
ment interactions are complex, achieving that understand-
ing requires analysis of a whole system of interactions. The
0016-7185/$ - see front matter Ó2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.geoforum.2007.02.010
*
Corresponding author. Address: School of Natural Resources and
Environment, University of Michigan, 440 Church Street, Ann Arbor, MI
48109-1041, United States.
E-mail address: danbrown@umich.edu (D.G. Brown).
www.elsevier.com/locate/geoforum
Available online at www.sciencedirect.com
Geoforum 39 (2008) 805–818
research reported in this paper seeks to understand the
interactions of human and social processes and land-cover
dynamics at and beyond the urban–rural fringe and how
complex dynamics affect these interactions. Ultimately,
the outcomes of interest are the measurable biophysical
landscape changes, but the explanations rest on the more
elusive interplay between humans and that same landscape.
Our research draws on a number of theoretical and
methodological themes from complexity science and the
study of complex adaptive systems (CAS), which are essen-
tial foundations for biocomplexity studies. We use the term
CAS to refer to systems that have (a) macro-level patterns
(i.e. aggregation), (b) autonomous decision making entities
(i.e., agents); (c) heterogeneity among those entities; (d)
social and spatial interactions that operate across multiple
scales; and (e) nonlinear feedback mechanisms (e.g., Axel-
rod and Cohen, 2000; Holland, 1995; Waldrop, 1992).
Reviews of these and other complexity science themes
and their relevance to geographical research have been well
done by Manson (2001), O’Sullivan (2004), and Manson
and O’Sullivan (2006). Rather than proposing a set of
hypotheses to be tested (e.g., as a theory) or making specific
ontological claims, the array of complexity science themes
offer a flexible ontology based on things (or actors) and
their relationships and makes epistemological claims about
how we can learn about systems using, primarily, simula-
tion modeling. In this paper, we use these themes to illus-
trate how empirical research has complemented
simulation modeling in our biocomplexity research, by
documenting system outcomes of interest, by parameteriz-
ing mechanisms encoded in models with data from South-
eastern Michigan region, and by responding to empirical
questions raised in the model development process.
We view exurban land-use change as an aggregate out-
come (Manson, 2001) that results from the interactions
of multiple actors, with each other and with the landscape,
and that produces the observable spatial and temporal pat-
terns of settlement and land cover. In our view the primary
set of actors affecting land-use and land-cover patterns at
the urban–rural fringe include the households that pur-
chase residential properties, the developers that make these
properties available to consumers, the farmers who use the
land for agriculture, and the local governments (including
their planning commissions and township boards) that reg-
ulate these transactions and provide infrastructure for the
new developments. Secondary actors also contribute to this
process by affecting the availability of credit (lending insti-
tutions), broader sets of infrastructure like the interstate
highway system (federal government), and environmental
regulations like those governing wetlands (state and federal
governments). In order to focus our attention on the effects
of land-use changes on the physical landscape, and because
of the importance of local governments in setting land-use
policy in the United States, we focused on the primary
actors, while recognizing that the others also play a role.
Because landscape characteristics (e.g., landscape aes-
thetic quality) and location relative to urban amenities play
a role in land-use decision making, and because those char-
acteristics are subsequently affected by residential loca-
tions, simple uni-directional models (i.e., in which
humans affect the landscape, but not vice versa) are inade-
quate. Feedbacks can be the essential ingredients that cre-
ate complex nonlinear dynamics and that can complicate
scaling from knowledge of agent behavior to understand-
ing system behavior, like overall landscape patterns
(Alberti, 2005). While a number of spatial land-use model-
ing frameworks exist in economic geography and regional
science (Briassoulis, 2000), only rarely do these models
consider landscape aesthetic characteristics or operate at
scales that would permit representation of the interactions
between land development and landscape patterns. More-
over, while many models of landscape aesthetic preferences
can be found in the landscape architecture, environmental
psychology, and cultural geography literatures (Nassauer,
1995; Daniel, 2001), these models rarely explicitly antici-
pate effects of changing aesthetic quality on urban land-
use change. Agent-based models have allowed us to repre-
sent the actors in this system and their interactions, and to
simulate landscape dynamics under multiple scenarios.
Spatially explicit agent-based models (ABMs) have become
popular tools for understanding land-use systems (Polhill
et al., 2001; Parker et al., 2003; Deadman et al., 2004),
but only a few have included these aesthetic feedbacks
(Irwin and Bockstael, 2004; Parker and Meretsky, 2004).
An agent-based model consists of a set of computational
objects, called agents, that interact in space and time
according to a set of rules. These agents can follow fixed
behaviors, e.g., buy the nearest piece of land, or they can
adapt their behavior, e.g., search k parcels and choose
the best, where the parameter k changes over time in
response to some feedback to the agent. These models pro-
duce system-level outcomes from the bottom up (Brown,
2005; Page, 2005). A pressing challenge in using these mod-
els to support understanding and scenario development is
to link them more closely with empirical research (Janssen
and Ostrom, 2007).
While computer models necessarily generalize a given
process, and are the products of theory, empirical data
refer to characteristics of specific cases (Brown et al.,
2005c). Referring to a fundamental tension in complexity
research between focusing on general system characteristics
and specific examples of actors within the system, Manson
and O’Sullivan (2006, p. 682) point out that ‘‘researchers
walk a fine line between holism and reductionism’’. The
mechanisms that support understanding a complex system
are at the micro-level (i.e., reductionism), even though the
goal of the analysis and modeling is to understand a system
or macro-phenomenon (i.e., holism). By using real-world
measurements to support the various processes encoded
in the models, we hope to better represent the processes
that give rise to observable patterns of land-use and land-
cover. ‘‘A fuller understanding of the relationship between
pattern and process ... seems most likely to be arrived at
by relatively abstract modeling of spatial systems combined
806 D.G. Brown et al. / Geoforum 39 (2008) 805–818
with considerable empirical grounding’’ (Manson and
O’Sullivan, 2006, p. 685).
Although any given model of a particular system can be
criticized as incomplete, as they certainly all are, we argue
that the process of building simple representations of sys-
tems can help us understand the implications of our con-
ceptual models and to identify areas where we need to
learn more. By formalizing our conceptual models as com-
putational models we uncover specific empirical needs
answerable with traditional observational approaches.
The results of our empirical investigations can then be used
to calibrate the essential mechanisms in our computational
models and discover alterations that must be made to the
initial conceptual model. We are therefore describing an
iterative process of conceptual model formation, parallel
efforts of empirical analysis and agent-based (i.e., com-
puter) model building, and conceptual model refinement,
but focusing on the roles of empirical observations in the
process.
This paper focuses on the empirical activities that have
supported our modeling work aimed at understanding
exurban land-use and land-cover change in Southeastern
Michigan. We structure the discussion of empirical activi-
ties around several key theoretical themes that are used
in complexity science: (a) macro-level patterns, (b) the
autonomous decision making entities (i.e., agents); (c) het-
erogeneity in agents and the environment; (d) interactions
that are structured socially and spatially and operate across
multiple scales; and (e) feedbacks. The remainder of the
paper discusses, first, our conceptual models based on
expert opinion and preliminary pattern analyses. We
focused on two of the key decision makers structuring
demand for developed land in exurban landscapes, i.e., res-
idents and land developers. We then formalized our con-
ceptual models as pilot ABMs to help us identify our
empirical needs, but leave out a detailed discussion of these
models to place emphasis on the empirical work. Next we
describe our empirical research that has been focused on
(1) quantifying the macro-patterns of land-use and land-
cover change that we want to understand using remote
sensing based measures; (2) quantifying the behavior and
heterogeneity in the actors and the landscape, and (3) defin-
ing the interactions among agents and between agents and
the landscape. Finally, though feedbacks can be thought of
as a type of interaction and can be the direct result of inter-
actions in a model or system, we discuss separately our
attempts to identify feedbacks within this system. We dis-
cuss the findings with respect this particular system, and
the general implications of an approach to understanding
complex systems that iterates between modeling and empir-
ical observation.
2. Study area
Our study focuses on the 10 counties in Southeastern
Michigan that comprise the Detroit, Ann Arbor, and Flint
metropolitan areas. This is an excellent region to study
land change at the exurban fringe. The region was home
to about 5.5 million people in 2000. Although the total
number of households residing in the four major cities of
the region (i.e., Detroit, Flint, Ann Arbor, and Pontiac)
declined from 494,374 to 455,099 between 1990 and 2000
– Ann Arbor was the only one of these cities that experi-
enced an increase in the number of households – the num-
ber of households in the ten-county region increased from
1.92 million to 2.08 million over the same period (US
Bureau of the Census, 2001). This disparity reflects a gen-
eral deconcentration of the population within the region
during the 1990s. At the same time (1992–2002), the
amount of cropland in the region declined from 1.32 mil-
lion acres to 1.25 million acres (US Department of Agricul-
ture, 2002).
3. Two conceptual models
We started by building two conceptual models to repre-
sent key processes hypothesized to produce observable pat-
terns of land use and land cover, and to guide the
construction of computational agent-based models and
the collection of empirical data. The models were based
on theoretical considerations about relevant actors in the
system, their interactions, and nonlinear dynamics that
might emerge through feedbacks in the system.
3.1. A model of the urban system
Our first model focused on the role of locational prefer-
ences of residential households in determining the spatial
patterns of development in and around a city center. We
call the model SLUCE’s Original Model for Exploration
(SOME). The model also includes service centers that
locate near recent residential in-migrants after a minimum
number of new residents has entered. The inclusion of ser-
vice centers captures a positive feedback between residen-
tial development and urban service provision: residents
draw service centers, and service centers, in turn, attract
residents. Residents or service centers locate in cells on a
lattice; only one agent is permitted per cell. The residential
agents use a boundedly rational decision making approach
(Arthur, 1994) to select the cell at which to locate. In a
model based on perfect rationality, residents would select
sites from among all available locations and select the site
that optimizes their utility. Our model assumes that resi-
dents optimize a utility function that includes distance to
service centers and aesthetic quality as critical components,
but their rationality is bounded by incomplete information
about the real estate market (Brown et al., 2004; Brown
and Robinson, 2006). The utility function the agents use
is a variation of the Cobb–Douglas utility function
(Chiang, 1984) and takes the following multiplicative form:
urðx;yÞ¼Y
m
i¼1
ðciðx;yÞÞair ð1Þ
D.G. Brown et al. / Geoforum 39 (2008) 805–818 807
where u
r(x,y)
is the utility of location (x,y) for resident r;a
ir
is the weight the resident r places on factor i;c
i(x,y)
is the
value of factor iat location (x,y), and mis the number
of factors evaluated (in our initial models, m= 2, distance
to service centers and aesthetic quality).
Access to services is a critical component of the utility
function, reflecting the longstanding tradition of modeling
land change as a function of access to markets and jobs
(Briassoulis, 2000). Existing empirical literature supports
this tradition and shows that road patterns are critical
for determining patterns of settlement (e.g., the hedonic
analysis of Boarnet and Chalermpong, 2001). Some land-
use models (e.g., DUEM; Batty et al., 1999) include road
building as an endogenous feedback process. Road build-
ing affects and is affected by settlement patterns. However,
we assumed that the bulk of the road system in Southeast-
ern Michigan was established either before or very early in
the period of interest (i.e., 1960–2000). While road capacity
has surely changed, we assumed that the spatial pattern of
the road network, which we believe is an important deter-
minant of the pattern of settlement, was relatively stable.
For this reason, we ignored the road system in cases where
we modeled hypothetical landscapes, basing access calcula-
tions on Euclidean distance, and used a static representa-
tion of the road network from most recent data to
calculate distances in cases where our models were intended
to represent real places (Brown et al., 2005b).
The SOME model used either designed landscapes with
assumed patterns of aesthetic quality (Brown et al., 2004),
or maps of aesthetic quality for areas of Southeastern
Michigan based on (a) relief – greater variations in which
relate to more aesthetic quality; (b) land cover – more for-
est cover and other open space increases aesthetic quality;
and (c) water – greater proximity to open water increases
aesthetic quality (Brown et al., 2005b). A feedback is impli-
cit in this simple definition of aesthetic quality. As develop-
ment changes land-cover it, therefore, affects aesthetic
quality for subsequent residents.
We implemented this model using Swarm (http://
www.swarm.org) and explored its dynamics in the absence
of empirical data to (a) compare the functioning of the
model with an analytical model of the effects of a greenbelt
near a growing city (Brown et al., 2004); and (b) demon-
strate that the decision making of agents in the model
can generate distributions of developed cluster sizes that
compare well with the structural form of real-world cities
(Rand et al., 2003). Furthermore, the model has been used
to demonstrate new techniques for spatial model validation
that recognize the possibilities for path dependence in the
agent-based models (Brown et al., 2005b).
3.2. A model of an exurban township
Our second conceptual model focused on the evolution
of land cover as residential development expands within
exurban townships, and includes agents that represent
township policy boards, developers, farmers, and residents
that interact at a number of different scales. We call this
model Dynamic Ecological Exurban Development
(DEED). Developers assess township- and farm-level char-
acteristics before making heuristic decisions about where to
develop a specific type of subdivision. Residents decide
where to locate by assessing environmental characteristics
of lots within subdivisions. Different types of subdivisions
have differential appeal to residents and differential effects
on the land cover. The model consists of four agent types
(farm households, resident households, developers, town-
ships) and three spatial-object types (farms, subdivisions
and lots). Agents in the model have their own behavior;
objects may be created, used, or eliminated by agents in
the model but do not have their own behavior.
A township (representing an area of about 9350 ha) is
divided into farms, using either a regular grid pattern or
a GIS-based map. Farmers create and offer for sale rural
lots (Table 1) on some parts of their farms that are both
(a) near county roads; and (b) on poor quality soil. Farms
are then made available for sale with some probability each
year. Farms are labeled as suitable for one or more of three
subdivisions types–country, horticulture, and remnant—
depending on their environmental characteristics (Table
1). Each subdivision type has a minimum lot size and a
set of location rules. The developers determine if they are
available to create a new subdivision. Developers may be
constrained by a minimum lot size policy enforced by a
township. In our conceptual model, minimum lot-size reg-
ulations favor higher-priced homes with lots that are more
likely to contain natural features. Additionally, developers
respond to demand from residents for particular types of
lots. If a developer builds a subdivision of a given type
and a threshold percentage of lots remain unsold, there
are reasons to believe that they would be unable or unwill-
ing to obtain financing for another subdivision of the same
Table 1
Residential development types in our expert conceptual model
Lot sizes Effects on Tree cover Locations
Country subdivisions Smallest Decline Are developed where there is no change in elevation, no water, and no forest
Horticultural subdivisions Medium Constant May be developed in any location regardless of environmental amenities
Remnant subdivisions Largest Increase Are developed where variation in elevation, water, or forest exists
Rural lots Most variable Decline Near county roads and on non-prime soil
Each of these characteristics is subjected to threshold values (30%, 1%, and 10%, respectively) that govern the required amount to signify their existence at
a location. These thresholds and assumptions govern development behavior for farms greater than or equal to 160 acres, other heuristics apply for smaller
sized farms or rural lot developments.
808 D.G. Brown et al. / Geoforum 39 (2008) 805–818
type. If available, the developer develops the lots within the
subdivision and offers them for sale to residents.
Residents enter the model at a constant rate (e.g., 10 per
time step) and have variable preference weights for envi-
ronmental characteristics and heuristics describing which
subdivision type they prefer. Both preference weights and
subdivision choices are determined by their socioeconomic
characteristics, including income, parental status, and
whether or not they belong to an environmental group.
Membership in an environmental group is included as a
lifestyle indicator, recognizing that choices are not wholly
determined by life-stage factors. Residents randomly select
lots in a number of subdivisions for evaluation and move
into the most suitable subdivision or exit the model if no
suitable lot can be found (e.g., there are no available lots
or they are unaffordable). The utility calculation for resi-
dents is as follows:
Resident Utility :
¼aforest subfarea þapv subpv þarelief subrelief þawater subwarea
4
Subdivision Evaluation ð2Þ
where the a’s are the resident’s preference value for forest,
panoramic view (pv), change in relief (relief), and water.
The sub variables are the subdivision environmental feature
values that correspond to the resident preferences (area for-
est – farea, area water – warea, others as listed above). The
last term Subdivision_Evaluation represents a heuristic deci-
sion tree that determines the resident’s evaluation of a sub-
division based on its type and his/her socio-economic
characteristics (Fig. 1). All values in the utility equation
are scaled to a range of 0.0–1.0.
Each cell in the landscape begins with an initial amount
(proportion) of tree cover. The tree cover is modified over
time, according to the type of development that occurs. If
the cell is part of a country subdivision or a rural lot, all
trees are removed. If the cell is part of a horticultural sub-
division, the trees remain unchanged. If the cell is part of a
remnant subdivision, the tree cover is incremented in all
cells until the total tree cover in the subdivision is 20%,
or until the tree cover is 20% higher than it started with,
whichever amount is greater. These landscape changes
reflect the relative effects we expect the subdivisions to have
on tree cover. The initial regrowth rates were set to reflect
our understanding of tree-cover patterns within subdivi-
sion types.
With this model we have explored (a) the interactions
between public and private lands in creating habitats of
various kinds, focusing mainly on tree cover, in exurban
regions; and (b) the payoffs to townships seeking to plan
for an increased tax base, in the form of more and wealth-
ier residents, and increased ecological quality, in the form
of increased tree cover (Zellner et al., in review). Running
the model with two adjacent townships, each with the abil-
ity to set their own lot-size regulations, reveals the types of
policy games that can emerge as townships independently
seek their own payoffs within a regional context.
Is the lot part of a Country Subdivision?
Is the resident household
not part of an Environmental group
and has a low income or kids?
Is the resident
household part of an
environmental group
and has low income
or
has high income and
does not have kids
and is not part of an
environmental
group?
Value=1
Is the lot part of
a Remnant
Subdivision?
Yes
Yes
Yes
No
Yes No
No
Value = 0
Does the resident
household have a high
income?
Yes No
Is the lot part of
A Horticultural
Subdivision?
Value = 0.5 Value = 0
Value = 1
Is the resident household
part of an environmental
group?
Value = 0.5
Value = 0
Yes
Does the resident
household have a
high-inc?
Value = 0
Value =1
No
No
Yes No
Yes No
Fig. 1. Heuristic decision tree used by resident agents to evaluate the social and landscape characteristics of a subdivision.
D.G. Brown et al. / Geoforum 39 (2008) 805–818 809
4. Empirical elements of complexity
4.1. Macro-level patterns
An analysis of a complex system often begins with
patterns in space and/or time that are difficult to
explain with simple linear relationships. Our project focused
on both the spatial patterns of land development, the
subsequent effects on land-cover proportions, and the
temporal trends in those proportions. We use these obser-
vations as indicators of the ecological effects of land
change.
To describe the land-use and land-cover changes that
have occurred outside the core urban areas in Southeastern
Michigan, we compiled data on the land-uses of parcels
and the proportional composition of these parcels in multi-
ple land-cover classes (i.e., tree cover, impervious surface,
agriculture, other-natural covers) within selected townships
between 1950 and 2000. We selected 13 townships for anal-
ysis and to represent a range of conditions with respect to
the amount and timing of population growth and develop-
ment. Land-owner parcels within the townships were digi-
tized from plat books compiled in the late 1950s, 1960s,
1970s, 1980s, and 1990s (Rockford Map Publishers,
1956–1999). Land-uses and land-covers were interpreted
for each parcel by overlaying parcel boundaries on aerial
photographs that were selected from the nearest available
date (a maximum 2-year difference) to the parcel maps
and scanned at 2 m resolution.
Based on data from 11 of our 13 sample townships
(those for which sufficient data were available), tree cover
increased in our sample townships by an average of 1.8%
(or 29 ha) per year between the mid 1950s and late 1990s
(the range across the 11 townships was from +0.4% to
+3.5% per year). Because of the location of the region
within the Eastern temperate forest zone, increasing tree
cover is a possible indicator of ecologically beneficial
landscape changes. However, impervious surfaces also
increased, by an average of 11.1% (or 28 ha) per year (rang-
ing from 3.3% to +26.2% per year). Because of decreased
groundwater recharge, increased runoff, and consequent
potential for surface water pollution, such increases can
indicate negative ecological effects. These increases were
largely at the expense of agricultural land covers, which
declined an average 1.5% (or 70 ha) per year (ranging from
2.9% to 0.04% per year). It is these landscape changes,
as well as their effects on spatial patterns, that we seek to
understand through modeling.
The trajectories of land-covers in three illustrative town-
ships (Fig. 2) demonstrate the land-cover transitions the
townships are undergoing. The three townships, Tyrone,
Scio, and Rochester, are representative examples of town-
ships with low (9 people/10 ha), moderate (23 people/
10 ha) and high (83 people/10 ha) levels of population den-
sity in 2000, respectively. They each demonstrate a steady
decline in crop cover and steady increase in impervious
cover over a period of approximately 40 years. Crops are
being replaced in large measure by trees and other land
covers (including managed and unmanaged grasslands).
Each of these cases suggests a phase transition in the
amount of tree cover on the landscape. From a low level
of tree cover associated with agricultural landscapes to a
somewhat higher level associated with residential land-
scapes. Rochester has certainly completed that transition;
and changes is tree-cover in Scio Township also appear
to have slowed. Tree cover in Tyrone Township was
increasing rapidly during the 1990s, indicating that town-
ship is in the early stages of this transition.
In addition to these changes in land cover, we have used
spatial metrics of land-use and -cover patterns, for example
using the approach taken in the Fragstats software (McGa-
rigal and Marks, 1995), and maps that identify regions of
invariant and variant outcomes from the models (Brown
et al., 2005b) to provide multiple patterns for assessing
the usefulness of the model for experimentation and sce-
nario analysis. Following the pattern oriented modeling
(POM) approach described by Grimm et al. (2006),we
are interested in understanding how these spatial and tem-
poral patterns in land use and land cover come about as a
result of agent-level processes, and we have compared
agent-based model output to multiple pattern descriptions
to enhance our confidence in them (Rand et al., 2003;
Brown et al., 2005b).
Tyrone Township, Livingston County
0%
20%
40%
60%
80%
100%
1954 1971 1978 1989 1998
Area (%)Area (%)Area (%)
Scio Township, Washtenaw County
0%
20%
40%
60%
80%
100%
1957 1967 1 979 1989 1999
Rochester, Oakland County
0%
20%
40%
60%
80%
100%
1956 1970 1977 1988 1998
Water
Other
Agri
Imperv
Trees
Water
Other
Agri
Imperv
Trees
Water
Other
Agri
Imperv
Trees
Fig. 2. Changes in proportions of land covers in three sample townships.
810 D.G. Brown et al. / Geoforum 39 (2008) 805–818
4.2. Micro-level decision making
The ways agents make decisions are central to determin-
ing the overall functioning of an agent-based system. Deci-
sion models specify what information the agents use and
how they combine that information with their own prefer-
ences to decide which specific actions to take. They also
determine how agents interact with each other and/or with
the environment. In the SOME model, residential agents
make decisions about where to locate themselves after
gathering information about the characteristics (i.e., near-
ness to services and aesthetic quality) of a sample of sites
and calculating a utility for each that combines that infor-
mation with weights representing their preferences for cer-
tain characteristics (Eq. (1)). The DEED model includes
developer agents that make decisions about where to locate
developments of a given type based on the landscape and
location characteristics of available farms. Residential
agents then locate themselves within subdivisions using a
utility calculation (Eq. (2)) that considers the physical land-
scape characteristics associated with different subdivision
types.
4.2.1. Residential locations
Empirical support for the decision-making models can
take many different forms (Robinson et al., 2007; Janssen
and Ostrom, 2007). Empirical challenges include (a) evalu-
ating if the structure of the decision model is correct; (b)
assessing which factors and inputs the agents consider in
making the decisions; and (c) determining the appropriate
weights to assign for each agent and each factor (i.e., a
ir
).
We used survey data to answer the last two of these empir-
ical imperatives, but did not evaluate the first.
For independent data on the distributions of residential
preferences in the population of the region, we turned to
survey research on residential preferences (Marans, 2003).
The data were derived from household surveys conducted
in the Detroit metropolitan area during the spring and
summer of 2001. In part of the survey, each respondent
was asked about the relative importance of factors influ-
encing their decision to move to their current house. A
four-point importance scale ranging from ‘‘very impor-
tant’’ to ‘‘not at all important’’ was used for the following
12 factors: close to work; good schools; housing costs and
good value; convenient to places such as shopping and
schools, lots of recreational opportunities; attractive
appearance of neighborhood; community size; people sim-
ilar to me; appearance and layout of the dwelling; familiar
with area; close to natural areas (woods, ponds, streams,
etc.); openness and spaciousness of area; and close to fam-
ily and friends.
We analyzed these data to evaluate the correlations
among responses to the different factors (using factor anal-
ysis), the clustering of responses around particular patterns
of responses (using cluster analysis), and the relationships
between responses and characteristics of the households,
e.g., education, income, race, marital status and age (Fer-
nandez et al., 2005). The analysis was limited to respon-
dents who had moved to an exurban location within the
last ten years. t-tests (detailed by Fernandez et al., 2005)
revealed the following relationships: having children under
18 in the household resulted in stronger preference for
nearness to work and good schools, and weaker preference
for residential aesthetic concerns; households headed by
married couples had stronger preference for nearness to
work and good schools; households with income greater
than $75,000 had weaker preference for social comfort fac-
tors (e.g., near family and friends, familiar with area, peo-
ple similar to me); and respondents over 40 years old had
weaker preference for nearness to work and good schools.
Neither college education nor membership in a minority
group were significantly related to differences in stated
preferences. Overall, the preferences of residents were only
weakly related to the measured household characteristics,
and these characteristics serve as only partial surrogates
for location preferences. It is likely that additional lifestyle
factors are needed to explain preferences, which is why we
included membership in an environmental group in the
DEED model.
To identify the relative importance of landscape factors
in our conceptual model of homeowner decisions, which
includes landscape characteristics at the regional, subdivi-
sion, and lot scales, we conducted a web-based choice
experiment with 494 homeowners in Southeastern Michi-
gan who lived in areas that were dominated by large-lot
zoning as identified by municipal boundaries. This web-
based survey invited homeowners to ‘‘shop’’ for a new
home, neighborhood, and yard within the price range of
the home they currently owned. It presented a range of eco-
logically beneficial and conventional designs for subdivi-
sions and finer-scale residential development features
(e.g., open spaces, streets, and yards) and allowed respon-
dents to choose their most preferred new house within a
large-lot exurban residential development. Survey respon-
dents also provided information on other demographic
and behavioral variables that we believe are potential cor-
relates with landscape preferences across scales.
The web-based survey results indicated that increased
neighborhood tree cover is positively correlated with home-
owner preference in new exurban subdivisions, accounting
for 52.1% of variance in preference for neighborhoods and
for 56.2% of variance in preference for residential streets.
Responding to views of exurban open spaces, homeowners
strongly disliked large areas of turf, including athletic play-
ing fields, and preferred wooded areas; these two factors
together accounted for 74.6% of variance in preference
for views of exurban open spaces. These results support
Eq. (2), which demonstrates that those who can afford to
do so would prefer to live in wooded exurban subdivisions
near forested open spaces, like the more expensive remnant
subdivisions in the conceptual model.
Although survey data can provide valuable information
about the distributions of agent characteristics, and of sta-
ted preferences, they provide no indication of the processes
D.G. Brown et al. / Geoforum 39 (2008) 805–818 811
by which agents make decisions. In our simple agent-based
model, we assumed that all residents optimize a utility
function that was of the same form for each agent. The nat-
ure of this utility function, and the ways in which agent
rationality is bounded, are elements of our conceptual
model that remain to be tested against empirical or exper-
imental cases. Hedonic price models can reveal parameters
that describe variations in price, and can be used to infer
variations in agent preferences (e.g., Geoghegan et al.,
1997), but they are similarly challenged to provide justifica-
tion for the structural form of the utility function.
Although we have used sensitivity tests to evaluate the
implications of different utility models (Rand et al.,
2002), additional work, using field and lab experiments
for example, is needed to evaluate how residents actually
decide, and how heterogeneous are the ways people actu-
ally make land-use and land-cover decisions (Evans et al.,
2006).
4.2.2. Subdivision locations
Initially our conceptual model used heuristics, informed
by expert opinion, to describe how the development types
are located on the landscape (Fig. 1,Table 1). We opera-
tionalized the conceptual model as a pilot ABM, i.e.,
DEED. We then empirically evaluated the location of
development types by testing the relationships between
development events and locational attributes, including
environmental, geographical, and socio-economic variables
(An et al., submitted for publication). The environmental
variables we measured for each sampled subdivision were
soil quality, slope, and initial tree cover (i.e., that observed
in the decade before the subdivision was built). The geo-
graphical variables included distances to the nearest city
in each of three hierarchical levels (Detroit, five mid-level
cities, and small urban areas), along with distances to near-
est water feature (i.e., lake or river), highway, and county
road. All of the environmental and geographical variables
were measured at the most recent time for which data were
available. The time-varying socio-economic characteristics
of townships in which each development fell were collected
from US census data. Factors measured for each decade
and each township included population density, population
change rate, median age, and education level.
We used survival analysis to understand relationships
between development of each type and the explanatory
variables. An important advantage of survival analysis is
its ability to account for time-varying factors that affect
establishment of the development types and the inherent
inaccuracy in measuring the timing of the events. A sur-
vival model takes the following general form:
log hiðtÞ¼aiðtÞþb1Xi1ðtÞþb2Xi2ðtÞþþbkXik ðtÞð3Þ
where h
i
(t) is the time-varying hazard rate (instantaneous
risk of being developed at a time) for parcel i,X
ik
(t)is
the value of explanatory variable X
k
for parcel iat time
t, and b
k
are the coefficient values for the kth variable. In
addition, b
k
can vary if an interaction term with time is
added in the model.
The analysis results confirmed several aspects of the
conceptual model regarding placement of developments
of different types (Table 2). Results of the survival analysis
supported our assumptions that the prevalence of remnant
subdivisions tended to increase with distance from county
roads, higher initial tree cover, higher slope, and proximity
to Detroit and five mid-level cities. As time went on, rem-
nant subdivisions occupied land of increasingly good soil.
Horticultural subdivisions were more likely to be located
in areas closer to the five mid-level cities and became
increasingly proximate to Detroit over time. Country sub-
divisions tended to be near county roads and far from
Table 2
Effects of landscape variables on the location of subdivision developments
Unit Country subdivision Horticultural subdivision Remnant subdivision
DEED
Pilot
Model
%Din
hazard rate
per unit D
Scaled
alpha
values
DEED
Pilot
Model
%Din
hazard rate
per unit D
Scaled
alpha
values
DEED
Pilot
Model
%Din
hazard rate
per unit D
Scaled
alpha
values
Distance from
county roads
1 km Strong
negative
0.65 0.00 No
influence
0.21 0.04 Positive 0.13 0.00
Distance from
water
1 km Positive 164.34 0.60 Negative 24.24 4.62 Negative 46.69 0.39
Percent tree
cover
1% No
influence
0.93 0.00 Positive 2.06 0.39 Strong
Positive
1.26 0.01
Soil quality
(prime
farmland)
0 or 1 N/A 48.55 0.18 N/A 3.05 0.58 N/A 69.67 0.58
Percent slope 1% Strong
negative
60.65 0.22 Positive 12.81 2.44 Positive 8.49 0.07
Distance from 5
city centers
1 km N/A 2.13 0.01 N/A 4.74 0.90 N/A 3.18 0.03
Distance from
Detroit
1 km N/A 2.64 0.01 N/A 5.60 1.07 N/A 2.64 0.02
812 D.G. Brown et al. / Geoforum 39 (2008) 805–818
water. They were located on good soils at early time peri-
ods, but this soil effect diminished over time.
Using the hazard-rate equations for the development of
each subdivision type, we extracted the relative difference
and direction of influence of each independent variable
on the hazard rate. Then we rescaled these relative weigh-
tings to empirically inform the preference weights in a util-
ity function used by developer agents to evaluate farms for
subdivision. Using the same form as Eq. (1), the indepen-
dent variables are weighted using the scaled values listed
in Table 2. It should also be noted that the survival analysis
identified three variables of influence (i.e. soil quality, dis-
tance to five mid-level cities, and distance to Detroit) that
were not included in the conceptual or pilot models, but
were significant for identifying the location of at least one
of the subdivision types. Because the DEED model was
implemented at the township level, the township variables
that had been included in the survival model were not used
in the computational model.
Though our empirical analysis does not confirm that the
factors used in the statistical model were, in fact, those con-
sidered by developers, it provides evidence that the devel-
opment process results in the development types being
spatially distributed in ways that our conceptual model
suggests. Furthermore, the identification of additional vari-
ables of significant influence from our empirical analysis
supports the need for modeling to be an iterative process
and refinement of conceptual design, empirical data collec-
tion, and model construction.
4.3. Heterogeneity
Systems composed of multiple agents that interact to
create feedbacks can be very sensitive to the actions of a
small number of agents that have particular characteristics.
For this reason, understanding the actions of average
agents is insufficient to explain observed patterns and it is
important to understand the nature of heterogeneity
among agents within a system. To better understand the
implications of the heterogeneity found in human systems,
more researchers are using agent-based techniques (e.g.
Parker et al., 2003). Because of their ability to both formal-
ize heterogeneity to a degree that more closely parallels real
systems and track or trace the behavior of an individual
agent and/or group of agents, agent-based tools can better
represent complex interactions among many heterogeneous
actors than traditional mathematical models of land-use
systems.
4.3.1. Resident preferences
To evaluate the effects of varying preferences on land-
use patterns in the urban growth model, we used character-
izations of heterogeneity of preferences from survey
respondents directly, i.e., based on the factor analysis of
Fernandez et al. (2005). The SOME model was run with
agents having varying degrees of heterogeneity in prefer-
ences, including the following cases: (1) no heterogeneity
(i.e., homogeneous agents); (2) normal distributions of
preferences describing the factor scores that were based
on the survey results; (3) mean preferences for seven groups
or clusters, identified from cluster analysis of the survey
data; (4) seven different normal distributions of preferences
representing variability within each of the seven clusters;
and (5) uniform random distributions, set up with no infor-
mation from the survey as a null model for comparison.
Results of model runs that included the heterogeneous res-
ident preferences, whether drawn from the survey or drawn
randomly, exhibited more sprawling and fragmented pat-
terns than did runs of the same model with average agent
preferences; and agents were able to achieve higher levels
of average utility when their preferences varied (Brown
and Robinson, 2006). These patterns can be attributed to
the importance of residential agents with a very strong
preference for aesthetic quality, relative to other factors,
and the feedbacks in our model that involve stochastic
placement of service centers to serve existing residents.
The sensitivity of the results to heterogeneity supported
our assumption that agent-based models would be helpful
in understanding land-use systems.
4.3.2. Differentiation of development types
Our residential development typology defines four types
of exurban lots or subdivisions, each of which was defined
by observed land-cover proportions and patterns, street
patterns, and lot sizes. We hypothesized that these types
were different in terms of the effects they had on land-cover
changes, and therefore subsequent ecological effects. The
heterogeneity of these ecological effects is one mechanism
by which landscape patterns can be determined by agent
level actions. Empirical tests of the observed land-cover
changes associated with each type constitute a partial vali-
dation of the typology and of these mechanisms.
To test land-cover change impacts of developments, we
used to parcel maps obtained from eight townships,
together with recent aerial photographs, to identify subdi-
visions and label their type. We sampled approximately 4%
of the parcels in these townships (n= 854). Parcels were
merged to form larger polygons representing subdivisions.
Using the historical aerial photographs, we recorded the
land-cover characteristics within these polygons, focusing
on percent tree cover, in each decade between the late
1950s and late 1990s and also identified the decade during
which each subdivision was started.
For a stratified random sample (n= 427) of subdivi-
sions, we found significant differences between develop-
ment types and the proportional change in tree cover and
other natural cover. For each of the subdivisions, we mea-
sured the tree-cover percentages in the images taken imme-
diately before the development and the average (over
decades) percentages after the development, and then we
did a series of two-sample t-tests. We found that tree cover
of a location tended to increase after development of rem-
nant subdivisions (p< 0.05 for null hypothesis that the per-
centages before and after developing into remnant
D.G. Brown et al. / Geoforum 39 (2008) 805–818 813
subdivisions are equal). There was some, but less consis-
tent, evidence that tree cover tended to decrease after devel-
opment of country subdivisions (p< 0.05 for the same null
hypothesis), while no significant change in tree cover was
found in horticultural subdivisions (p= 0.30). This analysis
substantiated the directionality of our assumptions about
the differential landscape effects of subdivisions that we
had represented in the DEED model, in which tree cover
increased for remnant subdivisions, decreased for country
subdivisions, and stayed constant for horticultural sub-
divisions.
4.4. Interaction
Interactions between agents can include the means by
which they communicate with each other or the effects of
one agent on the environment (e.g., landscape), which
can affect the subsequent actions of other agents. Our mod-
els included a variety of interactions for which we
attempted to collect empirical data. The interactions
between developer and residential agents in the DEED
model operate across scales and were affected by the indi-
rect effects of developments on landscape characteristics
(described above), but also by the effects of costs of devel-
opments on the ability of residents to pay for the lots they
prefer. We validated these latter interactions by examining
differences in housing price within subdivisions in South-
eastern Michigan that we had labeled as remnant, horticul-
tural, or country subdivisions. Our analysis of the survey
data on residential preferences highlighted the potential
importance of social interactions, which we examined with
experiments using the SOME model. Finally, the rules gov-
erning the effects of development on tree cover in the
DEED model raised questions about possible spatial inter-
actions that were not accounted for in our simple rules for
tree-cover effects based on subdivision type alone. We
sought to answer these questions with remotely-sensed data
on the spatial patterns of tree-cover effects by subdivision.
4.4.1. Cross-scale interactions
In the empirical test of market valuation, we examined a
sub-sample of these subdivisions (n= 826), and looked at
the assessed market valuation of individual properties
within the subdivisions. We constructed a regression model
to evaluate the factors that relate to the market value of
homes (as the dependent variable). Included as indepen-
dent variables were house size, lot size, house age, and
dummy variables for the type of subdivision the house
was in (i.e., country, horticultural, and remnant subdivi-
sions). Controlling for the other variables, development
type was a significant predictor of the market valuation
of homes, predicting 8.4% of the variance in standard
equalized valuation (p< 0.000). House size was nearly
seven times more important in the model, while lot size
and house age were about 1.5 times as important. Houses
in remnant subdivisions were significantly more expensive,
holding other factors constant, than houses in the other
types of subdivisions. This cost effect is compounded by
the fact that remnant subdivisions tended to contain larger
houses on larger lots than the other types. This empirical
test supported the validity of our assumption that resi-
dents’ selection of different types of developments would
be influenced by the wealth of the agents.
4.4.2. Social interactions
Our initial conceptual and computational models of
SOME incorporated only aesthetic quality and distance
to service centers as landscape attributes influences residen-
tial location decisions. Results from our analysis of the sur-
vey suggested that residents also used a third factor in
deciding where to locate, which was labeled social comfort
(Fernandez et al., 2005). Some agents heavily weighted
their location decision based on such characteristics as if
the location was near family and friends, if a neighborhood
was populated by people like themselves, and if they were
familiar with the area. As a result of this analysis, we mod-
ified our conceptual model and the corresponding agent-
based model to include a consideration of neighborhood
similarity as a factor in the utility function (Eq. (1);Brown
and Robinson, 2006).
4.4.3. Spatial interactions
The DEED model includes a representation of the land-
cover effects of residential development, through changes in
tree cover. It does not currently have explicit mechanisms
for modeling the spatial patterns of tree-cover change,
either within or among subdivisions. To develop such
mechanisms, so that we can use the model to understand
the spatial patterns and amount of tree cover, we have pur-
sued two strategies. We have used the results of the web-
based survey to evaluate the importance of spatial interac-
tions in the land-cover decisions of residents, and the aerial
photography to evaluate the spatial adjacency effects in the
amounts of tree-cover within subdivisions.
The web-based survey included questions that allowed
us to evaluate the effect of neighbors’ landscape patterns
on homeowners’ preferences for their own landscapes.
When selecting a specific design for their yard from among
multiple alternatives, residents were shown simulations of
what the yards of their neighbors would look like in their
hypothetical home-shopping situation. The results indi-
cated a strong effect of neighboring yards on residents’
choices. Specifically, when respondents were shown neigh-
boring yards that had conventional designs they preferred a
conventional design for their own yard significantly more
than any other yard design (F= 42.73, p< 0.0001). How-
ever, when respondents were shown that all adjacent neigh-
bors’ yards used innovative ecological designs, contrary to
broad cultural conventions, respondents preferred an inno-
vative design for their own yard significantly more than
any other yard design (F= 51.83, p< 0.0001) (Nassauer
et al., in preparation).
Though our analysis (described in the section on Differ-
entiation of Development Types) demonstrated that the
814 D.G. Brown et al. / Geoforum 39 (2008) 805–818
subdivision types had differential effects on the dynamics of
tree cover that were consistent in relative terms with our
hypotheses, implementing our conceptual model and eval-
uating its ecological landscape effects required that we also
define the model in terms of the amount and pattern of
land-covers resulting from different development types.
To evaluate the differential effects of development on
land-cover patterns, we used the aerial photos together
with the parcel maps to quantify how the amount and pat-
tern of land-covers were related to land-use types under
various conditions. For example, the frequency of low-den-
sity residential developments in our 13-township sample
decreases with increasing percent tree cover with a mean
and standard deviation of 18.9% and 17.9%, respectively.
Using such a distribution, the amount of tree cover can
be assigned probabilistically by the agents using the land.
The amount of land cover assigned through this method
provides an initial estimate. Further refinement of this
approach will require testing for adjacency effects, i.e.,
how land cover in a subdivision might be related to the
land-covers of neighboring subdivisions, and whether these
adjacency effects can be observed to change over time.
Incorporation of adjacency effects can create the additional
nonlinear feedbacks within the system. We plan to evaluate
such effects using summaries from our land-cover data ana-
lyzed at the scale of subdivisions.
4.5. Feedbacks
Our conceptual models include a number of feedbacks
that we believe are important in exurban land-change pro-
cesses, including those between landscape aesthetic quality
and residential and subdivision locations and designs and
those between residential locations and the locations of
urban services. These feedbacks are an important controls
on the models of this system (as well as the system they rep-
resent), observed complex nonlinear dynamics, and sensi-
tivity to both initial conditions and to small
perturbations in the distributions of inputs (i.e., sensitivity
to heterogeneity). We addressed the effects of landscape
characteristics on residential preferences, an important
form of feedback, in our web-based survey (described
above).
A number of existing models address the important
feedback between land-use change and transportation sys-
tems (Wegener, 2005). Our assumption of a static road net-
work raised an important empirical question, about the
degree to which the road network in the region had chan-
ged over our study period as a result of infrastructure
investment. Using our compiled aerial photographs, we
mapped roads in each decade using a classification system
based on the National Functional Classification (NFC)
system developed by the US Federal Highway Administra-
tion (US FHWA, 1989). We calculated the sum of the
length of all roads across all 13 townships for each date
and each class (Fig. 3). The most substantial increase in
the length of principal, minor, or collector roads occurred
between 1950 and 1960, with the creation of the interstate
highways. After that period, the length of these road types
increased by only 13%, 4%, and 10% from 1960s to 2000,
respectively. Therefore, though the road system likely had
a significant impact on settlement patterns, the roads that
establish the accessibility of various locations on the land-
scape changed only a little after the initial decade of the last
half of the 20th century. On the other hand, the lengths of
local roads and, especially, residential roads increased quite
dramatically over the entire period (155% from the 1950s to
2000). These increases, along with the widening and
increased traffic on the major roads, may have directly
affected both ecological systems (Forman et al., 2003)
and accessibility within the urban system. Given our focus
on townships in Southeastern Michigan between 1950 and
2000, this analysis supports our decision to treat the system
of major roads and highways as exogenously determined,
but points out that ecological effects associated with resi-
dential road building might, nonetheless, have been
substantial.
5. Discussion and conclusions
In order to characterize the dynamics of a system that
couples human decision making about land use and land
management at the urban–rural fringe with biophysical
changes on the landscape, we have constructed conceptual
models that take an agent-based view of that complex sys-
tem. Our conceptual models incorporate a number of
themes from complexity science, including: (a) macro-level
patterns, (b) the autonomous decision making entities (i.e.,
agents); (c) heterogeneity in agents and the environment;
(d) interactions that are structured socially and spatially
and operate across multiple scales; and (e) feedbacks. The
conceptual models have led us in two epistemological direc-
tions simultaneously. First, we have used and constructed
agent-based computational models to represent key aspects
of the system. These computational models force us to for-
malize our knowledge of the system and allow us to evalu-
ate the implications of that knowledge in ways that other
analytical approaches do not (Lempert et al., 2002).
Secondly, we present in this paper empirical data col-
lected to describe both macro- and micro-level patterns
All Townships
0
500
1000
1500
2000
2500
1950 1960 1970 1980 1990
Years
Total Length (km)
Residental
Local
Collector
Minor
Primary
Fig. 3. Trends in road lengths by type of roads across 13 townships in
Southeastern Michigan.
D.G. Brown et al. / Geoforum 39 (2008) 805–818 815
and processes for the purposes of refining and validating
our conceptual models as well as providing patterns to cal-
ibrate our computational models. The divergent nature of
our empirical needs, described in this paper, has required
us to employ a range of methods, including survey
research, remote sensing, spatial analysis, and survival
analysis. This work leads us to some general conclusions
about the relationship between general models and specific
places, as well as specific conclusions about our own con-
ceptual models of land-use and land-cover change at the
urban–rural fringe.
Our work suggests that the distribution of preferences
and behaviors the various actors (e.g., residential land pur-
chasers and developers) can have significant effects on the
settlement patterns that result from the interactions of
those actors. We have no evidence that the distributions
vary from region to region, as our focus was exclusively
on Southeastern Michigan, but we do find significant vari-
ation among actors in the region and, to the degree that the
distributions of variation are different in different regions,
we might expect differences in land-use dynamics and spa-
tial patterns.
Variation in natural landscape characteristics might also
be expected to influence settlement patterns and subsequent
effects on landscapes and ecological processes. Our analysis
identified four types of developments that we observe in
exurban parts of Southeastern Michigan. One of these
types, i.e., remnant subdivisions, is associated with the
presence of environmental amenities that have the poten-
tial to provide environmental benefit. Because these subdi-
visions have significantly different effects on land cover
from the other types (i.e., tree cover increased on these,
but remained constant or decreased on the others), the
abundance and spatial distribution of environmental ame-
nities could have a significant influence on (at least the
potential) patterns of development and their effects. Fur-
thermore, although our analysis does not permit conclu-
sions on this question, it is possible that the relative
abundance of different development types, and even the
types themselves, could be different in different regional set-
tings, with different environmental amenities (e.g., moun-
tains) and different planning environments (e.g., growth
boundaries).
Our methodology is drawn from complexity science and
focused on understanding the formation of aggregate, or
macro-level, patterns from micro-level processes. To better
understand the macro-level patterns, we collected data on
the trajectories of land-cover change within 13 sample
townships in Southeastern Michigan. Many of the town-
ships have undergone in the last half of the twentieth
century, or are still undergoing, a transition from predom-
inantly agricultural land uses to predominantly residential.
The transition is manifested in a steady, and sometimes
rapid, decline in land covers associated with crops, a steady
increase in impervious surface, and a phase transition in
the amount of tree cover from a relatively low level to a
somewhat higher level. The computational models we have
built are intended to help us understand both the spatial
patterns of settlement (in the case of SOME) and the
amounts and distributions of tree cover (in the case of
DEED) in exurban Southeastern Michigan.
Our empirical data collection about micro-level pro-
cesses has focused on the decision making of and heteroge-
neity in the two primary actors, the residents who buy
residential lots and the developers who build them (repre-
sented by types of developments). Our analysis generally
confirmed several critical features of our conceptual mod-
els. First, residents consider both aesthetic quality and
proximity to urban services and jobs in their selection of
a residential location. Second, the relative importance res-
idents’ place of various location factors exhibits significant
variability. Third, the four development types (i.e., rural
lots, country subdivisions, horticultural subdivisions, and
remnant subdivisions) were significantly different on
dimensions that were relevant to our conceptual model,
i.e., their biophysical landscape effects, and their locations
were consistent with many of their hypothesized locational
characteristics.
An important challenge that an agent-based view of sys-
tems creates for empirical data collection is in the identifi-
cation and quantification of interactions among agents.
Our empirical work uncovered social, spatial and cross-
scale interactions that could prove significant for the func-
tioning of our agent-based models. First, some residents
indicated a great deal of importance of factors related to
social comfort or similarity in choosing where to live
(e.g., familiarity with the area, proximity to family and
friends). Because of this empirical result, we subsequently
included a social similarity factor into the SOME model.
Second, our web-survey of landscape preferences revealed
a strong spatial interaction, such that the preferences of
individuals for the look of their own yards were signifi-
cantly affected by the look of neighbors’ yards. We have
not yet incorporated this interaction into the DEED
model, but plan to do so. Third, we hypothesize that sim-
ilar spatial effects on landscape design will also be apparent
at the subdivision scale, but have not yet completed data
collection to test this. Finally, we identified ways in which
different types of agents interact, e.g., through effects of
subdivision types on price signals to residents.
Another empirical challenge raised by complexity sci-
ence is in the quantification of feedbacks. Some of the feed-
backs are built into the specific interactions represented in
our models, for example in the way in which landscapes
affect residential choices and residential choices affect land-
scapes. We also analyzed the spatial pattern of the road
network to evaluate the possible importance of a missing
feedback between the road network and land use patterns.
After 1960, we observed large increases in residential
streets and slight increases in major roads. While major
roads clearly played a large role in establishing the pattern
of development, this finding gives us some confidence that,
though road capacities have surely changed, their spatial
patterns were relatively well established by 1960. Note that
816 D.G. Brown et al. / Geoforum 39 (2008) 805–818
this finding is likely to be regionally and temporally
specific.
In addition to assisting the validation of our conceptual
models, we used the empirical data collection to provide
calibration of our models. This form of calibration differs
from the more common approach of tuning the value of
a parameter until the macro-scale patterns match observed
patterns. We aim to calibrate the micro-level processes
independently, to ensure that the process structure of our
models most accurately represented the structural mecha-
nisms in the target system. The surveys provide some cali-
bration of the relative importance of various factors in the
residential location. The survival analysis contributed to
the calibration of factors that influence the location of dif-
ferent types of residential developments. The analysis of
land-cover within land-use types from remote sensing pro-
vides calibration for the landscape effects (both quantity
and location/pattern) of various land-use conversions.
Empirically calibrating ABMs may facilitate the extension
of ABM applications over relatively large geographic
regions. ABMs are known to be data demanding, especially
when used in real applications. Multi-method and creative
approaches are needed to collect the data needed to cali-
brate these models.
Acknowledgements
We gratefully acknowledge financial support in the form
of Grants from the National Science Foundation Biocom-
plexity in the Environment Program (BCS-0119804) and
the Office of the Vice President for Research at the Univer-
sity of Michigan.
References
Alberti, M., 2005. The effects of urban patterns on ecosystem function.
International Regional Science Review 28 (2), 168–192.
An, L., Brown, D.G., Nassauer, J.I., Low, B. submitted for publication.
Timing, location and determinants of residential development types in
exurban Southeast Michigan.
Arthur, W.B., 1994. Inductive reasoning and bounded rationality. The
American Economic Review 84 (2), 406–411.
Axelrod, R., Cohen, M.D., 2000. Harnessing Complexity: Organizational
Implications of a Scienctific Frontier. Free Press, New York.
Batty, M., Xie, Y., Sun, Z., 1999. Modeling urban dynamics through GIS-
based cellular automata. Computers, Environment and Urban Systems
23 (3), 205–233.
Boarnet, M.G., Chalermpong, S., 2001. New highways, house prices, and
urban development: the case of toll roads in Orange County, CA.
Housing Policy Debate 12 (3), 575–605.
Briassoulis, H., 2000. Analysis of land use change: Theoretical and
modeling approaches. Regional Research Institute, West Virginia
University, Morgantown, WV <http://www.rri.wvu.edu/WebBook/
Briassoulis/contents.htm>.
Brown, D.G., 2005. Agent-based models. In: Geist, H. (Ed.), Our Earth’s
Changing Land: An Encyclopedia of Land-Use and Land-Cover
Change. Greenwood Press, Portsmouth, NH.
Brown, D.G., Robinson, D.T., 2006. Effects of heterogeneity in residential
preferences on an agent-based model of urban sprawl. Ecology
and Society 11 (1), 46 <http://www.ecologyandsociety.org/vol11/
iss1/art46>.
Brown, D.G., Pijanowski, B.C., Duh, J.-D., 2000. Modeling the relation-
ships between land-use and land-cover on private lands in the Upper
Midwest, USA. Journal of Environmental Management 59, 247–
263.
Brown, D.G., Page, S.E., Riolo, R.L., Rand, W., 2004. Agent based and
analytical modeling to evaluate the effectiveness of greenbelts. Envi-
ronmental Modelling and Software 19 (12), 1097–1109.
Brown, D.G., Johnson, K.M., Loveland, T.R., Theobald, D.M., 2005a.
Rural land use change in the conterminous US, 1950–2000. Ecological
Applications 15 (6), 1851–1863.
Brown, D.G., Page, S.E., Riolo, R., Zellner, M., Rand, W., 2005b. Path
dependence and the validation of agent-based spatial models of land
use. International Journal of Geographical Information Science 19 (2),
153–174.
Brown, D.G., Riolo, R.L., Robinson, D., North, M., Rand, W., 2005c.
Spatial process and data models: Toward integration of agent-based
models and GIS. Journal of Geographical Systems 7 (1), 1–23.
Chiang, A.C., 1984. Fundamental Methods of Mathematical Economics.
Mcgraw-Hill, New York, pp. 788.
Daniel, T.C., 2001. Whither scenic beauty? Visual landscape quality
assessment in the 21th century. Landscape and Urban Planning 54 (1–
4), 267–281.
Deadman, P.J., Robinson, D.T., Moran, E., Brondizio, E., 2004. Effects of
colonist household structure on land use change in the Amazon
Rainforest: an agent based simulation approach. Environment and
Planning B: Planning and Design 31, 693–709.
Evans, T.P., Sun, W., Kelley, H., 2006. Spatially explicit experiments for
the exploration of land use decision-making dynamics. International
Journal of Geographical Information Science 20 (9), 1013–1037.
Fernandez, L., Brown, D.G., Marans, R., Nassauer, J., 2005. Character-
izing location preferences in an exurban population: implications for
agent based modeling. Environment and Planning B 32 (6), 799–
820.
US FHWA, 1989. FHWA Functional Classification Guidelines. US
Federal Highway Administration, Washington, DC <http://
www.fhwa.dot.gov/planning/fcsec1_1.htm>.
Forman, R.T.T., Sperling, D., Bissonette, J.A., Clevenger, A.P., Cutshall,
C.D., Dale, V.H., Fahrig, L., France, R., Goldman, C.R., Heanue, K.,
Jones, J.A., Swanson, F.J., Turrentine, T., Winter, T.C., 2003. Road
Ecology: Science and Solutions. Island Press, Washington, DC.
Geoghegan, J., Wainger, L.A., Bockstael, N.E., 1997. Spatial landscape
indices in a hedonic framework: an ecological economics analysis using
GIS. Ecological Economics 23 (3–5), 251–264.
Grimm, V., Revilla, E., Berger, U., Jeltsch, F., Mooij, W.M., Railsback,
S.F., Thulke, H., Weiner, J., Wiegand, T., DeAngelis, D.L., 2006.
Pattern-oriented modeling of agent-based complex systems: Lessons
from Ecology. Science 310 (5750), 987–991.
Groffman, P.M., Bain, D.J., Band, L.E., Belt, K.T., Brush, G.S., Grove,
J.M., Pouyat, R.V., Yesilonis, I.C., Zipperer, W.C., 2003. Down by the
riverside: urban riparian ecology. Frontiers in Ecology and Environ-
ment 1 (6), 315–321.
Hansen, A.J., Knight, R., Marzluff, J., Powell, S., Brown, K., Hernandez,
P., Jones, K., 2005. Effects of exurban development on biodiversity:
patterns, mechanisms, research needs. Ecological Applications 15 (6),
1893–1905.
Holland, J.H., 1995. Hidden Order: How Adaptation Builds Complexity.
Massachusetts, Perseus Books, Cambridge.
Irwin, E.G., Bockstael, N.E., 2004. Land use externalities, growth
management policies, and urban sprawl. Regional Science and Urban
Economics 34 (6), 705–725.
Janssen, M.A., Ostrom, E., 2007. Empirically based agent-based model-
ling. Ecology and Society 11 (2), 37.
Lempert, R., Popper, S., Bankes, S., 2002. Confronting surprise. Source
Social Science Computer Review 20 (4), 420–440.
Manson, S.M., 2001. Simplifying complexity: a review of complexity
theory. Geoforum 32, 405–414.
Manson, S.M., O’Sullivan, D., 2006. Complexity theory in the study of
space and place. Environment and Planning A 38, 677–692.
D.G. Brown et al. / Geoforum 39 (2008) 805–818 817
Marans, R.W., 2003. Understanding environmental quality through
quality of life studies: the 2001 DAS and its use of subjective and
objective indicators. Landscape and Urban Planning 65 (1–2), 75–85.
McGarigal, K., Marks, B.J., 1995. FRAGSTATS: Spatial Pattern
Analysis Program for Quantifying Landscape Structure. General
Technical Report PNW-GTR-351, USDA Forest Service, Pacific
Northwest Research Station, Portland, OR.
Nassauer, J., 1995. Culture and changing landscape structure. Landscape
Ecology 10 (4), 229–237.
Nassauer, J., Dayrell, E., Wang, Z. in preparation. What will the
neighbors think? Cultural norms for ecological design.
O’Sullivan, D., 2004. Complexity science and human geography. Trans-
actions of the Institute of British Geographers, NS 29, 282–295.
Page, S.E., 2005. Agent-based models. In: Newman, P., Eatwell, J.,
Millgate, M. (Eds.), The New Palgrave Dictionary of Economics.
Palgrave MacMillan, New York.
Parker, D., Meretsky, V., 2004. Measuring pattern outcomes in an agent-
based model of edge-effect externalities using spatial metrics. Agricul-
tural Ecosystems and Environment 101 (2–3), 233–250.
Parker, D.C., Manson, S.M., Janssen, M.A., Hoffmann, M.J., 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, 314–337.
Pickett, S.T.A., Cadenasso, M.L., Grove, J.M., Nilon, C.H., Pouyat,
R.V., Zipperer, W.C., Costanza, R., 2001. Urban ecological systems:
Linking terrestrial ecological, physical, and socioeconomic compo-
nents of metropolitan areas. Annual Review of Ecology and System-
atics 32, 127–157.
Polhill, J.G., Gotts, N.M., Law, A.N.R., 2001. Imitative versus non-
imitative strategies in a land use simulation. Cybernetics and Systems
32 (1–2), 285–307.
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: Macal, C., Sallach, D. (Eds.), Pro-
ceedings, Agent 2002: Social Agents: Ecology, Exchange, and
Evolution. Argonne National Laboratory, Chicago, IL, pp. 149–
161.
Rand, W., Brown, D.G., Page, S.E., Riolo, R., Fernandez, L.E., Zellner,
M. 2003. Statistical validation of spatial patterns in agent-based
models. In: Proceedings, Agent Based Simulation, Montpellier,
France, April 2003.
Robinson, D.T., Brown, D.G., Parker, D.C., Schreinemachers, P.,
Janssen, M.A., Huigen, M., Wittmer, H., Gotts, N., Promburom, P.,
Irwin, E., Berger, T., Gatzweiler, F., Barnaud, C., 2007. Comparison
of empirical methods for building agent-based models in land use
science. Journal of Land Use Science, in press.
Land Atlas and Plat Books., 1956–1999. Rockford Map Publishers, Inc.,
Rockford, IL, USA.
US Bureau of the Census, 2001. Census of Population and Housing:
Summary File 1 United States. Bureau of the Census, US Washington,
DC.
US Department of Agriculture, 2002. Census of Agriculture, Geographic
Area Series. In: CD Cen V1 (6-04). USDA, vol. 1. National
Agricultural Statistics Service, Washington, DC.
Waldrop, M.M., 1992. Complexity: The Emerging Science at the Edge of
Order and Chaos. Simon and Schuster, New York.
Wegener, M., 2005. Urban land-use transportation models. In: Maguire,
D.J., Batty, M., Goodchild, M.F. (Eds.), GIS, Spatial Analysis, and
Modeling. ESRI Press, Redlands, CA, pp. 203–220.
Zellner, M.L., Page, S.E., Rand, W., Brown, D.G., Robinson, D.,
Nassauer, J., Low, B., in review. The emergence of zoning policy
games in exurban jurisdictions. Land Use Policy.
818 D.G. Brown et al. / Geoforum 39 (2008) 805–818

File (1)

Content uploaded by Derek T. Robinson
Author content
... Developments at one scale are contingent on linked developments at other scales (Chapman, 2004). For example, cities are linked economically not only to their regional hinterlands , but into the global economic 'ecosystem' (Brown et al., 2008). Lastly, cities can be resilient, although history demonstrates that this is not necessarily the case (Chelleri, 2012). ...
... This study also demonstrates how one integrative tool, the agent-based model, provides a framework to simulate geomorphological change. Although ABMs are already common in studies of social–ecological systems (e.g., Brown et al., 2007; An et al., 2014), geomorphologists have only begun to explore their use. Challenges remain to gather sufficient data for both the physical and human systems – especially in quantifying human decision-making and behavior – to fully inform application of these models to adequately capture the dynamics of human–landscape systems. ...
Article
Full-text available
As human interactions with Earth systems continue to intensify, understanding the complex relationships among human activity, landscape change, and societal responses to those changes becomes increasingly important. Interdisciplinary research centered on the theme of “feedbacks” in human-landscape systems serves as a promising focus for unraveling these interactions. This paper examines the specific case of the 2012 Waldo Canyon Fire of Colorado, where human responses after the fire to perceived threats of hydro-geomorphological hazards included construction of tall fences at the base of a burned watershed. These actions prompted feedbacks that promoted further landscape change that ultimately increased those hazards, rather than dampening the hydro-geomorphological effects of fire. Geomorphic analysis showed that the fences trapped particles that would naturally move through the system by flows with recurrence intervals greater than 3.3 years. With the particles blocked by the fences, the channel downstream became erosive, because it was devoid of large particles that produce substantial hydraulic resistance. Channel incision prompted a second human response to pave the eroding channel, which led to further incision downstream. This cycle of positive feedbacks between human decision-making and landscape change eventually led to a complete channelization of the stream channel downstream of the fences. The explanation for the transformation of the post-fire landscape therefore lies in the interacting human impacts and feedbacks, rather than the expected post-fire hydro-geomorphological adjustments. An initial agent-based model, capable of integrating social and hydro-geomorphological data, simulates these interacting impacts and feedbacks. Further refinement with more complete data input, especially pertaining to human decision making at individual or local levels, is required to fully demonstrate the utility and promise of this tool for application to geomorphic analysis.
... Of these five methods, sample survey is the most quantitative approach and can be carried out at a relatively large spatial scale [13]. Though qualitative information about agents' interactions and feedbacks could be obtained using different methods, it is still challenging to empirically identify and quantify social, spatial and cross-scale interactions among agents and feedbacks between agents' decisions and the biophysical processes [35]. ...
Article
Full-text available
We developed an agent-based model (ABM) to simulate farmers’ decisions on crop type and fertilizer application in response to commodity and biofuel crop prices. Farm profit maximization constrained by farmers’ profit expectations for land committed to biofuel crop production was used as the decision rule. Empirical parameters characterizing farmers’ profit expectations were derived from an agricultural landowners and operators survey and integrated in the ABM. The integration of crop production cost models and the survey information in the ABM is critical to producing simulations that can provide realistic insights into agricultural land use planning and policy making. Model simulations were run with historical market prices and alternative market scenarios for corn price, soybean to corn price ratio, switchgrass price, and switchgrass to corn stover ratio. The results of the comparison between simulated cropland percentage and crop rotations with satellite-based land cover data suggest that farmers may be underestimating the effects that continuous corn production has on yields. The simulation results for alternative market scenarios based on a survey of agricultural land owners and operators in the Clear Creek Watershed in eastern Iowa show that farmers see cellulosic biofuel feedstock production in the form of perennial grasses or corn stover as a more risky enterprise than their current crop production systems, likely because of market and production risks and lock in effects. As a result farmers do not follow a simple farm-profit maximization rule.
... This study also demonstrates how one integrative tool, the agent-based model, provides a framework to simulate geomorphological change. Although ABMs are already common in studies of social–ecological systems (e.g., Brown et al., 2007; An et al., 2014), geomorphologists have only begun to explore their use. Challenges remain to gather sufficient data for both the physical and human systems – especially in quantifying human decision-making and behavior – to fully inform application of these models to adequately capture the dynamics of human–landscape systems. ...
... The model performed well in comparison with an analytical model of the influence of a greenbelt on land-development patterns but was much more easily extended to two dimensions: a heterogeneous environment and heterogeneous agents (Brown et al., 2004). An additional model was developed using the RePast software libraries in Java, in order to add agent types to represent developers and township policy-making bodies, and resident decisions about land cover (Brown et al., 2008). This model relies on the identification of three different types of subdivisions that differ in terms of both characteristics that residents evaluate (cost and aesthetic quality) and the landscape patterns they produce. ...
Article
Full-text available
Rural migration is an integral component of land systems, altering land management both at the origin and destination of migration in regions of Europe and North America. We examine population moving in and out of rural areas in Ohio, U.S.A, between 2008 and 2016 and examine spatiotemporal patterns in land cover. Among families changing residential location, one-third of all household moving events are migration from or to rural areas. A PCA ordination and cluster analysis of rural migration revealed qualitatively distinct patterns. Urban peripheries grew due to in-migration from metro suburban areas, while agricultural and forested areas reflect intra-rural and micropolitan migration. Forested communities have lost population to other rural areas, and rural poverty pockets have gained low-income families from both rural and metro urban areas. Spatial analysis of rural migration patterns and associations with land use helps develop targeted land management strategies and reveals under-represented dynamics in prior studies.
Chapter
Full-text available
Amenity migration and associated exurbanization are changing the ways that rural areas are viewed, used, and managed, shifting the landscapes and social norms of places along the rural–urban interface. As such, exurbia, that particular form of growth or development that occupies the edges and borders between rural and urban spaces and places, is significant not just for its spatial patterns, but also for its characteristic social, cultural, and political dynamics. The functional shifts to land and society occurring in exurbia make understanding environmental management regimes both challenging and critical as the implications for land-use change in the exurban context are multiple and profound. In an effort to guide and frame these ongoing analyses, I develop and apply the concept of contested ecologies, wherein disagreements over the meaning, values, and/or function of land or resources lead to contrasting perspectives of a particular place, environment, or landscape. In this paper, I present three cases of land-use change in Calaveras County, California, and their contested ecologies, focusing, in particular, on the policy and political dimensions of the case of the Trinitas golf course. The cases serve to demonstrate the ecological consequences of various land-use debates. I conclude with an evaluation of the significance of divergent environmental and ideological perspectives for land-use decision-making and environmental management in the context of contested ecologies along the rural–urban edge.
Article
Full-text available
Intraurban migration—residential movement within a metropolitan area—defines the nature of urbanization. Housing location decision making is a complex process driven by the interactions between the housing market and home searchers. Researchers have paid much attention to the environmental, socioeconomic, cultural, and policy features of housing markets. In contrast, housing search has been relatively neglected due to challenges of theory, methodology, and data. This article addresses these challenges by presenting an agent-based model of intraurban migration featuring straightforward and empirically specified rules for housing search. This model is calibrated and validated against real-world housing vacancies and relocation origin–destination pairs extracted from parcel records for the Twin Cities of Minnesota, USA, for 2005–2007. Drawing on these unique data sidesteps a long-standing issue, the prohibitive costs of identifying, recording, and quantifying housing search activities for an entire metropolitan region. Conceptually, this model updates geographic theories of intraurban migration that focus on intervening opportunities and spatial bias. It also methodologically advances the agent-based modeling of urbanization with a high-resolution, empirically specified model that demonstrates how urban pattern emerges from simple rules and interactions. Overall, the model demonstrates that relatively straightforward housing search rules can simulate realistic patterns of intraurban migration.
Article
Full-text available
Reactive nitrogen loss from agricultural fertilizer use remains a crucial environmental problem in the United States, contributing to ecosystem degradation and global climate change. This intractable problem requires a coupled human and natural systems approach that combines biophysical, sociological, and economic knowledge into an integrative analysis. Much is known about the biogeochemistry of nitrogen and agricultural nitrogen loss; however, much is not known about how soil variability and climate change will affect farmer decisionmaking. Although it is widely understood that personal values and beliefs, social norms, economics, and policies influence farmer decisionmaking, very little is known about decisionmaking specific to fertilizer management. In addition, little is known about the socioeconomic influences on decisionmaking across scales and how ecological change is perceived and responded to. Combining sociological, economic, and biophysical knowledge can provide key insights regarding how these factors interact and can support more effective strategies to address this persistent problem.
Article
Full-text available
Surprise takes many forms, all tending to disrupt plans and planning systems. Reliance by decision makers on formal analytic methodologies can increase susceptibility to surprise as such methods commonly use available information to develop single-point forecasts or probability distributions of future events. In doing so, traditional analyses divert attention from information potentially important to understanding and planning for effects of surprise. The authors propose employing computer-assisted reasoning methods in conjunction with simulation models to create large ensembles of plausible future scenarios. This framework supports a robust adaptive planning (RAP) approach to reasoning under the conditions of complexity and deep uncertainty that normally defeat analytic approaches. The authors demonstrate, using the example of planning for long-term global sustainability, how RAP methods may offer greater insight into the vulnerabilities inherent in several types of surprises and enhance decision makers’ ability to construct strategies that will mitigate or minimize the effects of surprise.
Article
Full-text available
Riparian areas are hotspots of interactions between plants, soil, water, microbes, and people. While urban land use change has been shown to have dramatic effects on watershed hydrology, there has been surpris- ingly little analysis of its effects on riparian areas. Here we examine the ecology of urban riparian zones, focusing on work done in the Baltimore Ecosystem Study, a component of the US National Science Foundation's Long Term Ecological Research network. Research in the Baltimore study has addressed how changes in hydrology associated with urbanization create riparian "hydrologic drought" by lowering water tables, which in turn alters soil, vegetation, and microbial processes. We analyze the nature of past and cur- rent human interactions with riparian ecosystems, and review other urban ecosystem studies to show how our observations mirror those in other cities.
Article
Full-text available
This article presents an overview of multi-agent system models of land-use/cover change (MAS/LUCC models). This special class of LUCC models combines a cellular landscape model with agent-based representations of decision making, integrating the two components through specification of interdependencies and feedbacks between agents and their environment. The authors review alternative LUCC modeling techniques and discuss the ways in which MAS/LUCC models may overcome some important limitations of existing techniques. We briefly review ongoing MAS/LUCC modeling efforts in four research areas. We discuss the potential strengths of MAS/LUCC models and suggest that these strengths guide researchers in assessing the appropriate choice of model for their particular research question. We find that MAS/LUCC models are particularly well suited for representing complex spatial interactions under heterogeneous conditions and for modeling decentralized, autonomous decision making. We discuss a range of possible roles for MAS/LUCC models, from abstract models designed to derive stylized hypotheses to empirically detailed simulation models appropriate for scenario and policy analysis. We also discuss the challenge of validation and verification for MAS/LUCC models. Finally, we outline important challenges and open research questions in this new field. We conclude that, while significant challenges exist, these models offer a promising new tool for researchers whose goal is to create fine-scale models of LUCC phenomena that focus on human-environment interactions.
Article
This report describes a program, FRAGSTATS, developed to quantify landscape structure. Two separate versions of FRAGSTATS exist: one for vector images and one for raster images. In this report, each metric calculated by GRAGSTATS is described in terms of its ecological application and limitations. Example landscapes are included, and a discussion is provided of each metric as it relates to the sample landscapes. -from Authors
Article
In order to understand the magnitude, direction, and geographic distribution of land-use changes, we evaluated land-use trends in U.S. counties during the latter half of the 20th century. Our paper synthesizes the dominant spatial and temporal trends in population, agriculture, and urbanized land uses, using a variety of data sources and an ecoregion classification as a frame of reference. A combination of increasing attractiveness of nonmetropolitan areas in the period 1970-2000, decreasing household size, and de- creasing density of settlement has resulted in important trends in the patterns of developed land. By 2000, the area of low-density, exurban development beyond the urban fringe occupied nearly 15 times the area of higher density urbanized development. Efficiency gains, mechanization, and agglomeration of agricultural concerns has resulted in data that show cropland area to be stable throughout the Corn Belt and parts of the West between 1950 and 2000, but decreasing by about 22% east of the Mississippi River. We use a regional case study of the Mid-Atlantic and Southeastern regions to focus in more detail on the land-cover changes resulting from these dynamics. Dominating were land-cover changes associated with the timber practices in the forested plains ecoregions and urban- ization in the piedmont ecoregions. Appalachian ecoregions show the slowest rates of land- cover change. The dominant trends of tremendous exurban growth, throughout the United States, and conversion and abandonment of agricultural lands, especially in the eastern United States, have important implications because they affect large areas of the country, the functioning of ecological systems, and the potential for restoration.
Article
Urban ecological systems are characterized by complex interactions among social, economic, institutional, and environmental variables. These interactions generate complex human-dominated landscapes, which significantly influence the functioning of local and global earth ecosystems and the services they provide to humans and other life on earth. Urban development fragments, isolates, and degrades natural habitats; simplifies and homogenizes species composition; disrupts hydrological systems; and modifies energy flow and nutrient cycling. Urban areas also appropriate a large share of earth’s carrying capacity from other regions in terms of resource input and waste sinks. Change in ecological conditions that result from human actions in urban areas ultimately affect human health and well-being. In this article, the author reviews the empirical evidence on the effects that patterns of urban development have on ecosystem function. Urban development affects the spatial heterogeneity of the landscape (i.e., pattern of variation in land cover) and spread of disturbance (i.e., invasive species). The author proposes that alternative urban patterns generate differential ecological effects. The review reveals that the interactions between urban development patterns and ecosystem dynamics are still poorly understood. The author draws on an empirical study of the Puget Sound metropolitan region currently developed at the University of Washington to propose directions for future empirical research that can inform strategies to minimize urban impacts on ecosystems.
Article
We examine the link between highways and urban development by employing both hedonic analysis and multiple sales techniques to study the impact of the construction of toll roads in Orange County, California, on house prices. Urban economic theory predicts that if highways improve accessibility, that accessibility premium will be reflected in higher land prices. Our empirical analyses of house sales prices provide strong evidence that the toll roads, the Foothill Transportation Corridor Backbone in particular, created an accessibility premium; home buyers are willing to pay for the increased access that the new roads provide. Such willingness to pay influences both development patterns and, potentially, induced travel (the association between increases in highway capacity and increases in vehicle miles of travel). The results are consistent with the idea that induced travel is caused, in part, by changes in urban development patterns that are linked to increases in highway capacity.
Article
Low-density rural home development is the fastest-growing form of land use in the United States since 1950. This ''exurban'' development (6–25 homes/km 2) includes urban fringe development (UFD) on the periphery of cities and rural residential development (RRD) in rural areas attractive in natural amenities. This paper synthesizes current knowl-edge on the effects of UFD and RRD. We present two case studies and examine the patterns of biodiversity response and the ecological mechanisms that may underlie these responses. We found that many native species have reduced survival and reproduction near homes, and native species richness often drops with increased exurban densities. Exotic species, some human-adapted native species, and species from early successional stages often in-crease with exurban development. These relationships are sometimes nonlinear, with sharp thresholds in biodiversity response. These effects may be manifest for several decades following exurban development, so that biodiversity is likely still responding to the wave of exurban expansion that has occurred since 1950. The location of exurban development is often nonrandom relative to biodiversity because both are influenced by biophysical factors. Consequently, the effects on biodiversity may be disproportionately large relative to the area of exurban development. RRD is more likely than UFD to occur near public lands; hence it may have a larger influence on nature reserves and wilderness species. The ecological mechanisms that may underlie these responses involve alteration of habitat, ecological processes, biotic interactions, and increased human disturbance. Research on the patterns and mechanisms of biodiversity remains underdeveloped, and comparative and experimental studies are needed. Knowledge resulting from such studies will increase our ability to understand, manage, and mitigate negative impacts on biodiversity.
Article
Complexity science has attracted considerable attention in a number of disciplines. However, this perspective on scientific understanding remains ill defined. In this paper, ideas and approaches from complexity science are reviewed. It appears that complexity science fundamentally is driven by ontological decisions on the part of the investigator. This is a result of the epistemological approach fundamental to complexity as it is currently studied, which is based on the construction of computer simulation models of reality. This methodology requires that researchers decide what exists and is important enough to represent in a simulation, and also what to leave out. Although this points to serious difficulties with complexity science, it is argued that the approach nevertheless has much to offer human geography. Drawing on complexity science, renewed engagements between physical and human geography, and between both and geographical information science seem possible, based on clearly shared concerns with the representation of geographical phenomena. In conclusion, it is suggested that seeing models as a source of geographical narratives may be a useful way to promote constructive engagement between different perspectives in the discipline.
Article
This paper analyzes quantification error versus location error in a comparison between two cellular maps that show a categorical variable. Quantification error occurs when the quantity of cells of a particular category in one map is different from the quantity of cells of that category in the other map. Location error occurs when the location of a category in one map is different from the location of that category in the other map. The standard Kappa index of agreement is usually not appropriate for map comparison. This paper offers alternative statistics: (1) proportion correct with perfect ability to specify location, (2) proportion correct with perfect ability to specify quantity, (3) Kappa for no ability, (4) Kappa for location, and (5) Kappa for quantity. These statistics can help scientists improve classification. This paper applies these theoretical concepts to the validation of a land-use change model for the Ipswich Watershed in Massachusetts.