5i. .«0

fl

 

     

 

    

      

T ._ ..:

£1.35. :3 ‘ z “

     

.-r&§..fi_.ifi.£.. u m
.

 

£2 ,.

.1.

 

; *5. ($..
‘ s'lkb'n

()
1‘ " \ -
2| .

52/891630

 

 

LIBRARY
Michigan State
University

 

 

 

This is to certify that the
thesis entitled

VALIDATION OF THE LAND TRANSFORMATION
MODEL (LTM)
TO PREDICT LAND USE CHANGES IN GRAND BLANC
TOWNSHIP, MICHIGAN

presented by

Matthew H. Malone

has been accepted towards fulfillment
of the requirements for the

 

 

MS. degree in 4 Resource Development

    

 

Major Erofessor's Signature

53/9/ flap 2
Date

MSU Is an AMnnaflva Action/Equal Opportunity Institution

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJClRC/DaleDuopBS—p. 1 5

VALIDATION OF THE LAND TRANSFORMATION MODEL (LTM)
TO PREDICT LAND USE CHANGES IN GRAND BLANC TOWNSHIP, MICHIGAN

By

Matthew H. Malone

A THESIS

Submitted to
Michigan State University
in partial fitlfillment of the requirements
for the degree of

MASTER OF SCIENCE
Department of Resource Development

2003

ABSTRACT

VALIDATION OF THE LAND TRANSFORMATION MODEL (LTM)
TO PREDICT LAND USE CHANGES IN GRAND BLANC TOWNSHIP, MICHIGAN

By

Matthew H. Malone

The research examined the predicted patterns of urbanization in the Charter
Township of Grand Blanc, Michigan using the Land Transformation Model (LTM). The
LTM modeling engine utilizes an artificial neural network to predict urbanization trends
through the input of land use data and “driving” variables shown to influence land use
change. Land use data from 1978 and 2000 along with six driving variables were input to
predict urbanization patterns for the year 2000 and future urbanization patterns at
intervals ranging from five to one-hundred percent. These predicted urbanization
patterns were compared with current land use data to evaluate the predictive accuracy of
the LTM via comparisons of predicted versus actual areas transitioning to urban uses.
Township officials were then interviewed to solicit their opinions and reactions regarding
the accuracy of the predictions of 2000 urbanization and future urbanization patterns

depicted at increasing percentages.

ACKNOWLEDGMENTS

I wish to express sincere thanks to my committee members, Dr. Gerhardus
Schultink, Dr. Stuart Gage, and Dr. Mike Thomas for their valuable insight and guidance
over the past several years. Thanks also to Dr. Bryan Pij anowski and Snehal Pithadia for
their generous assistance in producing the model results.

Finally, I’d like to thank all my loved ones, family and friends for their
unwavering support and encouragement over the last five years. They believed in my
ability to overcome the obstacles that life presented me, and without them this thesis

would not have been possible.

iii

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................. vi
LIST OF FIGURES .......................................................................................................... vii
CHAPTER I
INTRODUCTION AND PROBLEM ANALYSIS ............................................................. 1
1.1 Urban Sprawl in Southeast Michigan and Smart Growth ................................. 1
1.2 Land Use Forecasting Tools, Planning Support Systems and
Geographic Information Systems ..................................................................... 4
1.3 Perception and Interpretation of Land Use Forecasting Tools ......................... 6
1.4 Problem Statement ............................................................................................ 6
1.5 Research Questions ........................................................................................... 7
1.6 Study Significance and Target Audience .......................................................... 8
CHAPTER II
LITERATURE REVIEW .................................................................................................. 11
2.1 Overview of Planning Support Systems and
Geographic Information Systems ................................................................... 11
2.2 The Land Transformation Model .................................................................... 13
2.3 Map Interpretation and Perception of Accuracy ............................................. 15
2.4 Prevalence of GIS Data and Usage in Local Governments ............................ 19
CHAPTER HI
RESEARCH DESIGN AND METHODS ......................................................................... 21
3.1 Study Location ................................................................................................ 21
3.2 Data Compilation and Geoprocessing Strategy .............................................. 22
3.3 LTM 1978 to 2000 Urban Transition Run ...................................................... 28
3.4 LTM Predictive Runs with Five Percent Increases in Urbanization ............... 30
3.5 Solicitation of Opinions on LTM Output Maps by Local Officials ................ 31
3.6 Methods Utilized for Assessing LTM Accuracy ............................................ 36
CHAPTER IV
RESEARCH RESULTS .................................................................................................... 34
4.1 LTM Validation - 1978-2000 Analysis ........................................................... 34
4.1.2 Relative Influence of Driving Variables on Overall Model Output .......... 37
4.2 Township Officials’ Opinions and Reactions to LTM Forecast Results ........ 42
4.2.2 Opinions on LTM 1978-2000 Urbanization Pattern .................................. 43
4.2.3 Opinions on LTM Extrapolation — Five Percent Projections ..................... 48
4.3 Conclusions and Recommendations ............................................................... 50
4.3.2 LTM 1978 — 2000 Validation .................................................................... 50
4.3.3 Township Officials’ Opinions and Reactions Analysis ............................. 53
4.4 Future Research and the LTM ........................................................................ 57

iv

REFERENCES .................................................................................................................. 59

APPENDICES ................................................................................................................... 64
APPENDIX A

MAP SERIES #2, FIVE PERCENT NCREMENT MAPS

FIVE THROUGH FIFTY PERCENT ............................................................................... 64
APPENDIX B

MAP SERIES #3, FIVE PERCENT INCREMENT MAPS

F IF TY-FIVE THROUGH ONE HUNDRED PERCENT ................................................. 75
APPENDIX C

FIVE PERCENT INCREMENT SPREADSHEET CALCULATIONS ........................... 86
APPENDIX D

UCRIHS ORAL CONSENT FORM ................................................................................. 88

LIST OF TABLES

Table

1. Driving variable raster grids utilized in the Land Transformation Model ............... 27

2. Existing 2000 land use categories and urban/non-urban reclassification ................ 3O

3. Year 2000 urbanization prediction results gn'd ........................................................ 34

4. Existing 2000 land use categories and the number of cells correctly predicted to
transition to that land use type. ................................................................................ 36

5. Summary table of opinions and confidence expressed regarding the three map
series utilized during interviews with Grand Blanc Township officials .................. 44

vi

Figure

10.
ll.
12.
l3.
14.
15.
l6.
17.
18

19.

LIST OF FIGURES

Grand Blanc Township study region in southeast Lower Michigan ........................ 22
Driving variables for the Grand Blane Township study region. .............................. 25
Driving variable study region surrounding Grand Blane Township ........................ 27
LTM predictions of 2000 urbanization pattern change ............................................ 29
Grand Blane Township goodness of fit metric versus window size ........................ 39
Grand Blane Township goodness of fit metric versus window size showing

relative influence of the six driving variables included in the LTM ........................ 40

Grand Blanc Township quadrants used to categorize respondents

opinions on LTM map series results. ....................................................................... 46
Five percent increase in conversion to urban. .......................................................... 65
Ten percent increase in conversion to urban ............................................................ 66
Fifteen percent increase in conversion to urban. ..................................................... 67
Twenty percent increase in conversion to urban ...................................................... 68
Twenty-five percent increase in conversion to urban. ............................................. 69
Thirty percent increase in conversion to urban. ....................................................... 70
Thirty-five percent increase in conversion to urban. ............................................... 71
Forty percent increase in conversion to urban. ........................................................ 72
Forty-five percent increase in conversion to urban .................................................. 73
Fifty percent increase in conversion to urban. ......................................................... 74
F ifiy—five percent increase in conversion to urban. ................................................. 76
Sixty percent increase in conversion to urban. ........................................................ 77

vii

20.

21.

22.

23.

24.

25.

26.

27.

Sixty—five percent increase in conversion to urban. ................................................. 78

Seventy percent increase in conversion to urban. .................................................... 79
Seventy-five percent increase in conversion to urban. ............................................ 80
Eighty percent increase in conversion to urban. ...................................................... 81
Eighty-five percent increase in conversion to urban ................................................ 82
Ninety percent increase in conversion to urban. ...................................................... 83
Ninety-five percent increase in conversion to urban. .............................................. 84
One hundred percent increase in conversion to urban. ............................................ 85

viii

I. INTRODUCTION AND PROBLEM ANALYSIS

1.1 URBAN SPRAWL IN SOUTHEAST MICHIGAN AND SMART GROWTH

Elected officials, planners and decision makers are constantly striving to make
efficient and environmentally sound land use development decisions that positively affect
their respective municipalities. The continuing spread of new suburban development '
continues across the United States and in Michigan as older central city cores deteriorate
and decay. While city cores struggle to survive, conversion of farmland in the suburbs
for other uses has increased dramatically. The Michigan Farm Bureau reports that more
than 200 acres of farmland has been lost each day over the past fifteen years (The Detroit
News, 1999). Additional research indicates that between 1990 and 2020 the amount of
developed land in Michigan will increase from 63% to 87%, while the population will
only increase by 11.8% (Rustem, 1997). The Michigan Land Resource Project (Public
Sector Consultants, 2001) also indicates that between 1980 and 2040 the amount of built
land in Michigan will increase by approximately 178% while an additional 17% of
agricultural lands are lost

The term often used to describe this rapid rate of development is “urban sprawl”,
which is defined as the “rapid conversion and scattering of urban land uses away from a
centralized urban core (Machemer et al., 1999).” The rapid rate of suburban development
across southeast Lower Michigan makes land use decisions for local governments an
increasingly important process. Many townships across this portion of the state are
experiencing dramatic growth pressures as new residents and commuters in the
metropolitan Detroit area populate subdivisions on the metropolitan fringe. The outward

migration {Tom the Detroit area to these fringe zones led to a population shifi between

1995 and 1998 totaling 58,508 people, or 5.7% of the population of metro Detroit (The
Detroit News, 1999). Southeast Lower Michigan’s population is projected to increase
through natural increase (births greater than deaths) by an additional 810,820 inhabitants
between 1990 and 2030, when the population will reach an estimated 5.4 million
residents (SEMCOG, 2001). The growth in population and migration in metro Detroit
has been linked to multiple factors, including: available land, jobs within commuting
distance; and increases in job growth in the suburban fi'inge (SEMCOG, 2001). If the
population projections are correct, then approximately 251,000 acres of open land,
equaling over 10 townships in size, will be converted to other uses (SEMCOG, 1995).

Growth in these rural townships extends northward from the Detroit metro area
through northern Oakland County and southern Genesee County. One recent study
revealed that Holly and Groveland Townships in northern Oakland County along the
border with Genesee County, could experience an increase in population of 46% and 72%
respectively by the year 2020 (SEMCOG, 1995). Neighboring Genesee County has also
felt the impact of residential expansion. One recent land use study (Rusk, 1999) showed
that Genesee County's population had increased fi'om 198,000 people on 45 square miles
of urbanized land in 1950, to 326,000 people on 164 square miles of urbanized land in
1990. The study also indicated that while Genesee County's population grew by 65%
during the 1950-1990 period, the total urbanized land area had increased
disproportionately by approximately 266%.

To deal with this rapid growth, elected officials and planners across the United
State have utilized several tools to assist them in the planning process. These tools are

often categorized under the umbrella term of "smart growth" techniques. As a basic

definition, smart growth techniques are employed by municipalities to encourage
efficient and environmentally sound development patterns within their municipal borders
(Lee et a1. 1998). Only a handful of these growth management programs are in place
throughout the state of Michigan, which has allowed Michigan to rank 49th out of the 50
states in a survey measuring implementation and effectiveness of smart grth
techniques (Sierra Club, 1999).

Smart growth techniques encompass numerous strategies to assist with growth
management including: urban grth boundaries in Portland, Oregon and Contra Costa
County, California (Staley et a1. 1999); open space preservation through transfer of
development rights initiatives in New Jersey (New Jersey Pinelands Commission, 2003),
and purchase of development rights programs in Boulder, Colorado (County of Boulder,
Colorado, 2003); regional planning in Grand Rapids, Michigan (Grand Valley
Metropolitan Council, 2003) and Denver, Colorado (City of Denver, Colorado, 2000); as
well as comprehensive programs and laws that are hybrids of several growth management
techniques (Lee et a1. 1998). Genesee County municipalities do not currently employ any
of the aforementioned growth management techniques. However, two examples of urban
growth boundaries are found in other regions of Michigan. The most notable cases are in
the cities of Frankenmuth and Midland, Michigan (Michigan Department of
Environmental Quality, 1999).

In the Frankenmuth, Michigan case, a local survey was conducted to gauge the
community’s interest in cooperative planning between the City of Frankenmuth and
adjacent Frankenmuth Township. The survey results indicated that fifty-percent of the

community residents wanted to see growth in the City restricted, and future development

to take place outside of prime agricultural lands (Hamlin et al. 2000). The survey
resulted in the creation of a cooperative urban growth boundary between the
municipalities that designated future residential and commercial development districts
over the next forty years.

The City of Midland, Michigan utilizes an urban growth boundary to dictate
growth based on the availability of public water supplies. Midland has historically
controlled growth on its borders through the use of a “No Annexation/No Water” policy
(Hamlin et a1. 2000). The City can employ this technique to control growth since its
public water supply originates in nearby Lake Huron and is more palatable when
compared to the brackish groundwater supply found in municipalities surrounding the
City. Midland’s city boundary has thus slowly crept outward via annexation of
surrounding townships where water infrastructure has been extended.

1.2 LAND USE FORECASTING TOOLS, PLANNING SUPPORT SYSTEMS AND
GEOGRAPHIC INFORMATION SYSTEMS

Many tools and techniques exist for planners to assist them in managing and
analyzing development within their municipalities. One of these tools is called a build-
out analysis, which calculates in a tabular or graphical form the maximum extent and
amount of development that may occur in a region based on current zoning regulations
and other land use policies (U .8. EPA, 2003; Washtenaw County, Michigan, 2002).
Additional techniques include land suitability analysis, which utilizes a Geographic
Information System (GIS) in order to display areas suitable for development based on
factors including slope, soil composition, elevation, and other physical characteristics
(Swanson, 2003). The emerging usage of GIS at all levels of government allow elected

officials, planners, and decision makers to add GIS and land-use modeling as tools to

assist in smart growth planning. The increasing availability of GIS data that can be made
available at the local level enables these officials to more frequently utilize GIS-based
land use models and software applications in their decision making process. Many land
use modeling algorithms stand alone from a particular GIS package, but many have been
incorporated and renamed either a Decision Support System (DSS) or a hybrid GIS
known as a Planning Support System, or PSS (Harris and Batty, 1993; Hopkins, 1999;
Klosterrnan, 1997). The DSS and PSS are additional tools in the growth management
palette that are used to implement land use and growth management policies. One
capability of many land use based PSS techniques incorporates the power to extrapolate
and predict potential land use development patterns or regions ripe for transformation to
other land uses, e.g., urban or commercial (Heikkila, 1998; Lee et a1. 1998). Determining
the multitude of outcome scenarios that could result from land use policy decisions had
traditionally been a complex, costly, and lengthy procedure before the advent of GIS and
PSS modeling.

Another advantage of using a PSS approach to simulate potential land use
scenarios revolves around the inherent ability of a GIS to display visual information in
the form of a map. The ability for a map to inform a layperson in a clear and accurate
fashion allows a PSS to become a powerful tool to convey information. Recent PSS
models also stretch beyond traditional map images and provide realistic three-
dimensional visualization of the local area including existing buildings and vegetation
(Brail and Klosterrnan, 2001). A PSS is also advantageous in its ability to expedite the
creation of outcome scenarios. This timely production allows the user and the general

public quick access to land use scenarios generated by altering variables in the PSS. An

example would be to alter a township's minimum acreage requirements for a particular
zoning class such as “residential", and then visualizing the potential impact that a change
in this requirement could have on development patterns.

1.3 PERCEPTION AND INTERPRETATION OF LAND USE FORECASTING
TOOLS

The increase in usage of PSS and land use forecasting tools makes it necessary for
one to analyze the perception and interpretation of these potential outcomes by local
officials and decision makers. The relative ease and availability of these models have
made it imperative to examine the influence that maps of potential urbanization patterns
have on the local officials. After all, these officials will ultimately decide the
development direction for their municipalities through creation and enforcement of local
ordinances pertaining to zoning and land development. Much of the power of these
models lie in their ability to allow leaders to see the potential effects of policies,
practices, ordinances, and plans prior to their passage and implementation. They also
incite decision makers to evaluate the possibilities of future land development and to pose
rational questions regarding the potential firtures presented. The decision makers are then
better equipped to potentially set into motion changes (e. g., zoning ordinances, local
laws) to make these desired futures a reality.

1.4 PROBLEM STATEMENT

The advantages of accurate, timely, and increasingly user-friendly PSS’s and land
use models has not been lost on the planning community. However, despite the powerful
capabilities and increased usage of these PSS’s by planners and local officials, there have
been few studies that examine the accuracy of the potential land use scenarios generated

by these systems (Klosterrnan, 199%). Even if datasets, sofiware, and methodologies

employed in the creation the predictions are of high quality, the public groups viewing
the output fi'om a PSS may accept the potential land use outcomes without much thought
to the accuracy of these predicted scenarios. Often, these scenarios are demonstrated
within the context of a collaborative group such as a group of citizens or officials
concerned about local development (Klosterrnan, 1999a). These group sessions are more
fi'equently including the demonstration of PSS software using local GIS data to provide
powerful "what if“ scenarios that may occur under present zoning ordinances and statues.
The enhanced two and three-dimensional visualization of these potential outcomes, albeit
only potential, may still be utilized by elected officials or planners to assist them in
updating a master plan or altering land use and zoning ordinances to realize these
scenarios. These alterations to local land use zoning laws can have enormous
implications for future growth and development in a community. If they are based solely
or even partially on a PSS that has not been evaluated for accuracy then they risk the
potential to derail the favorable land use outcomes they had hoped to initiate.
1.5 RESEARCH QUESTIONS

The research addressed several aspects of the predictive capability of the Land
Transformation Model (LTM), a neural-network based land use modeling program
created at Michigan State University (MSU) (Pijanowski et a1. 1997). The LTM operates
through the input of GIS datasets in order to calculate the probability of an area to
convert to urbanized land from other uses. In many ways, the LTM typifies the modeling
“engine” used to produce alternative land use scenarios found in many PSS software

packages.

Several research questions were examined regarding the LTM’s predictive
accuracy in producing future urbanization scenarios. First, how accurately would the
LTM predict patterns of urbanization in a township? For example, which of the LTM’s
“driving variables”, which consist of the GIS layers used in modeling these urbanization
patterns, appear to have the most impact on model accuracy at a large scale? Should
additional variables be considered on a township-scale project?

Secondly, how would the general public, local officials and decision makers react
and interpret the results presented from the LTM? Would they take the results at face-
value under the assumption that they're correct due to the complex computer techniques
involved? What are their opinions on the LTM’s predictions of urbanization for 2000
and beyond? Do they believe there are areas in the township that are better "handled" by
the predictive algorithms of the LTM based on their local knowledge of the township?
1.6 STUDY SIGNIFICANCE AND TARGET AUDIENCE

The pervasiveness of planning support systems will continue to grow as
additional municipalities begin to harness the power of GIS technology. With the advent
of PSS software, it is reasonable to assume that more local governments may find the
systems useful to run predictive scenarios and interact with the public on important land
use issues such as master plan or zoning ordinance updates. Planners and consulting
firms will also likely begin to utilize the PSS as an essential element in their planning
toolbox. Focus groups, concerned citizens, and environmental organizations may also
find it useful to purchase PSS software and obtain the necessary GIS datasets themselves.

Many GIS datasets are available either free of charge through local government Internet

sites or obtainable through the Freedom of Information Act, making this type of analysis
a distinct possibility.

With these groups in mind, the importance of this study revolves around the
accuracy of PSS software in general, and the LTM in particular, to predict potential land
use outcomes. Modeling engines like the LTM are only as accurate as the data entry and
modeling assumptions built into their mathematical algorithms. Planners and local
officials who begin to use PSS output on a more frequent basis as part of their community
planning processes should have a firm basis for evaluating the scenarios generated by
these models. The general public tends to have a blind faith towards computer models,
simulations, and their map output to the point where they do not question the model or
the assumptions used to generate these future scenarios. The oft-repeated computer-
related phrase "garbage in, garbage out" may not be familiar to them. Despite the fact
that these systems display only potential land use scenarios, these outcomes may be seen
as potential "truth" to local officials. The literature is replete with journal articles
describing different PSS and land use models, but most of them describe the models
rather than evaluate their accuracy or examine their impacts on their intended audiences
(Harris et al. 1993; Putnam et al. 2001; Waddell, 2001). In the past, many of these
models were considered far too expensive or computationally intensive to be
implemented with existing computer power and finances of a local government.

The LTM could prove to be one of many powerful predictive models in the
growing field of planning support systems. LTM model runs have been completed for
metropolitan regions both inside the United States (Minneapolis/ St. Paul, Minnesota and

Detroit, Michigan) as well as an international trial in Kuala Lumpur, Malaysia

(Pijanowski et a1. 2001). Hamilton County Ohio, which includes the Cincinnati area,
added a PSS known as What If?, with predictive land use algorithms similar to the LTM
to its planning toolbox in January 2000 (Klosterrnan, 1999c). If Cincinnati, the birthplace
of the comprehensive plan, trusts in the capabilities of a PSS, then we owe it to the GIS
and planning communities to begin examining the output of these models' advantages and
disadvantages in their predictive capability as well as their impact on local decision

makers.

10

II. LITERATURE REVIEW

2.] OVERVIEW OF PLANNING SUPPORT SYSTEMS AND GEOGRAPHIC
INFORMATION SYSTEMS

Geographic Information Systems (GIS) have become very commonplace in a
myriad of professional pursuits since their genesis in 1963 with the birth of the Canada
Geographic Information Systems, CGIS (University of Wisconsin, 1999). Since that
time, GIS technology has advanced through the use of remote sensing technology and
innovations in the design of GIS software applications (University of Wisconsin, 1999).
Several definitions of GIS have been stated in the literature. Many of these definitions
take what could be deemed a "software-only" approach in defining a GIS. As is the case
in Aronoff (1989) who states, "A GIS is designed for the collection, storage and analysis
of objects and phenomena where geographic location is an important characteristic or
critical to the analysis." Or the definition that a GIS is a system that integrates hardware,
software, databases, and analyses to solve geographic problems (Maquire et al. 1991),
allowing map data (e. g., land use, road networks, zoning districts) to be displayed as a
digital map “layer." Some practitioners however, expand the definition to include all
aspects of running and staffing a GIS, including the personnel who control the system
and the data that resides in the system (United States Geological Survey, 1999).

Major developments in the history of commercially available GIS packages
stemmed fiom the creation of powerful GIS software at the Environmental Systems
Research Institute (ESRI). ESRI was founded in 1969 as a private company specializing
in land use analysis projects (ESRI, 2003). The company developed its flagship GIS
package, Arc/Info, in 1981, and has since expanded their GIS products to include

Internet-based and mobile, handheld GIS products. Many PSS software packages are

11

either software extensions of existing ESRI products or were constructed using ESRI-
based development tools. The datasets created for this research were produced by the
LTM, but the model results and maps were created using ESRI’s ArcGIS software. As a
testimony to the widespread usage of ESRI software, the 2003 ESRI International User
Conference attracted over 11,000 users from several dozen countries (ESRI, 2003).

The merging of the GIS and planning disciplines has been in existence since the
first use of computers for planning desired travel routes in the Chicago Area
Transportation Study back in the 1950's (CATS, 2003). The progression of planning
support systems has evolved from simple spreadsheet calculations of population trends to
highly complex analysis tools to visualize and extrapolate potential land use outcomes
(Lee et al. 1998). The future scenarios that can be created using available GIS data can
assist planners and local officials in determining the appropriate land use policies for
their jurisdiction.

The works of numerous authors have bolstered the definition of a planning
support system. Harris and Batty (1993) went to great lengths to define the usage and
purposes of a PSS. With regards to a fully fimctional PSS they state:

“A major consideration of planning is the avoidance of unintended
consequence while pursuing intended goals. Both intended and unintended
consequences arise out of the propagation of eflects throughout the system for
which planning has been undertaken overtime, space, and function. In order to
assess these consequences planning needs methods for making conditional
predictions based on alternative hypothetical decisions. Both the research
establishing the capability to make such predictions and the mechanisms by which
scores of predictions can be made and examined, call for extensive computational
resources and sophisticated simulations modeling. ”

Klosterman (1997) mentions that a PSS ". . .includes only the computer hardware,

software, and related information that are used for planning, i.e., the information

12

technologies planners use to perform their unique professional responsibilities as
planners." And states further, "The heart of any PSS will undoubtedly be a GIS. The
GIS will serve first as a display and communication device producing maps and charts
that describe past and present conditions and model outputs that suggest alternative
futures (Klosterman, 1997)." Some authors suggest that in addition to the standard
points, lines, and polygons that are used in a GIS, a PSS should focus on elements of
urban development such as, "actors, activities, flows, investments, facilities, regulations,
rights, issues, forces, opportunities, and constraints (Hopkins, 1999).”
2.2 THE LAND TRANSFORMATION MODEL

Many of the aforementioned PSS and DSS software have at their heart a
processing engine to produce alternative land use scenarios based on policy inputs as well
as predictive models that use existing GIS data. The LTM program utilized in this study
typifies one of these types of modeling algorithms used to run predictive analyses and
trends in land use change. The LTM operates through the input of “driving variable”
grids and exclusionary zones where land use transitions cannot occur. The LTM receives
data inputs from the driving variable datasets in order to predict the probability (also
deemed a “change likelihood value”) of a grid cell to transition from other uses to urban
in a region (Pijanowski et al. 2001). The LTM uses this data to train an artificial neural
network (ANN) simulator known as the Stuttgart Neural Network Simulator (University
of Stuttgart, 2003).

Artificial neural networks operate by simulating the brain’s activity and its
complex processes of transfen'ing information between neurons (Hinton, 1992). Through

simulation of neural processes acting in parallel, an ANN can achieve complex

13

calculations through a series of inputs into the system. In essence, the system “learns” by
analyzing data fed into the model through a process of repetition to train the system and
reduce potential output error (Pijanowski et al. 2001). The LTM utilizes a popular ANN
known as a multi-layer perceptron (MLP). The MLP contains input, hidden, and output
layers to process information through the system. Driving variable datasets are fed into
the system where the ANN calculates random weights for the input layers and propagates
the data forward through the nodes and then backwards through a training calculation
knows as back propagation (BP). These weighted inputs are then passed through the
system from input nodes to hidden nodes, and finally to an output node. The BP
compares the expected output with the model’s calculated output, then summarizes and
distributes the error across all nodes in the network. Quite often this process of error
calculation requires several hundred or thousand forward and backward iterations to
stabilize error throughout the network (Pijanowski et a1. 2001). The LTM runs for Grand
Blanc Township completed approximately one-thousand of these iterative cycles to
stabilize error in the model output.

The LTM has been implemented with geographic data at varying scales in both
the United States and foreign countries (Michigan State University, 2003). Applications
of the LTM include studies where model output assisted in studying the spatial-temporal
effects of land use change on groundwater (Wayland et a1. 2002) to the impact of
urbanization on multi-watershed regions along Lake Michigan (Pijanowski et al. 2002).
The LTM has also been the subject of many high-profile and peer-reviewed journals
within the planning and GIS profession. This proven track record of the LTM via

publications and the continuous refinements and enhancements of the model indicate the

14

LTM’s high potential to become one of the more popular and cost effective land use
modeling solutions available to municipalities.

In addition, the LTM requires relatively few GIS data layers such as current and
historical land use, roads and interstates, water bodies, and existing urban areas to begin
processing within the ANN for forecasting and predictive purposes. The main layers that
were required for the study were readily available for inclusion in the model using
existing GIS resources at Grand Blanc Township.

2.3 MAP INTERPRETATION AND PERCEPTION OF ACCURACY

The primary output from any planning support system is typically in the form of a
two or three-dimensional map product. Maps created by these systems are often
introduced as part of a moderated forum discussing a municipality’s new master plan,
parks and recreation plan, open space program, etc. Maps are Often presented to local
officials, developers, and concerned residents as a “truthful and objective” representation
of potential land use scenarios. However, the interpretation of maps produced fi'om a
PSS remains an element of the system that is far from reaching objectivity. The accuracy
and interpretation of the model results are the first step towards acceptance of the model
by the general public.

The terms “accuracy” and “precision” are often used to describe the output of a
particular land use model. In general terms, accuracy refers to “the degree to which
information on a map or in a digital database matches true or accepted values (Foote et al.
1995).” Precision refers to the consistency of that information and the level of exactness

used in its measurement. In keeping with this definition of accuracy, the research

15

attempted to evaluate the accuracy of the urbanization information presented on the
LTM’s predictive map with the “true and accepted values” of landuse fi‘om 2000.

Cartographic skill and mapmaking ability are vital to the accurate interpretation of
any map(s) created from a PSS. The difficulty in this process is that in order to transform
a three-dimensional model of the world into a two-dimensional representation of reality,
the procedure will invariably introduce errors. As stated by Monmonier (1991) in his
seminal cartographic work, “How to Lie with Maps,” “There’s no escape from the
cartographic paradox: to present a useful and truthful picture, an accurate map must tell
white lies.”

The complexities of learning and interpreting map data compounds the difficulty
of PSS-produced maps when the map readers are the general public. Maps are often
taken at face value as a reflection of reality and with an inherent trust that the
cartographer (and in this case, the modeler) are competent stewards of their craft. Map
users tend to not question the “authority” vested in the cartographer, and operate under
this assumption of competency until proven otherwise (Monmonier, 1991). Many of the
maps generated by PSS software are generated by planners and consultants who
thoroughly understand the model and its limitations, although the likelihood that they
have had any formal training in GIS or cartography is small. Therefore, the
comprehension level of these future scenario maps may be difficult if the consultant
chooses to have the maps produced by an individual with little training in geography or
cartographic output techniques. The predictive maps produced via a PSS model display
complex patterns of information, making the appropriate color schema, line weights, etc.,

important to avoid misunderstanding by the general public.

16

The learning process employed by map readers is often a complex psychological
issue, dealing with cognitive development, comprehension of graphical symbols, and
even gender-based differences (Postigo and P020, 1998; Kirby, 1994; Harrell et al. 2000).
The myriad ways that maps are interpreted remains a topic that is truly multi-disciplinary
in nature, covering both the fields of geography and psychology. Studies have focused
on differences in map learning as differentiated between age-based expert and novice
groups (Postigo and P020, 1998). In Postigo and P020 (1998) it was determined that
complex tasks, such as map reading and graphical comprehension, require the formation
of a detailed conceptual schematic of the region being analyzed. Verbal instructions
regarding the map compositions to subjects aided significantly in the learning ability of
the map reader. Adults tended to perform better on comprehension tests versus
adolescents, but contrary to their hypotheses, no significant differences existed between
two adult groups composed of “experts” and “novices.” They further state that

“...in the case of graphical material, the capacity demonstrated by expert
subjects cannot be explained solely by their possession of large quantities of
knowledge. It is the organization of this knowledge that distinguishes the experts
and determines their mental representations. ”

Ironically, the “experts” in this study were geography undergraduate students.

Additional studies of map perception focus on the link between spatial
representations learned via the use of computer models (Rossano and Moak, 1998). The
principle of cognitive load is introduced as an additional hurdle for the map reader. Map
readers are often confronted with questions posited by researchers that may hinder
problem solving skills if the sheer amount and complexity of questions reach a critical

mass. With regards to map interpretation, Rossano and Moak hypothesized that people

with “direct learning” of map knowledge, e.g., real-world familiarity with the map area,

17

would have an advantage over map readers learning the area of interest via a highly
realistic computer-based model’s map output. The rationale was that the cognitive load
of the map reader would be reduced if their map experience was gamered from viewing a
computer-based image versus “direct learning.” It was hypothesized that reading the
realistic computer-based map would “simulate” the direct learning experience as if the
reader physically visited the location. Previous studies have suggested that even a few
minutes of map experience can often lead to superior map knowledge versus subjects
with 10 years of direct learning experience (Lloyd, 1989).

Map interpretation and learning also requires development of an understanding of
the reason for the map’s existence. This understanding stems from the three principles of
cartographic maps described in Liben (2001): purpose, duality, and spatialization. It
must be made clear to the map viewer what the map represents (purpose); what it stands
for (duality); and its relation to the real-world (spatialization). The intended audience of
the map must be able to apply these three principles in order for a sound judgment and
interpretation of the map presented to them. The principle of spatialization also takes
into account the three additional effects of map viewing distance, angle, and azimuth.
Azimuth, the orientation of the map product, is an intriguing concept that studies the
effect of map interpretation based on the top edge of the map not necessarily representing
north. This representation ties itself directly to the aforementioned points concerning
inherent trust between the cartographer and map user. If “11p” does not always equal

“north,”, then map interpretation could drop substantially.

18

2.4 PREVALENCE OF GIS DATA AND USAGE IN LOCAL GOVERNMENTS

The usage of a PSS to determine alternative land use scenarios would not be
possible without the existence of data for that system. This data collection effort has
been bolstered by the GIS industry, which has expanded the creation and distribution of
geographic datasets at the federal, state, and local levels. In the past decade, the advent
of cheaper personal computers has allowed many local governments to harness and
develop a locally based GIS (Klosterrnan, 1999a). These local governments have also
been assisted financially by the creation of GIS based datasets by other regional and state
government agencies.

In order to centralize the archiving and distribution of statewide datasets, the State
of Michigan formally established the Center for Geographic Information (CGI) on April
26, 2002 (State of Michigan Center for Geographic Information, 2003). The CGI’S
Geographic Data Library currently contains over sixty (60) unique statewide GIS
datasets including land use, topography, and digital orthophotography. All datasets in the
GDL are free of charge and available in common GIS formats for expedited use in local
GIS projects. The GDL contains enough base map information for a municipality to
create their own GIS with little or no monies spent on GIS dataset development — the
costliest portion of any GIS project. The GDL also contains all datasets from the defunct
Michigan Resource Information System (MIRIS), which includes planimetric layers from
US. Geological Survey 1:24,000-scale topographic maps (State of Michigan Center for
Geographic Information, 2003), as well as statewide land use/land cover polygons fiom

1978. The invaluable MIRIS datasets have served as the primary source of GIS

19

information for numerous state and local GIS projects over the past decade, including this
research.

In addition to MIRIS data, locally generated GIS datasets exist at lower levels of
government. Grand Blanc Township in Genesee County, Michigan, like many other
townships, generates internal datasets within their own GIS department. They currently
supplement MIRIS data with extensive datasets that include zoning, land use, sewer and
water infrastructure, and parcel based information (Taylor 1999). The Charter Township
of Waterford in Oakland County also has a fully-fledged GIS with the enhanced
capability to display and print maps on the Internet (Charter Township of Waterford,
2003). In addition, southeast Lower Michigan is also the home to several existing
countywide GIS programs, including extensive programs in Oakland, Saginaw, and
Lapeer Counties (Oakland County, 2003; Saginaw County, 2003; Lapeer County, 1999),
as well as startup programs in Genesee County (Gonzales, 1999). The Oakland County
GIS project has also taken additional steps to ensure that the data that is created at the
county level will be distributed to every governmental entity in their borders to assist in
the creation of their own GIS (Oakland County, 2003). In addition to state and local
governments, regional agencies in Michigan such as SEMCOG distribute numerous GIS
datasets covering their seven-county region in southeast Lower Michigan (SEMCOG,
2003). The Great Lakes Information Network also distributes online GIS datasets

covering the entire Great Lakes drainage basin (Great Lakes Information Network, 2003)

20

III. RESEARCH DESIGN AND METHODS

Note: Images in this thesis are presented in color

In order to better comprehend the validity and usage of these burgeoning land use
models and PSS techniques, two aspects of PSS models were examined. First, the
research attempted to validate the land use transformation scenarios produced by
Michigan State University's Land Transformation Model. The LTM represents one of the
components found in a traditional PSS, which is the ability to predict and visualize
changes in land use patterns across a region. Secondly, the research examined the usage
of the LTM's predictive output within a local municipality to gauge the reaction, impact,
map interpretation skill, and understanding of the model results by local officials and
decision makers.

3.1 STUDY LOCATION

The study site for the research thesis was Grand Blanc Township in southern
Genesee County, Michigan (Figure l). Genesee County is experiencing growth pressures
from the Flint and Detroit metro areas (Rusk, 1999) and Grand Blanc Township is no
exception. Although not explicitly stated, the LTM implies that in a fashion similar to
other PSS software such as the What If? PSS, that its application is most appropriate for
"areas that are experiencing, or anticipating, rapid urbanization (Klosterrnan, 1999a)."
Grand Blanc Township contains many of the key elements often considered vital to the
expansion of a rapidly urbanizing area. The township’s road infrastructure includes two
major interstates, I-75 and I-475, which bisect the township allowing easy access for
Metro Detroit area commuters. Development in the township has also been assisted by

the widespread availability of public water and sewer infi'astructure throughout the

21

township. An additional incentive for development was the recent activation of a new
water booster station to service the needs of residents in the rapidly developing southeast
corner of the township (Riopelle, 2003). The township also led Genesee County in

residential building permits issued in 2001 and 2002 (Richardson, 2003).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

GRAND BLANC
TOWNSHIP
GENESEE :%
COUNTY
MICHIGAN a q
/
D /
--r
a" '3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 1. Grand Blanc Township study region in southeast Lower Michigan.
3.2 DATA COMPILATION AND GEOPROCESSING STRATEGY

Data access to Grand Blanc Township's historical land use and other GIS datasets
were critical to the success of the project. Fortunately, my position as GIS Coordinator
for Grand Blanc Township granted me unrestricted access to non-confidential materials
and GIS datasets necessary to complete the research. Grand Blanc Township's GIS
Department houses multiple geographic layers as part of a GIS implementation begun at

the township in 1996 (Taylor, 1999).

22

The process of utilizing the LTM was two-fold. First, historical land use data
from the DNR’s 1978 MIRIS land use dataset (State of Michigan Center for Geographic
Information, 2003) were entered into the LTM along with existing land use from 2000.
The LTM processed the data through a back-propagation procedure in the ANN to
forecast and display areas within the township that it predicted would become urbanized
in 2000. The existing land use data from 2000 was then used a second time after the
model run as a “truth” layer to compare the LTM’s predicted urbanized areas in 2000
with the actual urban areas that existed in 2000.

Secondly, the number of cells predicted to transition to urban in the 1978-2000
model results were then utilized an additional time in the LT M to predict areas converting
to urban after 2000. Using the year 2000 as the starting point, the model calculated areas
where additional urbanization would occur using five percent intervals. The interval
maps displayed the location of fiiture urbanization starting at five percent and increasing
to one-hundred percent change in urbanized area.

Data preparation for the LTM involved obtaining several key “driving variable”
datasets that are commonly utilized within the LTM’s ANN to assist in predicting
changes in urbanization patterns. The original GIS layers for these driving variables were
obtained from Grand Blanc Township’s GIS Department in a vector-based format known
as a shapefile, the proprietary data format used within the Environmental Systems
Research Institute’s (ESRI) ArcGIS 8.2 software product.

The driving variables datasets are composed of GIS layers that display the
Euclidean (straight-line) distance from several natural and man-made geographic features

that influence urbanization (Figure 2). Among the variables created for this study were

23

layers that measured the distance from road features including highways, residential
roads, and county roads. The distance to major roads typically influence development by
allowing easier and expedited access to areas with available vacant land. Water bodies
(including lakes and streams) also lead to potential development based upon the aesthetic
attraction of areas surrounding these natural features. The distance to urban regions
variable takes into account the outward expansion and development pressures exerted on
Grand Blanc Township by neighboring urban centers.

Along with the driving variables described above, an “exclusionary zone” was
integrated into the LTM. Exclusionary zones are areas within the study region where
development cannot occur due to several reasons that limit further urbanization. The
exclusionary zone layer was composed of three integrated GIS datasets which included
existing urbanized areas (e.g., residential, commercial, and industrial development),
public lands and parks, and water bodies (including wetlands). Areas in the study region
that met these criteria were inserted into the model as locations where no further
development could occur.

The vector data for each driving variable and the exclusionary zone was converted
through a rasterization process within ArcGIS using the Spatial Analyst software
extension to ESRI’s raster GRID format. The GRID format is an image format
commonly utilized in GIS software packages where images are comprised of rows and
columns of individual pixels where each pixel has a distinct value. Upon conversion to
the GRID format, each driving variable and the exclusionary zone were converted once
again to a standard ASCH format known as a pattern file (Pij anowski et al. 2002) for

input into the LTM.

24

 

Driving Variable Datasets

“I"TH -3» no.1
‘1’" I‘ll *4} Jr "‘
(L . .

4

5.3.432: ._

 

D. Distance from rivers

"‘“1 ‘. ,0, ‘ ._‘ . I
. . . . .l - _ ‘ . .
{vh’tf‘z ,‘_»‘. ,wl.’
“ Ii I" J ,(r-j r‘w‘yfi
“#:33'; I_ ' . ‘ '

E. Distance from residential streets F. Distance from urban

   

 

 

 

Figure 2. Driving variables for the Grand Blanc Township study region.

Driving variable grids were created for an area approximately 325 square miles

surrounding Grand Blanc Township (Figure 3). Despite the fact that Grand Blanc

Township only constitutes 32 square miles (with the remaining 4 square miles

25

residing within the City of Grand Blanc), the driving variables listed above also exert
development pressure from areas outside the township. This is especially relevant when
one considers the driving variable for “distance to existing urban.” The City of Flint
represents the largest urban center in Genesee County, and lies directly north of Grand
Blanc Township. Due to the proximity of this heavily populated urban center, Flint
exerts an obvious influence on urbanization trends throughout the county. One possible
influence is through outward migration of population from the City of Flint. Census
statistics indicate that between 1990 and 2000, Flint lost approximately 11% of its
population, while Grand Blanc Township led the county in population grth (The
Detroit Free Press, 2001). The exclusion of Flint fi'om the study area would likely skew
the LTM’s prediction as to where new urbanization would occur.

The driving variable grids and associated details are displayed in Table 1. It was
decided in conversations regarding previous LTM runs, that larger scale areas such as
Grand Blanc Township would likely require a correspondingly larger raster cell size of
20 meters (65.616 feet). The resulting raster GRID files were composed of 506 rows and
500 columns of 20 meter cells. LTM runs in smaller scale settings, including the Metro

Detroit region, utilized larger cell sizes of 100 x 100 meters (Pijanowski et al. 2001).

26

 

 

  
 

Driving Variable Study Region

Davlson

 

Flint Twp

 

Burton

 

Davlaon Twp

 

rand Blane T 1‘

fin
Mandy Twp 7 Atlas Twp
Grand Blanc

 

 

Lindon

Fenton Twp Holly Twp Groveland Twp

 

Fenton

 

 

MILES

 

 

 

Figure 3. Driving variable study region surrounding Grand Blanc Township.

 

Driving variable

Description

Raster grid cell

 

 

 

Water bodies Distance fiom lakes, rivers, and 65.61 feet (20 meters)
streams

Residential roads Distance from subdivision and 65.61 feet (20 meters)
minor non-arterial roads

County roads Distance from major arterial 65.61 feet (20 meters)

roads

 

Highways and Interstates Distance from major

expressways

65.61 feet (20 meters)

 

 

Urban areas

Distance from existing urbanized
areas surrounding the township

 

 

65.61 feet (20 meters)

 

Table 1. Driving variable raster grids utilized in the Land Transformation Model

27

 

3.3 LTM 1978 TO 2000 URBAN TRANSITION RUN

To gauge the overall effectiveness of the LTM’s predictive algorithms, land use
data from the 1978 MIRIS dataset were input into the LTM. This land use layer was then
integrated with the driving variable and exclusionary zone spatial datasets to create a
modeling run to predict the extent of urbanization in Grand Blanc Township in 2000.
After the model was run, an output layer was created to indicate areas predicted as
transitioning to urban from 1978 to 2000 (Figure 4).

The regions predicted by the LTM as transitioning to urban in 2000 were then
compared to the urbanized area depicted in the 2000 existing land use layer created as
part of Grand Blanc Township’s 2000 master planning efforts. The 2000 land use layer
contained several distinct classifications of land use found throughout the Township
(Table 2). These categories were examined to determine if they were considered an
urban or non-urban land use type. The land use types deemed as urban were then
reclassified and grouped together in the GIS as a single layer representing the 2000
urbanized area. The reclassification allowed comparison between the LTM’s predicted

urbanization layer and the actual urbanized area that existed in 2000.

28

 

i

E’

( L93. ‘ wig-gr I

r ;~. '. L éfii
‘ ' Jilly.

\ f. " V

. l‘ ,
3“?

f
. Fii'i‘d'h i.

‘1 ’ J»

' I

D
l
l
4

/

 

 

E W charge. charge (3)

 

 

 

,__ _:, A};
r

7 . O ‘ ‘ I; .j"
‘i 44M?
, i {‘37

 

0 0.25 0.5

MILES

 

Figure 4. LT M predictions of 2000 urbanization pattern change.

29

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Existing2000 Land Use Urban Land Use Type?
Agricultural Vacant No
General Commercial Yes
Heavy Commercial Yes
Heavy Industrial Yes
Hospital Yes
Light Industrial Yes
Multiple-Family Residential Yes
Neighborhood Commercial Yes
Office Yes
Parks and Recreation No
Public No
R Yes
Semi-Public Yes
Single Family Residential Yes
Two Family Residential Yes
Vacant No

 

 

 

Table 2. Existing 2000 land use categories and urban/non-urban reclassification

Calculations were run between the two layers to compare several aspects of the
predicted land use pattern. The raster cells were grouped into three categories: cells
where transition to urban was predicted, but no change occurred (red); cells where
transition to urban occurred, but was not predicted by the model (green); and cells where
transition to urban was predicted where change actually occurred (yellow).

3.4 LTM PREDICTIVE RUNS WITH FIVE PERCENT INCREASES IN
URBANIZATION

The second and third model runs of the LTM revolved around extrapolation of
urbanization patterns as they may potentially exist with five percent increases in
urbanized land with no time constraint. Initial plans were to extrapolate urbanization
patterns to the years 2020 and 2040, but initial model runs indicated that by 2040 the
LTM predicted all non-exclusionary cells would transition to urban. The decision was
made that the research should instead focus on the incremental spread of urbanization at

five percent intervals to display how this urbanization pattern would progress across

30

Grand Blanc Township. Creation of the five-percent increase maps were based on the
results of the 197 8 to 2000 modeling results. The number of raster cells predicted as
having changed to urban in the 2000 run, represented by categories two and three, totaled
50,060 20-meter cells. To calculate where additional increases in urbanization may
occur, these original 50,060 cells were multiplied by 5% increments and the resulting
number was inserted into the LTM modeling program. The number generated
represented the “threshold value” which designated the value that had to be exceeded in
order for an area to transition to urban. The total numbers of cells above these threshold
values were then selected in ArcGIS using the “raster calculator” included in the Spatial
Analyst module. Maps displaying the five-percent incremental increases were then
created and included as Map Series #2 (Appendix A) and (Appendix B) #3. The
spreadsheet utilized to calculate the threshold values is included in Appendix C. These
results were then presented to local officials to solicit their opinions, insights, and
comments regarding the patterns of urbanization depicted in the scenarios.

3.5 SOLICITATION OF OPINIONS ON LTM OUTPUT MAPS BY LOCAL
OFFICIALS

Upon completion of the map series run in the LTM, the second phase of the
research involved asking Grand Blane Township department heads and elected officials
about their opinions, comments, and reactions on the urbanization patterns depicted in on
the LTM-generated maps. The rationale for this portion of the research stemmed from
the fact that many of the local officials interviewed had thirty years of experience
working in Grand Blanc Township. Hence, the officials had accumulated a vast
knowledge of Township growth and development via their unique personal experiences

and professional perspectives over time. The interviews were meant as an avenue to

31

“test” the pattern of urbanization depicted from the LTM modeling runs against the
mental maps and development assumptions predicted by local officials.

The maps were broken up into three separate map series for interviews with the
local officials. Maps depicting the 1978-2000 LTM model run and the five percent
urbanization increases were formatted in an identical fashion and placed within a binder
for respondents to examine. The binder included the requisite background information
on the project via an abstract and the MSU University Council on Research in Human
Subjects (U CRIHS) approved oral consent form (Appendix D). Respondents were given
a brief introduction to the project and had the ability to ask questions concerning the map
series. Their attention was then directed to the three numbered questions included on the
oral consent form that corresponded to the three map series in the binder. Respondents
then were left to examine the urbanization patterns depicted in the maps for
approximately fifteen minutes or however long they felt was necessary to form their
opinions on the maps. Upon completion, the respondents were then solicited for their
opinions on the map series with their responses hand-written on a worksheet that
replicated the questions included on the oral consent form.

3.6 METHODS UTILIZED FOR ASSESSING LTM ACCURACY

In order to examine differences between the predicted land use transitioned cells
and cells where a land use change actually occurred, additional accuracy assessment
calculations were performed. The accuracy assessment technique utilized was to
determine the percent correct match (PCM), which was first described by Pijanowski et
al. (2001) with the following calculation:

# cells correctly predicted to change x 100.0
#cells actually transitioning

32

The second statistical measure employed was to examine how effectively the
LTM deals with placement of the predicted land use types. The assessment of this
accuracy was calculated using a table comparing the 2000 existing land use categories
with the LTM’s number of cells correctly predicted to transition to urban. The resulting
percentages can then point to relative strengths or weaknesses in the model when it

comes to the accuracy of predicting urbanized versus non-urbanized areas.

33

IV. RESEARCH RESULTS

4.1 LTM VALIDATION - 1978-2000 ANALYSIS
The validation of the 1978-2000 projection accurately predicted a conversion to
urban for approximately 62% of the raster cells in the Grand Blanc Township model

region. Comparison results are shown in Table 3.

 

 

 

 

 

 

 

Category number LTM Cell DEscription Number of cells
One Predicted Change / No Change 19,083
Two Change / Not Predicted 19,084
Three Predicted Change / Change 30,976
Total Cells 69,143

 

 

Table 3. Year 2000 urbanization prediction results grid.
The calculation was performed using the PCM method described in the previous section,
and is shown below:

# cells correctly predicted to change x 100.0
#cells actually transitioning

30,976 X 100

= 0
50,060 61 '8 /°

Resulting PCM rates for the township’s LTM run are significantly different from
runs performed in previous studies covering larger geographic regions. The PCM results
for model runs in the Metro Detroit and Minneapolis/T win Cities region equaled 18.6%
and 31.9% respectively (Pijanowski et al. 2001). Additional LTM runs in the Grand
Traverse County region in Michigan yielded a 49% PCM, and in Kuala Lumpur,
Malaysia a 78% PCM results was obtained. The high PCM result in Grand Blanc
Township may be attributable to factors such as the smaller geographic study region,

which allowed for a much smaller grid cell size of 20 x 20 meters. Another reason may

34

 

stem from the inclusion of the “correct” driving variables that weigh heavily in
determining urbanization patterns within the study region.

Grand Blanc Township’s PCM rate of approximately 62% leads to several
observations regarding the differences between the Township’s LTM run and previous
applications of the model. First, the model’s relatively large scale analysis region
allowed greater detail in terms of grid size for input into the LTM. Previous model runs
have focused on regional areas utilizing raster cell sizes of 100 x 100 meters. The larger-
scale approach taken in this study allowed the model’s refinement and usage of cells
approximately twenty-five times smaller at 20 x 20 meters. One could hypothesize that
smaller cell sizes allow for the model to more easily predict areas transitioning to urban.
However, when previous regional models have been tested on subsets of their regions,
the PCM has been shown to increase only slightly, keeping in mind that these subsets
were still many times larger than the multi-township region utilized in this study.

Comparisons with existing 2000 land use were then performed on the
urbanization pattern predicted for 2000 using ArcMap’s Spatial Analyst module to
perform a “clipping” operation on the existing land use layer. The clipping process is an
algorithm run within a GIS to select features fi'om one map layer using another layer as
the “clipping” layer. The analogy typically used is that of a cookie-cutter slicing out the
selected geographic area(s) on the map. In this analysis, the “cookie cutter” areas were
the LTM regions where change was predicted, and where it had actually occurred (value
3, yellow tone on Figure 4). This region then clipped the existing 2000 land use layer to
show the percentage of cells that fell within the clipped regions. Table 4 summarizes the

results.

35

The table shows several interesting insights into the nature of the LTM’s

prediction of 2000 land use in Grand Blanc Township. One interprets the table by first

examining the types of land use that were predicted to become urbanized by 2000, and

then seeing the actual percentage of those areas classified as urban by the model. For

example, approximately 1.38% of the cells predicted to become urban in 2000 fell in

areas shown to be classified as “vacant” in the 2000 existing land use layer. The second

column indicates if the land use category was classified as “urban” in 2000, thus allowing

comparisons of the strengths of the LTM’s prediction on a categorical basis. The table

output points to strengths in the model where particular land use categories were

correctly predicted to transition to urban. Residential development involving single and

multiple-family developments appeared as the predominant land use type predicted

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Existing Land Use 2000 Number of 20—meter cells Percent of Total
Agricultural Vacant 93.41 0.3 1
General Commercial 260.58 0.62
Heavy Commercial 235.21 0.77
Heavy Industrial 513.18 1.68
Hospital 549.37 1.80
Light Industrial 674.03 2.20
Multiple-Family

Residential 2,292.89 7.49
Neighborhood Commercial 42.46 0.14
Office 280.61 0.92
Parks and Recreation 19.06 0.06
Public 987.34 3.23
R 13.1 1 0.04
Semi-Public 1,039.90 3.40
Single Family Residential 23,025.63 75.25
Two Family Residential 149.94 0.49
Vacant 423.57 1.38
Total 30,600.29 99.76

 

Table 4. Existing 2000 land use categories and the number of cells correctly predicted to

transition to that land use type.

36

 

correctly. The two categories constituted approximately 7.5% and 75% respectively of
the total cells correctly predicted as transitioning to urban. Commercial and industrial
development was also correctly predicted at a combined rate of approximately 4% of the
total cells. The model only sparingly predicted a transition to urbanization in areas not
classified as “urban” in this study. When combining the 2000 existing land uses not
categorized as “urban,” which were comprised of the agricultural vacant, parks &
recreation, public, and vacant land use types, the total percentage is approximately 5%.
This indicates that of the 30,060 cells correctly predicted to change to urbanized uses;
only 5% were incorrectly shown as urban when they were in reality a non-urbanized use.

4.1.2 RELATIVE INFLUENCE OF DRIVING VARIABLES ON OVERALL
MODEL OUTPUT

The 1978 — 2000 LTM transition run resulted in a percent correct match rate of
61.8% of the cells correctly predicted to transition to urban. The overall proportion of the
number of cells that were predicted correctly can be determined by plotting a goodness of
fit metric versus window size (Pijanowski et al., 2002). The window size equals the
number of cells included in the width and height of the analysis, e.g., a window of 1 x 1
equals the native 20 x 20 meter cell size used in the model. Starting with this native
window size and a 0.618 (61.8%) goodness of fit, the model increases to a value of 0.637
(63.7%) at 100 x 100 meters (e.g., .0.1 km, 5 cells) and quickly rises to approximately
0.84 (84%) at 1000 x 1000 meters (e.g., 1 km, 50 cells) (Figure 5). Values level off in
the 0.85 to 0.86 range immediately after the 1 km window threshold. As the window size
gradually increases past the calculated 100 x 100 cell window (e.g., 2 km), the values

increase towards the expected value of 1.0. The increase in overall PCM rates as window

37

size increased was expected due to comparisons with previous research. However, the
high rate at which the curve flattened at approximately 85% was higher than expected

Analyses of the LTM’s goodness of fit were also performed in the Grand Traverse
County, Michigan region. The Grand Traverse study utilized a native grid cell size of
100 x 100 meters covering a multi-county region in northwest Lower Michigan. The
number of cells correctly predicted to transition to urban in the study equaled 46% at the
100 x 100 meter cell size (Pijanowski et al., 2002). In comparison, the Grand Blanc
Township percentage correct match rate equaled 63.7% at the same 100 x 100 meter
resolution. The Grand Traverse study yielded a 65% match at the 1 km window size,
versus an 84% match in this study. However, numerous differences exist between these
two modeling analyses. In the Grand Traverse study, the analysis area covered a multi-
county region consisting of dozens of townships, versus one township in Grand Blane.

In addition to study area size, driving variables considered critical to the LTM runs in
Grand Traverse (e.g., lakeshore distance), were not relevant in the Grand Blanc Township
area.

Another method utilized to analyze the relative performance of the LTM’s output
is through statistical analysis of the driving variables used in the model run. This step is
performed by removing one of the model’s six driving variables and running the model to
calculate the difference in predictive accuracy without that variable. Upon completion of
the run, the variable is added back into the LTM and the second driving variable is
removed. This process continues for all driving variables to calculate a goodness of fit
metric that indicates the relative influence each variable has on overall model

performance. These values are calculated over an increasing window size which assists

38

 

Goodnaaa of Flt Metric

1

0.9

0.8

0.7

0.0

0.5 —Overall Fit

Variance

0.4

0.3

0.2

0.1

 

0

15 913172125293337414649535761656973778185899397101
Wlndowsmmxzomcalla)

 

 

 

Figure 5. Grand Blanc Township goodness of fit metric versus window size.

in showing the level of influence each variable has on land use change in the region.
Each variable’s goodness of fit calculation is then subtracted from the full LTM model to
determine the relative contribution in model performance. The calculated values can be
either positive or negative in nature. Positive values indicate that the variable has a
higher influence on overall model performance, and negative values indicate that the
variable has a negative impact on the model’s performance (Figure 6).

Driving variable influence fluctuates widely over the varying window sizes
depicted in Figure 6. In small window sizes of less than 0.5 km (< 25 cell size), all six
variables exert a positive influence on the model ranking from high to low as follows:
distance to residential roads, distance to rivers and streams, distance to county roads,

distance to lakes, distance to existing urban, and distance to highways. At this scale, the

39

positive influence of the county roads, lakes, and existing urban variables were very
similar due to the geographically small study area. The influence of the variables
remains positive until the window size approaches 1 km where the distance to existing
urban variable exerts a negative impact on the model performance. Interestingly, the
variables tend to cluster together in their model influence at the 0.5 km window with the
obvious exception of the distance to residential roads layer. The residential roads
variable may exert a larger influence due to the complexity and density of the residential
road network in the model. Since the driving variable region extended to the City of Flint

to include the City’s effect on the model, the density of residential roads also increased

 

 

 

 

 

 

 

vm

dramatically.
Goodnua OI FII Metric
0.12
0.1 I}
/ I \
I
0.08 l
I
I 0.5 ”I! \
I I I -—County Roads
0.00 t I
I
I
I
l

0.04

 

 

 

 

0.02

 

 

 

 

 

-0.02

 

mmmuraoxzomwm

 

 

 

Figure 6. Grand Blanc Township goodness of fit metric versus window size showing
relative influence of the six driving variables included in the LT M.

40

In window sizes between 0.5 km and 1 km (25 — 50 cell size), five of the variables
tend to have a decreasing influence on model performance with the exception of the
distance to highways variable. The highways variable increases steadily in overall model
influence across all window sizes before leveling off near the 1.5 km cell size. One
possible explanation could be that the larger window size incorporates a higher
proportion of the three major highways that intersect the study region, including two
interstates (I-75 and I-475) that bisect Grand Blanc Township. Four of the six variables
increase in positive model influence after an inflection point at the window size of 1.2 km
(60 cell size), where the distance to urban influence exerts a negative influence on the
model before rebounding as window size increases. Once again, the inclusion of the
numerous urban centers surrounding Grand Blanc Township likely exerts a larger
influence as window size increases. The distance to urban influence would possibly be
much higher if the driving variable study region was expanded to include the Pontiac,
Michigan metropolitan area to the south.

In the Grand Traverse study, many of the same driving variables were utilized in
the LTM runs. Although using a larger grid cell size, the ten driving variables’ relative
influence on the model was comparable to the results found in Grand Blanc Township.

In particular, the distance to residential roads variable exerted the second highest
influence on the model at window sizes less than 0.5 km, where in Grand Blanc
Township the variable’s influence ranked first overall. The distance to highways variable
also exerted the third-highest positive influence on the Grand Traverse model. However,
in Grand Blanc Township the distance to highways influence ranked last and nearly

exerted a negative influence on the model at the original window size. Ironically, the

41

Grand Traverse highway variable begins with a high positive influence and decreases
until reaching the 0.5 km window, where it then steadily increases across all window
sizes much like the constant increase shown in Grand Blanc Township. The rationale for
the increasingly positive influence of highways may be related to the enhanced
accessibility that highways offer in terms of access to a region (Pijanowski et al., 2002).
Highways and interstates influence also hinges on the accessibility of the highways
themselves. Limited access highways typically have higher posted speed limits, thus
allowing expedited access for commuters and travelers. The number of entry points onto
these highways has been shown to influence commercial development through the
construction of gas stations, fast-food and lodging establishments, and other businesses
related to highway travel.

4.2 TOWNSHIP OFFICIALS’ OPINIONS AND REACTIONS TO LTM
FORECAST RESULTS

The three map series were generated and given to township officials and
department heads over a two-week period in June 2003. During that period of time,
interviews were conducted with seven township officials with an in-depth knowledge of
Grand Blanc Township. Six department heads were interviewed including assessing,
building, planning/zoning, public works, finance, police, and the township’s elected
treasurer. The average length of employment at Grand Blanc Township was
approximately twenty years among the seven respondents, with several individuals
having over thirty years experience. The interviews were set up at a time convenient for
the officials since the discussions were set during business hours. Typical interviews
involved 5-10 minutes of background information regarding the research and the format

of the questions to be asked regarding the maps. It was made evident that the interview

42

would serve a dual purpose. First, the interview would consist of a brief discussion
concerning the map series and any questions regarding map comprehension. Secondly,
the interview would serve as forum to solicit opinions and reactions based upon the
official’s knowledge of Grand Blanc Township.
4.2.2 OPINIONS ON LTM 1978-2000 URBANIZATION PATTERN

The first map series shown to the officials depicted the LTM’s prediction of land
converting to urban for the year 2000 (Figure 4). Respondents had the most difficulty
interpreting this map even with several minutes of explanation regarding the modeling
process. Fortunately, most respondents had some exposure to GIS via the township’s
GIS Department, but many were not familiar with GIS in a modeling capacity as their
main exposure was simply of GIS as a “computer mapping” tool. Several were curious
as to how a GIS-based model functioned in general and the LTM in particular. The
respondents were provided with a figure (Figure 2) showing the inputs used in this LTM
modeling run to assist in their interpretation. In order to help categorize and summarize
the respondent’s opinions and reactions on the pattern depicted on the map series, a map
was created breaking Grand Blanc Township into four quadrants of nine one-square mile
sections each (Figure 7). Results fi'om the interviews were then categorized by the level
of confidence in the maps and specific quadrants expressed by the official after

examining each map series (Table 5).

43

 

 

 

 

 

 

 

 

 

 

 

 

 

OPINION RESULTS SUMMARY TABLE Respondent’s Confidence Level
and luadrants Mentioned
Low Medium High
Map Series #1: 1978-2000 Map
Category 1-Red* 0 7 (SW, SE) 0
Category 2-Green* 2 (SW) 5 (SW) 0
Category 3-Yellow* 0 2 (SE) 5 (NE,
SW)
Map Series #2: 5%-50% 0 7 (SE, SW, 0
NW)
Map Series #3: 55% - 100% 0 6 (SE, SW) 1 (SE)
*Category Definitions:

Category 1 = Predicted Change/No Change
Category 2 = Change/No Change Predicted
Category 3 = Predicted Change/Change Occurred

 

Table 5. Summary table of opinions and confidence expressed regarding the three map
series utilized during interviews with Grand Blanc Township oflicials.

As expected, the respondents’ knowledge of the township and its pattern of
urbanization led to mixed reactions on the map presented to them. Initial reactions to the
LTM’s areas where change was predicted, and change occurred (Category 3 on Figure 4)
ranged from “quite amazingly accurate” to “fairly representative.” Most respondents felt
that the model adequately displayed areas where they knew urbanized land had existed in
the 2000. Specific areas where the officials opined that the LTM predicted 2000 urban
lands correctly were the NE and SW quadrants of the township. Development between
Perry Road and Hill Road in the NE quadrant grew rapidly in the late 1990s, and the

model correctly predicted change in those regions. The SW quadrant of the township has

 

had many new subdivisions platted between Baldwin Road and Cook Road to the north,
which were also correctly identified according to several respondents.

Criticisms of the model included the overwhelming response that the model
completely overlooked the development of section 36 in the SE quadrant of the township.
The development known as the Woodfield (Figure 7) Planned Unit Development (PUD)
encompasses almost all of section 36 of Grand Blanc Township as well as extending into
the northern half of section 1 in Holly Township to the south. Much of the Woodfield
PUD, which was built in 2000 and includes several hundred single-family homes and
condominiums intertwined with golf courses, was not predicted as urbanized land. It
should be mentioned that despite the larger number of residences, the Woodfield PUD is
classified as an “open-space development” replete with an 18-hole golf courses and
walking trails. Several respondents also expressed concern over the model’s prediction
of urbanized land around Halsey Road in the SW quadrant. Some respondents mentioned
that in their opinion the areas around Halsey Road could not be considered “changed” to
urban in 2000.

The respondents then examined Category 2 (green tone on Figure 4) on the LTM
map where change occurred, but was not predicted by the model. Many respondents
weighed in less on the areas where the LTM did not predict change, but more on the fact
that these areas had not become urbanized by 2000. The comments on the areas depicted
as having changed to urban mirrored those mentioned above concerning the Halsey Road
corridor. In essence, many respondents felt that the areas shown as urbanized in 2000

were not entirely correct. However, the respondents felt that these were relatively

45

isolated and confined mainly in the SW quadrant in the Halsey Road corridor. These

opinions did not detract from their overall opinions on the accuracy of the projections.

 

Township Officials' Reaction Quadrants

 

 

 

 

 

Figure 7. Grand Blanc Township quadrants used to categorize respondents
opinions on LTM map series results.

46

Additional concerns focused on their opinions of areas where the LTM “missed
the boat” (according to one observer) in not predicting urbanization. Many of the
comments focused on the influence of Interstate 75 (I-75) which bisects the township in a
SE-NW direction. Respondents felt that the model should have accounted more for the
influence of I-75 on development in the township. The fact that distance to highways
was included as a driving variable in the model was reiterated to the respondents. Upon
hearing this several respondents felt that it should be “weighted” more in the model
analysis. This was an interestingly intuitive response considering that only one of the
seven respondents had any prior experience or knowledge of GIS modeling and
weighting of layers. An additional area of concern was the lack of change prediction for
an area known as the Preserve (Figure 7), which is also in the SE quadrant of the
township. The Preserve stands as the highest-density development in the township, but
was not shown as changing to urban in the LTM’s 2000 model run.

Finally, respondents focused on Category one (red tone on Figure 4) in the model
where change was predicted by the LTM, yet no change occurred. Several respondents’
interest again focused on the Halsey Road corridor in the SW quadrant. In particular, the
LTM predicted change for areas east of Halsey Road near the Baldwin Road intersection.
This area is currently occupied by several rural homes with no platted subdivisions. The
additional areas shown incorrectly according to the respondents runs north-south along
Vassar Road in the SE quadrant of the Township. The area was not developed in 2000,

but many pointed out that several subdivisions are currently being looked at in the region.

47

4.2.3 OPINIONS ON LTM EXTRAPOLATION — FIVE PERCENT
PROJECTIONS

After examining the predicted urbanization pattern in 2000 depicted in Map
Series #1, the respondents proceeded to Map Series #2 (Appendix A) and #3 (Appendix
B), which displayed areas predicted as transitioning to urban in five-percent increments
of five to one-hundred percent. The map series were broken up in two different sections.
Map Series #1 displayed the result maps showing five-percent incremental increases from
five to fifty percent, while Map Series #2 displayed maps showing increases from fifty-
five to one-hundred percent. Although the maps portrayed identical types of data, it was
decided that the respondents would be more apt to offer opinions if the maps were broken
up in two distinct map series. This allowed the respondent to digest all the information
presented to them in smaller amounts often maps per series versus twenty maps total for
both series.

The respondents first examined Map Series #2, which was met with mixed results.
Many respondents expressed concerns similar to those mentioned during examining the
1978-2000 predictive run. Namely, many felt that the Woodfield PUD in the SE
quadrant should indicate an increase in urbanization. Respondents also felt that the rapid
encroachment of urbanization around the Halsey Road corridor was unlikely based on
their knowledge of the region. One respondent even categorized the Halsey Road
prediction as a “total miss.” However, it was pointed out by another respondent that the
model may be reacting to the Genesys Hospital development which lies on the north side
of Baldwin Road near I-75 and directly across fi'om the Halsey Road region. Another
opinion that was expressed regarding the Halsey Road region referenced the fact that

Halsey Road is still an unpaved road, and one of only several miles of dirt or gravel roads

48

in the township. The respondent’s point was that Halsey Road’s development will be
inexorably linked to the inevitable paving of the road, hence leading the respondent to
concede that the model may be more accurate than originally thought in the SW quadrant.
The northern border of the township is defined by Maple Ave, where one respondent
expected the LTM to indicate more development based on “cheaper housing and access
to Grand Blanc schools.” Another respondent expressed concern over the model showing
development in the NW quadrant along Porter Road. The respondent mentioned that
non-buildable wetland areas exist in the region were displayed as converting to urban.
The respondents then examined Map Series #3, which displayed urbanization
increases from fifty-five to one-hundred percent. Respondents were much more verbose
and detailed in their opinions of the maps, which allowed for interesting discussions
regarding the apparent ability of the LTM. Comments on the final series’ pattern of
urbanization ranged from “I don’t’ understand why Halsey would develop” (once again
referring to the Halsey Road corridor), to “Glad I’ll be retiring!” expressed eloquently by
one respondent after viewing the nearly total conversion to urban on the 100 percent
increase map. Many of the comments centered on concerns over the model’s predicted
lack of urbanization in areas the respondents were certain would develop over a
reasonable time frame. Once again, the Woodfield PUD was mentioned as a curious
anomaly in the model results. Several respondents felt that the model didn’t develop the
area adequately in future extrapolations of urbanization, which was a similar concern in
the 2000 prediction. The common theme was that respondents turned the map pages
anticipating that the model would eventually “get it right” in accordance to their opinion.

The focus on the Woodfield PUD in some ways became a litmus test for the overall

49

accuracy of the model, as 6 of the 7 respondents independently mentioned its lack of
development on the extrapolation. Several respondents also expressed their opinion that
perhaps the forested areas in the Woodfield PUD were somehow preventing the LTM
from increasing urbanization in that region. One respondent did point out that in some
ways the LTM handled Woodfield correctly in their opinion since predicted urban didn’t
cover all of the Woodfield PUD, thereby “preserving the ‘character’ of Woodfield” with
its golf courses and open space.

Additional comments on the maps referred to the lack of development in the SW
quadrant’s southwest corner. Specifically, three developments of over 300 homes each
have been approved or are under construction in the SW quadrant. Respondents had
hoped that the model would accurately show future development in that quadrant, as it
will invariably happen over the next five to ten years.

4.3 CONCLUSIONS AND RECOMMENDATIONS
4.3.2 LTM 1978-2000 VALIDATION

The inclusion of additional driving variables and expansion of the study region
could further refine the LTM and increase predictive accuracy. However, an increase in
accuracy would require the same driving variables at an increasing level of detail as the
study area grows larger in extent. It was determined early in the process that the
influences of existing urban centers necessitated a larger study region surrounding Grand
Blanc Township (Figure 2). Although the township was the main focus of the study with
its smaller grid cell size, the township’s developmental influences extend far beyond its
36-square mile border. Perhaps one of the reasons for the higher PCM rate for the

township resulted from the fact that the larger study region surrounding the township was

50

utilized for driving variable grid creation? As shown in figure 5 by the increasing PCM
rate over larger cell sizes, the expansion of the study area gradually increases the PCM to
a plateau after which the levels would likely decline with additional study region growth.
This begs the question, “How far is too far out?” when it comes to the study region. One
could ensure all urbanization influences were accounted for by pushing the study region
envelope to include county, regional, or even statewide areas. Statewide projections
using the LTM have shown that the model can be utilized over large geographic areas
using readily available land use data from MIRIS (Public Sector Consultants, 2001).
However, this would not likely be practical if highly detailed GIS data were required for
such a large region.

Driving variable grids chosen for the study could be expanded to include
suggestions made by township officials in their opinion interviews (many of which are
described in the next section). One suggestion was to include distance to school districts
as a driving variable in the model. This additional input would likely influence the model
results in the township if one considers the differences in schools from region to region
throughout Michigan on the basis of statistical measures such as the Michigan
Educational Assessment Program (MEAP). In the case of Genesee County, the
difference between school districts is very pronounced if one considers the disparities in
MEAP scores for 2001 between the Flint and Grand Blanc Public School Districts. In
high school MEAP tests taken in the fall of 2001, 88% of Grand Blanc high schools met
or exceeded statewide standards in mathematics versus 35% in Flint high schools
(Genesee County Intermediate School District, 2002). Grand Blanc schools substantially

outperformed Flint schools in the additional testing categories of reading, writing,

51

science, and social studies. Datasets of MEAP scores are readily available for school
districts statewide, many of which are in database formats that are easily converted to
GIS grids for usage in any future LTM runs.

The issue of parcel salability could also be a driving variable in large-scale LTM
runs. During many interviews with local officials, occasional comments would be made
regarding larger parcels of land that were pointed out as having “never sold” or “will
never be sold.” These types of observations stemmed fiom casual or personal knowledge
of the parcel’s owner and any information the respondent might have regarding the
owner’s intention to sell their property for future development. If GIS data were
available for a municipality, a more objective driving variable could be created that takes
into account the distance from these large parcels based on several factors: a “longevity
factor” that examines time since last sale or parcel split; parcel size; access to
infiastructure; property tax rates; and current zoning classification. Creation of these
datasets could influence the model in unforeseen ways if data were obtainable.
Unfortunately, even with a more objective criterion for development of the variable,
access to GIS data across a study region similar in size as the one used in this study
would prove difficult without a county or regional GIS authority in place. This type of
analysis will hopefully be possible in the near future as more county, regional and state
GIS entities such as the State of Michigan’s Center for Geographic Information, and
Grand Rapids’ Grand Valley Metropolitan Council begin to centralize and distribute their
GIS datasets.

Creation of a distance to infrastructure driving variable may also be very

influential on a larger-scale model. The pattern of development across a township often

52

follows the location of access to sewer and water infrastructure, and Grand Blanc
Township is no exception. If data were available, a driving variable could be constructed
using the distance to existing sanitary sewer or water mains or a combination of the two.
This variable may even work on a regional context if regional water or sewer systems
were taken into account. One example would be the Detroit Water System, which winds
its way parallel to the I—69 corridor from Lake Huron into Flint’s water distribution plant
which distributes water throughout the Flint Metropolitan area. However, the ability to
tap into the water line across these counties is limited, so any regional modeling with the
water infrastructure variable would have to be examined closely to ensure the correct
influence of the variable on model results.
4.3.3 TOWNSHIP OFFICIALS’ OPINIONS AND REACTIONS ANALYSIS

Map reading and interpretation is very subjective in its very nature. People often
enter into a map reading situation with preconceived notions of what constitutes “correct”
and what constitutes “erroneous.” These opinions are unique to each respondent since
their individual life experiences and collective knowledge influence how they look at the
world. The subjective nature of PSS modeling and the often volatile nature of the
emotions surrounding the topics of development and urban sprawl, indicates that a high
level of confidence must be demonstrated for any predictive model. The LTM provides a
solid and proven foundation for land use modeling, but the critical step in testing any GIS
model output or map in general is to examine its impact on the intended audience. In
many cases this audience is composed of the uninitiated laymen when it comes to topics
such as raster-based GIS modeling or neural networks. However, the same audience is

often composed of knowledgeable community leaders, planners, and citizens who

53

constitute the “expert panel” when it comes to local knowledge of a region and its
history. These individuals come as interested observers and examiners of the end result,
and their confidence is then placed into the hands of the session’s facilitator, modeler,
urban planner, and ultimately of the cartographer. The amount of cartographic skill that
is required to properly display the model output is critical to map interpretation by the
end user. Recent trends in the field of GIS have seen many educational institutions
implementing GIS programs that consist less of formal geographic (and cartographic)
training, than they do of “technical institute” style curricula focused on learning popular
GIS software packages. Hopefully these institutions will enhance or supplement their
programs with formal training in geographic principals, lest the future cadre of GIS
practitioners head off to produce maps that may add to the difficulties of the map reader.
In the research, both the technological modeling aspect and the “human element”
regarding the response to the model’s cartographic output were analyzed. One of the
human elements that were taken into consideration relates to the cognitive load on the
respondent. During the interviews concerning the map series showing incremental
increases, it was observed that the maps were perused very quickly by respondents as
they rushed to see the “end result.” Respondents were increasingly talkative when
examining Map Series #3. This is likely due to the increase in reactions stemming hour
the spreading pattern of urbanization depicted in the third map series. The number of
opinions generated runs counter to the rationale for breaking the maps up into separate
map series, which was to ease the cognitive load of the respondent. The data from the
respondents also display the most detailed responses and reactions expressed after

viewing Map Series #3. Map Series #2 did not generate the response anticipated, as

54

respondents appeared anxious to proceed to last map in Map Series #3 to “see what
happens” to the township.

The research and opinion interviews with township officials often took the
research in unforeseen directions. One intriguing new direction which evolved was that
during and after many of the interviews regarding the urbanization pattern in the
township, the conversation often became a forum for the official to express their opinions
and rationale as to what drives suburban development. Since the opinion of each
respondent on the causes of urban development exerts an obvious influence on their
opinion of the LTM results, several interesting points were raised which could lead to
future refinement of the model. Many of the additional driving variables suggested for
future LTM research such as distance to quality schools, existing infrastructure, and
parcel salability, were all “discovered” during these interview sessions.

Several respondents pointed to the fact that the model tended to show
development in areas near the fringe of the township, and speculated that access to Grand
Blanc Schools were the influencing factor. It was pointed out that the model inputs did
not include distance to particular schools or school district regions, but acknowledged
that the addition of a school distance driving variable might exert influence on the model.
One way this factor could be included would be through the addition of a weighted
distance to school district driving variable that could be based upon statistical measures
such as retention rates and MEAP scores, which are readily available from the State of
Michigan website (State of Michigan Department of Education, 2003). Grand Blanc
Community Schools would undoubtedly influence the model as its schools are highly

regarded with some of the highest MEAP scores in the State of Michigan

55

Many respondents also stated that in the case of Grand Blanc Township,
development trends point to an influx of population and spreading of development in a
south to north fashion. Grand Blanc Township fits the definition of a “bedroom
community” with little commercial development but rapid development of residential
subdivisions along the southern border with Oakland County. Respondents felt that the
model inaccurately displayed heavier concentrations of urbanization heading south fi'om
the Flint metropolitan area. Ironically, much of the development on the northern border
of the township has cropped up in recent years due to the aforementioned proximity to
the Grand Blanc School District. Respondents were informed that the model included
distance to urban centers in the nine-square township study region including local cities.
One observation is that perhaps the scale of the study area needs to be larger to account
for this driving variable. Especially if one tries to extrapolate urbanization trends in a
large-scale area such as Grand Blanc Township.

The factor would also depend on the proximity to larger urban centers. In the
case of this research, it was decided to include the Flint metropolitan region, but not the
Pontiac metropolitan region due to F lint’s sharing of a portion of the township’s northern
border. The results of the LTM’s urbanization pattern on the southern fringe may have
been much different if the study area included the Metro Detroit region where much of
the outward migration to suburban areas originates. However, individual model runs will
have to be evaluated on a case-by-case basis to determine the geographic extent
necessary to include all relevant factors in the model, which will evolve with future

studies of the LTM and its predictive outputs.

56

4.4 FUTURE RESEARCH AND THE LTM

The research results presented in this study indicate the power and ease of use
inherent in the LTM. The small number of required datasets consisting of physical
driving variables such as water bodies and roads, and the common GIS firnctions
employed to create these datasets, has led to several applications of the LTM worldwide.
One of the goals in this research was to determine the opinions and reactions of local
officials on the output presented by the LT M. Although the research employed my own
cartographic techniques to display the model results, the intention was to examine the
reactions of officials when exposed to the output fi'om technologies as advanced as an
artificial neural network. Research has been relatively sparse when examining how end
users utilize the maps produced by any PSS or modeling algorithm. Future applications
of the LTM should attempt a more dichotomous approach focusing on model refinement
and research, as well as the “hmnan element” involving interpretation and mapping of the
model output.

When compared to other planning support systems, it’s apparent that the LTM
exists simply as artificial neural network “engine.” Many PSS applications include an
advanced graphical user interface (GUI) from which to perform a variety of planning and
GIS-based routines. The LTM exists in a “raw” form only, using a command-line
interface without the GUI familiarity expected by most computer users. Although the
researcher may be familiar with the GIS inputs necessary for the system, the lack of an
interface or input “wizard” to feed data into the LTM could create difficulty for users

unfamiliar with the system.

57

The future of the PSS in the opinion of this researcher should be to invest in an
effort to transform the LTM from a stand-alone modeling package to a fully-functioning
PSS on par with other software in this burgeoning field. It is recognized that it might not
be the goal of the developers to take the package beyond a purely theoretical research
too]. However, the relative ease of use and power of the model engine could be a
valuable addition to the suite of tools available in other PSS packages. Integration of the
model into an existing PSS could complement tools such as policy analyzers and three-
dimensional visualization routines to assist the decision-making capabilities of the end
user. The result may be beneficial to both the developers of the LTM to receive wider
usage of their model to assist in sensible planning, as well as to planners and officials

seeking tools to visualize the impact of planning decisions on their local communities.

58

REFERENCES

Aronoff, S. 1989. Geographic Information Systems, A Management Perspective.
Ottawa, Canada: WDL Publications. 294 pp.

Brail, R.K. and R. Klosterman. 2001. Planning Support Systems, Integrating
Geographic Information Systems, Models, and Visualization Tools. ESRI Press.
443 pp.

Charter Township of Waterford. "The Charter Township of Waterford, Michigan-GIS
Department Home Page." [Online] Available.
http://www.twp.waterford.mi.us/gis/, 2003.

Chicago Area Transportation Study. “History of CATS.” [Online] Available.
http://www.catsmpo.com/aboutcats/history.htm, 2003.

City of Denver, Colorado. 2000. Denver Comprehensive Plan 2000. Denver, Colorado.
245 pp.

County of Bolder, Colorado. “Boulder County Comprehensive Plan — Agricultural
Element.” [Online] Available.
http://www.co.boulder.co.us/lu/bccp/agriculture.htm, 2003.

Environmental Research Systems Institute. “History of ESRI.” [Online] Available.
http://www.esri.com/company/about/history.html, 2003.

F oote, K.E., and DJ. Huebner. “Error, Accuracy, and Precision —— The Geographer’s Craft
Project.” [Online] Available.
http://www.colorado.edu/geography/gcraft/notes/error/error_£htrnl, 2003.

Genesee County Intermediate School District. “Genesee County MEAP High School Test
Results.” [Online] Available.
http://www.gisd.k12.mi.us/gisd/hst_winter02.pdf, 2003.

Gonzalez, D. 1999 Genesee County Environmental Health Department, Flint, Michigan.
Personal Communication.

Grand Valley Metropolitan Council. “Grand Valley Metropolitan Council.” [Online]
Available.
http://www.gvmc.org/administration/index.shtrnl, 2003.

Great Lakes Information Network. “GIS Data Sets by Topic.” [Online] Available.
http://www.glin.net/gis/data/topic.html, 2003.

59

Hamlin, R., J. Thomas, P.L. Machemer, and K Kolakowski. 2000. Urban Growth
Boundaries. A Policy Brief for the Michigan Legislature. East Lansing:
Michigan State University Urban and Regional Planning Program. 10 pp.

Harrell, W.A., J.W. Bowlby, and D. Hall-Hoffarth. 2000. “Directing Wayfinders With
Maps: The Effects of Gender, Age, Route Complexity, and Familiarity With the
Environment.” The Journal of Social Pscyhology. 140:169-178.

Harris, B. and M. Batty. 1993. "Locational models, geographic information and
planning support systems."Journal of Planning Education and Research 12: l 84-
198.

Heikkila, E.J. 1998. "GIS is dead; long live GIS!" Journal of the American Planning
Association 64:350-360.

Hinton, GE. 1992. “How Neural Networks Learn fi'om Experience.” Scientific American
267:145-151.

Hopkins, L.D. 1999. "Structure of a planning support system for urban development."
Environment and Planning B: Planning and Design 26:333-343.

Kirby, J .R. 1994. “Comprehending and Using Maps: Are there Two Modes of Map
Processing?”, in: GE. Stelmach and RA. Vroon (Eds.). Comprehension of
Graphics. North Holland. pp. 63-77.

Jensen, JR. 1986. Introductory Digital Image Processing, A Remote Sensing Perspective.
Prentice-Hall. 379 pp.

Klosterrnan, RE. 1997. "Planning support systems: A new perspective on computer-
aided Planning. "Journal of Planning Education and Research. 17:45-54.

Klosterrnan, R.E. 1999a. "New perspectives on planning support systems." Environment
and Planning B: Planning and Design 26:317-320.

Klosterrnan, R.E. 1999b. Personal Communication, Electronic Mail, 27 October.

Klosterrnan, R.E. 1999c. "The What If? collaborative planning support system."
Environment and Planning B: Planning and Design 26:393-408.

Lapeer County, Michigan, "Lapeer County Government-Equalization-G.I.S." [Online]
Available. http://www.county.lapeer.org/Equal/index.html, 1999.

Lee, J ., L.J. Erickson, and TD. Kulikowski. 1998. "Analyzing growth-management

policies with geographical information systems." Environment and Planning
B: Planning and Design 25:865-879.

60

Liben, LS. 2001. “Thinking through Maps”, in: M. Gattis (Ed.) Spatial Schemas and
Abstract Thought. MIT Press. pp. 45-77.

Lloyd, M. 1989. “The estimation of distance and direction from cognitive maps.”
American Cartographer 16:109-122.

Machemer, P.L., M.D. Kaplowitz, and TC. Edens. 1999. Managing Growth and
Addressing Urban Sprawl: An Overview. Michigan State University. 27 pp.

Maquire, D., M.F. Goodchild, and D. Rhind. 1991. Geographic Information Systems:
Principles and Applications volumes 1 and 2. New York, John Wiley.

Michigan Department of Environmental Quality, Michigan Coastal Management
Program. 1999. "Local Tools & Techniques to Achieve Smart Growth." 20 pp.

Michigan Department of Natural Resources, Land and Mineral Services Division. "Rmap
Home Page." [Online] Available.
http://www.dnr.state.mi.us/RED/catalogue/index.htrn, 1999.

Michigan State University. "LTM Projects.” [Online] Available.
http://www.ltm.msu.edu/projects.htm, 2003.

Monmonier, M. 1991. How to Lie with Maps. University of Chicago-Press. 207 pp.

New Jersey Pinelands Commission. “NJ Pinelands Commission — Density Transfer.”

[Online] Available.
http://www.state.nj.us/pinelands/density.htm, 2003.

Oakland County, Michigan. "Oakland County-GIS." [Online] Available.
http://www.co.oakland.mi.us/gisl, 2003.

Pijanowski, B., D.T. Long, S.H. Gage, and WE. Cooper. 1997. “A Land Transformation
Model: Conceptual Elements, Spatial Object Class Hierarchies, GIS Command
Syntax and an Application to Michigan's Saginaw Bay Watershed.” Research
presented at Land Use Modeling Workshop, Sioux Falls, SD. June 1997.

Pijanowski, B., B.A. Shellito, M.E. Bauer, and KB. Sawaya. 2001. “Using GIS, Artificial
Neural Networks and Remote Sensing to Model Urban Change in the
Minneapolis-St. Paul and Detroit Metropolitan Area.” ASPRS Proceedings.

St. Louis.

Pijanowski, B., B.A. Shellito, S. Pithadia, and K. Alexandridis. 2002. “Forecasting and

assessing the impact of urban sprawl in coastal watersheds along eastern Lake
Michigan.” Lakes & Reservoirs: Research and Management. 7:271—285

61

Postigo, P. and J. Pozo. 1998. “The Learning of a Geographic Map by Experts and
Novices.” Educational Psychology 18:65-80.

Public Sector Consultants. 2001. Michigan Land Resource Project. Lansing, Michigan.
135 pp.

Putnam. S. H. and S. Chan. 2001. “The METROPILUS Planning Support System: Urban
Models and GIS”, in: R.K. Brail and RE. Klosterman (Eds) Planning Support
Systems. ESRI Press. pp. 99-128.

Richardson, K. 2003. Charter Township of Grand Blanc, Grand Blanc Township,
Michigan. Personal Communication.

Riopelle, N. 2003. Charter Township of Grand Blanc, Grand Blanc Township, Michigan.
Personal Communication.

Rusk, D. 1999 "Urban sprawl effects on Genesee County, Michigan." Research presented
at Urban Sprawl Forum, Flint, Michigan. February 1999.

Rustem, W. 1997. Research presented at Michigan State Land Use Forum, East Lansing,
Michigan. 1997.

Rossano, M. and J. Moak. 1998. “Spatial representations acquired from computer
models: Cognitive load, orientation specificity and the acquisition of survey
knowledge.” British Journal of Psychology 89:481-497.

Saginaw County, Michigan. "Saginaw County GIS." [Online] Available.
http://www.saginawcounty.com/gisnf.html, 2003.

Sierra Club. "1999 Sprawl Report." [Online] Available.
http://www.sierraclub.org/sprawl/report99/, 1999.

Southeast Michigan Council of Governments. 1995. "2020 regional development
forecast: summary report." Detroit, Michigan. 45 pp.

Southeast Michigan Council of Governments. 2001. “2030 Regional Development
Forecast for Southeast Michigan.” Detroit, Michigan. 44 pp.

Southeast Michigan Council of Governments. “SEMCOG Maps and GIS/Digital Data.”
[Online] Available. http://www.semcog.org/Products/MapsGIS/DigitalData.htrn,
2003.

Staley, S.R., J.G. Edgens, and G.C.S. Mildner. 1999. A Line in the Land, Urban Growth

Boundaries, Smart Growth, and Housing Aflordability. Reason Public Policy
Institute.

62

State of Michigan Department of Education. "MEAP Information." [Online] Available.
http://www.meritaward.state.mi.us/mma/meap.htm, 2003.

State of Michigan Center for Geographic Information. "Center for Geographic
Information is Established.” [Online] Available.
http://www.michigan.gov/documents/cgi_release_20997_7.pdf, 2003.

Swanson, E. 2003. Michigan Land Use Leadership Council. Geographic Information
Systems (GIS) Information Enhanced Land Use Planning. Lansing, MI 9 pp.

Taylor, J. 1999. Charter Township of Grand Blanc, Grand Blanc Township, Michigan.
Personal Communication.

The Detroit Free Press. "Metro Detroit Sprawl: Suburban transplants alter landscape,
identity of Fenton." [Online] Available.
http://www.freep.com/news/metro/fenton16_20010416.htm, 2001 .

The Detroit News. "Public seeks to tame sprawl." [Online] Available.
http://www.detnews.com/ l999/specials/suburbs/lead/lead.htrn, 1999.

United State Environmental Protection Agency. “EPA Green Communities- Build Out
Analysis.” [Online] Available.
http://www.epa.gov/greenkit/build_out.htm., 2003.

United States Geological Survey. "Geographic Information Systems-GIS." [Online]
Available. http://www.usgs.gov/research/gis/titlehtml, 1999.

University of Stuttgart. "SNNS-Stuttgart Neural Network Simulator." [Online] Available.
http://www-ra.informatik.uni-tuebingcn.de/SNNS/, 2003.

University of Wisconsin. "History of GIS." [Online] Available.
http://feature.geography.wisc.edu/sco/gis/gis_hist.htrnl, 1999.

Waddell, P. 2001. “Between Politics and Planning: UrbanSim as a Decision-Support
System for Metropolitan Planning”, in: R.K. Brail and RE. Klosterrnan (Eds.)
Planning Support Systems. ESRI Press. pp. 201 -228.

Washtenaw County, Michigan. 2002. Build-out Analysis for Washtenaw County
Communities. Ann Arbor, Michigan. 28 pp.

Wayland, K.G., D.W. Hyndman, D. Boutt, B. Pijanowski, and D.T. Long. 2002.
“Modeling the Impact of Historical Land Uses on Surface Water Quality Using
Ground Water Flow and Solute Transport Models. Lakes and Reservoirs:
Research and Management. 7: 189-199.

63

APPENDIX A
Map Series #2
Five Percent Increment Maps
Five through Fifty Percent

 

 

 

 

Landuse Type {:1 Conversion to urban (5% increase)
- Urban! Built-Up N Roads N
- Agricultural w E
- Non-forested lands
- s
Forest lands
- Water bodies 0 0.25 0.5 1
-w°“a"°‘ MILES

 

 

 

 

 

 

 

 

 

Figure 8. Five percent increase in conversion to urban.

65

 

 

 

Landuse Type [:] Conversion to urban (10%)
- Urban/ Built-Up /\/ Roads

- Agricultural

m Non-forested lands

- Forest lands

- Water bodies

- Wetlands

 

 

 

 

 

0 0.25 0.5 1

MILES

 

 

 

Figure 9. Ten percent increase in conversion to urban.

66

 

 

 

 

1

0 0.25 0.5

i=1
MILES

 

C] Conversion to urban (15%)

N Roads

 

 

- Urban! Built-Up
- AgncuIIuraI

- Non-forested lands
- Forest lands

Landuse Type

- Water bodies
- Wetlands

 

 

 

 

 

 

 

 

Figure 10. Fifteen percent increase in conversion to urban.

67

 

 

 

 

 

 

Landuaa Typo C] Comerslon to urban (20%)
- Urban! Bullt-Up N Roads
[3 Agricultural N
L] Non-forested lands W$r
- Forest lands
- was: bodies 0 0.25 0.5 1
[J Wetlands
MILES

 

 

 

 

 

 

 

Figure 11. Twenty percent increase in conversion to urban.

68

 

 

 

 

Landuse Typo

- Urban/ Built-Up
- Agricultural

- Non-forested lands
- Forest lands

- Water bodles

- Wetlands

 

[:1 Conversion to urban (25%)
N Roads

 

 

 

0 0.25 0.5 1

MILES

 

 

 

Figure 12. Twenty-five percent increase in conversion to urban.

69

 

 

 

 

 

 

Landuso Type [:1 Conversion to urban (30%) N
- Urban/ Built-Up /V Roads w a
- Agricultural

- Non-M '5 H mm 0 0.25 0.5 1
- Forest lands | I | |
- Wder bodies MILES

- Wetlands

 

 

 

 

 

 

 

 

 

Figure 13. Thirty percent increase in conversion to urban.

70

 

 

 

Landuse Type [:1 Conversion to urban (35%)
- new Edit-Up N Roads

 

 

 

 

“My «gap,
. g' “q 1,.
it“s“. ‘1 -','.I'
“W ‘T ‘ i
., fr,“-

.14".
:‘gg.

 

 

 

 

Figure 14. Thirty-five percent increase in conversion to urban.

71

 

 

 

 

 

 

Landuso Typo C] Conversion to urban (40%) N

 

- Urbarv Bull-Up N Roads w E
- Agricultural

NON“ law 0 025 0.5 1
- Forest lands
-Waer bodles MILES
- Wetlmds

 

 

 

 

 

 

 

 

Figure 15. Forty percent increase in conversion to urban.

72

 

 

4

7: ._. at
J

“WIN.

 

 

Landuss Type C] Conversion Io urban (45%) u
- Urban! Bill-Up N Roads V®E
[1 Agricultural s
[j Non-remind lands 0 025 0.5 1
- Forest lmds
- WIer bodies
[_-:I Wetlands

 

 

 

 

 

 

 

 

 

Figure 16. Forty-five percent increase in conversion to urban.

73

 

 

 

 

Landuse Type C] Conversion to urban (50%) g
- Urban/ Built-Up N Roads w E
- Agricultural
- Non-forested lands 0 025 0.5 1
- Forest lands
- Water bodies muss
- Wetlands

 

 

 

 

 

 

 

 

 

Figure 17. Fifty percent increase in conversion to urban.

74

APPENDIX B
Map Series #3
Five Percent Increment Maps
Fifty-five through One Hundred Percent

75

 

 

 

 

 

Landuse Type [:3 Conversion to urban (55%)

- Utban/ Built-Up N Roads g

- Agricultural

Non—forested lands W®E
- Forest lands 5

- Water bodies 0 0.25 0.5 1
- Wetlands MILES

 

 

 

 

 

 

 

 

 

Figure 18. Fifty-five percent increase in conversion to urban.

76

 

 

“militia,

3212113- *9

 

 

Landuse Type [3 Conversion to urban (00%)
- Urban! Built-Up N Roads g

- Agricultural

- Non-forested lands W$E

- Forest lands .

- “’3‘" WI” 0 0.25 0.5 1
-m|am MILES

 

 

 

 

 

 

 

 

 

Figure 19. Sixty percent increase in conversion to urban.

77

 

 

 

 

Landuse Type :1 Conversion to urban (65%)

- Urban! euiIr-up N Roads

L] Micultutal N

[j Non-forested lands $5
- Forest lands 5

- winter bodes 0 0.25 0.5 1
IL] wen-m“ MILES

 

 

 

 

 

 

 

 

 

Figure 20. Sixty-five percent increase in conversion to urban.

78

 

 

 

 

 

 

Landuse Type C] Conversion to urban (70%)

- Urban/ Built-Up N Roads g

- Agricultural

- Non-forested lands W$a
- Forest lands

- Water bodies 0 0.25 0.5 I
- Wetlands MILES

 

 

 

 

 

 

 

Figure 21. Seventy percent increase in conversion to urban.

79

 

 

: 5"“: It.
, (l .7

1:

 

 

 

 

Landuse Type (:1 Conversion to urban (7595)
- Urban! Bull-Up N Roads g
- Agncunural
- Non-forested lands “$8
- Forest lands 3
- Water bodies 0 0.25 0.5 1
- Wetlands
MILES

 

 

 

 

 

 

 

Figure 22. Seventy-five percent increase in conversion to urban.

80

 

 

3.3-L

1:} WE“ fig
., $3,“ I g kg“,

 

 

Landuse Type I: Conversion to urban (00%)
- uni-iv Beau, xv Rm g
E_] Mrlcullml
E3 Non-rum |Inds w$n
- For.“ lm
- Mt m 0 0.25 0.5
U W

MlLEs

 

 

 

 

 

 

 

 

 

Figure 23. Eighty percent increase in conversion to urban.

81

 

 

 

 

 

 

Landuse Type [:3 Conversion to urban (35%)
- Urban/ Bull-Up N Roses g
m Agrlcultural
[:1 Non-forested lends “$8
- FOI'OSI I“! s
- WlIel' bodies 0 0.25 0.5 1
u Wetlands

MILES

 

 

 

 

 

 

 

 

Figure 24. Eighty-five percent increase in conversion to urban.

82

 

 

 

 

;- wuor bodies 0 0.25 0.5 1

 

 

 

MLES

 

 

 

 

 

 

Figure 25. Ninety percent increase in conversion to urban.

83

 

 

I

f...

 

' ‘5‘. 40*": WP
I ~4 ‘
L .~. i 9,“ ’4

 

 

 

Lambs. Typo m Conversion to urban (95%)

- UM gum /\/ ROI“ N

LJ Mm w E

[ng Non-roleslod lands

- Forest lands s

- more: bodies 0 0.25 0.5 1
W

'3 MILES

 

 

 

 

 

 

 

Figure 26. Ninety-five percent increase in conversion to urban.

84

 

 

 

 

Landuso Typo [Conversion to urban (100%)
- Urban! Built-Up N Roads N
- Agnew-n! w®g
Non-forested lands 3
- Fons: [mas
-waer bodies 0 0.25 0.5 1
- Wetlands

MILES

 

 

 

 

 

 

 

 

 

Figure 27. One-hundred percent increase in conversion to urban.

85

APPENDIX C. FIVE PERCENT INCREMENT SPREADSHEET
CALCULATIONS

86

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Urbanization Increase Number of Cells (out of Change Threshold
Pereentgg 50,060) Value

0 50,060 43,698

5 52,563 42,076
10 55,066 40,501
1 5 57,569 38,894
20 60,072 37,514
25 62,575 36,239
30 65,078 34,962
35 67,581 33,862
40 70,084 32,787
45 72,587 31 .834
50 75,090 30,937
55 77,593 30,145
60 80,096 29,429
65 82,599 28,769
70 85,1 02 28,060
75 87,605 27,428
80 90,108 26,782
85 92,61 1 25,951
90 95,1 14 24,651
95 97,61 7 22,922
100 100,120 20,994

 

 

 

87

 

APPENDIX D. UCRIHS ORAL CONSENT FORM

88

ORAL CONSENT FORM
FOR MASTERS THESIS RESEARCH

Study Title: Validation of the Land Transformation Model (LTM) to Predict Land Use
Changes in Grand Blane Township, Michigan

Abstract: The issues of sprawl and suburban development have focused attention on the
necessity of sound land use planning practices through the use of growth management
techniques. One of these techniques involves the use computer modeling through the
usage of Geographic Information Systems (GIS). GIS allows communities to utilize
existing geographic data to aid collaborative planning efforts through the extrapolation,
prediction, and allocation of potential land use scenarios for a municipality. These
scenarios may then assist in amending or implementing changes in zoning ordinances in
order to realize these hypothetical futures. However, no empirical analyses have taken
into account the predictive accuracy of these models in determining future land use
patterns. In this study, Grand Blanc Township, Michigan was analyzed using the Land
Transformation Model (LTM) created by Michigan State University. Historical land use
data, circa 1980, was entered into the LTM and used to simulate land use patterns for the
year 2000, as well as projections simulating urbanization patterns at increasing
percentages across the township.

Confidentiality and Anonymity: Your privacy regarding this study will be protected to
the maximum extent allowable by law. Participation in this research project is voluntary.
You may refuse to answer specific questions or withdraw from this study at any time
without penalty. No direct quotations will be used in the aggregate data collected in this
study. All data and observations will be destroyed after study completion.

Contact Information: If you have any questions about this study, please contact the
investigator at the following:

Project Investigator: Matthew H Malone

Address: PO Box 190, Grand Blanc, MI 48439
Phone: 810-5 13-0282
Email: malone@twp.grand—blanc.mi.us

 

If you have questions or concerns regarding your rights as a study participant, or are
dissatisfied at any time with any aspect of this study, you may contact — anonymously, if
you wish — Ashir Kumar, M. D., Chair of the University Committee on Research
Involving Human Subjects (UCRIHS) by phone: (517) 355-2180, fax: (517) 432-4503,
email: ucrihs@msu.edu, or regular mail: 202 Olds Hall, East Lansing, MI 48824.

 

89

RESEARCH QUESTIONAIRRE
FOR MASTERS THESIS RESEARCH

Estimated Time of Completion: Thirty (30) minutes.

Oral Instructions for Subjects: Please examine the three maps presented to you as
output from the Land Transformation Model (LTM).

1.

The first map represents the LTM’s prediction of urbanized areas in Grand Blanc
Township in the year 2000. These predictions were made utilizing GIS data
representing 1978 land use patterns. What is your opinion on the accuracy of the
pattern of urbanization depicted on the map based upon your knowledge of Grand
Blanc Township?

The second map series represents the LTM’s prediction of urbanized areas as
projected using five (5) percent intervals in urbanization in Grand Blanc
Township. These predictions were made by extrapolation of urbanization patterns
from the year 2000 and depict future urbanization patterns at five (5) percent
intervals ranging from five (5) percent to fifiy (50) percent. What are your
opinions, insights, or comments on the pattern of urbanization depicted on this
map series for Grand Blanc Township?

The third map series represents the LTM’s prediction of urbanized areas as
projected using five (5) percent intervals in urbanization in Grand Blanc
Township. These predictions were made by extrapolation of urbanization patterns
from the year 2000 and depict future urbanization patterns at five (5) percent
intervals ranging from fifiy—five (55) percent to one hundred (100) percent. What
are your opinions, insights, or comments on the pattern of urbanization depicted
on this map series for Grand Blanc Township?

90

hill". lal.
III