university of california santa barbara...university of california santa barbara the effectiveness of...
TRANSCRIPT
UNIVERSITY OF CALIFORNIA
Santa Barbara
The Effectiveness of the Williamson Act: A Spatial Analysis
A Dissertation submitted in partial satisfaction of the
requirements for the degree Doctor of Philosophy
in Geography
by
Jeffrey Alan Onsted
Committee in charge:
Professor Keith Clarke, Chair
Professor Helen Couclelis
Professor David Carr
Professor David Cleveland
September 2007
The dissertation of Jeffrey Alan Onsted is approved.
_____________________________________________ David A. Cleveland
_____________________________________________ David Carr
_____________________________________________ Helen Couclelis
_____________________________________________ Keith Clarke, Committee Chair
June 2007
ii
The Effectiveness of the Williamson Act: A Spatial Analysis
Copyright © 2007
by
Jeffrey Alan Onsted
iii
ACKNOWLEDGEMENTS
It would have been difficult indeed to conduct the research required to complete this
dissertation without the assistance of others. Though many have offered their
valuable time and expertise, the first and most obvious person upon which to bestow
my gratitude would be my advisor and mentor, Professor Keith Clarke. Despite being
on sabbatical during the final year of my research he still made himself easily
accessible, seemingly annihilating the distance between us. His guidance, support,
and unflagging enthusiasm and positivity were bulwarks against the tide of despair
that can so often assail those of us in the trenches of their dissertation. Professor
Helen Couclelis has been another mentor of mine. She has kept me honest by
keeping me mindful of the importance of theory as well as the wealth of criticism
leveled at the seemingly ubiquitous assumption that GIS and modeling can explain all
things in the world. For his effusive comments, generous offers of time, and an
inspiring example of courage in the face of even the worst of adversity, I would like
to thank Professor David Carr. I must also thank Professor David Cleveland since he
has brought a valuable perspective beyond the field of Geography and his great
attention to detail has been invaluable in my revisions.
As far as others who have assisted me, I must thank Ms. Kristin Hart. Her brilliant
and productive assistance with the Assessor’s database editing and the creation of
PowerPoint animations proved essential. Of course, I would also like to thank all of
those public servants who have provided me with data and sound advice. Though this
iv
list is numerous I must especially single out Mr. Michael Hickey and Mr. Roland
Hill.
Other graduate students have also proved an essential resource and, in a manner only
they can provide, inspiration as well. These include Dr. Noah Goldstein, Mr. Nick
Gazulis, and Dr. Chuck Dietzel. Mr. Thomas Pingel owes my thanks for his
inexhaustible wisdom in all areas technical and his constant willingness to help me.
Mr. Michael Vergeer should also be singled out for his unswerving idealism and
commitment to teaching, which always inspired me when I was in doubt.
Of course, my family: Mom, Ron, Laura, Gerry, Tony, Angie, and all the kids have
served as an anchor of love and support all throughout graduate school. Without
them, and all my dear friends too numerous to list, I could never have made it.
Thanks also to new extended family: Wayne, Fran, and Maggie for giving me a home
away from home. As for my fiancée Kiki, I can only say that her love, loyalty, and
constant sweetness to me, especially during the tumultuous final throes of my
dissertation, has humbled me and offered me fresh hope for what love can be.
Lastly, I would like to dedicate this dissertation to my late Father, William G. Onsted.
He was the greatest man I ever knew and, though he has long since passed, I still
consult his remembered wisdom and take inspiration from his zeal for life. And since
he was also a strong supporter of education, I dare hope that he would be proud of
me.
v
Jeffrey A. Onsted
Doctoral Candidate [email protected]
805-681-9278
EDUCATION Ph.D. in Geography. UC Santa Barbara. “Effectiveness of the Williamson Act: A
Spatial Analysis” (July, 2007) M.A. in Geography. UC Santa Barbara. “SCOPE: A Modification and
Application of the Forrester Model to the South Coast of Santa Barbara County (June, 2002)
B.A. in Urban Studies and Planning with a Minor in Environmental studies. UC San Diego. (June, 1995)
EXPERIENCE Assistant Professor, Florida International University. Joint Appointment between
Department of Environmental Studies and Department of International Relations and Geography. (August, 2007 – Present)
Trainee and Graduate Student Researcher, NOAA Sea Grant Program. Through USC, working with Ms. Gail Osherenko and Dr. Keith Clarke examining retention of Coastal Zone agriculture in California. (April, 2004 – June, 2007)
Graduate Student Researcher, UC Santa Barbara, Department of Geography. Worked in the GeoVisualization Lab. Duties included: analyzing 2-D studies already created, and creating two new-3D models for use in experiments. One is a constellation of points using Vizard and the other an urban landscape using ArcScene. (September 2003-March 2004)
Teaching Assistant, UC Santa Barbara, Department of Geography. Spring 2003 - Physical Geography: Land Surface Processes (Sundbeck) Spring 2001 - Introductory Human Geography (Montello) Fall 2000 - Urban Geography (Couclelis) Spring 2000 - Introductory Human Geography (Proctor) Winter 2000 - Geography Planning and Policy Making (Couclelis) Fall 1999 - Geography of the Information Society (Couclelis)
Field Researcher, UC Santa Barbara, Department of Geography. Conducted GPS and GIS research in Belize and Guatemala with the Meso-American Research Center. Duties included reconciling different GIS layers, using GPS in the field, assisting with editing of grant proposals and reports, as well as participating in meetings and workshops with stakeholders in the Mayan Forest Region. (May 2002)
Graduate Student Researcher, UC Santa Barbara, Department of Geography.
vi
Worked with Dr. Keith Clarke, assisting him with the South Coast Outlook and Participation Experience (SCOPE) for Santa Barbara County. Duties included: porting the initial model (designed at Prescott College) from PowerSim to Stella modeling languages, reducing size and complexity of model, challenging and changing its assumptions and parameters appropriately, adding additional elements to model, preparing and presenting PowerPoint Presentations to County stakeholders, create necessary environment for model to be simulated in a user-friendly interface through the Internet (currently off-line), and the writing up of the final report. Work included the continuing improvement of the model, outreach to the community, and discovering methods of coupling this model with the Clarke Urban Growth Model (SLEUTH). This culminated in the Regional Impacts to Growth Study (RIGS), the results of which came directly from the SCOPE model. Authored half the RIGS Appendix and helped write and edit sections of report dealing with model and its results. Have appeared on local television explaining the model’s workings and its results. Continuing to provide ad hoc support for the ongoing modeling work. (June 2000 – September 2003)
Environmental Policy Analyst, Science Applications International Corporation (SAIC), Tysons Corner, VA. Environmental Health and Sciences Group. Collaborated in the construction of an EPA urban modeling questionnaire designed for distribution to the various businesses and institutions that designed these models (i.e., MEPLAN, DRAM/EMPAL,etc.) Worked with the EPA's National Agricultural Compliance Assistance Center. Duties included: Internet searches for a wide variety of information useful to the Center; assisting with creation and editing of both the Crop Sector Notebook and the Livestock Sector Notebook; many other miscellaneous tasks including the composing of reports. Assisted with the Title V Air Permit Project by visiting state environmental offices, collecting copies of permits, and entering data. Played a major role in completing the deliverable draft of EPA's Internal Audit Policy Survey. Role included participation in data entry, Q/A, Access Queries, and Report generation. Provided financial research and analytical support for Local Government Sector Profile. Worked on EPA's Oil Refinery Compliance Study. Duties included: visiting state and regional environmental protection offices to procure relevant documents; identifying and analyzing compliance information concerning the constituent refineries, and entering information into a database. (August 1997 – September 1999)
Quality Assurance Specialist, CACI (Information Technology Services and Products). Arlington, VA. Responsible for ensuring correct document analysis for a large commercial litigation suit. Assisted Project Manager with the generation of reports and provided ideas for the evolving methodology of CACI's approach to capturing and presenting the information. (January 1997 – August 1997)
Research Assistant, Renew America, Washington, DC. Worked with the Program
vii
Director to research, identify, and verify successful environmental programs. Other duties included: assisting with outreach to environmental, business, government, and community groups; and assisting with overall coordination of awards program. (June 1996 – December 1996)
Intern, San Elijo Lagoon Conservancy, Encinitas, CA. Assisted the Executive Director by taking notes at meetings, procuring and photocopying documents from the county, calling appropriate government officials to secure information involved with various issues, speaking before the City Planning Commission, and laboring within the lagoon itself (which included removing invasive species and building channel catches to reduce erosion.) (Winter 1995)
PUBLICATIONS, PRESENTATIONS, POSTERS, AND REPORTS Wu, X., Y. Hu, H. He, R. Bu, J. Onsted, and F, Xi. 2007. “Using multiple methods
to evaluate the performance of SLEUTH in the Shenyang metropolitan area.” (in Revision)
Onsted, J. and K. Clarke. 2007. “The Impact of Policy on Land-Use and Land Cover Change.” Paper, Framing Land Use Dynamics II. Utrecht, The Netherlands. April 18– 20.
Onsted, J. 2006. “California’s Coastal Zone Management Program: Retaining Agricultural Land in the Face of Urban Growth” Poster Presentation, California and the World Ocean’s Conference. Accepted but not shown. September 18.
Onsted, J. 2006. “Assessing the long-term viability of the Williamson Act in California.” The International Conference on The Future of Agriculture: Science, Stewardship, and Sustainability Sacramento, CA. August 7-9.
Onsted, J. 2006. “Suburbs: Farming on the Fringe: Can Tax Incentives Save California’s Farmlands?” The Next American City 11: 23-25
Onsted, J. 2006. “California’s Coastal Zone Management Program: Retaining Agricultural Land in the Face of Urban Growth” Presentation, Sea Grant Trainee Symposium. USC. Los Angeles, CA. April 10.
Onsted, J. 2006. “Effectiveness of a Differential Tax Assessment Program for Farmland Conservation in Tulare County, California”. Presentation, Association of American Geographers. Chicago, IL. March 7 – 11.
Onsted, J. 2004. “Effectiveness of the Williamson Act: An Exploratory Study” Presentation, Association of American Geographers. Philadelphia, PA. March 14-19. Santa Barbara Economic Community Project. 2003. South Coast Regional Impacts of Growth Study. (I authored the appendix and provided most of the modeling support)
Onsted, J. 2002. “SCOPE: The South Coast Outlook and Participation Experience.” Poster, Association of American Geographers. Los Angeles, CA. March 19-23.
AWARDS/HONORS Awarded Dangermond Travel Grant for Conference in Europe, 2007 Sea Grant Traineeship (competitive fellowship) Awardee 2004-2007 Received 4th place in AAG Geography Bowl, 2002
viii
Department of Geography Out-of-State Fee Fellowship (UCSB) 1999, 2000, 2001, 2002
Department Block Grant Fee Fellowship (UCSB) 1999, 2000, 2001, 2002 Department of Geography Teaching Fellowship (UCSB) 1999, 2000 Provost's Honor List, Muir College (UC San Diego), 1994 Offered Winslow Scholarship (University of Maryland at College Park), 1991 National Merit Scholar Commendation (High School) 1990
GRANTS NOAA Sea Grant (I helped edit proposal) (2005 – 2006) CPRC (California Policy Research Center) (I was primary author) (2004 – 2005) SPF (Shoreline Preservation Fund)(I was primary author) (2005) NIMA (National Image and Mapping Agency) (2003 – 2004) NSF, UCIME (National Science Foundation’s Urban Change Integrated Modeling
Environment) (2000 – 2003)
SERVICE Reviewer, Environmental Management Reviewer, International Journal of Geographic Information Science Reviewer, Environment and Planning B Session Chair, AAG Annual Meeting, Philadelphia, PA (2004) Guest Lecturer, Geography 176 C (GIS Design and Applications) at UC Santa Barbara (2006) Guest Lecturer, Western Kentucky University’s Geography Department. Gave lecture on Folk Culture versus Popular Culture to a Human Geography Class. (2006) Invited Lecturer, Kansas State University's Geography Department. Talk entitled, "Examining the Effectiveness of California's Land Conservation Act". (2006) Guest Lecturer, Economic Geography Class at Santa Barbara City College. (2005,
2006) Participant, in inaugural meeting of the prospective UC Environmental Politics/Policy Institute at UC Santa Cruz. (2005) Guest Lecturer, Goleta Sustainable Food Systems for the Environmental Studies Department at UC Santa Barbara. (2004, 2005, 2006) Guest Lecturer, Coastal and Ocean Law and Policy class for the Environmental
Studies Department at UC Santa Barbara (2004) Guest Lecturer, Urban and Regional Modeling and Planning class (Geography 184 C)
for Geography Department at UC Santa Barbara. (2003) Representative, for the South Coast modeling team, with presentation and
demonstration at GIS Day, held at the Santa Barbara County Government Building. Presentation was broadcast on local government television. (2002)
Lecturer. Presented hour-long presentation and demonstration of South Coast model
ix
and local planning issues to over 100 local residents during an adult education class at Santa Barbara City College. (2001)
Representative, of the Geography Department to the Graduate Student Association. (2001-2002)
Mentor. Initiated an intra-departmental mentoring program at UCSB Geography for incoming graduate students. Administered this program ever since and have personally been a mentor to five graduate students. (2001 – Present) Mentor. Helped guide students who are placed on academic contract after their first
quarter at UCSB. Duties included assisting with self-assessment, motivation skills, time organization, positive thinking, future goals, and assistance in accessing campus resources. (2001)
Volunteer, with Campus-Wide Orientation for incoming graduate students at UCSB. Entertainment Committee, for Geography Department. (2000 – present) Volunteer, with Summer Bridge program for incoming freshman from under-
represented high schools at UCSB. (2000)
AFFILIATIONS Association of American Geographers
-Coastal Specialty Group -GIS Specialty Group -Graduate Student Affinity Group -Population Specialty Group -Regional Development and Planning Specialty Group -Rural Geography Specialty Group -Spatial Analysis and Modeling Specialty Group -Urban Geography Specialty Group North American Cartographic Information Society American Congress on Surveying and Mapping National Geographic Society Subscriber: Journal of Land Use Science
x
ABSTRACT
The Effectiveness of the Williamson Act: A Spatial Analysis
by
Jeffrey Alan Onsted
The Williamson Act is the flagship conservation program protecting California's
world-renowned farmland. This act, though, is an incentive-based, voluntary
program that is easily compromised by the landowners' prospects of cashing in on
development dollars. This differential tax assessment method for protecting land
exists nationwide but it still cannot compete with the vast amounts of money that
developers can offer these landowners. This temptation is exacerbated by difficulties
farmers on the periphery of urban areas already face. These include suburban
complaints of their operations, trespassers, and marauding dogs and cats, to name a
few. This paper outlines just how effective the Act has been in protecting these
farmlands near urban areas by tracking parcels’ entry into and exit from this Act in
the path of urban expansion, then using this data to create a future scenario of
xi
Williamson Act lands using a cellular automata (SLEUTH) model. These results
create a probabilistic future land availability landscape that is used, in turn, as an
excluded layer for urban growth scenarios in the future.
By using SLEUTH’s built-in metrics and comparing them with similar modeling
efforts, the value of integrating Williamson Act data, rather than ignoring it or
treating it coarsely, is demonstrated. Exploring different urban growth scenarios in
the future that correspond to Williamson Act policy options offers not only image
outputs but additional metrics as well. This innovative use of SLEUTH allows for a
more robust urban growth simulation and also allows policy makers to explore
various angles of Williamson Act implementation. For instance, though continuing
the current method of Williamson Act administration still results in continual loss of
farmland, abolition of the Act results in far more. Scenario exploration, therefore,
suggests that permanent contracts offer the best protection for farmland. This
dissertation has applications to any geographic region employing differential
assessment programs and can lead to further exploration of exclusion layer
forecasting as well as concomitant coupling opportunities with traditional urban
models.
xii
TABLE OF CONTENTS
I. Statement of the Problem ......................................................................................... 1
A. Significance of Project .......................................................................... 1
B. Project Outline ...................................................................................... 4
1. Research Questions……………………………………..…4
2. Methods……………………………………………………5
II. Context of the Study................................................................................................ 8
A. Historical Land Use…………………………………………………...8
B. California Farmland………………………………………………….10
C. Farmland Protection Program………………………………………..18
D. Williamson Act Implementation and Eligibility……………………..22
E. Williamson Act Criticism……………………………………………27
F. Theories of Growth…………………………………………………..30
G. Modeling Theory and Criticism……………………………………..35
H. Modeling Approaches……………………………………………….39
I. SLEUTH Applications………………………………………………47
III. Methods and Data………………………………………………………………59
A. Data Background…...………………………………………………59
B. Data Acquisition……………………………………………………61
C. Data Rendering……………………………………………………..66
D. Modeling…………………………………………………………...70
E. Coefficients……………………………………………………….…78
xiii
F. Growth Rules………………………………………………………..80
G. Self-Modification Parameters………………………………………82
IV. Results………………………………………………………………………….87
A. Part 1: (Text)………………………………………………………..87
1. Introduction…………………………………………..….87
2. Examining the Past………………………………………89
3. Williamson Act Forecasting………………...…………..91
4. Urban Forecasting…………………………..…….……..97
5. Conclusion…………………………………..…………..104
B. Part 2 (Maps, Figures, and Tables)…………………………………105
1. Part 1:Exploratory Maps………………………………..106
2. Part 2: Input Images…………………………………….113
3. Part 3: Output Images…………………………………...117
4. Scenario Images…………………………………………127
a) Tulare…………………………………………….127
b) StanMerc…………………………………………134
V. Conclusions........................................................................................... ..……...141
A. Research Questions Revisited…………………………………..…141
B. Future Research…………………………………………………….145
References………………………………………………………………………….149
xiv
LIST OF FIGURES
Figure 2-1. FMMP Survey Area-2002 ................................................................... ..14
Figure 2-2. Number of Farms and total agricultural acreage in CA: 1950-2005....... 17
Figure 2-3. California Government Code Section 51220…………………………....23
Figure 3-1. Uncertainty in urban models over time………………………………….63
Figure 3-2. FMMP Definitions of Important Farmland Categories………………….66
Figure 3-3. FMMP Definitions of Important Farmland Categories………………….72
Figure 3-4. Metrics output for prediction runs using SLEUTH……………………..86
Figure 4-1. FMMP Survey Area-2002 ................................................................. ..105
Figure 4-2. Tulare excluded.wac ......................................................................... …113
Figure 4-3. Tulare.urban.2002.wac.…………………….………………………….114
Figure 4-4. Tulare.excluded.c.……………………………………………………..115
Figure 4-5. Tulare.excluded.nowac……………….………………………………..116
Figure 4-6. 2003 Tulare County Williamson Act run………………………………117
Figure 4-7. 2030 Tulare County Williamson Act run ………..…………………….118
Figure 4-8. Excluded.bauc………………………………………………………….119
Figure 4-9. Stanmerc.Excluded.Wa………………………………………………...120
Figure 4-10. Stanmerc.urban.2002.wa……………………………………………...121
Figure 4-11.
Stanmerc.excluded.new……………………………………………….122
Figure 4-12. Stanmerc.excluded.nowanew………………………………………...123
Figure 4-13. StanmercWA 2003……………………………………………………124
xv
Figure 4-14. Stanmerc.wa.2030…………………………………………………….125
Figure 4-15. Stanmerc.excluded.baunew2............................................................... 126
Figure 4-16. a) Tulare in 2003…………………………...…………………………127
b) Strict Adherence to WA, 2030……………………………………..127
c) Business As Usual, 2030…………………………………………...128
d) Abolition of the WA, 2030……………………… ……………...…128
Figure 4-17 a) Tulare County WA Classification, 2002……………………………130
b) Tulare County WA Classification, 2030 .......................................... 130
Figure 4-18 a) Tulare County Land Use, 2002........................................................ 131
b) Tulare County Land Use, Strict Adherence to the WA, 2030………131
c) Tulare County Land Use, Business As Usual, 2030…..……..……...132
d) Tulare County Land Use, Abolition of the WA,
2030…………..…..132
Figure 4-19. Tulare County Land Converted to Urban by type and
by scenario, 2030………………………………………………………………133
Figure 4-20. a) Stanislaus and Merced Counties, 2003.....…………………………134
b) Strict Adherence to WA, 2030…………………………………….134
c) Business As Usual, 2030…………………………………………..135
d) Abolition of the WA, 2030……………………… …………….….135
Figure 4-21 a) Stanislaus and Merced Counties WA Classification, 2002………..137
b) Stanislaus and Merced Counties WA Classification, 2030 ............ 137
Figure 4-22 a) Stanislaus and Merced Land Use, 2002..................................... .…138
b) Stanislaus/Merced Land Use, Strict Adherence to the WA, 2030...138
xvi
c) Stanislaus/Merced Land Use, Business As Usual,
2030…..……….139
d) Stanislaus/Merced Land Use, Abolition of the WA, 2030………….139
Figure 4-23: Stanislaus and Merced Counties Land Converted to Urban by type and
by scenario, 2030………………………………………………………140
LIST OF TABLES
Table 2-1: FMMP Acreage Changes California: 2000-2002..................................... 17
Table 2-2: Williamson Act Enrollment Acreage ....................................................... 29
Table 3-1: WA Modeling Layers............................................................................... 69
Table 3-2: Urban Modeling Layers ........................................................................... 70
Table 3-3:Metrics That Can Be Used to Evaluate the Goodness of Fit of SLEUTH .75
Table 3-4: Routines and Results for Calibrating SLEUTH for Tulare County……...76
Table 3-5: Stanislaus and Merced Counties…………………………………………77
Table 4-1: Summary Statistics Table for Tulare County......................................... 129
Table 4-2: Summary Statistics Table for Stanislaus / Merced Counties ................. 136
LIST OF MAPS
Map 2-1: Visalia 1986 Metropolitan Area Land Use (Tulare County) ..................... 18
Map 2-2: Current and Former Williamson Act Lands in Visalia Metropolitan Area.26
Map 4-1: Western Tulare County, 2002 .................................................................. 106
Map 4-2, Visalia and Tulare 1986 ........................................................................... 107
xvii
xviii
Map 4-3, Visalia and Tulare 2002………………………………………………….108
Map 4-4: Stanislaus and Merced Counties, 1984…………………………….……109
Map 4-5: Stanislaus and Merced Counties, 2002………………………………….110
Map 4-6: Modesto Metropolitan Area, 1984 ........................................................... 111
Map 4-7: Modesto Metropolitan Area, 2002 ........................................................... 112
Chapter 1: Statement of the Problem
Significance of Project
When Americans reflect upon California and its economy, thoughts usually turn to
Hollywood and the high-tech industries of Silicon Valley, rather than to the State’s
acres of avocados and endless miles of cattle ranches. Many visualize the Midwest as
the nation’s most important agricultural region, but it is California that is the most
agriculturally productive state in the union. In spite of this, California also has one of
the fastest growing populations. Given these two colliding realities, the protection of
California’s agricultural lands becomes paramount. Although there are a number of
Federal, State, and local programs in place to protect farmland in California, the
Williamson Act is responsible, by far, for the most acreage. But how effective has this
land conservation act been over its 40-year history? This research seeks to more fully
understand the impact the Williamson Act has had on the protection of agricultural
lands over time, and also to assess the Act’s continuing long-term viability in the face
of tremendous population pressure.
As far as permanent protection measures go, California does not lead the way in
agricultural conservation easements. That is why it is important to investigate the
effectiveness of the Williamson Act. If it fails to really halt urban sprawl and protect
farmland then nearly all of the
1
State’s farmland could eventually be at risk. Also, since nearly every state uses
something similar to the Williamson Act1 it is vital to discover just how effective
voluntary, incentive-based programs are in a general way. Of course, every locality
offers a different zoning and ordnance setting and this should be addressed when
trying to analyze the act. It should be noted, however, that current zoning may make
less difference when predicting growth than one might assume (See Chapter 3).
It is also essential to explicitly examine just what the California legislature’s
reasoning is concerning the formation, promulgation, and enforcement of the
California Land Conservation Act. After all, one must be sure what the Act was
intended to accomplish before assessing its success or failure. Under Section C of the
Act’s declaration (See Figure 2-3), the legislature discourages the “premature and
unnecessary conversion of agricultural land to urban uses.” Therefore the
Williamson Act never intended to halt all development on agricultural lands.
Nevertheless, the wording of the Act, “premature and unnecessary”, is placed sphinx-
like before policy makers both state and local, as well as academics because,
ultimately, this vague raison d’etre must be realized through the specific policy
provisions of the Act as well as oversight and regulation. As the State is not the true
implementer of the Williamson Act, this is interpreted differently in different counties
and at different times. This dissertation, consequently, is not equipped to answer the
question: “Has the Williamson Act prevented the premature and unnecessary
1 Referred to by the American Farmland Trust as “Differential Assessment” programs (AFT website).
2
conversion of agricultural lands to urban uses?” This is a normative judgment and,
therefore, unanswerable in a quantitative sense. What this dissertation aims to
achieve is a framework of past maps and future scenarios through which others,
perhaps even policy makers, may explore this normative question. If they find the
answer troublesome, then perhaps they may initiate change.
A thorough investigation of the Williamson Act bears much more than academic fruit
since, periodically, the State considers whether or not to terminate its subvention
payments to the localities enrolled. This is particularly tempting given the current
belt-tightening climate2. If subvention payments cease, the financial burden for
continuing the program falls on the local governments, many of which are unlikely to
continue it unsubsidized. Also, since there is a great deal of land and money at stake,
any policy decision concerning the act should be as informed as possible. Currently,
there are about 16.6 million acres under Williamson Act contract, well over half of all
private agricultural land. Although this is actually more acreage than was in the act
ten years ago, that fact alone can be misleading. Understanding the geographic
history of adoption and withdrawal from the act offers a more vivid picture (See Map
2-2). Though many new parcels have joined, many others have left and some
developments even sit atop former Williamson Act lands. The several counties
studied in this research reflect this, as much of the prime irrigated farmland tends to
3
be on the flat plains surrounding the burgeoning urban areas and therefore lies in the
path of development. This presents a conundrum for the administration of the
Williamson Act. First, those areas most likely to be developed, ceteris paribus, are
those areas near other developed areas3. Second, Prime Farmland tends to be closest
to these developed areas and, thus, are the very parcels in the greatest danger
(Kuminoff, et. al., 2001; Sokolow and Spezia, 1992). Third, those Williamson Act
parcels nearest to developed areas are more likely to non-renew4 (See Map 2-2). So,
if all this is true, then the Williamson Act is best at protecting those parcels that are
farthest from developed areas which are not only under less threat of development but
tend to be less valuable lands in terms of their agricultural worth.
Project Outline
Research Questions
With the preceding in mind, the following research questions and hypotheses will
guide this dissertation.
• Question: Is the Williamson Act (both specifically and as representative of
other differential assessment programs) useful for modeling of urban spread?
2 At the time of this writing (2007) the Governor of California has offered a revised budget to the CA Legislature that suggests removal of the $39 million in Subvention payments to local governments, an act that would all but destroy the Williamson Act. 3 See SLEUTH Model publications: E. A. Silva and K. C. Clarke, 2002; Clarke, K. C., and L. Gaydos, 1998; Clarke, K. C., Hoppen, S. and L. Gaydos, 1997 4 Although not always for reasons of development, see California Department of Conservation, “The Williamson Act: 1991-1993 Status Report”, California Resources Agency, July, 1994.
4
• Hypothesis: The Williamson Act’s inclusion in urban growth modeling results
in greater accuracy than its exclusion.
• Question: Do spatial variables predict a parcel’s removal from the WA?
• Hypothesis: The same geographic phenomena that predictably apply
development pressure on undeveloped lands also apply pressure on
landowners to leave the Williamson Act.
• Question: How can knowledge of these spatial variables be used to model the
future of WA termination?
• Hypothesis: By designating Former WA lands as urban for the purposes of
SLEUTH’s nomenclature, a future landscape of the WA can be forecast with
accuracy comparable to urban growth forecasting.
• Question: How can WA forecasts be used to influence urban growth
modeling?
• Hypothesis: A probabilistic excluded landscape can be created in a WA
modeling run and fed into a traditional urban modeling routine.
Methods
To answer these questions, the following tasks were performed:
5
► A comparison of the development patterns historically for three counties:
Tulare, Stanislaus, and Merced (See Figure 2-1 for reference).
By populating a Geographic Information System with maps from the California
Department of Conservation’s Farmland Mapping and Monitoring Program (FMMP)
the net growth of urbanized lands and other metrics were evaluated including: edges,
clusters, cluster sizes, and spread. This shed much needed light on the effect the
Williamson Act has on urban growth. These maps also reflect the value of the
farmlands in these counties according to the FMMP’s hierarchical classification
system: Prime Farmland, Farmland of Statewide Importance, Unique Farmland,
Farmland of Local Importance, and Grazing Land.
► An analysis of each county to determine what effect urban growth has on the
adoption, non-adoption, non-renewal, and cancellation of parcels in the
Williamson Act
► Creation of a tool that will allow the Department of Conservation, the State
Lawmakers, as well as local interests and academics to see animated maps of
different farmland conversion scenarios according to various policy options.
This research offers both methodological and practical advances. First, this work has
successfully forecast future Williamson Act patterns so that appropriate
administrators may see the direction in which this voluntary measure is heading in
their respective jurisdictions. Second, building upon the former, this dissertation
utilizes a cellular automata model to create a probabilistic excluded layer that can
6
then be used for urban growth modeling. Such a technique has, to the author’s
knowledge, never been performed.
In the chapters that follow many of the points mentioned in this Introduction will be
expanded. In Chapter 2, a thorough contextualization is offered so that the audience
may see the evolution of theories and techniques that have lead to this dissertation as
well as the role it plays in the field overall. The third Chapter offers an account of the
methods applied and the data utilized for the realization of this research. Chapter 4
engages the specific results of the work along with pertinent additional analysis.
Finally, the fifth Chapter examines how effective this dissertation has been in
answering its research questions and what the implications of this project are for
future modeling investigations as the future of the Williamson Act.
CHAPTER 2: Context of the Study
7
Historical Land Use
Urban sprawl is a relatively new phenomenon. For several thousand years, cities and
towns, at least in Europe, were lonely islands in the broad swaths of countryside that
abruptly appeared outside the city gates. This is, of course, no longer the case
everywhere, particularly in the US. During the last several hundred years and at an
accelerating pace, cities have been advancing and the farmland and open spaces have
been falling back (Vitousek et. al., 1997). Ramankutty and Foley report that, world-
wide, 2.6 million square miles of natural lands have been converted to other uses
(mostly agricultural) over the last few centuries (1999). The détente between city and
country is over. The war though, if it may be called that, is not. There remains a
great deal of agricultural land in the world and developed areas still make up only a
tiny fraction of the Earth’s surface, though much of the world’s land is neither
picturesque nor of much direct use to human beings. Consequently, flat well-drained
lands, comprising only a modest proportion of the Earth’s land, are in direct
competition with expanding cities since these lands are also ideal for development.
Though agriculture has been responsible for the greatest land cover change, altering
what was once wildlands and forest, in the last one hundred years there has not only
been a tremendous increase in population but also a reorienting of the world’s
population from predominantly rural to half urban (Knickerbocker, 2007). This shift
has had a corresponding effect on land use. Cities have expanded and built upon the
lands that formerly surrounded them. This transition is particularly dramatic in the
8
United States, where roughly 80% of the population now lives in urban areas
(Acevedo, 1999). This was not always so.
Since colonial times, the Eastern portion of the US has undergone several
transformations. First, its forests gave way to farms that were settled by early
European pioneers. Later, many of these farms were reclaimed by forests as
populations became more dependent on secondary and tertiary economic activities
and agriculture migrated westward. This is particularly true in New England and
other portions of the Northeast (New York Department of Environment
Conservation). The western portion of the US, on the other hand, still exhibits a
landscape that reflects a great deal of primary economic activity. In fact, agricultural
development has altered much of western North America. For instance, the draining,
irrigation, and damming of California’s Central Valley has transformed it from
wetlands to now mostly croplands (Golden State Museum).
As the remaining arable lands are plowed or protected as open space, the days of vast
agricultural expansion in California are drawing to a close. However, finding the right
balance of farmland, wildlands, open space, and developed land is the new challenge.
There is disagreement, obviously, on just how to achieve this. Many contend, as
discussed later, that it is of the greatest urgency that we protect farmland, particularly
in California, while others claim the rate of conversion is so slow there is no reason
for panic (Plaut, 1980; Fischel, 1982; Gustafson and Bills, 1984; Kuminoff et. al,
2001).
9
California Farmland
California’s vanishing farmland is not an unpopular topic of study. With its 2002
market of agricultural products valued at nearly $26 billion, California is the most
productive agricultural region in the country, if not the world. It is not surprising,
therefore, that the conversion of these fertile lands to urban uses has many groups--
from policy-makers and non-profit organizations to academics and private citizens--
alarmed, and with good reason. The state has already lost well over 11 million acres
of farmland since its peak in the 1950’s5 6. What these numbers don’t tell us,
however, is the way agriculture has been redistributed throughout California in the
last 50 years. After World War II, a building boom began squeezing farmland out of
the coastal regions into new lands colonized in the Central Valley (Sokolow and
Spezia, 1992). This boom continues today and now even threatens this agricultural
cornucopia, since recent studies suggest the Central Valley’s ability to keep the
agricultural economy afloat will be greatly challenged by tremendous population
pressure and low density sprawl (Teitz, et. al., 2005). Unfortunately, as this new
farmland disappears there is no new valley to rescue California’s agricultural
economy (Sokolow, 1997).
More input-intensive farming methods and greater labor availability have
compensated for some of this lost acreage (Medvitz, 1999). However, it is now being
5 But not a concomitant loss in agricultural productivity, see Kuminoff et. al, 2001 and Medvitz, 1999.
10
suggested that productivity will eventually decline as the best farmland continues to
be nibbled away by sprawl and as water becomes an increasingly scarce commodity
(Medvitz, 1999). In fact, new farming operations have become impossible to begin in
some areas of the state since the cost of installing a new water meter has become
prohibitively expensive (Lane, 2005).
Also, farming on the urban fringe is more difficult than in the hinterlands and this
exacerbates the problem. Homeowners complain about farming noises, odors, and
chemicals, while farmers have to deal with trespassers, theft, and mischievous pets
(Handel, 1999).
Even with Right-to-Farm ordinances in place, constant bickering with neighbors is
often the last straw for edge farmers (Hvolboll, 2005). This disconnect has much to
do with expatriated urbanites valuing farmland for its open space aesthetic rather than
for its more utilitarian functions as a place of business and food production. It has
been observed, for instance, that the suburban public raises the alarm about the loss of
farmland not for economic or food security issues but for the open space that these
vanishing parcels afford them (Logan and Molotch, 1987).
The money gained by selling their farms to developers coupled with the difficulty
involved in farming on urban peripheries act as both carrot and stick to pull and push
6 The total loss of farmland is greater since this is only net loss (lost agland – new agland)
11
farmers over the line and into selling their land. As Eric Hvolboll (2005), a Santa
Barbara County avocado farmer and eighth generation Californian put it, “There are
not many farmers so invested in their work that they can justify the opportunity cost
of not selling…Why risk your life, doing a dangerous job, working 60 hours a week,
when you could make more money doing nothing?”
To be fair, though, not all farmers near urban areas are giving up on their trade, even
in Santa Barbara County. In urban Goleta, for instance, there are a number of
working farms still in business. Though none are in the Williamson Act, by
intensifying their agricultural practices in the form of more greenhouses and less row
crops, or by direct marketing, many of the farmers have been able to maximize the
use of their land and stay afloat financially (Santa Barbara County Planning and
Development, 2002). Another advantage pointed out by both the owners and the
managers of the farms has been the low transport costs and a proximal advantage to
the many farmers’ markets stretching along the Coast from Santa Barbara to Los
Angeles (Ibid). Fairview Gardens, one of the oldest organic farms in Southern
California, now surrounded by development, is a conservation easement that not only
proves that urban agriculture can exist but actually utilizes its twelve acres as an
educational farm, teaching the public about the benefits of urban agriculture.
Nevertheless, even these savvy and persistent edge farmers admit their farms may not
last the test of time (with the exception of Fairview Gardens) and cite development
pressures as a constant concern.
12
This pressure is not confined to California, but is a problem throughout much of the
Country. To systematically examine the issue, the American Farmland Trust (AFT)
has identified the 127 Major Land Resource Areas (MLRAs)7 containing high quality
farmland in areas of rapid development (Sorensen et. al, 1997). AFT also ranks these
MLRAs according to an index that takes into account both the area’s agricultural
value as well as local development pressure. California, unfortunately, has three
MLRA’s that are among the top 20 in this index. These include the Central Valley
(considered number one), the central California Coastal Valleys (15th), and Imperial
Valley (17th). The most recent data show an average yearly loss of 260,000 acres
between 1992 and 20028. (National Agricultural Statistics Service (NASS))
Furthermore, recent modeling efforts predict the Central Valley’s population will
triple by 2040, with as much as 1,035,000 acres of additional farmland converted to
urban uses, including over 600,000 acres of prime or statewide important quality
land9 (Bradshaw et. al., 1998. See Figure 2-1 for an overall look at California’s land).
Figure 2-1: FMMP Survey Area-2002 (CA Dept. of Conservation)
7 The US Department of Agriculture has divided the US into 181 Major Land Resource Areas (MLRAs) 8 This covers conversions of every type, not just urbanization. 9 Farmland of statewide importance is the second most valuable farmland type, according to the USDA’s classification system. This category is similar to prime but with minor defects. Go to http://www.consrv.ca.gov/DLRP/fmmp/mccu/map_categories.htm for more details.
13
It should be noted, though, that the majority of agricultural parcels in California are
not currently on the urban fringe and are consequently not in imminent danger of
conversion to urban uses (Sokolow and Spezia 1992; UC Davis 1989; Kuminoff, et.
al., 2001). In fact, at first glance the urbanization rates appear so low that some
argue there may really be no urgency at all. From 1992 to 2002 there were
14
approximately 280,000 acres converted from agricultural uses to urban uses10 (CA
Dept. of Conservation). Considering that this is out of a 1992 pool of 30 million acres
of agricultural land11, that is an average annual conversion rate of 0.093 percent. If
this is in fact the case, is the Williamson Act, or any agricultural preservation
measures, needed at all? The answer seems to be yes, for two reasons. First, these
values don’t suggest the acceleration over this decade in urbanization. The period
1992 to 1994 saw 30,000 acres paved over while the 2002-2004 figures report 94,000
acres converted, a three-fold increase in rate (Ibid. See Table 2-1 and Figure 2-2
below). This is an alarming trend for those with an interest in California’s agricultural
future. Second, most of the best farmland rings urban areas and is in the direct path of
expanding land-hungry cities (See Map 2-1). California's Farmland Mapping and
Monitoring Program's statistics reveal that over 85,000 acres of prime farm land were
urbanized during this decade and, given a base of 4.3 million acres in 199212, prime
land averaged an urbanization rate of nearly 0.2 percent a year, almost three times
faster than the rate for more remote grazing lands (Ibid). Therefore, those farms
closest to urban centers, which tend to be the most prime and productive, are
disproportionately threatened by urban sprawl than more distant farms.
10 There is no universally agreed upon figure. The FMMP and the NASS give different figures for California. Also, there are other categories to which agricultural lands can be converted besides urban and these are not reflected here. 11 NASS’s 1992 figure. Since FMMP does not inventory all lands the NASS figure for total farmland is given instead. FMMP’s total comes out to 26 to 28 million acres. 12 FMMP’s subcategory figure.
15
While the urbanization of farmland is arguably the most irrevocable conversion, the
overall transition of farmland to other uses (including non-productive lands, open
space, or extreme low-density rural developments not considered “urbanized”) has
removed a great deal of these lands from use in the last 50 years. During the 1950s
(1950 – 1960) there was actually an increase of 1.3 million acres of agricultural lands
in California as the investment of new farmlands actually eclipsed the elimination of
others. The ‘60’s saw a reversal of this trend as California experienced an average
loss of 220,000 acres a year. The 1970’s saw even more eradication, with 2.8 million
acres lost. This trend accelerated in the 1980’s as 300,000 acres a year were, on
average, removed from agriculture. The nineties saw a slight deceleration, with 2.8
million acres lost. However, as the first decade of the 21st century has progressed,
there has been an average of 320,000 acres lost per year, on track to be the most
devastating decade for farmland removal since records have been kept (See Figure 2-
2 for all years from 1950 to 2005).
Table 2-1: FMMP Acreage Changes California: 2000-2002
16
Figure 2-2: Courtesy of NASS
Number of Farms and total agricultural acreage in California: 1950-2005
0
20,000
40,000
60,000
80,000
100,000
120,000
140,000
160,000
1950
1956
1962
1968
1974
1980
1986
1992
1998
2004
Number of farmsAcres (in thousands)
Map 2-1: Visalia 1986 Metropolitan Area Land Use (Tulare County)
17
Farmland Protection Programs
There are a number of arguments for the preservation of farmland in any locality.
First, prime farmland is disappearing across the Earth due both to poor farming
practices, which erode topsoil, as well as from expanding land-hungry cities. Urban
areas are often found in flat, well-drained areas that are ideal for agriculture as well,
causing competition between the two uses. Underdeveloped nations are, tragically,
expected to become ever-increasing consumers of US agricultural products as this
18
dearth becomes compounded by staggering population growth, another economic and
moral incentive to protect our farmland. Also, with dangerous diseases such as the
Avian Influenza looming on the horizon, food security at multiple geographic scales
seems prudent. Third, farms tend to cost local governments less than the taxes levied
on them while housing developments cost more, given the high cost of infrastructure
and services that must be provided (California State Assembly, 2001). Fourth, farms
often provide an attractive viewshed and can even promote tourism, whether it’s the
wine country of California or the Amish country of Pennsylvania. Fifth, farms are
places of business and they provide jobs as well as support ancillary services such as
agricultural equipment suppliers, veterinarians, feed and fertilizer consultants, even
accountants. Lastly, though they use a tremendous amount of water and oftentimes
pesticides as well, many farms provide services to the environment by protecting
topsoil, providing open space, preventing runoff, and even acting as critical pieces of
habitat, services not easily rendered by yet another big box mall (Pennsylvania
Farmland Preservation Association). In fact, Boody et. al. (2005) have found that not
only does agriculture have less of an economic impact on the taxpayer but, when
employing environmentally friendly techniques, can improve the local ecology
without incurring additional expenses, in some cases even improving farm
profitability.
Despite the relatively recent trend toward urban sprawl, agricultural conservation has
been pursued for decades, beginning during the Depression when the Federal
Government’s 1936 Soil Conservation Act took an active interest in protecting
19
farmland (USDA website). The 1960’s saw many state governments following suit,
including California. Today, conservation measures extend from the federal level all
the way down to the local level. A short tour of them allows for a greater
understanding of the policy context within which the Williamson Act functions.
At the Federal level a suite of programs under the aegis of the 2002 Farm Bill form
the central source of money. Rather than being hands-on, this bill instead offers
money to those who apply for conservation easements. The Farmland Protection
Program, for instance, provides $597 million in matching funds to qualified entities
(states, counties, etc.) over 6 years. The previous Funding Bill from 1996 only
provided $52 million. There are also the Wetland Reserve and the Grasslands Reserve
programs of the new Farm Bill that provide funds for farmers to protect these
portions of their property. Although there are other aspects of the Farm Bill that assist
farmers in their environmental efforts, these are the only programs designed to
directly and permanently protect farmland and other sensitive private lands. Whatever
the reasons, though, for the promulgation of farmland conservation measures, those
that are in place must actually accomplish their goals to justify their expense.
California has several state programs of its own that are designed to conserve
farmland. The first, Purchase of Agricultural Conservation Easements (or PACE), is a
loose description of many different states’ attempts to protect farmland directly by
paying farmers to permanently preserve their land. California’s version at the state
level is called the California Farmland Conservancy Program, or CFCP, and it has
20
yielded roughly 24,000 acres thus far (California Department of Conservation (CDC)
website). Similar efforts at the local level have acquired approximately 53,000 more
(AFT website). This may seem like a good deal of acreage, but when put in contrast
with California’s total private farmland acreage of 30 million acres13, this provides
only about 0.26 percent guaranteed perpetual protection of California’s agricultural
lands. Some acreage is counted twice, as it receives both local and state funds, taking
that percentage down even further. In contrast, the state of Maryland in 2003 had 2.1
million acres of farmland (NASS) with 282,000 acres in a similar state-level program
and 209,000 acres held in local programs, (AFT website) yielding perhaps as much as
23 percent permanent protection (Ibid). California’s Williamson Act, on the other
hand, covers a great deal more acreage than the CFCP but, as shall be seen, this
protection is not permanent.
Williamson Act Implementation and Eligibility
The California Land Conservation Act of 1965 (See Figure 2-3), better known as the
Williamson Act, permits local authorities to enter into contracts with private
landowners with the goal of restricting the land use of specific parcels to agriculture
or open space. Under the Act, landowners receive lower use-based property tax
13 NASS’s 1992 figure. Since FMMP does not inventory all lands the NASS figure for total farmland is given instead. FMMP’s total comes out to 26 to 28 million acres.
21
assessments and not the full market value of the land. Local governments then receive
an annual compensation for lost tax revenues from the state via the Open Space
Subvention Act of 1971. Since the financially strapped State periodically examines
the Act for elimination,14 it is pertinent to ask the questions: what are the benefits of
the Williamson Act; have the conservation goals of the Act been met (in light of the
caveats above); and how does this relate to where urbanization and land development
have taken place in the State? More broadly, what is the long-term viability of
voluntary incentive-based conservation programs in the face of inexorable
development pressure? Since they are not unique to California’s policy culture and, in
fact, are used in many of the United States (CDC, 2003), a thorough knowledge of the
specific geographical manifestations of these applications of policy is important.
These effects are difficult to predict, though, since the Williamson Act attempts to
conserve land not through coercion, but through incentives. Nevertheless, as Chapter
3 will demonstrate, though it is difficult to predict a future Williamson Act regulatory
landscape, this dissertation offers an important contribution in the form of a
methodological solution for not only forecasting the Williamson Act but also using
these results to affect urban growth modeling.
Figure 2-3, from California Government Code Section 51220:
51220. The Legislature finds: (a) That the preservation of a maximum amount of the limited
14 At the time of this writing (2007) the Governor of California has offered a revised budget to the
CA Legislature that suggests removal of the $39 million in Subvention payments to local governments, an act that would all but destroy the Williamson Act.
22
supply of agricultural land is necessary to the conservation of the state's economic resources, and is necessary not only to the maintenance of the agricultural economy of the state, but also for the assurance of adequate, healthful and nutritious food for future residents of this state and nation. (b) That the agricultural work force is vital to sustaining agricultural productivity; that this work force has the lowest average income of any occupational group in this state; that there exists a need to house this work force of crisis proportions which requires including among agricultural uses the housing of agricultural laborers; and that such use of agricultural land is in the public interest and in conformity with the state's Farmworker Housing Assistance Plan. (c) That the discouragement of premature and unnecessary conversion of agricultural land to urban uses is a matter of public interest and will be of benefit to urban dwellers themselves in that it will discourage discontiguous urban development patterns which unnecessarily increase the costs of community services to community residents. (d) That in a rapidly urbanizing society agricultural lands have a definite public value as open space, and the preservation in agricultural production of such lands, the use of which may be limited under the provisions of this chapter, constitutes an important physical, social, esthetic and economic asset to existing or pending urban or metropolitan developments. (e) That land within a scenic highway corridor or wildlife habitat area as defined in this chapter has a value to the state because of its scenic beauty and its location adjacent to or within view of a state scenic highway or because it is of great importance as habitat for wildlife and contributes to the preservation or enhancement thereof. (f) For these reasons, this chapter is necessary for the promotion of the general welfare and the protection of the public interest in agricultural land.
To summarize Figure 2-3 above, the Williamson Act allows both the state and the
parcel to “enter into a contract in which each accepts certain costs in return for other
benefits” (USDA website). Enrollment, though voluntary both for the locality as well
as the landowner, is much easier than a withdrawal. For example, in a participating
county or city, a potential candidate need only file an application with the local
government, usually the county planning department. Then, if the land is eligible, it is
usually accepted. Under normal conditions the contract automatically renews each
23
year. To withdraw from the program requires an initiation of non-renewal on the part
of the landowner, setting into motion a nine-year phase out period whereby taxes are
gradually shifted back from use value to market value rates.
All agricultural lands in the state are eligible for Williamson Act contracts, provided
the landowners have enough acreage to meet the minimum parcel requirements and
the county (or city) has designated the land under which the parcel falls as an
agricultural preserve. Every county also has its own rules and regulations regarding
the size and type of land that is eligible. However, the State mandates (except in very
unusual circumstances) that each parcel not only be located in an area zoned for
agricultural preserves by the local government but that these zones must also be no
less than 100 acres. This acreage can be counted across two or more contiguous
parcels or even two non-contiguous parcels as long as they are under the same
ownership.15 It is further stipulated that within these preserves prime agricultural land
must be in parcels of no less than 10 acres and non-prime land in parcels of no less
than 40 acres. However, additional open space lands are also acceptable for contracts.
Rather than carefully designated, Agricultural zones are usually assigned ad-hoc as
various farmers request Williamson Act exemptions (Ken Trott, personal
communication). As far as subvention payments, the State does not simply pay back
the exact amount that the locality has lost. Rather, the local governments are
compensated according to a uniformly standard and simple system: $5 per acre for
15 For more information see the following website:
24
prime land16 outside the three mile sphere of influence for each incorporated city and
$1 for each non-prime acre of land outside this sphere; and $8 per acre for all
Farmland Security Zones within the incorporated area or its three mile sphere of
influence (CA Government Code Section 16140-16154). The totals these
subvention payments bring to counties rarely, if ever, equal the tax revenue these
local governments have foregone. California counties are relatively fortunate,
however, since most states offer no subventions whatsoever to localities that offer
similar programs (AFT, 2003).
Map 2-2: Current and Former Williamson Act Lands in Visalia Metropolitan Area
25
http://www.consrv.ca.gov/DLRP/lca/basic_contract_provisions/index.htm
Although the overall quantity of land is well known, just where these contracted lands
are located is of even greater importance (See Map 2-2). Land use lawyer Susan
Petrovich claims, “the Williamson Act is very effective for non-prime parcels like
ranches out in remote areas because it keeps development from leapfrogging into the
rural areas and keeps it closer to already developed areas” (personal communication).
A formal study conducted in 1989 lends great support to her claim (Williamson Act
Study Group, 1989). Consequently, there is less conversion in lands distant from
urban clusters, but those areas have always offered less attraction for speculators or
developers than those on the city fringe, and arguably require less strenuous measures
of protection. Unfortunately, these distant parcels are mostly comprised of less
valuable grazing lands while the best farmland tends to be closer to the city center
16 Prime farmland is defined by meeting two conditions: one, the land has been irrigated in the last 4 years and two, it meets the various soil criteria as defined by the USDA. More information can be
26
(See Map 2-1). Thus, the lands just outside of California’s cities are prime both for
farmland and real estate, a competition that agriculture just cannot win. As Goleta
farmer John Lane, manager of Lane Farms says, “Once a neighborhood is built next
to a farm it is only a matter of time before the farm succumbs to development”
(2005).
Williamson Act Criticism
Differential assessment programs, such as the Williamson Act, are supposed to
redirect these tendencies of farmers to act in their own economic best interest by
making farming less expensive on the urban periphery and elsewhere. Taxes are low
for those who wish to keep farming while for those who do not wish to continue or do
not have children or other family interested in working the land, no amount of tax
breaks can compel them not to sell. But there is much middle ground for those who
can be swayed one way or the other. This is a crucial piece of understanding, at least
rhetorically. Where is the breaking point for each farmer? How many tax breaks are
necessary for farmers to remain farming rather than sell out, statistically speaking?
This work lends itself to this continuing debate concerning farmers as agents and how
they act in times of growth. It also weighs in on appropriate policy structuring for
remedying the issue of prime farmland loss.
found at http://www.consrv.ca.gov/DLRP/fmmp/overview/prime_farmland_fmmp.htm
27
Differential Assessment programs are in place to help prevent farmland conversion
but many have pointed out they are not a long-term solution. In “A Panacea That
Wasn’t,” John Dean argued that prime farmland has the least comparative tax
advantage since the Williamson Act taxes farmers based on their farm income and not
on the value of their land (1975). Though prime land farmers are taxed more than
ranchers, it is also important to point out the real estate worth of these lands is usually
higher due to their usual nearness to urban areas and a topography ideal for
development. They may pay more than ranches under the Williamson Act but they
certainly pay less than they would without the Act. The more important comparison
lies between the advantage of paying taxes on productive farmland or selling that
land for tremendous profit (Dresslar, 1979). This comparison often renders edge
farmers a greater advantage than remote ranchers. Distant intensively used
agricultural lands would, however, fall victim to Dean’s cited disadvantage. All who
have studied the Williamson Act recognize this major dilemma. Nevertheless,
despite the criticisms, solutions have been intractable (Sokolow, 1990).
The lack of centralized administration is another fault found by scholars. John
Dresslar (1979) called this a major failing and insisted the State take steps to alleviate
the problems resulting in local regulation of the Williamson Act. Though this
recommendation was made nearly thirty years ago it still has not been implemented,
though some of his others were.
28
Alvin Sokolow, widely recognized as one of the leading scholars studying the
Williamson Act, gave the Act a mixed endorsement in 1990. On the one hand, he did
admit that it helped keep certain farming operations solvent. On the other hand,
however, he claimed it did not have an appreciable overall effect on the conversion of
agricultural lands to urban lands. Williamson Act enrollment numbers, therefore, can
be misleading (See Table 2-2) since there are many new and distant farmlands joining
to make up for those on the urban fringe leaving, most likely as a precedent to
development.
Table 2-2: Williamson Act Enrollment Acreage
Total Reported Acreage* Fiscal Year Calendar Year Total Reported Acreage*
1990-91 1990 15,969,159 1991-92 1991 15,946,783 1992-93 1992 15,942,758 1993-94 1993 15,952,365 1994-95 1994 15,952,144 1995-96 1995 15,908,538 1996-97 1996 15,812,511 1997-98 1997 15,889,804 1998-99 1998 15,925,301 1999-00 1999 15,977,116 2000-01 2000 15,936,437 2001-02 2001 16,344,433 2002-03 2002 16,504,721 2003-04 2003 16,560,132
*Totals include both continuing term and nonrenewal Williamson Act contracted land, as well as a small amount of other enforceably restricted non-Williamson Act acreage. Note: “nonrenewal” refers to land that is undergoing the process of non-renewal at that time step only, it does not include those lands that have completed this process sometime in the past.
29
Perhaps Peter Brand says it best, “On the world stage California is a major
agricultural power, but at the local level the agricultural industry offers almost no
resistance to the forces that may eventually destroy it (1995).”
In light of these criticisms, this research has important applications to policy. For
instance, by examining the following attributes of each parcel for each time step:
opting in or opting out; distance from urban area; prime farmland or other; we can
begin to understand how the particular implementation of this policy might be altered
with a view towards greater more efficient realization of its stated goals. By making
the tax breaks greater for prime farmland and less for non-prime farmland, for
instance, the state could help channel growth away from the prime farmland and
towards non-prime farmland.
Theories of Growth
The phenomenon shown in Map 2-1, prime farmland surrounding cities, is not a
recent trend. Even two hundred years ago, Johann Heinrich von Thunen, (1783-
1850) noted that more intensive agricultural lands tend to be closer to the city while
ranching and other less intensive agricultural activities tend to be located farther
away (Von Thunen, 1966).
He theorized that, in an equal plain stretching in all directions, there is an observable
pattern of rings radiating outward from the city where the intensity of agriculture
decreases with distance from the city center. Though geographically applicable, his
30
theory is essentially economic: as one travels farther from the city center land prices
decrease and therefore, less intensive uses of the land are made affordable. As the
thinking goes, a 5000 acre cattle ranch on the border of a large city is just not
economically feasible. However, a five-acre greenhouse parcel that concentrates on
expensive flowers would be. It is natural then to speculate on what could be more
profitable than a five-acre greenhouse parcel. The answer is most likely the creation
of an urban parcel, perhaps a grocery store where agricultural goods are distributed
on a comparatively small lot (perhaps an acre) to a large population base. Since Von
Thunen’s model of land use is temporally static this work can help build upon his
original ideas. However, if it were not, and we continue to expect farmers to act as
economic agents, then we would see those parcels on the urban fringe eventually
becoming urbanized. This would widen the urban footprint and then a wave of more
intense land use could ripple across the landscape. Of course, farmers do not always
act as economically rational agents and many farmers love their life’s calling.
Nevertheless, it is easy to imagine a childless farmer retiring, selling the farm to
developers, and retiring in luxury somewhere.
Thomas Malthus (1798) proffered a theory of population and agriculture even before
von Thunen. He claimed the human population will continue to grow exponentially,
exploiting more land and growing increasing amounts of food until room and
resources are exhausted (Ibid). There would then be an “adjustment” period, which
would involve a great famine and a plummeting of the population to sustainable
levels. His offer of abstinence as an alternative to this dystopia has, thus far, found
31
too few adherents throughout the world to appreciably affect the high rate of
population growth. Nevertheless, his theory was generally accepted, with only a few
objections, until Ester Boserup’s landmark work, The Conditions of Agricultural
Growth (1965). Boserup argues against Malthus’s reasoning and makes a
fundamentally opposite claim. Rather than agricultural innovation driving population
growth, as Malthus contends, it is population growth that creates agricultural
ingenuity. People do the least amount of work necessary to get the food they need,
she asserts. This least effort involves the greatest amount of land (Boserup has
divided the world’s farming systems into 5 hierarchical classes based on the amount
of time land remains fallow). It is only when population increases and available land
is all in use that people, in general, begin to intensify farming practices on the land.
Over 40 years ago Alonso (1960), revising von Thunen, pointed out the peculiarity of
Western urbanization. Instead of the rich living on the inside of the city, with the
poor living in surrounding squatter settlements, like we see in much of the developing
world, the wealthy in the West live on the urban periphery, where land is cheapest
and they can live at lower densities. This Western hunger for the urban-rural
boundary and the associated lower density living, that Americans in particular equate
with a high quality of life, is a large driver of suburbanization in the US. Of course,
our reliance on the automobile as our principle means of travel cannot be extricated
from our hunger for the urban periphery and the resulting and relatively even spread
of American cities, as opposed to along subway routes, represent this. Brian Berry
(1970) argues that the automobile and the abundance of American roads have not
32
caused urban sprawl but simply have facilitated our demand for low-density
residential land. Jackson (1985) disagrees and claims that our suburbanization is
based on our car culture, with Los Angeles as the epitome of automobile-directed
sprawl. Regardless of which came first, the chicken or the egg, everyone agreed by
the mid-1980s that America was a suburban nation (Ibid). Between 1950 and 1970
nearly 11 million people in America moved to the suburbs, while many others were
simply born there (Ibid).
As far as agricultural impact, it was during the late 1970s that the Department of
Agriculture first announced that three million acres of prime farmland were being lost
each year to suburban sprawl. In that time D. Berry (1978) suggested that much
farmland, in anticipation of conversion, would cycle out of production, causing a
ripple effect. Thus the emergence and recognition of the “impermanence syndrome”
(Lopez et. al., 1988), whereby farmers amidst the suburbs, realizing their farming
days are numbered, begin speculating about future development on their land.
Consequently, they reduce investment into their farming infrastructure and/or do not
replace failing or broken equipment, accelerating the land’s productive obsolescence,
and therefore strengthening the farmer’s desire to sell. Also, market forces can,
without a mitigating regulatory regime like the Williamson Act, drive property taxes
so high as to make farming the land an untenable use (Ibid). However, one aspect of
the market somewhat assuages this phenomenon. As the city advances towards the
farms, transportation costs to and from the market served by the farmer are reduced
(Ibid). Nevertheless, this is rarely enough, regardless of the regulatory regimes and
33
endemic market realities, to offset all of the previously mentioned difficulties and
temptations faced by edge farmers. One need only observe the disappearance of
farms along the urban periphery to confirm this, a retreat without which sprawl often
could not proceed.
R. F. Muth (1961) attempted to systematically examine this dynamic when he created
his now well-known urban-rural conversion model. Essentially, inside the urban
boundary, the market value of land exceeds the use value of land for agricultural
purposes and vice versa. However, by incorporating these aspects of speculation,
Lopez et. al., (1988) have suggested the boundary may advance more quickly than
Muth’s model explicitly allows. For instance, if speculators expect the urban-rural
line will be advancing then they anticipate an increase in the market value of their
land. If this expectation is higher than the current use value of their land, then they
will hasten this conversion more quickly, much in the same way that the perception of
a stock market crash can actually cause one.
Modeling Theory and Criticism
In geography as well as economics, models have often been built and consulted to
forecast things to come. In this dissertation, therefore, to tease out the patterns of
Williamson Act termination in the context of urban growth and to extend this into the
future, a model will be needed. Since this research does not simply use regression, as
this lacks a sophisticated understanding of the dynamic variables involved, a more
34
data intensive and spatially explicit tool is used instead. However, whenever scholars
deign to model a geographical phenomenon, they must face the many criticisms of
such an approach. These warnings were very nearly successful in burying modeling
of this type. Once Douglas Lee, Jr. wrote his “Requiem for Large-Scale Models” in
1973, many believed urban and land use modeling had finally spent the momentum it
gained from the quantitative revolution of the 1950s. Lee insisted these ambitious
models were nothing more than black boxes, were not based on soundly applied
urban theory, and were so complicated that even their originators did not understand
the connections between what went in and what came out. Indeed, modeling did
seem to whither under the continuous attacks from alternative approaches, at least in
the United States. However, many of these criticisms, particularly Lee’s, have been
attenuated by great leaps in technological innovation and a certain degree of
advancement in urban and land use theory, though many would argue not nearly
enough. In 1985, for instance, Harris officially claimed that Lee’s criticisms were
outdated and, in fact, fallacious from the start (1985). Undaunted, model critics have
continued to insist that GIS and urban modeling is an antediluvian attempt to
continue a now largely discredited quantitative revolution (Tuan, 1971; Pickles,
1999). Or, as Mercer and Powell (1972) claimed, the decades long devotion to
positivism stripped geography of scholars, leaving behind only computer technicians
and number crunchers. Couclelis, mindful of these attacks, has suggested theoretical
constructions of space (proximal) and implementations (geo-algebra) that would
render GIS and associated modeling environments more relevant to planners (1991,
1997). Wegener, in answer to model critics, insists that the proof of GIS and urban
35
model relevance lies in their continued application to real-world scenarios (1994).
Some have taken a middle path, validating Lee’s and others criticisms not by
abandoning modeling but by building a more thoughtful mousetrap. John Landis, in
the creation of his CUF model (1995) as well as Alberti (1999) would fall into this
category. Tayman (1996), for better or worse, argues that large-scale models are here
to stay not only because of greater advances in computing but because policy-makers
now insist upon their creation.
Though Lee’s requiem did not sound the death knell for large-scale modeling,
disappointing many humanist scholars, it did officially usher in the era of skepticism
in their use. Nevertheless, cellular automata (CA) models answer much of this
critique since they are usually based on simple rules that only through their
application result in complexity and emergence. CA models are also visually useful
and easily explainable and it is for these reasons, along with others, that they have
eclipsed not only systems dynamics models (Forrester, 1969) for geographic use in
general but for other related fields. Cellular automata models, unlike many of the
models being attacked in the past, do not necessarily suffer from large-scale black
box issues. First, CA are spatially explicit, and the model outcome is the totality of
many small and relatively simple micro-simulations. Second, unlike the
comprehensive model undertakings of the past, which provoked so much vitriol from
Lee (1973) and others, CA have rules that are usually simple but result in complexity
only in their application. Much like the game of chess has simple rules that allow for
tremendous complexity in strategy and tactics, CA’s simple rules allow for greater
36
transparency and understanding of results. Third, CA is easily rendered probabilistic,
rather than deterministic, such as a Systems Dynamics approach. This stochasticity
offers greater flexibility in results and allows for greater statistical analysis resulting
from the various conditions the modeler creates. Therefore, since CA is the model
platform of choice for this project, a proper understanding of micro-verus-macro
simulation as well as cellular automata versus agent based modeling is appropriate.
Beginning with theory, Helen Couclelis wrote a paper nearly twenty years ago
entitled, “Cellular worlds: a framework for modeling micro-macro dynamics” (1985)
where she confronts the problem of assuming that all macro-level phenomena emerge
solely from an aggregation of micro-level phenomena. She also cites the allocation
of macro effects amongst the agents as one other great difficulty. For example, if
there is a known increase in population (perhaps supplied by a macro-scale
population sub-model) how should housing demand be distributed amongst the
individual parcel managers across the extent of the modeling environment? Surely,
distributing this demand without knowing exactly which areas are increasing in
population dilutes the model’s accuracy. This pervasive problem “concerns our
capacity to explain the relationship between the constitutive elements of social
systems (people) and emergent phenomena resulting from their interaction (i.e.
organisations, societies, economies)” (Goldspink, 2004). Nevertheless, urban
modelers are determined to bridge this gap between the micro and the macro. Before
this is addressed, though, let us examine just what these two approaches involve in
more detail.
37
One of the main reasons for the nature of these two different approaches concerns the
disciplines that chiefly utilize them. Micro-simulation has evolved from the
approaches of the social sciences and, therefore, has yielded models that deal with the
behavior of individuals and how these aggregate into land use changes (Verburg
et.al., 2005). Two of the most commonly used micro-level approaches are agent-
based simulations (ABS) and micro-economic approaches. An agent is “a real or
abstract entity that is able to act on itself and on its environment; which can, in a
multi-agent universe, communicate with other agents; and whose behavior is the
result of its observations, its knowledge and its interactions with other agents”
(Sanders et al. 1997, as excerpted from Verburg, et al. 2005). Agents are given
certain motivations and other such endowments that affect their behavior and are then
set loose in an environment. ABS’s, when constructed effectively, can yield
fascinating results in the form of emergent properties, i.e., higher level effects that
cannot be observed or predicted from watching these agents in isolation. And not all
of these effects are positive. See Couclelis (1989) for an example where the micro
actions of individuals maximizing their utility collectively emerge into negative
externalities. It should be noted here that cellular automata (CA) are not necessarily
agents, though agents are automata. According to Benenson and Torrens (2005),
however, all CA fail to meet the criteria of agency since a) they do not have their own
agenda and b) they do not anticipate the future. However, if cells represent decision-
making landowners then this discussion may need to be refined further. These
characterizations concerning agents versus automata are still being debated by
38
modelers and theorists (Clarke and Couclelis, personal communication).
Nonetheless, it is contended here that most cellular automata are a form of non-agent
micro-simulation since they act individually from effects in local neighborhoods. It is
also possible to witness emergent behavior since it may not be obvious from the rules
programmed into the cells what kind of results will be produced on the macro level.
This leads to micro-economic models of land use change. These usually take on the
perspective of landowners trying to maximize profit or utility from their land.
However, as Verburg and his associates point out (2005) these models do not scale up
very well to the macro level.
Modeling Approaches
A good example of a micro-level model would be the NELUP (National
Environmental research council Land Use Programme) extension (Oglethorpe and
Calaghan, 1995). According to Agarwal et. al. (2000), this model is at a high level
of Human Decision-Making (HDM) complexity since it treats individual farmers as
micro-economic agents who are trying to maximize their profit and, in so doing,
affecting land use.
Macro-level models, on the other hand, tend to either employ macro-economic theory
or use the systems approach. SCOPE, a model that this author helped to build, is an
example of a macro-level model (Onsted, 2002). Built in STELLA, it uses a systems-
dynamics approach to model changes in jobs, population, housing, and even quality
39
of life throughout the South Coast of Santa Barbara County. What it could not do
though, which was of frustration to many audience members in a long-winded tour
amongst the populace, was allocate growth to particular locales within the South
Coast. This is, of course, a major drawback. Macro-scale models do have the
advantage, however, of incorporating more modules with greater ease since they can
be mathematically related within such frameworks as system dynamics. Which is not
to claim making them accurate and validly constructed is an easy task. In the end,
SCOPE is limited in its lack of specificity. It offers many different outputs, but none
of them are easy for a user to personalize (i.e., what will happen in my
neighborhood?) For planners, these model results, if they are to be believed, are
challenged to transform them into useful planning-relevant tools (See Couclelis,
1991).
It should be noted here, though, that not all macro-level land use models are
aspatial. For example, the Mertens and Lambin (1997) univariate spatial model
relates a series of landscape metrics to the pattern of deforestation in Cameroon and
then uses statistical correlation to make predictions about future deforestation. By
overlaying a remotely sensed image of land cover with another GIS layer of cultural
and natural variables this correlation is structured. Although highly mathematical, as
in the case of systems models, it does involve spatial inputs and outputs. This model
should not be mistaken, however, for a cellular automata, even though it uses raster
images and pixel outputs, since it employs a very different methodology.
40
There are also models that try to take the best of both worlds by integrating both
the micro-level simulation techniques as well as the macro-level simulation
techniques. This is difficult, though, for several reasons. First, if the macro-level
model is aspatial then distributing its results across a spatial field involve specious
and arbitrary decisions. Second, macro-level spatial models often have units of
measurement or resolutions that are too coarse to tease out processes that micro-level
simulations can discover. Third, micro-simulations often use too small an extent to
account for the processes existing at the macro-level and, therefore, fail to properly
account for the context in which their model is running. Fourth, it is dangerous to
aggregate separate micro-simulations together and proclaim a theory of their totality
since there may be emergent properties for which the simple sum of the sundry parts
can’t account (Verburg et. al. 2005).
A germane example of a model that attempts this integration is explained in a
paper written by Dawn Parker and Vicky Meretsky (2004). Their paper intends to
address the gap between macro-level land use change models and empirically derived
micro-level agent based models (ABM). As they point out, macro-oriented models
take such factors as number of jobs and population growth and use this to allocate a
new land use composition with new acreages listed across multiple categories.
Micro-scale models, on the other hand, have the advantage of precise spatial
allocation but when trying to account for broad macro-forces must use coarse
measures that usually fail to “capture the dynamic feedbacks between land-use
41
patterns, spatial location, and land-use composition” (Parker and Meretsky, 2004,
section 1.2).
Their model uses an endogenously derived rent for urban land based on the
quantity of developed land and its pattern. This pattern is derived from micro-scale
decisions of parcel owners. This, in turn, is then accounted for in the endogenously
generated price-mechanisms and this feeds back to the parcel agents. One of the most
innovative aspects of their model is their metrics of edge-effect externalities, which
they use to feed into the macro-level aspects of the model. Similar to SLEUTH, these
measurements take into account such things as: landscape composition, number of
patches/mean patch size, the average deviation of patch shape from the minimum
edge/area ratio (a circle), as well as others.
Though the possibility of land pattern metrics as the conducting rod between macro-
level modeling and micro-level modeling is intriguing, (Mertens and Lambin, 1997)
this dissertation required only a viable CA with which to answer its questions.
Nevertheless, there are other models that could be employed other than SLEUTH or
any CA for that matter. LUCAS, for instance, is possible because it employs
transition probability matrices (LUCAS website) (similar to SLEUTHs deltatrons
(Dietzel and Clarke, 2004)) and is spatially explicit. However, LUCAS has certain
extraneous elements for agricultural conversion purposes (like habitat impact) that
could prove cumbersome. CUF is also a good choice because it has various
42
submodels that allow for greater complexity and it includes zoning (Landis, 2001).
One of CUF’s major drawbacks, though, is it its lack of temporal complexity. “What
If?” is a GIS-based software package where users explore different development
scenarios and see the probable growth patterns and socio-economic consequences that
result (Klosterman, 2001). METROPILUS is actually a combination of Putman's
DRAM (which locates households) /EMPAL (which locates employers/employees)
along with an ArcView GIS output and is currently used in six major metropolitan
areas (Putman, 2001). CUF II, also by Landis, goes one step further with multiple,
rather than binary, land uses, allows different land uses to compete against each other
for sites, accounts for redevelopment and infill, and has been historically calibrated
(Landis, 2001). CURBA, again by Landis, is the California Urban and Biodiversity
Analysis Model (Ibid). While CUF II focuses on urban land uses, CURBA is more
concerned with the habitats in rural areas (Ibid). INDEX (developed by Criterion
Planners/Engineers) is a model that uses GIS to estimate the impact certain land use
decisions will have on community indicators (Allen, 2001). Finally, Li and Yeh
(2000) used a constrained CA they developed themselves to model sustainable urban
development in China, particularly the preservation of prime farmland. It is
interesting to note that different regulatory regimes may be better suited to different
approaches. For instance, China enjoys more centrally controlled planning and,
therefore, could have a better chance of implementing from the top-down the
sustainable ramifications the model suggests. In the United States, on the other hand,
land use planning is so decentralized that each locality suffers from a certain tragedy
of the commons, where each local government must invest in the conviction that
43
compact development and preservation of prime farmland, even when inconvenient,
will be to everyone’s benefit, including those not under its jurisdiction. Despite these
tempting alternatives then, SLEUTH, as an explicit HDM agricultural conversion
model, is more than adequate for the task at hand.
Kuminoff and Sumner’s “Modeling Farmland Conversion with GIS Data” (2001) is
also an approach worth expanding. In their study, they used both econometrics as
well as GIS tools to discover what SLEUTH (Silva and Clarke, 2002; Clarke and
Gaydos, 1998; Clarke et. al., 1997) bears out in a methodical fashion: the greatest
predictor of agricultural conversion is proximity to the edge of urban lands as well as
population growth. It has little to do with aggregate stock values of land. This shows
the importance of geographical specificity when conducting policy analysis,
particularly land use policy. In other words, geography matters. Of the six variables
they analyzed in their search for positive correlations with agricultural conversion,
the economic variables either showed negative or negligible correlation. They also
wrote that local zoning laws seem statistically irrelevant to the nature of urban
expansion and agricultural conversion17. SLEUTH, also, has been validated without
regard to specific zoning laws, apart from lands that are completely off limits (i.e., an
excluded layer). Instead, it relies upon organic growth rules that are universally
applicable to any region (Silva and Clarke, 2002; Clarke and Gaydos, 1997; Clarke
and Gaydos, 1998). The great success of SLEUTH’s applicability both domestically
and internationally speaks well for its employment in this project but also discredits
44
the idea that zoning, with its ad-hoc relevance, has any long-term effect on the shape
and direction of growth (SLEUTH’s operational rules are discussed at greater length
in subsequent chapters). As those who study such matters can attest, zoning is either
changed to fit the political exigencies of the time or, even more simply, exceptions
are made for convenience without any change to the master plan. Well-meaning
though they are, zoning efforts are no match for the forces of urban growth and
sprawl. A simple test of this fact can be made by examining old master plans for a
community and comparing these with what is found there today; many places zoned
for agriculture or other less intensive land uses are now either residential
neighborhoods or big box malls. The only zoning classification that SLEUTH
previously incorporated is outright protection, i.e., an excluded layer such as a park or
other easement. This project explores a more nuanced excluded landscape based on
the particular nature of the Williamson Act.
As for Kuminoff and Sumner, it is unclear how their development restrictions
quantitatively took the Williamson Act into account. Although there could be many
reasons why aggregate numbers could lend the appearance of impotence to the Act,
details must be explored spatially before it can be determined whether or not
restrictions are making a difference, and in what way.
17 Although they provide caveats to this surprising claim. See Kuminoff and Sumner, 2001.
45
This dissertation picks up where they left off. In their conclusions section they
remark: “The importance of edge effects as a determinant of farmland conversion and
of increased urbanization may be of particular interest to city planners and farmland
preservation organizations.”
This call to additional investigation is answered in this project since their research
lends credence to the assertion that no single factor is as significant in farmland
conversion than propinquity. This dissertation demonstrates this by showing the
feedback between urban growth and Williamson Act termination. The cellular
automata platform, in this case SLEUTH, has been indispensable in accomplishing
this.
SLEUTH Applications
Since SLEUTH is indeed the CA being used for this project, it is important to render
it in a broader context of modeling via a framework. C. Agarwal, et. al. (2000) have
designed just such a framework that compares land use/ land change models across
three dimensions: time, space, and human-decision making. These three dimensions
are ascribed two attributes: scale and complexity. Time scale refers to both time step
and duration. The time step is the shortest period of time between observable
phenomena. In SLEUTH’s case this would be one year. The duration is the length of
time that the model runs. The space scale refers to resolution and extent: SLEUTH’s
46
resolution is one cell size (for example, 30 meters by 30 meters) and the extent is the
area being modeled (the South Coast, for example).
These concepts are relatively easy to digest. Agarwal et. al., though, proposed
something new in the examination of the human-decision making: “agent” and
“domain.” (p.5) These terms are meant to be analogous to resolution/time step and
extent/duration. The agent is the unit that will be making decisions and the domain is
the universe in which the agent can act. For instance, in SLEUTH, the agent (though
not necessarily a “true” agent as discussed on page 39) is a cell and the domain would
be the zone being examined.
Now that the attribute of scale across the three dimensions is understood attention
should be focused on complexity, which, across time, is measured by the number of
time steps and model duration. The greater the number of time steps and the longer
the duration, the greater the complexity. Also, the highest complexity involves time
lags and feedback. SLEUTH has many time steps and a long duration (but no time
lags) so it gets a fairly high score on this dimension. Spatial complexity is measured
by how “spatially explicit” (p. 2) a model is. Spatial interaction, where one agent (or
cell in this case) is influenced by neighboring agents, is the most complex and this is
what is seen in SLEUTH.
Human-decision making (HDM) complexity spans the range from none (like in the
case of a systems dynamics model) all the way to high complexity where not only are
47
the agents making decisions within their domains but multiple agents can aggregate
their behaviors to affect higher level processes and higher level properties affect the
HDM of the lower-level agents. SLEUTH is relatively low on this scale since it does
not explicitly account for HDM. Rather, cell behavior is determined through Monte
Carlo simulation based on neighboring cells. Although all urban development
implies HDM, some models are more overt in this accounting than others.
The authors state “the ultimate goal of human-environment dynamic modeling (is to
be) high in all three dimensions.”(p. 10) With this in mind, there are several desirable
traits that an agricultural land conversion model should have. Most of these are
already contained in SLEUTH which, by itself, would be relatively adequate for
modeling farmland conversion. However, the modifications undertaken in this
research now more precisely account for the fluid nature of agricultural protection
policies and how they are shaped by farmers’ choices. It is further averred that these
modifications have improved the model in several ways.
First, and most importantly, SLEUTH is imbued with greater spatial complexity since
each county undergoes an additional modeling cycle analogous to an urban growth
run: Williamson Act termination and spread. SLEUTH currently has an excluded
zone that is off limits to development but this zone is static throughout the duration of
the model run. Although SLEUTH currently has the capacity to have “weighted
resistance” to development this still must be programmed in by the user and is
difficult to relate to policy (Dietzel and Clarke, 2004). The greater spatial complexity
48
results from addition of new rules. Cells that are currently in the Williamson Act are
excluded. However, the proximity of urban edges influences these cells to leave the
Williamson Act. Once they leave, they are then subject to the same phenomena that
the original SLEUTH model bears (Slope, Land use, Exclusion, Urban,
Transportation, Hillshade).
The greatest improvement made in SLEUTH is the addition of HDM. During the
Williamson Act modeling (see Chapter 3 for more details), cells reflect the individual
decisions of landowners to leave the Williamson Act, which needs no approval from
government agencies. These agents simulate HDM by acting in their own interests to
leave the Williamson Act, often in the hopes of selling their farmland to developers.
This builds the pathway to this research’s most important contribution to knowledge:
a technique for not only forecasting future differential assessment landscape’s (such
as the WA) but also a way of using these forecasts to construct probabilistic excluded
layers for use in traditional urban growth models.
SLEUTH is not an untested application. In fact, it has been employed by researchers
all over the world: from Albuquerque, New Mexico to Yaounde, Cameroon
(Gigalopolis website). The use of SLEUTH to forecast changes in the Williamson Act
builds upon the work of these innovators. The model was designed by Keith Clarke of
UC Santa Barbara, who was inspired by the modeling of fire spread. His partnership
with Len Gaydos of USGS, along with others who helped write the code, allowed for
SLEUTH to not only become well-known amongst researchers but amongst decision-
49
makers at all levels of government. National television even displayed some of
SLEUTH’s results on programming concerned with the explosion of urban sprawl
across the American landscape. Nevertheless, and much to SLEUTH’s credit, an
exhaustive accounting of its numerous applications is not necessary since they have
varied more along a geographic dimension rather than a methodological one. Also, a
full explication of SLEUTH’s operational underpinnings will take place in Chapter 3.
What is prudent, though, is a tour of the model’s evolution and how various
innovations by both practitioners and designers have laid important groundwork for
the work realized in this dissertation.
As of this writing there have been at least 32 cities or regions around the world that
have benefited from SLEUTH forecasts. Though for the most part SLEUTH has been
applied as a straightforward urban growth or land use change model (it can be used in
either capacity), there have been those who have created new uses for SLEUTH while
applying them to their region.
Arthur et al., for instance, coupled SLEUTH with an urban runoff model when
running simulations for Chester County, PA (2000; Arthur, 2001). Specifically, the
land use change portion of SLEUTH was run and each year of land use output was
then used to generate urban runoff responses. However, this coupling, rather than
being mutual, is actually unidirectional since these runoff impacts do not inhibit or
encourage urban growth. Nevertheless, Arthur’s work is considered one of the first
couplings of SLEUTH with a physical module capable of exploring land use change’s
50
impact on the environment. Arthur also took inspiration from SLEUTH’s Monte
Carlo uncertainty outputs and claimed that, ideally, a similar probability grid for
evapotranspiration and stormwater runoff could be created. However, she did not
fully implement a tenable methodology for Monte Carlo incorporation and was
therefore left with choosing only the final map output for the final Monte Carlo
simulation as input for her hydrology module or using coarse coupling methods that
made aggregate use of the Monte Carlo runs. As will be seen later, this dissertation
discovers a way to use probabilistic Monte Carlo simulations for its Williamson Act
termination landscape.
SLEUTH has also been coupled with other urban growth models in order to more
accurately determine biodiversity loss from various growth scenarios. Cogan et. al.,
(2001) compare urban growth forecasts amongst Landis’ model (California Urban
Futures or CUF), SLEUTH, and a coarse simplistic model (GAP boundary) to
determine differences in threat to both species and habitat. This work is important
not only for its ability to tease out important planning ramifications of model choice
but also for illustrating the importance of scale selection in both data input and model
output during result interpretation. Again, these two models are more loosely
coupled than tightly bound since SLEUTH urban growth and land use change are
used not to alter the operation of CUF but to examine its effect on CUFs attributes,
e.g., species habitat. Though more a combination of data than code, this effort did
much to demonstrate the benefits of data combination. This dissertation assembles
51
data from several different sources as well, which will be explained further in Chapter
3.
Solecki and Oliveri (2004), similar to Arthur, coupled SLEUTH with air quality
impact assessment models. However, in this case, they downscaled several narrative
scenarios garnered from the Intergovernmental Panel on Climate Change’s Special
Report on Emissions Scenarios (IPCC’s SRES) to create SLEUTH’s growth
parameters. Broken down into a pessimistic scenario (replete with increasing reliance
on automobiles for transportation and low-density sprawl) as well as a more
optimistic alternative (infill, higher-density, less reliance on automobiles) these two
possibilities drove two different parameterizations of SLEUTH. There was also a
third, based on SLEUTH’s normal calibration routine, resulting in a business-as-usual
forecast. These maps are then output as storytelling components to the SRES in the
New York Metropolitan region. Though the authors admit the difficulty in using
macro-scale narrative descriptions to adjust parameters on a local scale, (the possible
specious reasoning plaguing such attempts is discussed above) nevertheless SLEUTH
provided an appropriate tableau on which to realize visually and locally the various
futures of the IPCC. For instance, the authors used SLEUTH’s probabilistic
Excluded Layers in order to decrease infilling for the sprawl scenario. Though they
did this through trial and error rather than rigorous calibration, these sort of dynamic
excluded probabilities can be approached more systematically, which will be
discussed further in Chapter 3.
52
Porto Alegre City, Brazil was used by Leao et. al. (2004) to couple SLEUTH with a
multi-criteria model of landfill suitability assessment. The authors here make an
argument that SLEUTH’s CA functionality answers Couclelis’ (1997) observation
that GIS and urban models have a theoretical gulf between them that makes
interoperability problematic. This boils down to different constructs of space: GIS
uses absolute space and urban models use relative space. Couclelis suggests that
proximal space could bridge this gap with its strong use of neighborhoods (as found
in CA) as they are invested both with properties of absolute and relative space.
Their methodology makes use of SLEUTH’s ability to not only forecast an amount of
urban growth but also its spatial allocation component. This allows the waste
disposal submodule to output an amount of landfill needed to serve a population at
any given time but SLEUTH’s results also help to decide where to locate these new
sites. They have even integrated the NIMBY phenomenon along with more common
distance decay parameters (transport costs) for this suitability analysis.
Faculty and students at UC Santa Barbara have been on the forefront of not only
innovative uses of SLEUTH but also its experimentation and improvement. The most
important endeavor in that regard is the addition of the Land Use Change Modeling
component. With their use of “deltatrons,” Candau and Clarke (2000) took SLEUTH
from a binary urban growth model to a system that can simultaneously predict urban
growth and the change of one land use class to another (A full diagnostic of
SLEUTH’s functions will follow in the next Chapter). Herold et. al., (2003) also
53
incorporated into SLEUTH greater and more varied spatial metrics from which to
evaluate model fit with past data. They also provide a more sophisticated and caveat-
laden discussion of the proper use (and misuse) of metrics and their sensitivity to both
temporal and spatial scale. They even suggest the possibility of using metrics as a
policy tool for intelligent growth, once a better understanding is gained from current
research. SLEUTH was even used to more accurately understand past growth, rather
than forecasting future growth (Goldstein, et. al., 2004). The authors also compared
SLEUTH against an agent-based simulation to examine which bore the greater utility
in understanding past growth. They determined that SLEUTH (as representative of a
CA) had greater flexibility than an ABS, though the ABS approach also proved
useful.
As for the future use of and research in SLEUTH and other CA, there are already
emerging realms of inquiry. First, since model calibration (discussed at length in
Chapter 3) is as much art as science this makes systematization difficult (Benenson
and Torrens, 2005). Many are at work to remedy this (Dietzel and Clarke, 2004). The
lack of agreed upon conventions of metric evaluation is, in this author’s opinion, the
major compounding factor. Other novel uses include the injection and substitution of
“urban objects” into the field (Benenson and Torrens, 2005). These can include
houses, streets, or other objects of infrastructure and can alter the properties of
surrounding cells and even alter their shape. The agglomeration of cells together with
the same properties is another field of budding research. This is of particular use
when using CA at the cadastral level where parcels are of different size and shape
54
(Sembolini, 2000). Though this particular approach was not used here, due to
technical limitations, it could be of great use in the future. For instance, by directing
all cells of a parcel to leave the WA at the same time (or to join it) a more accurate
calibration is achievable as well as a more accurate forecast. Future efforts, which
this author intends to undertake, will hopefully utilize such cutting edge
methodology.
By broadening the use of space, Benenson and Torrens (2005) suggest the possibility
of employing CA at different spatial scales. This would allow different phenomena
and states to be reactive according to different sized neighborhoods. By redefining
the use of time, asynchronous CA, where each cell runs on its own time clock and not
all cells are available for state changes simultaneously, has the potential to greatly
influence the results of an otherwise synchronous CA and, in fact, may lead to greater
possibilities of emergence (Ingerson and Buvel, 1984; Schonfisch and de Roos,
1999).
Though this project builds upon all of the creative and innovative struggles of those
that preceded it there is one, in particular, upon which this undertaking most strongly
builds. Michel B. Tietz, Charles Dietzel (of UCSB), and William Fulton, under the
auspices and funding of the Public Policy Institute of California (PPIC) wrote a report
entitled, “Urban Development Futures in the San Joaquin Valley” (2005). By
consulting planners and decision-makers, the team designed four different scenarios
that were then run in SLEUTH. The scenarios were: 1. Accommodating Urban
55
Development, which is the baseline scenario. 2. Prime Farmland Conservation 3.
High-Speed Rail and 4. Automobile-Oriented Growth.
Though this dissertation does not explore transit-based scenarios it does examine
alternative farmland futures, though with a different methodology. Their Prime
Farmland Conservation alternative simply excludes all Prime Farmland from
development, without taking into “explicit account…such factors as zoning, local
growth control, or other public policy.” (p. 42) However, this reflects no shortcoming
of the team but the nature of SLEUTH itself. This CA uses organic growth rules in
order to predict future urban growth and land use change and the only regulations or
zoning for which it accounts are excluded lands, i.e., lands off limits to development.
Though, as discussed earlier, there is a functionality to create probabilistic, rather
than binary, excluded layers there is no easily quantifiable correspondence between
zoning and the creation of these probabilities.
Teitz et. al. also did not take into account the Williamson Act protected lands, though
they comprise tremendous acreage. Nevertheless, the team indeed calibrated the San
Joaquin Valley with past data and, therefore, their approach is sound. However, by
adding not only a richer, excluded, dataset but also an approach and use of SLEUTH
that allows for a projected change in Williamson Act lands, this dissertation can offer
a more sophisticated analysis of not only urban growth and land use change, but
alternative futures of protected farmland itself. Since policies similar to the
Williamson Act are by no means endemic to California, this dissertation can offer this
56
nascent yet effective method of dynamic excluded layers for use by others, to more
accurately reflect the possible urban growth landscape.
This dynamic excluded layers approach, being more time intensive than the static
approach, can only be justified if it produces more accurate results. A formal
comparison, then, between the work performed in the PPIC Report (2005) and this
dissertation is offered in the following Chapter. County by county, the legitimacy
and efficacy of this methodology against that used by the PPIC team will be
demonstrated.
57
CHAPTER 3: Methods and Data
This undertaking required not only a great deal of data but also a great deal of time to
effect as well. After the data were collected and rendered they had to be tested and
repaired when necessary. The lengthy process of calibration, and in some cases
recalibration, was then conducted before prediction could take place. After
prediction the results were analyzed and the writing process began. By far, though,
the greatest amount of time spent was on the collection and rendering of the data, so
it is that phase which should be addressed first in this chapter.
Data Background
The amount of data necessary to conduct this research should not be surprising since
most models are data-hungry and the quantity and quality of data can greatly
influence results (Lee, 1973). Therefore, before setting out in earnest to collect these
data, the different counties of California were explored and the author considered not
only the possible data they had to offer, but the particular status of their WA
programs and the growth pressure they have been under in past decades. The first
criterion was GIS shapefiles that offered fields relevant to the Williamson Act. There
was great variability amongst counties along these lines so this exploration led to a
selection of Tulare, Stanislaus, and Merced Counties. All three are in the Central
58
Valley, which has become the new locus of population growth in California, drawing
would-be homeowners priced out of the coastal market as well as newcomers from
out of state (Teitz, et. al., 2005). Also, all three counties had comprehensive data
concerning the Williamson Act, as was discovered from an initial exploration of the
Williamson Act data collected by Dr. Charles Dietzel. Though these initial
examinations were not as valuable as Assessor’s data, they were actually culled from
the same database and therefore gave an approximation of the kind of temporal and
spatial detail that might be expected from actual Assessor’s data. There were more
than three counties that fell into this category within the Central Valley. However,
the final three were settled on because a) Tulare County contains Farmland Security
Zones (FSZs), allowing a geographical examination of this new option, which many
counties have not yet employed; b) Merced County has only recently adopted the
Williamson Act and so urban growth before the Williamson Act as well as after could
be observed; and c) Stanislaus County has no Farmland Security Zones but has
employed the Williamson Act for decades and also shares a long border with Merced
County, allowing a contrast between one county in the Williamson Act and one
county not. Other counties in the Central Valley were not chosen because they did
not offer comprehensive data regarding the WA, either in format or content. By
gravitating towards those counties with more meticulous data and sophisticated
methods of recordkeeping, such as GIS shapefiles instead of paper maps, it is possible
the results could be biased. Therefore, although it is not suggested the results from
this dissertation are meaningless for other counties or regions employing the WA,
every region and jurisdiction has its own “digital DNA” and therefore variability in
59
results should be expected. Nonetheless, though the results may vary across
geographic regions, it is here averred that the approach offered in this research and
the difference between scenarios has relevance for not only any county in California
employing the WA, but any area utilizing differential assessment programs as tools
for land conservation.
Data Acquisition
Since the Land Use module of SLEUTH is being used for the purposes of this
dissertation the requisite land use data was collected. Though land can, of course, be
classified along a number of dimensions the Williamson Act orients itself with a
concern towards farmland typologies so likewise has the landscape in this manner
been classified according to farmland production. This was made more convenient by
the fact the California Department of Conservation (DOC) not only offers an online
repository of shapefiles classified by farmland types, found with its Farmland
Mapping and Monitoring Program (FMMP), but they also administer the Williamson
Act itself (CA DOC, 2006). By using this FMMP classified data, therefore, the risk of
classifying lands in categories that are not recognized by FMMP or are not
compatible with the rubric set forth in the Act itself, e.g., Prime and Non-prime, etc.
was eliminated.
The FMMP shapefiles exist biennially, generally from 1984 to 2002, the latest year
used in this research. Tulare County, however, only has data reaching back to 1986
60
Once all of the shapefiles were collected for the years 1984 through 2002, at biennial
intervals for all three counties, one could see urban growth throughout the years and
its conversion of surrounding farmlands. By creating a PowerPoint slide with each
map layer occupying the same place on each slide, an animation of sorts was created,
which is an excellent canvass for exploring map data with the most powerful tool
available, human eyes (Clarke, personal communication).
Since this research models significantly into the future it should also be based on
actual data reaching significantly into the past. Therefore, in this research, SLEUTH
runs as far into the future as data reaches into the past, realizing a certain temporal
symmetry. This is not based on the assumption of linear rates of change. If that
were the case then a simple regression model would be all that’s needed to forecast
growth. Instead, this temporal symmetry is needed to control for uncertainty. The
further one either hindcasts or forecasts from one single data point at the present, the
greater the uncertainty (Goldstein et. al, 2004). This uncertainty can be mitigated,
however, by using multiple data points as well as by modeling into the future
moderately (Ibid). These concepts are best exemplified in figure 3-1 below,
borrowed from Goldstein et. al., (2004). The reader should notice less uncertainty
over the same time period in Fig. 3-5b, due to the presence of multiple data points.
The area between data points is curved to account for the gradual increase in
uncertainty that occurs when leaving one data point but approaching another. Since
there are no data points in the future the uncertainty increases linearly and is not
attenuated by approaching data points.
61
Fig. 3-1: Uncertainty in urban models over time. In (a), only data for the present are included. For (b), three historical data sets are included in the modeling. (Figure and caption from Goldstein et. al., (2004))
Therefore, to create another even more distantly past layer for these counties, data
provided by Charles Dietzel as part of a Public Policy Institute of California (PPIC)
project (Teitz et. al., 2005) was attended to. The details of their data collection
methods can be found in their report (Ibid). However, their data provided a) the road
layers necessary to run SLEUTH, b) the DEM providing Slope information and c) the
Hillshade layer for display purposes. Dietzel also provided an urban layer dating
back to 1974, but no land use. Easier comparisons between results is another
advantage that lay in using the same data as the PPIC team. Of course, FMMP data
was mixed with their data in order to allow for the exploitation of the Land Use
portion of the model.
62
Land use is only half of the story to be told in this dissertation, though. The other
half concerns outlining which lands are in the Williamson Act and which are not, a
characteristic abstruse for even the most powerful satellites. By exploiting a
combination of Assessor’s data as well as Department of Conservation files, a portrait
of present and past Williamson Act landscapes could be painted (See Map 4-3 for an
example). This was accomplished by matching Assessor Parcel Numbers (APNs)
between the documents and the Assessor’s database. This record was then edited to
reflect a) whether the parcel is or was in the WA b) the parcel’s entry date into the
WA c) the parcel’s exit date from the WA and d) the parcel’s reason for exiting the
act. The documents declare a number of various mechanisms by which a landowner
may exit the Act. These sundry methods were aggregated into three main categories.
First, and most common, was non-renewal. This aspect of the Act, discussed in
Chapter 2, allows landowners to phase out their enrollment over a nine-year period.
During this time, as they are slowly ramped back up to normal taxing levels, they still
may not develop their land. The second, less common for number of parcels but
sometimes comprising tremendous acreage, is what has been termed by the author
“involuntary removals.” These include cases of eminent domain, public acquisitions,
rescissions, and other singular incidents that consist of the government brokering a
termination of Williamson Act enrollment for the property. The last category is
cancellation. This is initiated on the part of a landowner who is unwilling to wait
nine years in order to develop his or her land. These landowners must pay an
enormous fee but are rewarded with instant freedom from the Act’s restrictions.
63
Many APNs listed on the older State documents no longer exist in current Assessor’s
rolls. This mismatching increased the further back in time the documents reached.
Of course, this should be expected since county assessor’s offices often must, for
either new bookkeeping reasons or, more often, for reasons of parcel subdivision,
revise APNs. Attempts were made to track down these missing APNs.
Unfortunately, despite being availed of all the resources at the various Assessors’
offices, there remains a number of APNs that could not be located. In total, there
were 3938 acres of former Williamson Act lands in Tulare County unaccounted for
and 10,582 acres in Stanislaus County that could not be found. Merced County,
being new to the Act, had no missing APNs.
This underestimation, unfortunately, affected the Former Williamson Act (FWA)
model runs. More FWA blobs would mean greater possibilities for additional
growth. Therefore, even though the model runs suggest a great deal of parcels
leaving the WA in the future, if all of the FWA were tracked and integrated into input
data there would be even greater termination of contracts. It is somewhat safe,
however, to assume the missing acreage is located mostly near urban areas since
APNs are often revised due to either subdivisions or other lot-line adjustments taking
place (See Chapter 4). Therefore, since much of the recovered missing acreage lies
on currently developed land then it is likely that many of these unaccounted for
parcels are also inside city limits. Because these areas tend to already be bereft of
current WA contracts, then much of the ability of these parcels to encourage WA
termination may be greatly decayed by distance. Therefore, it is assumed that these
64
missing acres, if found, would not radically alter the model results, particularly since
their location may remain a permanent mystery. When examining maps of former
Williamson Act lands shown in Chapter 4, though, this missing acreage should be
kept in mind. Nonetheless, with the Assessor’s data edited properly for both counties
the already assembled land use data was ready to be combined with the Williamson
Act data.
Data Rendering
FMMP data and the Assessor’s data from all three counties were in different
coordinate systems and projections. Therefore, the Assessors’ data, which were all in
the State Plane projection, were converted into the California Albers Coordinate
projection used by the FMMP as well as the PPIC team for their data. Of course, the
Assessor’s layers do not line up perfectly well but with a little spatial adjustment they
were more than adequate (See Chapter 4 for various maps displaying these overlays).
Once this was accomplished the data could be displayed. This allowed different
patterns of both urban growth and Williamson Act change to become apparent.
Figure 3-2, FMMP Definitions of Important Farmland Categories (CA DOC)
Important Farmland Categories
About 90% of the FMMP's study area is covered by US Department of Agriculture (USDA) modern soil surveys. A classification system that combines technical soil ratings and current land use is the basis for the Important Farmland Maps of these lands. In areas where no soil survey is available, a series of Interim Farmland definitions have been developed to allow land use monitoring until soils data becomes available.
65
IMPORTANT FARMLAND MAP CATEGORIES
The colors and letters above are used to depict categories described below. The minimum mapping unit for all categories is 10 acres unless specified. Smaller units of land are incorporated into the surrounding map classifications. Prime Farmland (P) Farmland with the best combination of physical and chemical features able to sustain long term agricultural production. This land has the soil quality, growing season, and moisture supply needed to produce sustained high yields. Land must have been used for irrigated agricultural production at some time during the four years prior to the mapping date. Download information on the soils qualifying for Prime Farmland. More general information on the definition of Prime Farmland is also available. Farmland of Statewide Importance (S) Farmland similar to Prime Farmland but with minor shortcomings, such as greater slopes or less ability to store soil moisture. Land must have been used for irrigated agricultural production at some time during the four years prior to the mapping date. Download information on the soils qualifying for Farmland of Statewide Importance. Unique Farmland (U) Farmland of lesser quality soils used for the production of the state's leading agricultural crops. This land is usually irrigated, but may include nonirrigated orchards or vineyards as found in some climatic zones in California. Land must have been cropped at some time during the four years prior to the mapping date. Farmland of Local Importance (L) Land of importance to the local agricultural economy as determined by each county's board of supervisors and a local advisory committee. Download a complete set of the Farmland of Local Importance definitions in PDF format. Grazing Land (G) Land on which the existing vegetation is suited to the grazing of livestock. This category was developed in cooperation with the California Cattlemen's Association, University of California Cooperative Extension, and other groups interested in the extent of grazing activities. The minimum mapping unit for Grazing Land is 40 acres. Due to variations in soil quality, smaller units of Grazing Land may appear within larger irrigated pastures. Urban and Built-up Land (D) Land occupied by structures with a building density of at least 1 unit to 1.5 acres, or approximately 6 structures to a 10-acre parcel. This land is used for residential, industrial, commercial, construction, institutional, public administration, railroad and other transportation yards, cemeteries, airports, golf courses, sanitary landfills, sewage treatment, water control structures, and other developed purposes. Other Land (X) Land not included in any other mapping category. Common examples include low density rural developments; brush, timber, wetland, and riparian areas not suitable for livestock grazing; confined livestock, poultry or aquaculture facilities; strip mines, borrow pits; and water bodies smaller than forty acres. Vacant and nonagricultural land surrounded on all sides by urban development and greater than 40 acres is mapped as Other Land. Beginning in 2002, the pilot Rural Land Mapping Project provides more detail on the distribution of various land uses within the Other Land category in four San Joaquin Valley counties.
66
Water (W) Perennial water bodies with an extent of at least 40 acres. The main purpose for creating these maps initially was to explore them visually.
Since this project began with a simple interest to explore maps such as these, it was
helpful to finally examine the dynamic interplay between urban growth and
Williamson Act change. A series of gifs were created from the GIS project that
corresponded to the two-year intervals used by the FMMP (First and last years are
displayed in Chapter 4). For the now edited Assessor’s database, snapshots were
captured for each biennial period for the Williamson Act as well. These display both
the parcels that are in the WA, the parcels that are not, as well as those parcels that
were formerly in the WA. It should here be noted that the Assessor’s data for each
County reflects the year it was made and, therefore, in these maps from the past the
parcel outlines are not entirely accurate. In general, the further back in time from the
date of the Assessor’s data the fewer number of parcels actually existed. It was
deemed prohibitively difficult to find old Assessor’s maps that have been digitized, if
any even exist. Also, in those cases where subdivision has taken place on a former
WA parcel, all of the constituent parcels are treated as former WA parcels.
Though animation is difficult to convey in the pages of a dissertation, in this research
the gifs were carefully placed in a PowerPoint presentation so they lined up exactly
with each other. Then, by advancing through the presentation the maps would
animate and one could see the changes taking place. As the next Chapter illustrates,
it appeared visually that, indeed, those parcels leaving the WA were not randomly
distributed (Clarke, personal communication) and that, in fact, a case could be made
67
that the same phenomena cited by Clarke et. al. (1996a, 1996b, 1997, 1998) as the
causal factors in urban growth (Slope, proximity to urban areas, proximity to roads)
are also relevant for the spread of former Williamson Act (FWA) lands. The
differences in degree that each factor has on FWA parcels can, of course, be
discovered and quantified with SLEUTH. Armed with properly rendered data it was
time to delve into the modeling proper.
Modeling
The most time-intensive aspects of SLEUTH, and indeed in much of modeling, are
the data acquisition, rendering, and input. Since SLEUTH accepts data only in the
form of grayscale GIFs, each GIF must be derived from grids all in the same
projection and with the same number of rows and columns. In order to achieve this
all the necessary layers needed to be built. Tables 3-1 and 3-2 below display all of
the different GIFs used. The organization of the table has been borrowed from Teitz
et. al., (2005).
Table 3-1: WA Modeling Layers
Data Type Source Collection Method Description Resolution
Slope California Spatial Information Library (www.gis.ca.gov) (As provided by Dr. Dietzel)
Clipped and reclassified from San Joaquin for Tulare and Stanislaus / Merced
Digital Elevation Map that provides Slope information for one time period
30 m2
Land Use County Assessor’s data and DOC documents
Editing of Assessor’s data with DOC documents and research at Assessor’s offices
Offers 9 different WA status classifications for two time
+/- 40 feet
68
periods for 3 counties
Urban FMMP and USGS (offered by Dietzel) and Assessor’s data (for WA lands)
Aerial Photography (FMMP) Download (Me). MSS Imagery for USGS. Editing of Assessor’s data with DOC documents and research at Assessor’s offices.
Urban areas and Former WA lands are joined together to act and grow as an urban layer
1.5 buildings / acre for FMMP 30 m2 for USGS
Excluded Assessor’s data Areas not in WA are marked via editing of Assessor’s data
Lands not in the WA
+ / - 40 feet
Transportation Paper maps from Earth Science and Map Library or UC Berkeley. (Provided by Dietzel)
Subtraction from current Cal Trans GIS layer via date matching
Road networks for 4 time-periods classified according to accessibility
100 m2
Reciprocal Excluded Layer Background (Instead of Hillshade)
California Spatial Information Library (www.gis.ca.gov) (Provided by Dietzel) Assessor’s data for WA Lands.
Downloaded by PPIC team and then provided by Dietzel. Then clipped and reclassified from San Joaquin for Tulare and Stanislaus / Merced
This begins as a typical Excluded Layer (Publicly owned lands) but it also has WA lands accounted. It is overwritten with a probabilistic WA removal landscape that is then used for the urban runs.
100 m 2 for Publicly owned lands + / - 40 feet for WA lands
Table 3-2: Urban Modeling Layers Note: Originally aggregated by PPIC team in one-degree blocks. Then merged for all of California then clipped for San Joaquin Valley. Then clipped again and reclassified for both Tulare and then Stanislaus / Merced. Year Source Collection Method Description Resolution Slope California Spatial
Information Library (www.gis.ca.gov) (As provided by Dr. Dietzel)
Clipped and reclassified from San Joaquin for Tulare and Stanislaus / Merced
Digital Elevation Map that provides Slope information for one time period
30 m2
Land Use FMMP Aerial Photography Offers 6 10 acre
69
(FMMP) Download (Me)
possible Land Use classes for each county for two time periods
MMU
Urban FMMP and USGS (offered by Dietzel)
Aerial Photography (FMMP) Download (Me). MSS Imagery for USGS
Displays urban areas for the three counties for four time periods
1.5 buildings / acre
Excluded California Spatial Information Library (www.gis.ca.gov) (Provided by Dietzel) Assessor’s data for WA Lands.
Downloaded by PPIC team and then provided by Dietzel and WA. For calibration, 2002 public lands and WA lands used. For prediction, excluded landscape created through WA modeling module
Publicly Owned Lands for one time period as well as WA lands for calibration. Publicly owned lands and probabilistic modeled WA landscape for prediction
100 m 2 for Publicly owned lands + / - 40 feet for WA lands
Transportation Paper maps from Earth Science and Map Library or UC Berkeley. (Provided by Dietzel)
Subtraction from current Cal Trans GIS layer via date matching
Road networks for 4 time-periods classified according to accessibility
100 m2
Hillshade California Spatial Information Library (www.gis.ca.gov) (As provided by Dr. Dietzel)
Clipped and reclassified from San Joaquin for Tulare and Stanislaus / Merced
Digital Elevation Map that provides Slope information for one time period
30 m2
There are two tables displayed above, rather than one, because there are two aspects
being modeled. First, WA removal is simulated, and then the results are used to
create a probabilistic excluded layer for the urban growth and land use modeling
effort.
70
By examining the assembled maps and displaying simultaneously the growth of
development and the spread of Former Williamson Act (FWA) parcels it became
clear that the spread of FWA parcels, rather than being random, actually appeared to
follow a similar pattern to urban growth. In particular, proximity to urban areas and
transportation corridors appeared to have strong correspondence with leaving the Act
(See Maps in Chapter 4). Therefore, it is justified to use SLEUTH in order to
calibrate the data and offer a metrics output that would quantify what was being
observed. A score comparable to an urban growth run in the same area would
indicate a similar response to the urban growth stimuli of slope, urban areas, and
transportation corridors. For instance, Teitz. et. al., (2005) yielded a high Lee-Sallee
score of .32958 in their coarse calibration of Tulare County (See Figure 3-3 below).
Figure 3-3: The Lee-Sallee index (excerpted from Clarke et. al., 1996) shown below measures the average difference of growth between the control data years and the simulated years. This is specifically expressed by the ratio: the average of the intersection of known urban extent and simulated urban extent over the average of their union.
An FWA coarse calibration conducted in this dissertation yielded a Lee-Sallee high
of .44139, significantly higher than theirs (See Tables 3-4 and 3-5). However, it
should be noted that, in order to realize an accurate simulation of SLEUTH’s edge
effect, all Former WA lands as well as urban lands were treated as one monolithic
71
class. This was done in order to faithfully recreate the conditions that appear to
influence FWA spread. Since proximity to urban areas is one aspect that has been
asserted to indeed be important in FWA spread then the only way to model this,
within the confines of SLEUTH’s architecture, is to treat them the same. In this way,
new FWA lands can be created near urban areas as well as other FWA lands. This,
of course, affects the results, but refusal to combine them enervates urban areas’
effect on the creation of FWA parcels.
In a sense, the author has sought to fool SLEUTH into thinking it is simply modeling
urban growth and land use during the WA runs. But instead of modeling urban
growth it is modeling FWA spread. It would take a great deal of time to actually alter
SLEUTH’s underpinnings in such a way as to avoid the unfortunate necessity of this
conflation. Nevertheless, the combination of these two elements allows for a full
expression of SLEUTH’s growth rules. It should also be made clear that there is a
radically different Excluded Layer in use for the Williamson Act run (See Figures 4-2
and 4-9). In this case, only Williamson Act lands are available for FWA spread and
all other lands (including urban lands) are off-limits. Consequently, this can be
thought of as a “reciprocal excluded layer” since it is somewhat complementary with
the more traditional excluded layer (see Chapter 4) used in an urban growth
simulation. All those Williamson Act lands unavailable for development in the urban
growth simulation are, in this case, the only lands available for spread. By examining
output images in Chapter 4 it can be seen that FWA lands have spread tremendously.
72
Although Lee-Sallee was employed in order to compare metrics with the PPIC Report
(Teitz, et. al., 2005) a different metric was actually used to select between calibration
rounds. In fact, PPIC co-author Dr. Charles Dietzel offered a workshop whereby he
suggested using a product of seven of the built-in metrics and so it was decided to
utilize his research in this project ((2004), See Table 3-3 below for all the metrics
SLEUTH provides for use in calibration). After each round of calibration would be
complete the control stats output was converted into an Excel File. Then, what will
be termed the “Dietzel” (2004) product field was added and then sorted in a
descending order according to that field. However, for the sake of comparison,
Tables 3-4 and 3-5 display side-by-side the Lee-Sallee scores of each round of
calibration in order to compare with Teitz et. al. (2005). There was also another
difference in the approach. The PPIC team (Teitz et. al., 2005), though they
themselves used hierarchical resolution steps during calibration, gave commentary on
this approach’s problems. Since research has shown this method to “lead to
parameter sets that do not as accurately describe the growth of the system as a
calibration at full data resolution (Dietzel, 2004)” (excerpted from Teitz et. al., 2005,
p.94) the author opted to persist with full resolution throughout the entire calibration
process for each county, WA and urban simulation.
Table 3-3
73
Metrics That Can Be Used to Evaluate the Goodness of Fit of SLEUTH (as excerpted
from Teitz et. al, 2005) Note: Bold indicates additional metric added by author.
Metric Name
Description
Product
All other scores multiplied together
Compare
Modeled population for final year/actual population for final year, or IF Pmodeled > Pactual { 1 – (modeled population for final year/actual population for final year)}
Pop Least squares regression score for modeled urbanization compared to actual urbanization for the control years
Edges Least squares regression score for modeled urban edge count compared to actual urban edge count for the control years
Clusters Least squares regression score for modeled urban clustering compared to known urban clustering for the control years
Cluster size
Least squares regression score for modeled average urban cluster size compared to known average urban cluster size for the control years
Lee-Sallee
A shape index, a measurement of spatial fit between the model’s growth and the known urban extent for the control years
Slope Least squares regression of average slope for modeled urbanized cells compared to average slope of known urban cells for the control years
% urban Least squares regression of percentage of available pixels urbanized compared to the urbanized pixels for the control years
X-mean Least squares regression of average x_values for modeled urbanized cells compared to average x_values of known urban cells for the control years
Y-mean
Least squares regression of average y_values for modeled urbanized cells compared to average y_values of known urban cells for the control years
Rad Least squares regression of average radius of the circle which encloses the urban pixels
“Dietzel” Product of Compare, Pop, Edges, Clusters, Slope, X-Mean, Y-Mean (Dietzel, 2004)
Table 3-4 Routines and Results for Calibrating SLEUTH for Tulare County
PPIC Report (Teitz et. al., 2005)
WA Runs Urban Runs integrating 2002 WA into Excluded Layer
Coarse: Monte Carlo iterations = 3 Total no. of simulations = 3,125
Growth Parameters Range Step Diffusion 1–100 25 Breed 1–100 25 Spread 1–100 25
Coarse: Monte Carlo iterations = 4 Total no. of simulations = 3,125 Growth Parameters Range Step Diffusion 1–100 25 Breed 1–100 25 Spread 1–100 25
Coarse: Monte Carlo iterations = 3 Total no. of simulations = 3,125 Growth Parameters Range Step Diffusion 1–100 25 Breed 1–100 25 Spread 1–100 25
74
Slope resistance 1–100 25 Road gravity 1–100 25
Resulting Metrics Lee-Sallee = 0.32958
Slope resistance 1–100 25 Road gravity 1–100 25 Resulting Metrics Lee-Sallee = 0.44139
Dietzel Score = 0.616536
Slope resistance 1–100 25 Road gravity 1–100 25 Resulting Metrics
Lee-Sallee = 0.67246 Dietzel Score = 0.86165
Fine: Monte Carlo iterations = 5 Total no. of simulations = 2,160 Growth Parameters Range Step Diffusion 1–10 5 Breed 1–15 5 Spread 15-35 5 Slope resistance 1–50 10 Road gravity 1–50 10
Resulting Metrics Lee-Sallee = 0.32786
Fine: Monte Carlo iterations = 7 Total no.of simulations = 7776 Growth Parameters Range Step Diffusion 1–50 10 Breed 1–50 10 Spread 75-100 5 Slope resistance 25-75 10 Road gravity 1-50 10
Resulting Metrics Lee-Sallee = 0.3491 Dietzel Score = 0.619798
Fine: Monte Carlo iterations =5 Total no. of simulations = 7776 Growth Parameters Range Step Diffusion 75–100 5 Breed 5-100 5 Spread 0-25 5 Slope resistance 0-50 10 Road gravity 50-75 5 Resulting Metrics
Lee-Sallee = 0.58525 Dietzel Score = 0.905062
Final: Monte Carlo iterations = 7 Total no. of simulations = 3,456 Growth Parameters Range Step Diffusion 1–3 1 Breed 1–3 1 Spread 20-25 1 Slope resistance 1–50 10 Road gravity 1–10 2
Resulting Metrics Lee-Sallee = 0.31787
Final: Monte Carlo iterations = 8 Total no. of simulations = 7776 Growth Parameters Range Step Diffusion 1–25 5 Breed 1–25 5 Spread 85-100 3 Slope resistance 50-75 5 Road gravity 1-25 5 Resulting Metrics Lee-Sallee = 0.33685 Dietzel Score = 0.637348
Final: Monte Carlo iterations=8 Total no. of simulations = 7776 Growth Parameters Range StepDiffusion 85-100 3 Breed 80-90 2 Spread 0-10 2 Slope resistance 30-40 4 Road gravity 50-60 2 Resulting Metrics Lee-Sallee = 0.5786 Dietzel Score = 0.895966
Highest Values in Final Calibration:
Unknown
Highest Values in Final Calibration: Growth Parameters Final Value Diffusion 1 Breed 25 Spread 97 Slope resistance 65 Road gravity 1
Highest Values in Final Calibration: Growth Parameters Final Value Diffusion 85 Breed 82 Spread 1 Slope resistance 46 Road gravity 52
Self-Modified Parameter Value (SMP) (All Cases) Growth Parameters Final Value Diffusion 2 Breed 4 Spread 45 Slope resistance 1 Road gravity 2
Self-Modified Parameter Value (SMP) Growth Parameters Final Value Diffusion 1 Breed 33 Spread 100 Slope resistance 43 Road gravity 3
Self-Modified Parameter (SMP) Value Growth Parameters Final Value Diffusion 99 Breed 96 Spread 1 Slope resistance 1 Road gravity 57 Resulting Metrics Lee-Sallee = 0.56208
75
Dietzel Score = 0.080813
Table 3-5 Stanislaus and Merced Counties
PPIC Report (Teitz et. al., 2005) Stanislaus
PPIC Report Merced
Stanmerc WA Urban Runs integrating 2002 WA into Excluded Layer
Coarse: Monte Carlo iterations=3 Total no. of simulations = 3125 Growth Parameters Range Step Diffusion 1–100 25 Breed 1–100 25 Spread 1–100 25 Slope restnc 1–100 25 Road gravity 1–100 25
Resulting Metrics Lee-Sallee = 0.34441
Coarse: Monte Carlo iterations = 3 Total no. of simulations = 3125 Growth Parameters Range Step Diffusion 1–100 25 Breed 1–100 25 Spread 1–100 25 Slope restnc 1–100 25 Road gravity 1–100 25
Resulting Metrics Lee-Sallee = 0.26033
Coarse: Monte Carlo iterations=4 Total no. of simulations = 3125 Growth Parameters Range Step Diffusion 1–100 25 Breed 1–100 25 Spread 1–100 25 Slope restnc 1–100 25 Road gravity 1–100 25 Resulting Metrics Lee-Sallee = 0.52815 Dietzel Score = 0.505561
Coarse: Monte Carlo iterations = 4 Total no. of simulations = 3125 Growth Parameters Range Step Diffusion 1–100 25 Breed 1–100 25 Spread 1–100 25 Slope restnc 1–100 25 Road gravity 1–100 25 Resulting Metrics Lee-Sallee = 0.63475 Dietzel Score = 0.683694
Fine: Monte Carlo iterations=5 Total no. of simulations = 4500 Growth Parameters Range Step Diffusion 1–20 5 Breed 1–20 5 Spread 15-35 5 Slope restnc 25-50 5 Road gravity 50-100 10 Resulting MetricsLee-Sallee = 0.35935
Fine: Monte Carlo iterations = 5 Total no. of simulations = 5400 Growth Parameters Range Step Diffusion 1–25 5 Breed 1–25 5 Spread 15-35 5 Slope restnc 1-100 25 Road gravity 1-25 5 Resulting Metrics Lee-Sallee = 0.26585
Fine: Monte Carlo iterations=7 Total no. of simulations = 7776 Growth Parameters Range Step Diffusion 50-100 10 Breed 50-100 10 Spread 0-50 10 Slope restnc 0-25 5 Road gravity 0-50 10 Resulting Metrics Lee-Sallee = 0.51495 Dietzel Score = 0.767879
Fine: Monte Carlo iterations = 7 Total no. of simulations = 7776 Growth Parameters Range Step Diffusion 25-75 10 Breed 0-50 10 Spread 0-20 4 Slope restnc 0-25 5 Road gravity 50-100 10 Resulting Metrics Lee-Sallee = 0.73359 Dietzel Score = 0.768089
Final: Monte Carlo iterations=7 Total no. of simulations = 3456 Growth Parameters Range Step Diffusion 1–10 2 Breed 1–10 2 Spread 20-30 5 Slope restnc 40-50 2 Road gravity 80-100 5 Resulting Metrics Lee-Sallee = 0.34541
Final: Monte Carlo iterations = 7 Total no. of simulations = 3456 Growth Parameters Range Step Diffusion 1–5 1 Breed 1–5 1 Spread 18-23 1 Slope restnc 25-100 15 Road gravity 1-25 5
Resulting Metrics Lee-Sallee = 0.27036
Final: Monte Carlo iterations=8 Total no. of simulations = 3456 Growth Parameters Range Step Diffusion 50-75 5 Breed 50-75 5 Spread 5-15 2 Slope restnc 5-15 2 Road gravity 25-50 5 Resulting Metrics Lee-Sallee = 0.45644 Dietzel Score = 0.829079
Final: Monte Carlo iterations= 8 Total no. of simulations = 5184 Growth Parameters Range Step Diffusion 25-50 5 Breed 0-25 5 Spread 0-4 1 Slope restnc 0-15 3 Road gravity 75-100 5 Resulting Metrics Lee-Sallee = 0.73405 Dietzel Score = 0.777984
Highest Values in Final Calibration: NA
Highest Values in Final Calibration: NA
Highest Values in Final Calibration: Growth Parameters Value Diffusion 60 Breed 65 Spread 13 Slope resistance 7 Road gravity 40
Highest Values in Final Calibration: Growth Parameters Value Diffusion 40 Breed 10 Spread 3 Slope resistance 8 Road gravity 85
SMP Value (All Cases) Growth Parameters Final Value Diffusion 2 Breed 7 Spread 54
SMP Value (All Cases) Growth Parameters Final Value Diffusion 2 Breed 2 Spread 41
SMP Value (All Cases) Growth Parameters Final Value Diffusion 78 Breed 85 Spread 17
SMPValue (All Cases) Growth Parameters Final Value Diffusion 37 Breed 9 Spread 3
76
Slope resistance 29 Road gravity 100
Slope resistance 35 Road gravity 15
Slope resistance 1 Road gravity 44
Slope resistance 5 Road gravity 86 Resulting Metrics Lee-Sallee = 0.67033 Dietzel Score = 0.380748
Before continuing with an accounting of the particular approach used in this
dissertation it is now incumbent upon the author to briefly address SLEUTH’s
architecture. In order, the following pillars of its infrastructure will be addressed:
coefficients used by SLEUTH, its growth rules, and finally its self-modification
utility. Please note that greater detail and equations can be found on the Gigalopolis
website (from which the explanations below also borrow).
Coefficients
SLEUTH uses five different coefficients to describe and measure growth.
These are the very same coefficients that took so much time to calibrate and
re-calibrate, though this outcome depends greatly on the SMPs used in the
process. The coefficients include dispersion (initially referred to as
“diffusion” in the SLEUTH literature), breed, spread, slope, and road gravity.
The dispersion value affects two aspects of growth, spontaneous and road-
influenced. For spontaneous growth, dispersion controls the number of times
a pixel will be selected randomly for possible urbanization. At the same time,
the dispersion value also determines the maximum pixel distance from a cell
that can be searched for a road, allowing for further urbanization.
77
The breed coefficient influences both new spreading center growth and
road-influenced growth. A pixel already selected for spontaneous growth is
assigned an additional probability of becoming a new spreading center with
the breed variable. It also determines the number of searches that will be
made from an urbanized pixel during the road influenced growth portion of a
SLEUTH run.
Spread affects just edge growth by assigning a probability that a pixel in
the 3 x 3 neighborhood of a spreading center will generate additional urban
pixels. The slope coefficient, on the other hand, affects all types of growth by
forcing the model to consider the slope each time it examines a pixel for
possible urbanization. (There is a critical slope value, above which
urbanization is impossible, but this addressed in the SMPs.) The higher the
slope coefficient value, the less likely steeper slopes are to urbanize, and vice
versa. In essence, it describes the importance of slope in the urbanization
process. High values declare its importance while a value of one would
render it almost totally irrelevant. Finally, road gravity assigns a maximum
search distance from a pixel for a road from which to generate further
urbanization.
Growth Rules
78
As was discussed in the tour of coefficients above, there are four growth steps for
urban growth and two for “deltatron” dynamics used in land use change, which will
be addressed last. Spontaneous growth is the simple random urbanization of any
pixel anywhere on the lattice that is not already urbanized or in an excluded area. It
is controlled by the dispersion coefficient as well as the slope coefficient per the
explanation given previously. New Spreading Center Growth, controlled by the
breed coefficient and the slope coefficient, is next and determines whether or not any
of the newly urbanized cells created from the preceding spontaneous growth step will
become new spreading centers. It should be noted that this is conditioned by
availability of cells around the original cell for urbanization. Step three is edge
growth, whereby the pixels adjacent to all spreading centers, new and old, are
assessed a probability for urbanization based on the spread coefficient but also
conditioned, as always, by the slope coefficient. The final step, road influenced
growth, is determined by the breed coefficient, the dispersion coefficient, slope
coefficient, and, of course, the road gravity coefficient. Essentially, the road gravity
coefficient determines the maximum road search radius for an urbanized pixel. If a
road is found, then the closest pixel to the urbanized cell, but also adjacent to the
road, is temporarily urbanized. From this location a search is conducted along the
road, a “road trip”, with distance examined determined by dispersion. If any
available pixels are discovered then they are subject to random urbanization. Finally,
if this newly urbanized pixel has two adjacent neighbors that are also available then
they too become urbanized.
79
Although the land use change mode of SLEUTH was used for this dissertation, both
in the WA simulation and conventional land use change, this was more for the
purposes of examining the types of land lost rather than explicitly forecasting
Anderson level I land use change throughout the test counties. This was explained in
the beginning of this Chapter. Nevertheless, a brief explanation of the land use
change mode and its requisite use of “deltatrons” is in order. With two land-use maps
of the same area but two different time periods we can generate a difference matrix
for each pixel. SLEUTH will also calculate the average slope for each land class. As
far as predictive modeling, this is accomplished through the use of “deltatrons”,or
bringers of change (Candau, et. al., 2000), which don’t only initiate change but also
act as placeholders, marking where change has taken place and what kind. They also
keep track of lifecycles for a pixel’s land class so they may enforce spatial and
temporal autocorrelation. By using the urban growth aspect of SLEUTH as the driver
for the degree of land use change, the deltatrons execute their functions during two
phases of change. In phase one, random pixels are selected, with the number
contingent upon urban growth results, and the slope of these pixels are compared to
the slope of two other randomly selected land use classes. The two closest in slope
are then put into a transition probability matrix based on the difference maps created
and this is used to create land cover change. In step two, secondary land use changes
are created as a direct result of the earlier random changes. Clarke describes an
example thusly, “A single pixel may change from forest to urban. In the next step,
land immediately adjacent to the buildings may be cleared of trees for gardens and
farming (1997).” Therefore, through the search of each altered pixels neighborhood a
80
new transition probability matrix is created for associated growth and once again
change is enforced upon the prediction output.
Self-Modification Parameters
Self-modification parameters (SMPs) are the last aspects of SLEUTH’s growth
architecture that need to be addressed. Along with “boom and bust”, explained
below, there are three other additional user-controlled aspects of growth found in the
scenario-file along with the more dynamically functional SMPs. These are: a) slope
sensitivity, b) critical slope and c) road gravity sensitivity. Critical slope, briefly
addressed earlier, is simply the percent slope beyond which urbanization is
prohibited. The model does not discover this number on its own so the modeler must
explicitly define this. Consulting planning commission documents is one method for
achieving this step. However, like in the case of Tulare, slope is invoked (Tulare
County General Plan) in a manner to guide types of development rather than
forbidding development altogether. For instance, on page 5-8,
“Maximum Density: 1 DU/5 Acre if average cross slope is less than 30 percent. 1 DU/10 Acres if average cross slope is 30 percent or greater.”
Therefore, given that there is very little development beyond 30 percent or even
lower (especially given that the MMU for FMMP urban classification is 1.5
dwellings per acre), it was decided to defer to precedent, particularly as set by Clarke
(personal communication). Their critical slope was set at 25 percent, through a
process of trial and error, and, consequently, this is the value used here as well.
81
Slope sensitivity and road gravity sensitivity are simply additional multipliers to be
used with the their respective coefficients. Unfortunately, as discussed earlier in this
Chapter, there is, as of yet, no scientific process embedded within SLEUTH to
calibrate and discover these values, other than trial and error of course. Again,
therefore, Clarke (personal communication) and his values were deferred to. This
dissertation, it should be stated, is hopefully part of a trend of greater divulgence
concerning SMPS. In the literature review conducted in this research, no accounting
of these values in other applications was found. Since they greatly affect the
calibration as well as the prediction, offering coefficients without the concomitant
SMPs renders fellow researchers hamstrung, whether in the search for precedent in
their own efforts, or their ability to capitalize and expand upon the work of the
original researcher. It was fortunate, therefore, that at least one application (Clarke,
personal communication) was found upon which to justify the SMP selections.
However, as stated previously, these figures were not formally published but obtained
upon request personally.
The most important components of the SMPs are the “boom and bust.” These were
designed to simulate more realistically the growth cycles of regions. In the early
portion of a growth cycle, when there is an abundance of developable land,
urbanization is more rapid and, when it exceeds a certain level (defined by the critical
high) a “boom” can begin, which is a positive feedback that encourages even more
rapid growth. On the other hand, when a system is saturated due to lack of land or is
82
depressed with low growth for other reasons the growth rate can drop below the
critical low, at which time its growth is depressed even further by a multiplier less
than 1.0. Without these self-modification parameters SLEUTH could not achieve the
S curve that typifies actual growth in the world. Linear growth would result instead.
Once the WA runs were completed and a probabilistic excluded layer based on
patterns of Williamson Act termination was created (see Chapter 4 for images),
predictions according to three basic scenarios were executed: Strict Adherence to
current Williamson Act contracts, Business As Usual, and the Abolition of the
Williamson Act. These three scenarios are differentiated only by their respective
Excluded Layers. Output images for each of the scenarios are displayed in Chapter 4.
Strict Adherence to the Williamson Act (Strict for short-hand) is a scenario that
provides an extreme end of one spectrum. This future assumes that every WA
contract in existence in the year 2002 will persist until the end of the modeling runs in
2030. Therefore, the Excluded layer used for urban growth in this scenario is the
same used for calibration.
Business As Usual is the most sophisticated of the scenarios and exemplifies the
greatest contribution to knowledge offered in this work, since it utilizes the
probabilistic excluded layer created in the WA runs. The WA runs themselves,
incidentally, consist of one scenario only, since they have only one Excluded layer.
In fact, the WA runs’ greatest purpose is the realization of this business as usual
83
excluded layer for urban growth. Of course, they also offer a future regulatory
landscape in their own right.
The Abolition of the Williamson Act (or No WA for short) is actually quite a simple
scenario. This excluded layer is identical to that used by Teitz et. al. (2005). It
should be remembered, however, that the coefficients were derived using a different
excluded layer and, therefore, the application of the excluded layer used for the PPIC
report results in a great deal more land being available for development than was
assumed in calibration. One advantage, therefore, of WA integration during the
calibration phase is the ability to explore WA policy alteration during prediction.
Figure 3-3: Metrics output for prediction runs using SLEUTH (as excerpted from Gigalopolis website) run: a run consists of a single set of coefficient values and is executed MONTE_CARLO_ITERATIONS number of times from start to stop year year: the representative date for a growth cycle index: control year number sng: the number of new urban pixels generated from spontaneous growth sdg: the number of new urban pixels generated from new spreading center growth sdc: relic data type no longer used og: the number of new urban pixels generated from edge growth rt: the number of new urban pixels generated from road influenced growth pop: the total number of urban pixels area: the total number of urban pixels (same as pop) edges: the total number of urban/non-urban pixel edges clusters: the total number of urban clusters xmean: the average urban pixel column value ymean: the average urban pixel column value rad: the radius of the circle which encloses the urban area: (pow ((area / pi), 0.5)) slope: average slope of urbanized cells cl_size: average urban cluster size diffus: dispersion_coefficient value spread: spread_coefficient value breed: breed_coefficient value slp_res: slope_coefficient value rd_grav: road_gravity_coefficient value
84
%urban: Percent of the number of urban pixels divided by the total number of pixels in the study area (nrows*ncols) minus the number of pixels that are completely excluded from urban growth: ((100.0 * urbancount) / (total_pixels - (noncount+road pixels)) %road: Percent of the number of road pixels divided by the total number of pixels in the study area (nrows*ncols) minus the number of pixels that are completely excluded from urban growth: ((100.0 * roadcount) / (total_pixels - noncount)) grw_rate: Percent of the new urban pixels in one year divided by the total number of urban pixels: (100 * num_growth_pix / pop) leesalee: a shape index, a measurement of spatial fit between the model's growth and the known urban extent for the control years. In predict mode this value will always be zero (0):
where A is modeled and B is actual urban area. grw_pix: total number of new urban pixels
SLEUTH’s built-in metrics (see Figure 3-3) are then displayed to offer a sense of
total growth as well as dispersion, cluster, etc. Also, total land use acreages are
compared across the different scenarios. All of these output images, tables, maps,
and metrics are divulged in the following chapter.
CHAPTER 4: Results Part 1 (Text)
Introduction
The maps, figures, images, and tables that comprise this Chapter are the marrow of
the results gathered in this dissertation. Together, they continue the story begun in
the previous Chapter. Since incorporating the Williamson Act into the modeling
exercises was there shown to improve output metrics, then modeling future scenarios
85
could proceed. Preceding the images in this Chapter is explanatory text that provides
a companion narrative to the materials offered in the second half of the Chapter. The
most important images to examine are Figures 4-8 and 4-15 as they are a distillation
of the most essential contribution to knowledge offered in this dissertation: a method
for forecasting future excluded layers and using them to forecast urban growth.
In order to explore the various urban growth scenarios it was first necessary to
conduct a forecast of Williamson Act growth and terminations, the methodological
details of which were offered in Chapter 3. Though in the future there could be more,
in this case, there is only one Williamson Act scenario that required a SLEUTH
forecast: Business As Usual. This situation is predicated upon the assumption that
current Williamson Act administration and regulations will continue on to the year
2030. (See Figures 4-6 thru 4-8 and 4-13 thru 4-15.) The other two Williamson Act
scenarios either removed all Williamson Act protections (Abolition of the WA) or
assumed permanent protection of existing parcels (Strict Adherence). Each of these
three scenarios yielded an Excluded Layer that was then used in the traditional urban
modeling process.
To avoid confusion, though there are indeed three different scenarios involving the
WA, these scenarios are used to alter the excluded landscape of the urban growth
runs, not the WA itself. Of course there could be changes made to the WA excluded
layer (not to be confused with the urban growth excluded layer, see Figures 4-2 and
86
4-9 for WA excluded layers) in order to secure forecasts that correspond to different
policy options. For instance, if only certain WA parcels (like those of a certain
acreage threshold) were allowed to terminate their contracts then a different excluded
layer could be used for WA termination forecasting. Though this was not explored in
this dissertation, it is an area ripe for future research. For urban growth runs, each
scenario corresponds to a different excluded layer. There is no other difference
amongst the three forecasts. All were run with the same SMPs as well as growth
coefficients, within the respective counties.
This chapter offers a number of different maps, image outputs, and tables. Since each
figure and table was crafted carefully to fit on each page, maximizing the size of the
image yet efficiently taking up page space, the explanatory text for the results will
precede the maps, images, and tables, which are found at the end of the Chapter.
Examining the Past
As Chapter 3 explained, this dissertation research necessarily went through an
exploratory phase, where the first maps were created for the different time periods.
These maps’ value lay in the power of displaying both urban growth as well as WA
growth and termination (See Maps 4-1 through 4-7). They display what the author
suspected all along: WA termination is non-random. Using the most important data
exploration tool known, our eyes, (Clarke, 2000) it could be seen that those areas that,
87
according to SLEUTH, are more likely to urbanize also seemed to be valuable
predictors for WA contract termination. Also interesting were those parcels that were
tracked and displayed from protected farmland, to unprotected farmland, to
developed land. This land cycle speaks directly to the long-term efficacy of the WA.
Since animation is difficult to display on paper, only the years 1984 (1986 in Tulare’s
case) and 2002 are shown. This is enough, however, to notice the differences over
time. Several things should be easily discovered through their perusal. First, a
significant amount of urban growth occurred in both geographic areas from 1984 to
2002. Second, there was both a significant amount of parcels leaving the WA and, in
some cases, a considerable number of lands joining the WA. The simultaneity of
these processes gives the impression, when only examining overall numbers, that the
act enjoys a static presence in these counties. On the contrary, there is a great deal of
coming and going and the conveyance of this through these maps gives credence to
the far greater utility derived from the examination of phenomenon geographically
rather than through aggregate numbers alone. Third, those lands leaving the WA tend
to be near urban areas as well as roads. Though there are exceptions to this, many of
these are in fact due to extenuating circumstances. For instance, there exists an
enormous parcel in the extreme East of Tulare County (not displayed in Map 4-1 but
revealed in Figure 4-2). This parcel, owned by the National Forest Service and in an
area extremely remote, was nevertheless in the WA for a number of years. It left the
WA, but not for the purpose of housing developments. It is still currently owned by
the NFS and is not developed. Also, in Stanislaus County (See Map 4-5) the large
and numerous red parcels in the Southwest of that County were owned by the State of
88
California and removed from the Act through eminent domain for the purpose of
creating Henry W. Coe State Park. Hence, it was not a typical landowner decision
but a government decision and it was not for the purpose of development but a
necessary step in order to create the park.
Therefore, those parcels that most seem to confound the theory and the statistics that
WA parcel termination shows a predictability very similar to urban growth, are those
parcels most likely to not only be publicly acquired but also to not be destined for
urban growth. In this research, however, these outliers were kept and allowed to alter
the results. This was decided upon because it was necessary to create a WA
landscape in the future, both those parcels that left through landowner decisions as
well as those that left due to public acquisition or other such involuntary measures.
Only modeling and forecasting voluntary removals would paint an incomplete picture
of the WA. However, in the future a different approach could be employed that
parses out the different reasons for leaving and results in a more fastidious accounting
of the various mechanisms for contract termination.
Williamson Act Forecasting
Being convinced that there was an opportunity to model WA termination based on the
same criteria for urban growth, despite the difficulties associated with government
removals cited above, it was time to supply SLEUTH with the input images it needed
to run. Not all of the images are shown in this Chapter though a representative few
89
have been selected for display. In particular, only the urban inputs as well as the
excluded layers are shown. Figures 4-2 and 4-9 reveal the excluded layer for the WA
termination forecasting. Since only those parcels currently in the WA can actually
terminate their contracts, all non-WA lands are excluded from Former WA growth
(FWA). In these images, the color black, corresponding to 0 in the grayscale, is open
for contract termination and all grayscales with a value 100 or greater are excluded.
All black shown in this image are 2002 WA lands. The next images (Figures 4-3 and
4-10) are the 2002 “urban” input images for the WA runs. The word urban is used
even though, in actuality, the gray in this image is both urban land and FWA land in
the year 2002. The two were combined for reasons cited in Chapter 3. In the urban
layer, the value 0, appearing black, indicates non-urban land. The urban excluded
layer for Tulare County bears a certain reciprocity with the WA excluded layer for
Tulare County, hence the term “reciprocal excluded layers.” Specifically, those lands
not available for FWA growth are the only lands available for urban growth and those
lands off-limits for urban growth are those available for FWA growth. In the future
these reciprocal excluded layers (RELs) could be used to more tightly couple these
two models so they may exchange excluded layers each year, rather than through
final results only.
Figures 4-4 and 4-11 are the urban growth excluded layers for the strict adherence
scenario. The black (0 in the grayscale) are those lands not currently in the WA (in
2002). The two other shades of gray are both off-limits to development in this
scenario but are presented as different colors in order to differentiate WA lands from
90
National Forest lands and parklands. The Abolition of the WA scenario uses the
excluded layers shown in Figures 4-11 and 4-12. These figures simply ignore the
WA and so only treat parks and National Forest lands as off-limits to development.
The final scenario, Business As Usual, cannot proceed until the FWA growth
modeling occurs.
As discussed above, the Williamson Act termination modeling, or FWA growth
modeling, has only one excluded layer (Figures 4-2 and 4-9, respectively)
corresponding with its one scenario. For Tulare County, Figures 4-6 and 4-7 show
one Monte Carlo simulation using the coefficients generated through the calibration
process described in the previous chapter. The bright pastel colors, though somewhat
lacking aesthetically, were used to strongly differentiate these outcomes from the
traditional urban modeling outcomes, which use more subdued colors. One hundred
Monte Carlo simulations, similar to that shown in Figure 4-7, were run and the
probabilities were output in grayscale on top of the urban excluded layer
corresponding to strict adherence of the WA (i.e., all lands in the WA as well as all
public and parklands are shown as off-limits). This is displayed in Figure 4-8. The
fact that much of the image is as white as the background does not disrupt the
modeling process, since white (or 255 in the grayscale) is considered excluded. This
process creates the excluded layer for the third urban growth scenario: Business As
Usual, or the continuation of the current WA administration into the future. As
explained in Chapter 3, the number of times in the 100 Monte Carlo simulations a
particular parcel leaves the WA gives it a darker shade, i.e., a lower value on the
91
grayscale. This lower value then corresponds with greater availability for
development, with 0, or pitch black, representing no resistance to development
whatsoever, aside from possible Slope incompatibilities. For Tulare County, a
comparison of the Western portion of the County, as shown in figures 4-4 and 4-8,
reveals significant differences. Many areas currently under WA contract but near
urban areas left in nearly all of the 100 Monte Carlo simulations in Figure 4-8. There
is a distance decay effect moving away from the roads and urban areas that appears as
fuzziness between the black and the more solid outlying gray areas. Stanislaus and
Merced counties (hereafter referred to as StanMerc) had more dispersed WA
termination forecasting, resulting in wider patches of mid-level grays, rather than the
quicker decays found in Tulare County (Compare Figures 4-11 and 4-15). Also, the
diminished levels of WA termination in these two counties when compared to Tulare
can be attributed to the fact that Merced County has had zero terminations as of 2002,
affecting the overall results for the combined two counties. The reason for Merced’s
peculiar WA persistence is its very recent participation in the Act itself. It began its
administration of the program in 1998.
As for the actual WA forecasting itself, one Monte Carlo run is offered for the year
2030 for both geographic areas. This gives a general idea of what 2030 could look
like, but that the fact that the image is only one Monte Carlo simulation is important
to keep in mind. Therefore, the specific locations of the forecast FWA lands for this
one Monte Carlo simulation should be thoughtfully compared with the overlaid
probabilistic Excluded Layer that was created from 100 Monte Carlo simulations.
92
Nevertheless, even one simulation can offer insight. For Tulare County, Figure 4-7
presents a possible future of the WA. Though there is considerable acreage that is
added to the WA program, represented by dark green turning to light green and tan
turning to baby blue, a great deal is removed, as portrayed by the spread of red pixels.
As in urban growth, this spread is primarily near other urban and FWA areas as well
as along roads. Figures 4-17 a) and b) give a graphical representation of the amounts
of land in these two images in the form of a pie chart. Interestingly, when viewing
this land classification change with aggregate numbers alone, as was discussed in
Chapter 1, it would appear by looking at WA non-prime farmland that there was little
change. In fact, there was a dynamic interplay of non-prime acreage both leaving and
entering the WA in this forecast, rather than the stasis that these numbers alone might
suggest. As for FWA and Urban land, that more than triples between 2002 and 2030.
By examining the land use charts in Figure 4-18, the amount of FWA land alone can
be deduced as going from approximately 1.8% of Tulare County’s land to roughly
10% of all of Tulare County’s land. That is more than a five-fold increase. These
numbers were derived by assuming that the amount of urban land at the start of the
modeling cycle (Figure 4-18 a)) can be subtracted from the FWA and urban acreage
found in Figure 4-17 a), yielding 1.8%. The amount of Urban land in the Business As
Usual scenario for 2030 (Figure 4-18 b)) is 4.5 % that, when subtracted from 14.4%
is nearly 10%. This approach is inexact as these two modeling processes were run
separately, but it is a convenient way of differentiating between the FWA and urban
land that, as discussed earlier, were combined for methodologically expedient
reasons. It is also important to note that over the 28 years displayed in the two pie-
93
charts the amount of farmland that is neither in the WA nor was ever in the WA is an
ever-shrinking pool of acreage. Over half of the unprotected non-prime farmland in
2002 joined the WA while over half of the unprotected prime farmland joined as well.
As for which types of farmland left the WA, the charts reveal that this is
overwhelmingly prime farmland. Although 1.1% of all of Tulare County’s land in
2030, around 34,000 acres, is newly contracted prime WA farmland, the amount
leaving the WA far exceeds this value. The total amount of protected prime farmland
drops by nearly a third: 200,000 acres. That is only the net loss, however, since the
34,000 acres above were added during this process. The gross loss is 234,000 acres.
Only one fourth of the total 320,00 acres leaving the WA are non-prime, with a very
small fraction being other Land.
StanMerc, on the other hand, bears high dispersion values, but low spread values, and
very low slope resistance. Consequently, there is a very scattered effect for WA
contract termination (Refer to Tables 3-4 and 3-5 in Chapter 3 for reference). An
examination of 4-14 and 4-15 demonstrates the effect these coefficients have on the
growth of FWA land. Tulare County FWA results (see figures 4-7 and 4-8), on the
other hand, portray a vastly different pattern of growth. First, Tulare County FWA
has a maximum spread of 100 (Table 3-4) while also maintaining low dispersion and
considerable slope resistance. These last two factors, in combination, have a
straitening effect on FWA growth, and along with high spread, cause a very
concentrated growth of FWA land around areas already urban or FWA.
94
The total amount of FWA and urban land is more than doubled for StanMerc
throughout the modeling cycle (Figures 21 a) and b)). Of course, much of Merced
County was unprotected at the beginning of the modeling cycle and these lands are
excluded from becoming FWA. This also has a limiting effect on growth. Though
unprotected prime acreage in StanMerc declined very slightly, 74,000 prime acres
also left the WA during this time. Again, as with Tulare County, some of this FWA
and urban land was, in 2002, developed land already. By comparing the difference
between Figure 4-21 a) and 4-22 a) StanMerc is revealed to be 5.5% urban (120,000
acres) and 3.6% FWA (80,000 acres) in the beginning of the modeling cycle. By
2030, using the same method described above for Tulare County, StanMerc’s urban
acreage is 130,000 acres and its FWA acreage is 279,000 acres. Though the gains in
developed land were modest, FWA acreage more than tripled. Of those lands leaving
the WA, 126,000 were non-prime while 74,000 were prime. However, because many
parcels join the WA during this time, the net loss of protected non-prime farmland is
46,000 acres and only 4000 acres for prime. Also, since so much land was
unprotected in the beginning of the modeling run, even by 2030 fully 36% of
StanMerc’s agricultural land is forecast to not be protected nor to have ever been in
the WA (i.e., also not part of the FWA category). The WA added, but did not net,
80,000 acres of non-prime farmland and 71,000 acres of prime farmland. As with
Tulare County, the loss of other land is minor.
95
Urban Forecasting
As useful as modeling WA change may be, it still does not necessarily describe
changes on the ground. However, it does create a regulatory landscape that can affect
land cover, which will now be discussed. To reiterate Chapter 3, there are three
different scenarios under discussion for each geographic area. The only difference
between the three scenarios is the excluded layer. These scenarios are: 1. Business
As Usual (Continuance of current administration of the WA, excluded figures 4-8 and
4-15) 2. Strict Adherence to the WA (Present lands are frozen into the future,
excluded figures 4-4 and 4-11) and 3. Abolition of the WA (all current WA lands
become open for development, excluded figures 4-5 and 4-12). The final Monte
Carlo run for the years 2003 and 2030 for each of these scenarios and geographic
areas are displayed in Figures 4-16 a) – d) in the case of Tulare County and 4-20 a) –
d) for StanMerc.
Tulare County quite clearly shows increased growth, particularly in the Business As
Usual as well as the Abolition of the WA scenario. Strict Adherence, not
surprisingly, shows the least amount of difference with 2003. Profound growth along
major roads is another striking feature of these output images. By referring back to
Table 3-4 in the previous chapter, the final coefficients derived through calibration
can be seen for Tulare County urban growth. Using what has been discussed
concerning the nature of growth in SLEUTH and the particular role each of these
coefficients play, the output images for the different scenarios can be contextualized.
First, the tremendous growth along the roads is not only a function of the moderately
96
high road-gravity coefficient (57) but also the breed coefficient, bearing a very high
value of 96, and the dispersion (or diffusion) value, which is 99. Together, these
three values have a multiplicative effect on development along the roads. To review
Chapter 3, dispersion determines the maximum distance from a cell that a road search
can be conducted. Breed determines the number of searches that can be made from
an urbanized pixel during road influenced growth while road gravity itself determines
the maximum search distance from a pixel to search for a road, where further
urbanization can take place. The high dispersion coefficient is also the explanation
for the large number of clusters to be found during the spontaneous growth phase.
This becomes less apparent with greater WA retention since far-flung dispersed
development is not possible with all of those lands excluded. An examination of
Table 4-1 allows a comparison of Tulare County’s three different scenarios with the
base year of 2002, along a number of SLEUTH’s built-in metrics. As far as clusters,
even with Strict Adherence, the number of clusters more than doubles, while the
average cluster size is reduced in half. However, the radius for all the urban areas
only increases by 12 %. In the Business As usual scenario the number of clusters
increases more than seven fold, with average clusters less than a third the size they
were in 2002. The radius, meanwhile, increases by 42%. Finally, in the Abolition of
the WA scenario, the number of urban clusters increases twelve-fold with their
average size nearly one-sixth what they were in 2002. The urban radius increases
62% in this forecast.
97
Figures 4-18 a) thru d), offer pie charts of the different land breakdowns for not only
the three scenarios but for the base year of 2002 as well. Future urbanization ranges
from 18,000 acres for the Strict scenario to 111,000 for the Abolition scenario.
Business As Usual is closer to Abolition in its acreage lost to urbanization, with
71,000 acres, than it is to Strict enforcement. In the Strict Scenario, urban land
consumes 10,500 acres of non-prime farmland, 6500 acres of prime farmland, and
1200 acres of other Land. In Business As Usual, non-prime farmland is slightly more
favored as it comprises 47% of the lost acreage, with prime farmland making up 37%
and other Land equaling 15%. For reasons explained in Chapter 3, the various
farmland categories were kept static between 1984 and 2002, while only urbanization
changed. Therefore, there are no transitions between these various land use classes.
Consequently, all land use change is unidirectional, from the other three classes to
urban.
Figure 4-19 offers a two-dimensional examination of the land classes lost to
urbanization across the three different scenarios. It is helpful to simultaneously
peruse the 2030 image outputs of each of these scenarios while examining this figure,
in order to maximize the benefits of both the geographical and mathematical. In each
scenario, non-prime farmland is the greatest land use class lost to urbanization.
However, the percentage lost is different for each scenario. In the Strict scenario the
breakdown for prime, non-prime, and other land is as follows: 36%, 57%, and 7%,
respectively. For Business As Usual, these figures are: 36%, 48%, and 16%,
respectively. Finally, for the Abolition of the WA, the following results: 36%, 56%,
98
and 8%, respectively. The consistency of prime farmland’s proportion of land lost in
each scenario is the most significant fact to be ascertained from these figures. It
would appear that the proportion of available prime farmland, given the different
excluded layers as well as slope conditions, remains approximately equal across the
different scenarios. There is greater variability, on the other hand, between non-
prime farmland and other Land across the scenarios. Since much of the other Land
that is not public lands is actually in the WA and somewhat distant from urban areas,
the Strict scenario does not allow for much of its conversion. Therefore, the low
figures both in absolute numbers and percentages should not be surprising. Also, as
explained in the previous chapter, SLEUTH was designed to follow an S-curve of
urban growth. As available land disappears, growth slows down as it asymptotically
approaches full build-out. With such a decrease in available land, the actual rate of
urbanization is less.
The difference between Business as Usual and the Abolition of the WA concerning
the conversion of other Land may seem unexpected since Business As Usual actually
has more Other Land converted than the Abolition of the WA. However, this may be
due to a reduction in the amount of farmland available for development in the BAU
scenario, causing a greater demand for Other Land. Also, the variance inherent in
stochastic modeling may cause an aberrant figure due to the nature of that particular
Monte Carlo run.
99
Stanmerc, throughout all three scenarios, displayed far more restrained growth.
Figures 4-20 a) thru d) offer the output images for 2003 as well as the three scenarios
while Table 4-2 summarizes the output statistics. The cluster increase for Business
As Usual as well as Strict is very modest. The Abolition of the WA, on the other
hand, offers a tripling of the number of clusters. As far as average cluster size, the
Strict scenario leaves this value virtually unchanged from 2002. Business As Usual
shows a slight decrease in average size while the Abolition of the WA results in an
average cluster size only half as large as 2002.
In comparison to Tulare County, an examination of Figures 4-22 a) thru d),
demonstrates StanMerc’s more moderate growth, even during the Abolition of the
WA scenario. With the exception of the Abolition of the WA, the differences
between the scenarios and 2002 are not extreme. Even in the Abolition scenario,
however, developed land does not even double. Tulare County, on the other hand,
demonstrated more than double the urbanized land for even the Business As Usual
scenario when compared with its baseline 2002 year.
Figure 4-23 portrays the breakdown of acreage lost to urbanization across the three
scenarios. The most important figures to notice are the strikingly similar acreages
lost between the Strict Scenario and the Business As Usual scenario. As with Tulare
County, the Strict scenario has slightly more Other Land lost than Business As Usual.
This is also most likely due to similar reasons. A revisiting of the WA termination
modeling, along with an understanding of StanMerc’s urban growth coefficients,
100
sheds light on this issue. Since StanMerc’s WA terminations had a great dispersion
throughout the current WA landscape, the corresponding lowered resistance to the
forces of development was also dispersed (See figures 4-14 and 4-15). WA
terminations had a low spread value so WA lands near urban areas were not selected
for termination anywhere near as often over the course of the 100 Monte Carlo
simulations as they were for Tulare County. By comparing figure 4-15 with Figure 4-
8, the differences can be seen visually. While Tulare County has a great deal of
alloyed development resistance near urban areas, this falls off sharply with distance
from urban areas as well as roads. StanMerc, on the other hand, reveals its dispersed
WA termination nature through somewhat darker shades of gray spread more evenly
across the landscape. This works in tandem with StanMerc’s significantly lower
urban growth coefficients (See Tables 3-4 and 3-5). Consequently, StanMerc’s low
spread value (3) coupled with a relatively unenervated excluded layer near urban
areas causes minimal new growth around current development. Also, though
StanMerc has a reasonably strong dispersion value (37), most of the undeveloped
areas still offer reasonable resistance to development. Therefore, many of these cells
selected for possible development have an excluded value strong enough to prohibit
this from taking place. As a result, the difference between Strict Adherence and
Business As Usual is rather negligible. The Abolition of the WA, however, does
indeed encourage growth in StanMerc. As discussed earlier, this is not only because
lands previously selected for possible urbanization but denied due to exclusion are
now selected and available, but also because the greater availability of land affects
101
SLEUTH’s “boom and bust” self-modification parameters. Thus, greater growth is
encouraged due to the tremendous supply of available land.
Conclusion
The sum total of all the maps, figures, and tables offered in this Chapter shed light on
the past growth of these two regions of the Central Valley as well as offer possible
glimpses of their future. The effect of the WA on urban growth, whether it is
completely removed or softened over time by gradual contract termination, is clear in
both areas. With what has been examined in Chapter 3 as well as this Chapter, it is
hoped that SLEUTH has now been shown to be extremely valuable for modeling the
WA over time. The following and final Chapter will revisit the questions invoked in
Chapter 1 and explore how well they have been answered with these results as well as
possible future work that could build upon this dissertation.
102
Chapter 4: Results Part 2 (Maps, Figures, and Tables) Figure 4-1:
103
Part 1: Exploratory Maps
Map 4-1: Western Tulare County, 2002
104
Map 4-2, Visalia and Tulare 1986
105
Map 4-3, Visalia and Tulare 2002
106
Map 4-4: Stanislaus and Merced Counties, 1984
107
Map 4-5: Stanislaus and Merced Counties, 2002
108
Map 4-6: Modesto Metropolitan Area, 1984
109
Map 4-7: Modesto Metropolitan Area, 2002
110
Part 2: Input Images
111
Figure 4-2: Tulare excluded.wac. (Used for the Williamson Act Excluded Layer)
Gray are non-WA lands and are excluded. Black are lands in the WA. White is
water.
Figure 4-3: Tulare.urban.2002.wac
112
Gray is urban and Former WA lands. Black are lands that are neither urban nor
Former WA lands.
113
Figure 4-4: Tulare.excluded.c (Excluded Layer used for Strict Adherence Scenario)
Dark Gray are public lands. Light Gray are WA lands. Both are excluded. Black
are lands already developed or available for development. White is water.
114
Figure 4-5: Tulare.excluded.nowac (Excluded Layer used for Abolition of the WA scenario)
Gray are public lands and other protected areas off-limits to development. Black
are those lands open to development. White is water.
Part 3 Output Images:
115
Figure 4-6: 2003 Tulare County Williamson Act run
LANDUSE_CLASS= 0, Unclass is black LANDUSE_CLASS= 1, Former Williamson Act and Urban is red LANDUSE_CLASS= 2, Non-Prime non-WA is tan LANDUSE_CLASS= 3, Prime non-wa is dark green LANDUSE_CLASS= 4, Williamson_Act Non-Prime is light blue LANDUSE_CLASS= 5, Williamson_Act Prime is light green LANDUSE_CLASS= 6, Farmland Security Zone is light purple LANDUSE_CLASS= 7, Water is dark blue
LANDUSE_CLASS= 8, Other Land is dark purple
Figure 4-7: 2030 Tulare County Williamson Act run
116
LANDUSE_CLASS= 0, Unclass is black LANDUSE_CLASS= 1, Former Williamson Act and Urban is red LANDUSE_CLASS= 2, Non-Prime non-WA is tan LANDUSE_CLASS= 3, Prime non-wa is dark green LANDUSE_CLASS= 4, Williamson_Act Non-Prime is light blue LANDUSE_CLASS= 5, Williamson_Act Prime is light green LANDUSE_CLASS= 6, Farmland Security Zone is light purple LANDUSE_CLASS= 7, Water is dark blue LANDUSE_CLASS= 8, Other Land is dark purple
Figure 4-8: Excluded.bauc. Also, Tulare_cumcolor_urban_2030. This output becomes the excluded layer for the input.
117
White is completely off-limits. Solid Gray is WA lands. Fuzzy Gray is land with probability between 0 and 100 for being open to development. The darker the gray the more available the land is. Figure 4-9: Stanmerc.Excluded.Wa
118
Gray are non-WA lands and are excluded. Black are lands in the WA. White is water. Figure 4-10: Stanmerc.urban.2002.wa
119
Gray is urban and Former WA lands. Black are lands that are neither urban nor Former WA lands. Figure 4-11: Stanmerc.excluded.new (The excluded layer used for the Strict adherence scenario)
120
Dark Gray are public lands. Light Gray are WA lands. Both are excluded. Black
are lands already developed or available for development. White is water. Figure 4-12:Stanmerc.excluded.nowanew (The excluded layer used for the Abolition of the WA scenario)
121
Gray are public lands and other protected areas off-limits to development. Black are those lands open to development. White is water. Figure 4-13: StanmercWA 2003
122
LANDUSE_CLASS= 0, Unclass is black LANDUSE_CLASS= 1, Former Williamson Act and Urban is red LANDUSE_CLASS= 2, Non-Prime non-WA is tan LANDUSE_CLASS= 3, Prime non-wa is dark green LANDUSE_CLASS= 4, Williamson_Act Non-Prime is light blue LANDUSE_CLASS= 5, Williamson_Act Prime is light green LANDUSE_CLASS= 6, Farmland Security Zone is light purple LANDUSE_CLASS= 7, Water is dark blue LANDUSE_CLASS= 8, Other Land is dark purple Figure 4-14: Stanmerc.wa.2030
123
LANDUSE_CLASS= 0, Unclass is black LANDUSE_CLASS= 1, Former Williamson Act and Urban is red LANDUSE_CLASS= 2, Non-Prime non-WA is tan LANDUSE_CLASS= 3, Prime non-wa is dark green LANDUSE_CLASS= 4, Williamson_Act Non-Prime is light blue LANDUSE_CLASS= 5, Williamson_Act Prime is light green LANDUSE_CLASS= 6, Farmland Security Zone is light purple LANDUSE_CLASS= 7, Water is dark blue LANDUSE_CLASS= 8, Other Land is dark purple Figure 4-15: Stanmerc.excluded.baunew2
124
White is completely off-limits. Solid Gray is WA lands. Fuzzy Gray is land with probability between 0 and 100 for being open to development. The darker the gray the more available the land is.
SCENARIO IMAGES
125
126
TULARE Figure 4-16: a) Tulare in 2003. brown is urban, yellow is prime, orange is non-prime farmland and green is other land.
b) Strict Adherence, 2030 brown is urban, yellow is prime, orange is non-prime farmland and green is other land.
c) Business As Usual, 2030 brown is urban, yellow is prime, orange is non-prime farmland and green is other land.
d) Abolition of the WA, 2030 brown is urban, yellow is prime, orange is non-prime farmland and green is other land.
127
Table 4-1: Summary Statistics Table for Tulare County
year sng sdg og rt pop area edges cluster
s xmean y Base200
2 200
2 6845 6845 3295 308 193.7
NOWA 203
0 87 15
110
7 211789
21789
2 1306
8 3630 192.1
BAU 203
0 41 70 64 151391
61391
6 9291 2263 193.6
Strict 203
0 1 0 5 2 8633 8633 4744 777 196
Table Key: (as excerpted from Gigalopolis website) sng: the number of new urban pixels generated from “spontaneous” growth sdg: the number of new urban pixels generated from “new spreading center” growth og: the number of new urban pixels generated from “edge” growth rt: the number of new urban pixels generated from “road influenced” growth pop: the total number of urban pixels area: the total number of urban pixels (same as pop) edges: the total number of urban/non-urban pixel edges clusters: the total number of urban clusters xmean: the average urban pixel column value ymean: the average urban pixel column value rad: the radius of the circle which encloses the urban area: (pow ((area / pi), 0.5)) slope: average slope of urbanized cells cl_size: average urban cluster size
Figure 4-17 a) Tulare County WA Land Classification, 2002
128
129
Tulare WA Status: 2002
4.0%
4.5%
2.0%
17.9%
18.9%
0.2%
0.1%
52.3%
FWA and Urban
Non-WA Non-PrimeNon-WA Prime
WA Non-Prime
WA Prime
FarmlandSecurity ZonesWater
Other Land
b) Tulare County WA Land Classification,
Tulare WA Status: 2030
14.4%
2.0%
0.9%
17.8%
12.4%
52.2%
0.1%0.1%
FWA and UrbanNon-WA Non-PrimeNon-WA PrimeWA Non-PrimeWA PrimeFarmland Security ZonesWaterOther Land
Figure 4-18: a) Tulare County Land Use 2002
Tulare Land Use: 2002
Non-Prime Farmland
29.4%
Other Land56.0%
Water0.2%
Prime Farmland
12.3%
Urban2.2%
Figure b) Tulare Land Use Strict Adherence to WA, 2030
Tulare Land Use: 2030 (Strict)
Urban2.8%
Prime Farmland
12.0%
Non-Prime Farmland
29.1%
Other Land55.9%
Water0.2%
c) Tulare Land Use Business As Usual, 2030
130
Tulare Land Use: 2030 (BAU)
Urban4.5%
Prime Farmland
11.4%
Non-Prime Farmland
28.3%
Other Land55.6%
Water0.1%
d) Tulare Abolition of the WA, 2030
Tulare Land Use: 2030 (No WA)
Urban5.8%
Prime Farmland
10.9%
Non-Prime Farmland
27.4%
Other Land55.7%
Water0.2%
Figure 4-19: Tulare Land Converted to Urban by type and by scenario, 2030
131
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
Tulare Land Urbanized by Type and by Scenario: 2030
StrictBAUNo WA
Strict 6,532 10,435 1,264BAU 25,656 34,044 11,398No WA 40,475 61,987 8,649
Prime Farmland
Non-Prime Farmland Other Land
132
STANMERC Figure 4-20: a) Stanislaus and Merced Counties, 2003 brown is urban, yellow is prime, orange is non-prime farmland and green is other land.
b) Strict Adherence to WA, 2030 brown is urban, yellow is prime, orange is non-prime farmland and green is other land.
c) Business As Usual, 2030 brown is urban, yellow is prime, orange is non-prime farmland and green is other land.
133
d) Abolition of the WA, 2030 brown is urban, yellow is prime, orange is non-prime farmland and green is other land.
134
Table 4-2: Summary Statistics Table for Stanislaus / Merced Counties
year sn
g sd
g og rt pop areaedge
s cluster
s xmean ymean
2002 629
2629
2 3215 503 382.5 257.5 NoW
A 203
0 35 11 58 8949
5949
5 5381 1537 387.8 267.2
BAU 203
0 1 0 11 0673
8673
8 3422 566 383.6 259.3
Strict 203
0 1 0 10 0669
3669
3 3415 546 383.4 259.4
Table Key: (as excerpted from Gigalopolis website) sng: the number of new urban pixels generated from “spontaneous” growth sdg: the number of new urban pixels generated from “new spreading center” growth og: the number of new urban pixels generated from “edge” growth rt: the number of new urban pixels generated from “road influenced” growth pop: the total number of urban pixels area: the total number of urban pixels (same as pop) edges: the total number of urban/non-urban pixel edges clusters: the total number of urban clusters xmean: the average urban pixel column value ymean: the average urban pixel column value rad: the radius of the circle which encloses the urban area: (pow ((area / pi), 0.5)) slope: average slope of urbanized cells cl_size: average urban cluster size
135
136
Figure 4-21 a) Stanislaus and Merced Counties WA Classification, 2002
StanMerc WA Status: 2002
9.1%
31.2%
11.6%
31.7%
12.2%
1.0%
3.1%
FWA and UrbanNon-WA Non-PrimeNon-WA PrimeWA Non-PrimeWA PrimeWaterOther Land
b) Stanislaus and Merced Counties WA Classification, 2030
StanMerc WA Status: 2030
18.7%
27.6%
8.4%
29.6%
12.0%
1.0%
2.7%
FWA and UrbanNon-WA Non-PrimeNon-WA PrimeWA Non-PrimeWA PrimeWaterOther Land
Figure 4-22 a) Stanislaus and Merced Land Use, 2002
StanMerc Land Use: 2002
Urban5.5%
Prime Farmland
24.2%
Non-Prime Farmland
61.0%
Other Land8.2%
Water1.0%
b) Stanislaus/Merced Land Use, Strict Adherence to the WA, 2030
137
StanMerc Land Use: 2030 Strict
Urban5.9%
Prime Farmland
24.2%
Non-Prime Farmland
60.8%
Other Land8.0%
Water1.0%
c) Stanislaus/Merced Land Use, Business As Usual, 2030
StanMerc Land Use: 2030 BAU
Urban6.0%
Prime Farmland
24.2%
Other Land8.1%
Water1.0%
Non-Prime Farmland
60.8%
d) Stanislaus/Merced Land Use, Abolition of the WA, 2030
138
StanMerc Land Use: 2030 No WA
Urban8.4%
Prime Farmland
23.7%
Non-Prime Farmland
59.2%
Other Land7.6%
Water1.0%
139
Figure 4-23: Stanislaus and Merced Counties Land Converted to Urban by type and by scenario, 2030
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
40,000
StanMerc Acres Urbanized by Type and Scenario: 2030
StrictBAUNo WA
Strict 438 4,475 3,599BAU 458 5,784 3,079No WA 9,931 39,268 13,111
Prime Farmland
Non-Prime Farmland Other Land
140
CHAPTER 5: CONCLUSIONS
Research Questions Revisited
With the results offered in the previous chapter, along with the information presented
in Chapters 1 through 3, a number of different conclusions can be drawn. Also, as in
much academic investigation, new questions have been invoked that will stimulate
further research.
In the first chapter, questions were asked, and the subsequent chapters toured the
context, methodologies, and results involved in answering them. A brief recounting
of the questions and hypotheses is in order before turning to the conclusions proper.
• Question: Is the Williamson Act (both specifically and as representative of
other differential assessment programs) useful for modeling of urban spread?
• Hypothesis: The Williamson Act’s inclusion in urban growth modeling results
in greater accuracy than its exclusion.
By comparing the metrics used in this research with those used in the PPIC report
(Tietz et. al, 2005) those used in this dissertation proved closer to 1.0. These metric
values were higher even though different metrics were being used to selected
coefficient ranges between rounds of calibration (See Tables 3-4 and 3-5). Most
importantly, the only difference between the two approaches was this author’s
141
inclusion of the Williamson Act in the Excluded layer used in the rounds of
calibration.
• Question: Do spatial variables predict a parcel’s removal from the WA?
• Hypothesis: The same geographic phenomena that predictably apply
development pressure on undeveloped lands also apply pressure on
landowners to leave the Williamson Act.
A piece of land’s proximity to urban areas and to roads, as well as its slope, are all
considered to play a role in applying pressure on this land to develop. The creation
and examination of maps displaying these phenomena as well as the WA over time
demonstrates that indeed those leaving this voluntary Act tend to do so according to a
manner predictable by these stimuli. Since the calibration process used for the
Former WA modeling resulted in metric values greater than those used for urban
growth by the PPIC team (Tietz et. al, 2005), using the variables above to model this
landscape is justified. (See Tables 3-4 and 3-5)
• Question: How can knowledge of these spatial variables be used to model the
future of WA termination?
• Hypothesis: By designating Former WA lands as urban for the purposes of
SLEUTH’s nomenclature, a future landscape of the WA can be forecast with
accuracy comparable to urban growth forecasting.
142
Since it was determined that the same geographical phenomena that apply pressure on
lands to develop also apply pressure on lands to leave the WA, an urban growth
model, such as SLEUTH, is well-suited to forecast this future landscape. The
creation of appropriate excluded layers as well as naming Former WA lands as urban
lands, per SLEUTH’s requisite naming conventions, allows this process to proceed.
• Question: How can WA forecasts be used to influence urban growth
modeling?
• Hypothesis: A probabilistic excluded landscape can be created in a WA
modeling run and fed into a traditional urban modeling routine.
Historically, the “H” in SLEUTH has represented the Hillshade input layer. This
image has always been inert and serves only as the background for urban growth
displays. It can easily be removed and replaced with other images. Therefore, by
replacing the Hillshade layer with the Excluded layer used in the traditional urban
modeling calibrations, as well as creating a gray scale probability output in the
scenario file, a composite image is created during the FWA forecast over the 100 MC
iterations. This image can then be used as an Excluded layer for a Business As Usual
scenario for urban growth forecasts. Also, simply removing the WA from
consideration as well as assuming total persistence of current lands creates two other
scenarios based on Williamson Act administration variability.
As for the future of the Williamson Act itself, the results suggest that the voluntary
nature of the Act is both a blessing and a curse. It is a blessing because the lack of
143
coercion creates no friction with farmers and is often the only way for many of them
to remain solvent. Therefore, participation rates are enormous. Nevertheless, it is a
curse because farmers can still leave when they choose, though under most
circumstances it takes nine years. The future scenarios suggest that the rate of non-
renewal will eventually eclipse new enrollments leading to less and less protection of
those lands most vulnerable to development. In Tulare County, this is most readily
apparent. In the StanMerc region, this is not as clear since there are still so many
lands that have yet to join the WA. Nonetheless, it would only be a matter of time,
though further in the future than Tulare, that saturation will have taken place and then
contract terminations will begin to reduce the acreage every year thereafter.
Despite the eventual ineffectiveness of the Williamson Act that this model forecasts,
there still remains a remarkable difference between the Business As Usual scenario
and the Abolition of the WA scenario. Though the results suggest an eventual
weakening of the Act, its Abolition causes profoundly more farmland loss. As
mentioned in Chapter 2, the California Governor’s budget revisions have suggested
discontinuing subvention payments for the Williamson Act. This would effectively
destroy the Act as only a few extremely wealthy counties would assume the cost
themselves. As for the Strict Adherence scenario, though unlikely, there is precedent
for such permanence. In Eastern Long Island the differential assessment programs in
place there are enduring and transfer from owner to owner (Major M. Peguero,
personal communication). The tenability of implementing similar permanence in
California is, at least given the current endemic political and economic realities,
144
unlikely. Nevertheless, this dissertation can allow those in power to examine its
contents and perhaps make decisions armed with greater information.
Future Research
Any new piece of research, particularly as a graduate student, will usually contain a
certain degree of trial and error as well as a steep learning curve. This dissertation
was no exception. Consequently, throughout the course of this research a number of
lessons were learned as well as interesting questions raised that could lead to further
investigation.
First, there are issues of resolution that, if addressed, could help advance the accuracy
of these approaches. In particular, there is a parcel-to-pixel problem associated with
the Former Williamson Act modeling. Unlike urbanization, which can and does
occur one building at a time, Williamson Act parcels are either in or out of the Act
and these occurrences can happen instantly. Though the model is calibrated with data
that reflect these changes, future forecasts of a Williamson Act landscape are less
blocky and more refined than a true future cadastral landscape would be. (See
Figures 4-6 and 4-7 for comparison.) Therefore, future approaches would do well to
experiment with a more coarse resolution for the FWA forecasting than for the actual
urban growth modeling.
Second, there are time-delay effects that exist in the administration of the WA that are
more easily abjured in traditional urban modeling. For instance, though it is true that
145
a decision is made to build a home on a piece of land before the home is actually
built, this sort of time-delay does not need to be specifically addressed in a cellular
automata model such as SLEUTH for two reasons. First, SLEUTH is not an explicit
Human Decision Making model since it is specifically trying to capture urban growth
rather than the decision to urbanize amongst actors. Second, there is no universally
agreed upon time-delay between a decision to urbanize and actual development. The
Williamson Act, on the other hand, has an explicitly defined Non-Renewal condition
of termination. To review, this states that once a landowner requests to leave the
WA, there is a nine-year phase out period until that piece of land is officially
removed from the WA and available for urbanization. Since there are also other
methods of leaving the WA more or less instantly (cancellation, eminent domain,
public acquisition, etc.) it is challenging to combine these two types of WA
termination in the same modeling environment. SLEUTH, in this case, does not have
a user-defined option to allow for this sort of time-delay limbo between states.
Therefore, in the case of Non-Renewal, the code would need to be changed to
explicitly account for this time-delay between the decision to leave the Act and actual
removal. Also, experimenting with modeling non-renewal separately from other
forms of removal could not only help isolate this issue but also result, in some cases,
in greater accuracy since many remote WA parcels are purchased for the purposes of
park creation or other non-urban amenities, and therefore serve to confound the
rationale for using SLEUTH’s geographical phenomena (urban proximity, roads, etc.)
as criteria for WA removal.
146
This dissertation’s use of a cellular automata to forecast a regulatory landscape that
controls which lands may or may not be developed is, to this author’s knowledge,
novel. This could, however, be only the beginning of a new line of inquiry and use
for cellular automata models. Let us begin with the idea that, over an infinite period
of time, all lands on Earth will either be developed or protected from development. If
this is true than forecasting urban growth alone is asymmetrical. In every community
there is competition over land between environmentalists and developers, in a race to
develop or protect before the other side secures the fate of the land. By
systematically discovering, if at all possible, those factors relevant for the forecasting
of newly protected land then we may correct this current imbalance in modeling
priorities. In the case of the WA, this forecasting was made more amenable due to
the great similarity between factor’s influencing both Williamson Act removal and
urban growth. Nevertheless, a new cellular automata approach that addresses this
competition between protection and development would allow for a more refined
look at the future. A modeling environment that allows these two phenomena to
influence each other at each time step would be better still, whether they be coupled
models or housed in the same program.
Most modelers strive to make their tools more relevant to planners and planners
endeavor to improve their communities. Moving towards a modeling tool that offers
a myriad of scenarios, based on soundly tested theory and techniques could be
salutary to both. This dissertation is a step in that direction and this author intends to
continue research towards that end.
147
References
Acevedo, W. (1999). Analyzing Land Use Change in Urban Environments. USGS Fact Sheet. (http://landcover.usgs.gov/urban/info/factsht.pdf)
Agarwal, C., G. L. Green, M. Grove, T. Evans, and C. Schweik (2000). “A Review and Assessment of Land-Use Change Models” Dynamics of Space, Time, and Human Choice. Bloomington, IN: Center for the Study of Institutions Population, and Environmental Change, Indiana University.
Alberti, M. (1999). "Modeling the Urban Ecosystem: A Conceptual Framework." Environment and Planning B: Planning and Design 26: 605-630.
148
Allen, E. (2001). "INDEX: Software for Community Indicators" in Planning Support System: Integrating Geographic Information Systems, Models and Visualization Tools. Redlands, CA: Environmental Systems Research Institute and New Brunswick, NJ: Center for Urban Policy Research, (2001) (Edited with Richard K. Brail).
American Farmland Trust Website, a) “Fact Sheet: The Status of Local PACE Programs” b) “Fact Sheet: The Status of State PACE Programs” American Farmland Trust Website, http://www.farmland.org “The Farmland Protection Toolbox” August, 2003. Arthur, S. T. (2001). A satellite based scheme for predicting the effects of land cover change on local microclimate and surface hydrology. PhD Dissertation, Department of Meteorology, Pennsylvania State University. Arthur, S.T., T.N. Carlson, and D.A.J. Ripley (2000). “Land use dynamics of Chester County, Pennsylvania, from a satellite remote sensing perspective”. Geocarto International 15:25-35.
Benenson, I. and P. Torrens, (2005). Geosimulation. Chichester: John Wiley and Sons.
Berry, B. J. L. (1970). The geography of the United States in the year 2000. Transactions of the Institute of British Geographers 51: 21–53.
Berry, D. (1978) “Effects of Urbanization on Agricultural Activities.” Growth and Change 9.2-8
Boody, G., B. Vondracek, D. A. Andow, M. Krinke, J. Westra, J. Zimmerman, and P. Welle. (2005). Multifunctional agriculture in the United States. Bioscience 55:27-38
Boserup, E. (1965) The Conditions of Agricultural Growth: The Economics of Agrarian Change under Population Pressure. London, G. Allen and Unwin,; Chicago: Aldine.
Bradshaw, Ted K. and B. Muller, (1998) “Impacts of rapid urban growth on farmland conversion: Application of new regional land use policy models and geographical information systems”. Rural Sociology, 63(1): 1-25 Brand, Peter S. "Putting Agricultural Values into Dollars and Cents." (1995). California Coast and Ocean 11.3
California Department of Conservation:
149
a) http://www.consrv.ca.gov/DLRP/cfcp/stories/index.htm California Department of Conservation (2003), “The California Land Conservation Act: Status Report 2002” California Resources Agency. California: Government Code Section 51220 California: Government Code Section 16140-16154 California State Assembly, "Growth Challenges Facing the Golden State: A Series of Informational Hearings," ed. Smart Growth Caucus (Sacramento, 2001).
Candau, J., and K.C. Clarke. (2000). “Probabilistic land cover modeling using deltatrons.” Proceedings of the 38th Annual Conference of the Urban Regional Information Systems Association. Orlando, FL.
Candau, J., S. Rasmussen, and K.C. Clarke. (2000). Structure and Dynamics of a Coupled Cellular Automaton for Land Use/Land Cover Change. 4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4). Banff, Alberta, Canada, September.
Clarke, K. C. (1997). Land Transition Modeling With Deltatrons, Web Paper at: http://www.ncgia.ucsb.edu/conf/landuse97/
Clarke, K. C., and L. Gaydos (1998) "Loose Coupling A Cellular Automaton Model and GIS: Long-Term Growth Prediction for San Francisco and Washington/Baltimore" International Journal of Geographical Information Science, vol. 12, no. 7, pp. 699-714.
Clarke, K.C., Hoppen, S., Gaydos, L. (1996). "Methods and techniques for rigorous calibration
of a cellular automaton model of urban growth," Third International Conference/Workshop on Integrating GIS and Environmental Modeling, Santa Fe, New Mexico, January 21-25, 1996.
Clarke, K. C., Hoppen, S. and L. Gaydos (1997) "A self-modifying cellular automaton model of historical urbanization in the San Francisco Bay area" Environment and Planning B: Planning and Design, vol. 24, pp. 247-261.
Clarke, K. C. and Gaydos, L. (1998). "Long term urban growth prediction using a cellular automaton model and GIS: Applications in San Francisco and Washington/Baltimore", International Journal of Geographical Information Science. Cogan, C.B., F.W. Davis, and K.C. Clarke. (2001). “Application of urban growth models and wildlife habitat models to assess biodiversity losses.” University of California-Santa Barbara Institute for Computational Earth System Science. U.S.
150
department of the Interior, US geological Survey, Biological Resources Division, Gap Analysis Program, Santa Barbara, CA.
Couclelis H., (1985) "Cellular worlds: a framework for modeling micro -macro dynamics" Environment and Planning A 17 585 –596.
Couclelis, H., (1989) “Macrostructure and microbehavior in a metropolitan area” Environment and Planning B: Planning and Design 16 141-154.
Couclelis, H. (1991). "Geographically Informed Planning: Requirements for Planning-Relevant GIS." Papers in Regional Science 70(1): 9-19.
Couclelis, H. (1997). "From Cellular Automata to Urban Models: New Principles for Model Development and Implementation." Environment and Planning B: Planning and Design 24: 165-174.
Dean, John B. (1975): "A Panacea That Wasn't." Cry California 10.3 18.
Dietzel, C. and K. Clarke. 2004. “Determination of Optimal Calibration Metrics through the use of Self-Organizing Maps” The future of land use. Institute for Environmental Studies, Amsterdam. October 30th, 2004. Dietzel, Charles K., (2004) “Spatio-Temporal Difference in Model Outputs and Parameter Space as Determined by Calibration Extent,” in P. Atkinson, G. Foody, S. Darby, and F. Wu (eds), Geodynamics, CRC Press, Boca Raton, Florida.
Dresslar, John. "Agricultural Land Preservation in California: Time for a New View." Ecology Law Quarterly 8.2 (1979): 303.
Fischel, W.A., “Urbanization of Agricultural Lands: A Review of the National Agricultural Lands Study.” Land Econ. 58(1982):236-59. Forrester, J. W. (1969). Urban Dynamics. Cambridge, Massachusetts and London, England, The M.I.T. Press. Gigalopolis Website: http://www.ncgia.ucsb.edu/projects/gig/project_gig.htm
Golden State Museum website: http://www.ss.ca.gov/museum/intro.htm
Goldspink, C. 2004. Proceedings from Workshop conducted at University of Queensland, Australia. Available at: http://www.business.uq.edu.au/events/speakers/cgoldspink_paper.pdf
151
Goldstein, N.C., J.T. Candau, and K.C. Clarke. 2004. “Approaches to simulating the “March of Bricks And Mortar””. Computers, Environment and Urban Systems 28:125-147.
Gustafson, G.C. , and N.L. Bills, (1984) “U.S. Cropland, Urbanization, and Landownership Patterns.” AER-520. Washing DC: US Department of Agriculture, Econ. Res. Serv., November, 1984.
Handel, Mary E., “Conflict on The Urban Fringe,” in California Farmland and Urban Pressures, Albert Medvitz, Alvin Sokolow, and Cathy Lemp, eds. Agricultural Issues Center, UC Davis, 1999.
Harris, B. (1985). "Urban Simulation Models in Regional Science." Journal of Regional Science 25(4): 545-567.
Herold, M., N.C. Goldstein, and K.C. Clarke. 2003. “The spatio-temporal form of urban growth: measurement, analysis and modeling.” Remote Sensing of Environment 86:286-302.
Hvolboll, Eric. 2005. Goleta Farmer. From interview conducted by Noelle Boucquey.
Ingerson, T.E. and R.L. Buvel (1984). “Structure in asynchronous cellular automata.” Physica D 10: 59-68.
Klosterman, R. (1999). "The What if? Collaborative Planning Support System." in Planning Support System: Integrating Geographic Information Systems, Models and Visualization Tools. Redlands, CA: Environmental Systems Research Institute and New Brunswick, NJ: Center for Urban Policy Research, (2001) (Edited with Richard K. Brail).
Knickerbocker, B. (2007), Christian Science Monitor, January 12, 2007 (http://www.csmonitor.com/2007/0112/p25s02-wogi.html)
Kuminoff, Nicolai V., Alvin D. Sokolow and Daniel A. Sumner, “Farmland Conversion: Perception and Realities”, University of California, Davis, Agricultural Issues Center, Issues Brief no. 16, 2001.
Kuminoff, Nicolai V., and Daniel A. Sumner, “Modeling Farmland Conversion with new GIS Data”, University of California, Davis, Agricultural Issues Center, Paper prepared for the annual meeting of the American Agricultural Economics Association, Chicago, August 5-8, 2001. Landis, J. D. (1995). "Imagining Land Use Futures: Applying the California Urban Futures Model." Journal of the American Planning Association 61(4): 438-457.
152
Landis, J.D. (2001). "CUF, CUF II, and CURBA: A Family of Spatially Explicit Urban Growth and Land-Use Policy Simulation Models " in Planning Support Systems: Integrating Geographic Information Systems, Models and Visualization Tools. Redlands, CA: Environmental Systems Research (2001).
Lane, John. 2005. Lane Farms. From interview conducted by Noelle Boucquey.
Leão, S., I. Bishop, and D. Evans. 2004. “Spatial-temporal model for demand allocation of waste landfills in growing urban regions.” Computers Environment and Urban Systems 28: 353-385.
Lee, D. B., Jr. (1973). "Requiem for Large-Scale Models." AIP Journal 39(3): 146-178.
Li, X., & Yeh, A. (2000). Modeling sustainable urban development by the integration of constrained cellular automata and GIS. International Journal of Geographical Information Science, 14(2), 131-152.
Logan, John R., and Harvey L. Molotch. 1987. Urban Fortunes. Berkeley: University of California Press.
Lopez, R. A.; A. O. Adelaja; M. S. Andrews (1988) “The Effects of Suburbanization on Agriculture”, American Journal of Agricultural Economics 70(2): 346-358.
Machado, E. A., D. M. Stoms, F. W. Davis, and J. Kreitler. 2006. Prioritizing farmland preservation cost-effectively for multiple objectives. Journal of Soil and Water Conservation 61: 250-258.
Malthus, Thomas Robert, An Essay on the Principle of Population. J. Johnson. 1798. Library of Economics and Liberty. 24 January 2007 Medvitz, Albert G., “Population Growth and Its Impacts on Agricultural Land in California: 1850 to 1998”, in California Farmland and Urban Pressures, Albert Medvitz, Alvin Sokolow, and Cathy Lemp, eds. Agricultural Issues Center, UC Davis, 1999.
Merced County Assessor’s Office
Mercer, D.C., and Powell, J.M. (1972): Phenomenology and related non-positivistic viewpoints in the social sciences. Clayton, Victoria, Australia: Monash Publications in Geography, No. 1. Mertens, Benoît; Lambin, Eric F. 1997. “Spatial modeling of deforestation in southern Cameroon.” Applied Geography. 17(2): 143–162.
153
Muth, R.F., (1961) “Economic Change and Rural-Urban Land Conversions, “ Econometrica 29(1961):1-23.
New York State Department of Environmental Conservation: http://www.dec.state.ny.us/
National Agricultural Statistics Service (NASS), Agricultural Statistics Database website: http://www.nass.usda.gov/Data_and_Statistics/Quick_Stats/
Oglethorpe, D. R.; O’Callaghan, J. R. 1995. “Farm-level economic modelling within a river catchment decision support system.” Journal of Environmental Planning and Management. 38(5): 93–106
Onsted, J. 2002. “SCOPE: A modification and application of the Forrester model to the South Coast of Santa Barbara.” Master’s Thesis, UC Santa Barbara. http://zenith.geog.ucsb.edu/title.html
Parker, D. and V. Meretsky. 2004. “Measuring pattern outcomes in an agent-based model of edge-effect externalities using spatial metrics”. Agriculture, Ecosystems, and Environment. 101(2-3) p. 233-250.
Pennsylvania Farmland Preservation Association: http://www.pafarmland.org/why_preserve_farmland.htm
Pickles, J. (1999). “Arguments, Debates, and Dialogues: the GIS-social Theory Debate and the Concern for Alternatives” In P. A. Longley, M. F. Goodchild, D. J. MacGuire, and D. W. Rhind (eds.). Geographical Information Systems: Principles, Techniques, Applications, and Management (2nd ed.). New York: John Wiley and Sons.
Plaut, T.R.,(1980) “Urbanization Expansion and the Loss of Farmland in the United States: Implications for the Future.” American J. Agr. Econ. 62(1980): 537-42.
Putman, S. (2001). "The METROPILUS Planning Support System: Urban Models and GIS" in Planning Support System: Integrating Geographic Information Systems, Models and Visualization Tools. Redlands, CA: Environmental Systems Research Institute and New Brunswick, NJ: Center for Urban Policy Research, (2001) (Edited with Richard K. Brail).
Ramankutty, N., and J. A. Foley. 1999. “Estimating Historical Changes in Global Land Cover: Croplands from 1700 to 1992.” Global Biogeochemical Cycles 13(4):997–1028.
154
Sanders, L., Pumain, D., Mathian, H., Guerin-Pace, F., and Bura, S. 1997. SIMPOP: a multiagent system for the study of urbanism. Environment and Planning B 24, 287-305.
Santa Barbara County Planning and Development. 2002. Goleta Valley Urban Agricultural Newsletter. http://www.countyofsb.org/plandev/comp/programs/Newsletters/guan/default.asp
Schonfisch, B. and A.M. de Roos (1999). “Synchronous and asynchronous updating in cellular automata.” Biosystems 51: 123-143.
Sembolini, (2000). The growth of an urban cluster into a dynamic self-modifying spatial pattern. Environment and Planning B – Planning and Design 27(4); 549-564.
Silva, E. A. and K. C. Clarke (2002) “Calibration of the SLEUTH urban growth model for Lisbon and Porto, Portugal.” Computers, Environment and Urban Systems, Volume 26, Issue 6, November 2002, Pages 525-552.
Sokolow, Alvin D. Ed. "The Williamson Act: 25 Years of Land Conservation." The Resources Agency of California: Sacramento, 1990. Sokolow, Alvin D., 1997. “Farmland Policy in California’s Central Valley: State, County, and City Roles." CPS Brief, v. 9, n. 4 (October 1997).
Sokolow, Alvin D. and Julie Spezia, “Farmland Protection Policy”. California Policy Choices. Volume 8, pp. 151-168. 1992.
Solecki, W.D., and C. Oliveri. 2004. “Downscaling climate change scenarios in an urban land use change model.” Journal of Environmental Management 72:105-115.
Sorensen, A. Ann, Richard P. Greene, and Karen Russ, “Farming on the Edge”, American Farmland Trust, Center for Agriculture in the Environment, Northern Illinois University, Dekalb, Illinois, March 1997.
Stanislaus County Assessor’s Office
Tayman, J. (1996). "The Accuracy of Small-Area Population Forecasts Based on a Spatial Interaction Land-Use Modeling System." APA Journal 62(1): 85-98.
Teitz, M., Dietzel, C., Fulton, B. (2005). Urban development futures in the San Joaquin Valley. Public Policy Institute of California: San Francisco.
Tuan, Yi-Fu (1971): “Geography, phenomenology, and the study of human nature.” The Canadian Geographer 15, 181-92.
155
156
Tulare County Assessor’s Office Tulare County General Plan http://www.westplanning.com/docs/tulare/documents/draft_gpr/tulareco_draft_gpr_component_b_2006_11.pdf)
USDA Website: “A History of American Agriculture 1776-1990“ http://www.usda.gov/history2/text11.htm
Verburg, Peter H. and others. “Land use change modelling: current practice and research priorities.” GeoJournal [Dordrecht] Vol. 61. No. 4. 2004. p. 309-324.
Vitousek, P. M., H. A. Mooney, J. Lubchenco, and J. M. Melillo. 1997. “Human Domination of Earth’s Ecosystems”. Science 277(15 July):494–499.
Von Thunen, J. H. Der Isolierte Staat. Translated by C.M. Wartenberg. In P. Hall, ed., Von Thunen’s Isolated State (Elmsford, N.Y.: Pergamon, 1966). Wegener, M. (1994). "Operatonal Urban Models: state of the art." AIP Journal 60(1): 17-30.
Williamson Act Study Group, “Preserving Agricultural Land in California: A Short History of the Williamson Act”, University of California, Davis, Agricultural Issues Center, 1989, pp. 13-15. California Department of Conservation , http://www.consrv.ca.gov/DLRP/fmmp/stats_reports/conversion_tables_historic.htm