United States
Environmental Protection
Agency
EPA/600/R-08/076F | June 2009 | www.epa.gov
Land-Use Scenarios:
National-Scale Housing-Density
Scenarios Consistent with
Climate Change Storylines
Office of Research and Development, Washington, DC 20460
National Center for Environmental Assessment
-------
EPA/600/R-08/076F
June 2009
Land-Use Scenarios:
National-Scale Housing-Density Scenarios Consistent with Climate
Change Storylines
Global Change Research Program
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Washington, DC 20460
-------
DISCLAIMER
This document has been reviewed in accordance with U.S. Environmental Protection
Agency policy and approved for publication. Mention of trade names or commercial products
does not constitute endorsement or recommendation for use.
ABSTRACT
Climate and land-use change are major components of global environmental change with
feedbacks between these components. The consequences of these interactions show that land use
may exacerbate or alleviate climate-change effects. Based on these findings it is important to use
land-use scenarios that are consistent with the specific assumptions underlying climate-change
scenarios. The Integrated Climate and Land-Use Scenarios (ICLUS) project developed land-use
outputs that are based on the Intergovernmental Panel on Climate Change (IPCC) Special Report
on Emissions Scenarios (SRES) social, economic, and demographic storylines and adapted these
to the United States. ICLUS outputs are derived from a pair of models. A demographic model
generates population estimates that are distributed by the spatial allocation model as housing
density (HD) across the landscape; land-use outputs were developed for the four main SRES
storylines and a base case. The model is run for the conterminous United States and output is
available for each scenario by decade to 2100. In addition to maps of HD across the
conterminous United States, this project also generated maps of impervious surface (IS) cover
based on the HD projections. This report describes the modeling methodology for, some initial
analyses using ICLUS outputs, and recommendations for further research.
Preferred Citation:
U.S. Environmental Protection Agency (EPA). (2009) Land-Use Scenarios: National-Scale Housing-Density
Scenarios Consistent with Climate Change Storylines. Global Change Research Program, National Center for
Environmental Assessment, Washington, DC; EPA/600/R-08/076F. Available from: National Technical
Information Service, Springfield, VA, and online at http://www.epa.gov/ncea.
11
-------
CONTENTS
LIST OF TABLES v
LIST OF FIGURES vi
LIST OF ABBREVIATIONS AND ACRONYMS ix
PREFACE x
AUTHORS AND REVIEWERS xi
ACKNOWLEDGMENTS xi
EXECUTIVE SUMMARY xii
1. INTRODUCTION 1-1
2. ADAPTING THE SRES TO THE UNITED STATES 2-1
2.1. OVERVIEW OF THE SRES STORYLINES 2-2
2.2. INTERPRETING AND ADAPTING THE SRES STORYLINES 2-5
2.2.1. Al Storyline Adapted for the United States 2-5
2.2.2. Bl Storyline Adapted for the United States 2-6
2.2.3. A2 Storyline Adapted for the United States 2-6
2.2.4. B2 Storyline Adapted for the United States 2-6
3. DEMOGRAPHIC PROJECTIONS 3-1
3.1. OVERVIEW 3-1
3.2. INITIAL POPULATION 3-3
3.3. FERTILITY AND MORTALITY 3-4
3.4. NET INTERNATIONAL MIGRATION 3-5
3.5. DOMESTIC MIGRATION AND GRAVITY MODELING 3-6
3.5.1. 1995-2000 Migration Data 3-7
3.5.2. County Attribute Data 3-9
3.5.3. Distance Matrix 3-9
3.5.4. Model Specification 3-11
3.5.5. Stepwise Regression Results 3-15
3.5.6. Model Flow 3-15
3.5.7. Gravity Model and U.S.-Adapted SRES Storylines 3-16
3.6. MODEL RESULTS 3-16
3.7. COMPARISON OF DEMOGRAPHIC MODEL WITH EXISTING
PROJECTIONS 3-18
4. SPATIAL ALLOCATION MODEL 4-1
4.1. RATIONALE FOR THE SELECTION OF SERGoM 4-1
4.2. METHODOLOGY 4-3
4.3. INCORPORATING U.S.-ADAPTED SRES INTO SERGoM 4-6
4.4. INTEGRATION OF DEMOGRAPHIC, SERGoM, AND IMPERVIOUS
MODELS 4-8
in
-------
CONTENTS (continued)
5. IMPACTS AND INDICATORS ANALYSIS 5-1
5.1. RATES OF GROWTH IN DIFFERENT REGIONS 5-1
5.2. HOUSING DENSITY TRENDS 5-8
5.3. IMPERVIOUS SURFACES 5-10
5.3.1. Impervious Surface Calculations 5-11
5.3.2. Percent of Watersheds Over 5% Impervious 5-11
5.4. OPTIONS FOR FUTURE STUDY 5-23
6. DISCUSSION AND CONCLUSIONS 6-1
REFERENCES R-l
APPENDIX A: MAPS FOR ICLUS SCENARIOS A-l
APPENDIX B: DEMOGRAPHIC MODEL SENSITIVITY TESTING B-l
APPENDIX C: STATISTICAL RELATIONSHIP BETWEEN HOUSING DENSITY
AND IMPERVIOUS SURFACE COVER C-l
APPENDIX D: REGIONAL POPULATION GROWTH RATES AND PROJECTIONS
BASED ON EPA REGIONS D-l
APPENDIX E: COMPONENT AND COHORT MODEL DATA E-l
APPENDIX F: SUMMARY OF MAJOR MODEL INPUTS AND ASSUMPTIONS F-l
IV
-------
LIST OF TABLES
2-1. Qualitative demographic assumptions in global SRES storylines 2-3
3-1. Summary of qualitative adjustments to the demographic projections 3-3
3-2. Gravity model results 3-13
3-3. Ranking of the contribution of independent variables to explanatory power of the
models 3-14
4-1. Summary of adjustments to SERGoM v3 for SRES scenarios 4-7
5-1. U.S. Census regions 5-1
5-2. Projected regional populations and growth rates 5-2
5-3. Projected urban and suburban area increases in modeled scenarios, 2000-2100
(km2) 5-9
5-4. Projected area (km2) effects of urban, suburban, and exurban housing densities on
NLCD 2001 land cover types in modeled scenarios for 2050 5-10
5-5. Impervious surface estimates for SRES scenarios 5-12
D-l. EPA regions D-2
D-2. Projected population and growth rate by scenario and EPA region D-3
E-l. Fertility rates (births per 1000 women) E-2
E-2. Mortality rates (lifespan equivalent) E-6
E-3. Projected international migration E-7
F-l. Major demographic model inputs and assumptions F-2
F-2. Major SERGoM inputs and assumptions F-3
-------
LIST OF FIGURES
1-1. Model and information flow within the ICLUS project 1-2
2-1. Illustration of the SRES families along two dimensions that indicate the relative
orientation of the storylines along the axes of global or regional development
and economic or environmental concerns 2-3
3-1. Demographic model flow 3-2
3-2. Projected net international migrations under U.S. Census Bureau Scenarios 3-6
3-3. Total population under five ICLUS scenarios 3-17
3-4. Comparison of California and ICLUS (base case) 2030 projections 3-19
3-5. Comparison of Colorado and ICLUS (base case) 2030 projections 3-19
3-6. Comparison of Florida and ICLUS (base case) 2030 projections 3-20
3-7. Comparison of Minnesota and ICLUS (base case) 2030 projections 3-20
3-8. Comparison of New Jersey and ICLUS (base case) 2025 projections 3-21
4-1. SERGoM functional flow 4-5
5-1. Base case population by Census region 5-3
5-2. Al storyline population by Census region 5-3
5-3. A2 storyline population by Census region 5-3
5-4. Bl storyline population by Census region 5-4
5-5. B2 storyline population by Census region 5-4
5-6. Base case annual population growth rates by region 5-5
5-7. Al storyline annual population growth rates by Census region 5-5
5-8. A2 storyline annual population growth rates by Census region 5-5
5-9. Bl storyline annual population growth rates by Census region 5-6
5-10. B2 storyline annual population growth rates by Census region 5-6
5-11. Northeast region population by storyline 5-7
VI
-------
LIST OF FIGURES (continued)
5-12. Midwest region population by storyline 5-7
5-13. South region population by storyline 5-7
5-14. West region population by storyline 5-8
5-15. Urban/suburban housing land-use trends for ICLUS SRES storylines 5-9
5-16. Impervious surface area estimates, 2000-2100 5-12
5-17. 2050 estimated percent impervious surface, base case 5-13
5-18. 2000-2050 relative change in impervious surface, base case 5-14
5-19. 2050 impervious surface, Al storyline 5-15
5-20. 2000-2050 relative change in impervious surface, Al storyline 5-16
5-21. 2050 impervious surface, A2 storyline 5-17
5-22. 2000-2050 relative change in impervious surface, A2 storyline 5-18
5-23. 2050 impervious surface, Bl storyline 5-19
5-24. 2000-2050 relative change in impervious surface, Bl storyline 5-20
5-25. 2050 impervious surface, B2 storyline 5-21
5-26. 2000-2050 relative change in impervious surface, B2 storyline 5-22
A-l. Base case, year 2010 housing density map A-2
A-2. Base case storyline, year 2050 housing density map A-3
A-3. Base case, year 2100 housing density map A-4
A-4. Al storyline, year 2010 housing density map A-5
A-5. Al storyline, year 2050 housing density map A-6
A-6. Al storyline, year 2100 housing density map A-7
A-7. A2 storyline, year 2010 housing density map A-8
A-8. A2 storyline, year 2050 housing density map A-9
vii
-------
LIST OF FIGURES (continued)
A-9. A2 storyline, year 2100 housing density map A-10
A-10. Bl storyline, year 2010 housing density map A-ll
A-ll. Bl storyline, year 2050 housing density map A-12
A-12. Bl storyline, year 2100 housing density map A-13
A-13. B2 storyline, year 2010 housing density map A-14
A-14. B2 storyline, year 2050 housing density map A-15
A-15. B2 storyline, year 2100 housing density map A-16
B-l. Effect of fertility rate on national population B-2
B-2. Effect of international migration on national population B-3
B-3. Comparison of abroad range of scenarios B-4
B-4. Effect of mortality on national population B-4
C-l. Full regression tree backbone (58 terminal nodes) without labels C-4
C-2. Cross validation results for the full regression tree C-4
C-3. The relationship between percent impervious and housing density C-5
C-4. Top ten terminal nodes within full regression tree with housing density labels
and percent impervious estimates (terminal nodes) C-5
C-5. Estimated national impervious surface, 2000 C-7
C-6. Difference in impervious surface, United States C-9
C-7. Difference in impervious surface, Colorado C-10
C-8. Difference in impervious surface, Mid Atlantic region C-ll
C-9. Estimated impervious surface, base case 2030 C-12
Vlll
-------
LIST OF ABBREVIATIONS AND ACRONYMS
Al The Al storyline in the Special Report on Emissions Scenarios
A2 The A2 storyline in the Special Report on Emissions Scenarios
B1 The B1 storyline in the Special Report on Emissions Scenarios
B2 The B2 storyline in the Special Report on Emissions Scenarios
EPA U.S. Environmental Protection Agency
GCM General Circulation Model
GHG greenhouse gas
GIS Geographic Information System
HD Housing Density
HUC Hydrologic Unit Code
ICLUS Integrated Climate and Land-Use Scenarios
IPCC Intergovernmental Panel on Climate Change
IS impervious surface
ISpui percent urban impervious surface
MIGPUMA Migration Public-Use Microdata Areas
NCHS National Center for Health Statistics
NLCD National Land Cover Database
NRCS Natural Resources Conservation Service
NRI Natural Resources Inventory
PUMA Public-Use Microdata Areas
PUMS Public-Use Microdata Samples
SERGoM Spatially Explicit Regional Growth Model
SRES Special Report on Emissions Scenarios
UPPT unprotected, private protected, public protected, and tribal/native lands
USD A U.S. Department of Agriculture
VMT vehicle miles traveled
IX
-------
PREFACE
This report was prepared jointly by ICF International, Colorado State University, and the
Global Change Research Program (GCRP) in the National Center for Environmental Assessment
(NCEA) of the Office of Research and Development (ORD) at the U.S. Environmental
Protection Agency (U.S. EPA). The report has undergone internal, public, and external peer
review. Changes and edits between the draft and final report reflect comments made by
reviewers. The report describes the methodology used to develop and modify the models that
constitute the ICLUS project. The scenarios and maps resulting from this effort are intended to
be used as benchmarks of possible land-use futures, focusing on residential housing, that are
consistent with socioeconomic storylines used in the climate change science community. The
two-way feedbacks that exist between climate and land use are not yet fully understood and have
consequences for air quality, human health, water quality, and ecosystems. This report lays the
foundation for characterizing and assessing the effects of these feedbacks and interactions by
developing residential HD scenarios and deriving IS cover from them. These outputs facilitate
future integrated assessments of climate and land-use changes that make consistent assumptions
about socioeconomic and emissions futures. Outputs will also be distributed through the Web
and a Geographic Information System (GIS) tool that allows some customization of scenarios.
EPA's intention is to use the results of this first phase of modeling to inform and facilitate
investigation of a broader set of impacts scenarios and potential vulnerabilities in areas such as
water quality, air quality, human health, and ecosystems. More specifically, this research will
enable more sophisticated model runs that will evaluate the effects of projected climate changes
on demographic and land-use patterns and the results of these changes on endpoints of concern.
-------
AUTHORS AND REVIEWERS
The GCRP, within NCEA, ORD, is responsible for publishing this report. This document
was prepared by ICF International under Contract No. GS-10F-0234J, U.S. EPA Order No.
1101. Dr. Chris Pyke and Dr. Britta Bierwagen served as successive Technical Project Officers.
Drs. Pyke and Bierwagen provided overall direction and technical assistance, and Dr. Bierwagen
contributed as an author.
AUTHORS
U.S. EPA
Britta Bierwagen, John Thomas, Chris Pyke*
ICF International, Washington, DC
Anne Choate, Jonathan Cohen, Philip Groth
Natural Resource Ecology Lab, Colorado State University, Fort Collins, CO
David M. Theobald
REVIEWERS
This report benefited from the comments and suggestions of the following reviewers.
EPA Reviewers:
Cynthia Gage, Ellen Cooler, Sandy Bird, Laura Jackson, Matt Dalbey, Brooke Hemming,
Henry Lee
External Reviewers:
Dan Brown, University of Michigan; Steven Manson, University of Minnesota; Dawn Parker,
George Mason University; David Skole, Michigan State University.
ACKNOWLEDGMENTS
The authors thank the Global Change Assessment Staff in NCEA, particularly Chris
Weaver and Philip Morefield, for their input, advice, and assistance on this project. Randy Freed
also provided valuable input and advice. We also thank John Norman for his assistance.
Comments on earlier versions of this report by EPA and external reviewers greatly improved the
final product.
*Present affiliation: CTG Energetics, Washington, DC
xi
-------
EXECUTIVE SUMMARY
Climate and land-use change are major components of global environmental change.
Assessments of impacts associated with these changes often show interactions and two-way
feedbacks between climate and land use. The consequences of these interactions also show that
land use could exacerbate or alleviate climate-change effects. Based on these observations it is
important to use land-use scenarios that are consistent with the assumptions underlying
recognized international climate-change scenarios.
The ICLUS project developed outputs that are based on the IPCC SRES social,
economic, and demographic storylines. We adapted these storylines to the United States and
modified U.S. Census Bureau population and migration projections to be consistent with these
storylines. ICLUS outputs are derived from a pair of models: a demographic model that
generates population projections and a spatial allocation model that distributes projected
population into housing units across the landscape. The demographic model is at the county
scale and the spatial allocation model is at the 1 ha pixel scale (though results of analyses could
be aggregated to larger scales). Each scenario is run for the conterminous United States to 2100
by decade.
The results of this first phase of the project provide a foundation for evaluation of a
broader set of impacts scenarios and potential vulnerabilities in areas such as water quality, air
quality, human health, and ecosystems. More specifically, these scenarios will underlie more
sophisticated model runs that will evaluate the effects of projected climate changes on
demographic and land-use patterns and the results of these changes on endpoints of concern.
The products generated in this first phase are consistent with socioeconomic storylines used by
the climate change modeling community, but climate-change variables are not integrated into the
models in this phase. Outputs are also not specifically designed to feed into climate change
models.
The ICLUS project uses the SRES storylines because these storylines are utilized as
direct inputs into general circulation models developed by the climate change science
community. These storylines were selected to facilitate future, more integrated assessments of
climate and land use at national or regional scales, because the broad underlying assumptions are
the same. The SRES describe storylines along two major axes, economic versus
environmentally-driven development (A-B) and global versus regional development (1-2); the
four quadrants defined by these axes comprise the four storylines, Al, A2, Bl, and B2. We
adapted these storylines for the United States to inform our changes to variables such as fertility,
domestic and international migration, household size, and travel times from the urban core.
xn
-------
The demographic model is composed of five components, fertility, mortality, domestic
in-migration, domestic out-migration, and net international migration, which are calculated using
a cohort-component model and a gravity model. The population is divided into cohorts that are
age-, gender-, and race/ethnicity-specific. Changes due to these five components of change are
estimated over time as each cohort is tracked separately. The gravity model is used to track
domestic in- and out-migration by county. Components of the gravity model include certain
county amenities and functional distance that connects counties based on population locations
and transportation infrastructure. The resulting county-based population projections are the
inputs to the spatial allocation model.
The spatial allocation model distributes the population into housing units across the
country at a 1 ha pixel scale. The model used in this project is the Spatially Explicit Regional
Growth Model (SERGoM). SERGoM uses five main base datasets: housing units and
population based on the 2000 U.S. Census, undevelopable lands, road and groundwater well
density, commercial/industrial land use from the National Land Cover Database, and county
population projections from the demographic model described above. Household size and travel
times are adjusted to reflect the assumptions of the different U.S.-adapted SRES storylines.
These modifications result in different spatial allocations among the scenarios such that the
B storylines show more compact growth focused around urban areas and the A storylines show
less compact growth overall and more housing in suburban and exurban densities.
The scenarios result in a range of projected increases in urban and suburban area across
the United States. The smallest increase is 60% for the Bl scenario and the largest increase is
164% for A2. These increases in housing can be translated to changes in IS cover, which can be
used to examine impacts on water quality, for example. Our results show that there could be a
significant increase of watersheds (8-digit Hydrologic Unit Codes [HUCs]) that are likely to be
stressed from IS coverage of at least 5%. These changes will vary regionally across the country.
This report describes the modeling methodology for the ICLUS project and some initial
analyses using the outputs. There are many additional modifications that are possible to explore
additional land-use futures and there are many options for further research. Model modifications
can be made to further explore policy and planning alternatives such as Smart Growth or other
low-impact development patterns. The demographic and spatial outputs can be used in
numerous analyses examining potential future impacts on air quality and human health, water
quality, traffic and associated emissions, and regional growth rates.
Xlll
-------
1. INTRODUCTION
Climate change and land-use change are global drivers of environmental change. Impact
assessments frequently show that interactions between climate and land-use changes can create
serious challenges for aquatic ecosystems, water quality, and air quality. In many cases, it is
impossible to determine the impact of climate change without consideration of land use and land
cover dynamics. While land use can exacerbate climate impacts, land-use planning, policy, and
management can also create important adaptation opportunities to increase the resilience of
sensitive socioeconomic or ecological systems.
Integrated assessments of both climate and land-use changes currently are limited by
fragmented information on potential future land use. In many cases individual municipal areas
have conducted extensive analyses, but it is impossible to place results in regional or national
contexts. Moreover, the studies are often based on inconsistent or poorly documented
socioeconomic storylines. The motivation for the U.S. Environmental Protection Agency (EPA)
Integrated Climate and Land-Use Scenarios (ICLUS) project was derived from the recognition of
the complex relationships between land-use change and climate change impacts and the absence
of an internally consistent set of land-use scenarios that could be used to assess climate change
effects.
This report describes the first phase of this project, which is designed to provide a
foundation for the evaluation of a broader set of impacts scenarios and potential vulnerabilities in
areas such as water quality, air quality, human health, and ecosystems. More specifically, the
land-use scenarios described here can be used as inputs to more sophisticated model runs that
evaluate the effects of projected climate changes on demographic and land-use patterns and the
results of these changes on endpoints of concern. The products generated in this first phase are
consistent with socioeconomic storylines used by the climate change modeling community, but
this first phase does not explicitly integrate climate change variables into the models. Outputs
are also not specifically designed for incorporation into climate change models.
The ICLUS project developed scenarios of housing density (HD) and derived impervious
surface (IS) cover from these scenarios, for the entire conterminous United States for each
decade through 2100. These scenarios are based on the Intergovernmental Panel on Climate
Change (IPCC) Special Report on Emissions Scenarios (SRES) social, economic, and
demographic storylines (Nakicenovic, 2000), because these storylines are used as direct inputs
into general circulation models (GCMs) used by the climate change science community. This
link facilitates any future, more integrated assessments of climate and land use at national or
regional scales, because the broad underlying assumptions are the same. The ICLUS scenarios
are developed using a combination of models representing demography, including domestic and
1-1
-------
international migration, and spatial allocation of housing (Figure 1-1). The resulting scenarios
(1) will enable us, our partners, and our clients to conduct assessments of both climate and
land-use change effects across the United States; (2) provide consistent benchmarks for local and
regional land-use change studies; and (3) can be used to identify areas where climate-land-use
interactions may exacerbate impacts or create adaptation opportunities.
Initial county population
and SRES storylines
Repeat
1
Demographic model
(cohort component & migration)
I
Total population per
^ , , • f ... . county per time step
Population for next time step I
Housing allocation
Figure 1-1. Model and information flow within the ICLUS
project.
EPA's Global Change Research Program in the National Center for Environmental
Assessment of the Office of Research and Development began investigating the availability of
state and county-level population projections in 2004. Initial efforts evaluated the availability,
sources, and extent of state and county-level population projections with an emphasis on
identifying projections for the time period from 2050 through 2100. These efforts yielded
numerous datasets, but very few sources projected populations to 2050 and beyond, particularly
at the county level. Demographic projections for these later years are particularly relevant when
considering the impacts of climate change on ecosystems, water infrastructure, transportation
infrastructure, and land protection efforts. Population and land-use projections based on
economic factors such as regional income and employment growth can shift dramatically over
time. Projections based on fundamental demographic drivers such as fertility and mortality are
somewhat more stable, particularly over longer time frames. Therefore, the ICLUS project uses
demographic projections as the basis for modeling changes in HD.
The modeling framework (Figure 1-1) presented in this report uses demography to drive
the number and migration of people, while a spatial allocation model governs the distribution of
1-2
-------
people on the landscape in housing units. The demographic model has two parts, a
cohort-component model and a gravity model. The spatial allocation is conducted using an
established Geographic Information System (GlS)-based model, Spatially Explicit Regional
Growth Model (SERGoM) (Theobald, 2001, 2003, 2005). The study was designed to provide
county-level demography and HD for the conterminous United States with HD allocations for
each 1 ha pixel for each decade through 2100.
With countless variables and unforeseen events affecting demographic and land-use
change patterns, it is impossible to quantify the level of uncertainty associated with these
projections. In addition to general concerns about model accuracy, this approach does not
capture some important attributes of land-use change that will likely confront the U.S. landscape
in the future, such as regional development restrictions, local zoning laws, potential regional
growth limitations due to water scarcity, changes in national immigration policy, and unforeseen
socioeconomic limitations to development. Therefore, the scenarios presented in this study are
designed to explore a range of possible futures. Given the scope of this task, the authors
acknowledge the limitations and possibilities for additional evaluation in future analysis.
1-3
-------
2. ADAPTING THE SRES TO THE UNITED STATES
The socioeconomic storylines in the SRES are derived from anticipated demographic,
economic, technological, and land-use changes data for the 21st century, and are highly
aggregated into four world regions (Nakicenovic et al., 2000). The SRES describe linkages
between physical changes in climate and socioeconomic factors, because they link development
pathways with greenhouse gas (GHG) emissions levels used as inputs to GCMs (Rounsevell et
al., 2006). There have been other scenario-development exercises to assess future impacts on a
range of endpoints, including the Millennium Ecosystem Assessment (MEA, 2005), which also
describes economic and environmental conditions in the future. The benefit of using the SRES
storylines is their direct link to GCMs. This enables us to link our land-use projections to
emissions and future climate scenarios for integrated assessments of the effects of land-use
change and climate change in a consistent way.
Modeling and projecting human activity into the future is a challenge for many reasons.
Any attempt to project demographic and economic changes over time must contend with large
degrees of uncertainty. Furthermore, taking discontinuous events into consideration—ones that
would have a profound effect on any anticipated trajectory—is even more difficult.
Nevertheless, a forward-looking approach to environmental and economic problems encourages
us to look into the future to attempt to better understand the challenges that lie ahead, and to
better prepare our society to confront those challenges.
By taking a scenario approach to such modeling efforts, we acknowledge this inherent
uncertainty and consider a variety of possible trajectories. This approach results in a range of
outputs. No single output may be the "right" one, but together they paint a picture of a likely
range of possible futures. The primary challenge to the scenario approach lies in developing a
reasonable range of scenarios that can be used in multiple modeling efforts.
The emissions storylines in the SRES cover a wide range of possible paths for the
primary social, economic, and technological drivers of future emissions. These storylines have
since become the standard input of socioeconomic information for GCMs and other land-use
change modeling efforts (e.g., Solecki and Olivieri, 2004; Reginster and Rounsevelle, 2006;
Rounsevelle et al., 2006) providing a reasonable set of scenarios to bound the potential futures
with respect to climate change. By using the SRES storylines as the basis for the scenarios
investigated by this project, the results may be put into the context of widely available and
peer-reviewed climate-change model output (e.g., IPCC, 2001, 2007).
2-1
-------
2.1. OVERVIEW OF THE SRES STORYLINES
The development of SRES consisted of three main steps, beginning with qualitative
"storylines" that describe broad economic, environmental, technological, and social development
patterns that could unfold over the 21st century (Nakicenovic et al., 2000). Next, particular
quantitative paths for the fundamental driving forces of emissions, including population and
gross domestic product, were selected for each storyline. Finally, six different modeling teams
produced quantitative interpretations of the storylines, using the quantitative paths for driving
forces as inputs, resulting in 40 different storylines for energy use, land use, and associated GHG
emissions over the next 100 years (Nakicenovic et al., 2000). The SRES describe storylines
along two major axes, economic versus environmentally-driven development (A-B) and global
versus regional development (1-2), which make up the four combinations of storylines, Al, A2,
Bl, and B2 (Figure 2-1). There are between 6 and 18 emissions scenarios within each of the
four SRES storylines. Table 2-1 provides a summary of the qualitative fertility, mortality, and
migration assumptions made by the SRES authors for each storyline for the industrialized
country regions and the developing country regions; these qualitative assumptions served as the
framework for the more quantitative inputs for the scenarios.
Based on descriptions in the SRES report (primarily in Sections 4.3 and 4.4 of the SRES
report), a summary of the reasoning for these assumptions is provided below:
• In the Al storyline, rapid economic development, associated with improved education
and reduced income disparities, is assumed to drive a relatively rapid fertility decline in
the high fertility regions. Global population is expected to rise until peaking in the
middle of the century, after which fertility is generally below replacement level. Fertility
in industrialized regions is assumed to follow a medium path at least in part so that,
relative to the developing regions, the storyline is consistent with the assumption that
social and economic convergence will lead to demographic convergence as well. For
mortality, it is assumed that the conditions leading to low fertility are also consistent with
relatively low mortality, so mortality is assumed to be low in all regions. No explicit
discussion of migration is provided, although the projection eventually adopted assumes
medium migration levels.
• In the A2 storyline, the regional orientation and slower rate of economic growth, limited
flow of people and ideas across regions, and orientation toward family and community
values was judged to be consistent with a relatively high fertility in all world regions.
Mortality was assumed to be high as well, based on the assumption that conditions
leading to high fertility would also lead to relatively high mortality in all regions.
Although the storyline describes a limited flow of people across regions, the storyline
authors assumed medium migration flows, as in all other storylines.
2-2
-------
SRES Scenarios
Global
Regional
Figure 2-1. Illustration of the SRES families along two dimensions
that indicate the relative orientation of the storylines along the axes
of global or regional development and economic or environmental
concerns (reprinted from Nakicenovic et al., 2000).
Table 2-1. Qualitative demographic assumptions in global SRES storylines
Storyline
A1/B1
A2
B2
Fertility
IND: medium
DEV: low
IND: high
DEV: high
IND: medium
DEV: medium
Mortality
IND: low
DEV: low
IND: high
DEV: high
IND: medium
DEV: medium
Migration
IND: medium
DEV: medium
IND: medium
DEV: medium
IND: medium
DEV: medium
Projection source
Lutzetal., 1996
Lutzetal., 1996
UN, 1998
IND = Industrialized country regions; DEV = Developing country regions.
The "high", "medium", and "low" descriptions are interpreted as relative to the overall outlook within each region
(i.e., high fertility in the IND region means the high end of the plausible range for that region, but may in fact be
lower than low fertility paths for the DEV region, which occupy the low end of the plausible range for the DEV
region).
2-3
-------
• The Bl storyline shares the same population projection as the Al storyline, although for
somewhat different reasons. Rapid social development, particularly for women, and an
emphasis on education drives a relatively rapid decline in fertility in developing country
regions (as opposed to the Al storyline, in which economic development is seen as the
main driver). Reasoning for fertility in industrialized countries, and for mortality and
migration assumptions, are the same as in Al.
• In the B2 storyline, economic development is moderate, particularly in the developing
country regions. However education and welfare programs are pursued widely and local
inequity is reduced through strong community support networks. The mix of moderate
economic development and strong but heterogeneous social development results in an
assumption of medium fertility and mortality paths. Migration is again assumed to be
medium, with no explicit discussion of this choice.
Demographic assumptions in SRES are intended to be consistent with storylines and with
other driving force assumptions. Consistent relationships among these factors mean that the
demographic assumptions occur in the context of other socioeconomic trends in a way that is not
at odds with established theory, the weight of historical experience, or current thinking in the
literature on determinants of demographic trends. For example, a key factor differentiating
population assumptions across SRES storylines is the assumed speed of the transition from high
to low fertility in regions with relatively high current fertility. The transition occurs faster in the
Al and Bl storylines, and slowest in the A2 storyline. These choices are based on the rationale
that there are a range of conditions that contribute to fertility transitions, including economic
development, education, labor force opportunities for women, and the spread of ideas about
modern lifestyles (Lee, 2003). These factors are present, and stronger, in the Al and
Bl storylines (in different combinations) and absent, or weaker, in the A2 storyline. Thus, in this
case, there is a clear notion of consistency in which storyline elements can be said to favor
preferentially a particular demographic outcome.
However, consistency does not mean that the assumed demographic trends are the only
possible outcomes, or in some cases even the most likely outcomes, conditional on a particular
storyline. In some cases, storylines serve only as weak constraints on demographic futures, and a
wide range of demographic assumptions might all be consistent with the broader development
trends. For example, the demographic transition reasoning just described applies only to
countries with relatively high fertility (e.g., substantially above replacement level of about
two births per woman). These conditions occur for only about half the current population of the
world, and for only the first half (or even less) of the 21st century, according to projections (Lee,
2003). Once the transition to low fertility is complete, there is little theoretical basis for linking
subsequent fertility changes or cross-country differences to particular socioeconomic trends; a
wide range of outcomes is possible (O'Neill, 2005a). Yet SRES storylines link the pace of
2-4
-------
economic growth with fertility outcomes (i.e., demographic transition-type reasoning) for all
regions of the world, and for the entire century. It is certainly possible that rapid economic
growth will be associated with relatively low fertility (and slow growth with high fertility) in
post-transition societies, and it is not inconsistent in the sense of contradicting established theory
(though this is at least partly because there is no single established theory to contradict).
Completely opposite associations (e.g., low fertility with slow economic growth) are equally
plausible (and equally consistent) for post-transition societies.
Thus in considering the implications of SRES storylines for demographic outcomes at the
national and sub-national level, it must be kept in mind that it is possible that a wide range of
different assumptions could be judged to be consistent with SRES. Indeed other studies have
developed alternative demographic assumptions for SRES storylines, with different quantitative
outcomes (Hilderink, 2004), or have quantified a range of plausible outcomes associated with
storylines (O'Neill, 2004, 2005a,b).
2.2. INTERPRETING AND ADAPTING THE SRES STORYLINES
The SRES storylines do not provide a clear blueprint for downscaling to the local or even
the national level. In incorporating the SRES storylines into county-level projections for the
United States, an effort was made to be consistent in qualitative terms with the global SRES
storylines. Given the wide range of potential interpretations, this consistency was understood to
imply that the qualitative trends do not contradict established theory, historical precedent, or
current thinking. It was also a goal to model a wide a range of assumptions, while remaining
consistent with the SRES and U.S. demographic patterns. Rationales connected to SRES
storylines are discussed briefly for each scenario. For each of the storylines adapted to the
United States, the fertility assumptions are exactly consistent with the global assumptions, while
domestic and international migration patterns leave more room for interpretation and are more
specifically adapted to the United States. The low U.S. Census scenario for mortality was
chosen for all storylines used in the modeling (see Appendix B for more information). These
model inputs were varied to develop the different scenarios rather than to investigate the relative
importance of each of the inputs.
2.2.1. Al Storyline Adapted for the United States
Al represents a world of fast economic development, low population growth, and high
global integration. In this storyline fertility is assumed to decline and remain low in a manner
similar to recent and current experience in many European countries (Sardon, 2004). A plausible
rationale would be that the rapid economic growth in this storyline leads to continuing high
participation of women in the workforce, but it becomes increasingly difficult to combine work
2-5
-------
with childbearing due to inflexibilities in labor markets. At the same time, social changes in
family structures lead to increasing individuation, a rise in divorce rates, a further shift toward
cohabitation rather than marriage, later marriages and delayed childbearing, all of which
contribute to low fertility. Substantial aging resulting from the combination of low birth rates
and continued low death rates raises the demand for immigration. Meanwhile, economic growth
throughout the world and an increasingly unified global economy encourage the free movement
of people across borders. Domestic migration is anticipated to be relatively high as well, as
economic development encourages a flexible and mobile workforce.
2.2.2. Bl Storyline Adapted for the United States
This storyline represents a globally-integrated world similar to Al, but with greater
emphasis on environmentally sustainable economic growth. Like Al, fertility is assumed to be
low due to higher incomes and economic development. International migration is expected to be
high due to widespread economic development and freer global flows. Domestic migration
however is lower due to a combination of factors. First, an increased focus on sustainability
leads to a reduction of subsidies for development in previously rural counties with significant
natural amenities. Second, the information oriented economy increases demand for specialized
labor pools and increases the number of high paying jobs in traditional large urban centers.
2.2.3. A2 Storyline Adapted for the United States
The A2 storyline represents a world of continued economic development, yet with a more
regional focus and slower economic convergence between regions. Fertility is assumed to be
higher than in Al and Bl due to slower economic growth, and with it, a slower decline in
fertility rates. International migration is assumed to be low because a regionally-oriented world
would result in more restricted movements across borders. Domestic migration is high because,
like in Al, the continued focus on economic development is likely to encourage movement
within the United States.
2.2.4. B2 Storyline Adapted for the United States
The B2 represents a regionally-oriented world of moderate population growth and local
solutions to environmental and economic problems. Fertility rate is assumed to be medium,
while international and domestic migration are low due to the local emphasis, focus on
sustainability, and increasing number of jobs in urban centers. International migration is low due
to the regional orientation, as with A2. Domestic migration is low due to the more
environmental orientation, as with Bl.
2-6
-------
The SRES storylines adapted to the United States were integrated into our modeling
framework. We modified both the demographic and spatial allocation models for each of the
four scenarios.
2-7
-------
3. DEMOGRAPHIC PROJECTIONS
3.1. OVERVIEW
The ICLUS demographic model utilizes a cohort-component methodology. The
cohort-component methodology is a common technique for projecting population on the basis of
five independent variables or components of population change: fertility, mortality, domestic in-
and out-migration, and net international migration. The population is divided into cohorts that
are age-, gender-, and race/ethnicity-specific. Changes due to these five components of change
are estimated over time as each cohort is tracked separately, hence the term "cohort-component"
(Siegel and Swanson, 2004).
The methodology is flexible in that different assumptions can be applied to each
component of population change. For example, fertility — or the number of births — is estimated
by multiplying cohort-specific fertility rates (births per woman) times the population of each
cohort of women. Different rates can be applied to women of different ages and ethnicities.
Furthermore, these rates can change over time or between different scenarios. In all cases, the
fundamental method stays the same while changes in the rates can be used to simulate different
SRES storylines.
The population of a county in any year (f) as estimated by the model is determined using
Equation 3-1:
Pt = Pt-l+B-D + NDM + NIM (3-1)
Where:
Pt = Population in year t
Pt-i = Population in the previous year
B = Births in year t
D = Deaths in year t
NDM = Net Domestic Migration in year t
NIM= Net International Migration in year t
Beginning with an initial set of populations, annual components of change are applied in
the following process, which are repeated annually until the desired end year is reached.
1) Add births by cohort
2) Deduct deaths by cohort
3-1
-------
3) Add net international migration
4) Add net domestic migration by combining domestic in-migration and out-migration
(Estimated every fifth year, as discussed below.)
5) Age population one year and repeat for the next year
This methodology is illustrated in Figure 3-1 below. The cycle begins with an initial Year 2005
population and is repeated until reaching Year 2100.
Initial Population in Year
Apply Components of Change
I. Add
Births
2. Subtract
Deaths
I
3 . Add Net International
Migration
4. Add Net Domestic
Migration
Age Population
Population in Year t + 1
Figure 3-1. Demographic model flow.
Table 3-1 provides a qualitative description of how we modified the global storylines to
the United States. As a general observation, note that the proposed fertility and mortality
assumptions follow the SRES assumptions, with the exception that the Al storyline assumes low
fertility (rather than medium as in SRES). In addition, the international migration assumptions
assume high and low immigration in storylines Bl and A2, respectively (rather than medium as
in SRES). These choices are made in order to explore a fuller range of demographic trends for
the United States than in SRES, since the SRES storylines contained only a limited range of
projections for the North America region (O'Neill, 2005a). It should also be noted that these
assumptions are not designed to cover the widest possible range of population size outcomes;
e.g., the combinations resulting in the highest and lowest population sizes are not explored here.
In all cases the mortality rate was kept constant across all scenarios. This was partly due to data
availability—the U.S. Census did not release alternate scenarios for mortality rates. Experiments
3-2
-------
with adjusting these mortality rates (shown in Figure B-4 in Appendix B) demonstrated
relatively little change in total national population due to variations in the mortality rate.
Table 3-1. Summary of qualitative adjustments to the demographic
projections
Storyline
Al
Bl
A2
B2
Baseline
Demographic Model
Fertility
Low
Low
High
Medium
Medium
Domestic migration
High
Low
High
Low
Medium
Net international migration
High
High
Medium
Medium
Medium
In the following sections, the methods and data used for the initial population and each
component are discussed in greater detail. A summary of the data sources and the major
assumptions used in the demographic model is presented in the Appendix in Table F-l.
3.2. INITIAL POPULATION
In order to use the rates for components of change provided by the U.S. Census Bureau
(U.S. Census Bureau, 2000) (discussed below), it was necessary to begin with an initial
Year 2005 population dataset that was disaggregated using the same cohorts. These cohorts in
the rates data are divided into two genders (male and female), 100 age groups (0-99 in one year
increments), and five racial/ethnic—or "bridged race"—groups (Hispanic, non-Hispanic White,
non-Hispanic Black, non-Hispanic American Indian or Alaskan Native, and non-Hispanic Asian
or Pacific Islander).1 This represents 1,000 distinct population cohorts (2 genders x 100 ages x
5 bridged race groups).
County populations using bridged race and one-year age cohorts were most readily
accessible using the Bridged-Race Vintage 2006 dataset for July 1, 2005 provided by the
:In general, the U.S. Census Bureau considers the primary racial categories to be American Indian/Alaskan
Native, Asian/Pacific Islander, Black, and White. "Hispanic origin" is considered an ethnic category. The race and
ethnicity categories used by the Census have changed over time. In the 2000 Census, participants were allowed to
identify with two or more racial groups for the first time. This project utilized the "bridged race" categories listed
above as a way of making data collected with one set of categories consistent with data collected using another set
of categories.
3-3
-------
National Center for Health Statistics.2 After downloading and parsing the data, two
manipulations were required. First, the dataset provided eight bridged race categories—the
non-Hispanic populations of the four race groups and the Hispanic populations of the four race
groups. The Hispanic populations were summed and combined into a single bridged race group.
This group, along with the four non-Hispanic race groups, comprised the five bridged race
groups used throughout the model.
Second, the 2000 Census dataset provided one-year age groups from ages 0-84 and
combined all others in the 85+ age group. In order to extend the age distribution through
99 years, the 85+ population data was disaggregated into ages 85-99. We used the national
age 85-99 populations by race and gender from the 2000 Census to allocate the 85+ age group
from the NCHS data. We assumed an identical race- and gender-specific age distribution in each
county for the Year 2005 base population. Although this step likely led to some distortions, the
effects are not long-lasting in the model as the initial 85-99 age groups "age out" of the model
by 2015.
3.3. FERTILITY AND MORTALITY
Fertility and mortality followed a simple methodology. For fertility, the number of
children born equals the number of women in a given cohort times the average number of
children born annually to every 1,000 women in that cohort divided by 1,000. Because virtually
all births occur to women between the ages of 10 and 49, only those cohorts are considered in
this model. These births are summed by race and used to create a new age zero cohort. To
allocate births between males and females, we calculated a historic ratio of 1,046 males born for
every 1,000 females born and assumed that this ratio holds steady (Matthews and Hamilton,
2005).
Similarly, mortality is estimated by multiplying the number of people in a given cohort
times the cohort-specific mortality rates. The resulting number of deaths is then subtracted from
the cohort. Unlike fertility, all cohorts are subject to mortality. Therefore, mortality rates are
applied to each cohort. Although an increasing number of Americans is living to the age of 100
or more, the model assumes 100% mortality after age 99 for the sake of computational
efficiency. Even with continued rates of survivorship past this age, the 100+ age group will
remain a miniscule portion of the population.
2 NCHS (National Center for Health Statistics). (2007) Bridged-race Vintage 2006 postcensal population
estimates for July 1, 2000 - July 1, 2006, by year, county, single-year of age, bridged-race, Hispanic origin, and sex.
Released August 16, 2007. Centers for Disease Control and Prevention (CDC), Atlanta, GA. Available online at
http://www.cdc.gov/nchs/about/major/dvs/popbridge/datadoc.htm#vintage2006.
3-4
-------
For fertility and mortality rates, the U.S. Census Bureau's "Component Assumptions of
the Resident Population by Age, Sex, Race, and Hispanic Origin" were used (U.S. Census
Bureau, 2000). These are the same data used in Census projections. These components of
change are associated with the 1990 National Projections and are used in both the 1990 State
Projections (Campbell, 1996) and the 2000 National Projections (Hollman et al., 2000). While it
would be preferable to use more recent data, at this time components of change based on the
2000 Census have not yet been released. While the rates are national averages, county
differences that arise in fertility and mortality are a reflection of each county's unique age, sex,
and racial composition.
For both fertility and mortality, the so-called Middle Series of component information
was used as the baseline. Fertility rates are provided in a single file; mortality rates for each
component are provided in three different tables, for the years 1999-2010, 2015-2055, and
2060-2100. Projected fertility rates are provided for each year to 2100, but beginning with
2010, mortality rates are provided in five year increments only. We assumed that 2010 mortality
rates held steady from 2010-2014, 2015 mortality rates held steady from 2015-2019, and so on.
The Census Bureau also provides a low and high scenario for fertility. Alternative
scenarios for mortality were not available. While the middle series was used in this "base case,"
the low and high series were used when developing projections specific to the SRES storylines.
The low data set was used for the Al and B1 scenarios, the medium set was used for the
B2 scenario, and the high set was used for the A2 scenarios.
3.4. NET INTERNATIONAL MIGRATION
The projections for net international migration utilized a simple method based on the U.S.
Census Bureau's international migration projections for the entire country. These files contain
the projected net international migration for each gender, age, and race cohort for the years
2000-2100. Like the fertility and morality rates, these data are provided by the Census (U.S.
Census Bureau, 2000).
Since the tables "Foreign-born Net Migration to the United States" contain only national
level data, it was necessary to allocate the national migrants to the counties. Using 2000 Census
data (U.S. Census Bureau, 2007, Summary File 3, Table P22), we determined each county's
share of the total population of recent immigrants (i.e., those who entered within the last
five years). These county shares were then used to allocate each cohort of immigrants among the
nation's counties. The estimated number of immigrants in each cohort was then added to the
existing county population of each cohort. This method assumes a constant distribution of recent
immigrants based on Year 2000 immigration patterns. While we anticipate that many of the
current "gateway counties" will continue to draw a large share of new immigrants, it is likely
3-5
-------
that new settlement patterns will develop in the future. A more complex international migration
component that allows for changes in the distribution pattern of immigrants may be utilized in
future improvements to the model.
The Census Bureau provides a low, medium, and high series for net international
immigration. In the base case, the middle series was used. The medium and high series were
used when developing SRES storyline-specific projections. The high series was used for the Al
and Bl storylines, while the medium series was used for the A2 and B2 storylines. The low set
was not used because the values projected under the U.S. Census Bureau's low set were so far
below current levels of immigration (Figure 3-2, U.S. Census Bureau, 2008) that they were
considered to be unrealistic for these purposes.
Low Projection
Middle Projection
High Projection
Estimated Historical Data
2000 2010 2020 2030 2040 2050 2060 2070 2080 2(KM> 2100
Year
Figure 3-2. Projected net international migrations under U.S. Census Bureau
Scenarios.
3.5. DOMESTIC MIGRATION AND GRAVITY MODELING
Domestic migration is the most complex component of change included in this
population model. In contrast to straightforward processes like fertility and mortality, which
may be predicted in aggregate with reasonable confidence, domestic migration is much more
difficult to predict. People move within the United States with relative frequency—according to
3-6
-------
did in 1995 (Perry and Schachter, 2003). These migrations occur for a wide variety of personal
and economic reasons that are difficult to predict. Unlike international immigration, which can
be thought of as a single external source of immigrants, domestic migration works two
ways—migrants must choose to leave one place and resettle in another place.
To estimate domestic migration, this project utilized a "gravity model" approach.
Gravity models are a common type of spatial interaction model in which a dependent variable (in
this case, migration flow between a pair of counties) is estimated as a function of a series of
independent variables (Rodrigue et al., 2006). Common independent variables include
population and distance (Haynes and Fotheringham, 1984); natural amenities, such as climate
variables and bodies of water, have also been shown to affect migration patterns, particularly in
rural areas (McGranahan, 1999). Among these, temperate summers and warm winters have the
strongest correlation with population growth (R2 = 0.38 and R2 = 0.27, respectively), a finding
that holds true even when correcting for macro-trends such as the general movement from the
Northeast and Midwest to the South and West (McGranahan, 1999). Other work has also shown
that amenity-rich mountain and coastal areas are highly successful at attracting migrants,
particularly those who are affluent and/or retired (Manson and Groop, 2000). While a wide
variety of independent variables were tested, the final gravity model used in this effort included
the following independent variables: population of the origin and destination counties, selected
climate variables of the origin and destination counties, water surface area of the origin and
destination counties, ratio of 2000 county population to 1980 population of the destination
county, and the distance between counties. This first phase of the modeling does not attempt to
project changes in climate variables over time.
The final model form and model coefficients were estimated by analyzing 1995 to 2000
migration data reported in Public-Use Microdata Samples (PUMS) (U.S. Census Bureau, 2003).
Using county-to-county migration data as the dependent variable and the various county
attributes as dependent variables, we ran a series of stepwise regression analyses to select the
most statistically significant set of dependent variables and estimate the model coefficients.
Below, we discuss how each data source for the regression analyses was developed, the resulting
model form, and the way this model was implemented in the overall demographic model.
3.5.1. 1995-2000 Migration Data
The 2000 Census PUMS data provides detailed records of individual domestic migrations
and characteristics of these migrants between 1995 and 2000 (U.S. Census Bureau, 2003).
However, the raw data does not readily provide county-to-county migration counts. Because the
migrations are organized by the destination state, in-migrants to any given state can be analyzed
using that state's PUMS file, but analysis of out-migration requires analyzing data from all
3-7
-------
50 states, the District of Columbia, and Puerto Rico. This project used a national migrant file
created by the New York State Data Center from the individual state and area files.3
The second challenge with the PUMS data is that it is organized by Public-Use Microdata
Areas (PUMAs). Counties and PUMAs overlap in a non-systematic way. Large urban counties
consist of multiple PUMAs, while small rural counties are often combined into a single PUMA.
To further complicate the issue, the spatial area that PUMS uses to capture migration origins
(where respondents lived in 1995), the so-called Migration PUMA (MIGPUMA), is not the same
spatial area as the one used to capture where they lived in 2000 (the PUMA). The MIGPUMA is
a larger area, and there can be one or more PUMAs contained within the MIGPUMA.
Because we required a database that contains migration, attraction factors, production
factors, and distance information for pairs of areas, introducing the MIGPUMA into our spatial
framework would have required re-calculating all of the other variables at the MIGPUMA level,
and aggregating the migration data (destinations) to the MIGPUMA level, losing considerable
local detail. Instead, we elected to proportionately assign MIGPUMA origins to its constituent
PUMAs, on an average basis.
Furthermore, retaining a county-based framework is practical because the amenity data is
available only for counties and counties are a more popularly understood and conceptualized
units than PUMAs. For the sake of consistency and simplicity, we then proportionately assigned
PUMA origins and destinations into counties based on the proportion of each PUMA's
population belonging to one or more counties. In the case of large metropolitan and suburban
counties, several PUMAs were typically aggregated to a single county, while in smaller suburban
counties and rural areas, PUMAs were often split among two or more counties. Because the
largest migration flows involve large urban and suburban population centers where PUMAs were
aggregated up to the county level, it was assumed that this step did not introduce significant error
into the migration analysis. By converting from MIGPUMA to PUMA to county, the migration
data set was used to estimate county-to-county flows needed for this study.
These county-to-county flow data were then aggregated into various age groups. Since
migration behavior is hypothesized to vary for different age groups, different gravity models
were calibrated for different age cohorts. In initial runs, the population was divided into ten-year
cohorts, from 0-9 to 90 and up, and a different gravity model was estimated for each. We
observed some broad differences in behavior between the older (over 50) age groups and the
younger age groups, presumably due to differences in migration patterns due to retirement.
However, the differences between the ten individual models were not found to be significant
3The website for the New York State Data Center is
http://www.empire.state.ny.us/nysdc/default.asp.
3-8
-------
enough to warrant ten different age groups. As a result, the dataset was receded into just two age
groups: 0-49 and 50+.
3.5.2. County Attribute Data
After receding the migration data set from the original geographies to counties, we then
added the county data that would be included as independent variables in the analysis. These
variables include county environmental amenities, population, and 1980-2000 population
growth, which serves as a proxy for economic growth.
The source for the amenity information used in the regression was found in a database
compiled by the U.S. Department of Agriculture (USDA) (McGranahan, 1999), which utilizes
data originally collected from the Center for National Health Statistics, U.S. Department of
Health and Human Services. The amenity index includes a range of climatic and amenity factors
that are thought to influence migration. The data used from this set include:
• Mean temperature for January, 1941-70;
• Mean hours of sunlight for January, 1941-70;
• Mean temperature for July, 1941-70;
• Mean relative humidity for July, 1941-70; and
• Percent water area.
County population data were drawn from the U.S. Census Bureau's City and County
Data Book, 2000 edition. The population over age 50—used in analyses of the 50-99 age
group—was determined from data downloaded from the 2000 U.S. Census (Summary File 1,
Table P12). The 2000 county population data, combined with 1980 and 1990 county populations
provided by the Census Bureau (U.S. Census Bureau, 1992), were used to calculate the growth
rate of counties between 1980 and 2000. This term was added after initial model runs resulted in
declining populations in medium-sized counties that exhibited strong growth in the 1980s and
1990s.
3.5.3. Distance Matrix
A full county-to-county distance matrix was the final data input used in developing the
gravity model. Typical methods of migration movement between geographic locations (e.g., city
to city or county to county) assume that interaction can be estimated as straight-line county
centroid-to-centroid distance (Conley and Topa, 2002). Two main deficiencies are that:
(1) often the geographic centroid of a county poorly represents the population-weighted centroid
3-9
-------
of a county; and (2) humans do not move around on the ground optimally using a straight-line
strategy, rather we are constrained to transportation infrastructure which in turn has evolved in
response to major topographic and water features.
Because of these deficiencies, we developed a "functional" estimate of county-to-county
interaction. The main assumptions incorporated into the county-to-county functional distance
calculations are that the amount of interaction, and more specifically migration, between counties
are better approximated by examining: (a) where in a county population is located, and (b) how
much ground- (and water-) based transportation infrastructure is in place.
The geographic centroids of each of 25,150 U.S. Census Bureau-defined places (within
conterminous United States) were found and assigned to the county in which they were located.
The population (2000 Census) was then used to weight each point, and the population-weighted
centroid for each county was determined. For roughly a half-dozen counties, the
population-weighted centroid was calculated to be outside the boundaries of its respective
county. For these, the centroid was manually moved back into its respective county. We also
generated a centroid for a couple of counties in Nevada that did not have places in them (visually
in the center).
We generated a cost- weight surface using major roads such that the weights were
assigned the assumed average speed limit assigned by road type. Cost distance was then
computed using the population-weighted centroids as the "seeds," computing minutes travel time
(7) along the roads. For each adjacent (first-order neighbor) county-county pair (y) we adjusted
the travel time to account for k multiple roads (multiple connections between population
centroids) to compute an interaction weight (W). Equation 3-2 shows the calculation for the
interaction weight (W) to generate the cost-weight surface for county pairs.
(3-2)
Where:
Wjj = Interaction weight for the county-county pair /', j
Tk = Travel time for k multiple roads
We then generated a network and used a network-based least-cost path algorithm to
compute effective distances along paths of pair- wise segments using the FunConn tools
(Theobald et al., 2006). We developed a distance matrix of pairwise functional distances (Wij)
computed in minutes travel time. We exported this matrix from the GIS database in a list of
3-10
-------
from-county, to-county, weight format (over 9 million records for 3109 counties times
3108 destination counties) to the gravity model data table. We also manually added linkages
between counties that are served with regular ferry service as delineated in the U.S.
Transportation Atlas 2006 (e.g., across Lake Michigan and Nantucket Island Sound).
3.5.4. Model Specification
After collecting the above data, it was reorganized into a single data table. Each record in
the data table contained the number of migrations from one county to another, the attributes of
the origin county, the attributes of the destination county, and the functional distance between
them. In keeping with the traditional functional form of gravity models, which states migration
is proportionate to attraction and production variables, and inversely proportional to distance, the
functional form of the model used is shown in Equation 3-3:4
Fv= a x PP1 + P2 x ^ x JtV x V4 x St?5 x Sh?6 x Wf1 x Py*8 + P9 x ™ x Jtf™ x (3-3)
Where:
Fjj = the migration of the population from county /' to county j from 1995 to
2000
P = the population of a county in 2000. For the 0-49 age group, total
population was used. For over-50 age group only the over-50
population was used
Jt = the mean January temperature for a county
Js = the mean January hours of sunlight for a county
St = the mean July temperature for a county (S = summer)
Sh = the mean July humidity for a county (S = summer)
W = the percent water area for a county
G = county growth rate, expressed as the ratio of 2000 population to 1980
population
dy = the functional distance between / and7
a, pi, ... P16 = parameters estimated by the regression model.
4The parameters estimated by the regression model (provided below) are positive for those variables that
are directly proportional to migration (e.g., population) and negative for those variables that are inversely
proportional to migration (e.g., distance).
3-11
-------
In order to estimate this equation using multiple regression, a logarithm of both sides is taken,
which yields Equation 3-4, the linear form of Equation 3-3.
= log(a) + P 1 x log(P,.) + P2 x p, x i0g(P,.) + p3 x log(^) + (3-4)
P4 x \og(Jsj) + P5 x log(^) + P6 x log(Sh,) + P? x log(^) +
P8 x log(Py) + P9x/»y x log(P,) + PIO x log(./(j) + pi 1 x log(./Sj) +
P12 x log(^j) + P13 x \og(Shj) + P14 x log(fFy) + P15 x log(Gy) +
P16 x iog(4)
We also ran a variety of regionally-specific experimental runs to consider if different
effects were observed on a regional basis. The four Census-defined regions are the Northeast,
South, Midwest, and West. Although some differences were observed between the different
regions, the marginal improvement in model fit offered by regionally-specific models was
outweighed by the complexity of developing a separate model for each pair-wise combination of
regions.
In the equations above, population is treated differently from other variables. Initial
model runs used the form log(migration) = a + b x log(population) + other terms. Thereby,
migration is proportional to (population)6. Under this model, log(migration) always increases
with log(population) at the same rate (assuming b > 0), all else being equal. When running the
model over many decades, this caused the model to predict that most of the population of many
small counties would migrate to nearby large counties. However, such a scenario is unrealistic
as large counties grow increasingly crowded and the differential between land prices in the urban
core and land prices in suburb an/exurb an areas grows. We then modified the model so that the
slope (b) is not constant but varies slowly with the size of the population. The new model used
b = c + d x population, providing the following form: log(migration) = a + c x log(population) +
d x population x log(population) + other terms. With this revised equation, migration is
proportional to (population)(c + dx p°Pulatlon) AS expected, the fitted c coefficient was positive
while the d coefficient was small and negative, so that for small populations, the new model
looks like the old model, but for large populations, log(migration) increases with log(population)
at a slower rate. This model resulted in projections more consistent with expectations: with
more modest growth, the largest populations did not grow in extreme proportions, while the
suburban and exurban counties (particularly those in between two relatively close cities) grew
the fastest.
3-12
-------
Table 3-2. Gravity model results
Variable
Adj.R2
Intercept [log(a)]
Production Population
(Production Population)productlon Populatlon
Production January Temp.
Production January Sun
Production July Temp.
Production July Humidity
Production Water Surface
Attraction Population
(Attraction Population)Attractlon Populatlon
Attraction January Temp.
Attraction January Sun
Attraction July Temp.
Attraction July Humidity
Attraction Water Surface
Attraction 1980-2000 Growth Rate
Distance
Age group3
0-49
0.665
5.74014
0.7756
-9.12E-09
0.07301
0.04666
-1.28061
-0.4482
0.01199
0.85084
-1.58E-08
0.01362
0.06784
-0.78317
-0.38647
0.0192
0.30131
-0.98919
50+b
0.6591
3.14215
0.78429
-5.63E-08
N/A
-0.15858
-0.85029
-0.22306
-0.00422
0.84382
-8.83E-08
0.09291
0.09263
-0.78248
-0.28732
0.01334
0.54938
-0.83684
aThe variable result values correspond to the p-values in Equations 3-3 and 3-4.
Population over age 50 was used in place of total population in the Age 50+ model.
All parameters were significant at the p< 0.0001 level, with the exception of Production January Temp, in
the 50+ model, which was not significant.
3-13
-------
Table 3-3. Ranking of the contribution of independent variables to explanatory power of the models
Rank
1
2
o
J
4
5
6
7
8
9
10
11
12
13
14
15
16
Ages 0-49
Attraction Population
Production Population
Distance
Production July Humidity
Attraction July Humidity
(Attraction Population)Attractlon Populatlon
Production July Temp.
Attraction 1980-2000 Growth Rate
(Production Population)productlon Populatlon
Attraction July Temp.
Attraction Water Surface
Production January Temp.
Production Water Surface
Attraction January Sun
Production January Sun
Attraction January Temp.
Partial
R-square
for each
variable
0.3150
0.2130
0.1231
0.0052
0.0030
0.0023
0.0010
0.0010
0.0007
0.0004
0.0001
0.0001
0.0001
0.0000
0.0000
0.0000
Cumulative
model
R-square
0.3150
0.5280
0.6511
0.6563
0.6593
0.6615
0.6625
0.6636
0.6643
0.6647
0.6648
0.6649
0.6650
0.6650
0.6650
0.6650
Ages 50+*
Attraction Population
Production Population
Distance
Attraction 1980-2000 Growth Rate
(AttractionPopulation)Attactlonpopulatlon
Attraction July Humidity
(Production Population)productlon Populatlon
Production July Temp.
Production July Humidity
Production January Sun
Attraction July Temp.
Attraction January Temp.
Attraction Water Surface
Attraction January Sun
Production Water Surface
Partial
R-square
for each
variable
0.3315
0.1976
0.1143
0.0062
0.0023
0.0025
0.0020
0.0016
0.0007
0.0002
0.0001
0.0001
0.0001
0.0001
0.0000
Cumulative
model
R-square
0.3315
0.5291
0.6433
0.6496
0.6518
0.6543
0.6563
0.6579
0.6586
0.6588
0.6589
0.6590
0.6591
0.6591
0.6591
*Population over age 50 was used in place of total population in the Age 50+ model.
-------
3.5.5. Stepwise Regression Results
Table 3-2 provides a summary of the stepwise regression results, including the model R2
and the individual parameters. The R2 indicates the overall goodness-of-fit of the regression
models, or what percentage of the variance of the logarithm of the flow is explained by the
independent variables. The estimates for each parameter in the models are also provided. The
regression modeling was performed using a stepwise approach, such that at each step a term was
added if the significance level was <0.05 and terms were removed if the significance level was
>0.05. For these models, all the regression parameters except for one regression parameter in the
50+ model were statistically significant at each step (p < 0.0001), i.e., the improvement in
goodness-of-fit was statistically significant. The relative explanatory power of each variable is
provided in Table 3-3. These results show that population and distance represent the majority of
the model's explanatory power. While certain amenity variables may be more important in rural
counties, when considered with much larger urban and suburban counties, the additional
contribution of these variables is smaller, though still significant.
We also ran a correlation on each of the independent variables to determine the
prevalence of collinearity among these variables. In nearly all cases, all pairs of variables were
found to be statistically collinear at thep < 0.0001 level, although many of the estimated squared
correlations (R2) were small. With such an extremely large sample size (n = 2,397,007, or the
number of migration records in PUMS), it is relatively easy for even very low R2 values to be
statistically significant. In this case, having a very significants-value simply suggests that the
data are inconsistent with a true correlation of zero, but they could be consistent with some very
small, but non-zero, correlation.
3.5.6. Model Flow
Having estimated the gravity model parameters, the gravity model was incorporated into
the overall demographic model to estimate domestic migration. Because the underlying
migration data used in this analysis measured migration over a five-year period, the migration
was estimated at five-year intervals.
Using a pair of nested loops, the model cycles through each pair of counties, estimating
the migration from each county to every other county. Using the same amenity and distance data
discussed in Section 3.5.1-3.5.3 and the current estimated county populations as calculated by
the model, it enters the terms into the gravity model equation provided above for each of sets of
model parameters for each of the two age groups.
Because the gravity model does not consider race, gender, or age beyond the two broad
age groups, every estimation of migration from county A to county B must be applied to the
individual age, gender, and race groups. The model assumes that all groups within one migration
3-15
-------
model are equally likely to move, so it allocates the total migrants from one county to another
based on the relative populations of each group in the origin county. In reality, it is likely that
migration patterns vary by factors such as age and race. Immigrants are also found to display
different migration patterns than native-born citizens (Rogers and Henning, 1999; Perry and
Schachter, 2003). However, the initial data set used here did not provide the level of specificity
needed to add greater detail to the gravity model analysis.
Once migrants from one county to another are estimated by age, race, and gender, these
migrations are then subtracted from the origin county and added to the destination county. This
process is then repeated for every pair of counties. One major exception is that for counties that
fall in the least populated quintile of counties (approximately 620 counties), with a maximum
population in 2000 of 9,350, the gravity model is not applied, and no domestic migration is
assumed for those counties. The counties were included in the regression modeling, and
analyses indicated that the smallest counties are less likely to conform to the modeled behavior.
This higher error rate, combined with the powerful attraction of large counties, caused many
small counties to decline precipitously. Although excluding them from the gravity model (i.e.,
effectively estimating zero net domestic migration for these counties) likely leads to some
distortions in these counties, collectively they account for roughly 1% of the national population.
Therefore it is unlikely that their exclusion has much effect on the remainder of U.S. counties.
3.5.7. Gravity Model and U.S.-Adapted SRES Storylines
The gravity model provides many potential levers for adjusting migration patterns to
account for various future scenarios. To model the high and low migration patterns needed for
the SRES storylines adapted to the United States, total migrations were scaled in order to
estimate greater flows of people. In the Al and A2 storylines, where domestic migration is
assumed to be higher than in the base case, all migrations were increased by 50%. In the Bl and
B2 storylines, where domestic migration is assumed to be lower than in the base case, all
migrations were reduced by 50%. Appendix B includes a discussion on testing other migration
assumptions.
3.6. MODEL RESULTS
The demographic model was run for a base case (all parameters set to "medium") and for
the four SRES-compatible scenarios. A variety of sensitivity tests with a wider array of model
inputs were also carried out. These tests and the results are discussed in greater detail in
Appendix A. Model runs calculated population on an annual basis for 2005-2100, with county
totals reported for five-year intervals. These outputs were then used as inputs to the spatial
3-16
-------
allocation model (Chapter 4). While the detailed model output is too large to present here, a
summary discussion of the demographic model results follows.
Figure 3-3 shows the total population for the conterminous United States for the years
2005-2100, as modeled in the base case and four SRES-compatible scenarios. A2, with high
fertility and high net international migration represents the highest population scenario. The base
case and scenario B2 are the middle scenarios, with medium fertility and medium international
migration. The difference between these two scenarios lies in the domestic migration, where the
base case assumes middle-of-the-road migration flows, while B2 assumes low domestic
migration flows. As a result of this distinction, the county populations in urban and suburban
areas generally grow faster than in rural areas in the base case, but the experiences of individual
counties vary. Al and Bl, with low fertility and high international migration are the lowest of
the population scenarios. The primary difference between these scenarios occurs at the domestic
migration level, with an assumption of high domestic migration under Al and low domestic
migration under B1. The effect of different migration assumptions becomes evident in the
spatial model when the population is allocated into housing units across the landscape. A more
extensive discussion of the regional differences in population growth is included in Section 5.1.
800,000,000 -
700,000,000 -
600,000,000 -
o
I 500,000,000 -
Q.
O
Q_
400,000,000 -
300,000,000 -
200,000,000 -
j^
X
.0'
.0' r-^J
r^r~x*£*^ ^'^'~i* '
LOLnLnLnLnLnLnLnLOLn
— • — Base Case
--B--A2
- ,. B1
- .-B2
oooooooooo
C\IC\IC\IC\IC\IC\IC\IC\IC\IC\I
Figure 3-3. Total population under five ICLUS scenarios. Scenario B2 and the base case
have the same population trajectories, as do scenarios Al and Bl.
3-17
-------
In general, the need for additional development is directly proportional to population
growth. As shown in Table 5-3 below, the growth in the extent of urban and suburban areas is
greatest in the scenario with the highest projected population growth (A2). The growth in urban
and suburban areas in the other scenarios is analogous to their population growth: B2 and the
Base Case are in the middle, while Al and Bl are the lowest. With ISs, the results are somewhat
different (Table 5-5), indicating that the allocation and housing parameters adjusted in SERGoM
such as household size and commute travel time are a greater driver of changes in IS than
population alone. In this case the Al and A2 scenarios have the highest growth in IS, while the
Bl and B2 scenarios have the lowest. This suggests that decisions about which lands are utilized
are as important as actual population growth in driving land-use impacts. The impacts of the
population scenarios on land use and the methodologies used to model this are explored in the
following chapters in greater detail.
3.7. COMPARISON OF DEMOGRAPHIC MODEL WITH EXISTING PROJECTIONS
In order to substantiate the results of the demographic model, we compared the projected
county populations for five states with population projections developed by the states
themselves. The five states used were California, Colorado, Florida, Minnesota, and New
Jersey. All state projections were for the year 2030, except for New Jersey, which was for the
year 2025. These states were selected based on data availability and regional diversity.
Differences between the state-developed projections and the ICLUS projections were
anticipated. The ICLUS projections were developed using a single national model to estimate
county populations, while the individual state methodologies, while not reviewed for this report,
were likely developed using state-specific methods and information not available on the national
scale.
The results of these comparisons are shown in Figure 3-4 through Figure 3-8. For each
of the five states, the graphs depict the difference between the ICLUS projections and the state
projections expressed as a percentage of the state projection on the Y-axis and the log of the
county population on the X-axis. Each data point represents one county. While the results are
mixed, the position of each trend line below the X-axis indicates that ICLUS estimates were
lower than state estimates, on average. This is likely due to a combination of several factors:
• Higher estimated fertility, domestic in-migration, or international in-migration in the state
estimates than in the ICLUS estimates;
• Differences in the state methodology that likely do not fully estimate migration patterns
in other states; and
3-18
-------
• Local knowledge about specific areas targeted for new development that was not
included in the ICLUS demographic model.
cs
01
-2
cs
rence from s
£
•3
0
80%
60% -
40% -
20% -
0% -
-20% -
-40% -
-60% -
-80% -
• *
^
o 1,00* io>ooo ««*. ifrijO n n ^» i u*m nllli tinnnnnnn ioo,oc
« ********
**
)0,000
Log of population (state estimates)
Figure 3-4. Comparison of California and ICLUS (base case) 2030 projections.
80% n
60% -
(X)
1 40% -
fi
a 90%
Qj A:\J /O
03
g 0% -
| -20% -
Ig -40% -
5?
-60% -
-80% -
4
* •
« ^
•
o ^ 1,000 4 \^ »io,ooo 100,000 \,oyi
* * * »4 * *
» ***» * ***
*v
Log of Population (state estimate)
,000
Figure 3-5. Comparison of Colorado and ICLUS (base case) 2030 projections.
3-19
-------
oU /o "
60% -
is 40% -
1 20% -
S
as
S 0% -
| -20% -
=g -40% -
-60% -
-80% -
*
* *
0 1,000 IM^Qj) * 10^,0*0 »J*»*»00"«""H 10,00
«v* *
Log of Population (state estimate)
0,000
Figure 3-6. Comparison of Florida and ICLUS (base case) 2030 projections.
10,000,000
Log of Population (state estimate)
Figure 3-7. Comparison of Minnesota and ICLUS (base case) 2030 projections.
3-20
-------
80% -
60% -
is 40% -
1 20% -
2
VI
S 0% -
g -20% -
| -40% -
-60% -
-80% -
•
*
* * *2*^9>
o 1,000 10,000 loojytf*****'*^ ^,$0,000 10,00
<*»
Log of Population (state estimate)
0,000
Figure 3-8. Comparison of New Jersey and ICLUS (base case) 2025 projections.
A second important observation is the slight positive slope in each trend line. This
indicates that the ICLUS model is likely to underestimate the population of smaller counties
more than larger counties. This result was anticipated because of the large power of population
in the gravity model—in original runs, large counties grew enormous and small counties were
reduced to nothing over several decades. As a result, the lowest quintile of counties were
excluded from the gravity model and a limiting term was added that helped slow growth in the
largest counties. The downside of this is that the ICLUS model is not able to predict population
growth due to migration in small rural counties with high natural amenities. McGranahan (1999)
discusses the role of natural amenities in driving rural population growth, and while the ICLUS
model included natural amenities so as to model this trend, the stronger predictive power of
population overwhelmed the impact of natural amenities.
In terms of overall population though, the trend line indicates that the larger the county,
the more likely it was to be closer to the X-axis. This indicates that the percentage difference
between state projections and ICLUS projections is smaller for large counties, where a majority
of the population lives. With the exception of Florida, where the difference between the total
state population projection and ICLUS was 11%, all of the other state projections discussed here
differed from ICLUS by less than 7%. Given the heterogeneity of data and methods inherent in
the state-specific projections as compared to the national approach applied at the county level for
ICLUS, we were satisfied with the performance of the ICLUS demographic model.
3-21
-------
4. SPATIAL ALLOCATION MODEL
We selected a spatial allocation model to distribute the population into housing units
across the country. SERGoM (Theobald, 2005) was used to develop the land-use projections for
this effort. This section begins with a discussion of the reasons for selecting SERGoM, and
continues with a discussion of SERGoM's methodology. We then conclude this section with a
discussion of how the demographic model outputs were incorporated into SERGoM and how the
model was adjusted for the SRES-compatible scenarios.
4.1. RATIONALE FOR THE SELECTION OF SERGoM
SERGoM, unlike the majority of land-use change models, allocates a full continuum of
HD, from urban to rural. This allows a more comprehensive examination of growth patterns,
since exurban/low-density development generally has a footprint 10 times as large as urban areas
and is growing at a faster rate than urban areas (Theobald, 2005). Hence, it is an important
aspect of possible future growth scenarios. Other modeling approaches have recognized the
importance of including land-use changes beyond the urban fringe, but these are based on
modeling assumptions that require very detailed (spatially and thematically) data at the parcel
level, such as zoning and socioeconomic considerations. A comparison of the differences
between existing models and SERGoM is provided in Theobald (2001, 2003, 2005), but we
provide a brief comparison to other existing models here for completeness. Other modeling
approaches use county-level data (such as Natural Resources Conservation Service [NRCS]
Natural Resource Inventory) and are useful for very coarse-level planning, but they do not
provide spatially explicit forecasts of land-use change (Alig et al., 2004; Nowak and Walton,
2005). Cellular automata modeling is a particularly popular form of modeling land-use change
(e.g., Batty, 1997; Theobald and Hobbs, 1998), and in a direct comparison between SLEUTH
(Clarke and Gaydos, 1998) and SERGoM in the Chesapeake Bay Watershed, no practical
difference between the models were found (even after making the SERGoM model coarser to fit
the level of resolution for the SLEUTH model that was needed because of computational limits;
Claggett et al., 2004). There are other more spatially-explicit modeling approaches, particularly
from the economic geography literature, but the data to parameterize these models simply do not
exist in an easily available format across the United States (e.g., Landis and Zhang, 1998; Irwin
and Bockstael, 2002; Waddell, 2002; Jantz et al., 2003)—which is the primary reason why there
are no other national-extent, spatially-explicit maps of forecasted HD and/or land use. Other
national-extent models, mostly from Europe, are useful approaches, but have data and
computational limits that preclude them from being applied to a large extent such as the United
States (Verburg et al., 2002, 2006). The land-use change modeling field has been very active
4-1
-------
and new methods continue to be developed and explored, including Engelen et al.'s (2003)
coupling of cellular automata model with SRES storylines and Verburg et al.'s use of CLUE-S
model in Europe (2006). Parker et al. (2008) overviews a number of theoretical issues that are
important for stronger inclusion of human agency into models.
An advantage of modeling all types of HD, especially low density rural development, is
that links to GHG emissions by HD can be made. In addition, SERGoM forecasts housing
development by establishing a statistical relationship between neighboring HD, population
growth rates, and transportation infrastructure (Theobald, 2005). The model is hierarchical,
being parameterized at national, state, and county scales. It is dynamic in that as new urban core
areas emerge, the model re-calculates travel time from these areas. Although for our model runs
we assumed a static transportation infrastructure, the travel times from urban areas do change as
a function of the emergence of new urban cores. For this modeling effort the any changes in
functional connectivity that could result from such emerging urbanization were not fed back into
the functional connectivity calculations used to calculate domestic migration (Section 3.5.3
above). SERGoM also incorporates a detailed layer of developable/un-developable areas that
incorporates public protected lands as well as private protected (e.g., through conservation
easements) lands. Finally, SERGoM was designed to forecast HD growth for large, broad
(regional to national) extents. Population forecasts are a principal driver of SERGoM; in the
model, population growth is converted to housing units, which are spatially allocated in response
to the spatial pattern of previous growth and transportation infrastructure. An important
technical advantage of this model is that it produces seamless, nationwide maps at 1-ha
resolution. The benefit of this approach is that there are fewer (internal to conterminous United
States) discrete differences across artificial analytical boundaries imposed by "piecing"
individual model runs into a nationwide map, although the allocation of new housing units is
restricted to counties. Growth rates and many model parameters are specified
spatially-explicitly, so different regions (even census tracts or neighborhoods) have different
parameters. Although not exercised in this project, some additional model parameters could be
made spatially-explicit so they too could vary regionally—these might include different HD
thresholds for "urban" or "exurban" and relative changes in household size.
SERGoM has been evaluated by comparing SERGoM-based projections to estimated
"real" conditions from 1990 and 2000 (starting from "real" 1980 conditions) and summarized in
Theobald (2005). The evaluation used multi-resolution methods advocated by Pontius (2002)
and had reasonably good performance (from 79% to 99% accuracy). It is exceedingly difficult to
evaluate land-use change models, and Verburg et al. (2006) provide some useful considerations.
For example, they differentiate the inductive versus deductive role of theory, introduce pixel
versus agent-based representations of space, and consider the key issues of scale and level of
4-2
-------
analysis. They conclude that among the most pressing issues that need to be addressed are:
validation of models and particularly multi-agent models; stronger linkage between process and
pattern-based modeling approaches; and deeper understanding of the interactions and
sensitivities of model results to scale of spatial data. It is important to remember that the goal of
the land-use change modeling effort here is to project a range of likely changes under certain
assumptions, rather than to predict the future. Compared to many other modeling approaches,
SERGoM is based on very simple assumptions that incorporate only basic land-use drivers
(population growth and transportation infrastructure)—but the trade-off is that an initial estimate
of forecasted housing patterns has been provided. A number of other models also consider what
types of cover (e.g., forest, agriculture, etc.) are being converted to residential land use.
4.2. METHODOLOGY
The spatial database generated by SERGoM provides historical, current, and future
estimates of HD for the conterminous United States. That is, it represents residential land uses,
which are the major types of development and intensification of land use related to urbanization
in the United States (note that we also recognize and map commercial/industrial land
use—however, these are static layers and we do not explicitly represent other "development" in
the form of cropland or forestry developments). HD (number of housing units per acres) was
computed for each 1-ha cell (100 m x 100 m raster; 2.47 acres). There are five main input spatial
datasets used to estimate HD. These inputs are discussed below, while a summary of the data
sources and the major assumptions used in SERGoM is presented in the Appendix in Table F-2.
1. 2000 Census. Data were compiled from the 2000 Census on the number of housing units
and population for each block and the geography or polygon boundary for each census
block using the 2000 Census geography (from the SF1 dataset). Block-groups, which are
a coarser-level aggregation of block polygons, and attributes of the number of housing
units built by decade were used to estimate the historical number of housing units in each
block. An operating assumption in estimating historical housing units is that they have
not declined over time, so that the number of housing units in any past decade (back to
1940) did not exceed the number of units in any subsequent decade (up to 2000).
Reservoirs, lakes, and wide rivers that were identified as "water blocks" were also
removed, so that no housing units were placed in these undevelopable areas.
2. Undevelopable lands. Spatial data on land ownership were compiled from a variety of
sources to create the most current and comprehensive dataset—called the UPPT
(unprotected, private protected, public protected, and tribal/native lands). The UPPT
dataset was generated by starting from the Conservation Biology Institute's PAD v4
database (CBI, 2008). We updated the PAD dataset with more current data for 21 states.
The operating assumption is that housing units do not occur on publicly owned lands
(e.g., national parks, forests, state wildlife areas, etc.) or on privately-owned, protected
4-3
-------
lands. Some state lands in the western United States (the so-called "school lands"
sections, but not "stewardship" lands) were kept in the developable category because they
are in practice sold to generate revenue for state school systems. Also, tribal lands are
often considered federal (public), but here we included tribal lands as developable
(except for known tribal parks). The portions of blocks that overlapped with public (and
other non-developable lands) were deleted to create a modified or refined block. All
housing units associated with each block are then assumed to be located in the refined
(developable) portion of the blocks. Housing units were apportioned within the refined
block using a dasymetric mapping approach described below. The final product is a
raster dataset that represents the developable and undevelopable lands for SERGoM,
which is called dev20080611 and is available through the ICLUS tools.
3. Road and groundwater well density. The existence of major roads (interstates, state
highways, county roads) was used to better allocate the location of housing units within a
block. In a previous version of SERGoM (vl, v2), housing units were spread evenly
throughout the refined blocks. Here, in v3 of SERGoM, housing units were
disproportionately weighted to areas according to fine-grained land use/cover data from
the 2001 National Land Cover Database (NLCD 2001). Because major road
infrastructure is included in the NLCD 2001 (actually burned in as values 21, 22, and
some 23), road density per se was not included. Also, in the western United States where
the rural blocks are particularly large, groundwater well density was included to refine
the allocation of units. Also, the analytical hierarchy process (Saaty, 1980) was used to
provide an estimate of logical consistency during the development of the weights (the
consistency index was 0.035, which is less that then 0.15 threshold, showing that these
estimates were logically consistent). Note also that these weights are applied in a
relative, not absolute context. That is, the number of units that will be distributed in a
given area is specified by the census block and so units are allocated in proportion to the
weights found within a given block. This is robust in the face of potential
misclassification of land cover types, because all the known housing units will be
allocated to a given block, regardless of the land cover (but note that the
undevelopable—water and public lands—portions of the blocks are excluded). That is,
the number of units are specified at the block level, and the allocation of the units are in
proportion to the land cover weights. This guarantees that the correct number of units
will always be allocated within each block, although the spatial distribution of the units is
controlled by other factors such as roads, land cover, and groundwater well density.
4. County population projections. Population projections for each county, discussed in
Chapter 3, are used to drive the growth forecasts. Additional housing units were
computed by determining the number of new housing units needed to meet the needs of
the additional population, assuming the same (in 2000) population to housing unit ratio in
each tract, using 2000 U.S. Census of Housing data.
5. Commercial/industrial land use. We also mapped locations with land uses that would
typically preclude residential development (increased HD), especially commercial,
industrial, as well as transportation land uses. Using urban/built-up categories of NLCD
2001 (not open space developed), we identified locations (1-ha cells) that had >25%
4-4
-------
urban/built-up land cover but that had also had lower than suburban levels of HD
(because high-density residential areas would otherwise be included in the urban/built-up
land cover categories). Although some re-development of central business districts
("gentrification") is occurring, SERGoM works from the operating assumption that these
are relatively smaller portions of the landscape and typically brown-field settings.
The functional flow of SERGoM using data from these five sources is illustrated in Figure 4-1.
Figure 4-1. SERGoM functional flow.
HD for each decade from 2010 to 2100 was forecast using the SERGoM v3. SERGoM is
a demand/allocation/supply model, where the number of new housing units needed for the next
decade is computed to meet the demands of the projected population, computed here for each
county (but could be other analytical unit boundaries). The average growth rate for each
state-HD class is computed from the previous to current time step (e.g., 1990 to 2000). These
average growth rates are computed using a moving neighborhood (radius =1.6 km, 500-m cell)
for 12 development classes. These classes are formed by overlaying three HD
classes—urban/suburban, exurban, and rural—with four accessibility classes measured as travel
time (minutes one way) from the nearest urban/suburban core along (existing) major roads:
0-10, 10-30, 30-60, and >60 minutes. The resulting combination creates a "surface" of raster
values that reflect historical patterns of growth—called allocation weights—and are used to
allocate the new housing units for a given time step.
Based on the Census definition of urban areas, we defined urban housing densities as less
than 0.1 ha per unit and suburban as 0.1-0.68 ha per unit. We defined exurban density as
4-5
-------
0.68-16.18 ha per unit (to 40 acres) to capture residential land use beyond the urban/suburban
fringe that is composed of parcels or lots that are generally too small to be considered productive
agricultural land use (though some high-value crops such as orchards are a notable exception).
Rural is defined as greater than 16.18 ha per unit where the majority of housing units support
agricultural production.
The strength of SERGoM is that it provides a comprehensive, consistent, and nationwide
estimate of HD. It uses the most fine-grained data set currently available, has performed
reasonably well in assessments (79 to 99% accuracy rates), and compares favorably to
parcel-level and aerial photography data during ad hoc analyses in a variety of locations in the
United States (Theobald, 2005). It assumes that growth rates and patterns are likely to be similar
to recent times (1990s to 2000). The SERGoM outputs provide much more spatial detail as
compared to the USD A Census of Agriculture,5 which are county-based and only provide data
on land in farms. One other common datatset used to estimate the extent and trend of
urbanization in the United States is the NRCS's Natural Resources Inventory (NRI).6 It is based
on relatively fine-grained aerial photo analysis, but because they are sampled data, they are
aggregated up to coarse analysis units (either county, watershed 8-digit Hydrologic Unit Codes
[HUCs], or Major Land Resource Area). NRI also categorizes urban areas into only two
classes—either as "small" or "large" development—resolving housing densities at urban and
roughly 1 per 10 acre densities.
4.3. INCORPORATING U.S.-ADAPTED SRES INTO SERGoM
In addition to changes in population that resulted from the various demographic
assumptions associated with each SRES-compatible scenario developed for the ICLUS project,
the spatial location of growth was modified using SERGoM in two ways, through household size
and travel time (Table 4-1). With SERGoM, household size is expected to reflect demographic
changes due to changes in fertility and socioeconomic changes that affect household formation.
Travel time from urban "central city" locations is used to help express how the evolution of the
urban form might by affected by changing priorities and increases in the cost of transportation.
Table 4-1 also shows how travel times are translated into this urban form. Housing units are
allocated to a surface and the allocation is weighted by the accessibility to the transportation
network, thereby influencing urban form over time to create a more "compact" form of
development when allocations near urban centers are weighted more favorably.
5http://www.agcensus.usda.gov/.
6http ://www. ncgc .nrcs .usda. gov/products/nri/.
4-6
-------
Table 4-1. Summary of adjustments to SERGoM v3 for SRES storylines
Storyline
Al
Bl
A2
B2
Baseline
Household size
Smaller (-15%)
Smaller (-15%)
Larger (+15%)
NC
NC
Travel time (minutes)
<5; 10; 20; 30; 45; >45
Weighting (in percent)
75; 75; 85; 90; 95; 100
90; 95; 85; 90; 95; 100
75; 75; 85; 90; 95; 100
90; 95; 85; 90; 95; 100
75; 75; 85; 90; 95; 100
Form
NC
Slight compact
NC
Slight compact
NC
NC = No change from the U.S. Census Bureau's estimates.
The first SERGoM modification changed assumptions about households, particularly
household size (roughly family size), defined as the number of people living in a single housing
unit. Population projections from the U.S. Census assume that the ratio of population per unit,
computed at the tract level from the 2000 U.S. Census data, is static. We modified this ratio to
reflect assumptions in the SRES storylines to adjust for assumed changes in demographic
characteristics. For example, SRES Al and Bl assume smaller household sizes (reduction by
15%), whereas scenarios B2 and baseline are not changed and A2 assumes a 15% increase in
household size (Jiang and O'Neill, 2007). The changes in household size correspond to changes
in fertility rates that are assumed under the different storylines. Under Al and Bl, where fertility
is lowest, smaller average household sizes are also expected. Conversely, A2 has the highest
fertility rates, so an increase in household sizes is expected. In B2, which uses the medium
fertility rates, household sizes are not changed.
The second modification involves changing travel times by adjusting weighting values as
a function of distance away (travel time) from urban cores. Urban area (<5 minutes) weights can
be lowered by a given percentage to reflect a carrying capacity or saturation of an area, specified
by zoning perhaps; or raised to reflect increased desire for urban living (lofts, gentrification,
etc.). Exurban area weights (-30-60 minutes) can be lowered to reflect assumptions of lower
rates of development due to increased fuel prices or can be used as a surrogate for lower land
availability because of increased conservation purchases (or easements). It can also be raised for
exurban areas to reflect increased "urban flight" of baby-boomer retirees and rural amenities.
This weighting surface is re-computed at each time step. We modified the weights of travel
times for the Bl and B2 storylines to model a "compact" growth scenario (see Table 4-1). Given
the environmental orientation of the Bl and B2 storylines, we assumed that growth patterns in
4-7
-------
these scenarios would place a greater emphasis on promoting denser growth patterns closer to
existing urban centers.
We parameterized SERGoM to reflect the SRES storylines in the following ways. First,
the Al and Bl scenarios were modeled to reflect a 15% decline in average household size. A2
was modeled to show a 15% increase in average household size. B2 was modeled with no
change in household size. This modification changes the resulting number of housing units
because housing units in SERGoM are a product of the population growth and household size.
That is, number of housing units depends on the number of new people needed to be housed,
along with the assumed number of people per housing unit. Second, to model the "compact"
growth scenarios for the Bl and B2 scenarios, SERGoM was run with modifications to the
spatial allocation of new housing units as a function of travel time from urban cores (Table 4-1).
That is, the allocation weights were modified by multiplying the modification factor (e.g., 75%
for <5 minutes travel time for Al) times the overall weight developed from past growth patterns
(as a function of HD and travel time away from urban areas).
4.4. INTEGRATION OF DEMOGRAPHIC, SERGoM, AND IMPERVIOUS MODELS
The demographic inputs from each of the U.S.-adapted SRES storylines were fed into
SERGoM, along with the SRES-specific adjustments in SERGoM for a given SRES storyline.
The outputs of SERGoM were then mapped for each decade from 2000 to 2100 (Appendix A).
We also used the IS model to convert the SRES HD estimates to the total percent IS cover
(Section 5.3). Chapter 5 discusses some preliminary analyses and potential future developments
of the ICLUS modeling framework.
4-8
-------
5. IMPACTS AND INDICATORS ANALYSIS
5.1. RATES OF GROWTH IN DIFFERENT REGIONS
The growth rates of the different regions of the United States under the various SRES
storylines provides some interesting insight into the potential relative growth patterns in the
coming decades. For this analysis of regional growth patterns, we used the U.S. Census regions
(listed in Table 5-1). A separate analysis using EPA regions is presented in 0. The populations
of each of the Census regions and scenarios for 2005, 2030, 2060, and 2090, as well as the
growth rates for each intervening period are presented in Table 5-2. These data are then
displayed graphically to compare the different regions and scenarios (Figure 5-1 to Figure 5-13).
Table 5-1. U.S. Census regions
Census region
Northeast
Midwest
South
West*
States
Connecticut, Maine, Massachusetts, New Hampshire, New Jersey, New
York, Pennsylvania, Rhode Island, Vermont
Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri,
Nebraska, North Dakota, Ohio, South Dakota, Wisconsin
Alabama, Arkansas, Delaware, District of Columbia, Florida, Georgia,
Kentucky, Louisiana, Maryland, Mississippi, North Carolina, Oklahoma,
South Carolina, Tennessee, Texas, Virginia, West Virginia
Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico,
Oregon, Utah, Washington, Wyoming
'Alaska and Hawaii were excluded from the West region in this analysis.
Figure 5-1 through Figure 5-5 compare the population in the four Census regions under
each of the four SRES storylines and the base case. In all five sets of population projections, the
South region remains the most populous region, while growing faster than most other regions in
most scenarios. The West region, which begins with approximately the same population as the
Northeast and Midwest regions, outpaces those two regions in all scenarios except for B1.
Across the board, all regions experience growth in all scenarios, with the exception of the
Midwest region in scenarios Al and Bl.
5-1
-------
Table 5-2. Projected regional populations and growth rates
Census
region
Population
2005
2030
2060
2090
Growth Rate (%)
2005-
2030
2030-
2060
2060-
2090
Base case
Northeast
Midwest
South
West
54,679,292
65,936,398
108,981,468
64,973,375
63,384,211
71,250,888
134,649,231
81,705,117
70,279,618
73,287,983
158,547,450
96,926,417
79,600,361
76,745,425
187,417,301
114,073,784
16%
8%
24%
26%
11%
3%
18%
19%
13%
5%
18%
18%
Al Storyline
Northeast
Midwest
South
West
54,679,292
65,936,398
108,981,468
64,973,375
66,910,792
69,265,132
140,717,741
82,571,676
73,137,911
65,378,806
162,287,975
91,877,085
76,159,296
59,355,485
174,443,647
94,284,391
22%
5%
29%
27%
9%
-6%
15%
11%
4%
-9%
7%
3%
A2 Storyline
Northeast
Midwest
South
West
54,679,292
65,936,398
108,981,468
64,973,375
64,718,449
71,810,622
141,466,245
85,129,358
78,041,141
80,335,490
186,892,292
111,597,232
101,339,355
98,476,101
262,941,111
153,721,857
18%
9%
30%
31%
21%
12%
32%
31%
30%
23%
41%
38%
Bl Storyline
Northeast
Midwest
South
West
54,679,292
65,936,398
108,981,468
64,973,375
68,481,437
72,973,243
136,176,035
81,468,120
76,906,738
72,166,976
152,179,001
90,471,745
82,327,874
67,641,537
159,582,292
93,288,362
25%
11%
25%
25%
12%
-1%
12%
11%
7%
-6%
5%
3%
B2 Storyline
Northeast
Midwest
South
West
54,679,292
65,936,398
108,981,468
64,973,375
64,097,466
73,134,039
132,513,626
81,182,232
72,006,459
77,070,686
153,888,715
96,254,307
82,638,846
82,236,752
179,498,472
113,554,597
17%
11%
22%
25%
12%
5%
16%
19%
15%
7%
17%
18%
5-2
-------
Northeast
Midwest
South
West
2005 2010 2020 2030 2040 2050 2060 2070 2080 2090
Year
2100
Figure 5-1. Base case population by Census region.
2005 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100
Year
Figure 5-2. Al storyline population by Census region.
2005 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100
Year
Figure 5-3. A2 storyline population by Census region.
5-3
-------
350 -,
Northeast
M idwest
South
West
2005 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100
Year
Figure 5-4. Bl storyline population by Census region.
350 -,
300
B 250
o
1 200
Northeast
- M idwest
- South
-West
2005 2010 2020 2030 2040 2050
Year
2060
2070
2080
2090
2100
Figure 5-5. B2 storyline population by Census region.
Figure 5-6 through Figure 5-10 compare the average annual growth rates during each
modeled decade under the different scenarios (e.g., a growth rate of 1.01 indicates growth of
1%). Under the base case and B2 storyline, population growth rates are highest during the first
time period, then level off for the remaining two periods. The Al and Bl storylines, by
comparison, generally decline in growth rate throughout the 21st century. The A2 storyline,
which has the highest overall population growth, is the only one that exhibits steady increasing
rates of population growth over the next century.
5-4
-------
1.015 -,
0.995
Time Range
Figure 5-6. Base case annual population growth rates by region.
1.015 -,
1.01 -
•S 1.005 -
1
O
1 -
0.995
2005-10 2010-20 2020-30 2030-40 2040-50 2050-60 2060-70 2070-80 2080-90 2090-
Time Range
Figure 5-7. Al storyline annual population growth rates by
Census region.
1.015 -i
2005-10 2010-20 2020-30 2030-40 2040-50 2050-60 2060-70 2070-80 2080-90 2090-
Time Range
Figure 5-8. A2 storyline annual population growth rates by
Census region.
5-5
-------
1.015 -,
1.01
•S 1.005
o
0.995
2005-10 2010-20 2020-30 2030-40 2040-50 2050-60 2060-70 2C
Time Range
Figure 5-9. Bl storyline annual population growth rates by
Census region.
1.015 -i
0.995
2020-30 2030-40 2040-50 2050-60 2060-70
Time Range
Figure 5-10. B2 storyline annual population growth rates by
Census region.
Figure 5-11 through Figure 5-14 provide comparisons of the storylines for each of the
four Census regions. A2 produces the highest population in each region, confirming our
expectations based on our interpretation of the SRES storylines. Al produces the lowest
population in the Northeast and Midwest regions, while B1 produces the lowest population in the
South and West regions. This is because domestic migration, which is expected to continue to
trend toward the South and West (and away from the Midwest and Northeast), is set to "high"
under Al and "low" under Bl. Again, all regions grow under all scenarios, with the exception of
the Midwest region under Al and B1.
5-6
-------
Basecase
Al
--B---A2
Bl
- - & - - B2
2030 2040 2
Year
2060 2070 2080 2090 2100
Figure 5-11. Northeast region population by storyline.
-El
2030 2040 2050 2060 2070 2080 2090 2100
Year
Figure 5-12. Midwest region population by storyline.
350 -,
Basecase
Al
- - & - - A2
Bl
- - D - - B2
Year
Figure 5-13. South region population by storyline.
5-7
-------
200 -,
1 80
1 60
X
o 140
g 120 -
S
^ 100 -
'•§ 80 -
c3
c- ou "
Q
0 -
s
Cl' X
^•n^'^
...n---sf",^n — °— -
^a^=^a^ Q- — *"" -• * —
Q— — Q-^""^ 0
A
— -H. - -
--D--
2005 2010 2020 2030 2040 2050 2060 2070 2080 ;
Year
. J3
Or'
•ct— —
Rnsecnse
Al
A2
Bl
B2
.090 2100
Figure 5-14. West region population by storyline.
5.2. HOUSING DENSITY TRENDS
The projected growth in population and HD is anticipated to lead to corresponding
impacts on environmental attributes such as water quality and air quality. Since the challenges
are expected to be greater in urban and suburban areas, we used the model outputs to estimate the
growth in these higher density areas. Under all modeled scenarios, urban and suburban areas are
expected to increase between 60 and 164%. For this analysis, urban and suburban areas are
defined as those areas with more than -0.6 units per acre (or less than 1.68 acres per unit). This
land class is expected to increase the most in the A2 scenario, adding over 193,000 km2 over the
next century, or 164% more than 2000 levels (about 118,000 km2) for a total of over
300,000 km2 of urban/suburban area in 2100 (Table 5-3). Other increases are expected to be
significant, but more moderate than the A2 scenario, with Bl having the smallest increase (60%)
(Table 5-3; Figure 5-15). The non-intuitive result that B2 has a higher amount of urban/suburban
area as compared to the base case may be the result of the net trade-off of negatively weighting
growth in regions beyond suburban areas (i.e., exurban and rural areas); this growth results in a
greater extent of the land surface containing urban/suburban densities, compared with the base
case where those housing units are more frequently in exurban or rural categories. Note that we
do not include in our estimates of urban/suburban housing densities the over 32,300 km2 of
estimated commercial/industrial land cover for 2000.
We also examined which broad land cover types were likely to be most affected by the
projected development patterns (Table 5-4). To do this, we quantified the spatial overlap of the
urban, suburban, and exurban housing densities (>1 unit per 40 acres) on the existing major land
cover type as characterized by NLCD 2001 Anderson Level I coding. We did this by
aggregating the 30-m resolution data to 90-m cells using the majority class filter. The largest
5-8
-------
impacts for all scenarios, both in terms of percentage and total area, are estimated to be on
agricultural (cropland) cover where approximately 33% of area is converted into housing in these
scenarios. Although wetlands cover less land area, our scenarios convert 30-36% of wetlands to
housing. Shrublands are similar in total area converted, although the scenarios present more of a
range in the amount converted (25-34%).
Table 5-3. Projected urban and suburban area increases in modeled
scenarios, 2000-2100 (km2)
Scenario
Base case
Al
A2
Bl
B2
2000
118,468
118,468
118,468
118,468
118,468
2050
177,066
194,312
192,878
174,063
181,857
2100
263,315
223,272
312,426
189,649
244,080
% Increase, 2000-2100
99%
88%
164%
60%
106%
350000
100000
2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100
Year
•D- -
•D- -
Figure 5-15. Urban/suburban housing land-use trends for ICLUS SRES storylines.
5-9
-------
Table 5-4. Projected area (km ) effects of urban, suburban, and exurban
housing densities on NLCD 2001 land cover types in modeled scenarios for
2050
Scenario
Current
Forest
505,280
Shrubland
56,258
Grassland
83,257
Agriculture
(cropland)
188,316
Wetland
51,493
Total area change between current and 2050
Base Case
Al
A2
Bl
B2
-54,656
-57,267
-55,867
-55,389
-50,560
-20,491
-21,276
-20,956
-19,161
-18,156
-18,959
-19,755
-19,549
-17,920
-17,132
-58,054
-59,188
-58,887
-57,613
-51,884
-18,035
-18,787
-18,852
-16,958
-16,466
Percent change between current and 2050
Base Case
Al
A2
Bl
B2
10.8%
11.3%
11.1%
11.0%
10.0%
36.4%
37.8%
37.2%
34.1%
32.3%
22.8%
23.7%
23.5%
21.5%
20.6%
30.8%
31.4%
31.3%
30.6%
27.6%
35.0%
36.5%
36.6%
32.9%
32.0%
5.3. IMPERVIOUS SURFACES
ISs such as pavement and roofs degrade water quality, and by collecting pollutants,
increasing run-off during storm events, and absorbing heat they also contribute to the heat island
effect (Frazer, 2005). Although there are a variety of ways that IS plays a role in emissions or
climate, in this document we pursued only the use of IS as a general indicator—not specifically
tied to possible changes in carbon cycling, emissions, or heat island effects.
To develop national estimates of likely future IS, we developed a statistical relationship
between 2000 HD and the NLCD 2001 Percent Urban Imperviousness (ISPUi) data set.7 We
aggregated the 30 m ISpui data set to 900-m2 cells and then resampled using bilinear
interpolation to 1-km2 resolution. We aggregated the SERGoM 1 ha HD up to 1 km2 as well.
Based on a spatially-balanced random sample of 200,000 points from across the conterminous
United States, we developed a relationship using the cv.tree function in S-Plus (Insightful
Corporation, Seattle, WA) between ISpui and HD2ooo- To generate a map of IS based on the HD
7 http://www.mrlc.gov/multizone_download.php?zone=4; accessed 12 February 2007.
5-10
-------
(!SHD), we converted the tree into a set of if-then-else conditional statements in ArcGIS. A brief
comparison of our modeled IS to estimates of "real-world" IS (computed from interpretation of
high-resolution aerial photography) provide evidence that our modeling effort worked reasonably
well—resulting in R2 between 0.69 and 0.96. For example, we obtained an R2 = 0.69 for 80
"real-world" points used by Elvidge et al. (2004). Also, we obtained an R2 = 0.69 and R2 = 0.96
using "real-world" data points generated in Frederick County, Maryland and Atlanta, Georgia
(Exum et al., 2005). Because our estimates of IS are produced on a per 1 km2 pixel basis, we
were able to roughly compare our estimates to Exum et al.'s (2005) by averaging pixels that fell
within their 10-12-digit HUCs. It would be useful to compare our results to other
aerial-photography-based estimates, but to our knowledge the spatial datasets of these estimates
of IS are not readily obtainable. The detailed methods and results are described in Theobald et
al. (2009) and in Appendix C.
Because we estimated IS as a function of HD, the modifications to urban form and
consequently HD computations for scenarios Bl and B2 carry through. That is, the estimated IS
is a function of the HD specifically, and not just simply the number of housing units. This
captures the non-linear (decrease) in IS percentage on a per housing unit basis.
5.3.1. Impervious Surface Calculations
The total percent IS (computed at 1 km2) for the United States in 2000 was 80,094 km2.
We developed a regression model (described in Appendix C) that relates HD estimates in 2000
to estimates from the ISPUi from the NLCD 2001 dataset. Based on that statistical relationship,
we were able to forecast likely changes to IS for different future patterns of land use that reflect
our SRES growth scenarios (Table 5-5; Figure 5-16). To aid interpretation and mapping of our
results, we classified our continuous, quantitative estimates of IS into five legend classes:
0-0.9% (unstressed), 1-4.9% (lightly stressed), 5-9.9% (stressed), 10-24.9% (impacted), and
>25% (degraded), following Slonecker and Tilley (2004) and Elvidge et al. (2007) and Theobald
et al. (2009). Note that the use of legend labels such as "stressed" are for cartographic purposes
only and should be used as a general indicator of the level of IS.
5.3.2. Percent of Watersheds Over 5% Impervious
Our estimates here do not include IS for known commercial/industrial lands (in
2000)—that added an additional 13,430 km2 of IS area. All HD classes were included when
estimating the IS. In 2000, urban/suburban areas (<1.68 acres per unit) comprised 49.4% of the
total IS (accounting for different percent IS), exurban areas (1.68-40 acres per unit) comprise
34.2%, and rural comprised 16.3%. We estimate that in 2000 there were 136 8-digit HUCs that
5-11
-------
were stressed or higher (at least 5% IS), and this will likely increase to between 208 to 231 in
2050 and to between 225 and 326 in 2100.
In general, there are fairly large differences between the amount of IS that likely will
result from different growth scenarios—from a 5.5% increase (Al) from base case to a 1.2%
decline (Bl) from base case in 2050. Figure 5-17 to Figure 5-26 show the IS patterns for each
U.S.-adapted SRES storyline and the relative differences between these scenarios and the base
case.
Table 5-5. Impervious surface estimates for SRES storylines
SRES
Base case
Al
A2
Bl
B2
Year 2050
Total impervious
surface (km2)
106,174
112,052
110,574
104,946
105,252
Number of
stressed 8-digit
HUCs(of2101)
221
231
229
208
217
Year 2100
Total
impervious
surface (km2)
124,835
122,533
149,754
110,949
122,933
Number of
stressed 8-digit
HUCs(of2101)
257
256
326
225
248
"Stressed" level is defined as at least 5% impervious surface.
75000
2100
Figure 5-16. Impervious surface area estimates, 2000-2100.
5-12
-------
Unstressed (<1%)
Lightly Stressed (1-5%)
Stressed (>5-10%)
Impacted (> 10-25%)
Damaged (>25%)
500
Miles
Figure 5-17. 2050 estimated percent impervious surface, base case.
-------
I
K^
-&.
Increase in Impervious Surface
.
1-10%
> 1 0-50%
>50-100%
>100%
500
Miles
Figure 5-18. 2000-2050 relative change in impervious surface, base case.
-------
Unstressed (<1%)
Lightly Stressed (1-5%)
Stressed (>5-10%)
Impacted (> 10-25%)
Damaged <>25%)
500
Miles
Figure 5-19. 2050 impervious surface, Al storyline.
-------
Increase in Impervious Surface
.
1-10%
> 1 0-50%
>50-100%
>100%
500
Miles
Figure 5-20. 2000-2050 relative change in impervious surface, Al storyline.
-------
Unstressed (<1%)
Lightly Stressed (1-5%)
Stressed (>5-10%)
Impacted (> 10-25%)
Damaged <>25%)
500
Miles
Figure 5-21. 2050 impervious surface, A2 storyline.
-------
I
K^
oo
Increase in Impervious Surface
500
Miles
Figure 5-22. 2000-2050 relative change in impervious surface, A2 storyline.
-------
Unstressed (<1%)
Lightly Stressed (1-5%)
Stressed (>5-10%)
Impacted (> 10-25%)
Damaged <>25%)
500
Miles
Figure 5-23. 2050 impervious surface, Bl storyline.
-------
Increase in Impervious Surface
500
Miles
Figure 5-24. 2000-2050 relative change in impervious surface, Bl storyline.
-------
to
Unstressed (<1%)
Lightly Stressed (1-5%)
Stressed (>5-10%)
Impacted {> 10-25%)
Damaged (>25%)
500
Miles
Figure 5-25. 2050 impervious surface, B2 storyline.
-------
to
to
Increase in Impervious Surface
.
1-10%
> 1 0-50%
>50-100%
>100%
500
Miles
Figure 5-26. 2000-2050 relative change in impervious surface, B2 storyline.
-------
5.4. OPTIONS FOR FUTURE STUDY
IS calculations and regional growth rates are just the first set of many possible analyses
using results from this project. The HD and population projections can inform modeling
exercises that consider such diverse areas of research as traffic volumes, air quality, and water
quality. As a part of this first phase of the ICLUS project we will distribute map products on the
Web and through an ArcGIS tool. This tool will allow users to customize maps to a certain
extent, by varying some of the allocation parameters (i.e., travel times) and by selecting smaller
spatial extents (i.e., several states or a single state).
In subsequent phases of this project the demographic and housing allocation models
could be further modified to incorporate climate change variables, incorporate additional factors
affecting population change and housing patterns, or consider specific policy responses such as
an emphasis on Smart Growth or other low-impact development patterns. While efforts here
benefit from a whole-country approach, regional differences in planning efforts and local zoning
laws are likely to have a significant impact on land use. A future expansion on this work could
include regional variation in how populations grow. The current set of scenarios also does not
reflect the effects of climate change on development patterns. Some climate change effects, e.g.,
sea level rise, are likely to have significant impacts on development patterns by the end of the
century. By integrating projected changes in sea level, projected population growth, and land
development patterns, the resulting scenarios would provide valuable information to planners
interested in coastal development, transportation infrastructure vulnerability, HD, water quality
(e.g., salt water intrusion), and other endpoints of concern. Such integration is a possible next
step in the development of this project in order to begin to incorporate climate change variables
into the models to facilitate more comprehensive assessments of the combined effects of climate
change and land-use change. Additional options for incorporating climate change variables
include modifications to the gravity model to allow climate variables to change over time.
Another possible modification might examine changes in HD regionally. Current HD
class ranges are static and the same throughout the country. This can be easily modified
geographically, particularly west versus east to explore different development patterns
regionally, such as larger ranches or farms in the western United States (e.g., 10-40 acres
exurban versus -2-10 acres exurban in the east). Also, the amount of developable land is
currently assumed to be static, but that could be modified to remove additional protected lands.
However, an approach would first need to be developed to identify where these specific lands
may be protected and what the relationship to new housing allocations and densities might be.
The driving factor in SERGoM is population, so that forecasts are exogenous variables
that are input to the model. Other population scenarios could also be modeled. In the current
version of SERGoM housing units do not move across boundaries if an analytical unit is
5-23
-------
saturated. One reason for this is that it would require coupling SERGoM with the demographic
model (that is, at each decade SERGoM would have to pass back to the ICLUS demographic
model the actual number of people/households that were placed in a county). Rather, we
endeavored to make the ICLUS demographic model more responsive to broader-scale (county
and up) trends by using the amenity-based gravity model. A future refinement could be to
couple the demographic model and SERGoM to ensure explicit distribution of people/houses.
Another way to handle this challenge, if data were available and assumptions were reasonable, is
to estimate the "build-out" (or carrying capacity) of a county based on zoning, for example.
HD currently is spread from one location to another (even across county or state
analytical units) as a function of distance away (travel time) from urban core areas. Urban "core
areas" are identified by a high HD level (e.g., urban) of a size/population/area to approximate
providing general services. In the ICLUS scenarios these parameters are specified by two
user-defined values that do not vary across the study area (e.g., nationwide model). These could
be easily changed so that they are spatially-explicit parameters, which would allow regional
variation to occur. Also note that as growth continues through time, eventually some new urban
core areas will "emerge", creating a new "hot spots" or concentrations of growth.
Some options for future research include:
• Identify new urban clusters. Currently, new urban core areas are modeled by SERGoM
in the spatial allocation process. An analysis of the detailed SERGoM outputs would
help identify new or expanded urban clusters. Such new urban clusters would likely
experience dramatic changes in air quality, water quality, and traffic. Identifying these
growth areas could help improve regional analyses and planning.
• Estimate traffic demands. Vehicle miles traveled (VMT) can be estimated for each grid
cell based on the number of housing units and available estimates on the number of
automobile trips generated per day that vary based on HD class. When combined with
average trip lengths, these projections can be used to estimate fuel consumption, travel
demand, and other factors. Combining the HD maps with the road network layer of
SERGoM can allow more sophisticated traffic analyses.
• Model air quality changes. The VMT outputs above, along with other layers of data
about stationary emissions sources, can be used with existing air quality models to
estimate projected air quality under the various scenarios.
• Analyze effect of IS on water quality. The IS analysis can be used to consider the
quality of stormwater runoff and its impact on water quality in key watersheds.
Groisman et al. (2005) suggest that one potential impact of climate change is an increase
in the intensity of individual storm events. Since these events are responsible for the
majority of impacts to water quality from stormwater runoff, examining the possible
5-24
-------
extent of ISs become even more important given the anticipated impacts of climate
change.
• Develop alternative Smart Growth scenarios. Alternative SERGoM runs could be
used to project HD under alternative development patterns that reflect Smart Growth
goals. One example would be to model denser development along existing transportation
infrastructure. The results could also be combined with some of the other suggested
analyses to estimate the performance of such strategies. Another use would be to
examine the amount of growth that could be accommodated in brown/grey field sites,
versus greenfield sites.
• Analyze effect of urban versus exurban growth on IS thresholds. Another interesting
question to ask in the future would be the degree to which growth in urban/suburban
versus exurban classes is the main cause for watersheds to cross over a threshold (e.g.,
>5%) into the stressed IS classification. It would also be interesting to explore the
consequences of assumptions about how population per housing unit would vary both
between urban and rural areas, and through time.
• Evaluate forecasted housing density patterns to more local/regional land-use
models. It would be useful to evaluate the SERGoM estimated housing density patterns
against more regional and structural models of land-use change (e.g., CLUE-S; SLEUTH;
etc.).
• Explore a stronger coupling of agricultural land conversion to residential. Typically
agricultural land (either rangeland or cropland) is converted when undergoing increases
in residential land use, which has important implications for environmental endpoints. It
would be useful in future efforts to explore a stronger coupling of residential land-use
change with agricultural land-use change in order to understand trade-offs between these
land uses.
5-25
-------
6. DISCUSSION AND CONCLUSIONS
The models and results described in this report represent a first step in creating national-
scale scenarios of HD patterns that are broadly consistent with IPCC emissions storylines. The
IPCC emissions storylines were adapted to the United States by modifying Census demographic
projections, by incorporating county-to-county connectivity and amenity variables, and by
modifying the spatial growth pattern of housing. The resulting scenarios provide benchmarks for
possible future HD patterns. The scenarios are unique in terms of presenting these results for the
conterminous United States to the year 2100 in a manner consistent with IPCC emissions
storylines. Other available land-use models are still spatially more constrained or require data
not available consistently at the national scale.
Validation of models projecting a future condition is generally challenging. We
compared our demographic projections at the county level with selected state estimates to show
that our model is in the range of other projections, and therefore can be used to generate
plausible population scenarios. The differences underscore that there are many approaches that
can generate scenarios, and that these approaches may use more detailed or finer-scale
information. However, the ICLUS methodology does produce demographic results within the
range projected by other efforts and is therefore useful in benchmarking scenarios within
established GHG emissions storylines. Subsequent model improvements will attempt to use
other datasets for additional validation studies, particularly of domestic migration patterns that
contribute to the demographic projections.
The preliminary results presented in this report call attention to the expected spatial
variability of land-use effects and their possible intersection with regional climate changes. For
examples, population growth rates in the South and West may increase the vulnerability of these
regions to water quantity and quality issues if precipitation decreases. This is one way that these
land-use scenarios can facilitate integrated assessments of climate change and land-use change.
While there is a high degree of uncertainty associated with the specific projections, the
comparison of the scenarios in these types of assessments may still be useful in highlighting
potential vulnerabilities.
Our preliminary results also show differences in the impacts on various land-cover
classes, which, along with increases in IS cover, are likely to translate to effects on air quality,
human health, water quality, and ecosystems. For example, IS cover may serve as a surrogate
for estimating certain types of emissions associated with buildings and roads and as an input to
stormwater runoff models. Further assessments of effects from changes in HD and IS cover,
including more detailed spatial analyses of which watersheds and regions may be more
vulnerable to these changes; which wetland types and in which watersheds and regions may be
6-1
-------
more impacted by land conversion; and which regions may benefit more from policies and
planning that includes Smart Growth development patterns, will be important next steps. The
explicit integration of climate change into the next phase of modeling will further facilitate these
assessments.
In conclusion, the U.S.-adapted SRES storylines produce a range of outcomes both in the
demographic model of the ICLUS project and in the spatial allocation model. The range of
population projections, housing densities, and IS cover allows for a broad examination of trends
and impacts to a variety of endpoints. The ICLUS methodology also allows for future
modifications that can incorporate more explicit climate-change information, feedbacks to
domestic migration patterns from emerging growth centers, and a variety of regional changes to
housing densities and allocation preferences. The current outputs from the ICLUS project can be
used in a variety of assessments that include effects on air quality, water quality, and any other
endpoints that are modeled using either population or land use as an input, especially when these
assessments compare differences among scenarios.
6-2
-------
REFERENCES
Alig, RJ; Kline, JD; Lichtenstein, M. (2004) Urbanization on the US landscape: looking ahead in the 21st century.
Landscape Urban Plan 69:219-234. Available online at
http://www.fsl.orst.edu/lulcd/Publicationsalpha_files/Alig_etal_2004_LUP.pdf.
Batty, M. (1997) Cellular automata and urban form: a primer. J Am Plan Assoc 63(2):266-274.
Breiman, L; Friedman, J; Olshen, R; et al. (1984) Classification and regression trees. New York: Chapman and Hall;
358 pp.
Campbell, PR. (1996) Population projections for states by age, sex, race, and Hispanic origin: 1995 to 2025" U.S.
Census Bureau. Population Division, PPL-47. Available online at
http://www.census.gov/population/www/projections/ppl47.html.
Claggett, PR; Jantz, CA; Goetz, SJ; et al. (2004) Assessing development pressure in the Chesapeake Bay watershed:
an evaluation of two land-use change models. Environ Monit Assess 94(1-3): 129-146.
Clarke, KC; Gaydos, LJ. (1998) Loose-coupling a cellular automaton model and GIS: long-term urban growth
prediction for San Francisco and Washington/Baltimore. Int J Geogr Inf Sci 12:699-714.
Conley, TG; Topa, G. (2002) Socioeconomic distance and spatial patterns in unemployment. J Appl Econom
17(4):303-327.
CBI (Conservation Biology Institute). (2008) Protected Areas Database v4. CBI, Corvallis, OR. Available online at
http://www.consbio.org/what-we-do/protected-areas-database-pad-version-4.
De'ath, G; Fabricius, KE. (2000) Classification and regression trees: a powerful yet simple technique for ecological
data analysis. Ecol., 81(11):3178-3192. Available online at http://161.lll.161.171/estadistica2004/e0081death.pdf.
Elvidge, CD; Milesi, C; Dietz, JB; et al. (2004) US constructed area approaches the size of Ohio. EOS Tran Am
Geophy Union 85(24):233-240. Available online at
http://www.co.suffolk.ny.us/stormwater/FAQs/ImperviousArticle.pdf.
Elvidge, CD; Tuttle, BT; Sutton, PC; et al. (2007) Global distribution and density of constructed impervious
surfaces. Sensors 7: 1962-1979. Available online at http://www.ngdc.noaa.gov/dmsp/pubs/ISAglobal_20070921-
l.pdf.
Engelen, G; White, R; De Nijs, T. (2003) Environment explorer: a spatial support system for the integrated
assessment of socioeconomic and environmental policies in the Netherlands. Integral Assess 4(2):97-105.
Exum, LR; Bird, SL; Harrison, J; Perkins, CA. (2005) Estimating and projecting impervious cover in the
southeastern United States. U.S. Environmental Protection Agency, Ecosystems Research Division, Athens, GA;
EPA/600/R-05/061. Available online at
http://www.epa.gov/athens/publications/reports/Exum600R05061EstimatingandProjectingImpervious.pdf.
Frazer, L. (2005) Paving paradise: the peril of impervious surfaces. Environ Health Perspect 113(7):A457-A462.
Friedl, MA; Brodley, CE. (1997) Decision tree classification of land cover from remotely sensed data. Rem Sens
Env 61(3):399-409.
Groisman, PY; Knight, RW; Easterling, DR; et al. (2005) Trends in intense precipitation in the climate record. J
Clim 18(9): 1326-1350.
R-l
-------
Manson, GA; Groop, RE. (2000) U.S. intercounty migration in the 1990s: people and income move down the urban
hierarchy. Prof Geogr 52(3):493-504.
Haynes, KE; Fotheringham, AS. (1984) Gravity and spatial interaction models. Beverly Hills: Sage Publications.
Hilderink, H. (2004) Population and scenarios: worlds to win? RIVM Report 550012001. Bilthoven, The
Netherlands. Available online at http://www.rivm.nl/bibliotheek/rapporten/550012001.pdf.
Hollman, FW; Mulder, TJ; Kalian, JE. (2000) Methodology and assumptions for the population projections of the
United States: 1999 to 2100. Population division working paper No. 38. Bureau of the Census. Washington, DC.
Available online at http://www.census.gov/population/www/documentation/twps0038.html.
Homer, C; Huang, C; Yang, L; et al. (2004) Development of a 2001 national landcover database for the United
States. Photogramm Eng Rem Sens 70(7):829-840. Available online at http://www.mrlc.gov/pdf/July_PERS.pdf.
IPCC (Intergovernmental Panel on Climate Change). (2001) Climate change 2001: impacts, adaptation and
vulnerability. Contribution of Working Group II to the Third Assessment Report of the IPCC. JJ; McCarthy, OF;
Canziani, NA; et al., eds., Cambridge, UK: Cambridge University Press. Available online at
http://www.grida.no/publications/other/ipcc_tar/?src=/climate/ipcc_tar/wg2/index.htm.
IPCC (Intergovernmental Panel on Climate Change). (2007) Climate change 2007: impacts, adaptation and
vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the IPCC. Parry, ML;
Canziani, OF; Palutikof, JP; et al., eds. Cambridge UK: Cambridge University Press; 976pp. Available online at
http://www.gtp89.dial.pipex.com/intro.pdf.
Irwin, EG; Bockstael, NE. (2002). Interacting Agents, Spatial Externalities, and the Endogenous Evolution of
Residential Land Use Pattern. Journal of Economic Geography 2(1), January: 31-54.
Jantz , CA; Goetz, SJ; Shelley, MK. (2003) Using the SLEUTH urban growth model to simulate the impacts of
future policy scenarios on urban land use in the Baltimore-Washington metropolitan area. Environ Plan B: Plan
Design 31 (2) :251-271.
Jiang, L; O'Neill, BC. (2007) Impacts of demographic trends on US household size and structure. Popul Dev Rev
33(3):567-591.
Landis, J; Zhang, M. (1998) The second generation of the California urban futures model. Part 1: Model logic and
theory. Environ Plan B: Plan Design 25:657-666.
Lee, R. (2003) The demographic transition: three centuries of fundamental change. JEconPersp, 17(4): 167-190.
Available online at http://www.soc.upenn.edu/courses/2005/spring/soc796_hpkohler/readings/lee03t.pdf.
Lutz, W; Sanderson, WC; Scherbov, S; et al. (1996) World population scenarios for the 21st century. In: Lutz, W, ed.
The future population of the world What Can We Assume Today? W. Lutz. London: Earthscan; pp. 361-396.
Matthews, TJ; Hamilton, B. (2005) Trend analysis of the sex ratio at birth in the United States. Natl Vital Stat Rep
53(20): 1-17.
McGranahan, D. (1999) Natural amenities drive rural population change. Agricultural Economic Report No.
AER781. U.S. Department of Agriculture. Available online at http://www.ers.usda.gov/Publications/AER781/.
MEA (Millennium Ecosystem Assessment). (2005) Ecosystems and Human Well-Being: Volume 2 Scenarios:
Findings of the Scenarios Working Group. Washington: Island Press. 596 pp.
R-2
-------
Nakicenovic, N; Alcamo, J; Davis, G; et al. (2000) Special report on emissions scenarios. Intergovernmental Panel
on Climate Change (IPCC). Cambridge UK: Cambridge University Press. Available online at
http://www.grida.no/publications/otlier/ipcc_sr/?src=/climate/ipcc/emission/.
Nowak, D; Walton, JT. (2005) Projected urban growth (2000-2050) and its estimated impact on the US forest
resource. J Forest. 103(8):383-389.
O'Neill, BC. (2004) Conditional probabilistic population projections: an application to climate change. Int Stat Rev
72(2)167-184. Available online at http://www.iiasa.ac.at/Research/PCC/pubs/sci-pol/ONeill2004JSR.pdf.
O'Neill, BC. (2005a) US socioeconomic futures. International network to advance climate talks, options for future
climate policy: transatlantic perspectives October 2005. Available online at
http://www.iiasa.ac.at/Research/PCC/pubs/dem-emiss/ONeill2005_INTACT_final.pdf.
O'Neill, BC. (2005b) Population scenarios based on probabilistic projections: an application for the Millennium
Ecosystem Assessment. Popul Environ 26(3):229-254.
Parker, D; Hessl, A; Davis, SC. (2008) Complexity, land-use modeling, and the human dimension: fundamental
challenges for mapping unknown outcome spaces. Geoforum 39:789-804.
Perry, M; Schachter, J. (2003) Migration of natives and the foreign born: 1995 to 2000. Census 2000 special
reports. U.S. Census Bureau. Available online at http://www.census.gov/prod/2003pubs/censr-ll.pdf.
Pontius, RG, Jr. (2002) Statistical methods to partition effects of quantity and location during comparison of
categorical maps at multiple resolutions. Photogramm Eng Rem Sens 68(10):1041-1049.
Reginster, I; Rounsevell, M. (2006) Scenarios of future urban land use in Europe. Environ Plann B: Plann Design
33(4):619-636.
Rodrigue, JP; Comtois, C; Slack, B. (2006) The geography of transport systems. New York: Routledge.
Rogers, A; Henning, S. (1999) The internal migration patterns of the foreign-born and native-born populations in the
United States: 1975-80 and 1985-90. IntMigratRev 33(2)403-429.
Rounsevell, MDA; Reginster, I; Araujo, MB. (2006) A coherent set of future land-use change scenarios for Europe.
Agri Eco Environ 114:57-68.
Sardon, JP; Robertson, DA. (2004) Recent demographic trends in the developed countries. Population
59(2):263-314.
Saaty, TL. (1980) The Analytic Hierarchy Process. New York, NY: McGraw-Hill.
Slonecker, ET; Tilley, JS. (2004) An evaluation of the individual components and accuracies associated with the
determination of impervious area. GISci Rem Sens 41(2):165-184.
Siegel, JS; Swanson, DA; eds. (2004) The methods and materials of demography, second edition. San Diego:
Elsevier Academic Press.
Solecki, WD; Olivieri, C. (2004) Downscaling climate change scenarios in an urban land-use change model. J
Environ Mod 74:105-115.
Theobald, DM. (2001) Land-use dynamics beyond the American urban fringe. GeogrRev 91(3):544-564. Available
online at http://www.amergeog.org/gr/jul01/Theobald-lst.pdf.
R-3
-------
Theobald, DM. (2003) Targeting conservation action through assessment of protection and exurban threats.
ConservBiol 17(6): 1624-1637.
Theobald, DM. (2005) Landscape patterns of exurban growth in the USA from 1980 to 2020. Ecol Soc 10(1):32.
Available online at http://www.ecologyandsociety.org/vollO/issl/art32/.
Theobald, DM; Hobbs, NT. (1998) Forecasting rural land-use change: a comparison of regression- and spatial
transition-based models. Geogr Environ Mod 2(l):65-82.
Theobald, DM; Norman, JB; Sherburne, MR. (2006) FunConnvl user's manual: ArcGIS tools for functional
connectivity modeling. Natural Resource Ecology Lab, Colorado State University. Available online at
http://www.nrel.colostate.edu/projects/starmap/FUNCONN%20Users%20Manual_public.pdf.
Theobald, DM; Stevens, Jr, DL; White, D; et al. (2007) Using GIS to generate spatially-balanced random survey
designs for natural resource applications. Environ Manag 40(1): 134-146.
Theobald, DM; Goetz, S J; Norman, J; et al. (2009) Watersheds at risk to increased impervious surface in the
conterminous US. J Hydrol Eng 14(4):362-268.
UN (United Nations). (1998) World Population Projections to 2150. Department of Economic and Social Affairs,
Population Division, United Nations, New York.
United States Census Bureau. (1992) 1980 to 1990 demographic components of change file of the U.S., states, and
counties. Available online at http://www.census.gov/popest/archives/1980s/.
United States Census Bureau. (2000) Component assumptions of the resident population by age, sex, race, and
Hispanic origin: lowest, middle, and highest series, 1999 to 2100. Available online at
http://www.census.gov/population/www/projections/natdet-D5.html.
United States Census Bureau. (2003) Census 2000: 5-Percent public use microdata sample (PUMS) files. Available
online at http://www.census.gov/Press-Release/www/2003/PUMS5.html.
United States Census Bureau. (2008) Population, population change and estimated components of population
change: April 1, 2000 to July 1, 2007 (NST-EST2007-alldata). Available online at
http://www.census.gov/popest/datasets.html.
United States Census Bureau. (2007) Census 2000. Available online at http://factfinder.census.gov.
Verburg, PH; Soepboer, W; Veldkamp, A; et al. (2002) Modeling the spatial dynamics of regional land use: the
CLUE-S model. Environ Manag 30(3):391-405.
Verburg, P; Kok, K; Pontius, RG. (2006) Modeling land-use and land-cover change. In: Lambin, E; Geist, H; eds,
Land-use and land-cover change: local processes, global impacts. New York, NY: Springer.
Waddell, P. (2002) UrbanSim: modeling urban development for land use, transportation and environmental
planning. J Am Plann Assoc 68:297-314.
R-4
-------
APPENDIX A
MAPS FOR ICLUS SCENARIOS
A-l
-------
>
*; /
/ ' '
• v.
;sp
•'**<•
500
Miles
Figure A-l. Base case, year 2010 housing density map.
-------
>
•• *;•<
•,:$*• : •*.**•£' ; *
- 1
• >'•;• '.m IP
' *1* - - • ! •«,
;i''v**3f-^' ' *"''•'
../,, ,:*feJ(0^ -
• aK -jLri. ..•** .». •TITT'Tv"
Undevelopable
Rural (>I6.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban/Suburban (<0.68 ha/unit)
Commercial/IndustriaJ
500
Miles
Figure A-2. Base case storyline, year 2050 housing density map.
-------
' *: V
V1' , ' ,
• . / '.
•
~£~* \m •
•"Bk" .•*»"•«
r- ;' » *
-^-L,< —V-
- - • mmm
.tiwSfF^ .* •• -sn
Undevelopable
Rural (> 16.18 ha/unit)
Exurban (16.18-0 68 ha/unit)
Urban'Suburban (• 0.68 ha/unit)
Commercial/Industrial
500
Miles
Figure A-3. Base case, year 2100 housing density map.
-------
r. A
i \
.
. * • . if !
•
. ,. ;?«*
•
Undevelopable
Rural (>16.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban/Suburban (O.68 ha/unit)
Commercial Industrial
i k
* "
500
Miles
Figure A-4. Al storyline, year 2010 housing density map.
-------
Undevelopable
Rural (> 16.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban/Suburban (<0.68 ha/unit)
Commercial/Industrial
• . • '' > *3P'I«Fv w-'
fe Mv-
500
Miles
Figure A-5. Al storyline, year 2050 housing density map.
-------
>
, •„
(V ?•». O"/ .*-.,< J
Undevelopable
Rural (>16.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban/Suburban (<0.68 ha/unit)
Commercial/Industrial
500
Miles
Figure A-6. Al storyline, year 2100 housing density map.
-------
oo
Undevelopable
Rural (> 16.18 ha/unit)
Exurban (16,18-0.68 ha/unit)
Urban Suburban (<0.68 ha/unit)
Commercial Industrial
• • *
•> .
•
:fc- ;
• .'^-;«R*^
. :*
!
500
Miles
Figure A-7. A2 storyline, year 2010 housing density map.
-------
./*"'. ' ' _//
:•:,*-
' "' *
' si ; '« h&
Undevelopable
Rural (> 16.18 ha/unit)
Exurbaii (16.18-0.68 ha/unit)
Urban/Suburban (<0.68 ha/unit)
Commercial/Industrial
*'• f
•.'» v •
*-• * *? v '
f
• .»v
)
500
Miles
Figure A-8. A2 storyline, year 2050 housing density map.
-------
o
» . 1*
.
• : - * i-<
'
» *.-LT fe %**ii^"'**>'' ? '•'*'""-\
'
Undevelopable
Rural (> 16.18 ha/unit)
Exurbaii (16.18-0.68 ha/unit)
Urban/Suburban (<0.68 ha/unit)
Commercial/Industrial
500
Miles
Figure A-9. A2 storyline, year 2100 housing density map.
-------
4.• r
' A
''t\ '
- -^r^^
,-
••
•:•
Undevelopable
Rural (>16.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban Suburban (-0.68 lia'iinit)
Commercial Industrial
A v
'
•
•
-------
K)
* . . • . .j-
/ . . .
\
Undevelopable
Rural (> 16.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban/Suburban (<0.68 ha/unit)
Commercial/Industrial
- .
» ;. '. -
500
Miles
Figure A-ll. Bl storyline, year 2050 housing density map.
-------
;_v *
k
r. •
Undevelopable
Rural (> 16.18 ha/unit)
Exurbaii (16.18-0.68 ha/unit)
Urban/Suburban (<0.68 ha/unit)
Commercial/Industrial
500
Miles
Figure A-12. Bl storyline, year 2100 housing density map.
-------
Undevelopable
Rural (>16.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban Suburban (-0.68 ha/unit)
Commercial Industrial
-&•' • '
;m .. - i
-
500
Miles
Figure A-13. B2 storyline, year 2010 housing density map.
-------
* ,
/
'K'-m:. .> i
v ' • • ' •" rW^m -
* '• \ ' * - ':*' .• i ^|
Undevelopable
Rural (>I6.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban Suburban (• 0 68 ha unit)
Commercial/Industrial
.'•»
•
500
Miles
Figure A-14. B2 storyline, year 2050 housing density map.
-------
i
V . . ,
-• -4
- 3
'-
.-• «7 "W; -
H»:- '
;sp^*.-- * •
Jp^: •
*'
Undevelopable
Rural (>I6.18 ha/unit)
Exurban (16.18-0.68 ha/unit)
Urban Suburban (• 0 68 ha unit)
Commercial/Industrial
500
Miles
Figure A-15. B2 storyline, year 2100 housing density map.
-------
APPENDIX B
DEMOGRAPHIC MODEL SENSITIVITY TESTING
B-l
-------
In order to explore the demographic model's response to changes to its inputs and a
variety of potential scenarios, we ran a series of sensitivity tests that paid particular attention to
the behavior of the gravity model. These tests were used to improve the model, as well as to
develop the downscaling approach to the Special Report on Emissions Scenarios (SRES). We
also compared the outputs for several states to county-level projections produced by those states.
While this project could not expect to match the level of detail and state-specific methodology
produced for each state's estimates, this did provide us with a useful benchmark for comparison
with the Integrated Climate and Land-Use Scenarios (ICLUS) outputs.
The first set of testing involved the fertility rate. International migration was held at
medium, while domestic migration was set to zero, since we were interested only in looking at
total national population at this time. The model was then run using the low, medium, and high
fertility rate scenarios provided by the U.S. Census Bureau. Figure B-l below compares the
impact of the fertility rate scenarios on national population. Under these settings, low fertility
predicts a mid-century peak in population followed by a small decline, with the medium and high
scenarios result in steadily rising population.
800,000,000
700,000,000
a 500,000,000
"3 400,000,000
B.
&
300,000,000
200,000,000
100,000,000
• Low fertility
-Medium fertility
High fertility
ooooooo oooooooooooo
Year
Figure B-l. Effect of fertility rate on national population.
Next, we considered the impact of international migration on total population. This time,
fertility was held constant at medium and domestic migration was set to zero. The model was
run using the low, medium, and high international migration scenarios provided by the U.S.
Census Bureau. Figure B-2 below displays the results of these scenarios. All scenarios result in
B-2
-------
steadily rising national population, but since medium fertility produces a small steady increase in
the population, this increase is at least partially due to fertility. Figure B-2, which presents a
wider range of scenarios, shows that the combination of low fertility and low international
migration presents the lowest possible population trajectory given our current inputs. In
Figure B-3, the outputs for the base case and four SRES-compatible scenarios are presented, as
well as the maximum and minimum scenarios (calculated using high fertility and high
immigration, and low fertility and low immigration, respectively).
600,000,000
500,000,000
400,000,000
3 300,000,000
B.
&
100,000,000
•Low
International
Migration
- Medium
International
Migration
High
International
Migration
OOOOOOOOOOOOOOOOOOO^
Year
Figure B-2. Effect of international migration on national population.
We also ran tests with the mortality rate, even though the only available data set for
mortality rates was the Census middle case, and we elected to use this set in all SRES storylines.
To explore the effect of mortality on national population, we used the Census middle case to
create additional sets of mortality rates by adjusting the Census medium scenario by +/-!%,
+1-1.5%, and +1-2% per decade so that by 2100, the high cases were 10, 15, and 20% higher
than middle and the low cases was 10, 15, and 20% lower than middle, respectively. Each was
run with fertility and international migration set to medium and domestic migration set to zero.
As Figure B-4 shows, the effect of even the strongest change was relatively small compared to
changes in the other components of change.
B-3
-------
900,000,000
800,000,000 -
700,000,000 -
600,000,000 -
B
'is 500,000,000 -
i£ 400,000,000 -
300,000,000 -
200,000,000 -
100,000,000 -
iriO^lO^lO^lO^lO^lO^lO^lO^lO^lO
ooooooooooooooooooo^
Year
»—Base Case
k— Al
J—A2
k—Bl
3—B2
|— Low fertility,
LowIM
|—High fertility,
HighIM
Figure B-3. Comparison of a broad range of scenarios.
500,000,000
400,000,000
"3 300,000,000
B.
&
100,000,000
•Mortality (-20%)
•Mortality (-15%)
•Mortality (-10%)
• Medium mortality
Mortality (+10%)
Mortality (+15%)
Mortality (+20%)
if) O if» O if» O if» O if» O <^> O <^> O <^> O <^> O if» O
0000000000000000000-H
Year
Figure B-4. Effect of mortality on national population.
B-4
-------
Due to the complexity of the gravity model, there were many possible adjustments that
could be made to change the magnitude of domestic migration. The simplest change involved
the introduction of a scaling factor. Under this adjustment, the gravity model calculations would
proceed as normal, but all calculated migrations would be scaled upward or downward. For
example, if the normal model estimated 10 migrations from county A to county B, with a 50%
scaling factor to cut migrations in half, only 5 migrations would occur. Other approaches could
involve adjusting the model coefficients and/or y-intercept. These approaches would allow more
fine tuning by increasing the attraction of large cities or increasing the friction of distance, for
example.
In this analysis, we ran the gravity model with the following nine alternative scenarios:
1) Scaling all migrations by 50%. This has the practical effect of reducing migrations.
2) Scaling all migrations by 150%. This has the practical effect of increasing migrations.
3) Increasing the production population coefficient by 20%. Since the production
population coefficient is positive, this has the practical effect of increasing migrations.
4) Decreasing the production population coefficient by 20%. Since the production
population coefficient is positive, this has the practical effect of decreasing migrations.
5) Increasing the attraction population coefficient by 20%. Since the attraction population
coefficient is positive, this has the practical effect of increasing migrations.
6) Decreasing the attraction population coefficient by 20%. Since the attraction population
coefficient is positive, this has the practical effect of increasing migrations.
7) Increasing the distance coefficient by 20%. Since the distance coefficient is negative,
this has the practical effect of reducing migrations.
8) Decreasing the distance coefficient by 20%. Since the distance coefficient is negative,
this has the practical effect of increasing migrations.
9) Increasing the 1980-2000 growth rate coefficient by 20%. Since the growth rate
coefficient is positive, this has the practical effect of increasing migrations. We did not
analyze a similar decrease because adding this coefficient was deemed to be a model
improvement and the areas of concern so far have been in the rapidly growing counties.
Because domestic migration can only be considered from a county perspective, we
compared the outputs from five states with state projections and the revised base case projections
from October. The five states—California, Colorado, Florida, Minnesota, and New
Jersey—were selected for their geographic diversity and availability of suitable county-based
B-5
-------
projections. We compared the 2030 state projections (2025 for New Jersey) with the ICLUS
base case projections and with outputs from the scenario tests.
We used the outputs of these tests to help refine the demographic projections. For
example, early runs showed that urban counties were growing much faster in the ICLUS
projections than anticipated by state projections. This led to changes in how we modeled the
attraction of migrants to urban centers (see Section 3.5.4). We also found that counties currently
identified by the states as fast-growing areas did not grow as quickly in the ICLUS model as they
did in the state projections. Since the ICLUS model was designed to be a relatively simple
national model, it was not possible to include some of the specialized local details that the states
included in their projections. Therefore, divergences from the state projections were expected.
This observation did lead us to include 1980-2000 growth rates as a term in the migration
model. As a result, those fast-growing areas continued their relatively rapid growth rates in our
projections.
B-6
-------
APPENDIX C
STATISTICAL RELATIONSHIP BETWEEN HOUSING DENSITY
AND IMPERVIOUS SURFACE COVER
C-l
-------
C.I. BACKGROUND
The goal of this analysis was to develop a model to statistically relate housing density
(HD) estimated by Spatially Explicit Regional Growth Model (SERGoM) to impervious surface
(IS) cover. To do this, we examined how IS from the NLCD 2001 (MRLC, 2001) related to HD
and ancillary variables including transportation (highways, secondary, local roads), and
neighborhood density of urban/built-up land uses. Statistical relationships were developed by
using regression tree models. This approach is useful because regression trees do not assume
linear relationships in the independent variables, can handle a range of response variable types,
are easy to construct and robust, and results are straightforward to interpret (De'ath and
Fabricius, 2000). Regression trees have been increasingly used for broad-scale remote sensing
and Geographic Information System (GIS) datasets (e.g., Friedl and Brodley, 1997; Homer et al.,
2004). We also investigated other regression-based approaches that estimate imperviousness
from land cover (e.g., NLCD 2001) and/or population data. We felt these were limited or not
appropriate for our purposes primarily because they use population data rather than housing data,
and because population is tied to primary residence, they underestimate the actual landscape
effects of housing units. We also explored adding the locations of commercial/industrial land
use from NLCD 2001 to the IS estimates for HD because they are likely to have very high IS
levels as well. Note that we developed our IS model as a function of estimated housing densities
from SERGoM to NLCD 2001 IS estimates. One alternative would be to use "real-world"
estimates developed from either field data or aerial photo interpreted data, but these data simply
do not exist (e.g., Elvidge et al., 2004 rely on 80 non-probability based sample points). Rather,
we used ad hoc datasets on "real-world" IS as a way to validate our model only, not to generate
the estimates of IS.
C.2. METHODS
We downloaded the Percent Urban Imperviousness (PUI) dataset from the
Multi-Resolution Land Characteristics Consortium website.8 The PUI dataset is produced using
a Regression Tree that includes satellite imagery and roads (Homer et al., 2004). We aggregated
the 30-m resolution to roughly 1-km2 resolution (990-m) to compute the average PUI for each
1-km2 cell. We then resampled the average PUI at 0.98-km2 cell to 1 km2 using bilinear
interpolation. The 1-km2 resolution is a commonly used resolution to develop national estimates
of imperviousness (e.g., Elvidge et al., 2004).
We aggregated the HD estimates for the year 2000 from the SERGoM (Theobald, 2005)
from 1-ha to 1-km2 resolution. This provided us with the average HD for each 1 km2. We
8http://www.mrlc.gov/multizone_download.php?zone=4; accessed 12 February 2007.
C-2
-------
generated a sample of 200,000 random points (generated in a spatially-balanced way using the
Reversed Recursive-Quadrant Randomized Raster algorithm; Theobald et al., 2007) from across
the conterminous United States. We extracted the values of both the PUT and HD at each point,
and used a Classification and Regression Tree model to develop a regression equation to develop
a relationship between PUT and HD.
We generated the percent IS for current HD using the tree function in S-Plus (Insightful
Corp, Seattle, Washington). See Brieman et al. (1984) for a review of categorical and regression
tree methods. The resulting regression tree (Figure C-l) was then evaluated using a cross
validation (cv.tr ee S-Plus function) technique to investigate if the tree over-fitted the data. As
the number of terminal nodes increase, the overall deviance decreases (Figure C-2), indicating
that the original tree is not over-fitting the data. If the deviance were to start increasing after
some point within the cross validation analysis, then the tree would need to be pruned to a size
that would minimize deviance. Because this was not the case, we decided to develop the percent
IS with the original tree. The large tree size (58 nodes) can be explained best by the relatively
poor non-parametric relationship between PUI and HD (R2 = 0.38), meaning that there was not a
simple, linear fit, as shown in Figure C-3.
The model resulted in a Residual Mean Deviance (which is the sum of the square
differences between the actual values and the predicted values divided by the sample size) of
4.671. This is equivalent to the standard error in linear regression, which is the spread of error
(in impervious units) for a given observation. The distribution of the residuals (error associated
with the IS observations) has a minimum value of-62.35, a mean of-1.132e-14 (about 0), and
a max value of 89.5. This distribution was not unexpected because there are areas where
impervious values (independent variable) had values of 0 but have positive values of HD
(dependent, response variable). Figure C-l shows the decision backbone of the full regression
tree with the length of a limb indicating deviance. The top ten nodes within the tree minimized
deviance the most, with the remaining nodes making small adjustments for non-parametric small
grain instances. Figure C-4 shows the top ten nodes within the full regression tree and the HD
thresholds used to estimate percent impervious.
C-3
-------
Figure C-l. Full regression tree backbone (58 terminal nodes) without
labels.
380000.0 1500.0 910.0
i i i i i i i i i i i i i i MI
530.0 330.0
i i i i i
190.0
ii MI
40.0
o
o
o
o
o
o
o
o
a
in
o
o
o
o
o
o
o
o
o
o
10
30
size
40
50
Figure C-2. Cross validation results for the full regression tree.
C-4
-------
0 5000 10000 15000 20000 25000 30000 35000 40000 45000 50000
Housing Density (units/ha)
Figure C-3. The relationship between percent impervious and
housing density.
bhd20QO
bhd2000.Li5<336.5
bhd2QOO.U5<1531
us<49.5.
0.2029
1.3430
5.8370
bhd2000lus<7062
bhd2000.iis<3292.5
15.3300 25.3300
40.8300
Figure C-4. Top ten terminal nodes within full regression
tree with housing density labels and percent impervious
estimates (terminal nodes).
C-5
-------
We then converted the tree into a "con" (conditional) statement that can be input into
ArcGIS as a Map Algebra statement. This statement then will then apply the regression model to
generate a spatially-explicit map (note that HD is in units of housing units x 1000 per hectare;
and "x" is HD for the cell being processed):
con(x < 1855, con(x < 410.5, con(x < 67.5, con(x < 1.5, 0.009477, con(x < 13.5, 0.358900, con(x <
49.5, con(x < 25.5, 0.529100, con(x < 41.5, con(x < 26.5, 0.706300, 0.633000), 0.592000)), 0.766400))),
con(x < 207.5, con(x < 138.5, con(x < 105.5, 0.950700, 1.191000), con(x < 158.5, 1.397000, 1.721000)),
con(x < 281.5, con(x < 220.5, 2.492000, 2.235000), con(x < 289.5, 3.624000, con(x < 321.5, 2.516000,
3.137000))))), con(x < 1087, con(x < 545.5, con(x < 417.5, 5.245000, con(x < 473.5, 3.968000, 4.651000)),
con(x < 647.5, con(x < 585.5, 7.035000, 5.772000), con(x < 918.5, con(x < 692.5, 8.319000, con(x <
900.5, con(x < 884.5, con(x < 859.5, con(x < 826.5, 6.770000, 8.867000), 5.696000), 9.420000),
5.504000)), 8.126000))), con(x < 1429.5, con(x < 1417.5, con(x< 1321.5, 9.220000, 11.580000),
5.473000), con(x< 1684.5, con(x< 1677.5, con(x< 1521.5, 13.830000, 11.730000), 31.220000),
11.010000)))), con(x < 6800.5, con(x < 3106.5, con(x < 2666.5, con(x < 2519, con(x < 2473.5, con(x <
1883.5, 16.710000, con(x< 1909.5, 11.010000, con(x< 2219.5, 15.140000, 13.710000))), 21.680000),
13.450000), con(x < 2990, 18.220000, 14.670000)), con(x < 4620.5, con(x < 3141.5, 31.230000, con(x <
3824.5, con(x < 3803, con(x < 3165.5, 10.520000, con(x < 3748, 20.880000, 26.350000)), 10.280000),
23.640000)), con(x < 4639.5, 41.870000, con(x < 6031.5, con(x < 4796, 29.470000, con(x < 5193.5,
24.020000, 26.770000)), 29.800000)))), con(x < 13884.5, con(x < 9541, con(x < 9405, con(x < 7383.5,
32.750000, con(x < 7463, 44.370000, con(x < 8153, 34.070000, 36.610000))), 25.960000), con(x < 13713,
con(x < 9619, 54.960000, con(x < 10690.5, 37.480000, 41.710000)), 24.620000)), con(x < 32918, con(x <
19997.5, 51.850000, 48.040000), 68.890000))))))
C.3. RESULTS AND DISCUSSION
Using 2000 SERGoM v3 HD, we estimated 80,094 km2 of IS (Figure C-5). Our
estimated extent of IS (80,094 km2) is fairly similar to other nationwide estimates. The NLCD
2001 Urban Imperviousness layer estimated an IS of 95,746 km2. The impervious correlation
coefficient against a ground-truth dataset was 0.83, 0.89, and 0.91 (Homer et al., 2004). Note
that we were not able to develop "bracket" or "bookend" models that incorporate under- and
over-estimations based on deviations around published error statistics because they were not
provided for the NLCD 2001 IS cover. Also, the precision of estimated PUI below 20% is
believed to diminish so that low % PUI estimates from PUI were difficult to obtain (Homer et
al., 2004).
Elvidge et al. (2004) estimated an IS area of 113,260 km2 (they actually report 112,610;
+/-12,725 km2) based on nighttime lights radiance, road density, and NLCD 1992 urban land
cover classes. Thus, we were within 12% of NLCD 2001. Because SERGoM has NOD ATA
values on public (non-developable) lands, there were some cases where IS occurs on
C-6
-------
,.^'
<
^ oW' & &'
-r 1' -7
500
Miles
Figure C-5. Estimated national impervious surface, 2000.
-------
non-developable lands, such as military bases, airports, developed portions (visitor centers) of
national parks, interstates, etc. When we filled in areas of the SERGoM-based IS that had no HD
(mostly public lands) with NLCD 2001 urban imperviousness, the total IS area increased slightly
to 83,846 km2. Thus, we recommend using this combination of datasets to better represent total
IS that gets at roads and commercial/industrial areas as well.
We also conducted an additional validation step by developing a simple linear regression
of the SERGoM-based IS against 80 data points generated from high-resolution aerial
photography of 1-km2 "chips" and used to generate the Elvidge et al. (2004) product. The
resulting R2 was 0.694. This result was better than expected, as the 80 data points were not
randomly selected, rather purposively targeted to capture a gradient of urbanization, and as a
result these points were selected to pick up much of the commercial and industrial land cover
types.
We also generated a difference map to compare the NLCD 2001-derived estimates
against the SERGoM estimates (Figure C-6 through Figure C-8). In general, NLCD 2001
estimated higher imperviousness in urban areas (shown in red), and under-represented
imperviousness in lower-density, suburban/exurban land-use areas (shown in blue). Our
estimates of imperviousness are likely underestimated in urban areas because they do not include
commercial and industrial land uses. Because it is difficult to identify low HD land uses beyond
the suburban fringe using NLCD 2001 data, it is likely that NLCD 2001 PUI slightly
underestimates IS in exurban and rural areas.
Finally, in Figure C-9 we show the results of using the SERGoM HD projections for
2030—base case and the estimated IS as a function of HD. We will need to consider how to
incorporate commercial/industrial contributions to IS estimates for future projections. For now,
we recommend reporting just HD-based IS, realizing that it is a conservative estimate, and that
future efforts should better represent urban/industrial land-use growth as a function of
population/HD growth.
-------
o
,
It
' *
• v*: *
i%
SH
y
j*'
1
P ere ent I mp ervious
Value
HI -62- -20
IBI-19--5
^H -4 - -1
:-0.9-1
^•20.1-95
* ;
-
Inbp created on 7 Jims 2007 for
EPA-ICLUS project tiy Darid
HisotaH, Colfindo Stals Umkrsrsiy
Figure C-6. Difference in impervious surface, United States.
-------
V
^ % "* /
1 ' ".Vlf
\
•*%* Jz^fc,
• *r\-jet;—•••^a*H
•- " ^-> *• -JEfe
>
;:^"?-,tj?W^.'
/ - - 'r^P -Jf^fi
i-
•/^ qr ^s.
s ^^
\ .ilaC %F%.r
\ ./> ' ....3^/
\
Percent Impendous
^| -=-10%
|-t
f
Hij c sit d cuT .nns 2(07 fa
EPA-ICLU! jm^ct V &ua
i;.:.]t.i.i.l.:. ght Uimnini
Figure C-7. Difference in impervious surface, Colorado.
-------
o
'* m: • 'm*
•i • W-
, ".-. *^;
30
IPk-KLTTS pK^
i 2007 £M
Figure C-8. Difference in impervious surface, Mid-Atlantic region.
-------
O
Miles
Figure C-9. Estimated impervious surface, base case 2030.
-------
APPENDIX D
REGIONAL POPULATION GROWTH RATES AND PROJECTIONS
BASED ON EPA REGIONS
D-l
-------
Table D-l below provides a list of the U.S. Environmental Protection Agency (EPA)
regions that were used for the analysis of regional differences in population growth. Table D-2
provides the projected populations for each of the regions and each of the modeled storylines for
2005, 2030, 2060, and 2090.
Table D-l. EPA regions
Region
Region 1
Region 2
Region 3
Region 4
Region 5
Region 6
Region 7
Region 8
Region 9
Region 10
Region
description
Northeast
Mid-Atlantic
Mid-east
Southeast
Mid-west
Southwest
Cornbelt
Mountain-west
Pacific-south
Northwest
States
Connecticut, Maine, Massachusetts, New Hampshire, Rhode
Island, Vermont
New Jersey, New Yorka
Delaware, District of Columbia, Maryland, Pennsylvania, Virginia,
West Virginia
Alabama, Florida, Georgia, Kentucky, Mississippi, North Carolina,
South Carolina, Tennessee
Illinois, Indiana, Michigan, Minnesota, Ohio, Wisconsin
Arkansas, Louisiana, New Mexico, Oklahoma, Texas,
Iowa, Kansas, Missouri, Nebraska
Colorado, Montana, North Dakota, South Dakota, Utah, Wyoming
Arizona, California, Nevadab
Idaho, Oregon, Washington0
"Puerto Rico and the U.S. Virgin Islands were not included in Region 2 for this analysis.
bHawaii, American Samoa, Guam, Northern Mariana Islands, and Trust Territories were not included in Region 9
for this analysis.
0Alaska was not included in Region 10 for this analysis.
D-2
-------
Table D-2. Projected population and growth rate by scenario and
EPA region
EPA
region
Population
2005
2030
2060
2090
Growth rate
2005-
2030
2030-
2060
2060-
2090
Base case
1
2
3
4
5
6
7
8
9
10
14,255,073
28,018,871
28,797,147
57,405,312
51,257,348
35,680,974
13,269,562
10,006,652
44,519,456
11,360,137
15,519,657
35,109,989
31,978,301
70,474,771
55,592,269
45,708,352
14,224,959
12,930,609
56,426,295
13,024,245
15,902,916
41,602,295
34,424,923
82,641,728
57,423,505
55,363,006
14,504,537
15,592,590
67,555,641
14,030,327
16,887,110
49,258,590
38,637,662
97,900,593
60,486,700
65,879,226
15,010,238
18,402,596
79,977,083
15,397,072
9%
25%
11%
23%
8%
28%
7%
29%
27%
15%
2%
18%
8%
17%
3%
21%
2%
21%
20%
8%
6%
18%
12%
18%
5%
19%
3%
18%
18%
10%
Storyline Al
1
2
3
4
5
6
7
8
9
10
14,255,073
28,018,871
28,797,147
57,405,312
51,257,348
35,680,974
13,269,562
10,006,652
44,519,456
11,360,137
15,141,889
39,117,688
32,682,229
73,800,430
54,146,181
47,647,066
13,780,804
13,473,510
56,021,194
13,654,349
14,141,988
46,902,172
34,693,103
85,111,046
51,539,278
55,559,004
12,800,038
15,807,016
61,616,885
14,511,247
12,659,207
51,792,819
36,115,978
92,041,933
47,200,686
59,071,696
11,445,472
16,909,652
62,219,659
14,785,715
6%
40%
13%
29%
6%
34%
4%
35%
26%
26%
-7%
20%
6%
15%
-5%
17%
-7%
17%
10%
10%
-10%
10%
4%
8%
-8%
6%
-11%
7%
1%
1%
D-3
-------
Table D-2 (continued)
EPA
region
Population
2005
2030
2060
2090
Growth rate
2005-
2030
2030-
2060
2060-
2090
Storyline A2
1
2
3
4
5
6
7
8
9
10
14,255,073
28,018,871
28,797,147
57,405,312
51,257,348
35,680,974
13,269,562
10,006,652
44,519,456
11,360,137
15,987,287
35,917,932
32,999,574
74,318,414
56,019,601
47,833,571
14,369,471
13,630,155
58,911,077
13,137,594
18,051,181
45,986,762
40,088,688
97,906,372
63,041,628
64,361,337
15,919,641
18,236,722
78,044,717
15,229,105
22,566,774
61,228,288
54,511,203
138,195,518
77,561,440
90,250,013
19,525,569
25,157,767
108,283,362
19,198,490
12%
28%
15%
29%
9%
34%
8%
36%
32%
16%
13%
28%
21%
32%
13%
35%
11%
34%
32%
16%
25%
33%
36%
41%
23%
40%
23%
38%
39%
26%
Storyline Bl
1
2
3
4
5
6
7
8
9
10
14,255,073
28,018,871
28,797,147
57,405,312
51,257,348
35,680,974
13,269,562
10,006,652
44,519,456
11,360,137
15,141,889
39,117,688
32,682,229
73,800,430
54,146,181
47,647,066
13,780,804
13,473,510
56,021,194
13,654,349
14,141,988
46,902,172
34,693,103
85,111,046
51,539,278
55,559,004
12,800,038
15,807,016
61,616,885
14,511,247
12,659,207
51,792,819
36,115,978
92,041,933
47,200,686
59,071,696
11,445,472
16,909,652
62,219,659
14,785,715
6%
40%
13%
29%
6%
34%
4%
35%
26%
20%
-7%
20%
6%
15%
-5%
17%
-7%
17%
10%
6%
-10%
10%
4%
8%
-8%
6%
-11%
7%
1%
2%
D-4
-------
Table D-2 (continued)
EPA
region
Population
2005
2030
2060
2090
Growth rate
2005-
2030
2030-
2060
2060-
2090
Storyline B2
1
2
3
4
5
6
7
8
9
10
14,255,073
28,018,871
28,797,147
57,405,312
51,257,348
35,680,974
13,269,562
10,006,652
44,519,456
11,360,137
15,543,670
35,422,606
32,086,103
69,018,096
57,089,003
45,225,069
14,552,312
12,703,387
55,960,255
13,326,862
15,997,050
42,541,498
34,216,490
79,590,169
60,336,151
54,516,450
15,227,145
15,085,697
66,985,991
14,723,525
17,077,923
51,216,096
37,647,265
92,831,550
64,684,730
64,720,598
16,041,584
17,639,212
79,576,422
16,493,287
9%
26%
11%
20%
11%
27%
10%
27%
26%
17%
3%
20%
7%
15%
6%
21%
5%
19%
20%
10%
7%
20%
10%
17%
7%
19%
5%
17%
19%
12%
D-5
-------
APPENDIX E
COMPONENT AND COHORT MODEL DATA
E-l
-------
The following tables provide summary values for the entire population; actual rates used
in the model vary by age, sex, and race/ethnicity.
Table E-l. Fertility rates (births per 1000 women)
Year
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
Low
2,036
2,032
2,026
2,021
2,015
2,009
2,002
1,996
1,990
1,984
1,978
1,971
1,964
1,958
1,951
1,944
1,937
1,931
1,924
1,917
1,910
1,903
1,896
1,888
1,881
1,873
Mid
2,048
2,055
2,063
2,070
2,077
2,083
2,090
2,097
2,103
2,110
2,117
2,123
2,129
2,135
2,140
2,146
2,152
2,157
2,163
2,169
2,175
2,180
2,186
2,191
2,197
2,202
High
2,059
2,080
2,100
2,120
2,140
2,161
2,181
2,201
2,221
2,241
2,261
2,280
2,299
2,318
2,337
2,355
2,374
2,392
2,411
2,430
2,448
2,467
2,485
2,503
2,522
2,540
E-2
-------
Table E-l (continued)
Year
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
Low
1,866
1,864
1,862
1,860
1,857
1,855
1,852
1,849
1,846
1,843
1,840
1,837
1,834
1,832
1,829
1,826
1,823
1,820
1,818
1,815
1,813
1,810
1,807
1,805
1,802
1,800
1,797
1,794
1,791
Mid
2,207
2,208
2,210
2,211
2,212
2,213
2,213
2,213
2,214
2,214
2,214
2,214
2,214
2,214
2,214
2,215
2,215
2,215
2,216
2,216
2,217
2,217
2,218
2,218
2,219
2,219
2,219
2,219
2,219
High
2,558
2,563
2,567
2,572
2,576
2,580
2,584
2,588
2,591
2,594
2,597
2,601
2,604
2,607
2,610
2,613
2,617
2,620
2,624
2,627
2,631
2,634
2,637
2,641
2,644
2,647
2,650
2,652
2,655
E-3
-------
Table E-l (continued)
Year
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
Low
1,789
1,786
1,783
1,780
1,776
1,773
1,770
1,767
1,764
1,760
1,757
1,754
1,751
1,747
1,744
1,741
1,738
1,734
1,731
1,728
1,725
1,721
1,718
1,715
1,711
1,708
1,705
1,701
1,698
Mid
2,219
2,219
2,219
2,218
2,218
2,217
2,217
2,216
2,216
2,215
2,215
2,214
2,214
2,213
2,213
2,212
2,212
2,212
2,211
2,211
2,210
2,209
2,209
2,208
2,207
2,206
2,206
2,205
2,204
High
2,658
2,660
2,662
2,665
2,667
2,669
2,671
2,673
2,675
2,677
2,679
2,682
2,684
2,686
2,688
2,690
2,692
2,695
2,697
2,699
2,701
2,703
2,705
2,706
2,708
2,710
2,711
2,713
2,714
E-4
-------
Table E-l (continued)
Year
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
Low
1,694
1,690
1,687
1,683
1,680
1,676
1,672
1,669
1,665
1,661
1,658
1,654
1,651
1,647
1,643
1,640
1,636
1,632
Mid
2,203
2,202
2,201
2,199
2,198
2,197
2,196
2,195
2,194
2,193
2,191
2,190
2,189
2,188
2,187
2,185
2,184
2,183
High
2,716
2,717
2,719
2,720
2,721
2,723
2,724
2,725
2,726
2,728
2,729
2,730
2,731
2,733
2,734
2,735
2,736
2,737
E-5
-------
Table E-2. Mortality rates (lifespan-equivalent)
Year
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2015
2020
2025
2030
2035
2040
2045
2050
2055
2060
2065
2070
2075
2080
2085
2090
2095
2100
Low
79.74
80.01
80.16
80.32
80.47
80.62
80.78
80.94
81.09
81.24
81.39
81.66
82.41
83.26
83.88
84.60
85.20
85.90
86.51
87.21
87.80
88.47
89.05
89.70
90.24
90.86
91.39
92.00
92.51
93.00
Mid
79.74
79.9
80.05
80.2
80.36
80.51
80.67
80.82
80.97
81.13
81.28
81.43
82.19
82.93
83.56
84.17
84.78
85.4
86.01
86.62
87.22
87.81
88.4
88.97
89.53
90.09
90.64
91.18
91.71
92.24
High
79.68
79.95
80.10
80.25
80.40
80.54
80.69
80.85
80.99
81.14
81.29
81.54
82.26
83.07
83.65
84.33
84.88
85.53
86.08
86.71
87.22
87.82
88.30
88.85
89.28
89.80
90.21
90.69
91.06
91.42
E-6
-------
Table E-3. Projected international migration
Year
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
Low
670,138
593,255
531,919
462,200
404,294
355,173
312,734
275,252
241,908
211,938
184,668
179,040
174,268
170,255
166,833
163,809
161,032
158,577
156,347
154,457
152,592
166,085
179,067
191,267
202,980
214,147
224,882
235,272
245,307
Mid
1,020,435
1,030,425
1,030,293
995,679
961,750
928,453
895,833
863,540
831,875
800,663
769,797
774,125
778,360
782,327
786,288
790,010
793,650
797,129
800,507
803,815
806,979
840,341
873,156
905,230
936,857
967,891
998,465
1,028,622
1,058,406
High
1,432,695
1,570,973
1,671,881
1,704,589
1,722,833
1,729,993
1,728,678
1,720,305
1,706,158
1,687,319
1,664,477
1,699,910
1,733,468
1,765,441
1,795,997
1,825,332
1,853,500
1,880,682
1,907,047
1,932,527
1,957,368
2,041,510
2,125,426
2,208,839
2,292,232
2,375,342
2,458,340
2,541,223
2,624,090
E-7
-------
Table E-3 (continued)
Year
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
Low
255,086
264,891
257,533
250,901
244,888
239,382
234,661
230,311
226,341
222,757
219,520
216,643
213,996
211,696
209,507
207,356
205,448
203,620
201,922
200,347
198,712
197,203
195,780
194,554
193,265
192,017
190,931
189,843
188,619
Mid
1,087,913
1,117,043
1,110,301
1,104,092
1,098,310
1,092,915
1,088,005
1,083,384
1,079,014
1,074,903
1,071,081
1,067,573
1,064,300
1,061,065
1,058,054
1,055,215
1,052,456
1,049,922
1,047,331
1,044,964
1,042,694
1,040,387
1,038,197
1,036,208
1,034,270
1,032,477
1,030,777
1,029,070
1,027,388
High
2,706,829
2,789,455
2,797,150
2,804,858
2,812,504
2,820,177
2,827,787
2,835,422
2,842,719
2,850,066
2,857,333
2,864,468
2,871,496
2,878,370
2,885,342
2,892,083
2,898,727
2,905,317
2,911,896
2,918,282
2,924,605
2,930,848
2,936,971
2,942,987
2,949,104
2,955,004
2,960,927
2,966,785
2,972,462
Eo
-O
-------
Table E-3 (continued)
Year
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
Low
187,498
186,448
185,302
184,102
183,076
181,927
180,848
179,608
178,444
177,282
176,119
174,912
173,632
172,478
171,308
170,122
169,012
167,811
166,788
165,628
164,501
163,329
162,359
161,216
160,066
159,026
158,066
156,995
155,981
Mid
1,025,691
1,024,149
1,022,613
1,021,041
1,019,540
1,018,071
1,016,574
1,015,188
1,013,731
1,012,362
1,011,015
1,009,672
1,008,378
1,007,139
1,005,888
1,004,724
1,003,494
1,002,407
1,001,342
1,000,147
999,100
998,085
997,098
996,136
995,099
994,178
993,261
992,333
991,460
High
2,978,058
2,983,567
2,989,148
2,994,495
2,999,780
3,004,997
3,010,217
3,015,235
3,020,298
3,025,208
3,030,105
3,034,978
3,039,655
3,044,480
3,049,252
3,053,972
3,058,708
3,063,253
3,067,923
3,072,497
3,076,939
3,081,488
3,085,910
3,090,373
3,094,759
3,099,047
3,103,452
3,107,792
3,112,064
E-9
-------
Table E-3 (continued)
Year
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
Low
154,984
154,078
153,015
152,160
151,203
150,290
149,360
148,521
147,543
146,699
145,840
145,004
144,195
143,407
Mid
990,630
989,894
989,122
988,353
987,576
986,934
986,237
985,542
984,815
984,223
983,650
983,071
982,546
982,038
High
3,116,245
3,120,601
3,124,829
3,129,030
3,133,229
3,137,378
3,141,620
3,145,634
3,149,846
3,153,959
3,158,109
3,162,141
3,166,233
3,170,286
E-10
-------
APPENDIX F
SUMMARY OF MAJOR MODEL INPUTS AND ASSUMPTIONS
F-l
-------
Table F-l. Major demographic model inputs and assumptions
Model input
Source
Major assumptions
Initial 2005 county
population
Bridged-Race Vintage 2006
dataset for 7/1/2005*
Population over age 85 in
each county distributed by
national over-85 population
Mortality rates
Fertility rates
Net international
migration
Component Assumptions of
the Resident Population by
Age, Sex, Race, and Hispanic
Origin (U.S. Census Bureau,
2000)
Birth ratio of 1046 males to
1000 females (Matthews and
Hamilton, 2005)
Distribution of
international
immigration to counties
2000 U.S. Census, SF3, Table
P22 (U.S. Census Bureau,
2007)
Distribution remains
consistent with 2000 pattern.
1995-2000 Domestic
migration
Public Use Microdata
Samples (U.S. Census
Bureau, 2003), organized by
the NY State Data Center
Migration patterns from this
time period are used to drive
future migration
Natural amenities
(January temperature,
January sunlight, July
temperature, July
humidity, water area)
USDA Natural Amenities
Database (McGranahan,
1999)
Amenities calculated for
1941-1970 stay constant
over time
County growth rate
(1980-2000)
2000 U.S. Census (SF1,
Table PI2), 1980 U.S. Census
(U.S. Census Bureau, 1992)
The 1980-2000 growth rate
remains a constant proxy for
economic growth
Functional
county-to-county
distance
Based on
population-weighted center of
each county (as of 2000
Census) and transportation
infrastructure of in the 2006
National Transportation Atlas
Function distance remains
static in the model, despite
changes in the geographic
distribution of the population
and anticipated changes in
infrastructure
* NCHS (National Center for Health Statistics). (2007) Bridged-race Vintage 2006 postcensal population
estimates for July 1, 2000 - July 1, 2006, by year, county, single-year of age, bridged-race, Hispanic
origin, and sex. Released August 16, 2007. Centers for Disease Control and Prevention (CDC), Atlanta,
GA. Available online at http://www.cdc.gov/nchs/about/major/dvs/popbridge/datadoc.htm#vintage2006.
F-2
-------
Table F-2. Major SERGoM inputs and assumptions
Model input
Source
Major assumptions
Housing units
2000 U.S. Census (U.S.
Census Bureau, 2007)
Future growth patterns are
likely to be similar to those
that occurred historically
1980-2000
Undevelopable lands
UPPT (unprotected, private
protected, public protected,
and tribal/native lands)
dataset generated with data
from the Conservation
Biology Institute's PAD v4
database (CBI, 2008).
Updated with current data for
21 states
Development cannot occur
on these lands
Road and groundwater
well density
Fine-grained land use/cover
data (NLCD 2001)
Commercial/Industrial
land use
Changes in transportation
infrastructure will follow the
footprint of existing
infrastructure
County population
Developed using
demographic model
See Table F-l
Cross-cutting assumptions:
• Drivers are hierarchical (national, state, regional, county, w/in county)
• New urban areas can be emergent (like cellular automata)
F-3
-------
Recycled/Recyclable
Printed with vegetable-based ink on paper that contains a minimum
of 50% post-consumer fiber content and processed chlorine free.
&EPA
United States
Environmental Protection
Agency
PRESORTED STANDARD
POSTAGE & FEES PAID
EPA
PERMIT NO. G-35
Of Ice of Research and Development
National Center for Environmental Assessment
Washington, DC 20460
Of;_cial Business
Penalty for Private Use
$300
------- |