trA EPA/600/R-16/366F | January 2017 | www.epa.gov/research
United States
Environmental Protection
Agency
Updates to the Demographic and Spatial
Allocation Models to Produce Integrated
Climate and Land Use Scenarios (ICLUS)
Version 2
Office of Research and Development
Washington, D.C.
-------
vvEPA
EPA/600/R-16/366F
January 2017
www.epa.gov/research
FINAL
Updates to the Demographic and Spatial Allocation
Models to Produce Integrated Climate and Land Use
Scenarios (ICLUS) Version 2
Office of Research and Development
U.S. Environmental Protection Agency
Washington, DC
-------
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
The first version of the Integrated Climate and Land Use Scenarios (ICLUS) project
modeled population, residential development, and impervious surface cover changes by decade
to the year 2100 consistent with four Intergovernmental Panel on Climate Change (IPCC)
emissions scenarios and a baseline. This report discusses improvements to the underlying
demographic and spatial allocation models of the ICLUS that result in version 2 (v2) consistent
with two of the new Shared Socioeconomic Pathways (SSPs) and two Representative
Concentration Pathways (RCPs). Improvements include the use of updated data sets, integration
of changing climate variables within the migration model, inclusion of transportation network
capacity and its increase over time, growth in commercial and industrial land uses, and the use of
population density-driven demands for residential housing, commercial development, and
industry. This report demonstrates the effect of these improvements by comparing national and
regional results among the SSP and RCP combinations and the two climate models selected.
ICLUS v2 shows differences in population migration patterns by including climate variables that
change over time rather than ones that are static. Additionally, changing commercial and
industrial land uses can drive patterns of new urban growth with consequences for many
environmental endpoints.
Preferred citation:
U.S. EPA (Environmental Protection Agency). (2016) Updates to the Demographic and Spatial Allocation Models
to Produce Integrated Climate and Land Use Scenarios (ICLUS) Version 2. National Center for Environmental
Assessment, Washington, DC; EPA/600/R-16/366F. Available from the 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 viii
LIST OF TECHNICAL TERMS ix
PREFACE xi
AUTHORS AND REVIEWERS xii
EXECUTIVE SUMMARY xiii
1. INTRODUCTION 1
2. UPDATES TO THE MIGRATION MODEL 3
2.1. UPDATING THE MIGRATION MODEL 4
2.1.1. Parameterizing Domestic Migration 5
2.1.2. Functional Connectivity 7
2.1.3. Historic Climate Amenities 7
2.1.4. Climate Change Model Selection 9
2.1.5. Redesign and Recalibration of the Migration Model 11
2.2. MIGRATION MODEL INTERPRETATION 14
3. UPDATES TO THE SPATIAL ALLOCATION MODEL 16
3.1. OVERVIEW OF THE UPDATED SPATIAL ALLOCATION MODEL 16
3.2. Creating the Initial Accessibility-Capacity Surface 18
3.3. ICLUS V2 LAND USE CLASSES 20
3.3.1. Quantifying Land Use Changes, 2000-2010 21
3.4. TRANSITION-PROBABILITY MODEL 24
3.4.1. Empirical Estimation of Transition Probabilities 27
3.5. LAND USE AND CAPACITY DEMAND MODELS 29
3.5.1. Updating the Accessibility-Capacity Surface 32
3.6. LAND USE PATCH ALLOCATION PROCESS 33
4. RESULTS 34
4.1. POPULATION PROJECTIONS 34
4.1.1. National Projections 34
4.1.2. Regional Projections 35
4.1.3. The Effect of Changing Climate Amenities 36
4.1.4. Subregional Projections 40
4.2. LAND USE PROJECTIONS 46
4.2.1. National Projections 46
4.2.2. Regional Projections 48
4.2.3. Subregional Projections 50
5. CONCLUSION 57
in
-------
CONTENTS (CONTINUED)
6. REFERENCES 61
APPENDIX A. REGIONAL LAND-USE CHANGES FOR 2000-2010 A-l
APPENDIX B. REGIONAL TRANSITION PROBABILITY MODELS B-l
APPENDIX C. LAND USE CLASS (LUC) AND CAPACITY DEMAND MODELS C-l
IV
-------
LIST OF TABLES
1. Migration model results 15
2. Land Use Classes used in the ICLUS v2 model 21
3. Goodness-of-fit test results comparing Land Use Classes in 2000 and 2010, nationally. 23
4. Land Use Classes transitions from 2000 (rows) to 2010 (columns) incorporated into
ICLUS v2 26
5. GLS model results 45
6. Cumulative change in developed LUCs for 2050 (top row) and 2100 (bottom row) by
Shared Socioeconomic Pathways (SSPs), Representative Concentration Pathways (RCPs)
and climate model 50
v
-------
LIST OF FIGURES
1. Comparison of ICLUS vl and ICLUS v2 2
2. ICLUS v2 spatial allocation flow diagram 3
3. ICLUS v2 geographic units include metropolitan statistical areas (MSAs), micropolitan
statistical areas, and stand-alone counties 5
4. Biplots used to select climate projections used in this report. Dashed lines show median
values 11
5. Proportion of total migration between MS As, micropolitan statistical areas, and rural
counties 12
6. Regions used in ICLUS v2. Region 1-West Coast; Region 2-Intermountain West;
Region 3-North Central; Region 4-South Central; Region 5-Great Lakes; Region 6—
Southeast; Region 7-Northeast 17
7. Land use comparisons between 2000 and 2010, nationally 24
8. Predicted transition probability by capacity class into LUCs in 2010 28
9. Predicted log transformed pixel or capacity density (km-2) (± 2 SE) by log transformed
population density (km-2) 31
10. Total population for the conterminous United States to 2100 showing projections for
ICLUS v2 35
11. Total population for each ICLUS region to 2100 under four scenario assumptions 36
12. The effect of climate change-induced domestic migration expressed as differences in
millions of people 38
13. The effect of climate change-induced domestic migration expressed as percentage
differences 39
14. Differences in population and climate change projections driven by FIO-ESM and
HadGEM2-AO under SSP5 and RCP8.5 assumptions for 2050 and 2100 41
15. Differences in population and climate projections driven by FIO-ESM and HadGEM2-
AO under SSP1 and RCP4.5 assumptions for 2050 and 2100 42
16. Average ICLUS GU 10-year population change by (A) starting population density and
ICLUS region from 2010-2050, (B) starting population density and ICLUS region from
2060-2100, (C) starting population density and SSP from 2060-2100. PI: <5.0; P2:
5.1-15.0; P3: 15.1-45.0; P4: 45.1-135.0; P5: >135.1 people per km2 44
vi
-------
LIST OF FIGURES
17. National land use projections from ICLUS v2 to 2100 46
18. Relative increases in the area of developed LUCs nationally at 2050 (top row) and 2100
(bottom row) 48
19. Land use change in the vicinity of the Portland, OR-Vancouver, WA Metro Area under
the SSP1-RCP4.5 (FIO-ESM) scenario: 2010, 2050, and 2100 52
20. Land use change in the vicinity of the Portland, OR-Vancouver, WA Metro Area under
the SSP5-RCP8.5 (HadGEM2-AO) scenario: 2010, 2050, and 2100 53
21. Land use change in the vicinity of the Springfield, MO Metro Area under the
SSP1-RCP4.5 (FIO-ESM) scenario: 2010, 2050, and 2100 54
22. Land use change in the vicinity of the Springfield, MO Metro Area under the
SSP5-RCP8.5 (HadGEM2-AO) scenario: 2010, 2050, and 2100 55
23. Land use change in the vicinity of the Washington-Arlington-Alexandria, DC-VA Metro
Area under the SSP1-RCP4.5 (FIO-ESM) scenario: 2010, 2050, and 2100 56
24. Land use change in the vicinity of the Washington-Arlington-Alexandria, DC-VA Metro
Area under the SSP5-RCP8.5 (HadGEM2-AO) scenario: 2010, 2050, and 2100 57
vii
-------
LIST OF ABBREVIATIONS AND ACRONYMS
BCSD
bias-correcting and spatial-downscaling
CI
confidence interval
CMIP
Coupled Model Intercomparison Project
df
degrees of freedom
DUA
dwelling units per acre
EPA
U.S. Environmental Protection Agency
FASOM
Forestry and Agricultural Sector Optimization Model
FIO-ESM
First Institute of Oceanography-Earth System Model
FIPS
Federal Information Procession Standard
FORESCE
FOREcasting SCEnarios
GAM
generalized additive model
GCM
general circulation model
GHG
greenhouse gases
GU
geographic unit
HadGEM2-AO
Hadley Global Environment Model 2 Atmosphere-Ocean
ICLUS
Integrated Climate and Land Use Scenarios
IIASA
International Institute for Applied Systems Analysis
IPCC
Intergovernmental Panel on Climate Change
IRS
Internal Revenue Service
LUC
land use class
MIGPUMA
Migration Public-Use Microdata Area
MSA
metropolitan statistical area
NLCD
National Land Cover Database
US-NLUD
National Land Use Dataset
OMB
Office of Management and Budget
OR
odds ratio
ORD
Office of Research and Development
P
population density, pixel
PUMA
Public Use Microdata Area
PUMS
Public Use Microdata Sample
RCP
Representative Concentration Pathway
SERGoM
Spatially Explicit Regional Growth Model
SRES
Special Report on Emissions Scenarios
SSP
Shared Socioeconomic Pathway
TIGER
Topologically Integrated Geographic Encoding and Referencing
USGS
U.S. Geological Survey
vl
version 1
v2
version 2
WCRP
World Climate Research Programme
Vlll
-------
LIST OF TECHNICAL TERMS
Climate Amenities: Climate variables in association with their perceived value and putative
influence on migration decisions. For example, the climate variables selected to represent
climate amenities in ICLUS v2 are average monthly humidity-adjusted temperature and average
seasonal precipitation for both summer and winter.
County-to-County Migration: The permanent or semipermanent change of residence from one
county to another. In ICLUS v2, many county-to-county migration flows are modeled in
aggregate based on U.S. Census statistical area definitions.
Emission Scenario: A storyline providing assumptions on future releases of greenhouse gases
(GHGs) and other aerosols or atmospherically active substances into the atmosphere based on
postulated economic patterns, used as input into a climate model and a precursor to Shared
Socioeconomic Pathways (SSPs).
End point: An outcome used to assess risk to ecosystems and environmental health.
GU (Geographic Unit): An important county-based spatial unit of analysis used in ICLUS v2.
Includes aggregations of counties (i.e., metropolitan statistical areas and micropolitan statistical
areas) defined by the U.S. Census Bureau. Many rural or less-populated counties are not
included in either of those statistical area types, and remain standalone units.
Land Use: The human-induced activities on a unit of land.
Land Use Change: A transition to a different human-induced activity. May include changes in
management or environmental condition.
Land Use Class (LUC): The primary land use at a given location. In ICLUS v2, land use is
defined and tracked at a spatial resolution of 90 m x 90 m.
Observed Climate: Climate variables measured and recorded by instruments. Includes
derivative or composite climate products.
IX
-------
LIST OF TECHNICAL TERMS (CONTINUED)
Projected Climate: Future climate variables produced by computer simulation models. These
simulations are based on a set of assumptions (e.g., future global emissions of greenhouse gases)
and do not perfectly replicate the observed climate.
Patch: A contiguous area composed of one or more pixels of the same land-use class. In ICLUS
v2, land use change is modeled by placing new patches on the existing landscape.
Pixel: A single nonoverlapping member of a uniform grid. In ICLUS v2, the conterminous
United States is represented as a grid where each pixel measures 90 m x 90 m.
Region: A geographic area defined by similar climatic, demographic, and land use patterns.
Representative Concentration Pathways (RCPs): Used by climate modelers to standardize
experiments, a set of four forcing pathways or trajectories for GHG concentrations based on
underlying socioeconomic assumptions.
Shared Socioeconomic Pathways (SSPs): Narratives that qualitatively describe future changes
in demographics, human development, economy and lifestyle, policies and institutions,
technology, and the environment.
Spatial scale: The geographic area at which a response associated with a process or pattern is
examined. For example, in ICLUS v2, climate variables are resolved at the scale of GU to
examine the influence of climate on human migration.
Static Climate Variables: Climate variables that remain constant throughout the modeling
period. ICLUS vl used only static climate variables, taken from observed climate records.
Dynamic Climate Variables: Climate variable that change overtime. ICLUS v2 uses dynamic
climate variables taken from projected future climate conditions.
x
-------
PREFACE
This report was prepared jointly by the Office of Research and Development (ORD) at
the U.S. Environmental Protection Agency (EPA), ICF International, Colorado State University,
and Conservation Science Partners. The report describes the updates to data sets and models that
constitute ICLUS version 2 (v2). Because this is an update to ICLUS version 1 (vl), many of the
concepts and models build on the original report (U.S. EPA, 2009). Users familiar with ICLUS
vl can use this report as a reference guide to understand what changes have been made and the
implications for the resulting data sets and maps. Output data sets and maps are intended to be
used in a scenario context to assess the risks, vulnerabilities, impacts, and adaptation options of
climate change.
XI
-------
AUTHORS AND REVIEWERS
The Global Change Assessment Staff within the National Center for Environmental
Assessment, Office of Research and Development, is responsible for publishing this report. This
document was prepared by ICF International under Contract No. GS-10F-0124J, EPA Order No.
EP09C000208. Dr. Britta Bierwagen served as Technical Project Officer and provided overall
direction and technical assistance.
AUTHORS
U.S. EPA
Britta Bierwagen, Philip Morefield
Oak Ridge Institute for Science and Engineering (ORISE) Fellow at U.S. EPA
Jonathan Witt1
ICF International, Washington, DC
Anne Choate, Jonathan Cohen, Philip Groth, Dana Spindler
Conservation Science Partners; Department of Fish, Wildlife and Conservation Biology,
Colorado State University, Fort Collins, CO
David M. Theobald
REVIEWERS
U.S. EPA Reviewers
William Kepner (ORD/NERL), James Wickham (ORD/NERL)
External Peer Reviewers
Simon Choi (Southern California Assoc. of Governments), Elena Irwin (Ohio State Univ.),
David Plane (Univ. of Arizona), Timothy Randhir (Univ. of Massachusetts, Amherst)
ACKNOWLEDGMENTS
The authors would like to thank Alicia Barnash, Angelica Murdukhayeva, Jennifer
James, and David Gibbs for their assistance visualizing data sets and outputs and addressing peer
review comments. The thoughtful comments from reviewers substantially improved this report.
1 Current affiliation: Fairfax County Stormwater Planning Division, VA.
xii
-------
EXECUTIVE SUMMARY
The Integrated Climate and Land Use Scenarios (ICLUS) version 1 (vl) furthered land
change modeling by providing nationwide housing development scenarios to 2100. ICLUS
version 2 (v2) builds on this modeling approach by updating population and land use data sets
and addressing two sets of limitations identified in ICLUS vl. This report documents the
changes made to the underlying data sets used for model parameterization and to the
demographic and spatial allocation models. The purpose is to address limitations encountered in
ICLUS vl and identified, in part, by the ICLUS user community.
The first set of limitations is within the ICLUS vl migration component of the
demographic model, which incorporated only five years of human migration data, only
road-based connections among counties, and a static climate variable. To address these
limitations, ICLUS v2 uses a data set from the Internal Revenue Service (IRS) of
county-to-county migration from 1991-2000 to parameterize the migration model. Intercounty
connectivity calculations include fixed mass transit as well as roads.
Another update to the migration model is the inclusion of dynamic climate variables as
part of the amenity parameters. ICLUS vl used static amenity variables, including older
county-level historical climate data. ICLUS v2 now parameterizes the model with updated
historical climate data (1981-1999) and includes projected climate variables for each time step
to 2100. Analyses in this report use two different climate models: (1) the First Institute of
Oceanography-Earth System Model (FIO-ESM) and (2) the Hadley Global Environment Model
2 Atmosphere-Ocean (HadGEM2-AO) to illustrate the effect of dynamic climate variables on
migration patterns. Specific climate variables include January and July humidity-adjusted
temperature and summer (June, July, August) and winter (December, January, February)
precipitation incorporated as the running average of the previous 10 years of climate model
output. Comparisons of the results with and without projected climate variables show that
differences in regional migration patterns occur when dynamic climate variables are included.
Differences in the distribution of population between model runs using climate variables and
results of model runs using static climate variables are more similar to each other than are
differences in migration patterns between combinations emissions and demographic scenarios.
Different fertility and migration rates in the scenarios exert much larger influences on the overall
migration patterns than changes in climate amenities.
Several additional changes in the ICLUS v2 demographic model resulted from updates of
data sets. The use of the 2010 U.S. Census Bureau's data in the demographic model results in
new national population projections for each of the scenarios documented in this report. Because
the IRS database does not contain demographic information, the migration model in ICLUS v2
xiii
-------
combines all age groups into a single population, whereas ICLUS vl recorded separate migration
information for populations under and over 50 years old.
The second set of limitations identified was within the ICLUS vl spatial allocation
model, which used population to calculate housing density based on household size, while all
other land use classes remained static. ICLUS v2 uses statistical relationships between
population density, road capacity, and land use classes to allocate new land uses at the next time
step based on the demands of the growing population. Demand calculations are done nationally
for each developed land use class and transition probabilities from one land use to another
incorporate differences in growth patterns for each of seven regions of the conterminous United
States, similar to U.S. Census Bureau divisions. In addition to residential housing classes,
commercial and industrial land uses also change at each time step.
The spatial allocation model also incorporates updated data sets for land use (a new U.S.
National Land Use Dataset [US-NLUD] based on the 2011 National Land Cover Database
[NLCD] and many other detailed data on land use), transportation (roads and fixed mass transit),
and developable area derived from the 2012 U.S. Geological Survey (USGS) Protected Areas
Database. The model uses land use transitions from 2000 to 2010 as the basis for all future land
use changes. The spatial allocation model projects transitions for five residential housing classes
and commercial and industrial land uses. The sequence of land use class changes is based on the
theory that the highest and best use prevails, generally as determined by land value. The spatial
allocation model uses output from the demographic model to calculate demand for each land use
class in relation to population density. New land uses are allocated as patches that reflect a
region-specific distribution of sizes and shapes. Patch placement is determined by the antecedent
land use class and accessibility, and placement of residential patches also takes into account
distance to commercial areas.
The resulting land use allocation replaces low-density residential development by higher
density land uses as a population grows within ICLUS geographic units. Low-density
development generally expands outward. The development of higher density residential,
commercial, and industrial classes levels off in terms of demand at high population densities,
exhibiting a threshold effect. This threshold shows that these land use classes are not rapidly
replaced once developed, and that there are observed limitations in the density of particular land
use classes in dense metropolitan areas. Similarly, transportation capacity also reaches a
threshold. Dense cities add new road capacity more slowly than do smaller cities.
The emergence of new socioeconomic and emissions scenarios (e.g., SSPs and RCPs)
utilized in ICLUS v2, as opposed to the previous emissions storylines used in ICLUS vl, limits
the usefulness of direct comparisons between outputs from both ICLUS versions. Instead, this
report compares results from the SSP-RCP combinations implemented in ICLUS v2.
xiv
-------
Improvements in ICLUS v2 allow discussions of results in terms of national changes, as well as
regional and subregional changes over time.
The output of the demographic model is similar to globally based population estimates
for the United States that are consistent with the SSPs. ICLUS v2 population estimates for the
United States by 2100 are slightly higher for both SSP1 and SSP5 than those derived by KC and
Lutz (2014) for SSP1 and SSP5 from the global estimates for the United States, although the
relative difference in population between the scenarios in 2100 is similar. The use of two
population estimates, one higher and one lower, allows for an interpretation of differences in
impacts between the two scenarios. By using population estimates that are consistent with the
SSPs, the resulting impacts can be put into a context consistent with other efforts using those
socioeconomic storylines. Regionally, differences in population growth are greater between the
SSPs than differences between climate models used with the same SSP. However, comparisons
between model runs with dynamic climate variables and static climate show regional differences
in population of up to 4%. Subregionally, there are additional differences that are reinforced by
the choice of climate model used in the migration model. These differences are more distinct at
higher population densities and during the last half of the century, especially when using SSP5,
which has higher fertility rates than SSP1.
The national-scale land use projections show nearly identical trends when comparing
outcomes under the same SSP assumption; the choice of climate model has no discernible effect
on the overall amount of projected development. However, there are differences in amount and
allocation of land uses when comparing between SSPs and examining changes regionally.
Regional allocation patterns reflect existing differences across the conterminous United States
that continue to shape patterns into the future. While nearly all developed land use classes
increase in nearly all regions, the magnitude of changes reflects current trends, such that
low-density residential classes continue to increase in the Intermountain West more so than in
other regions, and regions with higher densities continue to increase their urban land uses.
Overall, ICLUS v2 provides users with the ability to model population and land use
changes consistent with SSP and RCP scenarios and specific climate models to improve
integrated climate and land use assessments. While this report only uses SSP1-RCP4.5 (lower
population growth, lower emissions) and SSP5-RCP8.5 (higher population growth, higher
emissions) scenarios in conjunction with two climate change models, FIO-ESM and
HadGEM2-AO, to illustrate ICLUS v2 improvements, the model structure allows users the
flexibility to select any SSP, RCP, and climate model combination. The use of statistically based
transition and demand models also allows users to change parameters for further scenario
explorations that alter development pathways from current trajectories. Improvements in ICLUS
v2 facilitate the analysis of scenarios of climate change impacts, vulnerability, and adaptation
xv
-------
options, including the use of ICLUS v2 outputs in models projecting emissions from developed
land uses to determine consequences for water and air quality endpoints, as well as human
health.
xvi
-------
1. INTRODUCTION
Changes in climate and land use are global drivers of environmental impacts. The
interactions between climate and land use changes are complex and can result in challenges for
ecosystems and environmental health. The motivation for the U.S. Environmental Protection
Agency (EPA) Integrated Climate and Land Use Scenarios (ICLUS) project originated with the
recognition of this complex relationship and the absence of an internally consistent set of land
use scenarios that support national assessments of climate change effects. This report describes
updates to the ICLUS model data, methods, and outputs described in Land-Use Scenarios:
National-Scale Housing-Density Scenarios Consistent with Climate Change Storylines
(U.S. EPA, 2009).2 The goal of the current report is to describe the changes between the ICLUS
version 1 (vl) data sets and modeling approach and ICLUS version 2 (v2) that are intended to
improve on the demographic and spatial model outputs.
ICLUS vl developed future scenarios of population, housing density, and impervious
surfaces that were consistent with the Intergovernmental Panel on Climate Change (IPCC)
Special Report on Emissions Scenarios (SRES) storylines (Nakicenovic and Swart, 2000).
ICLUS vl integrated two main components: a demographic model and a spatial allocation model
(see Figure 1). ICLUS vl helped advance land change modeling by providing nationwide
development scenarios to 2100. ICLUS v2 builds on this modeling approach by addressing two
sets of limitations. First, the demographic model in vl incorporated a limited timeframe of
movement data, intercounty connectivity solely based on roads, and a static climate variable in
its migration model. Second, the spatial allocation model used population to calculate housing
density based on household size, while all other land use types remained static, including
commercial and industrial uses. In addition to addressing these limitations, ICLUS v2
incorporates updated data sets of population, land use, and land cover. The addition of dynamic
future climate variables draws on the most recent climate data, which use Representative
Concentration Pathways (RCPs) rather than SRES storylines (van Vuuren et al. 2011). The
RCPs are targets of greenhouse gas concentrations that general circulation models reach by the
year 2100 to depict a range of climate change outcomes. Thus, ICLUS v2 is now consistent with
the most recent suite of climate change scenarios, linking RCP-driven climate model output with
Shared Socioeconomic Pathways (SSPs; O'Neill et al. 2014). Though the effect of dynamic
future climate variables on migration is small, the cumulative changes yield different settlement
patterns that enable scenario-based analyses of impacts and vulnerabilities of environmental
endpoints.
2 Download the ICLUS version 1 report: https://cfpub.epa.gov/ncea/global/recordisplay.cfm7deicH203458.
1
-------
County Population:
2000
SRES Storylines
Impervious Surface
Migration
County Population
Fertility, Immigration
Housing Density
County Population:
2010
SSP Storylines
HCP Climate
Fertility, Immigration
Land Use
Migration
MSA & County
Population
Figure 1. Comparison of ICLUS vl and ICLUS v2. The two model versions
are conceptually very similar. ICLUS v2 reflects substantial updates to key inputs
as well as a modified geographic framework. Estimates of percentage impervious
surface change for ICLUS v2 will be produced subsequent to this report.
This report covers the updates to the demographic model in Section 2 and the spatial
allocation model in Section 3. Figure 2 provides an orientation of the flow of data and processes
within ICLUS v2. The output of the demographic model, the population data, is one input to the
data sets further described in Section 3. Section 4 focuses on model outputs, both demographic
and land use, and compares these outputs among the scenarios implemented. Descriptions of the
updates and analyses of v2 outputs are intended to assist users of the ICLU S data sets and maps
to understand which changes were made, why, and what the consequences for the outputs are.
2
-------
Repeat for each time step
Capacity Update
Function
Section 3.5.1 \
Land use
Capacity Surface
Section 3.2
Demand for Land
Uses
Section 3.5
Commercial Pixels t
Section 3.5.1
Population,
Section 2
Repeat for each land use class
land Use Transition
Probability Surface
Section 3.4.1
Figure 2. ICLUS v2 spatial allocation flow diagram. Land use change is
modeled by allocating new patches (comprised of 90 m x 90 m pixels) until the
demand for new pixels of each land use is satisfied. The likelihood of an existing
pixel converting to a new land use is a function of both transportation capacity
(i.e., the accessibility of the pixel) and the existing land use. For residential
classes, proximity to commercial pixels is treated as an amenity and will attract
new growth. Transportation capacity grows in relation to population density
(green boxes). Likewise, the demand for new pixels is driven by population
growth (blue boxes).
2. UPDATES TO THE MIGRATION MODEL
The ICLUS demographic model consists of a cohort-component model and a migration
model that project county-level population for the conterminous United States on an annual basis
from 2010 to 2100 for a number of socioeconomic scenarios and climate projections. The
cohort-component methodology projects fertility, mortality, and international migration. The
model also includes a submodel to project county-to-county domestic migration influence by
amenity variables such as climate (U.S. EPA, 2009).
3
-------
The baseline population and demographic characteristics in ICLUS v2 use the most
recent 2010 U.S. Census Bureau data (NCHS, 2011) but the same fertility and migration rates as
ICLUS vl. The combinations of demographic components of change (i.e., fertility, mortality,
migration) was revisited for ICLUS v2 to be consistent with published descriptions of RCPs and
SSPs (Samir and Lutz, 2014; van Vuuren and Carter, 2014). Model updates discussed in this
report use combinations of RCPs and SSPs that succinctly demonstrate a range of possible
ICLUS v2 projections with respect to climatic changes and population growth. We used a
peer-reviewed crosswalk of the SSPs and RCPs to the SRES scenario framework to identify
combinations of SSPs and RCPs (van Vuuren and Carter, 2014) that resemble the bounding
scenarios used to demonstrate the range of impacts explored with ICLUS vl (e.g., Bierwagen et
al., 2010; Voorhees et al., 2011; Georgescu et al., 2014). We selected the combination of SSP5
and RCP8.5 to represent a high emissions, high population-growth scenario, and the combination
of SSP1 and RCP4.5 as a lower emissions, lower population-growth scenario. Like ICLUS vl,
the population-growth scenarios were generated using projections of immigration, fertility, and
mortality produced by the U.S. Census Bureau (2000). Specifically, the SSP5-RCP8.5 scenario
uses the U.S. Census Bureau's high fertility, high domestic migration, and medium immigration
rates; SSP1-RCP4.5 uses medium fertility, high domestic migration, and medium immigration.
These combinations are qualitatively consistent with rates for high-income countries globally
(Samir and Lutz, 2014) and generally correspond to the SRES AlFi (high emissions) and B1
(low emissions) scenarios, respectively (van Vuuren and Carter, 2014).
The focus of the remainder of Section 2 is on implementing the migration model within
the cohort-component model. The following subsections describe changes to the migration
component of the ICLUS vl demographic model, including updates to domestic movements and
the incorporation of climate change projections. Section 4.1 shows the results of the updated
model and compares these to ICLUS vl outputs.
2.1. UPDATING THE MIGRATION MODEL
The demographic component in ICLUS vl included a migration model that simulated
domestic migration by estimating flows between pairs of counties. ICLUS v2 updates the
underlying data used to parameterize the migration model, adds new independent variables,
incorporates a county-to-county migration data set that covers a longer historical time period
than the data set in ICLUS vl, and aggregates some counties into metropolitan and micropolitan
statistical areas, defined as 50,000 people or more in an urban area and at least 10,000 but less
than 50,000 people, respectively. Finally, amenity variables use recent climate data for model
calibration and update these data each decade with model output of future climate variables.
4
-------
2.1.1. Parameterizing Domestic Migration
ICLUS v2 incorporates definitions of both metropolitan and micropolitan statistical areas
(OMB, 2010) and aggregates counties into geographic units accordingly.3 This change
effectively reduces the number of migration origin and destination locations and simplifies
analysis of the historic migration information by excluding many short-distance moves
(i.e., moves within metropolitan or micropolitan statistical areas). In addition, a small number of
independent cities that were not absorbed into metropolitan or micropolitan areas were merged
with an adjacent county. The resulting geographic framework consists of 2,256 units, composed
of metropolitan and micropolitan statistical areas and stand-alone rural counties, referred to
hereafter as ICLUS geographic units (GUs; see Figure 3).
Metropolitan Statistical Area ij^
Micropc
County
Micropolitan Statistical Area
Figure 3. ICLUS v2 geographic units include metropolitan statistical areas
(MSAs), micropolitan statistical areas, and stand-alone counties.
3 Metropolitan and micropolitan statistical areas are delineated by the U.S. Office of Management and Budget
(OMB) and are the result of the application of published standards to Census Bureau data. A metropolitan statistical
area contains a core urban area of population 50,000 or more, and a micropolitan statistical area contains an urban
core of at least 10,000 (but less than 50,000). Metro or micro areas represent larger regions to reflect broad social
and economic interactions (as measured by commuting to work) within the urban core.
5
-------
The ICLUS vl migration model used a temporally limited data set to parameterize
county-to-county movements across the conterminous United States, specifically the 1995 to
2000 Public Use Microdata Samples (PUMSs; U.S. Census Bureau, 2003). Although this data
set includes millions of migration records (n = 2,397,007), it covers just a single 5-year time
span. ICLUS v2 uses 10 years (1991 to 2000) of the IRS (2014) county-to-county annual
migration data to parameterize the migration model4. The values in the migration data set,
combined with specific county-level information, such as population size, growth rates, climate,
and connectivity to other counties, are used to parameterize the updated migration model. The
decade of IRS data chosen to calibrate the migration model captures relatively recent responses
to climate and overlaps with the climate data used in model parameterization described in
Section 2.1.3.
The IRS data set provides a full count of all income tax filers based on year-to-year
changes in or continuity of address reported on individual income tax returns. Data are
expressed in terms of inflows (the number of new residents who moved to a county and where
they originated) and outflows (the number of residents leaving a county and where they went).
The data set covers all counties in the United States but only reports county-to-county migrations
when 10 or more such migrations occurred.
The IRS data present multiple advantages over the PUMS data. First, unlike the PUMS
migration data used in ICLUS vl, the IRS migration data are true county-to-county records. The
PUMS migration data represent migrations between Migration Public-Use Microdata Areas
(MIGPUMAs). This required a two-stage conversion, from MIGPUMA to Public Use Microdata
Areas (PUMAs), and then from PUMAs to counties. Second, the IRS data represent full counts
of all income tax filers, while the PUMS data are based on a statistical sample. Third, and most
importantly, the IRS data used in this analysis are annual data for the years 1991-2000,
compared with a single 5-year period of PUMS data.
However, the IRS data has a different set of limitations not present in the PUMS data.
First, age is not included in the IRS data. The ICLUS vl migration model consisted of two age
groups (ages 0-49 years and ages 50 years and older). ICLUS v2, therefore, does not separate
the model into different age groups. Second, the IRS data are based on the number of income tax
filers and exemptions, not the number of people. The number of exemptions, however, closely
matches the number of people (IRS, 2014). Consequently, people who did not file income tax
returns are excluded from the IRS data, and their migrations would not be captured in ICLUS v2.
Third, in cases where fewer than 10 migrations were recorded between any county pair,
migration flows are aggregated in the IRS data. Flows of fewer than 10 migrants represent about
4 These data are available for public download: http://www.irs.gov/uac/SOI-Tax-Stats-Migration-Data.
6
-------
7% of total migrations but were not included in the analysis due to lack of specific
origin/destination pairing.
From the IRS data set, we extracted two key variables used in this analysis: (1) total
outflow expressed as a percentage of the county population and (2) individual county-to-county
migration records.
2.1.2. Functional Connectivity
ICLUS v2 also includes updated measures of connectivity. Like ICLUS vl,
population-weighted centroids were generated for each of the 2,256 geographic units. Centroids
for a few units were manually moved inside of their respective geographic boundaries. To
evaluate the connectedness of each geographic unit, a network-based travel time was calculated
for every possible origin-destination combination. Travel times were estimated using StreetMap
North America5 and the Network Analyst extension for ArcGIS 10.3. The population-weighted
centroids were snapped to the nearest network feature, including regular ferry routes where
applicable.
2.1.3. Historic Climate Amenities
Linkages between climate variables and human migrations are reported in the literature
(e.g., Alonso, 1971; Cragg and Kahn, 1996; Rappaport, 2007; Feng et al., 2010; Maxwell and
Soule, 2011; Sinha and Cropper, 2013) and form the basis for our exploration of including
changing climate variables in the migration model. The influence of climate on migration
decisions is only one of many possible amenity-based influences and is smaller than other factors
like jobs, housing costs, and family, which are implicitly represented in our migration model.
The explicit inclusion of climate variables allows for the development of land use scenarios that
incorporate climate change model output and are consistent with SSPs and RCPs. ICLUS vl
used a static set of 30-year average climate data based on 1941-1970 records (McGranahan,
1999). ICLUS v2 improves on the inclusion of a climate amenity value in two ways. First, the
historic climate data were updated to cover the 1981-1999 time period, which coincides with the
IRS migration data. Second, future projections of climate change are used to update these
amenity values at each time step of the migration model. Together, these improvements allow
the ICLUS v2 migration model to better reflect the human responses to climatic changes based
on the historical estimate of such responses.
5 http://resources.arcgis.eom/en/help/main/10.l/index.html#/About StreetMap North America/
001 z00000039000000/.
7
-------
In order to incorporate both observed and projected climate amenity values in the
migration model, data covering the observed historical period and future time period need to be
consistent. ICLUS vl used January temperature, January sunlight, July temperature, and July
humidity as the climate amenity variables. However, sunlight variables generally are not
available as output from general circulation models (GCM) used to model climate change and
therefore were not used in ICLUS v2. Furthermore, results from Sussman et al. (2014) suggest
that precipitation is a key climate amenity driving housing prices and should not be omitted in a
migration model. Results from Sussman et al. (2014) informed the ultimate selection of climate
variables to include in ICLUS v2.
Climate variables also need to be resolved at the spatial scale of ICLUS geographic units
(or smaller) for consistency with the migration model. While raw GCM output covers much
larger geographic areas, the use of downscaled products reduces the spatial resolution. Historical
and projected climate data are available for download from the World Climate Research
Programme's (WCRP's) Coupled Model Intercomparison Project Phase 5 (CMIP5) multimodel
data set with bias-correcting and spatial-downscaling (BCSD) methodology applied
(Wood et al., 2004; Maurer et al., 2007).6 The BCSD methodology uses statistical bias
correction to interpret GCM output over a large spatial domain based on current observations.
The principal potential weakness of this approach is an assumption of stationarity (i.e., the
relationship between large-scale precipitation and temperature and local precipitation and
temperature in the future will be the same as in the past). Thus, the method can successfully
account for orographic effects that are observed in current data, but not for impacts that might
result from the interaction of changed wind direction and orographic effects. A second
assumption included in the bias-correction step is that any biases exhibited by a GCM for the
historical period will also be exhibited in simulations of future periods.
The variables selected for use in the migration model were average monthly
humidity-adjusted temperature (January and July) and average seasonal precipitation (December
through February, or "winter," and June through August, or "summer"), although a number of
permutations were tested to maximize model fit. These included:
• Comparing the role of absolute temperature versus changes in temperature relative to the
mean;
• Comparing the role of absolute precipitation versus changes in precipitation relative to
the mean;
6 Bureau of Reclamation/Santa Clara University/Lawrence Livermore archive of downscaled IPCC model runs
available at http://gdo-dcp.ucllnl.org/downscaled cmip projections/.
8
-------
• Considering the impact of including temperature-squared and precipitation-squared terms
as quadratic terms;
• Comparing temperature versus humidity-adjusted temperature (a function of temperature
and humidity); and
• Considering alternative specifications of precipitation (monthly, seasonal, annual, etc.).
The precipitation variables used in ICLUS v2 were calculated from climate model output
downscaled using the BCSD methodology. Humidity-adjusted temperature is generally not
available as a downscaled climate model output. Instead, this variable was calculated using a
polynomial equation (eq 1-1) relating humidity-adjusted temperature to absolute temperature and
relative humidity (Rothfusz, 1990):
Humidity-adjusted temperature is calculated by:
Th = -42.379 + (2.04901523 X T) + (10.1433127 X RH) - (0.22475541 X T X RH)
- (0.00683783 X T2) - (0.05481717 X RH2) + (0.00122874 X T2 X RH)
+ (0.00085282 X T X RH2) - (0.00000199 X T2 X RH2)
(1-1)
Where:
Th = average monthly humidity-adjusted temperature
T = average monthly air temperature in degrees Fahrenheit
RH = average monthly relative humidity
Humidity-adjusted temperature (TH) was calculated only when absolute temperature (T) was
greater than 80°F and relative humidity was greater than 40%. When either of those conditions
was not met, unadjusted T was used.
2.1.4. Climate Change Model Selection
The selection of climate data for the migration model is another opportunity for
consistency with the SSP and RCP scenarios. For each of the RCP8.5 and RCP4.5 emission
scenarios, we identified two climate change projections that generally capture the range of
potential climate change for the contiguous United States. We constructed scatterplots of all
climate projections in the BCSD CMIP climate projection archive using climate amenity
9
-------
descriptions to form axes of "summer" and "winter" scatterplots and duplicated those scatterplots
for both emissions scenarios. As shown in Figure 4 below, the scatterplots provide a simple,
visual heuristic device to identify climate projections that bracket a broad range of future climate
change uncertainty. Using the plots in Figure 4, we selected projections from the HadGEM2-AO
and FIO-ESM climate models for the analyses included in this report. The selection of these two
climate models accomplished two goals: first, we wanted to represent the range of temperature
and precipitation changes in terms of a high and a low model, and second, we wanted to use the
same two models for both RCP 4.5 and RCP 8.5. Therefore, while the two climate models
selected do not always have the minimum or maximum temperature or precipitation values in the
scatterplot, they are the two models that balance our two goals most effectively.
10
-------
34 realizations of climate change from BCSD-CMIP5
Emissions Scenario: RCP45 Study area: conus
-------
migrations annually because the IRS data are based on single-year records. ICLUS vl was based
on 5-year migration records.
In addition, an important constraint was introduced to the updated migration model that
gives more reasonable population projections across the ICLUS v2 geographic framework. The
IRS migration records for 1991-2000 were grouped such that total migration between and
among metropolitan statistical areas (MSAs), micropolitan statistical areas, and stand-alone
(rural) counties could be quantified. The relative proportions shown in Figure 5 are used at each
annual time step to adjust the raw migrations calculated by the migration model. For example, at
each annual time step migration flows between MS As are rescaled to equal 70% of the national
migration total; total annual migration from rural counties to micropolitan statistical areas will
make up 1.6% of the national total, as depicted by each of the flows in Figure 5.
Proportion of migration by flow type
IRS files, 1991-2000
70%
3.2%
Metro
1.6%
Figure 5. Proportion of total migration between MSAs, micropolitan
statistical areas, and rural counties.
12
-------
The migratory flows shown in Figure 5 are consistent with U.S. Census Bureau data
reported for the 1995-2000 time period (U.S. Census Bureau, 2003). Incorporating these values
into the ICLUS v2 migration model provides two important advantages. First, we are able to
capture important macro-level trends, such as a net migration deficit for MSAs. Second, these
values serve as useful parameters for scenario exploration in future phases of ICLUS
development.
2.1.5.1. Revisedfunctional form and model statistics
The historical migration records and historical climate amenities discussed above were
combined so that a record in the migration data table contained the number of migrations from
one ICLUS Geographic Units (GU) to another, the attributes of the origin unit, the attributes of
the destination unit, and the functional distance between them. Attributes of the GUs include
population density, growth rate in the previous time period, developable area, and climate
variables. Equation 2-1 shows the variables used in the migration model. To estimate the
number of migrants, we used negative binomial regression with a natural log link. Predictor
variables were transformed as needed to control for skewness or heavy tails and were
standardized (Schielzeth, 2010). Because we expected that the amenity values associated with
temperature would depend on precipitation, we included interactions between those terms for
both summer and winter origin and destination units. As suggested by Dormann et al. (2013)
and to avoid the effects of collinearity, we used only predictor variables with absolute
correlations less than 0.70 (all correlations except those for summer and winter temperature were
less than 0.40). The migration model equation used balances theoretical considerations with
overall performance.
The migration model calculation is:
— Po + Pi x ln(^o) + [^2 x ln(Pi) + p3 x ln(^;')] +
[p4 X Gt 4 + p5 X Gj 4] + [p6 X ln04;) + p7 X ln(>l;-)] +
[p8 X SHt + /?9 X SHj] + [/?10 X WHt + /?n X WHj\ +
[ft2 X SPt + ft3 X SPj] + [ft4 X WPl12 + ft5 X WP*12] +
[/?16 X SHi X SPi + /?17 X SHj X SPj] +
[p18 X WHi X WPt1/2 + pig X WHj X WpV2]
13
-------
(2-1)
Where:
i = origin
j = destination
Fij = people migrating from unit i to unit j between year n and n + 1
pk = intercept or slopes quantifying the relationship between the parameters and number of
migrants
Dij = functional distance between unit i and j
P = population density
G = population growth rate, previous time step
A = developable land area
SH = mean summer (July) apparent temperature, 10 year running average
SP = mean summer (June, July, August) precipitation, 10 year running average
WH = mean winter (January) apparent temperature, 10 year running average
WP = mean winter (December, January, February) precipitation, 10 year running average
2.2. MIGRATION MODEL INTERPRETATION
The migration model parameters are derived from a generalized linear modeling
approach, so common measures of model performance are not available. However,
Nagelkerke's R2 was equal to 0.62 for the final model specification (Faraway, 2006).
Interpretation of the role of climate variables in the model is difficult, largely because
both origin and destination locations are affected simultaneously. Furthermore, migration is
calculated between all possible origin-destination pairs, meaning the observed net migration is
the difference between two opposing flows. Despite this complexity, the effects of variables in
the migration model may be characterized three ways.
First, the sign and magnitude of the coefficient indicates whether a variable will tend to
generally increase or decrease migrations. For example, winter temperature (WH) has a positive
coefficient for both the origin (WHi = 0.141) and destination (WHj = 0.207) locations. If all other
variables were held constant, more total migrations would occur between places with warm
winters, relative to places with cold winters in our model, though this is not a cause and effect
relationship of the climate variables. The magnitude of this influence is less than that of
population density (Pi = 0.530 andPj = 0.430), which exerts the largest influence on migration
(see Table 1).
14
-------
Second, comparing the origin and destination coefficients indicates the net directional
influence of that variable. For example, if all other factors are equal, the net flow of migrants
will be to locations with warmer winter temperatures (WHi < WHj) and less winter precipitation
(WPi > WPj\ see Table 1).
Table 1. Migration model results. Parameters are sorted by whether they
applied to origin or destination county (i or /), and matching pairs of
parameters share a row. Differences in slope estimates between matching
pairs of parameters are provided in the last column. Variables are defined in
Equation 2-1. /?£ is the estimate of the variable.
Parameter
Pk
P
Parameter
Pk
P
\Pki ~~ Pkj
Intercept
4.472
<0.0001
Dij
-1.048
<0.0001
Pi
0.530
<0.0001
Pj
0.430
<0.0001
0.100
Gi
0.027
<0.0001
Gj
-0.051
<0.0001
0.078
At
0.385
<0.0001
Aj
0.352
<0.0001
0.033
SHi
-0.080
<0.0001
SHj
-0.042
<0.0001
0.038
WHi
0.141
<0.0001
WHj
0.207
<0.0001
0.066
SPi
-0.088
<0.0001
SPj
-0.082
<0.0001
0.006
WPi
-0.077
<0.0001
WPj
-0.101
<0.0001
0.024
SHi x SPi
0.022
<0.0001
SHj x SPj
0.019
<0.0001
WHi x WPi
0.002
0.3040
WHj x WPj
0.040
<0.0001
Lastly, the relative contribution of each climate variable to net migration patterns is also
related to the absolute difference between the origin and destination coefficients (the last column
in Table 1), although we did not test the significance of this different. For example, winter
temperature is the most influential climate variable in the ICLUS v2 migration model, given both
the relative size of the absolute difference between the origin and destination coefficients and the
size of the coefficients relative to other climate variables. Summer temperature, winter
precipitation, and summer precipitation variables follow winter temperature in influence on net
migration.
15
-------
3. UPDATES TO THE SPATIAL ALLOCATION MODEL
ICLUS vl used the Spatially Explicit Regional Growth Model (SERGoM) to project
future increases of housing density at a relatively fine spatial resolution (Theobald, 2005;
Bierwagen et al., 2010). This update to the spatial allocation model addresses reviewers'
comments on ICLUS vl and incorporates advances in the literature on land use change
modeling. The new literature suggests that land use models should (1) incorporate spatial
dynamics7 and multiple sources of spatial heterogeneity, (2) explicitly describe transitional
dynamics of urban land use, (3) incorporate direct effects of market adjustments, (4) use
local-scale heterogeneity to determine urban spatial dynamics (Irwin, 2010), and (5) integrate
top-down and bottom-up methods that incorporate the effects of national and global drivers of
change while also accounting for local drivers of change and feedbacks (Sohl et al., 2010). For
ICLUS vl, SERGoM met the conditions for (1), (4), and partially (5). The revised allocation
model in ICLUS v2 addresses (2) by using a transition probability model, partially addresses (3)
by incorporating an assumption of maximum utility of land use (Alonso, 1964), and strengthens
(5) by using a finer spatial and thematic resolution. Another major change in ICLUS v2 is that
the land use modeled in ICLUS vl was the dynamic growth of a single (residential) land use in
ICLUS vl, although commercial and industrial lands were identified and held constant through
time. ICLUS v2 uses a deterministic demand-allocation approach, similar to SERGoM, which
assumes many aspects of future growth will resemble the recent past (i.e., 2000 to 2010), though
over time land use changes would result in different overall patterns. Different from ICLUS vl,
v2 sequentially allocates patches from seven of the 19 discrete land use classes (LUC) used in
ICLUS v2: five levels of residential, plus commercial and industrial. Thus, in ICLUS v2
commercial and industrial LUCs no longer remain constant.
3.1. OVERVIEW OF THE UPDATED SPATIAL ALLOCATION MODEL
The updated spatial allocation model incorporates information from multiple spatial
scales. At the national scale, all 2,256 ICLUS GUs (see Figure 3) in the conterminous United
States were used to construct a statistical model that generates local demands for new pixels
(90 m x 90 m) of land use based on changes in population density. The demand model captures
a log-log relationship that is consistent with a theory of city growth broadly relevant to many
aspects of city form and function (Bettencourt et al., 2007; Bettencourt, 2013; Batty, 2013).
Satisfying the demands for new land use involves using transition probabilities and land use
patch size and shape distributions that are region specific. Finally, local patterns of
7 A spatially dependent dynamic process is one in which a change over time at one location is dependent on the state
or changes in the state at other locations.
16
-------
transportation capacity and accessibility to commercial areas inform future spatial patterns of
growth. Figure 2 (see Section 1) illustrates the spatial allocation process and references specific
sections for each step in the flow diagram. The areas used to calculate regionally specific
distributions and demands are similar to U.S. Census Bureau regions (see Figure 6).
Figure 6. Regions used in ICLUS v2. Region 1-West Coast; Region 2-
Intermountain West; Region 3-North Central; Region 4-South Central; Region
5-Great Lakes; Region 6-Southeast; Region 7-Northeast.
The application of regions within ICLUS v2 is intended to maintain differences in land
use patterns across the country and over time. Within each region (see Figure 6), patterns of land
use change between 2000 and 2010 were summarized to form a land use transition matrix that
captured the probability of a given pixel converting to a specific land use category given (1) the
antecedent LUC and (2) the accessibility of the pixel.8 ICLUS v2 prioritizes pixels by transition
probability in order from highest to lowest. We similarly allocate new land use pixels beginning
with highest value land uses (e.g., Industrial, Commercial) and continuing in order to the lowest
value land uses (i.e., exurban-low). The process of allocating new land use pixels to the most
8 Accessibility is defined using the capacity (i.e., people per hour) of transportation infrastructure, and is updated at
each time step. This aspect of the model is elaborated later, in Sections 3.2 and 3.5.1.
17
-------
likely remaining location continues until demand for each LUC has been satisfied. While
somewhat simplistic, this approach nevertheless reflects classic land use theory, that is, a pattern
of transition to the highest and best use for a given location (Chisolm, 1962). New land uses are
then allocated as patches using a distribution of patch shapes and sizes for each LUC unique to
each region. These patches are used in ICLUS v2 at each time step such that the size, shape, and
frequency of new patches within a region reflect the new patches observed when comparing
2010 land use to 2000 land use.
The allocation of residential land use patches also considers accessibility to commercial
areas. This consideration holds to the precept that people will generally prefer to live close to
areas that offer employment opportunities, as well as the goods, services, and other amenities
associated with commercial development. A similar concept was used in ICLUS vl and yields a
modeling framework that is responsive to emergent urban areas.
3.2. CREATING THE INITIAL ACCESSIBILITY-CAPACITY SURFACE
When allocating new housing units, ICLUS vl used a nationwide surface of travel time
to preferentially weight new growth in areas most accessible to existing development, including
transitions to higher density development and new development. A key limitation of the ICLUS
vl model was that this travel time surface was static at each time step, and therefore was not
updated to reflect improvements to transportation networks. ICLUS v2 uses a more
sophisticated surface of accessibility that incorporates the capacity of roads and fixed mass
transit (i.e., people per hour), and is also updated at each time step. We refer to this as the
capacity surface.
The spatial allocation model is initialized with a capacity surface for the year 2010. To
generate the capacity surface, we follow methods outlined in Theobald (2008), which are
summarized here. Conceptually, this followed three steps, with details about each step in the
following paragraphs. First, we identified urban cores (e.g., central business districts) at multiple
resolutions. Second, we calculated the travel time to the nearest urban core for each pixel, which
reflects the polycentric nature of modern human settlements. For road infrastructure, we assume
travel speeds occur at typical speed limits for different road types, including fixed mass transit,
and for off-road pixels we used walking speeds. We then calculated the travel time from the
centroid of each urban core through the transportation infrastructure using cost-distance analysis
as determined by distance and travel time. This calculation was performed for each pixel.
Finally, we incorporated the different capacity of roads by increasing accessibility linearly by the
number of highway lanes.
Urban cores were built directly on the LUCs by converting developed LUCs to the
following weights: exurban high = 1; suburban and institutional (only where the National Land
18
-------
Cover Database [NLCD] identified developed areas with values of 23 or 24) = 5; urban low and
transportation = 8; and urban high and commercial = 10. These values were then aggregated by
summing their values to 270-m resolution. We then identified the upper half of values (greater
than mean of 136) and calculated a kernel density on these cells with a radius of 1 mile. Then,
we identified the cells resulting from the kernel-density operation that have values in the upper
half of values. To get at urban areas as a multiscale phenomenon, we generated urban core areas
at six spatial scales using the natural log of the number of cells. These areas of urban clusters
range from: 1.2, 3.2, 8.8, 23.9, 64.7, and 175.0 km2. We then identified the centroid of each of
these clusters at the six different scales and used these as the starting location from which to
calculate travel time. The benefit to this approach is that the centroid of the urban area is defined
by the land use pattern.
The next step was to create the cost weights that reflect the assumed travel speeds
through the transportation infrastructure. We assumed the same travel speeds as in ICLUS vl
but updated the transportation infrastructure to the U.S. Census Bureau Topologically Integrated
Geographic Encoding and Referencing (TIGER) 2010 roads.9 For each of the six urban cluster
starting locations, we generate a cost-distance layer that reflected the travel time from the urban
core through the infrastructure. We then combine the six time travel surfaces by averaging them
to generate a travel time surface.
The accessibility surface provides a platform on which to allocate new growth, but it does
not yet account for differences and changes in the capacity of the infrastructure. That is, most
infrastructure changes are simply to widen or increase the number of lanes on a given road,
rather than to generate a brand new highway through a roadless area. New, typically low-density
development can, and does, occur in large, roadless, previously undeveloped tracts of land even
though the model does not explicitly add roads to the landscape (further described below in
Section 3.4.1 and Figure 8).
To transform the travel time surface into a capacity surface (measured as passenger cars
per hour per lane), we calculated the number of cars that could be handled by converting travel
time to units of hours, then multiplying by the number of lanes of road. State and U.S. highways
and interstates that had information on the number of lanes in the National Transportation Atlas
Database10 were used, otherwise we assumed only a single lane (each way). We also accounted
for fixed mass transit (i.e., light rail). We assumed that a light rail system added the equivalent
in capacity as a single lane of interstate highway (roughly 2,000 passenger cars per hour per
9 ftp://ftp2.census.gov/geo/tiger/TIGER2010/RQADS/.
10 http://www.rita.dot.gov/bts/sites/rita.dot.gov.bts/files/publications/national transportation atlas database/
index, html.
19
-------
lane)11 because we did not have individual transit information on the number of cars, number of
passengers carried in each car, and other pertinent data.
We converted the continuous capacity surface into a series of eight capacity classes. To
identify the class thresholds, we calculated class breaks using the "Natural Breaks (Jenks)"
method in ArcGIS and then modified class breaks slightly using visual analysis of five
"representative" urban areas: San Francisco, Portland, Denver, Atlanta, and New York City. The
classes are at breaks of: 1 > 1,300; 2 = 900 to 1,300; 3 = 600 to 900; 4 = 300 to 600; 5 = 200 to
300; 6 = 150 to 200; 7 = 100 to 150; and 8 = < 100. We used these classes to compute the
transition probabilities of growth as a function of the broader neighborhood location of change,
rather than the more local scale that the strict LUC transitions provided. That is, for each land
use type, we found the transition probabilities for each capacity class independently (or jointly).
The capacity class values for time step t— 1 are combined with the land use surface from t - 1 to
yield a transition probability surface at time step t (see Figure 2).
3.3. ICLUS V2 LAND USE CLASSES
In ICLUS v2 land use is represented by 19 discrete categories delineated in the U.S.
National Land Use Dataset (US-NLUD; Theobald, 2014). The US-NLUD contains
high-resolution (90 m x 90 m pixels) land use information for the years 2000 and 2010 and
provides the statistical underpinnings for ICLUS v2 land use change probabilities. The
US-NLUD synthesizes data from multiple sources, including remotely sensed data, to map the
primary land use at a given location. Parameterization of ICLUS v2 is based on land use
transitions from 2000 to 2010, which may not remain constant over time. Changes to transition
probabilities are not explored in this report, but may be implemented in a scenarios context
within the ICLUS v2 framework.
From the US-NLUD, we retained four nonresidential land use categories (commercial,
industrial, institutional, and transportation) within the developed land use group, and further
subdivided the residential-urban and residential-rural subgroups to form five categories of
residential intensity. Urban residential uses are defined at the 1.6-dwelling units per acre (DUA;
3.95 units per hectare) threshold based on the U.S. Census Bureau definition of urban population
of 1,000 people per square mile (Theobald, 2001). The urban high category is greater than
10 DUA based on typical densities at which public transportation is viable (Ewing and Cervero,
2010). Suburban areas have residential densities below the urban low threshold but greater than
the 0.4 DUA threshold, which is commonly the density at which services such as municipal
sewer and water supply are provided. Lower densities are split into two additional categories
11 http://www.fhwa.dot.gov/ohim/hpmsmanl/appn2.cfm.
20
-------
with exurban high as 0.1-0.4 DUA and exurban low as 0.02-0.1 DUA. We also included nine
other land use/land cover categories that can be converted into developed land uses, such as
cropland, grazing, and timber. The complete list of LUCs used in ICLUS v2 is shown in
Table 2. Further detail on the entire US-NLUD can be found in Theobald (2014).
Table 2. Land Use Classes used in the ICLUS v2 model.
Code
Group
Class Name
0
Water
Natural water
1
Reservoirs, canals
2
Wetlands
3
Protected
Recreation, conservation
I4
W orking/production
Timber
5
Grazing
6
Pasture
I7
Cropland
8
Mining, barren land
9
Developed
Parks, golf courses
10
Exurban, low density
11
Exurban, high density
12
Suburban
13
Urban, low density
I14
Urban, high density
15
Commercial
16
Industrial
17
Institutional
18
Transportation
3.3.1. Quantifying Land Use Changes, 2000-2010
To examine relative changes in land use between 2000 and 2010, we estimated the
amount of land assigned to each of the seven developed LUCs and used the counts in 2000 and
2010 as observed values in chi-squared goodness-of-fit tests. For this analysis, the 90 m pixels
21
-------
were aggregated to 1 km pixels for computational efficiency. As the first step in the analysis,
both nationally and in each ICLUS region, we tested whether the total percentage of land in
developed LUCs increased from 2000 to 2010. Then we tested whether or not the percentage of
developed land assigned to the seven individual developed LUCs changed between 2000 and
2010. Only allowable transitions (see Table 3) were considered. The results of these statistical
tests show whether development increased significantly (p <0.05) between 2000 and 2010
nationally and for each region and whether development patterns (i.e., relative proportions of the
developed classes) changed significantly over the same period (p <0.05). Appendix A presents
results for each of the seven ICLUS regions.
If development patterns changed significantly, we examined the changes among the
seven developed classes for which ICLUS v2 models transitions. We first compared the odds
that a unit of land remained in the same developed LUC from 2000 to 2010 to the odds that it
transitioned to a different developed LUC for 2010. If the confidence interval (CI) of the
calculated odds ratio (OR) spanned zero, the percentage of developed land assigned to that
particular class did not change significantly between the two time periods. If the OR was
statistically significantly greater or less than zero, then the percentage of developed land assigned
to that particular class increased or decreased in 2010, respectively.
We compared the odds that a unit of land remained in the same residential LUC from
2000 and 2010 to the odds that it transitioned to the next most developed residential class for
2010. The five residential classes are only allowed to transition in one direction: progressively
from exurban low to urban high. This resulted in four comparisons: (1) exurban high versus
exurban low, (2) suburban versus exurban high, (3) urban low versus suburban, (4) urban high
versus urban low. If the CI of the calculated OR spanned zero, the relative amount of land
assigned to the two residential classes did not differ between 2000 and 2010 (i.e., the amounts of
the two residential classes were not distinguishable). If the OR was significantly greater or less
than zero, relatively more or less land, respectively, was assigned to the higher density
residential class in 2010. To correct for multiple comparisons and keep the family-wise error
rate at 0.05, confidence intervals of 98.3% were used. Furthermore, because the data were
aggregated at a 1-km2 resolution rather than at 8,100 m2 (the native resolution of the model), our
results should be considered conservative.
Combining the data from all ICLUS regions, both the percentage of land assigned to
developed use classes (%2 = 34,501.40, [degrees of freedom] df = 1 ,p <0.0001; Table 3, A;
Figure 7, C) and the relative amount of land assigned to each of the seven developed LUCs
(%2 = 276.07, df = 8, p <0.0001; Table 3, B) increased between 2000 and 2010. Among the
developed classes, the proportion of developed land in the urban low, commercial, and industrial
LUC decreased, the proportion of developed land in the exurban low, suburban, and urban high
22
-------
LUCs increased between 2000 and 2010, and the proportion of developed land in the exurban
high LUC did not change significantly between 2000 and 2010 (see Figure 7, A). Relative
growth in the urban high LUC was significantly larger than in the urban low LUC (see Figure 7,
B). Conversely, relative growth in the urban low LUC was significantly less than in the
suburban LUC. The relative growth in the suburban LUC tested statistically was not
significantly different than exurban high LUC, and growth in the exurban high LUC was not
significantly different than the exurban low LUC (see Figure 7, B).
Table 3. Goodness-of-fit test results comparing Land Use Classes in 2000
and 2010, nationally. Values are limited to developable area and Land Use
Classes that transition in the model. (A) Land assigned to developed and
undeveloped Land Use Classes. (B) Percentage developed land assigned to
the seven developed Land Use Classes.
(A) Land Use Type
2000
2010
Developed
12.60%
16.61%
Undeveloped
87.40%
83.39%
tf = 34,501.40
df: 1
/7-value: <0.0001
(B) Developed LUC
2000
2010
Exurban low
53.04%
53.33%
Exurban high
26.00%
26.14%
Suburban
8.93%
9.11%
Urban low
7.81%
7.48%
Urban high
0.44%
0.52%
Commercial
2.38%
2.22%
Industrial
1.40%
1.20%
-£ = 276.07
df: 8
/7-value: <0.0001
23
-------
1 40
1 20
(A)
1 00
0 80
0 60
Exurban Low Exurban High Suburban Urban Low Urban High Commercial Industrial
1.50
1.40
1.20
1.00
0.80
0.50
(B)
(C)
t
•
2
w
i
Exurban High vs. Suburban vs. Exurban Urban Low vs. Urban High vs. Urban Developed vs.
Exurban Low High Suburban Low Undeveloped
Figure 7. Land use comparisons between 2000 and 2010, nationally. (A)
Odds ratios (ORs) and confidence intervals that a unit of land stayed in the same
developed LUC fiom 2000 to 2010 compared to the odds that it switched to a
different developed LUC for 2010; (B) ORs and confidence intervals that a unit of
land stayed in the same residential LUC fiom 2000 and 2010 compared to the
odds that it switched to the next most developed residential class for 2010; and
(C) OR comparing developed and undeveloped LUCs.
3.4. TRANSITION-PROBABILITY MODEL
We calculated the transition probabilities between LUCs empirically from the baseline
change layers (i.e., 2000 and 2010 land use layers). We identified transitions that were plausible
24
-------
and then further identified transitions that were plausible but could not be supported by the
underlying data (see Table 4) to correct for spurious changes that resulted from artifacts in the
various data sets. For example, the institutional land use data set does not contain information
about the year that land use first appeared; therefore, we could not infer any change in the
institutional category. Furthermore, as in ICLUS vl, land uses transition to increasing intensity
and, therefore, "backwards" transitions are excluded (e.g., urban to suburban). Note that this
also requires generation of a modified land use data set for 2000, such that the classes are
consistent logically with 2010. In ICLUS v2, 2010 is the base year for future projections; thus,
the 2000 data set needed to be consistent with 2010 information.
25
-------
Table 4. Land Use Classes transitions from 2000 (rows) to 2010 (columns) incorporated into ICLUS v2. Filled
circles (•) denote transitions that were included in the model; shading is added for emphasis. Empty circles (o)
denote plausible transitions that were excluded for the purpose of model simplification. Hatches (x) denote
plausible transitions that were excluded from the model because temporal data were not available. Unmarked
transitions were excluded from the model because they were considered unlikely or infrequent and temporal
data were not available.
Water
Wetland
Rec Con
Timber
Graze
Pasture
Crop
Mining
Parks
Exurb L
Exurb H
Suburb
1-1
c
C3
a
UrbanH
Comm
Indust
Inst
Trans
Water
X
Wetland
X
X
X
X
X
X
X
•
•
•
•
•
•
•
X
X
Recreation and Conservation
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Timber
X
X
o
o
o
X
X
•
•
•
•
•
•
•
X
X
Grazing
X
X
o
o
o
X
X
•
•
•
•
•
•
•
X
X
Pasture
X
o
o
o
X
X
•
•
•
•
•
•
•
X
X
Cropland
X
o
o
X
X
•
•
•
•
•
•
•
X
X
Mining
X
X
X
X
X
X
X
X
X
X
X
Parks and Open Space
X
X
X
X
X
X
X
X
X
Exurban Low
o
o
o
o
X
X
•
•
•
•
•
•
X
X
Exurban High
X
X
o
•
•
•
•
•
X
X
Suburban
X
X
X
o
o
•
•
•
o
X
X
Urban Low
X
X
o
o
o
•
•
X
X
Urban High
X
X
o
o
o
o
•
X
X
Commercial
X
X
o
o
o
o
o
•
X
X
Industrial and Utility
X
X
X
o
o
o
o
o
o
X
X
Institutional
X
X
X
X
X
X
X
X
X
Transportation
X
X
26
-------
3.4.1. Empirical Estimation of Transition Probabilities
A series of multinomial generalized additive models (GAMs) were used to model LUC
transitions using the VGAM package in R (Yee, 2010; R Core Team, 2015). The GAMs predict
the probability that a pixel transitioned from one LUC to another between 2000 and 2010 by
transportation capacity class. Capacity class here is determined by binning raw capacity values
into eight ordinal values, 1-8, where lower values represent higher transportation capacity
(described in Section 3.2). The 53 possible transitions between LUCs were modeled in two
stages. First, for each ICLUS region, we modeled the probability that a pixel transitioned into
each LUC,/?(LUCj), by capacity class, where subscript j is the LUC in 2010. These seven
regional, "marginal" models had capacity class as their predictor variable and LUCj as a
categorical response variable (seven levels: exurban low, exurban high, suburban, urban low,
urban high, commercial, and industrial; see Figure 8). Second, seven "conditional" models for
the LUCj in each region model the probability that a pixel transitioned from a LUC in 2000
(represented as subscript i) if it transitioned into LUCj in 2010, /?(LUCiy), by capacity class.
Each of these models had capacity class as its predictor variable and LUCiy as a categorical
response variable (up to ten levels depending on the region and LUCj: wetland, timber, grazing,
pasture, cropland, exurban low, exurban high, suburban, urban low, urban high, and
commercial).
These sets of models are the basis for probability calculations that a pixel transitioned
from one LUC to another by multiplying the corresponding two model predictions together, that
is, for a given capacity class and region, the probability that a LUC transitioned from LUCj to
LUCj is /?(LUCij) =/?(LUCj) x p(LUCiy). Pixels that did not transition and response categories
with zero pixels were not included in the analysis and were given transition probabilities of zero.
The many transitions containing a small number of pixels required limiting the degrees of
freedom used in the smoothing function to three. Each multinomial GAM used a logit link, and
the LUC with the largest number of pixels was set as the reference category for comparison
purposes.
27
-------
Region 1 / Transitions Into LUC2010
Region 2 / Transitions Into LUC2010
— emraaniow — Urtanio* — ¦wusmai
fc«MtxjnHigh Uiwnwtf)
— SUunan — Canrr&uM
Enjruanuj* - UUianLw* mou-unal
fcnjrtanHiqh — UiMnMian
— - Suutur — Cocnmwcui
Capacity Class
Region 3 / Transitions Into LUC2010
Capacity Class
Region 4 / Transitions Into LUC2010
— fcxuran tow mwnLo* — ifxaiwia
~tufUMi.Mtqh URHR H<}t»
9UMMB — Comtwca
Capacity Class
Region 51 Transitions Into LUC2010
Capacity Class
Region 6 / Transitions Into LUC2010
o
co
o
to
o
©
o
o
2
3
4
5
6
7
8
o
CO
o
to
o
M
o
o
o
1
2
3
4
5
6
7
8
Capacity Class Capacity Class
Region 7 /Transitions into LUC2010
T
T
T
T
T
T
T
T
1 2 3 4 5 6 7 8
Capacity Class
Figure 8. Predicted transition probability by capacity class into Land Use
Classes in 2010. Each panel shows the probability of a pixel transitioning to each
of seven land uses based on observed 2000 to 2010 changes.
28
-------
Due to the large number of models, Tables B-l to B-7 (see Appendix B) show only
model outputs with statistically significant individual smoothing terms and global tests. Global
test results, which compare models with capacity class as a predictor variable to intercept only
models using the difference in the deviance and residual df between models, showed that
capacity class was a highly significant predictor of transition probability overall (p <0.0001 in all
cases, Tables B-l to B-7). Figure 8 shows the relationships between the probability of
transitioning into LUCj in 2010 and capacity class for the seven regional, marginal models.
Figures B-l to B-7 (see Appendix B) also show the full regional transitional probabilities,
created by multiplying the marginal and conditional models together. Generally, pixels were
more likely to transition into higher density residential classes at lower capacity class values and
vice versa with some regional variability on that overall pattern (see Figure 8 and Appendix B).
The intermediate density residential LUC showed unimodal responses, while the probability of
transitioning into urban high and exurban low monotonically decreased and increased in higher
capacity classes, respectively. Transitions into the commercial LUC displayed more regional
variability, with some monotonically decreasing with capacity class or displaying unimodal
behavior. Industrial transitions, however, were relatively low overall. The transition
probabilities do allow for irregular growth patterns that do not always follow the most likely
pattern (Appendix B). For example, the suburban LUC can transition into the commercial LUC,
which then becomes a new commercial core and an attractor for new residential growth.
Although the general pattern of transitions into the seven LUC held across the regions, we expect
the regional variability in observed transitions to produce different growth patterns over the
80-year projection period.
3.5. LAND USE AND CAPACITY DEMAND MODELS
To estimate LUC demands and changes in capacity, we created eight GAMs to predict
LUC density from population density within each ICLUS GU (see Figure 2 for representation of
ICLUS GUs). Capacity and each of the seven developed LUCs has its own GAM, created using
the mgcv package in R (Wood, 2004; R Core Team, 2015). Each model includes population
density, ln((people + 1) km-2), as its primary predictor variable and either capacity density per
km-2, Incapacity km-2) or LUC pixel density, ln((pixels + 1) km-2), as its response variable.
Density calculations use both 2000 and 2010 population data within each ICLUS GU and the
developable area for each ICLUS GU, estimated from the 2010 U.S. Census and USGS
Protected Areas Database of the United States (USGS, 2012), respectively.
Comparison of the difference in estimated number of pixels for each LUC or capacity,
the dependent variables of the eight GAMs, between adjacent time periods is the basis of the
demand calculation for each decade from 2020 to 2100. For example, 2050 demands were
29
-------
calculated by subtracting modeled 2040 from 2050 pixel counts or capacity. The pixel counts
and capacity for each ICLUS GU and decade were calculated by back transforming the value
%t + £1,2010, where j)/,? is the modeled response for a specified ICLUS GU and decade, and e/,2010 is
the raw residual associated with the 2010 measurement for that GAM and ICLUS GU. Adding
the raw residual for 2010 ensured that all ICLUS GU densities were scaled to their actual
densities in 2010, and that each GU followed a course parallel to the estimated density curve
over time on the log scale. In effect, this can be thought of as estimating proportional changes in
LUC density or capacity from proportional changes in population density. ICLUS v2 does not
generate LUC or capacity demands for counties that are projected to lose population, meaning
land use patterns in these counties do not change.
Table C-l (see Appendix C) presents summary results of the GAMs with a brief
overview presented here. Smoothing terms of the eight models are highly significant (p <0.0001
for all cases) and the adjusted R2 of the curves ranges from 0.550 for the exurban low model to
0.889 for the suburban model. Relationships between ln(population density) and ln(pixel
density) are displayed in Figure 9. For all exurban low and high classes, the relationship
between population and pixel density was unimodal, and monotonically increasing for all others.
This matched our expectation regarding urban land use succession (i.e., higher density pixels
should displace lower density pixels at high accessibility locations, while low-density pixels
displace nonurban land uses at the urban fringe). Urban high, the highest density class, continues
to increase rapidly with population density, while the rates of increase for other classes level off.
Generally, the persistence of high-density residential classes at high population densities
suggests urban areas are either better mixed (less likely to be replaced with growth) or that
expansion and replacement rates of these classes balance as cities expand outward. Similar to
high density residential classes, commercial and industrial classes tend to level off in counties
with high population densities. This leveling off indicates that these classes are not rapidly
replaced or that growth and replacement rates balance as counties grow. Similar to the urban
high residential class, transportation capacity initially increases approximately exponentially and
then linearly at higher population densities.
30
-------
E«urtjn Low ILUC10)
Exuiban Hsgti(LUC11I
»n estst< it—:
Suburban (OIC12)
iN F9pgU'i{*i .«—
Urban Low|LUC»3)
»N A»l,-4l»>r estst< It—J
Urban H>gh ILUC14)
iN F9pgU'i»*i &W»'IV *~-S •
Commeic>al (LUC15)
M
»N ^eCk'aW G9'iVE<
industrial (LUC 16)
iN Faun'i*"! .«*
Capacity
I £
Figure 9. Predicted log transformed pixel or capacity density (km 2) (± 2 SE)
by log transformed population density (km-2). Each panel shows a smooth
curve for a different Land Use Classes or for mean capacity nationally
31
-------
3.5.1. Updating the Accessibility-Capacity Surface
As shown in Figure 2, the surface of continuous capacity values at time step f — 2 is
updated and used to form a surface of land use transition probabilities at time step t. To
complete this update, we generate demand for new capacity units as a function of population
density and proportionally allocate that demand using region-specific weights calculated from
the 2000 and 2010 capacity surfaces. These weights are specific to each combination of LUC i
and region k First we calculated the sum of capacity units C by land use and region, averaged
across 2000 and 2010:
^*2000 i,k + ^2010 i,k
Li,k ~ ^
(3-1)
Next, we calculated a relative weight W for each LUC, where CMAX is the maximum
result from Equation 3-1 for region k and 0 < W^k < 1:
., j
wiik = -jr—
lmax
(4-1)
Equation 4-1 yields the final weights used to allocate new capacity units through time.
Each time the capacity update function is called, new capacity units Ufor pixel P are given as:
UD =
WF
Wr
x ZX
(5-1)
where Wp is the weight value from Equation 4-1; Wt is the sum of pixel weights for the entire
geographic unit being processed; and Dt is the countywide demand for new capacity units.
Equation 5-1 represents the culmination of the capacity update function.
32
-------
3.6. LAND USE PATCH ALLOCATION PROCESS
At each time step, the allocation of new land use pixels occurs on a county-by-county
basis. Industrial pixels are allocated first, based on the reasoning that fundamental services such
as water and electric utilities have the least flexibility in terms of location siting. Commercial is
allocated next, and the process continues iteratively through the urban, suburban, and exurban
residential classes following the highest-to-lowest order of land use intensity and value. After
the allocation of commercial patches, the model calculates a cost-distance surface such that each
pixel in the county is assigned a functional distance from commercial areas. All five residential
LUCs include this cost-distance surface as a spatial allocation weight for new patches. The order
in which LUCs are allocated, and the inclusion of accessibility to commercial pixels as an
amenity for residential classes, results in a land use change pattern that is generally consistent
with classic land use economic theory (Alonso, 1964). This process also allows new commercial
and urban centers to form that alter the cost-distance surface in the next time step.
ICLUS v2 uses the observed set of land use patches as an analog for future development
patterns. That is, for each LUC-region combination, a patch is drawn at random from the set of
patches that appeared between 2000 and 2010. That patch is compared against the transition
probability surface and placed at the location of the highest median probability, with the
constraint that all probabilities considered must be greater than zero. In the case of ties between
two or more locations, one location is selected at random. This process is repeated until the
demand for each land use is satisfied. If there are no remaining pixels with a greater-than-zero
probability of being converted, then any remaining demand is carried over to the next time step.
As in ICLUS vl, we assume that the vast majority of land use changes will be to a higher
intensity or value, and thus restrict new patches of land use from replacing pixels of a higher use.
There is no "undevelopment" in either ICLUS vl or ICLUS v2, although we recognize that in a
few urban areas (e.g., Detroit, Michigan) recent and unprecedented economic conditions have
resulted in conversion of higher density areas to less developed land uses.
The patch allocation process uses morphological functions from the Python programming
language,12 specifically the SciPy13 package (Jones et al., 2001). An important change in this
new version of the ICLUS model is the use of pseudorandom numbers at two stages of the patch
allocation process: (1) patch selection and (2) choosing between locations of equal probability.
It is not the goal of the ICLUS project to generate probabilistic forecasts of land use change;
therefore, stochastic processes were not incorporated into any phase of the model. Instead,
12 www. python, org.
13 The binary_hit_or_miss function from scipy.ndimage is used to identify valid locations for a new patch. The
medians/liter function is then used to identify the valid location(s) of the highest median transition probability.
33
-------
Python's random number generator was "seeded" at the start of the initial patch allocation
process for each county. For this, we used the integer version of the five-digit county Federal
Information Procession Standard (FIPS) code. This step ensures that, holding all other
parameters constant, consecutive runs of the model will yield identical results.14
4. RESULTS
This section discusses the consequences of the data set and model updates for ICLUS v2.
Similar to the overall model flow, Section 4.1 provides results for the demographic model, and
Section 4.2 describes land use changes.
The discussion of the demographic model begins at the national level, then examines
regional population trends including the effect of changing climate variables in the migration
model. This subsection delves into further detail on the influence of climate on domestic
migration by ICLUS geographic units in relation to climate variables. These maps demonstrate
the absolute and relative influence that climate change has on domestic migration in the
ICLUS v2 modeling approach. The discussion on migration concludes with an analysis of the
relative contribution of the different scenarios, climate models, and regions on migration
patterns.
The discussion of land use changes initially focuses on the addition of commercial and
industrial classes to the set of transitioning land uses. This section also examines growth in all
developed LUCs by region over time. Finally, comparisons of standardized LUCs between
ICLUS vl and v2 show the overall differences in output that result from all of the data set and
model updates.
4.1. POPULATION PROJECTIONS
4.1.1. National Projections
Figure 10 shows projections of total population for the conterminous United States.
Nationally, the ICLUS SSP5 scenario results in the highest total population because of higher
fertility rates than the ICLUS SSP1 scenario. The relative difference in population in 2100
between ICLUS SSP1 and 5 (229 million) is similar to the relative difference between the
International Institute for Applied Systems Analysis (HASA) SSP1 and 5 scenarios
(247 million), allowing qualitative comparisons and exploration of differences in impacts
14 Results shown in this report were generated on a computer using Windows 7 (64-bit) and Python version 2.7.10
and SciPy version 0.16. Executing the ICLUS v2 model on computers with different software will yield different
random number draws, despite the "seeding" process described above.
34
-------
between scenarios. Both SSP scenarios Ml within the range of the U.S. Census Bureau's 2000
projections (see Figure 10).
1,182
= 810
713
Q.
O
Q.
581
466
282
2000
2020
2040
2060
2080
2100
Elgure 10. Total population for the conterminous United States to 2100
showing projections for ICLUS v2. For comparison, historic and projected
population from the U.S. Census Bureau, and projected population from the
International Institute for Applied Systems Analysis (IIASA)15 are shown. The
most recent census projection (2014) aligns well with the SSP1 projection used in
this report through 2060.
4.1.2. Regional Projections
By region, ICLUS v2 total population projections are similar within the same SSP-RCP
combination but use different climate model output in the migration model (see Figure 11).
Even when climate change projections are selected to maximize differences, regional population
projections will largely reflect demographic parameters such as fertility rates, net immigration
assumptions, and so forth. Section 4.1.4 discusses differences between scenarios at the
subregional scale that arise from the spatial allocation model.
15 These population projections are available at https: // secur e. i i as a. ac. at/web- app s/ en e/S spD b.
Historic Census
Census (2000),
Census (2000),
Census(2014)
IIASA SSP5
IIASA SSP1
ICLUS SSP5
ICLUS SSP1
Highest Series
Lowest Series
National Population Projections
35
-------
250
SSP1, RCP4.5 (FIO-ESM)
SSP1, RCP4.5 (HadGEM2-AO)
200
150
100
E.
O 250 SSP5, RCP8.5 (FIO-ESM) SSP5, RCP8.5 (HadGEM2-AO)
J2
D
O 200
Q.
150
100
0
2020 2040 2060 2080 2100 2020 2040 2060 2080 2100
— Region 1 — Region 3 — Region 5 — Region 7
— Region 2 — Region 4 — Region 6
Figure 11. Total population for each ICLUS region to 2100 under four
scenario assumptions. SSP1 is low population growth, SSP5 is high population
growth; RCP 4.5 is low carbon emissions, RCP8.5 is high carbon emissions.
With respect to temperature increases over the United States, FIO-ESM and
HadGEM2-AO are among the models least and most sensitive to global
emissions, respectively.
4.1.3. The Effect of Changing Climate Amenities
A key feature introduced in ICLUS v2 is the integration of climate change as an amenity
(or tfa-amenity) in the migration model equation used to simulate domestic migration at each
annual time step (see Section 2.1.4). Using this additional information means that a wider range
of spatial patterns are theoretically possible with respect to population distribution because each
unique climate change projection should produce a unique pattern of domestic migration.
Moreover, small differences between two similar climate change projections could yield
36
-------
pronounced differences in migration patterns as the cumulative effect of simultaneously
adjusting amenity values for each geographic unit at each annual time step plays out over time.
Figure 12 shows the effect of climate change-induced migration by ICLUS region and
scenario relative to a migration model that, like ICLUS vl, holds climate amenity variables
constant over time for all scenarios. There were no entirely consistent patterns with respect to
population differences, as five regions (1, 3, 4, 6, and 7) had either higher or lower total
populations by 2100 depending on the scenario. The total population of Region 5 (Great Lakes)
was higher relative to the no climate change model regardless of scenario, while in Region 2
(Intermountain West) the opposite was true, especially under the ICLUS SSP5-RCP8.5
scenarios. Across all scenarios explored in this report, the effect of climate change-induced
migration on total population for any ICLUS Region was no more than about ± 2,500,000 people
(see Figure 12).
This diversity of outcomes is not surprising given the complexity of the underlying
model. Each climate change projection presents a unique spatiotemporal pattern of migration
model inputs. These patterns in turn alter the spatial distribution of population over time and
across the modeling domain and enhance or diminish migration feedbacks via other variables in
the migration model equation (i.e., population density or growth rate). While the relative net
effect of these interactions may total millions of people for a given region, we note that these
differences are a small fraction of total population. Figure 13 shows that, in relative terms, the
effect of climate change-induced migration is no more than —4% of the regional population, as
seen in Region 2 under SSP5-RCP8.5 using the HadGEM2-AO climate data. Most differences
are between ± 2% of the regional population regardless of scenario and climate model.
37
-------
Region 1
Region 2
8~5
s
Region 3
Region 4
a a q S f
3 8 8
A. v /
A
X X X
0 o a
Region 6
Region 5
o o
o o
8
Region 7
2020 2040 2060 2080 2100
o
O
o
SSP1, RCP4.5
O
o
SSP5, RCP8.5
A
X
X
SSP1, RCP4.5
X
SSP5, RCP8.5
2020
2040
2060
2080
2100
Figure 12. The effect of climate change-induced domestic migration
expressed as differences in millions of people. Differences in regional
population projection by emissions scenario and climate model are shown.
Values are expressed as the difference from a "no climate change" version of the
migration model. SSP1 is low population growth, SSP5 is high population
growth; RCP 4.5 is low carbon emissions, RCP8.5 is high carbon emissions.
With respect to temperature increases over the United States, FIO-ESM and
HadGEM2-AO are among the models least and most sensitive to global
emissions, respectively.
38
-------
4
Region 1
Region 2
0
0
c
CD
1
2
0
-2
4
4
2
0
-2
4
4
2
0
-2
4
Region 3
Region 5
MlSss
o
it-
Region 7
8 « fl X 3 X
8 8 8 8 g »
O
8 g
Region 4
. Q S ^ t
Q ® Q v '
A
X
X
S 6
Q
Region 6
2020
2040
2060
2080
U
v
0
x
y
* X X x
@ © © o
X
2100
O
X
SSP1. RCP4.5 (FIO-ESM)
SSP5, RCP8.5 (FIO-ESM)
SSP1, RCP4.5 (HadGEM2-AO)
SSP5, RCP8.5 (HadGEM2-AO)
2020
2040
2060
2080
2100
Figure 13. The effect of climate change-induced domestic migration
expressed as percentage differences. Relative differences in regional population
projection by emissions scenario and climate model are shown. Values are
expressed as the percentage difference from a "no climate change" version of the
migration model. SSP1 is low population growth, SSP5 is high population
growth; RCP 4.5 is low carbon emissions, RCP8.5 is high carbon emissions.
With respect to temperature increases over the United States, FIO-ESM and
HadGEM2-AO are among the models least and most sensitive to global
emissions, respectively.
39
-------
4.1.4. Subregional Projections
The effect of climate change on ICLUS v2 population projections can be further
illustrated with difference maps comparing climate variables derived from the FIO-ESM and
HadGEM2-AO climate projections and their respective population projections. An examination
and interpretation of the migration model is provided in Section 2.2; however, some general
spatial relationships between climate variable differences and population differences are
apparent.
For example, under the SSP5-RCP8.5 scenario assumptions, total population in Region 6
(Southeast) is generally higher when the migration model is driven by the HadGEM2-AO
climate projection (see Figures 14, A and B, green areas). In this comparison, all parameters and
assumptions are identical except for the annual climate amenity values; therefore, differences in
the spatial pattern of population are the cumulative result of migration differences where and
when the climate projections diverge.
The difference between the two climate models in terms of winter precipitation seems to
play an important role in this particular spatial pattern. While relatively warmer winter
temperatures are projected by the HadGEM2-AO model over most of the country, the
southeastern United States is one of the few areas to show relatively dryer winters by
HadGEM2-AO. The effect of markedly warmer winters (which would attract more migrants)
projected by HadGEM2-AO across the northern plains is difficult to discern because of generally
smaller, fewer, and more distant high-population areas relative to the southeastern United States.
In addition, relatively more winter precipitation would also work to slow migration into and
within the northern plains area.
A comparison of population difference maps in Figures 14 and 15 shows somewhat
larger migration differences under RCP4.5—the lower emissions scenario of the two. This
somewhat counterintuitive result is explained by the relatively larger difference between the two
climate models as shown in the climate maps. The spatial extent and magnitude of divergent
projections is clear for virtually all combinations of variables and years. The cumulative effect
of these comparatively larger differences in climate variables results in comparatively larger
migration differences.
These maps demonstrate some implications of the ICLUS v2 modeling approach;
however, care should be taken to avoid over-simplifying the apparent spatial relationship
between climate variables and population shown in Figures 14 and 15. The suite of interactions
and feedbacks present in the migration model extends beyond the figures presented here, and
cannot be exhaustively characterized by examples presented in this report.
40
-------
HADGEM2-AO(RCP85) minus FIO-ESM(RCP85): Summer precipitation
¦ ? 0 to 4 Jmmjday
HADGEM2-AO(RCP85) minus FIO-ESM(RCP85): Winter precipitation
mm, flay
[¦ 20 to 31 mnVday
HADGEM2-AQ(RCP85) minus F1Q-ESM(RCP85): Summer temperature
HADGEM2-AO(RCP85) minus FIO-ESM(RCP85): Summer precipitation
. rrnvday
HADGEM2-AOIRCP85) minus FIO-ESM(RCP85): Winter precipitation
HADGEM2-AO (SSP5, RCP85) minus FIO-ESM (SSP5, RCP85)
¦100.000
HADGEM2-AO (55P5. RCP85) minus FIO-ESM (SSP5, RCP85)
200.000,
1M.115 j
HADGEM2-AO(RCP85) minus FIO-ESM(RCP85): Winter temperature
HADGEM2-AO(RCP85) minus FIO-ESM(RCP85): Summer temperature
HADGEM2-AO(RCP85) minus F1Q-ESM(RCP85): Winter temperature
2100
Figure 14. Differences in population and climate change projections driven
by FIO-ESM and HadGEM2-AO under SSP5 and RCP8.5 assumptions for
2050 and 2100. (A) Population differences by ICLUS GU in 2050 and (B) in
2100; (C) differences in change in winter temperature in 2050 and (D) in 2100;
(E) differences in change in summer temperature in 2050 and (F) in 2100; (G)
differences in change in winter precipitation in 2050 and (H) in 2100; (I)
differences in summer change in precipitation in 2050 and (J) in 2100.
41
-------
HAOGEM2-AO(RCP45) minus F10-ESM(RCP45). Winter precipitation
HAD
-------
To further investigate the differences among scenario, climate model, and region, we
developed a fully factorial generalized least squares model to test the influence of these variables
and their interactions on mean 10-year changes in population density. The categorical
independent variables include ICLUS GU initial population density (people per km2 in five size
bins, PI: <5.0; P2: 5.1-15.0; P3: 15.1-45.0; P4: 45.1-135.0; P5: >135.1; see Figure 16), ICLUS
region (seven regions; see Figure 6), SSP (two scenarios: SSP1, SSP5), and climate model (three
levels: no climate change, FIO-ESM, HadGEM2-AO). The model also includes all possible
2-way, 3-way, and 4-way interactions among the variables. To meet the assumption of
homogeneity, we allowed each county size class to have its own residual variance (Zuur et al.,
2009). We ran separate models to look at population differences between 2010-2050 and
2060-2100 because results suggest higher divergence in populations by the end of the century
(see Figures 14 and 15).
During the initial decades modeled, 2010-2050, the magnitude of population change
depends on the initial population density, which varies by region (single 2-way interaction; see
Table 5, left). ICLUS GUs with higher initial population densities have larger increases in
population density overall and show the most distinct regional differences (see Figure 16, A). In
the second half of the century, 2060-2100, the magnitude of population change still depends on
the initial population density, but varies by both region and SSP (two 2-way interactions;
Table 5, right). As in the initial decades, ICLUS GUs with higher initial population densities
have larger increases overall and show the most distinct regional differences (see Figure 16, B).
Similarly, differences between SSPs are more distinct at higher population densities (see
Figure 16, C), in part because SSP5 uses a higher fertility rate and therefore has more people to
distribute across ICLUS GUs. The addition of the SSP variables in the late-century model shows
that the pathways diverge during this time period, but are similar during the first half of the
century.
43
-------
900
Q_
o
D_
CJ
Q.
o
D_
CJ
800
700
600
500
400
300
200
100
0
900
800
700
600
500
400
300
200
100
0
Mill
R1 R3 R5 R7 R2 R4 R6 R1 R3 R5 R7 R2 R4 R6 R1 R3 R5 R7
PI
P2
P3
P4
P5
I I
R1 R3 R5 R7 R2 R4 R6 R1 R3 R5 R7 R2 R4 R6 R1 R3 R5 R7
PI
P2
P3
P4
P5
£
(/>
c
O)
a
c
o
Q.
o
D_
CJ
900
800
700
600
500
400
300
200
100
. I I
SSP1 SSP5 SSP1 SSP5 SSP1 SSP5 SSP1 SSP5 SSP1 SSP5
PI
P2
P3
P4
P5
Figure 16. Average ICLUS GU 10-year population change by (A) starting
population density and ICLUS region from 2010—2050, (B) starting
population density and ICLUS region from 2060—2100, (C) starting
population density and SSP from 2060—2100. PI: <5.0; P2: 5.1-15.0; P3:
15.1-45.0; P4: 45.1-135.0; P5: >135.1 people per km2.
44
-------
Table 5. GLS model results. Model output includes degrees of freedom (df),
F-statistic, and significance (p). Nonsignificant terms (including interactions)
are included for completeness.
Change in population
density:
2010-2050
df
F
P
Change in population
density:
2060-2100
df
F
P
Initial population density
4
97.706
<0.0001
Initial population density
4
119.594
<0.0001
ICLUS region
6
20.639
<0.0001
ICLUS region
6
34.111
<0.0001
Socioeconomic pathway
1
0.183
0.6686
Socioeconomic pathway
1
1.534
0.2156
Climate model
2
0.055
0.9462
Climate model
2
0.024
0.9761
PyR
24
7.240
<0.0001
PyR
24
13.929
<0.0001
P*S
4
0.343
0.8493
P y S
4
3.750
0.0047
Ry-S
6
0.053
0.9994
Ry-S
6
1.441
0.1944
Py-M
8
0.071
0.9998
P y-M
8
0.084
0.9996
Ry-M
12
0.094
1.0000
R yM
12
0.246
0.9959
SyM
2
0.008
0.9919
SyM
2
0.007
0.9929
PyRy-S
24
0.050
1.0000
PyRy-S
24
0.501
0.9797
P y R y-M
48
0.034
1.0000
P y R yM
48
0.059
1.0000
Py-Sy-M
8
0.020
1.0000
Py-Sy-M
8
0.007
1.0000
RySy-M
12
0.017
1.0000
RySyM
12
0.028
1.0000
P y Ry S yM
48
0.008
1.0000
P yRySyM
48
0.009
1.0000
P: Initial population density
R\ ICLUS Region
S: Socioeconomic pathway
M: Climate model
45
-------
4.2. LAND USE PROJECTIONS
4.2.1. National Projections
The national-scale land use projections show nearly identical trends under the same SSP
assumption; the choice of climate model has no discernible effect on the overall amount of
projected development at the national level (see Figure 17). Relative to SSP 1, the larger national
population under SSP5 drives more development overall, particularly with respect to exurban
residential density (yellow wedge in all panels in Figure 17). By 2100, the area of developed
land in the conterminous United States increases by more than 80% of the 2010 value, yielding a
total of more than 1.6 million square kilometers under the SSP5 scenario. Under the SSP1
scenario, the increase is nearly 50%, and yields more than 1.3 million square kilometers of
developed land by 2100 (see Figure 17).
£
0)
"55
E
o
"O
c
C3
cn
3
6
SSP1, RCP4.5 (HadGEM2-AO)
SSP5, RCP8.5 (HadGEM2-AO)
1600
1200
800
400
1600
1200
800
SSP1, RCP4.5 (FIO-ESM)
SSP5, RCP8.5 (FIO-ESM)
400
2020
2040
2060
2080
2020
2040
2060
2080
EXURBAN
SUBURBAN
URBAN
COMMERCIAL
INDUSTRIAL
Figure 17. National land use projections from ICLUS v2 to 2100. Trends in
total area of exurban (exuiban low+ exurban high), suburban, urban (urban
low + urban high), commercial, and industrial lands are shown under four
scenarios.
46
-------
Differences in the percentage changes in each of the LUCs emerge nationally when
comparing SSP1-RCP4.5 and SSP5-RCP8.5 (see Figure 18). The SSP1-RCP4.5 projection using
FIO-ESM climate data has the smallest increases over time in terms of land use changes, as
compared to the SSP5-RCP8.5 projection using HadGEM2-AO climate data. These two
scenario combinations represent the extremes explored in ICLUS v2 in terms of demographic
and climatic change rates. For SSP1-RCP4.5, only the combined urban category increases by
more than 100% in 2100 and commercial land uses increase nearly that much. This scenario
consists of a relatively lower national population (SSP1) and lower anthropogenic perturbation of
the climate system (RCP4.5) modeled with a demonstrably less sensitive climate model
(FIO-ESM).
Conversely, the SSP5-RCP8.5 (HadGEM2-AO) projection models more than a 100%
increase in the extent of all developed LUCs already by 2050. The extent of urban land increases
by more than 200% by 2050 under this scenario, and more than quadruples by 2100. This
projection uses a very high population scenario (SSP5) and climate scenario of high
anthropogenic forcing (RCP8.5) modeled with a demonstrably more sensitive climate model
(HadGEM2-AO). This combination of model variables leads to greater changes in the extent of
developed lands than the SSP1-RCP4.5 (FIO-ESM) combination, even though the initial land use
demands and transition probabilities are the same. Changes in land use demands and transition
probabilities represent a future pathway to explore further differences among ICLUS v2
scenarios.
47
-------
SSP1. RCP4.5 (FIO-ESM) SSP5. RCP8.5 (HadGEM2-AO)
URBAN mm NDUSTRIAL MM DURBAN MB URBAN mm N0USTR1AL
COMMERCIAL mm SUBURBAN ¦-> COMMERCIAL
Figure 18. Relative increases in the area of developed LUCs nationally at
2050 (top row) and 2100 (bottom row). The left column shows results for
SSP1-RCP4.5 using FIO-ESM climate data; the right column shows results for
SSP5-RCP8.5 using HadGEM2-AO climate data. The SSP x RCP x climate
model combinations shown at the top of the graphs bracket the range of national
population projections, emissions scenarios, and climate model sensitivity,
respectively, of all combinations considered in this report.
4.2.2. Regional Projections
Summarizing the ICLUS v2 land use projections by region illustrates substantial
differences between the SSP1-RCP4.5 and SSP5-RCP8.5 scenarios. In almost every scenario
run, developed land use categories in all regions increase (see Table 6). The magnitudes of those
increases vary based on the SSP (i.e., population) assumption being considered.
ICLUS v2 projects a net decrease in the lowest residential density class (exurban-low) in
Region 7 (Northeast) by 2100 under the SSP5-RCP8.5 (HadGEM2-AO) scenario (see Table 6).
This singular instance of an extent decrease reflects the relatively high population density of the
northeastern United States, and the concomitant demand for higher density residential
development. In that case, the conversion of exurban-low pixels to other developed uses has
outpaced the demand for low-density residential pixels.
The urban-high LUC shows the greatest percentage increase by 2050 in both SSPs
considered, although smaller increases occur in Region 7 (Northeast) and, under SSP1-RCP4.5
(FIO-ESM), in Region 5 (Great Lakes). Substantial increases in commercial and industrial land
uses occur in Regions 2—4 under both SSPs by 2050, with more moderate increases in the
48
-------
remaining regions. Region 2 (Intermountain West), which is currently less densely developed
than most other regions, also has greater percentage increases in both exurban classes under both
SSPs by 2050. In 2100, this remains true for SSP1-RCP4.5 (FIO-ESM), although Regions 3 and
4 have the next highest percentage increases compared to the other regions, while the increases
in Regions 3 and 4 under SSP5-RCP8.5 (HadGEM2-AO) are more similar to Region 2 and larger
than the other regions (see Table 6). The overall regional pattern across both SSPs is that
urban-high increases sooner than lower density land uses, and that generally the pattern of
increases follows the density classes from urban-high to exurban-low.
49
-------
Table 6. Cumulative change in developed land use classes for 2050 (top row)
and 2100 (bottom row) by Shared Socioeconomic Pathways (SSPs),
Representative Concentration Pathways (RCPs) and climate model (in
parentheses). Values shown rep resent the change in square kilometers for
each LUC since 2010. Shading is used to describe that change as a
percentage, with the darkest gray indicating a >100% change, medium gray
50-100% change, ligjit gray 0-50% change, and peach <0% change.
SSPL RCP45(FIO-ESM)
2050
SSP5, RCP85) HadGEM2-AO)
2050
ICLUS
REGION
EXURBAN
LOW
EXURBAN
HGH
SUBURBAN
URBAN
LOW
URBAN
HGH
COMMERCIAL
INDUSTRIAL
1
30
321
245
584
112
101
58
2
1531
791
305
488
45
133
68
3
1344
804
305
492
30
157
91
4
2.859
1284
595
778
66
226
128
5
202
709
441
813
64
150
75
0
2.027
2.412
1.472
1347
122
238
144
7
147
223
190
230
87
42
15
2100
ICLUS
REGION
EXURBAN
LOW
EXURBAN
HGH
SUBURBAN
URBAN
LOW
URBAN
HGH
COMMERCIAL
INDUSTRIAL
1
131
497
437
1294
333
224
111
2
3
3.799
3280
1.005
1.720
712
712
1,153
1.125
130
99
292
341
139
195
4
7.(XX)
2.019
1242
1,735
184
484
252
5
751
1254
859
i.eee
170
301
138
0
4.044
4.109
2.804
2,787
348
607
261
7
51
202
395
497
257
94
32
EXURBAN
LOW
EXURBAN
HIGH
SUBURBAN
URBAN
LOW
URBAN
HIGH
COMMERCIAL
INDUSTRIAL
55
393
310
764
152
132
73
2,013
1590
3.881
993
933
1.663
383
338
767
642
588
1,031
62
44
92
171
182
2S6
86
105
164
300
887
518
982
83
181
89
2,608
2.788
1,717
1572
152
353
166
99
213
221
272
113
50
17
2100
EXURBAN
LOW
EXURBAN
HIGH
SUBURBAN
URBAN
LOW
URBAN
HIGH
COMMERCIAL
INDUSTRIAL
549
6,657
5615
12348
728
2.433
2592
4.082
651
1,082
1,072
2,027
1,889
1.885
1512
2978
645
251
183
381
366
461
525
813
161
209
293
393
1,761
1.719
1292
2.619
331
470
205
8.619
6380
4515
4.612
705
1.039
411
-21
270
530
851
506
164
56
Percent Change
< 0%
0-50%
50-100%
>100%
4.2.3. Sub regional Projections
Decadal land use maps show changes for three selected metropolitan areas (see
Figures 19-24). Net changes in other land uses classes (e.g., agriculture, recreation) are only
negative and only occur as a result of transitions into developed classes. For example, in the
Portland, OR-V ancouver, WA metropolitan area most of the growth in low-density urban land
uses results from conversion of suburban and exurban areas, although more conversions of
cropland to urban low occur in the decades from 2050-2100 than the earlier time period under
both SSPs (see Figures 19 and 20). Similar trends also occur in cities in other regions (e.g.,
Springfield, MO; see Figures 21 and 22). In contrast, some metropolitan areas that already have
multiple high-density urban centers throughout the area (e.g., Washington, DC metropolitan
area) and have high population growth convert more of the existing residential land uses to
50
-------
additional high-density urban areas under both SSPs (see Figures 23 and 24). These three
metropolitan areas exemplify changes nationally in such areas and illustrate the spatial patterns
produced using ICLUS v2.
51
-------
Figure 19. Land use change in the vicinity of the Portland, OR-Vancouver,
WA Metro Area under the SSP1-RCP4.5 (FIO-ESM) scenario: 2010, 2050,
and 2100.
52
-------
Wetlands
\
Exurban, low
Exurban, high
Suburban
Urban, low
Urban, high
Commercial
Industrial
Institutional
Transportation
Natural water
Reservoirs, canals
Conservation
Timber
Grazing
Pasture
Cropland
Mining, barren
Parks, open space
Figure 20. Land use change in the vicinity of the Portland, OR-Vancouver,
WA Metro Area under the SSP5-RCPS.5 (HadGEM2-AO) scenario: 2010,
2050, and 2100.
53
-------
2010
r
r
. L
kJ
1
y°Sr.
1 «*.S-
X*
. *<¦ IV
s_» g
J 3
H ' /¦
V ¦.
rurfflv^rl
¦
w. ...
U «
-/tVa
— "" ^'Vs " 1
J '
t.
Natural water
J Exurban, low
]] Reservoirs, canals
| Exurban, high
Wetlands
] Suburban
Conservation
J Urban, low
Timber
1 Urban, high
Grazing
] Commercial
Pasture
_J Industrial
| Cropland
1 Institutional
Mining, barren
1 Transportation
Parks, open space
2050
U" *,r
i'
"¦*1
*YJ?v
.<-¦¦•
f
Cl ¦ Jc£&
$s£ ^sm^m
M&mm
j**A
mmwwGf?" <*s
met
IrliL
2100
-' *•:
i'
> * «u
^383 1
-'l'"' S-Wr.f ^
.4Vs
_JT.
r -to
3 Miles
Figure 21. Land use change in the vicinity of the Springfield, MO Metro
Area under the SSP1-RCP4.5 (FIO-ESM) scenario: 2010, 2050, and 2100.
54
-------
2010
"V
- * * '•'?»
>*<
t >
•' > -i '
:?¦
w;.
—V-
Jtk£
•t.
p T 1. { •'
\ - ru
Natural water
Reservoirs, canals
Wetlands
Conservation
Timber
Grazing
Pasture
Cropland
Mining, barren
Parks, open space
Exurban, low
Exurban, high
Suburban
Urban, low
Urban, high
Commercial
Industrial
Institutional
Transportation
2050
r
, • * V s'v
v
tflh * i
¦J
• i -
2100
» N
" ">T
¦ "***
, V ' .•
10 l
a.
3 Miles
Figure 22. Land use change in the vicinity of the Springfield, MO Metro
Area under the SSP5-RCP8.5 (HadGEM2-AO) scenario: 2010, 2050, and
2100.
55
-------
Wetlands
Natural Water
Reservoirs, canals
Conservation
| Timber
~ Grazing
]] Pasture
| Cropland
Mining, barren
Parks, open space
Exurban, low
Exurban, high
Suburban
Urban, low
Urban, high
Industrial
Commercial
Institutional
Transportation
Figure 23. Land use change in the vicinity of the Washington-Arlington-
Alexandria, DC-VA Metro Area under the SSP1-RCP4.5 (FIO-ESM)
scenario: 2010, 2050, and 2100.
56
-------
Wetlands
\
Exurban, low
Exurban, high
Suburban
Urban, low
Urban, high
Industrial
Commercial
Institutional
Transportation
Natural Water
Reservoirs, canals
Conservation
Timber
Grazing
Pasture
Cropland
Mining, barren
Parks, open space
Figure 24. Land use change in the vicinity of the
Washington-Arlington-Alexandria, DC-VA Metro Area under the
SSP5-RCP8.5 (HadGEM2-AO) scenario: 2010, 2050, and 2100.
5. CONCLUSION
The updated data sets and underlying statistical and spatial methods result in realizations
of future land use changes that are substantially different from ICLUS v 1. The revisions made
for ICLUS v2 have many advantages, particularly for assessments of future climate change
impacts, vulnerabilities, and adaptation options. These advantages include the ability to
(1) develop future scenarios that include changes in commercial and industrial land uses,
57
-------
(2) examine the effect of changes in transportation capacity through additional lane miles or
added fixed mass transit, (3) examine trends in land use changes regionally, and (4) assess
differences among scenarios consistent with current socioeconomic and emissions storylines
(i.e., SSPs and RCPs). However, some of the updates have disadvantages. For example, the use
of the IRS migration data set requires collapsing all age classes from the cohort component
model into one population, compared with the two age groups used in ICLUS vl. The loss of
this demographic information theoretically results in less useful model outputs because the
assessment of future health impacts related to climate change typically is improved by using
segmented age groups. This limitation is somewhat mitigated by the fact that ICLUS vl only
retained two broad population segments, over 50 and under 50. An additional limitation of a
single population is that people of different ages move in different patterns (e.g., Voorhees et al.,
2011) and may respond differently to future climate. These behaviors are likely to have
repercussions in the population and land use patterns generated by ICLUS v2. Methods to add
more detailed demographic information back into the migration model would make the
population outputs from ICLUS v2 more useful for the health impacts communities, research on
vulnerable populations, and examinations of potential environmental justice issues.
ICLUS v2 represents significant progress in the development of land use change
scenarios that are consistent with emissions story lines and has the flexibility to adapt to other
emerging storylines from the climate change modeling community. For example, land use
transitions can be altered by changing the population density and land use demand relationships.
The current transitions are based on a limited temporal segment of land use data (2000-2010)
and remain constant over time. These transition probabilities may change over time, and this
change currently is not represented in the model. There are several options for exploring
changes in transitions over time. For example, new land use change information can be used to
compared predicted land uses to actual land uses in 2015. This would yield information on
deviations from near-term trends. Exploring longer term implications of changes in land use
transitions can employ a scenarios approach. Both of these approaches can inform on potential
trajectories and environmental impacts.
The current ICLUS v2 land use transitions follow an expected development path from
low to high densities, generally expanding outwards from population centers. Higher density
residential classes, commercial, and industrial development exhibit a threshold effect at high
population densities, such that these land uses generally are not replaced once they are
developed. This tendency has implications in terms of the continuity of urban form,
redevelopment patterns, creation of park and recreation areas, and other "undevelopment"
(e.g., transitions from higher land use classes to lower ones as a result of declining population),
which in turn influences subsequent development patterns. One potential consequence of not
58
-------
allowing LUCs to transition to lower density or nondeveloped uses is that these data sets
overestimate impervious surface cover and its impacts, even though such surfaces may remain
for many years following population loss from an area. Alternatively, some industrial sites may
be redeveloped into lower use classes such as residential housing, in this case also altering the
impervious surface cover estimates and population densities. While the current model does not
explicitly include these types of transitions, the model structure does allow for the future
exploration of these phenomena through scenarios.
ICLUS v2 also makes significant progress in providing future estimates of commercial
and industrial land use changes. These estimates serve as inputs to a variety of environmentally
relevant models that project changes in emissions and other air quality factors. Additional
research into the emergence of new commercial areas and densities and occurrences of mixed
commercial and residential buildings in urban areas would be useful inputs into future ICLUS
updates and land use change scenarios. Data on the emergence of new commercial and industrial
centers, as well as associated impervious surface cover, are critical for modeling future changes
in a variety of air and water quality endpoints, including emissions of criteria air pollutants,
greenhouse gases, and stormwater runoff.
Another important advancement of ICLUS v2 is the inclusion of future climate change
variables in the migration model. While climate variables represent a relatively small
instantaneous influence on migration, the cumulative effect of this influence through time on a
process as complex as human migration results in meaningful spatial variability of population
projections across the ICLUS GUs. The strength of this influence also can be explored through
scenarios that alter migration responses to climate change over time. Additionally, differences in
migration patterns can be explored as other climate model data are incorporated.
The use of changing climate variables in the migration model does produce some
differences in population distribution. Differences in regional populations between static and
dynamic climate variables are no more than approximately 4%. Most differences are ±2% of the
regional population, regardless of scenario and climate model combination. Nationally, the
choice of climate model has little effect on the overall development pattern. However, this
report only used two climate models as examples to implement the changes in the ICLUS v2
models. Other climate change models may have more extreme temperature or precipitation
values in certain regions that may exert larger influences on population migration. ICLUS v2
users can explore impacts of other climate change model values as part of scenario and
sensitivity analyses. However, as projected temperatures and precipitation amounts become
more extreme in some models, these values will be outside of the range of the data used to
parameterize the migration equation.
59
-------
The results presented in this report cover only two of the many possible GCMs and two
emissions scenario. Data from other climate change models can be incorporated easily into the
migration model. Additional emissions scenarios also can be explored. Transition probabilities
and land use and capacity class relationships can be modified to create land use patterns
consistent with SSP and RCP combinations not explored in this report.
As in ICLUS vl, this version focuses on developed land uses. It would be useful to
integrate ICLUS v2 with models using similar principles that change other land uses, such as
agriculture and forestry—particularly for more comprehensive assessments of impacts,
vulnerabilities, and adaptation options related to climate change. The composition of
agricultural, forest, and natural landscapes has changed and will continue to change over time in
response to human, climatic, and other factors. A large body of research exists that models
changes in various species distributions under the SRES storylines (e.g., Thomas et al., 2012).
These types of analyses can make use of the changing development patterns from the ICLUS
output, and provide feedbacks from changes in the undeveloped landscape that can be
incorporated into the ICLUS modeling structure. Several models exist that can easily integrate
ICLUS data and vice versa. For example, the FOREcasting SCEnarios of Land-use Change
model (FORESCE; Sohl et al., 2007) also uses scenario assumptions to examine changes in
forest composition in the future, while the Forestry and Agricultural Sector Optimization Model
(FASOM) can integrate changes in the available agricultural and forest land area to develop
projections of future markets based on population demands (Zhang et al., 2014). These types of
feedbacks and interactions among changes in land use and land cover are an active area of
research that are likely to improve future version of ICLUS output.
The data sets resulting from ICLUS v2 can serve as inputs for other models to further
investigate changes in environmental and human health endpoints. Many models use population
as a critical variable, and ICLUS v2 enables scenario-based explorations of the endpoints of such
models. These types of analyses also can explore such endpoints in the context of the global
SSPs and RCPs because of the consistency of the ICLUS v2 outputs with those scenarios. Other
models also use a combination of population and land use variables for which ICLUS v2 can
provide inputs. In some cases, the scenarios of land use change provided by ICLUS v2 can add a
novel forward-looking component to other models and further analyses of feedbacks among land
uses or influences from land use changes on specific endpoints. The range of data sets and the
consistency of the data sets with SSPs and RCPs facilitates the use of ICLUS v2 in many
applications.
60
-------
6. REFERENCES
Alonso, W. (1964) Location and land use. Cambridge: Harvard University Press.
Alonso, W. (1971) The system of intermetropolitanpopulation flows. [Working Paper No. 155], Prepared for the
National Commission on Population Growth and the American Future. Berkeley, CA: University of
California, Institute of Urban and Regional Development.
Batty, M. (2013) A theory of city size. Science 340:1418-1419.
Bettencourt, LMA. (2013) The origins of scaling in cities. Science 340:1438-1441.
Bettencourt, LMA; Lobo, J; Helbing, D; Kuhnert, C; West, GB. (2007) Growth, innovation, scaling, and the pace of
life in cities. Proc Natl Acad Sci 104(17):7301-7306.
Bierwagen, B; Theobald, DM; Pyke, CR; Choate, A; Groth, P; Morefield, P; Thomas, JV. (2010) National housing
and impervious surface scenarios for integrated climate impact assessments. Proc Natl Acad Sci
107(20887-20892).
Chisholm, M. (1962) Rural settlement and land use. London: Hutchinson.
Cragg, M; Kahn, M. (1996) New estimates of climate demand: evidence from location choice. J Urban Econ 42:261-
284.
Dormann, CF; Elith, J; Bacher, S; Buchmann, C; Carl, G; Carre, G; Garcia Marquez, JR; Gruber, B; Lafourcade, B;
Leitao, PJ; Munkemuller, T; McClean, C; Osborne, PE; Reineking, B; Schroder, B; Skidmore, AK; Zurell,
D; Lautenbach, S. (2013) Collinearity: a review of methods to deal with it and a simulation study
evaluation their performance. Ecography 36:27-46.
Ewing, R; Cervero, R. (2010) Travel and the built environment. J Am Plann Assoc 76(3):265-294.
Faraway, JJ. (2006) Extending the linear models with R: Generalized linear, mixed effects and nonparametric
regression models. Boca Raton, Fl: Chapman and Hall/CRC Press
Feng, S; Krueger, AB; Oppenheimer, M. (2010) Linkages among climate change, crop yields and Mexico-US cross-
border migrations. Proc Natl Acad Sci 107(32): 14257-14262.
Georgescu, M; Morefield, PE; Bierwagen, BG; Weaver, CP. (2014) Urban adaptation can roll back warming of
emerging megapolitan regions. Proc Natl Acad Sci 111(8):2909-2914.
IRS (Internal Revenue Service). (2014) SOI tax stats - migration data, [Website], http://www.irs.gov/uac/SOI-Tax-
Stats-Migration-Data. Last updated October 8, 2015.
Irwin, EG. (2010) New directions for urban economic models of land use change: incorporating spatial dynamics
and heterogeneity. JReg Sci 50(1):65-91.
Jones, E; Oliphant, E; Peterson, P; et al. (2001) SciPy: open source scientific tools for Python, http://www.scipy.org/
KC, S; Lutz, W. (2014) The human core of the shared socioeconomic pathways: Population scenarios by age, sex
and level of education for all countries to 2100. Global Environ Change.
doi: 10.1016/i.aloe.nvc.ha.2014.06.004 (online pub)
Maxwell, JT; Soule, PT. (2011) Drought and other driving forces behind population change in six rural counties in
the United States. Southeast Geogr 51(1): 133-149.
Maurer, EP; Brekke, L; Pruitt, T; Duffy, PB. (2007) Fine-resolution climate projections enhance regional climate
change impact studies. Eos Trans Am Geophys Union, 88(47):504.
61
-------
McGranahan, D. (1999) Natural amenities drive rural population change. [Agricultural Economic Report No.
AER781], Washington, DC: U.S. Department of Agriculture.
Nakicenovic, N; Swart, R; eds. (2000) Special report on emissions scenarios. Cambridge, UK: Cambridge
University Press.
NCHS (National Center for Health Statistics). (2011) April 1, 2010 Bridged race population estimates. Released
November 17, 2011. Atlanta, GA: Centers for Disease Control and Prevention (CDC).
http://www.cdc.gov/nchs/nvss/bridged race/data documentation.htm#april2010
OMB (Office of Management and Budget). (2010) Standards for delineating metropolitan and micropolitan
statistical areas. Fed Reg 75(129):37246-
39952. https://www.whitehouse.gov/sites/default/files/omb/assets/fedreg 2010/06282010 metro standard
s-Complete.pdf
O'Neill, BC; Kriegler, E; Riahi, K; Ebi, KL; Hallegatte, S; Carter, TR; Mathur, R; van Vuuren, DP. (2014) A new
scenario framework for climate change research: the concept of shared socioeconomic pathways. Clim
Change 122: 387-400.
R Core Team. (2015) R: A language and environment for statistical computing. Vienna, Austria: R Foundation for
Statistical Computing.
Rappaport, J. (2007) Moving to nice weather. Reg Sci Urban Econ 37:375-398.
Rothfusz, LP. (1990) The heat index "equation" (or, more than you ever wanted to know about heat index).
[National Weather Service Technical Attachment SR 90-23.] Fort Worth, TX: Scientific Services Division,
NWS Southern Region Headquarters, http://www.srh.noaa.gov/images/ffc/pdf/ta htindx.PDF
Samir, KC; Lutz, W. (2014) Demographic scenarios by age, sex and education corresponding to the SSP narratives.
Popul Environ 35:243-260.
Schielzeth, H. (2010) Simple means to improve the interpretability of regression coefficients. MethEcol Evol 1:103-
113.
Sinha, P; Cropper, ML. (2013) The value of climate amenities: evidence from US migration decisions. [Working
Paper 18756] Cambridge, MA: National Bureau of Economic Research.
Sohl, TL; Sayler, KL; Drummond, MA; Loveland, TR. (2007) The FORE-SCE model: a practical approach for
projecting land use change using scenario-based modeling. J Land Use Sci 2(2): 103—126.
Sohl, TL; Loveland, TR; Sleeter, BM; Sayler, KL; Barnes, CA. (2010) Addressing foundational elements of regional
land-use change forecasting. Landsc Ecol 25(2):233-247.
Sussman, F; Saha, B; Bierwagen, BG; Weaver, C; Morefield, P; Thomas, J. (2014) Estimates of changes in county-
level housing prices in the United States under scenarios of future climate change. Clim Change Econ
05(03):1450009
Theobald, DM. (2001) Land-use dynamics beyond the American urban fringe. Geogr Rev 91(3):544-564.
Theobald, DM. (2005) Landscape patterns of exurban growth in the USA from 1980 to 2020. Ecol Soc 10(1):32.
http://www.tetonwvo.org/compplan/LDRUpdate/RuralAreas/Additional%20Resources/Theobald2005.pdf
Theobald, DM. (2008) Network and accessibility methods to estimate the human use of ecosystems. Proceedings of
the 11th AGILE International Conference on Geographic Information Science 2008, University of Girona,
Spain. http://www.agile-online.org/Conference Paper/CDs/agile 2008/PDF/107 DOC.pdf
Theobald, DM. (2014) Development and applications of a comprehensive land use classification and map for the
US. PLoS ONE 9(4): e94628. doi:10.1371/journal.pone.0094628
62
-------
Thomas, KA; Guertin, PP; Gass, L. (2012) Plant distributions in the southwestern United States; a scenario
assessment of the modern-day and future distribution ranges of 166 species. [U.S. Geological Survey Open-
File Report 2012-1020, 83 p. and 166-page appendix], Washington, DC: U.S. Department of Interior, U.S.
Geological Survey, http://pubs.usgs. gov/of/2012/1020/
U.S. Census Bureau. (2000) Assumptions for the components of change. In Methodology and assumptions for the
population projections of the United States: 1999-2100. Washington, DC: U.S. Census Bureau,
Department of Commerce, http://www.census.gov/population/proiections/data/national/natproi2000.html
U.S. Census Bureau. (2003) Census 2000, Public Use Microdata Sample (PUMS). Washington, DC: U.S. Census
Bureau, Department of Commerce. https://www.census.gov/census2000/PUMS5.html.
U.S. EPA (Environmental Protection Agency). (2009) Land-use scenarios: National-scale housing-density scenarios
consistent with climate change storylines (Final Report). [EPA/600/R-08/076F], Washington, DC: National
Center for Environmental Assessment.
http://cfpub.epa.gov/ncea/risk/recordisplay.cfm?deid=203458&CFID=42769880&CFTOKEN=75743611
USGS (Geological Survey) (2012). Protected areas database of the United States (PADUS) version 1.3. National
Gap Analysis Program.
http://gapanalvsis.usgs. gov/padus/?s=+Protected+Areas+Database+of+the+United+States+PADUS+versio
n+13&submit=Go
van Vuuren, DP; Edmonds, J; Kainuma, M; et al. (2011) The representative concentration pathways: an overview.
Climatic Change 109: 5-31.
van Vuuren, DP; Carter, TR. (2014) Climate and socio-economic scenarios for climate change research and
assessment: Reconciling the new with the old. Clim Change 122:415-429.
Voorhees, AS; Farm, N; Fulcher, C; Dolwick, P; Hubbell, B; Bierwagen, B; Morefield. (2011) Climate change-
related temperature impacts on warm season heat mortality: a proof-of-concept methodology using
BenMAP. Environ Sci Technol 45(1450-1457.
Wood, AW; Leung, LR; Sridhar, V; Lettenmaier. DP. (2004) Hydrologic implications of dynamical and statistical
approaches to downscaling climate model outputs. Clim Change 62:189-216.
Wood, SN. (2004) Stable and efficient multiple smoothing parameter estimation for generalized additive models. J
Am Stat Assoc 99:673-686.
Yee, TW. (2010) The VGAM package for categorical data analysis. J Stat Softw 32:1-34.
Zhang, Y-Q; Cai, Y-X; Beach, RH; McCarl, BA. (2014) Modeling climate change impacts on the US agricultural
exports. J Integr Agricul 13(4):666-676.
Zuur, A; Ieno, EN; Walker, N; Saveliev, AA; Smith, GM. (2009) Mixed effects models and extensions in ecology
withR. New York, NY: Springer.
63
-------
APPENDIX A. REGIONAL LAND-USE CHANGES FOR 2000-2010
A.l. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 1
(PACIFIC) LAND USE CHANGES
In the Pacific region (Integrated Climate and Land Use Scenarios [ICLUS] Region 1), the
percentage of land assigned to developed use classes increased between 2000 and 2010 (see
Table A-l, A, Figure A-l, C). Over the same period, the relative amount of land assigned to
each of the seven developed land use classes (LUCs) also changed (see Table A-l, B). Among
the developed classes, the proportion of developed land in the urban low LUC decreased, while
the proportion of land in the urban high LUC increased between 2000 and 2010 (see Figure A-l,
A). The relative amount of developed land in the exurban low, exurban high, suburban,
commercial, and industrial LUCs did not change statistically significantly between 2000 and
2010. Relative growth in the urban high LUC was larger than in the urban low LUC (see
Figure A-l, B). The relative amount of growth in paired comparisons of exurban high with
exurban low, suburban with exurban high, and urban low with suburban LUCs show no
statistically significant differences.
A-l
-------
Table A-l. Goodness-of-fit test results comparing LUCs in 2000 and 2010 in
Integrated Climate and Land Use Scenarios (ICLUS) Region 1 (Pacific).
Values are limited to developable area and LUCs that transition in the model. (A)
Land assigned to developed and undeveloped LUCs. (B) Percentage developed
land assigned to the seven developed LUCs.
(A) Land Use Type
2000
2010
Developed
13.33%
15.72%
Undeveloped
86.67%
84.28%
X3: 873.48
DF: 1
/7-value: <0.0001
(B) Developed LUC
2000
2010
Exurban low
39.28%
39.11%
Exurban high
25.67%
26.44%
Suburban
11.10%
10.70%
Urban low
16.86%
16.27%
Urban high
1.23%
1.60%
Commercial
3.58%
3.76%
Industrial
2.28%
2.12%
X3: 47.74
DF: 8
/7-value: <0.0001
A-2
-------
1 60
1 40
.2 1 20
i
o 1 oo
0 80
0 60
(A)
i
l T
t . }
Exurban Low Exurban High Suburban Urban Low Urban High Commercial Industrial
1.50
1.40
1.20
1.00
0.80
0.50
(B)
(C)
i
i
f I
I 1
Exurban High vs. Suburban vs. Exurban Urban Low vs. Urban High vs. Urban Developed vs.
Exurban Low High Suburban Low Undeveloped
Figure A-l. Land use comparisons between 2000 and 2010 in Integrated
Climate and Land Use Scenarios (ICLUS) Region 1 (Pacific). (A) Odds ratios
(ORs) and confidence intervals comparing allocations among the seven developed
LUCs; (B) ORs and confidence intervals comparing adjacent residential LUCs
(high density versus low density); and (C) OR comparing developed and
undeveloped LUCs.
A.2. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 2
(INTERMOUNTAIN WEST) LAND USE CHANGES
In the Intermountain West region (ICLUS Region 2), the percentage of land assigned to
developed use classes increased between 2000 and 2010 (see Table A-2, A, Figure A-2, C).
A-3
-------
Over the same period, the relative amount of land assigned to each of the seven developed LUCs
also changed (see Table A-2, B). Among the developed classes, the proportion of developed
land in the exurban high, urban low, and industrial LUC decreased, while the proportion of
developed land in the exurban low and urban high LUCs increased between 2000 and 2010
(see Figure A-2, A). The relative amount of developed land in the suburban and commercial
LUCs did not change significantly between 2000 and 2010. Relative growth in the urban high
LUC was larger than the urban low LUC (see Figure A-2, B). However, relative growth in the
exurban high LUC was less than the exurban low LUC. The relative amount of growth in the
suburban LUC was not significantly different from the exurban high LUC, and the relative
amount of growth in the urban low LUC was not significantly different from the suburban LUC.
Table A-2. Goodness-of-fit test results comparing LUCs in 2000 and 2010 in
Integrated Climate and Land Use Scenarios (ICLUS) Region 2
(Intermountain West). Values are limited to developable area and LUCs that
transition in the model. (A) Land assigned to developed and undeveloped LUCs.
(B) Percentage developed land assigned to the seven developed LUCs.
(A) Land Use Type
2000
2010
Developed
3.41%
4.53%
Undeveloped
96.59%
95.47%
X2' 1,557.17
DF: 1
/7-value: <0.0001
(B) Developed LUC
2000
2010
Exurban low
37.39%
40.62%
Exurban high
29.45%
27.92%
Suburban
12.05%
11.67%
Urban low
13.86%
13.01%
Urban high
0.58%
0.77%
Commercial
4.39%
4.07%
Industrial
2.28%
1.94%
X2: 99.84
DF: 8
/7-value: <0.0001
A-4
-------
1 80
1 60
1 40
1 20
1 00
0 80
0 60
(A)
i
»
I
* i I i I
I
Exurban Low Exurban High Suburban Urban Low Urban High Commercial Industrial
2.00
1.80
1.50
1.40
1.20
1.00
0.80
0.50
(B)
(C)
J
i
I
Exurban High vs. Suburban vs. Exurban Urban Low vs. Urban High vs. Urban Developed vs.
Exurban Low High Suburban Low Undeveloped
Figure A-2. Land use comparisons between 2000 and 2010 in ICLUS Region
2 (Intermountain West). (A) Odds ratios (ORs) and confidence intervals
comparing allocations among the seven developed LUCs; (B) ORs and
confidence intervals comparing adjacent residential LUCs (high density versus
low density); and (C) OR comparing developed and undeveloped LUCs.
A.3. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 3
(NORTH CENTRAL) LAND USE CHANGES
In the North Central region (ICLUS Region 3), the percentage of land assigned to
developed use classes increased between 2000 and 2010 (%2 = 1,507.45, DF = \,p < 0.0001;
see Table A-3, A, Figure A-3, C). Over the same period, the relative amount of land assigned to
A-5
-------
each ofthe seven developed LUCs also changed (%2 = 149.09, DF = 8,p < 0.0001;
see Table A-3, B). Among the developed classes, the proportion of developed land in the
exurban high, suburban, urban low, and industrial LUC decreased, while the proportion of
developed land in the exurban low and urban high LUCs increased between 2000 and 2010
(see Figure A-3, A). The relative amount of developed land in the commercial LUC did not
change significantly for the same period. Relative growth in the urban high LUC was larger than
the urban low LUC (see Figure A-3, B). Conversely, relative growth in the exurban high LUC
was less than the exurban low LUC. The relative amount of growth in the suburban LUC was
not significantly different than the exurban high LUC, and the relative amount of growth in the
urban low LUC was not significantly different than the suburban LUC.
Table A-3. Goodness-of-fit test results comparing LUCs in 2000 and 2010 in
Integrated Climate and Land Use Scenarios (ICLUS) Region 3 (North
Central). Values are limited to developable area and LUCs that transition in the
model. (A) Land assigned to developed and undeveloped LUCs. (B) Percentage
developed land assigned to the seven developed LUCs.
(A) Land Use Type
2000
2010
Developed
4.05%
5.10%
Undeveloped
95.95%
94.90%
X2: 1,507.45
DF: 1
/7-value: <0.0001
(B) Developed LUC
2000
2010
Exurban low
47.01%
50.55%
Exurban high
27.72%
25.51%
Suburban
9.17%
8.61%
Urban low
9.93%
9.49%
Urban high
0.24%
0.32%
Commercial
3.62%
3.47%
Industrial
2.31%
2.05%
X2: 149.09
DF: 8
/7-value: <0.0001
A-6
-------
2 00
1 80
1 60
1 40
I,
20
(A)
i
»
i * t {
1 00
0 80
0 60
Exurban Low Exurban High Suburban Urban Low Urban High Commercial Industrial
2.00
1.80
1.50
1.40
1.20
1.00
0.80
0.50
(B)
(C)
i
I 1
i
Exurban High vs. Suburban vs. Exurban Urban Low vs. Urban High vs. Urban Developed vs.
Exurban Low High Suburban Low Undeveloped
Figure A-3. Land use comparisons between 2000 and 2010 in ICLUS Region
3 (North Central). (A) Odds ratios (ORs) and confidence intervals comparing
allocations among the seven developed LUCs; (B) ORs and confidence intervals
comparing adjacent residential LUCs (high density versus low density); and (C)
OR comparing developed and undeveloped LUCs.
A.4. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 4
(SOUTH CENTRAL) LAND USE CHANGES
In the South Central region (ICLUS Region 4), the percentage of land assigned to
developed use classes increased between 2000 and 2010 (see Table A-4, A, Figure A-4, C).
Over the same period, the relative amount of land assigned to each of the seven developed LUCs
A-7
-------
also changed (see Table A-4, B). In this particular region, the amount of developed land
allocated to the exurban low and exurban high LUCs was lower in 2000 than expected
(see Table A-4, B), and a large number of grazing land use pixels transitioned into these LUCs in
2010. However, values for the exurban low and exurban high LUCs were comparable to other
regions in 2010, which suggests the model had difficulty distinguishing between exurban and
agricultural classes in 2000. As a result, comparisons among the LUCs below are not
particularly meaningful, but are presented for completeness. Among the developed classes, the
proportion of developed land in the exurban high, suburban, urban low, urban high, commercial,
and industrial LUCs decreased, while the proportion of developed land in the exurban low LUC
increased between 2000 and 2010 (see Figure A-4, A). Relative growth in the exurban high
LUC was less than the exurban low LUC, relative growth in the suburban LUC was less than the
exurban high LUC, and relative growth in the urban low LUC was less than the suburban LUC
(see Figure A-4, B). The relative amount of growth in the urban high LUC was not significantly
different than the urban low LUC.
A-8
-------
Table A-4. Goodness-of-fit test results comparing LUCs in 2000 and 2010 in
Integrated Climate and Land Use Scenarios (ICLUS) Region 4 (South
Central). Values are limited to developable area and LUCs that transition in the
model. (A) Land assigned to developed and undeveloped LUCs. (B) Percentage
developed land assigned to the seven developed LUCs.
(A) Land Use Type
2000
2010
Developed
3.90%
11.52%
Undeveloped
96.10%
88.48%
X2: 41,129.98
DF: 1
/7-value: <0.0001
(B) Developed LUC
2000
2010
Exurban low
19.28%
54.43%
Exurban high
31.26%
25.62%
Suburban
16.43%
8.07%
Urban low
19.47%
7.19%
Urban high
0.95%
0.34%
Commercial
8.04%
2.78%
Industrial
4.56%
1.57%
X2: 17,949.23
DF: 8
/7-value: <0.0001
A-9
-------
(A)
Exurban Low Exurban High Suburban Urban Low Urban High Commercial Industrial
(B)
(C) *
T
t
m
•
•
Exurban High vs. Suburban vs. Exurban Urban Low vs. Urban High vs. Urban Developed vs.
Exurban Low High Suburban Low Undeveloped
Figure A-4. Land use comparisons between 2000 and 2010 in Integrated
Climate and Land Use Scenarios (ICLUS) Region 4 (South Central).
(A) Odds ratios (ORs) and confidence intervals comparing allocations among the
seven developed LUCs; (B) ORs and confidence intervals comparing adjacent
residential LUCs (high density versus low density); and (C) OR comparing
developed and undeveloped LUCs.
A.5. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 5
(GREAT LAKES) LAND USE CHANGES
In the Great Lakes region (ICLUS Region 5), the percentage of land assigned to
developed use classes increased between 2000 and 2010 (see Table A-5, A, Figure A-5, C).
A-10
-------
Over the same period, the relative amount of land assigned to each of the seven developed LUCs
also changed (see Table A-5, B). Among the developed classes, the proportion of developed
land in the exurban high and urban high LUCs increased, while the proportion of developed land
in the exurban low LUC decreased between 2000 and 2010 (see Figure A-5, A). The relative
amount of developed land in the suburban, urban low, commercial, and industrial LUCs did not
change significantly. Relative growth in the exurban high LUC was larger than in the exurban
low LUC, and relative growth in the urban high LUC was larger than the urban low LUC (see
Figure A-5, B). The relative amount of growth in the suburban LUC was not significantly
different than the exurban high LUC, and the relative amount of growth in the urban low LUC
was not significantly different than the suburban LUC.
Table A-5. Goodness-of-fit test results comparing LUCs in 2000 and 2010 in
Integrated Climate and Land Use Scenarios (ICLUS) Region 5 (Great
Lakes). Values are limited to developable area and LUCs that transition in the
model. (A) Land assigned to developed and undeveloped LUCs. (B) Percentage
developed land assigned to the seven developed LUCs.
(A) Land Use Type
2000
2010
Developed
20.12%
23.99%
Undeveloped
79.88%
76.01%
X2: 2,329.40
DF: 1
/>-value:<0.0001
(B) Developed LUC
2000
2010
Exurban low
53.02%
52.07%
Exurban high
25.52%
26.40%
Suburban
8.30%
8.31%
Urban low
9.04%
9.02%
Urban high
0.35%
0.47%
Commercial
2.22%
2.28%
Industrial
1.55%
1.44%
X2: 55.17
DF: 8
/7-value: <0.0001
A-11
-------
1 80
1 60
1 40
1 20
1 00
0 80
0 60
(A)
i
' I
t
1 1 1 1
Exurban Low Exurban High Suburban Urban Low Urban High Commercial Industrial
1.80
1.50
1.40
1.20
1.00
0.80
0.50
(B)
(C)
•
¦
• T
4 I
Exurban High vs. Suburban vs. Exurban Urban Low vs. Urban High vs. Urban Developed vs.
Exurban Low High Suburban Low Undeveloped
Figure A-5. Land use comparisons between 2000 and 2010 in ICLUS Region
5 (Great Lakes). (A) Odds ratios (ORs) and confidence intervals comparing
allocations among the seven developed LUCs; (B) ORs and confidence intervals
comparing adjacent residential LUCs (high density versus low density); and (C)
OR comparing developed and undeveloped LUCs.
A.6. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 6
(SOUTHEAST) LAND USE CHANGES
In the Southeast region (ICLUS Region 6), the percentage of land assigned to developed
use classes increased between 2000 and 2010 (see Table A-6, A, Figure A-6, C). Over the same
period, the relative amount of land assigned to each of the seven developed LUCs also changed
A-12
-------
(see Table A-6, B). Among the developed classes, the proportion of developed land in the
exurban low LUC decreased, while the proportion of developed land in the exurban high,
suburban and urban low, urban high and commercial LUCs increased between 2000 and 2010
(see Figure A-6, A). The relative amount of developed land in the industrial LUC did not change
significantly. Relative growth in all of the LUC comparisons were greater in 2010 than in 2000
(see Figure A-6, B).
Table A-6. Goodness-of-fit test results comparing LUCs in 2000 and 2010 in
Integrated Climate and Land Use Scenarios (ICLUS) Region 6 (Southeast).
Values are limited to developable area and LUCs that transition in the model. (A)
Land assigned to developed and undeveloped LUCs. (B) Percentage developed
land assigned to the seven developed LUCs.
(A) Land Use Type
2000
2010
Developed
27.38%
34.17%
Undeveloped
72.62%
65.83%
X3: 10,532.23
DF: 1
/7-value: <0.0001
(B) Developed LUC
2000
2010
Exurban low
61.97%
57.74%
Exurban high
24.18%
25.28%
Suburban
7.55%
9.22%
Urban low
3.87%
5.04%
Urban high
0.17%
0.28%
Commercial
1.47%
1.64%
Industrial
0.79%
0.81%
X2: 1,562.88
DF: 8
/7-value: <0.0001
A-13
-------
2 00
1 80
1 60
1 40
(A)
I,
20
1 00
0 80
0 60
Exurban Low Exurban High Suburban Urban Low Urban High Commercial Industrial
1.50
1.40
1.20
1.00
0.80
0.50
(B)
(C)
<
i
•
' 5 s
Exurban High vs. Suburban vs. Exurban Urban Low vs. Urban High vs. Urban Developed vs.
Exurban Low High Suburban Low Undeveloped
Figure A-6. Land use comparisons between 2000 and 2010 in Integrated
Climate and Land Use Scenarios (ICLUS) Region 6 (Southeast). (A) Odds
ratios (ORs) and confidence intervals comparing allocations among the seven
developed LUCs; (B) ORs and confidence intervals comparing adjacent
residential LUCs (high density versus low density); and (C) OR comparing
developed and undeveloped LUCs.
A.7. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 7
(NORTHEAST) LAND USE CHANGES
In the Northeast region (ICLUS Region 7), the percentage of land assigned to developed
use classes increased between 2000 and 2010 (see Table A-7, A, Figure A-7, C). Over the same
A-14
-------
period, the relative amount of land assigned to each of the seven developed LUCs also changed
(see Table A-7, B). Among the developed classes, the proportion of developed land in the
exurban low LUC decreased, while the proportion of developed land in the exurban high,
suburban, urban low, urban high, and commercial LUCs increased (see Figure 13, A). The
relative amount of developed land in the industrial LUC did not change significantly between
2000 and 2010 (see Figure A-7, A). Relative growth in all of the LUC comparisons was greater
in 2010 than in 2000, except for the urban low LUC, which was not significantly different from
the suburban LUC (see Figure A-7, B).
Table A-7. Goodness-of-fit test results comparing LUCs in 2000 and 2010 in
Integrated Climate and Land Use Scenarios (ICLUS) Region 7 (Northeast).
Values are limited to developable area and LUCs that transition in the model. (A)
Land assigned to developed and undeveloped LUCs. (B) Percentage developed
land assigned to the seven developed LUCs.
(A) Land Use Type
2000
2010
Developed
41.02%
46.97%
Undeveloped
58.98%
53.03%
X2: 2,248.51
DF: 1
/7-value: <0.0001
(B) Developed LUC
2000
2010
Exurban low
56.44%
54.21%
Exurban high
27.18%
27.90%
Suburban
8.33%
9.16%
Urban low
5.54%
5.89%
Urban high
0.66%
0.81%
Commercial
1.22%
1.37%
Industrial
0.64%
0.65%
X2: 178.10
DF: 8
/7-value: <0.0001
A-15
-------
1 60
1 40
1 20
1 00
0 80
0 60
1.40
1.20
1.00
0.80
0.50
(A)
i T
i
$
I
!
i
i
Exurban Low
Exurban High Suburban
Urban Low
Urban High Commercial
Industrial
(B)
(C)
Exurban High vs. Suburban vs. Exurban Urban Low vs. Urban High vs. Urban Developed vs.
Exurban Low High Suburban Low Undeveloped
Figure A-7. Land use comparisons between 2000 and 2010 in Integrated
Climate and Land Use Scenarios (ICLUS) Region 7 (Northeast). (A) Odds
ratios (ORs) and confidence intervals comparing allocations among the seven
developed LUCs; (B) ORs and confidence intervals comparing adjacent
residential LUCs (high density versus low density); and (C) OR comparing
developed and undeveloped LUCs.
A-16
-------
APPENDIX B. REGIONAL TRANSITION PROBABILITY MODELS
B.l. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 1
(PACIFIC) TRANSITION PROBABILITY MODELS
Table B-l. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial Generalized Additive Models
(GAMs). The top marginal model predicts the probability of transitioning into
each land use class (LUC)j in 2010, /?(LUCj), by capacity class, while the bottom
conditional models predict the probability of transitioning from each LUCi in
2000 given that they transitioned into a particular LUCj in 2010, /?(LUCiy), by
capacity class. RefLevel is the reference level land use from which transitions are
calculated.
Smoothing T erms
edf
x2
P
For Transitions into LUC2010 by Capacity Class (RefLevel: Exurban Low)
Capacity class (exurban high)
1.87
10,633.33
<0.0001
Capacity class (suburban)
1.73
5,543.97
<0.0001
Capacity class (urban low)
1.66
9,450.96
<0.0001
Capacity class (urban high)
1.48
855.19
<0.0001
Capacity class (commercial)
1.83
29,894.84
<0.0001
Capacity class (industrial)
1.97
604.42
<0.0001
Global test
16.55
1,463,202
<0.0001
From LUC2000 for Transitions into Exurban Low by Capacity Class (RefLevel: Grazing)
Capacity class (timber)
2.03
411.02
<0.0001
Capacity class (pasture)
2.01
2,039.43
<0.0001
Capacity class (cropland)
2.02
510.69
<0.0001
Global test
9.07
9,242.02
<0.0001
From LUC2000 for Transitions into Exurban High by Capacity Class (RefLevel: Exurban
Low)
Capacity class (timber)
2.00
221.42
<0.0001
Capacity class (grazing)
1.98
10,906.72
<0.0001
Capacity class (pasture)
2.03
215.47
<0.0001
Capacity class (cropland)
2.08
93.19
<0.0001
Global test
12.09
17,273.73
<0.0001
B-l
-------
Table B-l. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUCj in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUCi in 2000 given that they transitioned into a
particular LUCj in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing Terms
edf
x2
P
From LUC2000 for Transitions into Suburban by Capacity Class (RefLevel: Exurban
High)
Capacity class (wetlands)
1.94
11.08
0.0036
Capacity class (timber)
2.19
76.11
<0.0001
Capacity class (grazing)
2.12
85.46
<0.0001
Capacity class (pasture)
2.23
85.63
<0.0001
Capacity class (cropland)
2.37
363.75
<0.0001
Capacity class (exurban low)
2.07
594.51
<0.0001
Global test
18.91
15,207.84
<0.0001
From LUC2000 for transitions into Urban Low by Capacity Class (RefLevel: Suburban)
Capacity class (wetlands)
1.74
5.59
0.0469
Capacity class (timber)
2.24
86.71
<0.0001
Capacity class (grazing)
2.05
221.43
<0.0001
Capacity class (pasture)
2.39
44.67
<0.0001
Capacity class (cropland)
2.46
278.04
<0.0001
Capacity class (exurban low)
2.18
490.91
<0.0001
Capacity class (exurban high)
1.98
1,315.07
<0.0001
Global test
22.03
10,766.11
<0.0001
B-2
-------
Table B-l. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUCj in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUCi in 2000 given that they transitioned into a
particular LUCj in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing Terms
edf
x2
P
From LUC2000 for Transitions into Urban High by Capacity Class (RefLevel: Urban
Low)
Capacity class (wetlands)
1.42
2.36
0.1966
Capacity class (timber)
1.81
5.66
0.0486
Capacity class (grazing)
1.78
11.73
0.0021
Capacity class (pasture)
1.1
4.43
0.0409
Capacity class (cropland)
1.95
8.8
0.0115
Capacity class (exurban low)
1.98
25.38
<0.0001
Capacity class (exurban high)
1.83
103.49
<0.0001
Capacity class (suburban)
2.00
119.96
<0.0001
Global test
21.84
2,022.44
<0.0001
From LUC2000 for Transitions into Commercial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
2.07
270.23
<0.0001
Capacity class (suburban)
1.74
272.1
<0.0001
Capacity class (urban low)
1.73
378.23
<0.0001
Capacity class (urban high)
1.61
6.07
0.0317
Global test
11.14
20,068.51
<0.0001
From LUC2000 for Transitions into Industrial by Capacity Class (RefLevel: Exurban
High)
Capacity class (grazing)
0.72
5.79
0.0096
Capacity class (exurban low)
2.06
16.7
3.00 x 10"4
Global test
4.79
1,078.55
<0.0001
B-3
-------
Rbq ton 1J Combined Modelc
"francfttonc, from LUC2000 to Exurtiin Law
«•
o
'£•
O
o
o
o
Region 1Combined Module
"n-anctttor* from LUC20DD to Exurtoan HVgft
«•
o
o
¦4
o
0*
o
o
o
CaoacTj- Cass
Capacity Ca::
Reg ton 1Combined tiodele
TranchBont from LUC200Q to Suburban
«•
o
O
-------
Region 1 / Combined Models
Transitions from LUC2000 to Industrial
E*urb3n Lew E*urb3n High
Capacity Class
FigureB-l. Predicted transition probabilities by capacity class from LUCs
in 2000 to LUCs in 2010 in Integrated Climate and Land Use Scenarios
(ICLUS) Region 1 (Pacific). Each panel shows transitions into a particular LUC
in 2010. These combined probabilities are the product of corresponding marginal
and conditional models (i.e., for a given capacity class the probability of
transitioning from LUCi into LUCj is P(LUCij) = P(LUCj) X P(LUCijj).
B-5
-------
B.2. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 2
(INTERMOUNTAIN WEST) TRANSITION PROBABILITY MODELS
Table B-2. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a particular
LUC in 2010,/?(LUCiy), by capacity class. RefLevel is the reference level land
use from which transitions are calculated.
Smoothing Terms
edf
x2
P
For Transitions into LUC2010 by Capacity Class (RefLevel: Exurban Low)
Capacity class (exurban high)
1.89
3,226.33
<0.0001
Capacity class (suburban)
1.8
11,531.99
<0.0001
Capacity class (urban low)
1.7
14,389.11
<0.0001
Capacity class (urban high)
1.78
3,477.46
<0.0001
Capacity class (commercial)
1.9
35,289.11
<0.0001
Capacity class (industrial)
1.91
1,313.18
<0.0001
Global test
16.98
1,050,212
<0.0001
From LUC2000 for Transitions into Exurban Low by Capacity Class (RefLevel: Grazing)
Capacity class (timber)
1.94
1,030.15
<0.0001
Capacity class (pasture)
2
2,895.73
<0.0001
Capacity class (cropland)
2.01
1,152.97
<0.0001
Global test
8.95
5,582.88
<0.0001
From LUC2000 for Transitions into Exurban High by Capacity Class (RefLevel: Exurban
Low)
Capacity class (timber)
2.13
492.9
<0.0001
Capacity class (grazing)
1.96
3,483.53
<0.0001
Capacity class (pasture)
2
94.44
<0.0001
Capacity class (cropland)
2.04
180.76
<0.0001
Global test
12.13
21,938.08
<0.0001
B-6
-------
Table B-2. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing T erms
edf
x2
P
From LUC2000 for Transitions into Suburban by Capacity Class (RefLevel: Exurban
High)
Capacity class (wetlands)
1.91
12.11
0.0021
Capacity class (timber)
2.29
139.88
<0.0001
Capacity class (grazing)
2.02
39.53
<0.0001
Capacity class (pasture)
2.24
55.39
<0.0001
Capacity class (cropland)
2.33
168.36
<0.0001
Capacity class (exurban low)
2.02
481.68
<0.0001
Global test
18.81
16,786.93
<0.0001
From LUC2000 for Transitions into Urban Low by Capacity Class (RefLevel: Suburban)
Capacity class (wetlands)
1.85
11.91
0.0021
Capacity class (timber)
2.22
46.77
<0.0001
Capacity class (grazing)
2.12
912.79
<0.0001
Capacity class (pasture)
2.27
87.06
<0.0001
Capacity class (cropland)
2.29
454.97
<0.0001
Capacity class (exurban low)
2.21
619.65
<0.0001
Capacity class (exurban high)
1.99
1,638.98
<0.0001
Global test
21.96
12,595.07
<0.0001
From LUC2000 for Transitions into Urban High by Capacity Class (RefLevel: Urban
Low)
Capacity class (wetlands)
1.05
3.04
0.0874
Capacity class (timber)
1.26
6.91
0.0109
Capacity class (grazing)
2.07
219.32
<0.0001
Capacity class (pasture)
2.05
9.09
0.0113
Capacity class (cropland)
1.51
11.57
0.0016
Capacity class (exurban low)
2.23
56.06
<0.0001
B-7
-------
Table B-2. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing T erms
edf
x2
P
Capacity class (exurban high)
2.04
200.89
<0.0001
Capacity class (suburban)
1.88
83.19
<0.0001
Global test
22.09
1,850.01
<0.0001
From LUC2000 for Transitions into Commercial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
2.01
195.61
<0.0001
Capacity class (suburban)
1.95
521.53
<0.0001
Capacity class (urban low)
1.96
272.46
<0.0001
Capacity class (urban high)
1.70
5.78
0.013
Global test
11.62
11,500.45
<0.0001
From LUC2000 for Transitions into Industrial by Capacity Class (RefLevel: Exurban
High)
Capacity class (wetlands)
1.15
24.53
<0.0001
Capacity class (grazing)
1.48
20.35
<0.0001
Capacity class (pasture)
0.64
1.39
0.143
Capacity class (exurban low)
2.08
36.75
<0.0001
Global test
9.36
993.61
<0.0001
B-8
-------
Region 2 J Combined Hodelc
Tra ncltkmc from LUC2D0Q ia Exurbsn Low
Region 2.* Combined Mode-ic
Tranclttorw from LUC2DDD to Ex urban High
03
o
«>
o
o
Ch
c
o
o
Capacity C ass
Capacity Ca:;
Reg ion 2 J Combined Uodelc
Transitions from LUC2DQQ 1o Suburban
Region 2.' Combined ModeiB
Transitions frcm LUC2D0Q to Urban Low
w
o
•b
O
C-J
o
o
o
CapacTj' Class
Capadly Glass
Region 2 J1 Combined Hodels
Transitions from LUC20Q* to Urban High
Region 2.' Combined Modeic
Transitions frcm LUC20DD to Commercial
o
o
•£
O
O'
o
o
Capacir/ Class
Capacity Class
B-9
-------
Region 2 I Combined Models
Transitions from LUC2000 to Industrial
Grazing Exurban high
Exurb3n Lew
Capacity Class
FigureB-2. Predicted transition probabilities by capacity class from LUCs
in 2000 to LUCs in 2010 in Integrated Climate and Land Use Scenarios
(ICLUS) Region 2 (Intermountain West). Each panel shows transitions into a
particular LUC in 2010. These combined probabilities are the product of
corresponding marginal and conditional models, i.e., for a given capacity class the
probability of transitioning from LUCi into LUCj is P(LUCij) = P(LUCj) X
P(LUCjy).
B-10
-------
B.3. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 3
(NORTH CENTRAL) TRANSITION PROBABILITY MODELS
Table B-3. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a particular
LUC in 2010,/?(LUCiy), by capacity class. RefLevel is the reference level land
use from which transitions are calculated.
Smoothing Terms
edf
x2
P
For Transitions into LUC2010 by Capacity Class (RefLevel: Exurban Low)
Capacity class (exurban high)
1.86
4,754.47
<0.0001
Capacity class (suburban)
1.8
16,711.47
<0.0001
Capacity class (urban low)
1.74
13,319.25
<0.0001
Capacity class (urban high)
2.18
22,346.04
<0.0001
Capacity class (commercial)
1.85
9,804.43
<0.0001
Capacity class (industrial)
1.94
1,802.87
<0.0001
Global test
17.37
1,058,696
<0.0001
From LUC2000 for Transitions into Exurban Low by Capacity Class (RefLevel: Grazing)
Capacity class (timber)
2.02
22.43
<0.0001
Capacity class (pasture)
2.01
3,718.37
<0.0001
Capacity class (cropland)
2.01
237.31
<0.0001
Global test
9.04
24,725.45
<0.0001
From LUC2000 for Transitions into Exurban High by Capacity Class (RefLevel: Exurban
Low)
Capacity class (timber)
2.08
157.92
<0.0001
Capacity class (grazing)
1.96
4,738.75
<0.0001
Capacity class (pasture)
1.99
225.36
<0.0001
Capacity class (cropland)
1.99
801.72
<0.0001
Global test
12.02
9,249.58
<0.0001
B-ll
-------
Table B-3. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing T erms
edf
x2
P
From LUC2000 for Transitions into Suburban by Capacity Class (RefLevel: Exurban
High)
Capacity class (wetlands)
1.97
123.14
<0.0001
Capacity class (timber)
2.05
14.99
0.0006
Capacity class (grazing)
2.04
101.77
<0.0001
Capacity class (pasture)
2.11
216.86
<0.0001
Capacity class (cropland)
2.16
262
<0.0001
Capacity class (exurban low)
2.1
495.46
<0.0001
Global test
18.43
19,567.29
<0.0001
From LUC2000 for Transitions into Urban Low by Ca
>acity Class (RefLevel: Suburban)
Capacity class (wetlands)
1.74
22.39
<0.0001
Capacity class (timber)
1.87
3
0.2027
Capacity class (grazing)
2.07
214.66
<0.0001
Capacity class (pasture)
2.07
160.27
<0.0001
Capacity class (cropland)
2.2
36.16
<0.0001
Capacity class (exurban low)
2.2
282.27
<0.0001
Capacity class (exurban high)
2.03
955.3
<0.0001
Global test
21.17
8,447.26
<0.0001
From LUC2000 for Transitions into Urban High by Capacity Class (RefLevel: Urban
Low)
Capacity class (wetlands)
1.15
4.86
0.0343
Capacity class (grazing)
2.29
40.23
<0.0001
Capacity class (pasture)
1.74
11.91
0.0018
Capacity class (cropland)
0.97
5.85
0.0149
Capacity class (exurban low)
2.02
39.67
<0.0001
Capacity class (exurban high)
2.03
139.47
<0.0001
B-12
-------
Table B-3. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing T erms
edf
x2
P
Capacity class (suburban)
1.85
163.45
<0.0001
Global test
19.05
541.05
<0.0001
From LUC2000 for Transitions into Commercial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
1.91
319.34
<0.0001
Capacity class (suburban)
1.91
1,006.82
<0.0001
Capacity class (urban low)
2.12
1,607.49
<0.0001
Capacity class (urban high)
2.19
674.84
<0.0001
Global test
12.13
9,102.24
<0.0001
From LUC2000 for Transitions into Industrial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
2.21
31.74
<0.0001
Global test
3.21
37.65
<0.0001
B-13
-------
RpQtanS • Comtilnpd Modplc
"Il-anc rttoric from LUC2D00 1o E turban Law
Region i; Combined Mode-lc
Trancfttor* from LUC2DDD to Ex urb an H kgh
Cass
Reg ton 4 ' Comiblr>&d Uodelc
TranclHonc from LUCSOOOlo Urt-an High
Region 4.' Combined Hodelc
Tranctttonc frcm LUC200D to Corrmesxilail
-------
Region 3 / Combined Models
Transitions from LUC2000 to Industrial
E*urb3n Lew E*urb3n High
Capacity Class
Figure B-3. Predicted transition probabilities by capacity class from LUCs
in 2000 to LUCs in 2010 in Integrated Climate and Land Use Scenarios
(ICLUS) Region 3 (North Central). Each panel shows transitions into a
particular LUC in 2010. These combined probabilities are the product of
corresponding marginal and conditional models, i.e., for a given capacity class the
probability of transitioning from LUCi into LUCj is P(LUCij) = P(LUCj) X
P(LUCjy).
B-15
-------
B.4. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 4
(SOUTH CENTRAL) TRANSITION PROBABILITY MODELS
Table B-4. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a particular
LUC in 2010,/?(LUCiy), by capacity class. RefLevel is the reference level land
use from which transitions are calculated.
Smoothing Terms
edf
x2
P
For Transitions into LUC2010 by Capacity Class (RefLevel: Exurban Low)
Capacity class (exurban high)
1.93
1,729.42
<0.0001
Capacity class (suburban)
1.95
12,232.72
<0.0001
Capacity class (urban low)
1.89
7,131.93
<0.0001
Capacity class (urban high)
2.06
498.58
<0.0001
Capacity class (commercial)
1.90
3,650.58
<0.0001
Capacity class (industrial)
1.87
543.53
<0.0001
Global test
17.60
1,912,574
<0.0001
From LUC2000 for Transitions into Exurban Low by Capacity Class (RefLevel: Grazing)
Capacity class (timber)
1.99
341.64
<0.0001
Capacity class (pasture)
2.01
6,979.23
<0.0001
Capacity class (cropland)
2.00
277.28
<0.0001
Global test
9.00
50,282.78
<0.0001
From LUC2000 for Transitions into Exurban High by Capacity Class (RefLevel: Grazing)
Capacity class (timber)
2.03
499.96
<0.0001
Capacity class (pasture)
2
1,659.84
<0.0001
Capacity class (cropland)
2
261.75
<0.0001
Capacity class (exurban low)
1.99
806.3
<0.0001
Global test
12.03
19,077.08
<0.0001
B-16
-------
Table B-4. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing T erms
edf
x2
P
From LUC2000 for Transitions into Suburban by Capacity Class (RefLevel: Grazing)
Capacity class (wetlands)
2.07
190.8
<0.0001
Capacity class (timber)
2.2
236.43
<0.0001
Capacity class (pasture)
2.07
462.93
<0.0001
Capacity class (cropland)
2.1
224.46
<0.0001
Capacity class (exurban low)
2.04
22.22
<0.0001
Capacity class (exurban high)
1.91
302.98
<0.0001
Global test
18.39
5,044.57
<0.0001
From LUC2000 for Transitions into Urban Low by Capacity Class (RefLevel: Grazing)
Capacity class (wetlands)
1.91
3.03
0.2059
Capacity class (timber)
2.15
20.35
<0.0001
Capacity class (pasture)
2.18
20.45
<0.0001
Capacity class (cropland)
2.28
21.35
<0.0001
Capacity class (exurban low)
2.1
5.58
0.0676
Capacity class (exurban high)
1.94
19.43
0.0001
Capacity class (suburban)
1.95
545.75
<0.0001
Global test
21.51
5,170.4
<0.0001
From LUC2000 for Transitions into Urban High by Capacity Class (RefLevel: Urban
Low)
Capacity class (wetlands)
1.09
1.23
0.2935
Capacity class (timber)
1.98
1.1
0.572
Capacity class (grazing)
1.85
26.12
<0.0001
Capacity class (pasture)
2.37
5.1
0.107
Capacity class (cropland)
2.04
6.98
0.0317
Capacity class (exurban low)
1.68
2.26
0.2599
B-17
-------
Table B-4. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing T erms
edf
x2
P
Capacity class (exurban high)
1.84
3.09
0.1888
Capacity class (suburban)
1.85
1.45
0.4467
Global test
22.7
352.8
<0.0001
From LUC2000 for Transitions into Commercial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
2.15
32.86
<0.0001
Capacity class (suburban)
1.99
147.97
<0.0001
Capacity class (urban low)
2.06
60.21
<0.0001
Capacity class (urban high)
0.68
2.48
0.0706
Global test
10.88
1,823.53
<0.0001
From LUC2000 for Transitions into Industrial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
1.99
25.82
<0.0001
Global test
2.99
185.08
<0.0001
B-18
-------
R»Q ton4 • Combined Modelc
Trancttionc from LUC2P00 1a Exurbwi Law
trav —
—
4. 5 6
Capacity Ciass
Region 4/ Combined Modelc
Tranctttor* from LUC2DD0 to Ex urban High
PC
a
'£
O
CM
O
o
o
Capacity Class
Red or 4Combined Modelc
TranGlfionc. from LUC2D0Q 1o Suburban
unft-i ngr
era*
¦ Tuir*iHgr'
uiw.
liW.Ui
TO
O'
•€
O
O
o
"3
Cacao Cass
Cap at try Cass
B-19
-------
£
Region 41 Combined Models
Transitions from LUC2000 to Industrial
Eurtan Low Eurtan high
Capacity Class
FigureB-4. Predicted transition probabilities by capacity class from LUCs
in 2000 to LUCs in 2010 in Integrated Climate and Land Use Scenarios
(ICLUS) Region 4 (South Central). Each panel shows transitions into a
particular LUC in 2010. These combined probabilities are the product of
coiresponding marginal and conditional models, i.e., for a given capacity class the
probability of transitioning from LUC, into LUQ is P(LUCy) = P(LUQ) x
P(LUClb).
B-20
-------
B.5. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 5
(GREAT LAKES) TRANSITION PROBABILITY MODELS
Table B-5. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a particular
LUC in 2010,/?(LUCiy), by capacity class. RefLevel is the reference level land
use from which transitions are calculated.
Smoothing Terms
edf
x2
P
For Transitions into LUC2010 by Capacity Class (RefLevel: Exurban Low)
Capacity class (exurban high)
1.82
5,643.06
<0.0001
Capacity class (suburban)
1.74
3,016.31
<0.0001
Capacity class (urban low)
1.71
3,105.97
<0.0001
Capacity class (urban high)
1.72
568.65
<0.0001
Capacity class (commercial)
1.77
6,819.62
<0.0001
Capacity class (industrial)
1.98
125.28
<0.0001
Global test
16.74
1,646,923
<0.0001
From LUC2000 for Transitions into Exurban Low by Capacity Class (RefLevel: Grazing)
Capacity class (timber)
2.01
110.82
<0.0001
Capacity class (pasture)
2.02
5,294.72
<0.0001
Capacity class (cropland)
2.02
2,428.31
<0.0001
Global test
9.05
16,604.58
<0.0001
From LUC2000 for Transitions into Exurban High by Capacity Class (RefLevel: Exurban
Low)
Capacity class (timber)
2.22
909.1
<0.0001
Capacity class (grazing)
1.96
4,938
<0.0001
Capacity class (pasture)
2.03
99.87
<0.0001
Capacity class (cropland)
2.04
195.63
<0.0001
Global test
12.24
7,392.31
<0.0001
B-21
-------
Table B-5. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing Terms
edf
x2
P
From LUC2000 for Transitions into Suburban by Capacity Class (RefLevel: Exurban
High)
Capacity class (wetlands)
1.93
168.63
<0.0001
Capacity class (timber)
2.21
126.68
<0.0001
Capacity class (grazing)
1.98
112.65
<0.0001
Capacity class (pasture)
2.14
136.16
<0.0001
Capacity class (cropland)
2.3
299.09
<0.0001
Capacity class (exurban low)
2.14
324.83
<0.0001
Global test
18.7
14,336.52
<0.0001
From LUC2000 for Transitions into Urban Low by Ca
>acity Class (RefLevel: Suburban)
Capacity class (wetlands)
1.88
47.52
<0.0001
Capacity class (timber)
1.97
36.46
<0.0001
Capacity class (grazing)
1.95
108.77
<0.0001
Capacity class (pasture)
2.21
75.13
<0.0001
Capacity class (cropland)
2.29
77.92
<0.0001
Capacity class (exurban low)
2.1
348.34
<0.0001
Capacity class (exurban high)
1.99
741.13
<0.0001
Global test
21.4
8,664.14
<0.0001
From LUC2000 for Transitions into Urban High by Capacity Class (RefLevel: Urban
Low)
Capacity class (wetlands)
1.95
0
1
Capacity class (grazing)
2.15
8.44
0.0172
Capacity class (pasture)
1.50
3.13
0.1373
Capacity class (cropland)
1.79
5.13
0.0629
Capacity class (exurban low)
2.00
16.35
0.0003
Capacity class (exurban high)
2.05
20.07
<0.0001
B-22
-------
Table B-5. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing Terms
edf
x2
P
Capacity class (suburban)
1.94
104.79
<0.0001
Global test
20.37
1,324.68
<0.0001
From LUC2000 for Transitions into Commercial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
2.06
90.98
<0.0001
Capacity class (suburban)
1.76
36.55
<0.0001
Capacity class (urban low)
1.73
12.29
0.0015
Capacity class (urban high)
1.46
2.26
0.2156
Global test
11.01
12,256.77
<0.0001
From LUC2000 for Transitions into Industrial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
2.01
5.08
0.0799
Global test
3.01
278.67
<0.0001
B-23
-------
Regions ' Combined Uodelc
"n-ancrttonc from LUC2D0Q 1o Exurban Low
¦£"
O
•N
O
O
o
Region 5: Combined Models
Tranctttor* from LUC2D00 to Ex urban High
w
o
'£
O
Oi
o
o
o
Capacty Cm:
L-2oaGty Ca:s
Raglan € J Combined Models.
Trancltlanc from LUC20QQ 1a Suburban
Heglon 5.' Combined Models
Trancftk>r>t Item LUC2D0Q tD Urban Low
oo
o
¦o
o
CM
O
O
o
o
re
o
o
o
©
o
Cap act/ Class
Capacity C-a;s
Reasons.' Combined Models
Transitions from LUCSaae to Urban High
o
•SO
O
o
o
o
^30aery Ciaa
Raglan s; Combined Models
Translfctor* from LUC2DD0 to CornmecaJ:
o
a7
•-
•£
o
o
o
o
Capacity Class
B-24
-------
Region 5 / Combined Models
Transitions from LUC2000 to Industrial
E*urb3n Lew E*urb3n High
Capacity Class
FigureB-5. Predicted transition probabilities by capacity class from LUCs
in 2000 to LUCs in 2010 in Integrated Climate and Land Use Scenarios
(ICLUS) Region 5 (Great Lakes). Each panel shows transitions into a particular
LUC in 2010. These combined probabilities are the product of corresponding
marginal and conditional models, i.e., for a given capacity class the probability of
transitioning from LUCi into LUCj is P(LUCij) = P(LUCj) X P(LUCijj).
B-25
-------
B.6. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 6
(SOUTHEAST) TRANSITION PROBABILITY MODELS
Table B-6. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a particular
LUC in 2010,/?(LUCiy), by capacity class. RefLevel is the reference level land
use from which transitions are calculated.
Smoothing Terms
edf
x2
P
For Transitions into LUC2010 by Capacity Class (RefLevel: Exurban Low)
Capacity class (exurban high)
1.8
19,946.39
<0.0001
Capacity class (suburban)
1.63
40,462.82
<0.0001
Capacity class (urban low)
1.64
64,034.18
<0.0001
Capacity class (urban high)
1.78
27,887.66
<0.0001
Capacity class (commercial)
1.83
61,257.16
<0.0001
Capacity class (industrial)
1.93
11,358.44
<0.0001
Global test
16.62
8,232,880
<0.0001
From LUC2000 for Transitions into Exurban Low by Capacity Class (RefLevel: Grazing)
Capacity class (timber)
2.01
102.93
<0.0001
Capacity class (pasture)
2.02
2,185.1
<0.0001
Capacity class (cropland)
2.03
995.05
<0.0001
Global test
9.06
183,239.3
<0.0001
From LUC2000 for Transitions into Exurban High by Capacity Class (RefLevel: Exurban
Low)
Capacity class (timber)
2.02
325.96
<0.0001
Capacity class (grazing)
1.95
35,257.99
<0.0001
Capacity class (pasture)
2.04
676.23
<0.0001
Capacity class (cropland)
2.05
1,016.18
<0.0001
Global test
12.06
51,509.93
<0.0001
B-26
-------
Table B-6. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing Terms
edf
x2
P
From LUC2000 for Transitions into Suburban by Capacity Class (RefLevel: Exurban
High)
Capacity class (wetlands)
1.99
105.92
<0.0001
Capacity class (timber)
2.1
118.7
<0.0001
Capacity class (grazing)
1.96
3,812.02
<0.0001
Capacity class (pasture)
2.11
821.79
<0.0001
Capacity class (cropland)
2.17
931.83
<0.0001
Capacity class (exurban low)
1.99
4,745.53
<0.0001
Global test
18.32
94,724.94
<0.0001
From LUC2000 for Transitions into Urban Low by Ca
rncity Class (RefLevel: Suburban)
Capacity class (wetlands)
1.97
151.68
<0.0001
Capacity class (timber)
1.99
202.48
<0.0001
Capacity class (grazing)
2.00
1,279.8
<0.0001
Capacity class (pasture)
2.13
1,002.82
<0.0001
Capacity class (cropland)
2.21
756.81
<0.0001
Capacity class (exurban low)
2.07
6,374.45
<0.0001
Capacity class (exurban high)
1.97
7,737.2
<0.0001
Global test
21.34
56,526.73
<0.0001
From LUC2000 for Transitions into Urban High by Capacity Class (RefLevel: Urban
Low)
Capacity class (wetlands)
1.58
14.89
0.0003
Capacity class (timber)
1.29
8.3
0.0064
Capacity class (grazing)
2.01
219.79
<0.0001
Capacity class (pasture)
2.06
37.33
<0.0001
Capacity class (cropland)
1.87
36.2
<0.0001
Capacity class (exurban low)
1.99
342.86
<0.0001
B-27
-------
Table B-6. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class. RefLevel is the reference
level land use from which transitions are calculated, (continued)
Smoothing Terms
edf
x2
P
Capacity class (exurban high)
2.01
692.26
<0.0001
Capacity class (suburban)
2.00
357.40
<0.0001
Global test
22.81
3,211.57
<0.0001
From LUC2000 for Transitions into Commercial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
1.98
241.71
<0.0001
Capacity class (suburban)
1.90
2,772.21
<0.0001
Capacity class (urban low)
1.95
2,106.21
<0.0001
Capacity class (urban high)
1.85
257.98
<0.0001
Global test
11.68
34,302
<0.0001
From LUC2000 for Transitions into Industrial by Capacity Class (RefLevel: Exurban
High)
Capacity class (wetlands)
0.78
8.62
0.0022
Capacity class (grazing)
2.39
252.5
<0.0001
Capacity class (pasture)
1.89
59.64
<0.0001
Capacity class (cropland)
1.44
12.58
0.0008
Capacity class (exurban low)
2.11
129.01
<0.0001
Global test
13.61
2,721.24
<0.0001
B-28
-------
Reg ton* i Combined Wotielc
"francftianc from LUC2D0Q 1a E t urfc in Law
Region B.' Combined Modelt
Trancrttorw from LUG20DD to Ex urban Htgh
40
o
¦SO
O
O
¦M
O
o
o
ta
o
•r
o
-«*
o
o
o
o
CaDac-j- Ck:
Capadty Class
Ren Jon® J Combined lio«ielc
TrancftJanc from LUC2DQQ ta Suburban
(O
o
•c<
o
•N
o
o
o
kmism -i ngp.
*
£
6
Reglon ® / Combined Models
Tkwwlifont from LUC20D0 to Urban Low
«•
o
o
CH
o
o
o
C-aoac"/ Class
CapaCty Class
Ren ton® 1 Combined Wodelc
TranGlllanc from LUCEOaolo Urban Hugh
'30
o
¦o
o
o
o
o
1 Bit
amHtp ur
au .i5»
— r .a
iur
kAt'
A
5
Region ® / Combined Modelc
Ttanclttorc from LUC2DDD to Corrmerolail
CO
o
•r
o
¦*
o
c*
o
o
o
Cap acta C «:
Capacity C a :c
B-29
-------
Region 6 I Combined Models
Transitions from LUC2000 to Industrial
Grazing Exurban high
Exurb3n Lew
Capacity Class
FigureB-6. Predicted transition probabilities by capacity class from LUCs
in 2000 to LUCs in 2010 in Integrated Climate and Land Use Scenarios
(ICLUS) Region 6 (Southeast). Each panel shows transitions into a particular
LUC in 2010. These combined probabilities are the product of corresponding
marginal and conditional models, i.e., for a given capacity class the probability of
transitioning from LUCi into LUCj is P(LUCij) = P(LUCj) X P(LUCijj).
B-30
-------
B.7. INTEGRATED CLIMATE AND LAND USE SCENARIOS (ICLUS) REGION 7
(NORTHEAST) TRANSITION PROBABILITY MODELS
Table B-7. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a particular
LUC in 2010,/?(LUCiy), by capacity class.
Smoothing T erms
edf
x2
P
For Transitions into LUC2010 by Capacity Class (RefLevel: Exurban Low)
Capacity class (exurban high)
1.72
22,867.66
<0.0001
Capacity class (suburban)
1.61
5,812.7
<0.0001
Capacity class (urban low)
1.62
5,180.12
<0.0001
Capacity class (urban high)
1.63
1,221.28
<0.0001
Capacity class (commercial)
1.88
12,214.7
<0.0001
Capacity class (industrial)
1.93
1,653.35
<0.0001
Global test
16.39
2,190,093
<0.0001
From LUC2000 for Transitions into Exurban Low by Capacity Class (RefLevel: Grazing)
Capacity class (timber)
2.1
68.99
<0.0001
Capacity class (pasture)
2.01
13,104.18
<0.0001
Capacity class (cropland)
2.07
4,203.43
<0.0001
Global test
9.17
72,913.49
<0.0001
From LUC2000 for Transitions into Exurban High by Capacity Class (RefLevel: Exurban
Low)
Capacity class (timber)
2.01
327.83
<0.0001
Capacity class (grazing)
1.99
16,657.91
<0.0001
Capacity class (pasture)
2.04
263.96
<0.0001
Capacity class (cropland)
2.16
1,047.29
<0.0001
Global test
12.2
31,983.46
<0.0001
B-31
-------
Table B-7. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class, (continued)
Smoothing Terms
edf
x2
P
From LUC2000 for Transitions into Suburban by Capacity Class (RefLevel: Exurban
High)
Capacity class (wetlands)
1.98
600.77
<0.0001
Capacity class (timber)
1.95
51.98
<0.0001
Capacity class (grazing)
1.94
604.6
<0.0001
Capacity class (pasture)
2.07
32.79
1.00 X 10"4
Capacity class (cropland)
2.2
225.2
<0.0001
Capacity class (exurban low)
2
205.86
<0.0001
Global test
18.14
18,896.3
<0.0001
From LUC2000 for Transitions into Urban Low by Ca
>acity Class (RefLevel: Suburban)
Capacity class (wetlands)
1.83
14.17
0.0007
Capacity class (timber)
1.76
8.68
0.0099
Capacity class (grazing)
1.93
103.01
<0.0001
Capacity class (pasture)
2.19
47.36
<0.0001
Capacity class (cropland)
2.15
115.43
<0.0001
Capacity class (exurban low)
2.04
173.29
<0.0001
Capacity class (exurban high)
1.97
713.07
<0.0001
Global test
20.88
9,717.48
<0.0001
From LUC2000 for Transitions into Urban High by Capacity Class (RefLevel: Urban
Low)
Capacity class (wetlands)
0.97
3.08
0.0757
Capacity class (timber)
1.2
1.35
0.3004
Capacity class (grazing)
1.88
32.85
<0.0001
Capacity class (pasture)
1.55
3.21
0.1377
Capacity class (cropland)
1.21
16.2
0.0001
Capacity class (exurban low)
1.9
33.15
<0.0001
B-32
-------
Table B-7. Estimated nonlinear degrees of freedom (edf) and significance of
the smoothing terms for the multinomial GAMs. The top marginal model
predicts the probability of transitioning into each LUC in 2010, /?(LUCj), by
capacity class, while the bottom conditional models predict the probability of
transitioning from each LUC in 2000 given that they transitioned into a
particular LUC in 2010, /?(LUCiy), by capacity class, (continued)
Smoothing Terms
edf
x2
P
Capacity class (exurban high)
2.05
44.44
<0.0001
Capacity class (suburban)
1.98
73.51
<0.0001
Global test
20.74
1,673.8
<0.0001
From LUC2000 for Transitions into Commercial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
1.97
576.07
<0.0001
Capacity class (suburban)
1.82
80.49
<0.0001
Capacity class (urban low)
1.78
15.94
0.0003
Capacity class (urban high)
1.68
9.73
0.0052
Global test
11.25
11,010.72
<0.0001
From LUC2000 for Transitions into Industrial by Capacity Class (RefLevel: Exurban
High)
Capacity class (exurban low)
2.13
330.36
<0.0001
Global test
3.13
469.32
<0.0001
B-33
-------
s eg tor 7 J Combined Wocelc
TransttSorc. from LUCSDDQ 1a Exurban Law
Region 7i" Combined Modeic
Ttarctttorw from LUC200D ta Ex urban High
o
C aas
CO
o
o
CM
O
c
o
Capacity C a:s
Region 7/ Combined MoqpIe
Trancttlarc from LUC2D00 1a Suburban
Region 7." Combined Mode c
Trarclttor* from LUC2DD0 to Urlbar Low
¦K>
O
¦tt
O
¦IN
O
o
o
TO
O
•£
O
C-i
O
o
O'
Cap act/ -Class
Capacity Cass
Region 7 J Combined Mocslt
TrancHlanc from LUC20IK 1o Urban High
o
-------
Region 7 I Combined Models
Transitions from LUC2000 to Industrial
E*urb3n Lew E*urb3n High
Capacity Class
FigureB-7. Predicted transition probabilities by capacity class from LUCs
in 2000 to LUCs in 2010 in Integrated Climate and Land Use Scenarios
(ICLUS) Region 7 (Northeast). Each panel shows transitions into a particular
LUC in 2010. These combined probabilities are the product of corresponding
marginal and conditional models, i.e., for a given capacity class the probability of
transitioning from LUCi into LUCj is P(LUCij) = P(LUCj) X P(LUCijj).
B-35
-------
APPENDIX C. LAND USE CLASS (LUC) AND CAPACITY DEMAND MODELS
Table C-l. Generalized additive model (GAM) model output relating
natural log (In) transformed LUC density and capacity to In transformed
population density. Output includes an estimate of the intercept, estimated
degrees of freedom (edf) for the smoothing term, the adjusted R2 associated with
the model, the standard error (SE) associated with the estimate of the intercept and
T and F statistics associated with the significance of the intercept and smoothing
terms, respectively.
GAM Relating ln(Exurban Low Pixel Density) to ln(Population Density)
Parametric Terms
Estimate
SE
T
p
Intercept
0.744
0.027
22.77
<0.0001
Smoothing Terms
edf
F
P
Population density
5.899
773.2
<0.0001
Adjusted R2
0.550
GAM Relating ln(Exurban High Pixel Density) to ln(Population Density)
Parametric Terms
Estimate
SE
T
p
Intercept
0.455
0.011
41.63
<0.0001
Smoothing Terms
edf
F
P
Population density
8.177
2,214
<0.0001
Adjusted R2
0.812
GAM Relating ln(Suburban Pixel Density) to ln(Population Density)
Parametric Terms
Estimate
SE
T
p
Intercept
-0.745
0.007
-100.1
<0.0001
Smoothing Terms
edf
F
p
Population density
7.021
4,453
<0.0001
Adjusted R2
0.889
GAM Relating ln(Urban Low Pixel Density) to ln(Population Density)
Parametric Terms
Estimate
SE
T
p
Intercept
-1.365
0.010
-132.1
<0.0001
Smoothing Terms
edf
F
P
Population density
7.005
2,218
<0.0001
Adjusted R2
0.800
C-l
-------
Table C-l. Generalized additive model (GAM) model output relating
natural log (In) transformed LUC density and capacity to In transformed
population density. Output includes an estimate of the intercept, estimated
degrees of freedom (edf) for the smoothing term, the adjusted R2 associated with
the model, the standard error (SE) associated with the estimate of the intercept
and T and F statistics associated with the significance of the intercept and
smoothing terms, respectively, (continued)
GAM Relating ln(Urban High Pixel Density) to ln(Population Density)
Parametric Terms
Estimate
SE
T
p
Intercept
-5.559
0.013
-413.4
<0.0001
Smoothing Terms
edf
F
P
Population density
6.730
1,819
<0.0001
Adjusted R2
0.761
GAM Relating ln(Commercial Pixel Density) to ln(Population Density)
Parametric Terms
Estimate
SE
T
p
Intercept
-2.540
0.015
-175.1
<0.0001
Smoothing Terms
edf
F
P
Population density
5.479
1,675
<0.0001
Adjusted R2
0.713
GAM Relating ln(Industrial Pixel Density) to ln(Population Density)
Parametric Terms
Estimate
SE
T
p
Intercept
-23.182
0.018
-177.0
<0.0001
Smoothing Terms
edf
F
P
Population density
6.056
1,051
<0.0001
Adjusted R2
0.629
GAM Relating ln(Capacity Density) to ln(Population Density)
Parametric Terms
Estimate
SE
T
p
Intercept
9.505
0.004
2,292
<0.0001
Smoothing Terms
edf
F
P
Population density
7.939
1,109
<0.0001
Adjusted R2
0.682
C-2
-------
<8-EPA
United States
Environmental Protection
Agency
PRESORTED STANDARD
POSTAGE & FEES PAID
EPA
PERMIT NO. G-35
Office of Research and Development (8101R)
Washington, DC 20460
Official Business
Penalty for Private Use
$300
------- |