Content uploaded by Matt Hauer
Author content
All content in this area was uploaded by Matt Hauer on Oct 17, 2017
Content may be subject to copyright.
LETTERS
PUBLISHED ONLINE: 17 APRIL 2017 | DOI: 10.1038/NCLIMATE3271
Migration induced by sea-level rise could reshape
the US population landscape
Mathew E. Hauer
Many sea-level rise (SLR) assessments focus on populations
presently inhabiting vulnerable coastal communities1–3, but to
date no studies have attempted to model the destinations of
these potentially displaced persons. With millions of potential
future migrants in heavily populated coastal communities, SLR
scholarship focusing solely on coastal communities character-
izes SLR as primarily a coastal issue, obscuring the potential
impacts in landlocked communities created by SLR-induced
displacement. Here I address this issue by merging projected
populations at risk of SLR1with migration systems simulations
to project future destinations of SLR migrants in the United
States. I find that unmitigated SLR is expected to reshape the
US population distribution, potentially stressing landlocked
areas unprepared to accommodate this wave of coastal
migrants—even after accounting for potential adaptation.
These results provide the first glimpse of how climate change
will reshape future population distributions and establish a new
foundation for modelling potential migration destinations from
climate stressors in an era of global environmental change.
It is generally understood that sea-level rise (SLR) of 1–2m
(refs 4–6) could lead to widespread human migration2,7 as residents
of highly vulnerable coastal communities look to escape rising
water levels. With up to 180 million people directly at risk to
SLR in the world and over 1 billion living in the lower-elevation
coastal zone8,9, understanding the ramifications of these potential
migrants on destination communities is a priority for climate
change research10–13.
SLR assessments, identifying both the number and locations of
potentially displaced persons, fill the literature1,2,14 and are useful
for the deployment of critical infrastructure in coastal areas. Yet
questions of where the millions of potentially displaced persons
will go remain unanswered despite a general understanding that
SLR displaced persons are likely to have profound effects on
future population landscapes11,14. Only a few studies have put forth
general hypotheses regarding SLR migration11,15 , and this void has
prompted recent calls for additional migration modelling16,17. To
date, no studies modelling precisely how SLR-induced migration
will affect the population distribution exist. By focusing solely
on coastal communities without directly addressing SLR-induced
migration, we probably underestimate the scale and magnitude of
these impacts.
Relationships between environmental stressors and migration
are highly complex as press and pulse events trigger migration
responses that range from short-distance temporary migration to
permanent long-distance migration; some will move and others
will not18–23. SLR is unique among environmental stressors as the
conversion of habitable land to uninhabitable water is expected
to lead to widespread human migration without the deployment
of costly protective infrastructure2,7,11,15. It is unclear, however,
what will actually trigger future climate migrants: press events,
such as drought or SLR, or pulse events, such as tropical
cyclones. When climate effects are integrated over long periods
of time, it is likely that a combination of press and pulse events
will spur migration24 across pre-existing migration pathways19,21 ,
leveraging established networks of social capital and kin networks in
destination decisions25. This is because press and pulse events that
spur migration operate mostly independently of the kin networks
and social capital that drive destination decisions10. Thus, climate
migrants resulting from press stressors will probably constitute
‘enhanced’, or extra, normal out-migration.
I combine estimates of the populations at risk to SLR1within
a migration systems simulation to estimate both the number and
destinations of potential SLR migrants in the United States (US) over
the coming century. By focusing on the destinations of SLR migrants
I am able to more holistically describe the impacts of SLR. This study
aims to answer one fundamental question regarding SLR-induced
migration: What areas are likely to see the greatest in-migration
due to SLR? Local officials in landlocked communities can use these
results to plan for potential infrastructure required to accommodate
an influx of coastal migrants and could shift the conceptualization
of SLR from a coastal issue to a more ubiquitous issue.
To answer these questions, first I use published estimates of
county-level projected populations at risk to SLR for the years 2010
through 21001under the 1.8 m SLR scenario for 319 coastal US
counties. Hauer et al.1simulated expected changes in the mean
higher high water (MHHW) mark on areas that are hydrologically
connected to coastal areas without taking into account additional
land loss caused by other natural factors such as erosion or land
subsidence. They then projected the populations exposed to SLR
over the coming century in a dynamically assessed, spatially explicit
small-area population-environment projection model, based on
growth in the period 1940–2010, where populations under the
projected MHHW mark are assumed to be at risk of displacement.
They estimated a potential 13.1 million persons could be at
risk of migrating due to a SLR of 1.8m by 2100. These data
provide the number of persons likely to migrate, but not the
migration destinations.
I then integrated the projected populations within a projected
migration system for all coastal counties affected by SLR (n=319)
and possible destinations (n=3,113) based on the Internal Revenue
Services’ annual series of county-to-county migration flow data
for the years 1990–201326. This is the largest dataset of county-to-
county migration in the US and includes data for 95 to 98 per cent
of tax filers and their dependents. To capture the temporal variability
of the county-to-county migration flows I used unobserved
component modelling (UCM)27 for each individual dyad origin–
destination pair (n=46,203) to project shifts and changes in
the migration system over the coming century. Populations are
migrated based on the proportion of total outflows from the
originating county in the projected system.
Department of Geography, University of Georgia, Athens, Georgia 30602, USA. e-mail: hauer@uga.edu
NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 1
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTERS NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3271
>450,000
200,001−450,000
50,001−200,000
1−50,000
−49,999−0
−199,999 to −50,000
−449,999 to −200,000
<−450,000
>450,000
200,001−450,000
50,001−200,000
1−50,000
−49,999−0
−199,999 to −50,000
−449,999 to −200,000
<−450,000
ab
TX
CA
MT
MI
AZ
ID
NV
NM
CO
OR
IL
UT
WY
KS
MN
IA
SD
NE
FL
WI
ND
OK
WA
MO
GA
LA
AL
NY
NC
AR
IN
PA
MS
VA
TN
OH
KY
ME
SC
WV MD
VT
NJ
MA
NH
CT
DE
RI
Miami
Tampa
Boston
Austin
Dallas
Denver
Atlanta
Chicago
Seattle
New York
Charlotte
Los Angeles
San Francisco
Figure 1 | Estimated SLR net migrants (in-migrants minus out-migrants) for counties and core based statistical areas under the 1.8 m scenario and no
adaptation. a, US counties. b, Core based statistical areas. For b, I considered only counties located in CBSAs. Counties and CBSAs without expected SLR
in-migration are in white. States are abbreviated to standard two-letter codes.
−3.0
−2.5
−2.0
−1.5
−1.0
−0.5
0.0
0.5
1.0
1.5
2.0
TX
GA
NC
PA
NV
NY
AZ
IL
TN
CO
WA
MD
OH
MI
OR
DC
MO
HI
NH
WY
MS
KY
IN
UT
MN
AK
WI
AL
KS
OK
RI
AR
SD
NE
NM
ME
ID
WV
VT
IA
ND
MT
DE
CT
MA
SC
CA
VA
NJ
LA
FL
Millions
Figure 2 | Net change in population due to sea-level rise under the 1.8m scenario and no adaptation. I considered migration destinations for all 50 states
and the District of Columbia (DC) and migration origins for 22 states and the District of Columbia. These are the net changes in population due to both in-
and out-migration due to sea-level rise. States are abbreviated to standard two-letter codes.
It is possible that populations escaping SLR might migrate
to inland areas completely unaffected by SLR, as hypotheses
suggest11,15. However, not all coastal counties will be completely
inundated and many areas will still be suitable for human settlement
even with 1.8 m of SLR. To capture these possibilities, projected
migrants to each possible destination county are dynamically
adjusted based on the unaffected populations remaining in each
coastal county. In this way I model both migrations to inland areas
completely unaffected by SLR and migrations to coastal areas still
suitable for habitation. A detailed technical description is available
in the Methods.
It is likely that many communities will deploy a wide variety
of adaptation measures, including sea walls, beach and marsh
nourishment, pumps, or elevate homes and roads to protect
both people and property, and IPCC reports have increasingly
emphasized adaptation when discussing SLR28. Global estimates of
adaptive infrastructure for SLR could reach US$421 billion (2014
values) per year8and could cost upwards of US$1.1 trillion in the
US29. However, the deployment of adaptation measures is driven
by wealth for both cities30 and individuals10,31 . To approximate
this dynamic, I assume that households earning greater than
US$100,000 per year are likely to adapt to SLR in some manner,
and thus unlikely to migrate. This income threshold represents
approximately the top quartile and double the US median household
income, and is neither too restrictive nor too broad to capture the
range of individual adaptive measures. Detailed projections of SLR
2
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3271 LETTERS
FL
500
1,000
1,500
2,000
2,500
3,000
3,500
4,000
4,500
5,000
5,500
6,000
6,500
7,0 00
7,500
8,000
8,500
9,000
TX
500
1,000
1,500
2,000
NY
500
1,000
1,500
CA
500
1,000
1,500
2,000
LA
500
1,000
1,500
2,000
NJ
500
1,000
NC
500
GA
500
MA
500
SC
500
VA
500
MD
PA
WA
NV
AZ
CT
IL
TN
CO
MS
AL
DE
OR
OH
MI
RI
DC
NH
ME
MO
HI
WY
KY
IN
UT
MN
AK
WI
KS
OK
AR
SD
NE
NM
ID
VT
WV
IA
ND
MT
Figure 3 | Circular plot of bi-lateral SLR migration flows for US States under the 1.8 m scenario and no adaptation. Tick marks show the number of
migrants (inflows and outflows) in thousands. States are ordered clockwise by the size of inflows. The top ten outflow states are coloured; all other states
are in grey. States are abbreviated to standard two-letter codes.
migrants for all destination counties under both a ‘no adaptation’
scenario and a wealth-based adaptation scenario are also found in
the Supplementary Dataset.
I find that in the US, every state, 86% of US Core Based Statistical
Areas (CBSAs) (791 out of 915), and 56% of counties (1,735 out of
3,113) could be affected in some way by net migration (in-migration
minus out-migration) associated with 1.8 m of SLR (Fig. 1). Florida
could lose more than 2.5 million residents due to 1.8m of SLR,
while Texas could see nearly 1.5 million additional residents (Fig. 2).
Additionally, nine states could see net losses in their populations
due to SLR. Figure 3 demonstrates all origin–destination flows at
the US state level, demonstrating that the sheer magnitude of places
affected could alter the US population landscape. SLR migrants are
expected to comprise both intra- and inter-state migrations, and no
state is left untouched by SLR migration. Even accounting for the
deployment of adaptive infrastructure, millions of people could still
migrate (Table 1 and Supplementary Dataset).
My results also suggest that CBSAs such as Austin Texas,
Orlando Florida, Atlanta Georgia, and Houston Texas could
see more than 250,000 previously unforeseen future SLR net
migrants each (Table 1 and Supplementary Dataset). Thirteen
CBSAs could see more than 100,000 SLR net migrants by 2100
with 1.8 m of SLR. Conversely, ten CBSAs could lose more
than 100,000 residents due to SLR, with Miami Florida losing
over 2.5 million residents. Even accounting for those who could
adapt in place, many inland communities could see tens of
thousands of SLR in migrants, and many coastal communities
could lose tens of thousands of residents. Extended results for all
NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
3
LETTERS NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3271
Table 1 | Select core based statistical area (CBSA) destinations of SLR net migration (in-migration minus out-migration) in 2100
with and without adaptation under the 1.8 m scenario.
CBSA No adaptation With adaptation
Net migration +/−Rank Net migration +/−Rank
Austin-Round Rock, TX 818,938 243,821 1 625,627 179,186 1
Orlando-Kissimmee-Sanford, FL 461,411 62,665 2 369,120 38,834 2
Atlanta-Sandy Springs-Roswell, GA 320,937 131,984 3 248,684 68,868 3
Phoenix-Mesa-Scottsdale, AZ 100,524 12,851 13 73,935 1,949 13
Myrtle Beach-Conway-N Myrtle Beach, SC-NC 12,146 13,855 78 3,142 9,389 141
North Port-Sarasota-Bradenton, FL −208 25,057 727 2,128 21,717 168
Los Angeles-Long Beach-Anaheim, CA −3,140 51,590 737 13,181 22,200 61
New York-Newark-Jersey City, NY-NJ-PA −50,804 494,625 775 15,808 194,047 50
New Orleans-Metairie, LA −500,011 24,053 795 −373,283 10,733 795
Miami-Fort Lauderdale-West Palm Beach, FL −2,509,978 155,119 796 −2,009,263 95,845 796
+/−represents the 80th confidence interval. States are abbreviated to standard two-letter codes.
counties and CBSAs are available in the Supplementary Methods
(Supplementary Dataset).
With many projected migrants remaining in coastal communities
(Table 1 and Supplementary Dataset), SLR could generate millions
of ‘trapped’ people32 . Trapped populations are sometimes discussed
through the concept of involuntary immobility32, but there is also
those who do not desire to move and thus constitute voluntary
immobility. These results suggest that many people displaced by SLR
could find themselves or their descendants exposed to SLR, even
with migration as an adaptation, as sea levels continue to rise past
the year 2100 with migration that constitutes relocation to presently
safe, but ultimately vulnerable, coastal communities.
Additionally, infrastructure challenges required to protect
coastal communities are well documented1,2, but the infrastructure
challenges of accommodating millions of SLR migrants in largely
unprepared inland municipalities is virtually unexplored. For many
destinations, such as Riverside California, Phoenix Arizona, Las
Vegas Nevada, and Atlanta Georgia, already experiencing water
management and growth management challenges, the SLR migrants
who wash across the landscape over the coming century could place
undue burden in these places if accommodation strategies are left
unplanned. Studies of migration impacts do not solve the challenges
in these areas, but rather reveal a more holistic understanding of
SLR impacts and needed interventions.
SLR has been broadly conceptualized as a coastal issue or haz-
ard, as assessments have focused on the effects in coastal com-
munities1,2,8,9. With millions of potential future migrants in heavily
populated coastal communities, SLR scholarship focusing solely on
coastal communities endorses a narrative that characterizes SLR
as primarily a coastal issue, obscuring the potential impacts in
landlocked communities created by SLR-driven migration. My work
shows that this coastal conceptualization of SLRcreates a deceptively
small area of affect if relocation is left unaccounted. This work offers
the first glimpse of how SLR could alter the population distribution
of the US as both coastal and landlocked communities are likely
to be affected by SLR: directly in coastal areas due to SLR itself
and indirectly in landlocked areas through the influx of people
escaping SLR.
Furthermore, the migration approach for examining destinations
associated with climate change shown here allows for modelling
migration destinations of other climate change stressors. For
instance, it has been estimated that parts of the Middle East
and North Africa (MENA) could become uninhabitable by the
end of the century, potentially spurring an exodus of 500 million
people3. Future scholars could employ my approach to model
the destinations of these potential MENA migrants. There is
tremendous potential in coupling migration systems information
with climate change models to examine the implications of climate-
change-induced migration. This type of modelling requires detailed
origin–destination migration information, limiting the areas where
this approach could be used to those where data are available.
Migration models are only as good as the data underlying them.
The suppression of IRS migration data of flows with fewer than 10
migrants could systematically bias my results against rural areas far
from coastal communities, and the modelling approach undertaken
does not allow for new unforeseen migration pairs to emerge in
the future. However, the destination counties cover 93.6% of the
US population and over 250 destination counties are outside of
CBSAs (the typical distinction of urban/rural), limiting the scope
of the rural bias to mainly sparsely populated communities far from
coastal areas. New destinations for SLR migrants could still emerge
in rural areas, and if they do, my results of the geographic spread
of SLR migrants could be considered conservative. Environmental
migration scholars could further investigate the likelihood of
emerging rural destinations related to climate change migration to
better inform future modelling efforts.
Previous trends are not always indicative of future results.
Societal and economic shifts, population ceilings, local growth
ordinances, adaptive behaviour, and climate change itself could
all change future migrant destinations. Although I model the
potential destinations of SLR migrants, I do not precisely model
how other climate stressors or other factors might influence the
future migration systems. Our current understanding of the location
decisions of environmental migrants is still limited, and there
have been recent calls to better understand migration flows16.
My approach builds on the growing literature concerning migrant
destinations and environmental change10,11,15,25,33–35, and accounts for
potential future migration systems.
High potential exists for development of deeper and more
integrated examinations of the role of adaptation and other climate
stressors on SLR migration. Neglecting to account for adaptive
behaviour could lead to an overestimation of either flood risk36
or migration37. Future studies could examine varying adaptation
scenarios related to the 100-year flood plain or the lower-elevation
coastal zone creating differing scenarios of ‘stayers.’
If future migration pathways mimic past pathways, SLR is
expected to reshape the US population distribution and could
stress some landlocked areas unprepared for these migrations while
revitalizing others. SLR is currently framed as a coastal hazard,
but the migratory effects could ripple far inland. My results show
the importance of accounting for future migrations associated
with climate change in long-range planning processes for disaster
management, transportation infrastructure, land-use decisions, and
so on.
4
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3271 LETTERS
Methods
Methods, including statements of data availability and any
associated accession codes and references, are available in the
online version of this paper.
Received 23 August 2016; accepted 13 March 2017;
published online 17 April 2017
References
1. Hauer, M. E., Evans, J. M. & Mishra, D. R. Millions projected to be at risk
from sea-level rise in the continental United States. Nat. Clim. Change 6,
691–695 (2016).
2. Strauss, B. H., Kulp, S. & Levermann, A. Carbon choices determine US cities
committed to futures below sea level. Proc. Natl Acad. Sci. USA 112,
13508–13513 (2015).
3. Lelieveld, J. et al. Strongly increasing heat extremes in the Middle East
and North Africa (MENA) in the 21st century. Climatic Change 137,
245–260 (2016).
4. DeConto, R. M. & Pollard, D. Contribution of Antarctica to past and future
sea-level rise. Nature 531, 591–597 (2016).
5. Vermeer, M. & Rahmstorf, S. Global sea level linked to global temperature.
Proc. Natl Acad. Sci. USA 106, 21527–21532 (2009).
6. Levermann, A. et al. The multimillennial sea-level commitment of global
warming. Proc. Natl Acad. Sci. USA 110, 13745–13750 (2013).
7. IPCC in Climate Change 2014: Impacts, Adaptation, and Vulnerability
(eds Field, C. B. et al.) (Cambridge Univ. Press, 2014).
8. Nicholls, R. J. et al. Sea-level rise and its possible impacts given a ‘beyond
4◦C world’ in the twenty-first century. Phil. Trans. R. Soc. A 369,
161–181 (2011).
9. Neumann, B., Vafeidis, A. T., Zimmermann, J. & Nicholls, R. J. Future coastal
population growth and exposure to sea-level rise and coastal flooding—a
global assessment. PLoS ONE 10, e0118571 (2015).
10. Black, R., Bennett, S. R. G., Thomas, S. M. & Beddington, J. R. Migration as
adaptation. Nature 478, 447–449 (2011).
11. Döös, B. R. Can large-scale environmental migrations be predicted?
Glob. Environ. Change 7, 41–61 (1997).
12. Mueller, V., Gray, C. & Kosec, K. Heat stress increases long-term
human migration in rural Pakistan. Nat. Clim. Change 4,
182–185 (2014).
13. Hugo, G. Future demographic change and its interactions with migration and
climate change. Glob. Environ. Change 215, 521–533 (2011).
14. Ben, M. & Anders, L. Loss of cultural world heritage and currently inhabited
places to sea-level rise. Environ. Res. Lett. 9, 034001 (2014).
15. Curtis, K. & Schneider, A. Understanding the demographic implications of
climate change: estimates of localized population predictions under future
scenarios of sea-level rise. Popul. Environ. 33, 28–54 (2011).
16. Willekens, F., Massey, D., Raymer, J. & Beauchemin, C. International migration
under the microscope. Science 352, 897–899 (2016).
17. Hugo, G. Migration, Development and Environment (International
Organization for Migration Geneva, 2008).
18. McLeman, R. A. Climate and Human Migration: Past Experiences, Future
Challenges Report No. 1107022657 (Cambridge Univ. Press, 2013).
19. Gutmann, M. P. & Field, V. Katrina in historical context: environment and
migration in the US. Popul. Environ. 31, 3–19 (2010).
20. Collins, S. L. et al. An integrated conceptual framework for long-term
social-ecological research. Front. Ecol. Environ. 9, 351–357 (2010).
21. Hunter, L. M., Murray, S. & Riosmena, F. Rainfall patterns and US migration
from rural Mexico. Int. Migr. Rev. 47, 874–909 (2013).
22. Thiede, B. & Brown, D. Hurricane Katrina: who stayed and why? Popul. Res.
Policy Rev. 32, 803–824 (2013).
23. Entwisle, B. et al. Climate shocks and migration: an agent-based modeling
approach. Popul. Environ. 38, 47–71 (2016).
24. Arenstam Gibbons, S. J. & Nicholls, R. J. Island abandonment and sea-level
rise: an historical analog from the Chesapeake Bay, USA. Glob. Environ. Change
16, 40–47 (2006).
25. Findlay, A. M. Migrant destinations in an era of environmental change.
Glob. Environ. Change 21, S50–S58 (2011).
26. SOI Tax Stats—County-to-County Migration Data Files (IRS, 2016);
http://www.irs.gov/uac/soi-tax-stats-county-to-county-migration-data-files
27. Harvey, A. C. Forecasting, Structural Time Series Models and the Kalman Filter
(Cambridge Univ. Press, 1990).
28. Brown, S. et al. Shifting perspectives on coastal impacts and adaptation.
Nat. Clim. Change 4, 752–755 (2014).
29. Neumann, J. E. et al. Joint effects of storm surge and sea-level rise on US
Coasts: new economic estimates of impacts, adaptation, and benefits of
mitigation policy. Climatic Change 129, 337–349 (2015).
30. Georgeson, L., Maslin, M., Poessinouw, M. & Howard, S. Adaptation responses
to climate change differ between global megacities. Nat. Clim. Change 6,
584–588 (2016).
31. Adger, W. N., Arnell, N. W. & Tompkins, E. L. Successful adaptation to climate
change across scales. Glob. Environ. Change 15, 77–86 (2005).
32. Black, R., Arnell, N. W., Adger, W. N., Thomas, D. & Geddes, A. Migration,
immobility and displacement outcomes following extreme events. Environ. Sci.
Policy 27, S32–S43 (2013).
33. Smith, C., Kniveton, D. R., Wood, S. & Black, R. Climate Change and
Migration: A Modelling Approach. Adv. Glob. Change Res. 43, 179–201 (2011).
34. Black, R., Kniveton, D. & Schmidt-Verkerk, K. Migration and climate change:
towards an integrated assessment of sensitivity. Environ. Plann. A 43,
431–450 (2011).
35. Black, R. et al. The effect of environmental change on human migration. Glob.
Environ. Change 21, S3–S11 (2011).
36. Haer, T., Botzen, W. J. W., de Moel, H. & Aerts, J. C. J. H. Integrating household
risk mitigation behavior in flood risk analysis: an agent-based model approach.
Risk Anal. http://dx.doi.org/10.1111/risa.12740 (2016).
37. Kocornik-Mina, A., McDermott, T. K., Michaels, G. & Rauch, F. Flooded Cities
(Centre for Economic Performance, LSE, 2016).
Acknowledgements
I am grateful for the constructive comments from J. M. Byars, S. Holloway,
J. M. Shepherd, J. Evans, J. S. Pippin and J. Véron.
Additional information
Supplementary information is available in the online version of the paper. Reprints and
permissions information is available online at www.nature.com/reprints. Publisher’s note:
Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.
Competing financial interests
The author declares no competing financial interests.
NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
5
LETTERS NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3271
Methods
The methodology for projecting sea-level rise (SLR) migrant destinations is
outlined in this section. First, I describe the datasets and basic methodology for
creating my migration matrices. Second, the methodology for capturing migration
system uncertainty is discussed.
Data. I utilize two primary sources of data concerning the magnitude of flows in
the migration system and the migration system itself. The first, the magnitude of
flows, comes from published populations projected to be at risk to SLR1.
Hauer et al.1used the National Oceanic and Atmospheric Administration’s
(NOAA) 0 m, 0.9 m (3 feet) and 1.8m (6 feet) SLR datasets for twenty-two coastal
states and the District of Columbia. These datasets simulate expected changes in
the mean higher high water (MHHW) mark on areas that are hydrologically
connected to coastal areas without taking into account additional land loss caused
by other natural factors such as erosion. They projected populations using a
modified Hammer Method38 combined with the Housing Unit Method39 for
population estimation to create temporally contiguous sub-county boundaries over
a 70-year base period from 1940 to 2010, which were then used to project
populations at these same sub-county geographies through the use of a
linear/exponential extrapolation approach for projecting census block groups
(CBG) from 2010 to 2100. The populations at risk to SLR, aggregated to the US
county and available in their Supplementary Table 21, provide the magnitude of
out-migrants from 319 coastal counties.
For this research, I used the annual series of county-to-county migration
datasets26, produced by the Internal Revenue Service (IRS) in conjunction with the
US Census Bureau, as the basis for the migration system. The IRS datasets utilize
the IRS Individual Master File containing every Form 1040, 1040A and 1040EZ
processed by the IRS, and includes 95 to 98 percent of all individual tax filers and
their dependents. The Census Bureau identifies migrants when a current filing
years’ return is from a different location than the matched preceding years’ return.
These data capture only the tax-filing universe, but the spatio-temporal
stability40–42 coupled with the very large administrative sample make them
attractive for modelling large-scale migration patterns. The IRS does suppress
migration flows comprising fewer than 10 individual migrants, systemically
suppressing small rural migration flows. However, the long-term trend of rural
out-migration to urban areas43 is expected to continue in this century44.
Migration systems theory (MST) has been tied to environmental migration in
recent years45–47. MST is a branch of migration research that holistically examines
migration options by studying all origin–destination combinations rather than any
single origin–destination combination45,46,48. Migration decisions—not just the
decision to migrate, but also decisions on where to migrate—are often driven by
kin networks, employment opportunities, amenities, both natural and economic,
economic vitality, and so forth10,22,46,49–51 . This network of ‘pull’ factors embedded
within the migration system tends to drive locational decision-making due to
environmental, or other, ‘push’ factors10,13,17,18,25,52,53 .
To describe the complete migration system in the United States, let matrix M(x)
represent all possible county-level origin–destination combinations. The sum of
any given column and row in the matrix will equal the total number of migrants
into or out of any given county. For this analysis I am concerned only with the 319
coastal counties expected to experience some form of SLR inundation under the
1.8 m scenario and their connections to the other 3,113 US counties
(n=993,047 matrix cells). I created these matrices for each year of the IRS
migration data between 1990 and 2013. Supplementary Fig. 1 shows examples of
these systems into and out of three sample counties.
M(x)=
m1,1 ··· m1,3113
.
.
.....
.
.
m319,1 ··· m319,3113
=[mx
o,d](1)
where
o∈{1, ..., 319}
d∈{1, ..., 3113}
Migration system projection approach. I employed the use of an unobserved
components model (UCM) for forecasting equally spaced univariate time series
data27. UCMs decompose a time series into components such as trends, seasons,
cycles and regression effects, and are designed to capture the features of the series
that explain and predict its behaviour. UCMs are similar to dynamic models in
Bayesian time series forecasting54. All projections were undertaken in SAS 9.4
using the PROC UCM procedure.
The basic structural model (BSM) is the sum of its stochastic components. Here
I use a trend component µtand a random error componentεt, and it can be
described as:
yt=µt+εt(2)
Each of the model components are modelled separately with the random error εt
modelled as a sequence of independent, identically distributed zero-mean
Gaussian random variables. The trend component is modelled using the
following equations:
µt=µt−1+βt−1+ηt
βt=βt−1+ξt
ηt∼N(0, σ2
η)
ξt∼N(0, σ2
ξ)
(3)
These equations specify a trend where the level µtand the slope βtvary over time,
governed by the variance of the disturbance terms ηtand ξtin their equations. Here
all origin–destination dyadic pairs containing any migration information over the
series were modelled (n=46,203) in individual UCM models.
This approach allows for the projected evolution of the migration system as
dyadic pairs either strengthen or weaken over time, allowing for migration links to
wax or wane over the projection horizon. Empirical simulations of environmental
migration have proven very fruitful in the modelling of climate-change-induced
migration15,55,56, and here I build upon those efforts by projecting future
climate migration.
Our current understanding of the migratory response to sea-level rise is still
underdeveloped. Will displaced coastal populations relocate into the parts of
coastal communities unaffected by sea-level rise? Or will displaced persons migrate
to more inland areas free from the challenges of sea-level rise11,15? Many areas in
threatened coastal communities are still eligible for human settlement and could be
possible destinations for future SLR migrants. To capture both possibilities, I
employ a raking procedure to proportionally adjust in-migrants based on the
inverse of the proportion of the population affected by SLR and a redistribution of
those migrants to unaffected counties.
ˆ
Mt
od =Mt
od ∗(1−Dt
d/Pt
d)(4)
The adjusted number of in-migrants Mto destination county dfrom origin county
oat time tis equal to the number of migrants multiplied by one minus the
proportion of the population impacted by SLR in the destination county at time t.
For inland counties, the right-hand side of the equation equals one, yielding
no adjustment.
ˆ
Mt
od[i=U]=XMt
o−Xˆ
Mt
o∗ ˆ
Mt
od[i=U]
Pˆ
Mt
o[i=U]!(5)
However, the out-migrants from each affected coastal county must be raked to
equal the total population at risk to SLR expected to be displaced. This is
accomplished by redistributing the difference of the unadjusted migrants Mt
ofrom
origin ofrom the adjusted migrants from origin oand multiplying it by the
proportion of adjusted migrants from origin oto destination dto the unaffected
counties (i=U) from the total adjusted migrants from origin o. In this way, the
underlying migration system from each individual origin is preserved in the
raking procedure.
To assess how adaptation might impact future migration streams, the
proportion of the population in households earning greater than $100,000 per year
in the 2011–2015 American Community Survey data were assumed to
be non-migrants.
Projection uncertainty. Evaluation of migration system projection. Projection
intervals allow us to examine the feasibility of the future projected migration
systems and are typically employed in the evaluation of population projections57 .
Demographers have typically used the 2/3 or 66% projection interval to assess the
accuracy of a population projection58,59, representing ‘low’ and ‘high’ scenarios
that are ‘neither so wide as to be meaningless nor too narrow to be
overly-restrictive’60 .
To examine the feasibility of the migration system projections, I produce
projections based on the equations in the preceding section with the base period
1990–2003 and an evaluation period of 2004–2013. If less than 2/3 of the IRS
migration counts fall within the 2/3 projection interval then the results would
suggest less than ideal accuracy. However, if more than 2/3 of the IRS migration
counts fall within the 2/3 projection interval, it would suggest an ideal amount of
accuracy. I assessed the 2/3 interval for 10 years of projections for an evaluation of
base period 1990–2003 and a projection period of 2004–2013.
Supplementary Table 1 shows the overall number of IRS migration counts that
fall within the 2/3 projection interval for each year of the evaluation period.
Overall, the UCMs produce robust projections, as all projection years are above the
2/3 projection interval.
Data availability. The data that support the findings of this study have been
deposited in openICPSR (http://dx.doi.org/10.3886/E100413V3)61 .
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE |www.nature.com/natureclimatechange
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3271 LETTERS
References
38. Hammer, R. B., Stewart, S. I., Winkler, R. L., Radeloff, V. C. & Voss, P. R.
Characterizing dynamic spatial and temporal residential density patterns from
1940 to 1990 across the North Central United States. Landscape Urban Plan. 69,
183–199 (2004).
39. Hauer, M. E., Evans, J. M. & Alexander, C. R. Sea-level rise and sub-county
population projections in coastal Georgia. Popul. Environ. 37,
44–62 (2015).
40. Rogers, A., Little, J. & Raymer, J. The Indirect Estimation of Migration: Methods
for Dealing with Irregular, Inadequate, and Missing Data Vol. 26 (Springer
Science & Business Media, 2010).
41. Engels, R. A. & Healy, M. K. Measuring interstate migration flows—an
origin-destination network based on Internal Revenue Service records.
Environ. Plann. A 13, 1345–1360 (1981).
42. Franklin, R. S. & Plane, D. A. Pandora’s box: the potential and peril of
migration data from the American Community Survey. Intl Reg. Sci. Rev. 29,
231–246 (2006).
43. Lichter, D. T. & Brown, D. L. Rural America in an urban society: changing
spatial and social boundaries. Annu. Rev. Sociol. 37, 565–592 (2011).
44. Jones, B. & O’Neill, B. C. Historically grounded spatial population projections
for the continental United States. Environ. Res. Lett. 8, 044021 (2013).
45. DeWaard, J., Kim, K. & Raymer, J. Migration systems in Europe: evidence from
harmonized flow data. Demography 49, 1307–1333 (2012).
46. Fawcett, J. T. Networks, linkages, and migration systems. Int. Migr. Rev. 23,
671–680 (1989).
47. Fussell, E., Curtis, K. J. & DeWaard, J. Recovery migration to the City of New
Orleans after Hurricane Katrina: a migration systems approach. Popul. Environ.
35, 305–322 (2014).
48. Massey, D. S. et al. An evaluation of international migration theory–the
North-American case. Popul. Dev. Rev. 20, 699–751 (1994).
49. Lee, E. S. Theory of Migration. Demography 3, 47–57 (1966).
50. Pandit, K. Cohort and period effects in US migration: how demographic and
economic cycles influence the migration schedule. Ann. Assoc. Am. Geograph.
89, 439–450 (1997).
51. Haug, S. Migration networks and migration decision-making. J. Ethn. Migr.
Stud. 34, 585–605 (2008).
52. Schultz, J. & Elliott, J. R. Natural disasters and local demographic change in the
United States. Popul. Environ. 34, 293–312 (2013).
53. Gray, C. & Bilsborrow, R. Environmental influences on human migration in
rural Ecuador. Demography 50, 1217–1241 (2013).
54. Harrison, J. & West, M. Bayesian Forecasting Dynamic Models
(Springer, 1999).
55. Reuveny, R. & Moore, W. H. Does environmental degradation influence
migration? Emigration to developed countries in the late 1980s and 1990s. Soc.
Sci. Quart. 90, 461–479 (2009).
56. Feng, S. Z., Krueger, A. B. & Oppenheimer, M. Linkages among climate change,
crop yields and Mexico–US cross-border migration. Proc. Natl Acad. Sci. USA
107, 14257–14262 (2010).
57. Gerland, P. et al . World population stabilization unlikely this century. Science
346, 234–237 (2014).
58. Tayman, J., Smith, S. & Lin, J. Precision, bias, and uncertainty for state
population forecasts: an exploratory analysis of time series models. Popul. Res.
Policy Rev. 26, 347–369 (2007).
59. Swanson, D. A. & Beck, D. M. A new short-term county population projection
method. J. Econ. Soc. Meas. 20, 25–50 (1994).
60. Swanson, D. A. & Tayman, J. Emerging Techniques in Applied Demography
93–117 (Springer, 2015).
61. Hauer, M. Sea Level Rise Induced Migration could Reshape the Population
Landscape (Inter-university Consortium of Political and Social Research, 2017);
http://dx.doi.org/10.3886/E100413V3
NATURE CLIMATE CHANGE |www.nature.com/natureclimatechange
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.