ORE Open Research Exeter
TITLE
An observation-based constraint on permafrost loss as a function of global warming
AUTHORS
Chadburn, SE; Burke, EJ; Cox, PM; et al.
JOURNAL
Nature Climate Change
DEPOSITED IN ORE
22 February 2019
This version available at
http://hdl.handle.net/10871/36030
COPYRIGHT AND REUSE
Open Research Exeter makes this work available in accordance with publisher policies.
A NOTE ON VERSIONS
The version presented here may differ from the published version. If citing, you are advised to consult the published version for pagination, volume/issue and date of
publication
LETTERS
PUBLISHED ONLINE: 10 APRIL 2017 | DOI: 10.1038/NCLIMATE3262
An observation-based constraint on permafrost
loss as a function of global warming
S. E. Chadburn
1,2
*
, E. J. Burke
3
, P. M. Cox
2
, P. Friedlingstein
2
, G. Hugelius
4
and S. Westermann
5
Permafrost, which covers 15 million km
2
of the land surface,
is one of the components of the Earth system that is most
sensitive to warming
1,2
. Loss of permafrost would radically
change high-latitude hydrology and biogeochemical cycling,
and could therefore provide very significant feedbacks on cli-
mate change
3–8
. The latest climate models all predict warming
of high-latitude soils and thus thawing of permafrost under
future climate change, but with widely varying magnitudes of
permafrost thaw
9,10
. Here we show that in each of the models,
their present-day spatial distribution of permafrost and air
temperature can be used to infer the sensitivity of permafrost
to future global warming. Using the same approach for the
observed permafrost distribution and air temperature, we
estimate a sensitivity of permafrost area loss to global mean
warming at stabilization of 4.0
+1.0
−1.1
million km
2
◦
C
−1
(1σ confi-
dence),which is around 20% higher than previousstudies
9
. Our
method facilitates an assessment for COP21 climate change
targets
11
: if the climate is stabilized at 2
◦
C above pre-industrial
levels, we estimate that the permafrost area would eventually
be reduced by over 40%. Stabilizing at 1.5
◦
C rather than 2
◦
C
would save approximately 2 million km
2
of permafrost.
Permafrost, defined as ground that remains at or below 0
◦
C for
two or more consecutive years, underlies 24% of the land in the
Northern Hemisphere
12
. Under recent climate warming, permafrost
has begun to thaw, causing changes in ecosystems and impacting
northern communities, for example through collapse of roads and
buildings as the ground becomes unstable
13
. Large quantities of car-
bon are stored in organic matter in permafrost soils
14
, which starts
to decompose when the permafrost thaws, resulting in the emission
of greenhouse gases such as carbon dioxide and methane. In the
future, carbon release from permafrost thaw may have a signific ant
impact on the Earth’s climate
6
. Due to its global importance, nu-
merous modelling studies have assessed the rate of permafrost thaw
under future climate warming
9,10,15,16
. However, despite progress in
process-based modelling on local and regional scales, for example,
ref. 17, a lack of data avai lability and model limitations mean that
permafrost is still poorly simulated in global climate models, where
the historical simulations show a present-day permafrost area any-
where between 0.1 and 1.8 t imes the size of that observed
9
. Models
often have shallow soil columns, a limited representation of soil
properties, inadequate snow thermal and physical dynamics and
other missing processes
9
. Here we present a projection of large-scale
permafrost thaw that is based on observations, avoiding model bias,
and accounting for observational uncertainty.
Our approach is based on using the relationship between
mean annual air temperature (MAAT) and permafrost occurrence
to estimate permafrost extent. Permafrost is not exclusively
determined by air temperature, being strongly influenced by
landscape features such as topography, soil thermal properties, snow
depth and hydrology
18
. Nonetheless, it is possible to construct a
broad relationship between MAAT and the presence of permafrost,
defined in terms of the probability of finding permafrost at a given
air temperature
19
. Averaged over broad spatial scales, probability
translates to the areal fraction underlain by permafrost. We
derived a MAAT–permafrost relationship using a robust approach
that integrates the spatial distribution of permafrost from the
International Permafrost Associat ion (IPA) map of permafrost in
the Northern Hemisphere
20
.
The observation-based IPA map defines the spatial boundaries of
the permafrost zones: continuous, >90% coverage; discontinuous,
50–90% coverage; sporadic, 10–50% coverage; isolated patches,
0–10% coverage. We took the air temperatures at the spatial
permafrost boundaries and fitted them against the respective
permafrost fractions. The resulting relationship between MAAT
and permafrost fraction is shown in Fig. 1a. We also prov ide a
plausible range, which covers different sources of uncertainty.
Firstly, the range of air temperatures for a given permafrost fraction
indicates variability due to large-scale differences in snow depth,
soil moisture, landscape type and so on; secondly, uncertainties in
the IPA map are incorporated by including air temperatures from
100 km either side of each boundary. Detailed evaluation of this
relationship by validation against local field data and regional mod-
elling suggests t hat it is robust (see Supplementary Information).
We used this relationship between MAAT and permafrost to re-
construct the IPA permafrost map from WATCH reanalysis air tem-
peratures (using the 1960–1990 period, consistent with the IPA-map
observational window)
21,22
(Fig. 1b,c). The estimated permafrost
area is 15.5 million km
2
using this technique (12.0–18.2 million km
2
using minimum/maximum curves), which compares well to
15.0 million km
2
from observations
20
(12.6–18.4 million km
2
). A
spatial correlation between observed and estimated permafrost ex-
tent has an r
2
value of 0.85. Note that this area refers to the actual area
of permafrost, whereas the larger value given in ref. 12 includes the
total area of all permafrost zones (for example, including the whole
sporadic zone, of which only a small fraction is actually permafrost).
Figure 1d,e shows the maximum and minimum permafrost
distributions according to the limiting curves on Fig. 1a. It is
clear that any major discrepancy between the observed distribution
and our estimate is covered by t he maximum and minimum
distributions (see also Supplementary Fig. 8). One of the maj or
causes of such discrepancies is snow, which insulates the ground
in winter
23
. We included the influence of factors such as snow and
1
University of Leeds, School of Earth and Environment, Leeds LS2 9JT, UK.
2
University of Exeter, College of Engineering, Mathematics and Physical
Sciences, Exeter EX4 4QF, UK.
3
Met Oce Hadley Centre, FitzRoy Road, Exeter EX1 3PB, UK.
4
Stockholm University, Department of Physical
Geography, 106 91 Stockholm, Sweden.
5
University of Oslo, Department of Geosciences, PO Box 1047 Blindern, NO-0316 Oslo, Norway.
*
e-mail: s.e.chadburn@exeter.ac.uk
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/NCLIMATE3262
−15 −10 −5 0 5 10
0.0
0.2
0.4
0.6
0.8
1.0
a
MAAT (°C)
Permafrost fraction
Most likely curve
Plausible range
Observations
(ref. 20)
0.0
0.1
0.5
0.9
PF fraction
1.0
0.0
0.1
0.5
0.9
1.0
From air temperatures,
1960−1990
Best fit
MaximumMinimum
From air temperatures,
1960−1990
From air temperatures,
1960−1990
bc
de
Figure 1 | Defining the spatial distribution of observed permafrost as a
function of observed air temperature. a, Relationship between MAAT and
permafrost fraction or probability. The central curve gives the most likely
value, with upper and lower curves giving the plausible range. See
Supplementary Fig. 1 for parameter values. b–e, Permafrost distribution
estimated from reanalysis air temperatures and relationships in a (central
curve (c), lower curve (d), upper curve (e)) validated against the IPA
map (b)
20
.
ground ther mal properties in the limiting curves instead of spatially
resolving them
15,24
, to account for the full range of uncertainties in
future projec tions.
We applied this relationship (Fig. 1a) to make projections of
future permafrost extent. Our approach calculates the committed
permafrost distribution for each global mean temperature. During
a period of warming, the actual changes in permafrost area will
lag behind this quasi-equilibrium state, due to the long timescale
of warming for the deep ground. However, our analysis has high
relevance to international climate negotiations, which are framed
in terms of climate stabilization. We can, for example, estimate
the relative impacts of stabilizing at 1.5
◦
C or 2
◦
C above pre-
industrial levels
11
.
Coupled models provide the best available indication of whether
the relationship between MAAT and permafrost will fundamentally
shift in the future (for example, if there is a pan-Arctic-scale
change in snow depth relative to air temperature). We therefore test
this using the CMIP5 (Coupled Model Intercomparison Project
Phase 5) climate model ensemble
25
, w hich provides a large data set
of coupled simulations. For each model we derive a model-specific
MAAT–per mafrost relationship from the historical simulation
(Supplementary Fig. 3). The robustness of our approach depends
on the extent to which this relationship between permafrost area
and air temperature remains consistent under climate change. The
transferability of the MAAT–permafrost relationship was assessed
by comparing the relationship derived from the models for t he
historical perio d 1960–1990, to that for the period 2270–2300
(Supplementary Fig. 4). The MAAT–permafrost curves for these
two perio ds are generally very similar, and always within our un-
certainty bounds. This is one of the main reasons that our approach
is robust, as it is valid in every case despite the fact that the models
differ in their representation of the key processes and in the details
of t heir projections. We then estimated the future permafrost area,
using the historical MAAT–permafrost relationships and future air
temperatures from each model (Fig. 2a), including nine coupled
climate models used in the latest IPCC (Intergovernmental Panel
on Climate Change) assessment
26
, and two different emission
scenarios. The area is accurately estimated in every case.
We applied the same technique using the ‘true’ observationally
derived MAAT–permafrost curve (Fig. 1a) to make projections of
future permafrost area that are constrained by observations. To
be independent of specific climate models and emission scenarios,
we reduced the f uture air temperature changes down to just
two variables: global mean warming, and Arctic amplification
as a function of latitude. For this we used a pattern-scaling
technique, in which air temperatures are increased by the global
mean warming multiplied by the Arctic amplification. Arctic
amplification is the phenomenon caused by changing surface albedo
due to the melting of snow and ice, in which air temperatures in
the Arctic warm approximately twice as fast as the g lobal mean
26
.
We estimate the amplification factor as a function of latitude,
from the observed historical warming trend (1936–2012), using
the WATCH reanalysis air temperature data
21,22
(Supplementary
Table 2). The observed amplification factor differs substantially
from models
27
(Supplementary Fig. 5), which is a good reason for
using this approach rather than simulated f uture air temperatures
(see Methods for further discussion).
The CMIP5 models were used to test the consistency of this
technique. Using the same information from the mo dels that is
available for the real world (Arctic amplification derived from
historical simulations (1936–2012), and global mean warming), we
estimate the future air temperatures for each model. From thes e
we again use the model-specific MAAT–permafrost relationships
to estimate future permafrost area. This gives projections of future
permafrost area that agree with the simulated permafrost areas
within the uncertainty for all models (Fig. 2b).
We can therefore apply our met hodology using observational
data alone, namely observed present-day air temperature, historical
Arctic amplification, and the observed MAAT–permafrost relation-
ship (Fig. 1a), to estimate global permafrost loss for a given level of
future global warming.
Using our approach, the loss of permafrost under stabilization, as
a function of the global mean warming, is 4.0
+1.0
−1.1
million km
2 ◦
C
−1
(note that all uncertainties are quoted at 1σ level). Under a 1.5
◦
C
stabilization scenario, 4.8
+2.0
−2.2
million km
2
of permafrost would b e
lost compared with the 1960–1990 baseline (corresponding to the
IPA map, Fig. 1b), and under a 2
◦
C stabilization we would lose
6.6
+2.0
−2.2
million km
2
, over 40% of the present-day permafrost area.
Therefore, stabilizing at 1.5
◦
C rather than 2
◦
C could potent ially
prevent approximately 2 million km
2
of permafrost from thawing.
The loss of permafrost with warming is shown on Fig. 3 for a
wide range of scenarios. Our results indicate that for the high
warming scenarios (5 or 6
◦
C above pre-industrial—simil ar to the
warming in RCP8.5 by 2100
26
), the vast majority of permafrost
will thaw, leaving only 0.3–3.1 million km
2
under 5
◦
C of warming
and 0.0–1.5 million km
2
under 6
◦
C. Even accounting for the
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/NCLIMATE3262
LETTERS
0 5 10 15 20 25
0
5
10
15
20
25
a
0
5
10
15
20
25
Future permafrost area from models (million km
2
)
Estimated permafrost area using
local air temperature from models (million km
2
)
CanESM2
MIROC−ESM
MPI−ESM−LR
IPSL−CM5A−LR
CESM1−CAM5
GISS−E2−R
IPSL−CM5A−MR
HadGEM2−ES
NorESM1−M
RCP2.6 at 2300
RCP4.5 at 2300
0
5101520
25
Future permafrost area from models (million km
2
)
Estimated permafrost area using
global air temperature from models (million km
2
)
CanESM2
MIROC−ESM
MPI−ESM−LR
IPSL−CM5A−LR
CESM1−CAM5
GISS−E2−R
IPSL−CM5A−MR
HadGEM2−ES
NorESM1−M
RCP2.6 at 2300
RCP4.5 at 2300
b
Figure 2 | Comparison of our estimate of global permafrost area with that simulated by the CMIP5 models (stabilization runs at 2300). a, Using local air
temperature from the models and the model-specific MAAT–permafrost relationships. b, Using global temperature from the models, Arctic amplification
from each model’s historical simulation and the MAAT–permafrost relationships. Error bars show 2σ confidence.
0
5
10
15
20
Global warming at stabilization (°C)
Permafrost remaining (million km
2
)
0123456
Models
1961−1990 baseline
Figure 3 | Relationship between global warming stabilization scenario and
remaining permafrost area using our approach. Boxes show 1σ and
whiskers show 2σ uncertainty bounds. Zero warming corresponds to
pre-industrial climate (1850–1900 average). The red box corresponds to
the time frame of the IPA permafrost map (Fig. 1b). The ‘model’ points
represent individual CMIP5 climate model stabilization simulations
(permafrost area at 2300).
uncertainties due to heterogeneity in air temperature, snow and so
on, we have greatly reduced the range from the unconstrained model
ensemble (shown on Fig. 3).
Our approach also enables a broad spatial assessment of
permafrost vulnerability, which is difficult with Earth system
models due to problems with their simulation of the current
permafrost distribution
9
. Figure 4 shows the estimated spatial
pattern of high-latitude permafrost historically (1960–1990), and
the range of the zonal boundaries under 1.5
◦
C stabilization
(Fig. 4a) and 2
◦
C stabilization (Fig. 4b). Thawing permafrost has
direct impacts on people and infrastructure in the areas where it
thaws. Thirty-five million people live in the permafrost zone
28,29
,
including in three cities (population >100,000) built on continuous
permafrost (marked on Fig. 4). These cities, for example, would
most likely transition to the discontinuous permafrost zone under
2
◦
C of warming, putting their infrastructure at risk. Hydrological
impacts vary with the depth of thaw but would include localized
ground collapse, lake formation and soil drainage. Note that due
to the nature of our approach, only large-scale spatial patterns of
permafrost thaw are considered.
Previous estimates of permafrost sensitivity were generally given
in terms of high-latitude war ming, rather than global warming.
Previous published values are equivalent to 3.3 ± 1.2 million
km
2 ◦
C
−1
, based on the CMIP5 model simulations
9
, and
1.8–2.6 million km
2 ◦
C
−1
based on an ensemble of offline model
runs
16
. These are smaller than our value of 4.0
+1.0
−1.1
million km
2 ◦
C
−1
(although they fall within 1–2σ of our estimate). The published
values
9,16
are derived from transient simulations, so the difference
may be partly due to the transient effect, where permafrost thaw
‘lags’ behind the climate warming, especially under scenarios such
as RCP8.5 where the air temperature changes very quickly. Indeed,
a study using equilibrium permafrost models driven by CMIP5
model output
10
showed that the equilibrium response is typically
25–38% greater than the transient response, and in some models
the difference was even larger (up to 70%). The major advantage of
the approach adopted here is that committed permafrost loss, along
with its uncertainty, can be estimated for any policy-relevant global
warming scenario.
We estimate the committed permafrost loss over the whole twen-
tieth century to be 3.4
+2.2
−2.3
million km
2
(until 2003–2012
26
). Some
of this committed change will not yet be observable, because of
the lag between the equilibrium and transient response. However,
our estimate of permafrost sensitivity to war ming is consistent
with observations of changes in near-surface permafrost, which
are expected to be much closer to equilibrium (see Supplemen-
tary Figs 6 and 7 and Supplementary Discussion). There may be
longer-term transient effects, but these are relatively small (se e
Supplementary Fig. 2).
This is the first study to quantify permafrost loss under
policy-relevant climate stabilization scenarios, def ine d by the
global warming. In p articular we take an approach that is base d
on observations and independent of climate model projections,
reducing the problem of future sensitivity down to only two
key quantities: Arctic amplification and global mean temperature
change. Furthermore, our constraint includes a comprehensive
uncertainty b ound, specifically giving a sensitivity to global
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/NCLIMATE3262
Changes in permafrost, 1.5 °C warming
a
0.0
0.1
0.5
0.9
1.0
Cities built on continuous permafrost
Changes in permafrost, 2 °C warming
Maximum continuous permafrost
Minimum continuous permafrost
Maximum isolated patches
Minimum isolated patches
b
Figure 4 | Changes in spatial patterns of permafrost under future stabilization scenarios. a,b, The shaded areas show estimated historical permafrost
distribution (1960–1990), and contours show the plausible range of zonal boundaries under 1.5
◦
C stabilization (a) and under 2
◦
C stabilization (b).
warming of 4.0
+1.0
−1.1
million k m
2 ◦
C
−1
at the 1σ level. This provides
an important benchmark for process-based global modelling. Using
our approach we have analysed the difference between 1.5 and
2
◦
C stabilization, and shown that the committed permafrost loss is
nearly 30% smaller at the lower stabilization t arget, with relevance
to climate negotiations surrounding the Paris Agreement
11
.
Methods
Methods, including statements of data availability and any
associated accession codes and references, are available in the
online version of this paper.
Received 15 August 2016; accepted 8 March 2017;
published online 10 April 2017
References
1. Romanovsky, V. et al. in Arctic Report Card 2013 131–136
(NOAA Arctic Program, 2013); ftp://ftp.oar.noaa.gov/arctic/documents/
ArcticReportCard_full_report2013.pdf
2. Romanovsky, V., Burgess, M., Smith, S., Yoshikawa, K. & Brown, J. Permafrost
temperature records: indicators of climate change. Eos 83,
589–594 (2002).
3. Grosse, G., Goetz, S., McGuire, A. D., Romanovsky, V. E. & Schuur, E. A. G.
Changing permafrost in a warming world and feedbacks to the Earth system.
Environ. Res. Lett. 11, 040201 (2016).
4. Schaefer, K., Lantuit, H., Romanovsky, V. E., Schuur, E. A. G. & Witt, R.
The impact of the permafrost carbon feedback on global climate. Environ. Res.
Lett. 9, 085003 (2014).
5. Schneider von Deimling, T. et al. Estimating the near-surface
permafrost-carbon feedback on global warming. Biogeosciences 9,
649–665 (2012).
6. Schuur, E. A. G. et al. Climate change and the permafrost carbon feedback.
Nature 520, 171–179 (2015).
7. MacDougall, A. H., Avis, C. A. & Weaver, A. J. Signif icant contribution to
climate warming from the permafrost carbon feedback. Nat. Geosci. 5,
719–721 (2012).
8. Burke, E. J., Hartley, I. P. & Jones, C. D. Uncertainties in the global temperature
change caused by carbon release from permafrost thawing. Cryosphere 6,
1063–1076 (2012).
9. Koven, C. D., Riley, W. J. & Stern, A. Analysis of permafrost thermal dynamics
and response to climate change in the CMIP5 Earth System Models. J. Clim. 26,
1877–1900 (2013).
10. Slater, A. G. & Lawrence, D. M. Diagnosing present and future permaf rost
from climate models. J. Clim. 26, 5608–5623 (2013).
11. Adoption of The Paris Agreement FCCC/CP/2015/L.9/Rev.1 (UNFCCC, 2015);
http://unfccc.int/resource/docs/2015/cop21/eng/l09r01.pdf
12. Zhang, T., Barry, R. G., Knowles, K., Heginbottom, J. A. & Brown, J. Statistics
and characteristics of permaf rost and ground-ice distribution in the Northern
Hemisphere. Pol. Geogr. 23, 132–154 (1999).
13. Schaefer, K., Lantuit, H., Romanovsky, V. & Schuur, E. A. G. Policy Implications
of Warming Permafrost (United Nations Environment Programme, 2012);
http://wedocs.unep.org/handle/20.500.11822/8533
14. Hugelius, G. et al. Estimated stocks of circumpolar permafrost carbon with
quantified uncertainty ranges and identified data gaps. Biogeosciences 11,
6573–6593 (2014).
15. Anisimov, O. A. & Nelson, F. E. Permafrost zonation and climate change in t he
Northern Hemisphere: results from transient general circulation models.
Climatic Change 35, 241–258 (1997).
16. Chadburn, S. E. et al. Impact of model developments on present and future
simulations of permafrost in a global land-surface model. Cryosphere 9,
1505–1521 (2015).
17. Westermann, S., Østby, T. I., Gisnås, K., Schuler, T. V. & Etzelmüller, B.
A ground temperature map of the North Atlantic permafrost region based on
remote sensing and reanalysis data. Cryosphere 9, 1303–1319 (2015).
18. Jorgenson, M. et al. Resilience and vulnerability of p ermafrost to climate
change. Can. J. For. Res. 40, 1219–1236 (2010).
19. Gruber, S. Derivation and analysis of a high-resolution estimate of global
permafrost zonation. Cryosphere 6, 221–233 (2012).
20. Brown, J., Ferrians, O. J. Jr, Heginbottom, J. & Melnikov, E. Circum-Arctic Map
of Permafrost and Ground Ice Conditions (National Snow and Ice Data Center,
1998); http://nsidc.org/data/docs/fgdc/ggd318_map_circumarctic
21. Weedon, G. P. et al. The Watch Forcing Data 1958–2001: A Meteorological
Forcing Dataset for Land Surface- and Hydrological-Models WATCH Te chnical
Report 22 (WATCH, 2010); http://www.eu-watch.org/publications
22. Weedon, G. P. et al. The WFDEI meteorological forcing data s et: WATCH
forcing data methodology applied to ERA-interim reanalysis data. Wat. Resour.
Res. 50, 7505–7514 (2014).
23. Goodrich, L. The inf luence of snow cover on the ground thermal regime.
Can. Geotech. J. 19, 421–432 (1982).
24. Nelson, F. E. Permafrost distribution in central Canada: applications
of a climate-based predictive model. Ann. Assoc. Am. Geogr. 76,
550–569 (1986).
25. Taylor, K. E., Stouffer, R. J. & Meehl, G. A. A Summary of the CMIP5
Experiment Design PCMDI Tech. Rep. (WCRP, 2009);
http://cmip-pcmdi.llnl.gov/cmip5/docs/Taylor_CMIP5_design.pdf
26. IPCC Climate Change 2013: The Physical Science Basis (eds Stocker, T. et al.)
(Cambridge Univ. Press, 2013).
27. Xie, Y., Liu, Y. & Huang, J. Overestimated Arctic warming and underestimated
Eurasia mid-latitude warming in CMIP5 simulations. Int. J. Climatol. 36,
4475–4487 (2016).
4
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange
