Content uploaded by Farrah Powell
Author content
All content in this area was uploaded by Farrah Powell on Oct 03, 2017
Content may be subject to copyright.
LETTERS
PUBLISHED ONLINE: 19 JUNE 2017 | DOI: 10.1038/NCLIMATE3322
Global risk of deadly heat
Camilo Mora1*, Bénédicte Dousset2, Iain R. Caldwell3, Farrah E. Powell1, Rollan C. Geronimo1,
Coral R. Bielecki4, Chelsie W. W. Counsell3, Bonnie S. Dietrich5, Emily T. Johnston4, Leo V. Louis4,
Matthew P. Lucas6, Marie M. McKenzie1, Alessandra G. Shea1, Han Tseng1, Thomas W. Giambelluca1,
Lisa R. Leon7, Ed Hawkins8and Clay Trauernicht6
Climate change can increase the risk of conditions that exceed
human thermoregulatory capacity1–6. Although numerous stud-
ies report increased mortality associated with extreme heat
events1–7, quantifying the global risk of heat-related mortality
remains challenging due to a lack of comparable data on
heat-related deaths2–5. Here we conducted a global analysis
of documented lethal heat events to identify the climatic
conditions associated with human death and then quantified
the current and projected occurrence of such deadly climatic
conditions worldwide. We reviewed papers published between
1980 and 2014, and found 783 cases of excess human
mortality associated with heat from 164 cities in 36 countries.
Based on the climatic conditions of those lethal heat events,
we identified a global threshold beyond which daily mean
surface air temperature and relative humidity become deadly.
Around 30% of the world’s population is currently exposed
to climatic conditions exceeding this deadly threshold for at
least 20 days a year. By 2100, this percentage is projected
to increase to ∼48% under a scenario with drastic reductions
of greenhouse gas emissions and ∼74% under a scenario of
growing emissions. An increasing threat to human life from
excess heat now seems almost inevitable, but will be greatly
aggravated if greenhouse gases are not considerably reduced.
Sporadic heat events, lasting days to weeks, are often related to
increased human mortality1,2, raising serious concerns for human
health given ongoing climate change1–3,8–16 . Unfortunately, a number
of challenges have hampered global assessments of the risk of
heat-related death. First, heat illness (that is, severe exceedance
of the optimum body core temperature) is often underdiagnosed
because exposure to extreme heat often results in the dysfunction
of multiple organs, which can lead to misdiagnosis2,3,5,17 . Second,
mortality data from heat exposure are sparse and have not been
analysed in a consistent manner. Here we conducted a global
survey of peer-reviewed studies on heat-related mortality to identify
the location and timing of past events that caused heat-related
deaths. We used climatic data during those events to identify the
conditions most likely to result in human death and then quantified
the current and projected occurrence of such deadly climatic
conditions. Hereafter, we use ‘lethal’ when referring to climatic
conditions during documented cases of excess mortality and ‘deadly’
when referring to climatic conditions that are projected to cause
death. We make this distinction to acknowledge that climatic
conditions which have killed people in the past are obviously capable
of causing death, but whether or not they result in human mortality
in the future could be affected by adaptation. We do not quantify
human deaths per se because the extent of human mortality will
be considerably modified by social adaptation (for example, use of
air conditioning, early warning systems, and so on18–20 ). Although
social adaptation could reduce the exposure to deadly heat18–20, it
will not affect the occurrence of such conditions. Given the speed
of climatic changes and numerous physiological constraints, it is
unlikely that human physiology will evolve the necessary higher
heat tolerance21,22, highlighting that outdoor conditions will remain
deadly even if social adaptation is broadly implemented. Our aim is
to quantify where and when deadly heat conditions occur, which in
turn can provide important information on where social adaptation
will likely be needed.
We searched available online databases for peer-reviewed
publications on heat-related mortality published between 1980 and
2014 (see Methods). From over 30,000 relevant references, we
identified 911 papers that included data on 1,949 case studies
of cities or regions where excess mortality was associated with
high temperatures. Case studies were broadly grouped into those
focusing on temperature–mortality relationships in a specific city,
region, or country (1,166 cases from 273 cities across 49 countries)
and those focusing on heat-related mortality during specific
episodes (783 cases from 164 cities across 36 countries). Cases
were predominantly reported for cities at mid-latitudes, with the
highest concentration in North America and Europe (Fig. 1a), and
included well-documented heatwaves like those in Chicago in 1995
(∼740 deaths23), Paris in 2003 (∼4,870 deaths24), Moscow in 2010
(∼10,860 deaths25) and many other, less publicized events (list
of cases provided at https://maps.esri.com/globalriskofdeadlyheat).
While data on the number of deaths was inconsistently reported, all
studies provided information on the place and dates when climatic
conditions were lethal, which we used to identify the specific
climatic conditions resulting in heat-related mortality.
To identify the climatic conditions related to lethal heat events,
we assessed daily climatic data (that is, surface air temperature,
relative humidity, solar radiation, wind speed, and several other
metrics, Supplementary Fig. 1) for the duration of lethal heat
episodes reported in the literature and an equal number of non-
lethal episodes (that is, periods of equal duration from the same
cities but from randomly selected dates); then we used Support
1Department of Geography, University of Hawai’i at M¯
anoa, Honolulu, Hawai’i 96822, USA. 2Hawai‘i Institute of Geophysics and Planetology, University of
Hawai‘i at M¯
anoa, Honolulu, Hawai’i 96822, USA. 3Hawai‘i Institute of Marine Biology, University of Hawai‘i at M¯
anoa, K¯
ane‘ohe, Hawai’i 96744, USA.
4Department of Botany, University of Hawai‘i at M¯
anoa, Honolulu, Hawai’i 96822, USA. 5Department of Plant and Environmental Protection Sciences,
University of Hawai‘i at M¯
anoa, Honolulu, Hawai’i 96822, USA. 6Department of Natural Resources and Environmental Management, University of Hawai‘i
at M¯
anoa, Honolulu, Hawai’i 96822, USA. 7Thermal and Mountain Medicine Division, U.S. Army Research Institute of Environmental Medicine,
Natick, Massachusetts 01760, USA. 8National Centre for Atmospheric Science, Department of Meteorology, University of Reading, Reading,
Berkshire RG6 6BB, UK. *e-mail: cmora@hawaii.edu
NATURE CLIMATE CHANGE | VOL 7 | JULY 2017 | www.nature.com/natureclimatechange 501
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTERS NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3322
Deadly
01020304050
0
20
40
60
80
100
Average daily temperature (°C)
Average daily relative humidity (%)
a b
Figure 1 | Geographical distribution of recent lethal heat events and their climatic conditions. a, Places where relationships between heat and mortality
have been documented (red squares) and where specific heat episodes have been studied (blue squares). b, Mean daily surface air temperature and
relative humidity during lethal heat events (black crosses) and during periods of equal duration from the same cities but from randomly selected dates
(that is, non-lethal heat events; red to yellow gradient indicates the density of such non-lethal events). Blue line is the SVM threshold that best separates
lethal and non-lethal heat events and the red line is the 95% probability SVM threshold; areas to the right of the thresholds are classified as deadly and
those to the left as non-deadly. Support vectors for other variables are shown in Supplementary Fig. 2.
1950 2000 2050 2100
0
10
20
30
40
50
60
Global land area (%)
1950 2000 2050 2100
0
20
40
60
80
100
Global human population (%)
Historical
RCP 8.5
RCP 4.5
RCP 2.6
a
YearYear
Reanalysis data b
Figure 2 | Current and projected changes in deadly climatic conditions. a,b, Area of the planet (a) and percentage of human population (b) exposed to
climatic conditions beyond the 95% SVM deadly threshold (red line in Fig. 1b) for at least 20 days in a year under alternative emission scenarios. Bold lines
are the multimodel medians, black lines are the results from reanalysis data and faded lines indicate the projections for each Earth System Model. Time
series were smoothed with a 10-year-average moving window. Area of the planet and human population exposed to dierent lengths of time are shown in
Supplementary Fig. 4. Results correcting for climatological mean biases between the reanalysis data and each Earth System Model are shown in
Supplementary Figs 8 and 10.
Vector Machines (SVMs) to identify the climatic conditions that
best differentiated lethal and non-lethal episodes. SVMs generate
a threshold that maximizes the difference in the attributes of two
or more groups, allowing for classification of objects in either
group based on where their given attributes fall with respect to
the threshold. In our case, SVM was used to generate a decision
threshold that maximizes the difference in climatic conditions of
lethal and non-lethal episodes, with the conditions on one side of the
threshold being lethal and those to the other side being non-lethal
(for example, Fig. 1b). Among all possible pair combinations of the
variables analysed here (Supplementary Figs 1 and 2), the SVM
using mean daily surface air temperature and relative humidity most
accurately distinguished between past lethal and non-lethal heat
episodes (that is, 82%, blue line in Fig. 1b); accuracy was measured
as the ratio of the number of correctly classified lethal and non-
lethal cases to the total number of cases. Adding other variables
to the temperature–humidity SVM resulted in less parsimonious
SVMs with minimal increases in accuracy (for example, the SVM
model including all 16 variables analysed here was only 3% more
accurate, Supplementary Fig. 3). SVM also allows for estimation
of a classification probability that increases with the distance of an
observation to the decision threshold; the use of a 95% probability
for the temperature–humidity SVM (red line in Fig. 1b) resulted
in 100% accurate predictions of true positives (that is, only prior
lethal heat episodes were on the deadly side of the 95% probability
SVM decision boundary). While our analysis used data on local
climatic conditions, the resulting pattern between temperature
and relative humidity allowed us to accurately classify lethal heat
events of different cities worldwide using a single common SVM
threshold (Fig. 1b).
The fact that temperature and relative humidity best predict
times when climatic conditions become deadly is consistent with
human thermal physiology, as they are both directly related to
body heat exchange2–4. First, the combination of an optimum body
core temperature (that is, ∼37 ◦C), the fact that our metabolism
generates heat (∼100 W at rest) and that an object cannot dissipate
502
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | VOL 7 | JULY 2017 | www.nature.com/natureclimatechange
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3322 LETTERS
0 50 100 150 200 250 300 350
Number of days per year above deadly threshold
Historical
a
RCP 2.6
RCP 4.5
RCP 8.5
b
c
d
Figure 3 | Geographical distribution of deadly climatic conditions under
dierent emission scenarios. a–d, Number of days per year exceeding the
threshold of temperature and humidity beyond which climatic conditions
become deadly (Fig. 1b), averaged between 1995 and 2005 (a, historical
experiment), and between 2090 and 2100 under RCP 2.6 (b), RCP 4.5 (c)
and RCP 8.5 (d). Results are based on multimodel medians. Grey areas
indicate locations with high uncertainty (that is, the multimodel standard
deviation was larger than the projected mean; coecient of variance >1).
The expected lower number of deadly days at higher latitudes (Fig. 4) may
help explain the large variability among Earth System Models in the
projected number of deadly days at higher latitudes31 (for example, in the
case for New York (illustrated in Fig. 4j) the one model projects nine deadly
days by 2100; yet any other model projecting 18 days will double the
variability). The uncertainty presented in this figure should be interpreted
with that caution in mind.
heat to an environment with equal or higher temperature (that is,
the second law of thermodynamics22), dictates that any ambient
temperature above 37◦C should result in body heat accumulation
and a dangerous exceedance of the optimum body core temperature
(hyperthermia5). Second, sweating, the main process by which the
body dissipates heat, becomes ineffective at high relative humidity
(that is, air saturated with water vapour prevents evaporation of
sweat); therefore, body heat accumulation can occur at temperatures
lower than the optimum body core temperature in environments
of high relative humidity. These properties help to explain why
the boundary at which temperature becomes deadly decreases
with increasing relative humidity (Fig. 1b) and why in our results
some heat mortality events occurred at relatively low temperatures
(Fig. 1b). These consequences of temperature and humidity are why
both of these variables are included in traditional thermal indices
such as humidex26 and wet-bulb globe temperature22,27 .
To quantify the global extent of current deadly climatic
conditions, we applied the 95% probability SVM decision boundary
between mean daily surface air temperature and relative humidity
(red line in Fig. 1b, hereafter referred to as deadly threshold)
to current global climate data (see Methods). Using data from
a climate reanalysis (see Methods), we found that in 2000,
∼13.2% of the planet’s land area, where ∼30.6% of the world’s
human population resides, was exposed to 20 or more days
when temperature and humidity surpassed the threshold beyond
which such conditions become deadly (Fig. 2, extended results in
Supplementary Fig. 4). Comparatively, using climate simulations
for the year 2000 (that is, historical experiment) developed for
the Coupled Model Intercomparison Project phase 5 (CMIP5), we
found that ∼16.2% (±8.3% standard deviation, s.d.) of the planet’s
land area, where ∼37.0% (±9.7% s.d.) of the world’s population
resides, was exposed to 20 or more days of potentially deadly
conditions of temperature and humidity (results are multimodel
medians and standard deviations among Earth System Models;
Fig. 2). Both the reanalysis and historical CMIP5 data revealed
increasing trends in the area and population exposed to deadly
climates during the time period for which such datasets can be
compared, although the trends in the reanalysis data are slightly
weaker than in the Earth System Models (Fig. 2). Overall, there
was ∼3% mismatch in the area of the planet exposed to deadly
climates (∼6.4% in global population) between the reanalysis
and the multimodel median, and thus, results based on CMIP5
simulations should be interpreted with that error in mind. However,
the effects of this mismatch and the uncertainty among Earth
System Models were smaller than the predicted changes in deadly
days (Supplementary Fig. 10). It is worth noting that most scientific
publications on deadly heat events have focused in developed mid-
latitude countries (Fig. 1a); yet, deadly heat conditions also occur in
developing tropical countries (Fig. 3). This suggests that the risk of
deadly heat could be currently underestimated in tropical regions,
which has been noted in prior studies28.
To predict the global extent of future deadly climates, we applied
the deadly SVM threshold to mean daily surface air temperature and
relative humidity projections from the CMIP5 Earth System Models
under low, moderate, and high emissions scenarios (Representative
Concentration Pathways, RCPs, 2.6, 4.5, and 8.5, respectively). We
found that by 2100, even under the most aggressive mitigation
scenario (that is, RCP 2.6), ∼26.9% (±8.7% s.d.) of the world’s
land area will be exposed to temperature and humidity conditions
exceeding the deadly threshold by more than 20 days per year,
exposing ∼47.6% (±9.6% s.d.) of the world’s human population to
deadly climates (using Shared Socioeconomic Pathways projections
of future human population29 relevant to each of the CMIP5 RCPs,
see Methods). Scenarios with higher emissions will affect an even
greater percentage of the global land area and human population.
By 2100, ∼34.1% (±7.6% s.d.) and ∼47.1% (±8.9% s.d.) of the
NATURE CLIMATE CHANGE | VOL 7 | JULY 2017 | www.nature.com/natureclimatechange
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
503
LETTERS NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3322
Deadly
2 d 9 d 27 d 50 d
Deadly
Distance from
the deadly threshold (unitless)
Distance from
the deadly threshold (unitless)
Distance from
the deadly threshold (unitless)
Distance from
the deadly threshold (unitless)
01020304050
0
20
40
60
80
100
01020304050
0
20
40
60
80
100
01020304050
0
20
40
60
80
100
01020304050
0
20
40
60
80
100
Average daily
relative humidity (%)
01020304050
0
20
40
60
80
100
01020304050
0
20
40
60
80
100
01020304050
0
20
40
60
80
100
01020304050
0
20
40
60
80
100
Average daily
relative humidity (%)
Jakarta
−40 −30 −20 −10 0 10 −40 −30 −20 −10 0 10 −40 −30 −20 −10 0 10 −40 −30 −20 −10 0 10
−40
−20
0
20
40
60
525 25 5
50
aHistorical RCP 2.6 RCP 4.5 RCP 8.5
Days in a year (%)
Latitude (° N)
−40
−20
0
20
40
60
−40
−20
0
20
40
60
−40
−20
0
20
40
60
New York
+1.6 °C +2.7 °C +5.5 °C
11 d 117 d 274 d 365 d
+0.9 °C +1.7 °C +3.8 °C
Deadly
Average daily temperature (°C) Average daily temperature (°C) Average daily temperature (°C) Average daily temperature (°C)
Average daily temperature (°C) Average daily temperature (°C) Average daily temperature (°C) Average daily temperature (°C)
bcd
efgh
ijkl
Figure 4 | Latitudinal risk of deadly climates. a–d, Distribution of the percentage of days in a given year (that is, colour gradients), at each latitude, as a
function of their distance to the deadly threshold (red line in Fig. 1b). Displayed here are the last year in the historical experiment (that is, 2005; a) and the
year 2100 under RCP 2.6 (b), RCP 4.5 (c) and RCP 8.5 (d). These plots illustrate that higher latitudes have fewer days near the deadly threshold compared
with the tropics. e–l, As examples, we show mean temperature and relative humidity for each day in the year 2005 in the historical experiments and the
year 2100 for all the RCPs in Jakarta (e–h) and New York (i–l), with consecutive days connected by lines. The 95% SVM threshold is shown as a red line
with numbers on the upper right hand corner indicating the number of days that cross the threshold and the dierence in temperature between 2100 and
2005. Examples are based on a single simulation of a randomly chosen model (that is, CSIRO-Mk3-6-0).
global land area will be exposed to temperature and humidity
conditions that exceed the deadly threshold for more than 20 days
per year under RCP 4.5 and RCP 8.5, respectively; this will expose
∼53.7% (±8.7% s.d.) and ∼73.9% (±6.6% s.d.) of the world’s human
population to deadly climates by the end of the century (Fig. 2,
extended results in Supplementary Fig. 4).
The projected number of days per year surpassing the deadly
threshold increases from mid-latitudes to the equator (Figs 4a–c,
5a and Supplementary Fig. 5a,d,g). By 2100, mid-latitudes (for
example, 40◦N or S) will be exposed to ∼60 deadly days per year
compared to almost the entire year in humid tropical areas under
RCP 8.5 (Figs 3b–d, 4b–d and 5a). This latitudinal pattern was con-
sistent among all scenarios (Supplementary Fig. 5a,d,g) and is largely
determined by the fact that the number of days with temperatures
close to the deadly threshold declines with increasing latitude (that
is, due to greater seasonality; Supplementary Fig. 6b–d28). At mid-
latitudes (for example, New York, Fig. 4i–l) temperatures approach
the deadly threshold only during the summer, which represents
a smaller proportion of the year; compared to tropical locations
(for example, Jakarta, Fig. 4e–h), which have consistently warm
temperatures near the deadly threshold year-round (Supplementary
Fig. 6). Although tropical humid areas will experience less warming
than higher latitudes (Fig. 5b, see also ref. 30), they will be exposed
to the greatest increase in the number of deadly days over time,
because higher relative humidity in tropical areas requires lower
temperatures to cross the deadly threshold (Figs 4e–h and 5e); a
condition that could be further aggravated by projected increases
in relative humidity of tropical areas (Fig. 5a). Subtropical and
504
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | VOL 7 | JULY 2017 | www.nature.com/natureclimatechange
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3322 LETTERS
Latitude (° N)
a
1960 2000 2040 2080
−40
−20
0
20
40
60
Change in deadly days
−300 −200 −100 0 100 200 300
Change in temperature (°C)
−4 4−2 20
Change in relative humidity (%)
−10 10−5 50
Temperature (°C)
26 28 30 32 34
Relative humidity (%)
50 60 70 80 90 100
Year
Latitude (° N)
1960 2000 2040 2080
−40
−20
0
20
40
60
Year
Latitude (° N)
1960 2000 2040 2080
−40
−20
0
20
40
60
Year
Latitude (° N)
1960 2000 2040 2080
−40
−20
0
20
40
60
Year
Latitude (° N)
1960 2000 2040 2080
−40
−20
0
20
40
60
Year
de
bc
Figure 5 | Simulated spatio-temporal changes in deadly climatic conditions in Earth System Models. a, Average changes over time in the number of days
per year exceeding the deadly threshold. b,c, Changes in temperature (b) and changes in relative humidity (c) during those deadly days, relative to mean
values between 1995 and 2005. d,e, Mean temperature (d) and relative humidity (e) during deadly days. Results are grouped by latitude and are based on
the multimodel medians for the historical experiment, which runs from 1950 to 2005, and RCP 8.5, which runs from 2006 to 2100. Results for all scenarios
are shown in Supplementary Fig. 5.
mid-latitude areas will have fewer days beyond the deadly threshold,
but such deadly days will be much hotter in the future (Figs 4e–h
and 5b,d). This general variability in the climatic conditions of
deadly days (Fig. 5b–d and Supplementary Fig. 7) is probably related
to mean global climate patterns associated with the general cir-
culation of the atmosphere: equatorial convection (that is, warm,
moist air rising) produces high humidity in low latitudes whereas
subtropical atmospheric subsidence (that is, cool, dry air sinking)
creates low-precipitation, low-humidity zones, where high sensible
heat flux contributes to extreme high temperatures at mid-latitudes
(Supplementary Figs 5i and 7).
Our study underscores the current and increasing threat to
human life posed by climate conditions that exceed human ther-
moregulatory capacity. Lethal heatwaves are often mentioned as a
key consequence of ongoing climate change, with reports typically
citing past major events such as Chicago in 1995, Paris in 2003,
or Moscow in 20101–6. Our literature review indicates, however,
that lethal heat events already occur frequently and in many more
cities worldwide than suggested by these highly cited examples.
Our analysis shows that prior lethal heat events occurred beyond
a general threshold of combined temperature and humidity, and
that today nearly one-third of the world’s population is regularly
exposed to climatic conditions surpassing this deadly threshold. The
area of the planet and fraction of the world’s human population
exposed to deadly heat will continue to increase under all emission
scenarios, although the risk will be much greater under higher
emission scenarios. By 2100, almost three-quarters of the world’s
human population could be exposed to deadly climatic conditions
under high future emissions (RCP 8.5) as opposed to one-half
under strong mitigation (RCP 2.6). While it is understood that
higher latitudes will undergo more warming than tropical regions30 ,
our results suggest that tropical humid areas will be dispropor-
tionately exposed to more days with deadly climatic conditions
(Fig. 5a), because these areas have year-round warm temperatures
and higher humidity, thus requiring less warming to cross the deadly
threshold (Fig. 4 and Supplementary Fig. 6). The consequences
of exposure to deadly climatic conditions could be further aggra-
vated by an ageing population (that is, a sector of the popula-
tion highly vulnerable to heat2–4) and increasing urbanization (that
is, exacerbating heat-island effects2–4). Our paper emphasizes the
importance of aggressive mitigation to minimize exposure to deadly
climates and highlights areas of the planet where adaptation will be
most needed.
Methods
Methods, including statements of data availability and any
associated accession codes and references, are available in the
online version of this paper.
Received 2 June 2016; accepted 17 May 2017;
published online 19 June 2017
NATURE CLIMATE CHANGE | VOL 7 | JULY 2017 | www.nature.com/natureclimatechange
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
505
LETTERS NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3322
References
1. Patz, J. A., Campbell-Lendrum, D., Holloway, T. & Foley, J. A. Impact of
regional climate change on human health. Nature 438, 310–317 (2005).
2. Basu, R. & Samet, J. M. Relation between elevated ambient temperature and
mortality: a review of the epidemiologic evidence. Epidemiol. Rev. 24,
190–202 (2002).
3. Kovats, R. S. & Hajat, S. Heat stress and public health: a critical review.
Annu. Rev. Publ. Health 29, 41–55 (2008).
4. Leon, L. R. Pathophysiology of Heat Stroke Vol. 7 (Colloquium Series on
Integrated Systems Physiology : From Molecule to Function to Disease,
Morgan Claypool Life Sciences, 2015).
5. Ostro, B. D., Roth, L. A., Green, R. S. & Basu, R. Estimating the mortality effect
of the July 2006 California heat wave. Environ. Res. 109, 614–619 (2009).
6. Glaser, J. et al . Climate change and the emergent epidemic of CKD from heat
stress in rural communities: the case for heat stress nephropathy. Clin. J. Am.
Soc. Nephrol. 11, 1472–1483 (2016).
7. Robine, J.-M. et al. Death toll exceeded 70,000 in Europe during the summer of
2003. C. R. Biol. 331, 171–178 (2008).
8. Sillmann, J. & Roeckner, E. Indices for extreme events in projections of
anthropogenic climate change. Climatic Change 86, 83–104 (2008).
9. Meehl, G. A. & Tebaldi, C. More intense, more frequent, and longer lasting heat
waves in the 21st century. Science 305, 994–997 (2004).
10. Orlowsky, B. & Seneviratne, S. Global changes in extreme events: regional and
seasonal dimension. Climatic Change 110, 669–696 (2012).
11. Tebaldi, C., Hayhoe, K., Arblaster, J. M. & Meehl, G. A. Going to the extremes.
Climatic Change 79, 185–211 (2006).
12. Tebaldi, C. & Wehner, M. F. Benefits of mitigation for future heat
extremes under RCP4.5 compared to RCP8.5. Climatic Change
http://dx.doi.org/10.1007/s10584-016-1605-5 (2016).
13. Sterl, A. et al. When can we expect extremely high surface temperatures?
Geophys. Res. Lett. 35, L14703 (2008).
14. Huang, C. et al. Projecting future heat-related mortality under climate change
scenarios: a systematic review. Environ. Health Persp. 119, 1681–1690 (2011).
15. Guo, Y. et al. Global variation in the effects of ambient temperature on
mortality: a systematic evaluation. J. Epidemiol. 25, 781–789 (2014).
16. Luber, G. & McGeehin, M. Climate change and extreme heat events. Am. J.
Prev. Med. 35, 429–435 (2008).
17. Bouchama, A. & Knochel, J. P. Heat stroke. New. Engl. J. Med. 346,
1978–1988 (2002).
18. Bobb, J. F., Peng, R. D., Bell, M. L. & Dominici, F. Heat-related mortality and
adaptation to heat in the United States. Environ. Health Persp. 122,
811–816 (2014).
19. Gasparrini, A. et al. Temporal variation in heat–mortality associations:
a multicountry study. Environ. Health Persp. 123, 1200–1207 (2015).
20. Lowe, D., Ebi, K. L. & Forsberg, B. Heatwave early warning systems and
adaptation advice to reduce human health consequences of heatwaves. Int. J.
Environ. Res. Public Health 8, 4623–4648 (2011).
21. Hanna, E. G. & Tait, P. W. Limitations to thermoregulation and acclimatization
challenge human adaptation to global warming. Int. J. Environ. Res. Publ.
Health. 12, 8034–8074 (2015).
22. Sherwood, S. C. & Huber, M. An adaptability limit to climate change due to
heat stress. Proc. Natl Acad. Sci. USA 107, 9552–9555 (2010).
23. Whitman, S. et al. Mortality in Chicago attributed to the July 1995 heat wave.
Am. J. Public Health 87, 1515–1518 (1997).
24. Dousset, B. et al. Satellite monitoring of summer heat waves in the Paris
metropolitan area. Int. J. Climatol. 31, 313–323 (2011).
25. Shaposhnikov, D. et al . Mortality related to air pollution with the Moscow heat
wave and wildfire of 2010. Epidemiology 25, 359–364 (2014).
26. Barnett, A. G., Tong, S. & Clements, A. What measure of temperature is the
best predictor of mortality? Environ. Res. 110, 604–611 (2010).
27. Willett, K. M. & Sherwood, S. Exceedance of heat index thresholds for
15 regions under a warming climate using the wet-bulb globe temperature. Int.
J. Climatol. 32, 161–177 (2012).
28. Argüeso, D., Di Luca, A., Perkins-Kirkpatrick, S. & Evans, J. P. Seasonal
mean temperature changes control future heatwaves. Geophys. Res. Lett. 43,
7653–7660 (2016).
29. Jones, B. & O’Neill, B. Spatially explicit global population scenarios consistent
with the Shared Socioeconomic Pathways. Environ. Res. Lett. 11, 084003 (2016).
30. Diffenbaugh, N. S. & Field, C. B. Changes in ecologically critical terrestrial
climate conditions. Science 341, 486–492 (2013).
31. Mitchell, D. et al. Attributing human mortality during extreme heat waves to
anthropogenic climate change. Environ. Res. Lett. 11, 074006 (2016).
Acknowledgements
We thank the Gridded Human Population of the World Database and the National
Center for Environmental Prediction and Department of Defense reanalysis database for
making their data openly available and B. Jones for sharing human population
projections. We acknowledge the World Climate Research Programme’s Working Group
on Coupled Modelling, which is responsible for CMIP5, and thank the climate modelling
groups (listed in Supplementary Table 1) for producing and making available their model
outputs. We also thank D. Schanzenbach, S. Cleveland and R. Merrill from the University
of Hawai’i Super Computer Facility for allowing access to computing facilities and
Hawai’i SeaGrant for providing funds to acquire some of the computers used in these
analyses. Q. Chen, A. Smith, C. Dau, R. Fang and S. Seneviratne provided valuable
contributions to the paper. The opinions or assertions contained herein are the private
views of the authors and are not to be construed as official or as reflecting the views of the
Army or the Department of Defense. We thank R. Carmichael, M. Deaton, D. Johnson
and M. Smith in ESRI’s Applications Prototype Lab for the creation of the online mapping
application. This paper was developed as part of the graduate course on ‘Methods for
Large-Scale Analyses’ in the Department of Geography, University of Hawai’i at M¯
anoa.
Author contributions
All authors contributed to the design of the paper. C.M., B.D., I.R.C., F.E.P., R.C.G.,
C.R.B., C.W.W.C., B.S.D., E.T.J., L.V.L., M.P.L., M.M.M., A.G.S., H.T. and C.T. collected
data. C.M. and I.R.C. performed analysis. All authors contributed to the writing of
the paper.
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. Correspondence and requests for materials should be
addressed to C.M.
Competing financial interests
The authors declare no competing financial interests.
506
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | VOL 7 | JULY 2017 | www.nature.com/natureclimatechange
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3322 LETTERS
Methods
Survey of published cases of heat-related mortality. We searched for
peer-reviewed studies published between 1980 and 2014 on heat-related mortality
in Google Scholar, PubMed, and the Web of Science using the following keywords:
(human OR people) AND (mortality OR death OR lethal) AND (heat
OR temperature). We searched for papers primarily in English, but also included
papers in Spanish, French, Japanese and Chinese when found. We reviewed the
titles and abstracts of the first 30,000 citations in Google Scholar and all citations
from other databases and selected any peer-reviewed publications on heat-related
human mortality (we also searched for additional sources in the references). These
efforts resulted in 911 peer-reviewed papers from which we collected information
on the place and dates of lethal heat events. Several papers noted that human
mortality may have occurred beyond the dates in which the extreme climatic
conditions occurred (‘mortality displacement’); in those cases, we extracted the
dates for which the extreme climatic conditions were reported in the given studies.
Our goal was to identify the dates in which climatic conditions triggered human
mortality regardless of whether mortality was displaced or not.
Climatic conditions related to prior cases of heat-related mortality. For the cases
in the literature review that reported the place and time of lethal heat events, we
assessed information for 16 climatic metrics based on mean daily surface air
temperature, relative humidity, solar radiation, and wind speed (Supplementary
Fig. 1). For each of the lethal heat events, we also assessed the same climatic
variables for a paired ‘non-lethal’ event of the same duration and from the same
city but from a randomly chosen date. Climatic conditions were characterized
using daily data from an atmospheric reanalysis of past climate (NCEP-DOE
Reanalysis 2). We used the NCEP-DOE Reanalysis database because it is among the
most studied and is well characterized relative to newer databases. We used
Support Vector Machine (SVM) modelling to separate the climatic conditions
associated with prior lethal heat events from those associated with non-lethal
events. Using SVM, we generated a decision vector/threshold that maximized the
distance between lethal and non-lethal episodes, with the conditions on one side of
the threshold being lethal and those to the other side being non-lethal (for example,
Fig. 1b). We developed such SVM models for all combinations of the variables
collected and then compared the accuracy of models to choose the most
parsimonious and best performing one.
Projected occurrence of deadly climatic conditions. To quantify the number of
days in a year that surpass the threshold beyond which conditions become deadly
under alternative emission scenarios, we applied the 95% SVM probability
threshold between mean daily surface air temperature and relative humidity of
prior lethal heat events to daily climate projections of the same variables. We used
the 95% SVM probability threshold because it resulted in a much more accurate
classification of prior lethal heat events, and because it restricts projected lethal
heat events to much more extreme conditions, hence yielding more conservative
results. We used daily climate projections of mean surface air temperature and
relative humidity from 20 Earth System Models under four alternative emissions
scenarios developed for the recent Coupled Model Intercomparison Project Phase 5
(Supplementary Table 1). We used the ‘historical’ experiment, which includes the
period from 1950 to 2005 and the Representative Concentration Pathways 2.6, 4.5
and 8.5 (RCP 2.6, 4.5 and 8.5, respectively), which include the period from 2006 to
2100. The historical experiment was designed to model recent climate (reflecting
changes due to both anthropogenic and natural causes) and allows the validation of
model outputs against available climate observations (Supplementary Figs 8 and 9).
RCP pathways represent contrasting mitigation efforts between rapid greenhouse
gas reductions (RCP 2.6) and a business-as-usual scenario (RCP 8.5). All analyses
were run at the original resolution of each climate database and the results were
interpolated to a common 1.5◦grid cell size using a bilinear function.
Projections of global land coverage and risk to human populations from deadly
climatic conditions. To calculate the amount of land area and fraction of the
human population that are likely to be exposed to deadly climates each year, we
summed the land area and human population for all cells experiencing varying
numbers of days in a year beyond the deadly threshold (Fig. 2 and Supplementary
Fig. 4). We used the Gridded Population of the World from the Socioeconomic
Data and Applications Center (http://sedac.ciesin.columbia.edu/data/set/gpw-v3-
population-count-future-estimates/data-download#) to estimate human exposure
up to the year 2005 and human population projections consistent with the different
emission scenarios used in the CMIP5 to estimate exposure between 2006 and
2100. For the population projections, we specifically used the spatially explicit
global population scenarios consistent with the Shared Socioeconomic Pathways
(SSP) developed by Jones et al.29, pairing RCP 2.6 with SSP1, RCP 4.5 with SSP3,
and RCP 8.5 with SSP5.
Limitations. There are several potential limitations to our study. First, the lethality
of deadly climatic conditions can be mediated by various demographic (for
example, age structure), socio-economic (for example, air conditioning, early
warning systems) and urban planning (for example, vegetation, high albedo
surface) factors that were not considered in our study. Consideration of these
factors would improve the understanding of global human vulnerability to heat
exposure and may reduce the number of human deaths, but they are unlikely to
affect the occurrence of deadly climatic conditions, which is what we estimated.
Second, our survey of cases of heat-related mortality was restricted to the period
between 1980 and 2014, and any bias or temporal heterogeneity in the monitoring
of lethal heatwaves and epidemiological studies in this period may influence the
cases we studied and the resulting SVM model. Third, while general agreement
among models was found in the predictions of deadly climatic conditions in
tropical areas, greater variability among models was seen in such projections at
higher latitudes (grey areas in Fig. 3). Because deadly conditions are more rare at
higher latitudes (Fig. 4), a larger number of model ensembles might allow for more
definitive statements about the risk of deadly climates in such regions, as has been
suggested for similar cases of rare events31. Finally, it is possible that some lethal
heat events were not documented in peer-reviewed publications and, if the dates of
those undocumented events happened to be selected as part of the non-lethal
events in our analysis, this could affect the resulting SVM model. However, this
error is likely minimal because there is a low probability of randomly selecting such
rare and brief events from a 30-year period in the given cities.
Data availability. The data that support the findings of this study are available
from the corresponding author upon request.
NATURE CLIMATE CHANGE |www.nature.com/natureclimatechange
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
