VU Research Portal
Future climate risk from compound events
Zscheischler, Jakob; Westra, Seth; Van Den Hurk, Bart J.J.M.; Seneviratne, Sonia I.;
Ward, Philip J.; Pitman, Andy; Aghakouchak, Amir; Bresch, David N.; Leonard, Michael;
Wahl, Thomas; Zhang, Xuebin
published in
Nature Climate Change
2018
DOI (link to publisher)
10.1038/s41558-018-0156-3
document version
Publisher's PDF, also known as Version of record
document license
Article 25fa Dutch Copyright Act
Link to publication in VU Research Portal
citation for published version (APA)
Zscheischler, J., Westra, S., Van Den Hurk, B. J. J. M., Seneviratne, S. I., Ward, P. J., Pitman, A., Aghakouchak,
A., Bresch, D. N., Leonard, M., Wahl, T., & Zhang, X. (2018). Future climate risk from compound events. Nature
Climate Change, 8(6), 469-477. https://doi.org/10.1038/s41558-018-0156-3
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal ?
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
E-mail address:
vuresearchportal.ub@vu.nl
Download date: 10. Aug. 2022
PersPective
https://doi.org/10.1038/s41558-018-0156-3
1
Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland.
2
School of Civil, Environmental and Mining Engineering, University of
Adelaide, Adelaide, South Australia, Australia.
3
Royal Netherlands Meteorological Institute (KNMI), De Bilt, The Netherlands.
4
Institute for Environmental
Studies, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands.
5
Australian Research Council Center of Excellence for Climate Extremes and Climate
Change Research Center, University of New South Wales, Sydney, New South Wales, Australia.
6
Department of Civil and Environmental Engineering,
University of California, Irvine, CA, USA.
7
Institute for Environmental Decisions, ETH Zurich, Zurich, Switzerland.
8
Federal Office of Meteorology and
Climatology, MeteoSwiss, Zurich, Switzerland.
9
Department of Civil, Environmental and Construction Engineering and National Center for Integrated
Coastal Research, University of Central Florida, Orlando, FL, USA.
10
Climate Research Division, Environment and Climate Change Canada, Toronto, Ontario,
Canada. *e-mail: jakob.zscheischler@env.ethz.ch
I
n the summer of 2010, Russia was struck by an unprecedented
heatwave
1
. Below-normal precipitation in the first seven months
of the year induced a summer drought that contributed to the
exceptional magnitude of the heatwave
2
. The extremely dry and hot
conditions led to widespread wildfires
3
, which damaged crops and
caused human mortality
4
. The wildfires also induced large-scale air
pollution in cities such as Moscow
5
, adding to the death toll caused
by the heatwave
6
. The incidents in Russia in the summer of 2010
can be termed a compound event, involving the co-occurrence of
multiple dependent hazards: drought, heat, fire and air pollution.
In combination, these hazards caused devastating impacts in many
areas, at a scale well beyond that which any one of these hazards
would have caused in isolation.
One might think that the simultaneous occurrence of these haz
-
ards is bad luck or simply a low-probability occurrence. Indeed, the
extraordinary nature of the 2010 event in western Russia is clear and
could be viewed as too rare to be predictable. However, accounting
for dependencies between all relevant climate drivers and/or haz
-
ards increases the likelihood of such events considerably, and may
make events of the rarity of the Russian event foreseeable and to
some extent predictable, as illustrated in the following. Temperature
and precipitation are strongly negatively correlated in summer
over western Russia, increasing the likelihood of extremely hot and
dry summers by a factor of up to five compared with both vari
-
ables being independent
7
. Furthermore, in addition to rising global
temperatures
8
, low soil moisture in spring and summer strongly
increased the magnitude of the heatwave
2
, providing an opportu-
nity for increased predictability. Similarly, fire regimes are known
to interact closely with drought
9
, and high temperatures and low
humidity can be predictable precursors of intense fires
10
.
The interaction between multiple climate drivers and/or haz
-
ards can also play a major role in coastal extremes
11,12
. Hurricane
Sandy hit the metropolitan New York area in 2012, causing damages
in excess of US$50 billion and a total death toll of 233
13
. Sandy’s
significant impacts were due to its unusual path, which resulted
from multiple weather systems coinciding over the North American
continent and the north Atlantic. Atlantic hurricanes commonly
dissipate over the open ocean; however, a strong blocking high in
the mid- to high latitudes of the north Atlantic in combination with
a mid-latitude trough over Canada and the northeast United States
steered Sandy back towards the coast
14
, leading to substantial inland
rain and flooding. Coming almost directly from the east, the storm
caused the highest storm surge in at least 300 years
15
, and coinciding
with a high (spring) tide, the storm led to widespread flooding in
New York City and surrounding areas. The strong winds also pro
-
duced high waves along the sandy coasts of New Jersey, where they
could travel closer to shore without breaking because of the high
water levels from the storm tide, ultimately resulting in massive
coastal erosion
16
. The compounding effects from inland precipita-
tion (pluvial flooding), high wind speeds, storm surge and waves,
played an important role in exacerbating the impacts of the event.
In 2017 Hurricane Harvey provided another example of com
-
pound flooding. From a meteorological perspective, the simultane-
ous occurrence of a high-pressure system over the western United
States pushed the storm back into the Gulf of Mexico instead of
allowing it to follow the typical track further inland, where the
system would have dissipated much faster. Instead, Harvey circled
back and made landfall a second time in the greater Houston area.
The stationarity of the system for an extended period of time led
to extremely high accumulated precipitation over several days (ini
-
tial estimates suggest a return time between 100 and 2,000 years
17
).
At the same time, Harvey produced a storm surge along the coast
that was moderate in height, but affected an extremely long segment
of the coast, with elevated water levels over five days and multiple
tidal cycles, significantly reducing the inland freshwater drainage
capacity. From a climate perspective, unusually high sea surface
temperature additionally fuelled the tropical system
18
, and sea-level
rise has led to higher baseline ocean levels than a century ago. This
Future climate risk from compound events
Jakob Zscheischler
1
*, Seth Westra
2
, Bart J. J. M. van den Hurk
3,4
, Sonia I. Seneviratne
1
,
Philip J. Ward
4
, Andy Pitman
5
, Amir AghaKouchak
6
, David N. Bresch
7,8
, Michael Leonard
2
,
Thomas Wahl
9
and Xuebin Zhang
10
Floods, wildfires, heatwaves and droughts often result from a combination of interacting physical processes across multiple
spatial and temporal scales. The combination of processes (climate drivers and hazards) leading to a significant impact is
referred to as a ‘compound event’. Traditional risk assessment methods typically only consider one driver and/or hazard at
a time, potentially leading to underestimation of risk, as the processes that cause extreme events often interact and are spa-
tially and/or temporally dependent. Here we show how a better understanding of compound events may improve projections
of potential high-impact events, and can provide a bridge between climate scientists, engineers, social scientists, impact mod-
ellers and decision-makers, who need to work closely together to understand these complex events.
Corrected: Author Correction
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | VOL 8 | JUNE 2018 | 469–477 | www.nature.com/natureclimatechange
469
PersPective
NaTure ClimaTe CHaNge
highlights the range of spatial and temporal scales that can ulti-
mately lead to the extreme impacts: from long-term global warming
changing the background climate state, through to the occurrence
of Hurricane Harvey causing heavy wind and rain and finally to
the localized effects in terms of storm surges and flood events for
Houston and the surrounding area.
Extreme events with devastating impacts such as those described
above leave an imprint in public memory and are typically char
-
acterized by a complex chain of processes, often extending well
beyond the local event itself. The destruction of large amounts of
Russian crops led to a grain export ban until the end of 2010, affect
-
ing global wheat prices
19
and potentially contributing to instability
and uprising in Egypt
20
. Parallel to the Russian heatwave, a record-
breaking flood occurred in Pakistan, which affected more than 20
million people
21
. There is strong evidence that these features are
connected through atmospheric dynamics
22
.
Understanding compound events therefore requires an analy
-
sis of the complex causal chains that can lead to extreme impacts.
Multiple drivers and/or hazards have to be investigated because it
is their combination that renders an event exceptional and pushes
the impact to extreme levels. In many cases, however, the unusual
combinations of processes associated with the events makes them
difficult to foresee, in particular because they are so rare and may
not have observed historical analogues
23
. This issue is likely to be
exacerbated as a result of climate change and human activity affect
-
ing both the background climate state
24,25
and how the system func-
tions. Therefore, the historical record of compound events provides
incomplete information on how events may occur in the future.
Furthermore, as we consider more complex causative changes, the
likelihood that specific combinations and sequences of drivers and/
or hazards have occurred previously can rapidly approach zero
26
.
Apart from changing likelihoods of the contributing processes,
systematic climate change has the potential to change relationships
between drivers and hazards to create novel conditions that our
socioeconomic systems have not been designed to withstand
27
.
Multiple drivers, conditional dependencies, a complex chain of
processes and extreme return times; these are all characteristics of
extreme climate events that lead to devastating impacts. Common
practice based on highly idealized conceptual frameworks of mod
-
elling, scenario construction and statistical analysis each have dif-
ficulties in fully capturing these interrelationships. Given their
disproportionate impacts, however, improving our understanding
and modelling capabilities of such events is of crucial importance.
In this Perspective, we first introduce a new definition of com
-
pound events, which aims to establish a framework for compound
event research. We then argue that a paradigm shift is needed when
compound events are incorporated in climate impact analysis. We
further discuss how compound event research can improve risk
assessments of extreme events. We end with five recommendations
targeted to the climate science and impact modelling communities
to advance compound event research.
Defining compound weather and climate events
A particular challenge with understanding compound events is
that dependencies between drivers and/or hazards can make the
estimation of event probabilities more difficult than if all drivers
and hazards were independent
28,29
. Poor representation of these
dependencies can lead to an underestimation of the risk of cata
-
strophic impacts, given that risk is often much greater than a naive
independent combination of the individual components would sug
-
gest
7,12,30–32
. For instance, extreme storm surge and rainfall events
are often positively correlated along the coastlines of the United
States
12
, The Netherlands
32
and Australia
33
, increasing the probabil-
ity of coastal floods. Precipitation and wind extremes are also likely
to co-occur, augmenting the risk of infrastructure damage during
severe storms
34
. Likewise, because of land–atmosphere feedbacks
35
,
warm season temperature and precipitation are generally negatively
correlated, rendering an extremely hot and dry summer far more
likely than an extremely hot and wet summer
7
. Table 1 presents a
non-exhaustive list of climate and/or weather driver combinations
that can lead to hazards, and hazards combinations that are known
to cause large impacts.
Given that most previous climate-related studies of hazards focus
on single drivers, and given the evidence that the events that are
particularly worrisome are typically multivariate in nature as illus
-
trated by the examples in this manuscript, we encourage a deeper
focus on multivariate drivers and hazards of large climate-related
impacts. We therefore introduce the following definition.
Compound weather/climate events. We here define compound
compound weather/climate events as the combination of mul
-
tiple drivers and/or hazards that contributes to societal or envi-
ronmental risk (Box 1). Drivers include processes, variables and
phenomena in the climate and weather domain that may span
over multiple spatial and temporal scales. Hazards are usually
the immediate physical precursors to negative impacts (such as
floods, heatwaves, wildfire), but can occasionally have positive
outcomes (for example, greening in the Alps during the 2003 heat
-
wave in Europe
36
). Risk is defined as probability of hazards (events
or trends) × consequences (see Box 1 for definitions used in this
Perspective). In the simplest case, × represents multiplication
37
,
but more generally, it represents a convolution of the respective
distributions of probability and consequences. In that sense, inte
-
grating over a limited range that only includes highly frequent
low-impact events can result in risks that are comparable to the
risk associated with very rare high-impact events
38
. Furthermore,
in the tail of the event distribution, which is often associated with
the most catastrophic impacts, probabilities may not be quantifi
-
able and storyline approaches
39
are needed.
Table 1 | Non-exhaustive list of documented climate-
related hazards for which drivers are dependent as well as
combinations of dependent hazards with potentially large
impacts
Hazard(s) Climatic drivers Reference(s)
Drought Precipitation, evapotranspiration,
historic evolution of soil
moisture, temperature
35,77,78
Physiological heat
stress
Temperature, atmospheric
humidity, strongly dependent on
diurnal cycle
56
Fire risk Temperature, precipitation,
relative humidity, wind, lightning
55,79
Storm risk Wind speed, humidity, large
scale atmospheric circulation
94,95
Coastal flood River flow, precipitation, coastal
water level, surge, wind speed
11,12,30
Flood risk at river
confluences
Precipitation, water levels of
contributing rivers, large-scale
atmospheric circulation
31
Concurrent drought
and heat
Temperature, precipitation,
evapotranspiration, atmospheric
humidity
7,35
Concurrent wind
and precipitation
extremes
Wind speed, precipitation,
orography, large-scale
atmospheric circulation
34
Concurrent heat and
air pollution
Temperature, sulfur dioxide, NO
x
,
particulate matter (PM
10
)
6,76
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | VOL 8 | JUNE 2018 | 469–477 | www.nature.com/natureclimatechange
470
PersPective
NaTure ClimaTe CHaNge
Compound events can be embedded in the general risk
framework linking hazards, vulnerability and exposure (Fig. 1).
Changes in exposure and vulnerability, often related to human
development
40
, can strongly affect environmental risk. While we
acknowledge this contribution, we focus on climate-related haz
-
ards here.
Our definition of compound weather/climate events
generalizes the earlier definitions in the IPCC Special Report on
Climate Extremes (SREX)
41
, which introduced compound events
as a general concept to the climate sciences, and that of Leonard
and colleagues
28
, who suggested a definition of compound
events that refers only to extreme-impact events with dependent
drivers. Our definition aims to establish a clear framework for
ground-breaking research in the climate and impact science
communities.
A paradigm shift in climate impact analyses
The emphasis on combinations of drivers and/or hazards that lead
to societal or environmental risks highlights the importance of
understanding the nature of the risks before identifying the relevant
drivers and hazards. This suggests the use of bottom-up approaches
to help identify which drivers and/or hazards lead to large impacts.
Bottom-up approaches
42
usually start with a system or impact (such
as a disaster), and then identify all of the underlying variables, pro
-
cesses or phenomena that play a role in shaping the outcome. This
includes identifying which parts in the driver distribution lead to
large impacts, and is therefore highly appropriate for studying com
-
pound weather and climate events. For instance, understanding the
possible meteorological drivers of a power outage in a city might
require identification of the climate-sensitive elements of the energy
system, such as the combination of renewable resources (solar, wind
and hydroelectricity), together with the physical assets such as poles
and power lines that could be affected by heavy winds, lightning and
flooding. This in turn forms the basis for understanding the weather
and/or climate drivers and hazards that could influence that system.
This system-centric approach contrasts with top-down or sce
-
nario-led approaches
43–45
, whereby climate change scenarios are
generated using climate models and then incorporated into an
impact model. So far, potential impacts of climate extremes are pre
-
dominantly assessed via top-down approaches. In this way, flood
risk
46
as well as impacts on crop yields
47
and human health
48
have
been estimated based on individual drivers and/or hazards or an
(independent) combination of multiple drivers such as run-off,
temperature and precipitation. Top-down approaches require bias
correction and downscaling
43
, resulting in large increases in uncer-
tainty
45
, while their effects on the multivariate distribution of cli-
mate drivers and/or hazards are unknown. Furthermore, future
climate scenarios do not cover the full probability space of all pos
-
sible future conditions
49,50
; hence, such risk assessments are unlikely
to represent the ‘real’ risk. It is therefore unclear how well top-down
approaches capture impacts associated with multiple interacting
drivers and/or hazards.
The benefits of the bottom-up approach for compound events
is that it focuses attention on the combinations of drivers and/or
hazards that can cause a system to fail, and then works backwards
to identify lines of evidence that could provide insights into the
likelihood of such combinations. Bottom-up or ‘scenario-neutral’
approaches are therefore increasingly being used to understand
climate impacts and system resilience
42,44,51–53
. The shift from top-
down to bottom-up approaches is in essence comparable to the shift
from impact analysis to vulnerability analysis in socio-economic
studies of climate change risks
54
. Whereas impact analysis traces the
impacts of a single hazard (drought) to multiple outcomes (fam
-
ine, economic loss), vulnerability analysis characterizes the multiple
causes (low precipitation, poverty, lack of planning) of single out
-
comes (famine).
An additional advantage of bottom-up approaches is the poten
-
tial for studying the impacts of hazards and the climate drivers of
those hazards in separation. This is an effective way to study mul
-
tiple hazards and their driving mechanisms simultaneously. At the
same time, it avoids a biased view by focusing on the full distribu
-
tion of climate drivers instead of only the fraction that is relevant for
a particular hazard. This may turn out to be very effective, as differ
-
ent hazards based on the same set of drivers may vary along differ-
ent gradients in the climate driver space (Fig. 2). For instance, the
Chandler Burning Index (an index for fire risk
55
) and wet-bulb tem-
perature (an index for heat stress
56
) can both be expressed in terms
of temperature and relative humidity
55,57
. While dry and hot condi-
tions increase fire risk (Fig. 2a), dry and humid conditions increase
human mortality risk (Fig. 2b). As illustrated by this example, the
distribution of the climate drivers of a given hazard is in principle
independent of the direction in which the hazard intensity varies,
Box 1 | Definitions used in this Perspective
Risk. The “effect of uncertainty on objectives”
97
. According to the
IPCC
96
, risk is the potential for consequences when something
of value is at stake and the outcome is uncertain, recognizing
the diversity of values. Risks arise from the interaction between
hazard, vulnerability and exposure and can be described by
the formula:
=×Risk (probability of events or trends)consequences
where consequences are a function of the intensity of hazard
(event or trend), exposure and vulnerability. Here, we use the
term risk to refer to environmental and societal impacts from
weather and/or climate events.
Exposure. The presence of people, livelihoods, species or ecosys
-
tems, environmental functions, services, and resources, infra-
structure, or economic, social, or cultural assets in places and
settings that could be adversely affected
96,98
.
Vulnerability. The propensity or predisposition to be adversely
affected
96,98
. Vulnerability encompasses a variety of concepts and
elements including sensitivity or susceptibility to harm and lack
of capacity to cope and adapt.
Hazard. The potential occurrence of a natural or human-
induced physical event or trend or physical impact that may
cause loss of life, injury, or other health impacts, as well as dam
-
age and loss to property, infrastructure, livelihoods, service
provision, ecosystems and environmental resources
96
. Here, the
term hazard usually refers to climate-related physical events or
their physical impacts.
Compound weather/climate events. The combination of multiple
drivers and/or hazards that contributes to societal or environ
-
mental risk.
Weather and climate events. Events at spatial and temporal scales
varying from local weather to large-scale climate modes.
Drivers. These include climate and weather processes, variables
and phenomena. We refer to the term drivers throughout as
direct (climate and weather related) causes of climate-related
hazards.
Impacts. The effects of physical events on natural and human
systems
96
.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | VOL 8 | JUNE 2018 | 469–477 | www.nature.com/natureclimatechange
471
PersPective
NaTure ClimaTe CHaNge
supporting a separate analysis of hazards and drivers. Moreover,
the multivariate distribution of climate drivers may change over
time, for instance if one driver is affected by trends
25,58
(temperature
increase, sea-level rise
59
, trends in storm activity) or changes in the
distribution (changes in temperature variance
60
, shifts in precipita-
tion distribution
61
). Finally, the dependence between climate drivers
may change over time, which also affects the multivariate distribu
-
tion of drivers. For instance, the increase in concurrent extreme
storm surge and precipitation events for United States coasts has
been attributed to changes in the dependence between surge and
precipitation rather than to trends in either of these variables
12
.
Similarly, the dependence between summer temperature and pre
-
cipitation is expected to change under strong GHG forcing
7
. Note
that even if we can model the whole distribution of drivers or haz
-
ards based on observational data, estimating dependence in the
tails
62
, for instance between different hazards, may still be challeng-
ing if the sample size is not very large.
Identifying which multivariate constellations of climate variables
are associated with hazards allows climate model output to be inter
-
rogated for exactly these constellations. Assessing the likelihood
of such constellations in future projections will help to investigate
risk. Besides providing a tool for the assessment of hazard likeli
-
hoods, this approach will bring focus on those physical processes
that need to be better understood to represent hazards in dynamical
models, providing guidance on which variables and dependencies
between variables need to be simulated skilfully or bias-adjusted
63
to correctly quantify hazards. The impact research community and
the climate science community can both contribute to this effort
by working closely together, revising and integrating currently used
approaches and moving towards a multivariate perspective in all
compartments of model construction, bias adjustment and analysis.
Climate change processes and associated effects
The bottom-up approach helps to define the required scope of
the modelling of physical processes that give rise to a particular
compound event. How can we represent these processes adequately?
While spatial and temporal scales of compound events can vary
significantly, the impacts are commonly felt at the local scale over
relatively short timescales. However, local-scale events are often
embedded within larger-scale systems, which in turn are affected by
planetary-scale features such as shifts in the radiation balance and
associated changes in mean temperature, mean sea level, the loca
-
tion of the jet stream and others. Modelling approaches that rep-
resent these ranges of space and time scales are therefore needed.
The non-stationarity of most compound events — both because of
anthropogenic climate change and because of other more local-scale
changes in the land surface due to urbanization and other forms
of development — has significant implications for how compound
events should be modelled.
These implications can be understood through the example of
estimating the probability of flooding for a particular catchment.
In the past, if historical records of sufficient length were available, it
was common to use these records to estimate the exceedance prob
-
ability of a future event through a method called flood frequency
analysis. However, such methods are only appropriate when the cli
-
matic drivers of floods are stationary over time. Flood frequency
analysis is not appropriate as a basis for designing future infrastruc
-
ture under considerations of significant climate change, since the
historical statistics may no longer reflect flood hazard in the future.
This means that to estimate the probability of flooding, we need
to understand the nature of changes affecting extreme events much
more explicitly, leading to a widening of the system boundaries. For
many event-based hydrological models, antecedent moisture con
-
ditions are typically treated as calibration parameters, for example
through loss parameters of the hydrological model. However, under
future climate, extreme rainfall may increase at a faster rate than
average rainfall, and evapotranspiration may change as well, so that
the relationship between flood-producing rainfall and the catch
-
ment’s antecedent conditions is no longer stable and may need to be
modelled explicitly, for example by using a continuous hydrological
Vulnerability
RiskHazards
Exposure
Socioeconomic
processes
Socioeconomic
pathways
Natural
variability
Anthropogenic
climate change
Adaptation and
migration actions
Governance
Climate
Non-climatic
drivers
Non-climatic
drivers
Impacts
Emissions
and land-use change
?
?
Climatic drivers
Fig. 1 | Extended risk framework. Multiple climatic drivers cause one or multiple hazards leading to societal and environmental risk. The climate drivers
(which may vary from local-scale weather to large-scale climate modes, represented by yellow circles) and/or hazards may be mutually dependent. Non-
climatic drivers related to vulnerability and exposure may also contribute to risk. Background risk figure adapted from ref.
96
, IPCC.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE | VOL 8 | JUNE 2018 | 469–477 | www.nature.com/natureclimatechange
472
