Human influences on spatially compounding flooding and heatwave events in China and future increasing risks

Abstract

Attribution of high-impact weather events to anthropogenic climate change is important for disentangling longterm trends from natural variability and estimating potential future impacts. Up to this point, most attribution studies have focused on univariate drivers, despite the fact that many impacts are related to multiple compounding weather and climate drivers. For instance, co-occurring climate extremes in neighbouring regions can lead to very large combined impacts. Yet, attribution of spatially compounding events with different hazards poses a great challenge. Here, we present a comprehensive framework for compound event attribution to disentangle the effects of natural variability and anthropogenic climate change on the event. Taking the 2020 spatially compounding heavy precipitation and heatwave event in China as a showcase, we find that the respective dynamic and thermodynamic contributions to the intensity of this event are 51% (35–67%) and 39% (18–59%), and anthropogenic climate change has increased the occurrence probability of similar events at least 10-fold. We estimate that compared to the current climate, such events will become 10 times and 14 times more likely until the middle and end of the 21st century, respectively, under a high-emissions scenario. This increase in likelihood can be substantially reduced (to seven times more likely) under a low-emissions scenario. Our study demonstrates the effect of anthropogenic climate change on high-impact compound extreme events and highlights the urgent need to reduce greenhouse gas emissions.

Full text

Weather and Climate Extremes 42 (2023) 100616
Available online 11 October 2023
2212-0947/© 2023 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Human inuences on spatially compounding ooding and heatwave events
in China and future increasing risks
Cheng Qian
a
,
b
,
*
, Yangbo Ye
a
,
b
, Emanuele Bevacqua
c
, Jakob Zscheischler
c
,
d
a
Key Laboratory of Regional Climate-Environment for Temperate East Asia, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
b
University of Chinese Academy of Sciences, Beijing, China
c
Department of Computational Hydrosystems, Helmholtz Centre for Environmental Research-UFZ, Leipzig, Germany
d
Technische Universit¨
at Dresden, Dresden, Germany
ARTICLE INFO
Keywords:
Spatially compounding event
Extreme event attribution
Human contribution
Storyline approach
Risk-based approach
Carbon neutrality
ABSTRACT
Attribution of high-impact weather events to anthropogenic climate change is important for disentangling long-
term trends from natural variability and estimating potential future impacts. Up to this point, most attribution
studies have focused on univariate drivers, despite the fact that many impacts are related to multiple com-
pounding weather and climate drivers. For instance, co-occurring climate extremes in neighbouring regions can
lead to very large combined impacts. Yet, attribution of spatially compounding events with different hazards
poses a great challenge. Here, we present a comprehensive framework for compound event attribution to
disentangle the effects of natural variability and anthropogenic climate change on the event. Taking the 2020
spatially compounding heavy precipitation and heatwave event in China as a showcase, we nd that the
respective dynamic and thermodynamic contributions to the intensity of this event are 51% (35–67%) and 39%
(18–59%), and anthropogenic climate change has increased the occurrence probability of similar events at least
10-fold. We estimate that compared to the current climate, such events will become 10 times and 14 times more
likely until the middle and end of the 21st century, respectively, under a high-emissions scenario. This increase in
likelihood can be substantially reduced (to seven times more likely) under a low-emissions scenario. Our study
demonstrates the effect of anthropogenic climate change on high-impact compound extreme events and high-
lights the urgent need to reduce greenhouse gas emissions.
1. Introduction
Compound extreme events cause impacts that are often much more
severe than those of individual extreme events (Zscheischler et al., 2018,
2020). Scientic research to address whether and to what extent
anthropogenic climate change has altered the characteristics of a
particular extreme event—“event attribution”—has thus far focused
largely on univariate extremes (Herring et al., 2022). Event attribution is
a key aspect of understanding climate-change risks (Stott et al., 2016). In
particular, it is vital to inform society how climate change is worsening
extreme events and further guide the community to better prepare for
future increases in climate-related risks and to better rebuild cities and
infrastructure after disasters to be more resilient in an impending
climate-changed world (Stott, 2016; Qian et al., 2022a). Concepts for
the attribution of compound extreme events have only emerged recently
(Chiang et al., 2021; Zscheischler and Lehner, 2022; Bevacqua et al.,
2023).
Among the various types of compound extreme events, spatially
compounding events occur when multiple connected locations are
concurrently affected by the same or different hazards, thus inducing an
aggregated impact (Zscheischler et al., 2020). Attribution studies of
spatially compounding events are still rare, and limited to those
considering the same hazard (Vogel et al., 2019; Verschuur et al., 2021;
Zscheischler and Lehner, 2022). Attributing different types of spatially
co-occurring hazards is methodologically challenging because of the
different spatiotemporal scales often involved. It is a challenge, for
example, to design a compound index that is impact-relevant and can
easily be applied to model simulations.
The commonly used risk-based (or probability-based) framework for
event attribution treats an observed event as one of a class of similar
* Corresponding author. Key Laboratory of Regional Climate-Environment for Temperate East Asia, Institute of Atmospheric Physics, Chinese Academy of Sciences,
Beijing, China.
E-mail address: qianch@tea.ac.cn (C. Qian).
Contents lists available at ScienceDirect
Weather and Climate Extremes
journal homepage: www.elsevier.com/locate/wace
https://doi.org/10.1016/j.wace.2023.100616
Received 20 July 2023; Received in revised form 5 October 2023; Accepted 10 October 2023

Weather and Climate Extremes 42 (2023) 100616
2
events and helps provide guidance for future adaptation and disaster
recovery (Stott et al., 2004; Otto et al., 2016; Philip et al., 2020).
However, this approach does not provide much information on the event
itself. As an alternative, the storyline framework for event attribution
examines the contribution of various factors to the specic event itself
and is valuable for understanding the evolution of the event in response
to various drivers (Trenberth et al., 2015; Shepherd, 2016). The story-
line approach generally neglects possible changes in the conguration of
the dynamics behind the event, but it is possible for the dynamic
conguration to also be altered by anthropogenic forcings (Otto et al.,
2016; Shepherd, 2016). The two frameworks provide complementary
insights into a high-impact event (Shepherd, 2016; Qian et al., 2022a).
However, they are rarely used in combination (Ye and Qian, 2021; Qian
et al., 2022b).
In this paper, we develop a storyline–probability combined frame-
work for applying extreme event attribution to spatially compounding
events that involve different hazards, with the goal of increasing con-
dence in the attribution statement. We combine the storyline and risk-
based approaches to quantify the contribution of large-scale atmo-
spheric circulation resembling the observations of the event (dynamic
effect) and anthropogenic forcings conditional on the atmospheric cir-
culation (thermodynamic effect) to the intensity of the event and the
overall anthropogenic contribution to the intensity and the occurrence
probability of similar events.
We illustrate the proposed approach for a spatially compounding
event that occurred in China in 2020. A record-breaking persistent
heavy rainfall event struck the middle and lower reaches of the Yangtze
River in China during the Meiyu period (June–July) in 2020. At the same
time, South China suffered from a concurrent record-breaking heatwave
event (Ye and Qian, 2021) (Fig. 1). These two events contributed to
widespread severe ooding and drought, respectively, in the two re-
gions. Both areas are important economic centres with high population
densities. Attribution studies of the precipitation event alone have been
carried out based on climate model simulations (Zhou et al., 2021; Lu
et al., 2022; Tang et al., 2022), and these studies consistently concluded
that anthropogenic forcing has reduced the occurrence probability of the
extreme precipitation event in 2020 almost by half (Zhou et al., 2021; Lu
et al., 2022; Tang et al., 2022). Conditional attribution of the contri-
butions of climate change and atmospheric circulation to the precipi-
tation and temperature events has been estimated separately, based on
observational and reanalysis data (Ye and Qian, 2021). In that study, it
was found that atmospheric circulation explained about 71% and 44% of
the extreme precipitation event and the concurrent heatwave event,
respectively; and that compared with past climate under similar atmo-
spheric circulation conditions, the occurrence probability of an event
reaching or exceeding the 2020 Meiyu amount increased by about ve
times under the present climate, and heatwave events reaching or
exceeding a threshold of one standard deviation increased from 0.6%
under past climate conditions to 69% under the present climate. Both
events were driven by the same modulator, i.e. an intensied western
Pacic subtropical high (Ye and Qian, 2021), and strongly affected the
agricultural sector in both regions. According to statistics from the
Ministry of Emergency Management, the persistent heavy rainfall in the
middle and lower reaches of the Yangtze River affected approximately
3.5798 million ha of crops, including 893.9 thousand ha that experi-
enced crop failure, resulting in direct economic losses of 132.2 billion
Chinese Yuan for July alone (https://www.mem.gov.cn/xw/yjglbgzdt/
202101/t20210102_376288.shtml). The inundation of crops also
resulted in a phenological delay in crop growth in 2020 (Qin et al.,
2022) and a reduction in vegetable yields. The heatwave in South China
also affected the growth of vegetables there. As a result, the national
Consumer Price Index for Vegetables in July 2020 in China saw
year-on-year increases of 11.4% (https://sannong.cctv.com/2020/08/0
5/ARTIv08N1New9JFarvghwIqV200805.shtml). We therefore consider
this event to be a spatially compounding extreme event (Zscheischler
et al., 2020) and focus on a compound index that is, effectively, the
average of the standardized temperature and precipitation anomalies in
both regions. We conducted an event attribution analysis employing
both the storyline approach (Shepherd, 2016), which is conditional on
the large-scale atmospheric circulation, and the unconditional
risk-based approach (Stott et al., 2004).
We rst estimated the contribution of the atmospheric circulation
resembling the observations in 2020 (dynamic effect) and that of
anthropogenic forcings conditional on the atmospheric circulation
(thermodynamic effect) to the intensity of the event from a storyline
perspective to demonstrate the role of anthropogenic forcings. In the
storyline approach, we propose a novel constructed ow analogues
(CFA) method to evaluate the contribution of the dynamic effect. CFA
can construct analogues of atmospheric circulation that are almost the
same as the observed atmospheric circulation and, thus, better estimate
its contribution compared to the ow analogues method (Yiou et al.,
2007; J´
ez´
equel et al., 2018), which has limitations when the intensity of
the extreme event is too strong to nd suitable analogues. When esti-
mating the thermodynamic effect, we used the simulation from the latest
Met Ofce attribution system, the HadGEM3-GA6-N216 model (Cia-
varella et al., 2018), and conditioned the atmospheric circulation using
the CFA method to resemble that of 2020. This storyline approach can
help us better understand the compound event itself. We also took into
account the possible inuence of anthropogenic forcings to changes in
atmospheric circulation by integrating a risk-based analysis of the in-
tensity of similar events with the same return period based on the
HadGEM3-GA6-N216 model and compared the result with that from
phase 6 of the Coupled Model Intercomparison Project (CMIP6)
multi-model ensembles (Eyring et al., 2016). By using CMIP6, we further
estimated the role of historical emissions of greenhouse gases (GHGs)
and anthropogenic aerosols (AAs).
We then carried out a risk-based attribution in terms of the proba-
bility of occurrence risk of similar compound events. This risk-based
approach, which we also applied to future projections, takes into ac-
count the inuences of anthropogenic forcings on the atmospheric cir-
culation and involves the probability of similar compound events; thus,
it is valuable for future adaptation.
2. Data and methods
2.1. Observational data
We used the observed daily maximum temperature (T
max
) and 24-h
precipitation data (R) for the period 1961–2020 from the CN05.1
dataset (Wu and Gao, 2013) (resolution of 0.25◦×0.25◦), which was
obtained based on interpolation from over 2400 observing stations in
China. For atmospheric circulation data, we used the daily mean sea
level pressure (SLP), surface pressure (SP), geopotential height at 500
hPa (Z500), wind, and specic humidity (eight levels, from 1000 to 300
hPa) from the NCEP–NCAR Reanalysis I dataset (Kalnay et al., 1996)
with a resolution of 2.5◦×2.5◦for the period 1961–2020.
2.2. Model descriptions and evaluation
Model simulations with daily resolutions from CMIP6 were used (see
Supplementary Table S1 for details on model names, scenarios, and
number of ensemble members). Data include historical simulation ex-
periments with combined natural and anthropogenic forcing (Eyring
et al., 2016), Detection and Attribution Model Intercomparison Project
(DAMIP) experiments with individual forcing only (Gillett et al., 2016),
and the Shared Socioeconomic Pathway 1–1.9 (SSP1-1.9), 2–4.5
(SSP2-4.5), and 5–8.5 (SSP5-8.5) scenario experiments for future pro-
jections (O’Neill et al., 2016). We combined historical simulations with
corresponding SSP2-4.5 scenarios to extend the data to 2020 for attri-
bution analysis.
Additionally, we used model simulations with and without anthro-
pogenic forcings (historical and historicalNat, respectively) from the
C. Qian et al.

Weather and Climate Extremes 42 (2023) 100616
3
latest Met Ofce attribution system—the HadGEM3-GA6-N216 model
(Ciavarella et al., 2018). This model has 525 members for 2020 for each
experiment (historicalExt and historicalNatExt, respectively; hereafter,
Hist2020 and HistNat2020, respectively), forced by the observed sea
surface temperature/sea-ice concentration (SST/SIC) and external
forcings in 2020. Such a large number of ensemble members is relevant
given that a large sample size is required for robust attribution of
compound events (Bevacqua et al., 2023). Fifteen members of the
historical simulations for 1961–2013 were used to evaluate the model
performance.
We interpolated the observed and model-simulated precipitation and
temperature data to a common 1◦×1◦grid. Then, we calculated the
area-weighted average of the precipitation anomaly percentages (here-
after, the R% anomaly) in the middle and lower reaches of the Yangtze
River (120.0◦E|35.0◦N, 123.0◦E|30.0◦N, 108.0◦E|25.0◦N, 105.0◦E|
30.0◦N) and that of the T
max
anomaly in South China (120.0◦E|29.0◦N,
Fig. 1. Characteristics of the spatially compound event of 2020. (a) Spatial distribution of the observed precipitation anomaly in percentages (R% anomaly,
shading) and the column-integrated moisture ux anomaly from 1000 to 300 hPa (vectors; kg (s m)
−1) during June–July 2020. (c) Anomaly in the number of hot days
(days with a daily maximum temperature T
max
>35 ◦C) during June–July 2020. Time series of the (b) average precipitation percentage anomalies, (d) temperature
anomalies, and (f) compound index over target areas (shown as the black box in (a) and (c)) based on observations (1961–2020) and the HadGEM3-GA6-N216 model
ensemble means (1961–2013) under historical (red) and historicalNat (blue) experiments (shading indicates a range of 15 member simulations of the model). The
blue dashed line represents the linear trend. The red diamonds in (b), (d), (e) and (f) indicate the 2020 event (in (e) and (f), the second strongest compound event that
occurred in 2016 is also shown, in blue). (e) Bivariate distribution of the normalized R% anomaly and normalized T
max
anomaly based on observational data (red
dots) during 1961–2020, with isolines indicating equal levels of the compound index (Equation (1)). (For interpretation of the references to colour in this gure
legend, the reader is referred to the Web version of this article.)
C. Qian et al.

Weather and Climate Extremes 42 (2023) 100616
4
123.0◦E|22.7◦N, 109.0◦E|18.0◦N, 106.0◦E|24.3◦N) over June–July
(Fig. 1a–d; the two regions were delineated by Ye and Qian (2021), but
we exclude grid points with land portions that are less than 50%).
Following Zhang et al. (2020), we used the R% anomaly rather than the
R anomaly in order to reduce the effect of the model biases on the
climatological values and anomalies. In detail, we rst calculated the
area-weighted average of the precipitation (temperature) over
June–July and then calculated its anomaly percentage (anomaly). All
anomalies were calculated relative to the 1961–1990 climatology for all
datasets separately. For each model, the climatology is estimated via
averaging the ensemble mean of the historical simulations.
Model evaluation was divided into three parts, an approach similar
to that described in Zscheischler and Lehner (2022). Firstly, we assessed
whether the observed and model-simulated distributions for the T
max
anomaly and R% anomaly were signicantly different based on a Kol-
mogorov–Smirnoff (K–S) test. Secondly, we checked whether the
observed and model-simulated empirical copula distributions between
these two variables were signicantly different based on a Cram´
er-von
Mises (C-VM) test (Genest et al., 2009). Lastly, we assessed the models’
skill in simulating the correlation between these two variables. All
evaluations were conducted based on the period of 1961–2013 for the
HadGEM3-GA6-N216 model and 1961–2014 for the CMIP6 models. The
evaluation results are shown in Supplementary Table S2, and only
models passing the evaluation (i.e., the HadGEM3-GA6-N216 model and
four CMIP6 models) were included in the subsequent attribution and
projection analyses.
2.3. Index for the spatially compounding event
In order to quantify the magnitude of the spatially compounding
event, we dened a compound index:
Compound index =1
2(Tanom
σ
Tanom
+R%anom
σ
R%anom)(1)
Here, Tanom and
σ
Tanom represent the T
max
anomaly and climatological
standard deviation, respectively, for the area-weighted-average time
series in South China (Fig. 1d). R%anom and
σ
R%anom are the same, but for
the R% anomaly in the middle and lower reaches of the Yangtze River
(Fig. 1b). Note that each term in Eq. (1) is a standardized anomaly and
dimensionless quantity to allow comparability across variables (Daber-
nig et al., 2017). Hence, this index assumes that individual extremes in
temperature and precipitation and moderate extremes in both variables
at the same time are equally relevant to a potential impact. Other as-
sumptions on the functional relationship of the compound index could
be made, but without impact data to calibrate this relationship, they all
seem equally valid (Bevacqua et al., 2021). It should be noted that the
results obtained from the subsequent reconstructions from ow ana-
logues for 2020 based on reanalysis data were also normalized by the
observed climatological means and climatological standard deviations
so that they could be compared with the observed intensity; however,
the corresponding indices from model data were normalized by their
own climatological means and climatological standard deviations. The
standard deviation of a model was calculated as the multi-ensemble
mean of the standard deviation of each ensemble member of that model.
We analysed the linear trend in the observed index and its statistical
signicance using the nonparametric Wang and Swail (2001) iterative
method, considering repeated values in the signicance testing (Qian
et al., 2019). We regarded the linear trend as statistically signicant
using an alpha level of 5%.
2.4. Storyline analyses: dynamical and thermodynamical contribution to
the intensity of the 2020 compound event
Here, we express the intensity of the extreme event as:
M(E) = M(D) + M(ND)(2)
where M is the magnitude of the event itself (E), D is the dynamical
situation, and ND is the nondynamical situation (the complement of D).
With the storyline approach, we examine the role of the various factors
contributing to the event itself as it unfolded in a conditional manner
(Shepherd, 2016), but with the sole focus here on large-scale atmo-
spheric circulation and anthropogenic forcings. The rst term in Eq. (2)
is estimated via conditioning on atmospheric circulation conditions that
resemble the D that occurred in 2020. The second term in Eq. (2) in-
cludes (i) the thermal effect of anthropogenic forcings conditioning the
atmospheric circulation to resemble that of 2020, (ii) the possible in-
uence of anthropogenic forcings on the changes in atmospheric cir-
culation, and (iii) other effects. Here, we only focus on effect (i) and
leave effect (ii) in the risk-based attribution of the intensity of similar
events with the same return period. We did not consider effect (iii) in
this study.
We rst estimated the contribution of atmospheric circulation to the
intensity of the 2020 event as the ratio of the difference between the
reconstructed intensity under analogue and random atmospheric cir-
culation patterns to the observed intensity in June–July. In order to do
so, we developed a new method. Deser et al. (2016) proposed the con-
structed circulation analogue (CCA) approach to estimate the dynamical
contribution to winter surface air temperature trends over North
America during 1963–2012. We introduced this CCA method into the
event attribution to estimate the contribution of atmospheric circulation
and rened the CCA method to make it more appropriate for our pur-
poses by incorporating steps from the ow analogue method. We coined
this new method as the CFA approach. It involves picking out daily
analogues, combining them to obtain the reconstructed value of the
variable, and calculating the contribution of the atmospheric circulation
to the intensity of a target event. The main steps of this CFA approach
are shown in Supplementary Fig. S1. Specically, we selected the closest
Na analogues of the dynamic situation in 2020 according to the spatial
pattern of atmospheric circulation for each day in June–July 2020 from
a ±30-d window centred on the target day over the period 1961–2019.
We then randomly subsampled Ns analogues from the Na analogues to
compute their optimal linear combination that best t the target atmo-
spheric circulation eld for each day. The coefcients of the combina-
tion were calculated based on singular value decomposition. Then, the
corresponding T
max
or R% anomalies of the Ns analogues were also
combined linearly based on the same coefcients to obtain the recon-
structed anomalies conditional to the observed atmospheric circulation
for each day in June–July. This CFA approach can construct analogues of
atmospheric circulation that are almost the same as the observed spatial
patterns and thus better estimate the contribution of the atmospheric
circulation to extreme events than the original ow analogue, without
the cost of decreasing sample size. Note that the nonlinear trend of all
variables (estimated based on quadratic polynomial tting (Deser et al.,
2016; Ye and Qian, 2021) in this study) was removed prior to selecting
the analogues to minimize the impact of climate change (Deser et al.,
2016; J´
ez´
equel et al., 2018; Ye and Qian, 2021).
In this study, we determined Na =100 and Ns =25, according to the
resultant relatively small root-mean-square error (RMSE) (Supplemen-
tary Fig. S2). After sensitivity testing (Supplementary Figs. S3 and S4),
we determined that the suitable target area to calculate the ow
analogue was the small-sized area (102◦–123◦E, 16◦–35◦N) (shown in
Supplementary Figs. S1, S3a, and S4a), the suitable atmospheric circu-
lation variable was SLP (Supplementary Figs. S3b and S4b), and the
suitable method of measuring the similarity between atmospheric cir-
culation patterns was the spatial Pearson correlation coefcient (Sup-
plementary Figs. S3c and S4c). Note that the sensitivity testing in
Figs. S3 and S4 was carried out after removing the trends of all variables
prior to selecting the analogues, to minimize the impact of climate
change as stated in the previous paragraph. The underestimations reect
the positive contribution of climate change. We reconstructed the daily
T
max
anomaly and corresponding R% anomaly simultaneously over June
C. Qian et al.

Weather and Climate Extremes 42 (2023) 100616
5
and July based on the steps in Supplementary Fig. S1, and then calcu-
lated the bimonthly mean anomalies and the compound index. These
steps were repeated 10,000 times (Fig. 2a).
We estimated the contribution of atmospheric circulation (dynamic
effect) to the intensity of the 2020 event based on the above re-
constructions from the CFA approach (Fig. 2a). We combined some steps
from the ow analogues method (Yiou et al., 2007; J´
ez´
equel et al., 2018)
in calculating Control-1 and Control-M, which respectively represent
totally random atmospheric circulation patterns (reconstructed
completely randomly) and considering the persistence in the atmo-
spheric circulation (reconstructed randomly, but the analogues in the
adjacent M days were not repeatedly picked). Here, M represents the
number of days for which the atmospheric circulation persists and is
obtained from an autoregressive moving average model, as in Ye and
Qian (2021). The contribution of the dynamic effect to the intensity of
the event was then estimated by subtracting the median of Control-M
from the median of 10,000 reconstructed results obtained from the
CFA approach and then dividing the observed intensity of the event, as
in J´
ez´
equel et al. (2018).
We then estimated the contribution of the thermodynamic effect
(effect of anthropogenic forcings conditional on the atmospheric circu-
lation pattern) to the intensity of the 2020 event based on the CFA
approach (Fig. 2b). We conditioned the spatial pattern of the atmo-
spheric circulation to resemble that observed in 2020 in the Hist2020
and HistNat2020 simulations and then computed the difference between
the medians of the reconstructed compound index in these two simu-
lations divided by the observed compound index in 2020. The resultant
value was regarded as the contribution of the thermodynamic effect to
this compound event. One may argue that anthropogenic forcings may
affect the atmospheric circulation and thus counteract or enhance the
thermodynamics (Seneviratne et al., 2021); this effect is considered in
the subsequent unconditional risk-based analysis.
2.5. Risk-based analyses: intensity and probability
We conducted risk-based analyses based rst on intensity and then
on probability for different purposes. For the intensity analysis (Fig. 2c),
we estimated the contribution of the anthropogenic forcings to the in-
tensity of the events with the same return period as that observed in
2020 to complement the results of the storyline attribution. We used the
return level at the observed return period instead of the magnitude of
the event as a way to account for the model-simulated bias in the
magnitude. This way of accounting for model bias was also adopted by
the World Weather Attribution initiative (van Oldenborgh et al., 2021).
We tted and calculated return periods of the compound index from
observations and different model simulations based on the generalized
extreme value (GEV) distribution, after a goodness-of-t testing by
quantile–quantile plotting (Supplementary Fig. S5). For the
Fig. 2. The attribution results obtained from the combined approach. (a) Dynamic and (b) thermodynamic contributions to the intensity of the 2020 event based
on a conditional storyline approach. (a) The reconstructed detrended compound index is based on all days (left-hand boxplot), randomly subsampled days (sub-
sampled every six days to correct for serial dependence) (middle boxplot), and constructed ow analogues (right-hand boxplot). The red line represents the cor-
responding observed value in 2020. (b) Distributions of the reconstructed compound index based on HadGEM3-GA6-N216 model simulations with (Hist2020) and
without (HistNat2020) anthropogenic forcings for 2020. (c) Return periods tted using the GEV distribution based on the HadGEM3-GA6-N216 model. The dashed
area indicates the 5%–95% uncertainty range obtained from 1000 subsamples with data lengths the same as those of the observations. The horizontal and vertical
black dashed lines represent the observed intensity of the event and the corresponding return period, respectively. The horizontal red and blue dashed lines represent
the model-simulated intensity of the events with the same return period as in the observations based on the Hist2020 and HistNat2020 simulations, respectively. (d)
Dynamic and thermodynamic contributions of this event from the storyline approach conditioning the atmospheric circulation and contribution of the anthropogenic
(ANT) forcings to the intensity of 2020-like events based on a risk-based approach estimated from different models. Vertical bars denote the 5%–95% uncertainty
range obtained from 1000 instances of bootstrap resampling. The contribution of greenhouse gas (GHG) forcing and that of anthropogenic aerosol (AA) forcing are
also shown. (For interpretation of the references to colour in this gure legend, the reader is referred to the Web version of this article.)
C. Qian et al.

Weather and Climate Extremes 42 (2023) 100616
6
observations, we used a nonstationary-extremes tting model to esti-
mate the return period. In detail, we tted the GEV distribution of the
compound index during 1961–2020 that allowed the location (
μ
) and
scale (
σ
) parameters to scale with the 4-year smoothed global mean
surface temperature anomaly (T
′
) (section 4.3.2 in Philip et al., 2020):
μ
=
μ
0exp(
α
T
′
μ
o),
σ
=
σ
0exp (
α
T
′
μ
o)(3)
The t was performed using a maximum likelihood method varying
α
,
μ
0, and
σ
0. We then obtained the return period for the climate of 2020.
For the HadGEM3-GA6-N216 model, there are simulations for 2020 that
represent factual climate conditions (Hist2020) and counterfactual
climate conditions (HistNat2020). For the CMIP6 models, we used his-
torical simulations for the most recent 20 years (2001–2020) as samples
representing the near-present-day climate conditions, as in Christidis
and Stott (2015). Using 20 years of samples centred on the year of the
event is ideal; however, hist-nat simulations ended in 2020. Since the
hist-nat simulation is stationary in the long run, we used the entire
period (1850–2020) as the counterfactual climate conditions, as in
Christidis and Stott (2022). The intensity when the return period
reached the observed one in the Hist2020 (historical in CMIP6) simu-
lations was compared with that in the corresponding HistNat2020
(hist-nat in CMIP6) simulations (Fig. 2c), and their difference when
divided by the intensity of each model’s Hist2020 (historical in CMIP6)
simulation was regarded as the contribution of anthropogenic forcings
to the intensity of 2020-like events (Fig. 2d). Bootstrap resampling was
then used to estimate the 95% condence intervals of this contribution.
It should be noted that we pooled the high-performing CMIP6 (Sup-
plementary Table S2) multi-model ensemble simulations as though they
were from one model, as has commonly been adopted in previous event
attribution studies (Chiang et al., 2021; Min et al., 2022), because we did
not nd inconsistencies between the modelled variability of the com-
pound index and that in the observations when assessing their standard
deviations (Supplementary Fig. S6). In detail, we estimated the forced
response by rst averaging the individual CMIP6 model runs and then
averaging all the available models. The modelled variability was esti-
mated by the individual model simulation minus the forced response.
We then compared the standard deviation of the detrended compound
index in the observations with that in the modelled variability (Sup-
plementary Fig. S6).
For probability analysis (Fig. 3), we adopted the concepts of fraction
of attributable risk (FAR; Stott et al., 2004) and probability ratio (PR;
Fischer and Knutti, 2015) to carry out the event attribution and future
projection. FAR was dened as–1 −P0/P1, in which P1 indicated the
occurrence probability of similar events under factual climate condi-
tions, and P0 indicated that under counterfactual climate conditions
(Stott et al., 2004). PR was dened as P1/P0 (Fischer and Knutti, 2015).
We regarded the previous record-breaking value (in the year of 2016) as
the threshold, since 2020 was the single year that exceeded the previous
record (Fig. 1f), and then analysed the occurrence probability of similar
events (equal to or larger than the threshold). Using this threshold al-
lows us to examine whether the occurrence probability of experiencing a
more severe year than the record-breaking 2016 (such as 2020) has
Fig. 3. Attribution and projection of the compound extreme event based on the risk-based approach. (a) GEV distributions of the compound index in the
Hist2020 (red) and HistNat2020 (blue) simulations from the HadGEM3-GA6-N216 model. The dashed and solid lines represent the intensities of the 2016 and 2020
events, respectively. The grey line shows the fraction of attributable risk (FAR; see y-axis on the right side of the panel). (b) The FAR value (bottom axis, the
corresponding probability ratio (PR) is shown on the top axis) associated with the 2020-like event based on the HadGEM3-GA6-N216 (black) and CMIP6 (purple, the
multi-model ensemble is shown) models. Horizontal bars denote the 5%–95% uncertainty range (see Section 2.5). (c) The PR obtained from the GEV distributions
based on three different scenarios in CMIP6 (a 20-year moving window from 2021 to 2100) compared with the near-present-day climate (historical simulations of
2001–2020). Shaded areas denote the 5%–95% uncertainty range (see Section 2.5). (For interpretation of the references to colour in this gure legend, the reader is
referred to the Web version of this article.)
C. Qian et al.

Weather and Climate Extremes 42 (2023) 100616
7
changed. Selection of this threshold, rather than the value in 2020, is
also a way to reduce the selection bias and has commonly been used in
previous event attribution studies (Stott et al., 2004; Lewis and Karoly,
2013; King et al., 2014; Knutson et al., 2014). Using the
HadGEM3-GA6-N216 model, we tted the GEV distribution to com-
pound index values in the Hist2020 and HistNat2020 simulations and
calculated the corresponding FAR/PR values to show the effect of
anthropogenic forcings on the frequency of similar events (Fig. 3a and
b). We also compared the results with those from the CMIP6 models,
using historical simulations of 2001–2020 as near-present-day climate
conditions and hist-nat simulations of 1850–2020 as the counterfactual
climate conditions, as described earlier (Fig. 3b). The 95% condence
intervals of FAR/PR were calculated based on the Koopman method
(Koopman, 1984). This method regards the occurrence of events as
samples from a binomial distribution, which can be counted and then
used to calculate test statistics (Paciorek et al., 2018; Zscheischler and
Lehner, 2022).
To further explore the role of precipitation and temperature in
changes in the compound index, we used the normal kernel function to
construct univariate and bivariate probability density functions
(Bowman and Azzalini, 1997) to demonstrate changes such as that of the
R% anomaly and T
max
anomaly in different simulations (Fig. 4).
3. Results
3.1. Observed trends in the spatially compound event
Driven by an anomalous convergence of moisture ux, the observed
precipitation percentage anomaly (R% anomaly) averaged in the middle
and lower reaches of the Yangtze River over June and July 2020 was
record-breaking based on data going back to 1961 (Fig. 1a and b). The
magnitude of the R% anomaly in 2020 reached 95%, 1.7 times the
previous record in 1996, which was 56% (Fig. 1b). At the same time,
South China experienced an anomalously high number of hot days,
reaching up to 36 days for an individual grid box (Fig. 1c). The regional
average daily maximum temperature (T
max
) anomaly was also record-
breaking (Fig. 1d). The magnitude of the T
max
anomaly in 2020 was
1.6 ◦C, 1.14 times the previous record in 2016, which was 1.4 ◦C
(Fig. 1d).
To capture the magnitude of the spatially compounding event, we
employed a compound index, which is the average of the June–July
normalized precipitation anomaly in the middle and lower reaches of
the Yangtze River in percentages (box in Fig. 1a) and the normalized
temperature anomaly in South China (box in Fig. 1c) (see Section 2.3;
isolines in Fig. 1e). The compound index shows a strongly increasing
trend, with a magnitude of 0.22/decade (P =0.00014, section 2.3)
during 1961–2020 (Fig. 1f). Its intensity in 2020 was also record-
breaking, with an anomaly of four standard deviations (
σ
) (Fig. 1e and
f). Both the T
max
anomaly and R% anomaly in 2020 were very large
(2.5
σ
and 5.4
σ
, respectively) (Fig. 1e).
3.2. Attribution of the intensity of the 2020 event
To conduct the event attribution analysis, we used climate model
simulations. We evaluated the performance of the HadGEM3-GA6-N216
model and found that the historical simulations captured the temporal
evolutions of the R% anomaly, T
max
anomaly, and the compound index
(Fig. 1b, d, f). Furthermore, the model represents the distribution of the
R% anomaly, T
max
anomaly, and their bivariate distribution and corre-
lation well (Supplementary Table S2 and Section 2.2). We therefore
conclude that this model can be used in the subsequent analysis.
Based on a conditional storyline approach, we quantied the dy-
namic and thermodynamic contributions to the intensity of the 2020
compound event (Fig. 2). To achieve this, we identied atmospheric
circulation patterns that resembled the pattern of 2020, i.e., analogues,
and reconstructed the compound index using our newly developed CFA
method (see Section 2.4). We found that the dynamic contribution was
about 51% (95% condence intervals: 35–67%) (Fig. 2a, d), which
implied that the atmospheric circulation played a signicant role in the
occurrence of this event. There is no trend in the temporal evolution of
the number of analogues during 1961–2019 (Supplementary Fig. S7),
indicating that the occurrence of the 2020 dynamic conguration is
mainly controlled by the internal climate variability. Further calculation
revealed that atmospheric circulation contributed 56% (35–77%) to the
intensity of the R% anomaly and 41% (21–61%) to that of the T
max
anomaly in 2020. The thermodynamic contribution was estimated to be
about 39% (18–59%); this contribution was obtained via conditioning
the atmospheric circulations to resemble those of 2020 in the HadGEM3-
Fig. 4. Modelled changes in the univariate and bivariate distributions relevant to the event. (a) Univariate distributions of the precipitation anomaly in
percentages (upper panel) and the T
max
anomaly in ◦C (right panel) and their bivariate distributions (bottom left panel) based on Hist2020 (red) and HistNat2020
(blue) simulations from the HadGEM3-GA6-N216 model. The contour lines, from smallest to largest, represent 50%, 75%, and 95% of all data points. The dashed
black and red lines represent the intensities of the 2016 and 2020 events, respectively. (b) The same as (a), but for the CMIP6 model based on historical (red,
2001–2020), hist-nat (blue, 1850–2020), SSP1-1.9 (grey, 2081–2100), SSP2-4.5 (black, 2081–2100), and SSP5-8.5 (purple, 2081–2100) simulations. The contour
lines represent 50% and 95% of all data points. (For interpretation of the references to colour in this gure legend, the reader is referred to the Web version of
this article.)
C. Qian et al.

Weather and Climate Extremes 42 (2023) 100616
8
GA6-N216 model’s simulations with and without anthropogenic forc-
ings (Fig. 2b, d; see Section 2.4). This magnitude implied that the
radiatively forced component also played a considerable role in the
occurrence of the event.
The aforementioned thermodynamic contribution was compared
with that of the overall anthropogenic forcings to the intensity of events
with the same return period as the observed compound event in 2020,
allowing us to consider the possible effect of anthropogenic forcings on
changes in atmospheric circulation. This was conducted through a risk-
based attribution that did not condition the atmospheric circulation in
the HadGEM3-GA6-N216 model’s simulations (Fig. 2c and see Section
2.5). We found that the overall anthropogenic contribution was about
37% (31–42%), i.e., very close to and statistically indistinguishable from
the thermodynamic effect by using the storyline approach (Fig. 2d).
These similar estimates of the thermodynamic and overall anthropo-
genic contributions indicate that the anthropogenically driven change in
the frequency of the atmospheric circulation patterns is rather minor.
This is also in line with the nonsignicant trends found for the occur-
rence of ow analogues (Supplementary Fig. S7). If we draw conclusions
based on the result of the thermodynamic effect from the storyline
approach alone, one may argue that anthropogenic forcings can affect
atmospheric circulation too, and thus conditioning the atmospheric
circulation may overestimate or underestimate the effect of anthropo-
genic forcings, thereby leaving the conclusion of the attribution
assessment somewhat uncertain. In contrast, if we draw conclusions
based on the result of the overall anthropogenic contribution from the
risk-based approach alone, one may also argue that climate models may
not represent the atmospheric circulation very well, and thus induce
uncertainty in the contribution of anthropogenic forcings. Therefore,
combining the storyline attribution approach with the risk-based attri-
bution approach allowed us to build a complementary combined
framework, thereby enhancing our condence in the attribution
statements.
To more comprehensively estimate the anthropogenic climate
change effect on the event magnitude, we compared the aforementioned
contribution of the anthropogenic forcings to the intensity of similar
events with those from high-performing CMIP6 models (Fig. 2d; see
Section 2.2, 2.5, and Supplementary Tables S1 and S2). Note that, in
contrast to the HadGEM3-GA6-N216 model, the SST/SIC and radiative
forcings of 2020 are not prescribed in the CMIP6 simulations. We pooled
all available CMIP6 simulations, as has commonly been done in previous
event attribution studies (Christidis and Stott, 2022; Min et al., 2022).
This pooling of CMIP6 models can be justied because the models have
variability in the compound index that is consistent with that in the
observations (Supplementary Fig. S6 and Section 2.5); thus, they do not
require a bias correction (Christidis and Stott, 2022). The overall
anthropogenic contribution was 35% (32–37%), according to the CMIP6
multi-model ensemble. This magnitude is very close to that from the
HadGEM3-GA6-N216 model (Fig. 2d). This similarity suggests only a
minor inuence of the SST/SIC conditions in the intensity of 2020-like
events, regardless of the model differences. When we further sepa-
rated the effect of the GHG forcing and AA forcing, we found that the
contribution from GHG forcing was 70% (60–78%). This contribution
was partly cancelled out by AA forcing, which was −30% (−49% to
−22%). The magnitude of these two contributions is generally consis-
tent among the individual CMIP6 models, except for in the MIROC6
model under ANT (anthropogenic) and AA forcings, although there are
some inter-model differences (Supplementary Fig. S8a).
3.3. Attribution and projection of the likelihood of similar events
We quantied the contribution of anthropogenic forcings to the
occurrence probability of 2020-like compound events based on
HadGEM3-A-N216 and CMIP6 (Fig. 3a and b and Section 2.5).
HadGEM3-A-N216 simulates a distribution of the compound index
shifted toward higher values under historical anthropogenic forcing
compared to a scenario with no emissions (Fig. 3a). Anthropogenic
climate change has thus increased the likelihood of such extreme com-
pound events. The observed intensity of the 2020 compound index in the
model-simulated response to the current level of anthropogenic forcing
was unprecedented, indicating how unusual the 2020 compound event
was. Hence, we considered the intensity of the previous record-breaking
event (2016) as the threshold by which to quantify the probability of a
more intense event, such as that in 2020 (see Section 2.5 for more de-
tails). Both the T
max
and R% anomalies were very large in 2016 (2.2
σ
and 2.4
σ
, respectively), although not as strong as those in 2020 (Fig. 1e).
The FAR (Stott et al., 2004) of similar events is 0.99 (0.91, 1), equal to a
PR (Fischer and Knutti, 2015) of 442.4 (10.8, Inf) based on the
HadGEM3-GA6-N216 model (Fig. 3a and b). This value indicates a
substantial contribution of anthropogenic forcings to the occurrence
frequency of 2020-like events. The results based on the CMIP6
multi-model ensemble are similar, with a FAR of 0.98 (0.89, 1), equal to
a PR of 52.6 (9.5, Inf) (Fig. 3b). The magnitude of the increase in PR is
consistent among the individual CMIP6 models, except for the MIROC6
model (Supplementary Fig. S8b).
We quantied the projected occurrence probability of 2020-like
compound events because of the relevance for future planning of pro-
jected changes in the frequency of extreme events under climate change.
We considered a moving 20-year window from 2021 to 2100 using the
CMIP6 multi-model ensembles under three typical emissions scenarios
(Fig. 3c). The SSP5-8.5, SSP2-4.5, and SSP1-1.9 scenarios respectively
represent a very high GHG emissions scenario, an intermediate GHG
emissions scenario, and a scenario with very low GHG emissions and
CO
2
emissions declining to net zero around 2060 followed by net-
negative CO
2
emissions (IPCC, 2021). The SSP1-1.9 scenario will lead
to global warming below 1.5 ◦C in 2100, in line with the Paris Agree-
ment’s 1.5 ◦C target. The net-zero timing in the SSP1-1.9 scenario is
coincidently close to that in China’s carbon neutrality scheme. We found
that the future increase in the occurrence frequency of 2020-like com-
pound events (using the value in 2016 as the threshold too) is higher in
higher-emissions scenarios (Fig. 3c). Compared to near-present climate
conditions (2001–2020), the event will become 10 times more frequent
in 2050 and 14 times more frequent in 2090 under the SSP5-8.5 sce-
nario; however, under the SSP1-1.9 scenario, the occurrence frequency
for both years is only seven times greater (Supplementary Table S3).
To further understand the role of changes in precipitation and tem-
perature in changes in the compound index, we explored their univari-
ate and bivariate probability density functions estimated via the normal
kernel function (Fig. 4 and Section 2.5). We found that precipitation is
lower in simulations with anthropogenic forcing compared to those
without anthropogenic forcing conditions in both HadGEM3-GA6-N216
(using the distribution in 2020 for reference, Fig. 4a) and the CMIP6
multi-model ensemble (using the distribution of the near-present-day
climate as reference, Fig. 4b). This phenomenon is also noticeable in
the time series in Fig. 1b, perhaps due to the effect of AA forcing sur-
passing that of GHG forcing over this stage. However, precipitation is
projected to increase under future scenarios by the end of this century
(Fig. 4b). The intensity of the observed 2020 precipitation event is likely
an extreme value even in the SSP5-8.5 scenario (Fig. 4b), indicating the
severity of the 2020 event. In contrast, temperature always tends to
increase with anthropogenic forcing (Fig. 4a and b). In particular, the
magnitude of the observed 2020 temperature event will be fairly normal
by the end of this century under the SSP2-4.5 and SSP5-8.5 scenarios
(Fig. 4b).
4. Discussion and conclusions
This paper presents a comprehensive framework for compound event
attribution that combines a storyline approach with a risk-based
approach to reach complementary conclusions. We refer to this
approach as the storyline–probability combined approach. In the
storyline approach, we also developed a novel constructed ow
C. Qian et al.

Weather and Climate Extremes 42 (2023) 100616
9
analogue method to quantify the dynamic and thermodynamic contri-
butions to the intensity of an extreme event. This approach is exible
and adaptable to other regions and other types of compound events or
individual extreme events.
We found that the occurrence probability of spatially compounding
events such as the record-breaking 2020 event in China, which caused
quasi-simultaneous oods and extreme heatwaves over different regions
of the country, was increased by anthropogenic climate change and is
projected to further increase in the future. Floods and heatwaves over
different regions, which may both have negative repercussions for the
agricultural sector, are generally considered separately. For example,
recent studies revealed that extreme rainfall reduced one-twelfth of
China’s rice yield over the last two decades (Fu et al., 2023), and that
heat stress may cause a signicant reduction of rice yield in China under
future climate scenarios (Sun et al., 2022). However, the agricultural
impacts of the 2020 event in China suggest that co-occurring weather
hazards in different regions can compound one another and lead to
aggregated national impacts. Overall, our study demonstrates the rele-
vance of considering such spatially compounding events in attribution
studies to avoid underestimating climate risks and to support the
development of adaptation measures for the changing climate.
Furthermore, our results suggest that controlling GHG emissions can
reduce the occurrence risk of similar compound events, especially under
a carbon-neutral scenario. This effect takes place through both the
mitigation of warming and heavy precipitation, although the effect on
the former is stronger.
Author contributions
Cheng Qian conceived the research and designed the study with
Yangbo Ye, incorporating suggestions from Jakob Zscheischler and
Emanuele Bevacqua; Yangbo Ye performed the analyses; Cheng Qian
wrote the initial manuscript with input from Yangbo Ye; Jakob
Zscheischler and Emanuele Bevacqua improved the manuscript.
Declaration of competing interest
The authors declare the following nancial interests/personal re-
lationships which may be considered as potential competing interests:
Cheng Qian reports article publishing charges was provided by the Na-
tional Natural Science Foundation of China.
Data availability
Data Availability is declared in the manuscript
Acknowledgments
Cheng Qian and Yangbo Ye were sponsored by the National Natural
Science Foundation of China (grant no. 42341203) and the Jiangsu
Collaborative Innovation Center for Climate Change. Jakob Zscheischler
and Emanuele Bevacqua acknowledge the European COST Action
DAMOCLES (grant no. CA17109), the Helmholtz Initiative and
Networking Fund (Young Investigator Group COMPOUNDX; grant no.
VH-NG-1537) and the European Union’s Horizon 2020 research and
innovation programme under grant no. 101003469 (XAIDA). The au-
thors thank the two anonymous reviewers for their comments and
suggestions to improve this manuscript. Cheng Qian also thanks the
Climate Science for Service Partnership China (CSSP China) project,
sponsored by the UK–China Research and Innovation Partnership Fund
through the Met Ofce CSSP China as part of the Newton Fund, for
providing the HadGEM3-GA6-N216 simulation data.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.wace.2023.100616.
References
Bevacqua, E., De Michele, C., Manning, C., et al., 2021. Guidelines for studying diverse
types of compound weather and climate events. Earth’s Future 9, e2021EF002340.
https://doi.org/10.1029/2021ef002340.
Bevacqua, E., Suarez-Gutierrez, L., Jezequel, A., et al., 2023. Advancing research on
compound weather and climate events via large ensemble model simulations. Nat.
Commun. 14, 2145. https://doi.org/10.1038/s41467-023-37847-5.
Bowman, A.W., Azzalini, A., 1997. Applied Smoothing Techniques for Data Analysis.
New York.
Chiang, F., Greve, P., Mazdiyasni, O., et al., 2021. A multivariate conditional probability
ratio framework for the detection and attribution of compound climate extremes.
Geophys. Res. Lett. 48 https://doi.org/10.1029/2021gl094361.
Christidis, N., Stott, P.A., 2015. Extreme rainfall in the United Kingdom during winter
2013/14: the role of atmospheric circulation and climate change. Bull. Am.
Meteorol. Soc. 96, S46–S50. https://doi.org/10.1175/bams-d-15-00094.1.
Christidis, N., Stott, P.A., 2022. Anthropogenic climate change and the record-high
temperature of May 2020 in western Europe. Bull. Am. Meteorol. Soc. 103, S33–S37.
https://doi.org/10.1175/BAMS-D-21-0128.1.
Ciavarella, A., Christidis, N., Andrews, M., et al., 2018. Upgrade of the HadGEM3-A
based attribution system to high resolution and a new validation framework for
probabilistic event attribution. Weather Clim. Extrem. 20, 9–32. https://doi.org/
10.1016/j.wace.2018.03.003.
Dabernig, M., Mayr, G.J., Messner, J.W., et al., 2017. Spatial ensemble post-processing
with standardized anomalies. Quart. J. Roy. Meteor. Soc. 143, 909–916. https://doi.
org/10.1002/qj.2975.
Deser, C., Terray, L., Phillips, A.S., 2016. Forced and internal components of winter air
temperature trends over north America during the past 50 years: mechanisms and
implications. J. Clim. 29, 2237–2258. https://doi.org/10.1175/jcli-d-15-0304.1.
Eyring, V., Bony, S., Meehl, G.A., et al., 2016. Overview of the coupled model
intercomparison project phase 6 (CMIP6) experimental design and organization.
Geosci. Model Dev. (GMD) 9, 1937–1958. https://doi.org/10.5194/gmd-9-1937-
2016.
Fischer, E.M., Knutti, R., 2015. Anthropogenic contribution to global occurrence of
heavy-precipitation and high-temperature extremes. Nat. Clim. Change 5, 560–564.
https://doi.org/10.1038/nclimate2617.
Fu, J., Jian, Y.W., Wang, X.H., et al., 2023. Extreme rainfall reduces one-twelfth of
China’s rice yield over the last two decades. Nature Food 4, 416–426. https://doi.
org/10.1038/s43016-023-00753-6.
Genest, C., R´
emillard, B., Beaudoin, D., 2009. Goodness-of-t tests for copulas: a review
and a power study. Insur. Math. Econ. 44, 199–213. https://doi.org/10.1016/j.
insmatheco.2007.10.005.
Gillett, N.P., Shiogama, H., Funke, B., et al., 2016. The detection and attribution model
intercomparison project (damip v1.0) contribution to CMIP6. Geosci. Model Dev.
(GMD) 9, 3685–3697. https://doi.org/10.5194/gmd-9-3685-2016.
Herring, S.C., Christidis, N., Hoell, A., et al., 2022. Explaining extreme events of 2020
from a climate perspective. Bull. Am. Meteorol. Soc. 103, S1–S129. https://doi.org/
10.1175/BAMS-ExplainingExtremeEvents2020.1.
IPCC, 2021. In: Zhai, V.P., Pirani, A., et al. (Eds.), Summary for Policymakers. Climate
Change 2021: the Physical Science Basis. Contribution of Working Group I to the
Sixth Assessment Report of the Intergovernmental Panel on Climate Change,
Masson-Delmotte. Cambridge University Press.
J´
ez´
equel, A., Yiou, P., Radanovics, S., 2018. Role of circulation in European heatwaves
using ow analogues. Clim. Dynam. 50, 1145–1159. https://doi.org/10.1007/
s00382-017-3667-0.
Kalnay, E., Kanamitsu, M., Kistler, R., et al., 1996. The ncep/ncar 40-year reanalysis
project. Bull. Am. Meteorol. Soc. 77, 437–471. https://doi.org/10.1175/1520-0477
(1996)077<0437:Tnyrp>2.0.Co;2.
King, A.D., Karoly, D.J., Donat, M.G., et al., 2014. Climate change turns Australia’s 2013
big dry into a year of record-breaking heat. Bull. Am. Meteorol. Soc. 95, S41–S45.
Knutson, T.R., Zeng, F.R., Wittenberg, A.T., 2014. Seasonal and annual mean
precipitation extremes occurring during 2013: a U.S. Focused analysis. Bull. Am.
Meteorol. Soc. 95, S19–S23.
Koopman, P.A.R., 1984. Condence intervals for the ratio of two binomial proportions.
Biometrics 40, 513–517. https://doi.org/10.2307/253.
Lewis, S.C., Karoly, D.J., 2013. Anthropogenic contributions to Australia’s record
summer temperatures of 2013. Geophys. Res. Lett. 40, 3705–3709. https://doi.org/
10.1002/grl.50673.
Lu, C., Sun, Y., Zhang, X., 2022. The 2020 record-breaking mei-yu in the Yangtze River
valley of China: the role of anthropogenic forcing and atmospheric circulation. Bull.
Am. Meteorol. Soc. 103, S98–S104. https://doi.org/10.1175/BAMS-D-21-0161.1.
Min, S.-K., Jo, S.-Y., Seong, M.-G., et al., 2022. Human contribution to the 2020 summer
successive hot-wet extremes in South Korea. Bull. Am. Meteorol. Soc. 103, S90–S97.
https://doi.org/10.1175/BAMS-D-21-0144.1.
O’Neill, B.C., Tebaldi, C., van Vuuren, D.P., et al., 2016. The scenario model
intercomparison project (scenariomip) for CMIP6. Geosci. Model Dev. (GMD) 9,
3461–3482. https://doi.org/10.5194/gmd-9-3461-2016.
Otto, F.E.L., van Oldenborgh, G.J., Eden, J., et al., 2016. The attribution question. Nat.
Clim. Change 6, 813–816. https://doi.org/10.1038/nclimate3089.
Paciorek, C.J., Stone, D.A., Wehner, M.F., 2018. Quantifying statistical uncertainty in the
attribution of human inuence on severe weather. Weather Clim. Extrem. 20, 69–80.
https://doi.org/10.1016/j.wace.2018.01.002.
C. Qian et al.

Weather and Climate Extremes 42 (2023) 100616
10
Philip, S., Kew, S., van Oldenborgh, G.J., et al., 2020. A protocol for probabilistic
extreme event attribution analyses. Adv. Stat. Climatol. Meteorol. Oceanogr. 6,
177–203. https://doi.org/10.5194/ascmo-6-177-2020.
Qian, C., Zhang, X.B., Li, Z., 2019. Linear trends in temperature extremes in China, with
an emphasis on non-Gaussian and serially dependent characteristics. Clim. Dynam.
53, 533–550. https://doi.org/10.1007/s00382-018-4600-x.
Qian, C., Ye, Y.B., Chen, Y., et al., 2022a. An updated review of event attribution
approaches. J. Meteorol. Res. 36, 227–238. https://doi.org/10.1007/s13351-022-
1192-5.
Qian, C., Ye, Y.B., Zhang, W.X., et al., 2022b. Heavy rainfall event in mid-August 2020 in
southwestern China: contribution of anthropogenic forcings and atmospheric
circulation. Bull. Am. Meteorol. Soc. 103, S111–S117. https://doi.org/10.1175/
bams-d-21-0233.1.
Qin, X.L., Shi, Q., Wang, D.Z., et al., 2022. Inundation impact on croplands of 2020 ood
event in three provinces of China. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 15,
3179–3189. https://doi.org/10.1109/jstars.2022.3161320.
Seneviratne, S.I., Zhang, X., Adnan, M., et al., 2021. In: Zhai, V.P., Pirani, A., et al. (Eds.),
Climate Change 2021: the Physical Science Basis. Contribution of Working Group I to
the Sixth Assessment Report of the Intergovernmental Panel on Climate Change,
Masson-Delmotte. Cambridge University Press, p. 1611.
Shepherd, T.G., 2016. A common framework for approaches to extreme event
attribution. Curr. Clim. Change Rep. 2, 28–38. https://doi.org/10.1007/s40641-
016-0033-y.
Stott, P.A., Stone, D.A., Allen, M.R., 2004. Human contribution to the European
heatwave of 2003. Nature 432, 610–614. https://doi.org/10.1038/nature03089.
Stott, P.A., 2016. How climate change affects extreme weather events research can
increasingly determine the contribution of climate change to extreme events such as
droughts. Science 352, 1517–1518. https://doi.org/10.1126/science.aaf7271.
Stott, P.A., Christidis, N., Otto, F.E.L., et al., 2016. Attribution of extreme weather and
climate-related events. WIRES Clim. Chang. 7, 23–41. https://doi.org/10.1002/
wcc.380.
Sun, Q., Zhao, Y.X., Zhang, Y., et al., 2022. Heat stress may cause a signicant reduction
of rice yield in China under future climate scenarios. Sci. Total Environ. 818 https://
doi.org/10.1016/j.scitotenv.2021.151746.
Tang, H., Wang, Z., Tang, B., et al., 2022. Reduced probability of 2020 June July
persistent heavy mei-yu rainfall event in the Middle to Lower Reaches of the Yangtze
River basin under anthropogenic forcing. Bull. Am. Meteorol. Soc. 103, S83–S89.
https://doi.org/10.1175/bams-d-21-0167.1.
Trenberth, K.E., Fasullo, J.T., Shepherd, T.G., 2015. Attribution of climate extreme
events. Nat. Clim. Change 5, 725–730. https://doi.org/10.1038/nclimate2657.
van Oldenborgh, G.J., van der Wiel, K., Kew, S., et al., 2021. Pathways and pitfalls in
extreme event attribution. Clim. Change 166, 13. https://doi.org/10.1007/s10584-
021-03071-7.
Verschuur, J., Li, S.H., Wolski, P., et al., 2021. Climate change as a driver of food
insecurity in the 2007 Lesotho-south africa drought. Sci. Rep. 11 https://doi.org/
10.1038/s41598-021-83375-x.
Vogel, M.M., Zscheischler, J., Wartenburger, R., et al., 2019. Concurrent 2018 hot
extremes across Northern Hemisphere due to human-induced climate change.
Earth’s Future 7, 692–703. https://doi.org/10.1029/2019EF001189.
Wang, X.L., Swail, V.R., 2001. Changes of extreme wave heights in Northern Hemisphere
oceans and related atmospheric circulation regimes. J. Clim. 14, 2204–2221.
https://doi.org/10.1175/1520-0442(2001)014<2204:Coewhi>2.0.Co;2.
Wu, J., Gao, X.-J., 2013. A gridded daily observation dataset over China region and
comparison with the other datasets. in Chinese Chin. J. Geophys. 56, 1102–1111.
https://doi.org/10.6038/cjg20130406.
Ye, Y., Qian, C., 2021. Conditional attribution of climate change and atmospheric
circulation contributing to the record-breaking precipitation and temperature event
of summer 2020 in southern China. Environ. Res. Lett. 16, 044058 https://doi.org/
10.1088/1748-9326/abeeaf.
Yiou, P., Vautard, R., Naveau, P., et al., 2007. Inconsistency between atmospheric
dynamics and temperatures during the exceptional 2006/2007 fall/winter and
recent warming in Europe. Geophys. Res. Lett. 34, 52–94. https://doi.org/10.1029/
2007gl031981.
Zhang, W., Li, W., Zhu, L., et al., 2020. Anthropogenic inuence on 2018 summer
persistent heavy rainfall in central western China. Bull. Am. Meteorol. Soc. 101,
S65–S70. https://doi.org/10.1175/bams-d-19-0147.1.
Zhou, T., Ren, L., Zhang, W., 2021. Anthropogenic inuence on extreme meiyu rainfall in
2020 and its future risk. Sci. China Earth Sci. 64, 1633–1644. https://doi.org/
10.1007/s11430-020-9771-8.
Zscheischler, J., Westra, S., van den Hurk, B., et al., 2018. Future climate risk from
compound events. Nat. Clim. Change 8, 469–477. https://doi.org/10.1038/s41558-
018-0156-3.
Zscheischler, J., Martius, O., Westra, S., et al., 2020. A typology of compound weather
and climate events. Nat. Rev. Earth Environ. 1, 333–347. https://doi.org/10.1038/
s43017-020-0060-z.
Zscheischler, J., Lehner, F., 2022. Attributing compound events to anthropogenic climate
change. Bull. Am. Meteorol. Soc. 103, E936–E953. https://doi.org/10.1175/bams-d-
21-0116.1.
C. Qian et al.