You are on page 1of 29

Received: 7 May 2021 Revised: 28 August 2021 Accepted: 7 September 2021 Published on: 23 September 2021

DOI: 10.1002/joc.7386

RESEARCH ARTICLE

Differences in multi-model ensembles of CMIP5 and CMIP6


projections for future droughts in South Korea

Young Hoon Song1 | Shamsuddin Shahid2 | Eun-Sung Chung1

1
Faculty of Civil Engineering, Seoul
National University of Science and
Abstract
Technology, Nowon-gu, Seoul, Korea This study quantified the uncertainties of future drought projections in the
2
School of Civil Engineering, Faculty of South Korea case by means of the reliability ensemble averaging (REA) of
Engineering, Universiti Teknologi
10 GCM equivalents of the Coupled Model Intercomparison Project, Phases
Malaysia (UTM), Johor Bahru, Malaysia
5 and 6 (CMIP5 and CMIP6). Two meteorological drought indices, the Stan-
Correspondence dardized Precipitation Index (SPI) and the Standardized Precipitation-
Eun-Sung Chung, Faculty of Civil
Engineering, Seoul National University of
Evapotranspiration Index (SPEI) for Representative Concentration Pathway
Science and Technology, 232 Gongneung- (RCP) 4.5 and RCP8.5 of CMIP5 and Shared Socioeconomic Pathway (SSP)
ro, Nowon-gu, Seoul 01811, Korea. 2-4.5 and SSP5-8.5 of CMIP6 were considered. The GCMs' performances for
Email: eschung@seoultech.ac.kr
the historical period were evaluated, and their biases were corrected using qua-
Funding information ntile mapping. The multi-model ensemble (MME) means of the GCMs were
K-water, Grant/Award Number: D-W-004
generated using the entropy and TOPSIS methods. The results showed that the
TOPSIS MME estimated more intense droughts than the entropy MME. The
levels of SPI and SPEI severity estimated using the entropy MME were higher
than those from the TOPSIS MME. The SPI and SPEI outcomes of RCP4.5
were much more robust than those of SSP2-4.5. The projected drought severity
in the near future was much greater than in the far future, while the reliability
of drought projections in the far future was much higher than in the near
future. The reliability levels of drought projections for the SSP scenarios were
higher than for the RCPs for most durations. This study can support planning
and management of future droughts considering uncertainty variables.

KEYWORDS
CMIP6, entropy, reliability ensemble averaging, shared socioeconomic pathways, TOPSIS

1 | INTRODUCTION altered the spatial and temporal patterns of precipitation


and temperature, outcomes which in turn have increased
Because drought has a lasting impact on agriculture, the the frequency and severity of hydrological disasters
economy, and the environment and nature, it is often (Asdak et al., 2018; Shiru et al., 2018; Alamgir et al. 2019;
more devastating than any other natural hazard. Cumu- Ayugi et al. 2020; Khan et al., 2020; Salman et al., 2020).
lative agricultural production losses due to droughts Droughts are more sensitive to climate change among
amounted to nearly US$166 billion during 1983–2009 the hydrological disasters due to their complex relation-
(Kim et al., 2019). Conforti et al. (2018) estimated US$29 ship with multiple climate factors (Kim et al., 2021; Cook
billion losses in developing countries due to droughts et al., 2020; Berg et al., 2017; Lemordant et al., 2018;
during the period of 2005–2015. Global warming has Mankin et al., 2017). Therefore, climate change has

2688 © 2021 Royal Meteorological Society wileyonlinelibrary.com/journal/joc Int J Climatol. 2022;42:2688–2716.


10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2689

increased droughts globally over the past few decades than their CMIP5 counterparts. The CMIP6 GCMs also
(Song et al., 2020). differ from CMIP5 in enhanced approximation of micro-
Most climate model simulations indicate a future physical processes and incorporation of multiple Earth
increase in drought frequency across the globe due to cli- processes, like biogeochemical cycles and ice sheets
mate change. Wang et al. (2021) analysed global drought (Eyring et al., 2016). This has made CMIP6 more capable
characteristics using newly released global climate of reliable climate projections and estimating climate
models (GCMs). They showed increases in drought recur- change impacts on different hydrological processes like
rence, durations and areal coverage areas in most parts of drought. The release of the CMIP3 GCM in 2007 acceler-
the globe. Pokhrel et al. (2021) showed that global land ated drought research across the globe. Many studies
areas under extreme drought would increase from 3% at using CMIP3 GCMs projected an increase in drought fre-
present to 7% by the end of this century, meaning that 8% quency and severity levels in the future in different
of the global population would live in extreme drought regions (Sheffield and Wood, 2007; Hernandez and
conditions. There will be some geographical variability of Uddameri 2014). CMIP5 provided GCM simulations of
drought hot spots due to spatially varying changes in the radiative concentration pathway (RCP) scenarios to
global climate. Spinoni et al. (2020) showed East Asian assess future climates over a broader range than those
countries would be among the possible drought hot spots used in CMIP3 scenarios. Most studies using CMIP5 pro-
due to global climate change. Liu and Chen (2021) evalu- jected increasing droughts over different regions of the
ated the socioeconomic risk of drought due to climate globe due to continuously rising temperatures (Wang and
change and revealed East Asia as being among the high Chen, 2014; Shiru et al., 2020; Khan et al., 2020). CMIP6
socioeconomic risk regions of global climate change- projections differ from those of CMIP5 in terms of future
induced drought. scenarios. CMIP6 employs the shared socioeconomic
As a country in East Asia, South Korea has frequently pathway (SSP) for future climate projections.
experienced drought due to the variability of the East The SSP scenario considers future socioeconomic fac-
Asian monsoon. Droughts in South Korea arise on both tors such as ecosystems, resources, and populations to
shorter and longer time scales, but shorter periods of estimate future climates. They also include mitigation
drought are more recurrent. However, the country has and adaptation efforts concerning climate change
also experienced several droughts of longer durations due (O'Neill et al., 2016). Therefore, SSP scenarios can pro-
to the consecutive failure of summer rainfall for two or vide more realistic climate predictions than other scenar-
more years. The most recent drought in 2013–2015 led to ios. Currently, many studies on climate change based on
abnormal stress on the water resources of South Korea SSP scenarios are underway (Global: Kim et al., 2020;
(Kwon et al., 2016). Climate change effect on drought Fan et al., 2020; China: Wu et al., 2019, South Korea:
characteristics is already noticeable on the Korean penin- Song et al., 2021a; Kim et al., 2021, Africa: Almazroui
sula (Oh et al., 2014). Park et al. (2020) reported more et al., 2020, Iran: Zamani et al., 2020, and US: Srivastava
severe droughts in East Asia, including South Korea, et al., 2020). Several recent studies also used SSP scenar-
when estimating future emission scenarios. ios for drought projections (Ukkola et al., 2020; Su
Long-term water management plans and strategies et al., 2021). However, most of these previous studies on
are required to cope with the water crisis intensified by drought projections were mainly based on the outputs of
climate change. A better understanding of possible spa- CMIP3 and CMIP5, while studies considering CMIP6
tiotemporal changes in drought conditions is essential for future scenarios are insufficient. Therefore, drought
this purpose. Therefore, projecting the trends and sever- assessments using the SSP scenario and comparisons of
ity levels of future droughts is a hot topic in drought drought severity levels and trends relative to those in pre-
research. The emission scenario of coupled model vious versions can lead to a better understanding of
intercomparison projects (CMIPs) can provide a climate future droughts. The combination of RCP and SSP can
factor to assist with projections of future droughts. Previ- improve knowledge of future climate change (Eyring
ous studies showed higher resolution GCMs provide et al., 2016; O'Neill et al., 2016).
more realistic precipitation projections (Wehner The GCM combining method known as the multi-
et al., 2014). A comprehensive study of the extreme pre- model ensemble (MME) method is frequently used to
cipitation indices of the CMIP5 model showed a better project future climates. However, one issue when using
representation of the precipitation characteristics due to MME is the removal of errors due to the averaging of
their higher resolution than the CMIP3 models, while the GCM values (Knutti et al. 2010). Many studies have rec-
uncertainty was similar (Sillmann et al., 2013; Gibson ommended quantifying weights to assign or exclude low
et al., 2019; Taylor et al., 2012; Wang et al., 2014). The weights for relatively less relevant GCMs (Ahmed
average CMIP6 GCMs' resolution is higher resolution et al. 2019; Shiru et al. 2019). Therefore, appropriate
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2690 SONG ET AL.

performance metrics and decision-making techniques Administration (KMA) for this study. The basis of choos-
should be selected, followed by a discussion and compar- ing the stations is the availability of monthly data for a
ative analysis of the applied historical and projected cli- longer period without missing values. The selected sta-
mate trends. tions are exposed to various disasters, such as floods,
GCM projections are uncertain due to errors arising heavy snow and droughts. Therefore, drought projections
from the model structure, the scenario considered, and at the selected stations can provide a regional assessment.
the initial conditions (Woldemeskel et al., 2014). Ghosh In this study, the 22 stations selected were divided into
and Mujumdar (2007) asserted that the output of GCM is four regions (Northwest (NW), Northeast (NE), South-
a source of high uncertainty due to the incomplete west (SW), and Southeast (SE)), as shown in Figure 1.
knowledge of future scenarios and physical processes that
are uncertain. Therefore, predicted future droughts using
GCM necessarily contain uncertainty. The uncertainty in 2.2 | Dataset and sources
GCM-based drought projection is primarily a discrepancy
in the magnitudes and signs of precipitation changes and This study used 10 GCMs available in both CMIP5 and
the magnitude of warming (Trenberth et al., 2014). Some CMIP6. Table 1 presents the names of the GCMs, the model-
studies comparing drought predictions obtained using ling centres, and the resolutions. The resolutions of the
different drought indices found that the estimation selected GCMs range from 0.93 to 2.81 . These 10 pairs of
method for drought characteristics can lead to uncer- CMIP5 and CMIP6 GCMs were selected as they have projec-
tainty in drought predictions for future periods tions of all required variables for a drought analysis for differ-
(Keyantash and Dracup, 2002; Mo, 2008; Dai, 2010). ent SSP and RCP scenarios. Each GCM's monthly
The purpose of the present study is to compare the precipitation, maximum temperature, and minimum tem-
spatiotemporal changes pertaining to SPI and SPEI perature hindcast for South Korea were evaluated for a com-
droughts based on the MME outcomes of 10 common mon baseline period (1970–2005) of both CMIP5 and
CMIP5 and CMIP6 GCMs for the near (2025–2060) and CMIP6. The SSP scenarios include social and economic fac-
far (2065–2100) futures in South Korea. MME was con- tors with radiative forcing from RCP (O'Neill et al., 2017).
structed using entropy weights and the technique for The SenariosMIP, which improved the integrated assessment
order of preference by similarity to ideal solution model for multi-model climate projection, was based on CO2
(TOPSIS), based on two evaluation metrics. The reliabil- emission and land-use changes (Riahi et al., 2015). Differ-
ity ensemble average (REA) method was employed to ent SSP scenarios are proposed according to adaptation
analyse the uncertainty in projected droughts for near and mitigation intensity, such as SSP1-2.6, SS2-4.5,
(2025–2060) and far (2065–2100) futures relative to the SSP4-6.0, and SSP5-8.5 (O'Neill et al., 2016; Wang
historical mean. The results of this study can help to et al., 2018). The projections for the SSP scenario allow
develop confidence in outcomes from CMIP6 GCMs for estimations of future energy and land-use changes based
drought impact assessments and can also be used to on adaptation and mitigation levels (O'Neill et al., 2016).
streamline adaptation measures formulated based on Therefore, the SSP scenarios offer more realistic climate
RCP scenarios. projections. RCP4.5 assumes that gas emissions stabilize
by the middle of the 21st century, whereas RCP8.5 corre-
sponds to a nominal anthropogenic forcing of 8.5 Wm−2
2 | S T UDY A R E A AN D D AT A until 2,100 (Riahi et al., 2015). SSP2-4.5 and SSP5-8.5 con-
sider similar emission levels to RCP4.5 and RCP8.5,
2.1 | Study area respectively (O'Neill et al., 2016). The RCP scenarios,
RCP4.5 and RCP8.5, and their equivalent SSP scenarios,
South Korea (35 500 N, 127 000 W) lies between Japan and SSP2-4.5 and SSP5-8.5, were used for drought projections.
China. The country has four distinct seasons: winter
(DJF), spring (MAM), summer (JJA), and autumn (SON).
The average annual precipitation is between 1,000 and 3 | METHODOLOG Y
1,600 mm. The majority of precipitation occurs in sum-
mer, while winter is the driest season. Drought occurs 3.1 | Drought index
mainly in spring due to failure in summer precipitation
and gradually worsens in winter. The average tempera- A drought can be meteorological, hydrological, or agri-
ture of the country ranges from 10 to 15 C. Monthly pre- cultural. Most droughts occur due to a lack of water.
cipitation and average temperature data observed at Therefore, droughts are sensitive to meteorological fac-
22 stations were collected from the Korea Meteorological tors such as precipitation, evapotranspiration, and
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2691

F I G U R E 1 Locations of the stations in South Korea used in the present study. The colour gradient represents the spatial variation in the
annual average precipitation and temperature during 1970–2005 [Colour figure can be viewed at wileyonlinelibrary.com]

TABLE 1 Information about the GCMs used in this study

Resolution
Institute Models (longitude × latitude)
Commonwealth Scientific and Industrial CMIP5 ACCESS1-3 1.25 × 1.875
Research Organisation, and Bureau of CMIP6 ACCESS-ESM1-5 1.25 × 1.875
Meteorology
Canadian Earth System Model version 5, CMIP5 CanESM2 2.81 × 2.81
Canadian Centre for Climate Modelling and CMIP6 CanESM5 2.81 × 2.81
Analysis (Canada)
NOAA/Geophysical Fluid Dynamics CMIP5 GFDL-ESM2G 2.02 × 2.00
Laboratory, Climate Model version 4 (USA) CMIP6 GFDL-ESM4 1.25 × 1.00
NASA Goddard institute for Space Studies CMIP5 GISS-E2-R 2.0 × 2.5
CMIP6 GISS-E2-1-G 2.0 × 2.5
Institute for Numerical Mathematics, Russian CMIP5 INM-CM4 1.5 × 2.0
Academy of Science, (Russia) CMIP6 INM-CM4-8 1.5 × 2.0
Institut Pierre Simon Laplace CMIP5 IPSL-CM5A-LR 2.5 × 1.26
CMIP6 IPSL-CM6A-LR 2.5 × 1.26
Atmosphere and Ocean Research Institute, The CMIP5 MIROC5 1.4 × 1.4
University of Tokyo CMIP6 MIROC6 1.4 × 1.4
Max Planck Institute for Meteorology (MPI-M) CMIP5 MPI-ESM-LR 1.125 × 1.12
(Germany) CMIP6 MPI-ESM1-2-LR 1.125 × 1.12
Meteorological Research Institute (Japan) CMIP5 MRI-CGCM3 1.125 × 1.07
CMIP6 MRI-ESM2-0 1.125 × 1.125
Norwegian Climate Centre (Norway) CMIP5 NorESM1-M 1.89 × 2.5
CMIP6 NorESM2-MM 1.25 × 0.9375
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2692 SONG ET AL.

temperature, highlighting the importance of drought esti- a coefficient depending on I, and K is the latitude of the
mations based on meteorological factors. The complexity observed data. Equation (4) is the difference between pre-
and severity of droughts make it more difficult to identify cipitation and evapotranspiration, where k is the time
and assess drought characteristics. Therefore, drought scale (Vicente-Serrano et al., 2010).
indices have been popularly used to evaluate and monitor
drought events. Among them, SPI and SPEI can better X
k−1

represent various types of drought and reflect the charac- Dkn = Pn − i −PET n − i ð4Þ
i=0
teristics of droughts. SPI considers only precipitation and
is used to determine the frequency and severity of
droughts quantitatively (McKee et al. 1993). SPEI calcu- SPEI employs a procedure similar to that of SPI, using D
lates the difference between precipitation and potential instead of rainfall to estimate droughts. Table 2 shows
evapotranspiration to assess droughts and thus also the drought classifications based on SPI and SPEI.
includes soil moisture in drought estimations (Vicente-
Serrano et al., 2010). Therefore, many studies have used
both SPI and SPEI for drought analyses (Liu et al., 2021; 3.2 | Quantile mapping
Labudova et al., 2017; Wang et al., 2017).
This study estimated drought severity levels using The GCM raw data does not fit the observed data. These
both SPEI and SPI. The selected drought duration time- errors are due to the coarse spatial resolution, imperfect
scales in this study were 3, 6, 9, and 12 months. The process parameterization, and incomplete knowledge of
durations were selected considering the frequent occur- the climate system (Stevens and Bony, 2013; Teutschbein
rence of short-term droughts in the country. SPI, intro- and Seibert, 2012). Quantile mapping (QM) is a com-
duced by McKee et al. (1995), quantifies drought at a monly used bias correction methods (Ringard et al., 2017;
given timescale using a fitting distribution on precipita- Cannon et al., 2015). Therefore, the precipitation and
tion data. The two-parameter gamma probability distri- temperature outputs of CMIP5 and CMIP6 GCMs were
bution function (pdf) is commonly used for SPI bias-corrected at the observed locations using the QM
estimation, considering its flexibility in representing pre- method.
cipitation time series (Karavitis et al., 2011; Shiau, 2020). QM is a non-parametric transformation that esti-
Therefore, SPI generally fits a gamma pdf to observed mates the values of the empirical cumulative density
precipitation data, as shown in Equation 1. functions of observed and modelled time series for reg-
ularly spaced quantiles (Enayati et al., 2021). The
1 smoothing spline method of the non-parametric trans-
x α− 1e − β , x≥0
x
gðx Þ= α ð1Þ
β τ ðα Þ formation is fitted in a quantile to quantile manner
with the observed and modelled times series
Here, α and β are the shape and scale parameters, x is the (Kouhestani et al. 2016; Gudmundsson et al., 2012).
amount of precipitation, and τðαÞ is the gamma function. This study used the smoothing spline for bias correc-
The cumulative probability is transformed into the stan- tion, as shown below,
dard normal random variable z. The gamma distribution
is not defined for x =0 and q=Pðx =0Þ>0, where Pðx =0Þ
is the probability of zero precipitation. Therefore, the T A B L E 2 Classification of different drought severity levels
cumulative pdf can be expressed using Equation (2). based on SPEI and SPI

Drought index Classification of drought (SPI


H ðx Þ =q + ð1 −qÞ × gðx Þ ð2Þ range and SPEI)
>2.00 Extremely wet
SPEI initially estimates the difference (D) between precipi-
1.50 to 1.99 Very wet
tation and potential evapotranspiration (PET). In this
1.00 to 1.49 Moderately wet
study, PET was calculated using the Thornthwaite method,
as expressed by Equation (3) (Thornthwaite, 1948), 0 to 0.99 Near normal
−0.99 to 0
 m
16T −1.00 to −1.49 Moderately dry
PET = 16K , ð3Þ
I −1.50 to −1.99 Severely dry
−2.00 to −2.50 Extremely dry
where T is the monthly average temperature ( C), I is the
<−2.50 Very extremely dry
annual heat index obtained by summing the month, m is
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2693

Po =F o− 1 ðF m ðPm ÞÞ, ð5Þ X


N
E= − wi lnwi ð7Þ
i=1

where, m is GCM simulation data, o is observed data, Pm


is the simulated value in GCM, F m is the cumulative dis- where wi is the weight assigned to each GCM and N is
tribution of Pm , Po is the precipitation and temperature of the total number of ensemble members.
the station, and F o− 1 is the inverse of the cumulative dis-
tribution of Po . Afterwards, correction factors were
applied to correct the bias in the projected data (2020– 3.5 | TOPSIS
2100) at each station. The QM method has been
concluded as effective for bias corrections of GCM The technique for order of preference by similarity to
simulations in South Korea (Song et al., 2020; Song ideal solution (TOPSIS) is a multi-criteria decision analy-
et al., 2021b). sis (MCDA) method proposed by Hwang and Yoon (1981).
TOPSIS is based on the concept that an alternative
should have the shortest distance from a positive ideal
3.3 | Inverse distance weight solution and the greatest distance from a negative ideal
solution (Chen and Hwang, 1992; Raju and
Because GCM outputs are in the form of a grid, spatial Kumar, 2010). Equations (8–10) show the TOPSIS calcu-
downscaling is needed to correct the difference between lation process, as formulated below.
the GCM simulation and station points. The IDW
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
method is frequently used for climate data interpolation u l
uX
(Longley et al., 2005). It uses the concept that closer dis- Da+ = t ðwi f i ðaÞ−wi f i Þ2 ð8Þ
tances are more relevant than greater distances, is the j=1

basis of the IDW concept (Tobler, 1970). Equation (6)


was used to interpolate the climate factors at the GCM Equation (7) serves to calculate the Euclidean distance of
grids to the observed locations. each criterion from its positive ideal solution and sum-
ming the overall criteria for alternative a, where j is the
P
N normalized value for alternative a and w is the weight
ðwi × Z xi Þ
assigned to the criterion j.
Pi = i = 1 ð6Þ
P
N
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
wi u l
i=1
− t
uX
Da = ðwi f i ðaÞ− wi f i Þ2 ð9Þ
j=1

Here, Pi is the observed data, Z x i represents the GCM-


simulated outputs at the grid points surrounding the Equation (9) calculates the distance between the
observed location, and wi is the interpolation weight, esti- target alternative and negative ideal solution, summing
mated as the inverse of the distance between the observa- these for all criteria for the alternative a, where j is the
tion location and the surrounding GCM grid points. normalized negative ideal solution.

Da−
Ca =   ð10Þ
3.4 | Entropy weight Da− +Da+

Shannon's entropy (Shannon, 1948), a statistical measure Equation (10) calculates the degree of similarity to the
of randomness, was used to generate the GCM MME negative ideal solution; the closer the similarity value is
(Pomerol and Romero, 2000). The weights of the GCMs to 1, the closer it is to the best alternative, and the closer
were calculated using entropy to combine the GCM the value is to 0, the closer it is to the worst alternative.
ensembles of climate factors. The climate factor of each
GCM estimates inequality in a set of weights of individ-
ual GCMs based on the evaluation index results and dis- 3.6 | Evaluation metrics
tinguishes the relative performance between GCM
outputs. Entropy reaches its maximum value when the This study used two evaluation metrics to assess the per-
weights are equally distributed among ensemble mem- formance of the bias-corrected GCM, as presented in
bers, according to Equation (7). Equations (11) and (12), where X s and X o are GCM
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2694 SONG ET AL.

simulation and station observation, respectively, X o and 4 | RESULTS


X o are the means of GCM and observed data, and n is the
total number of data. 4.1 | Evaluation of GCM performances
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn It is generally assumed that higher performances of GCM
1
n ðXs − Xo Þ2 for the historical period indicate a better projection of the
i= 1
NRMS = ð11Þ future. Therefore, this study evaluated the performances
Xo
of 10 GCM pairs of CMIP5 and CMIP6. The precipitation
and average temperature of the CMIP5 and CMIP6
The normalized root mean square error (NRMSE) is an GCMs were downscaled and bias-corrected against the
error magnitude indicator that removes the scale differ- data from the 22 stations used here using IDW and QM,
ence by dividing the RMSE value by the observed mean. respectively. The GCMs' performances in simulating his-
The scale of NRMSE is between 0 and 1, and the closer to torical precipitation and temperature outcomes, based on
0 is the optimal value. NRMSE and Pbias, are presented in Figures 2 and 3,
respectively.
P
n
Figure 2 shows that most CMIP6 GCMs performed
ðXo − Xs Þ
i= 1 better than CMIP5 in simulating historical precipitation,
Pbias = ð12Þ
P
n except for GISS and MIROC in terms of NRMSE. The
Xo
i=1 NRMSE medians for GISS-E2-R and GISS-E2-1-G were
close to the optimal value, but the spread of GISS-E2-1-G
The percent bias (Pbias) is the deviation of GCM esti- was wider than that of GISS-E2-R. Pbias indicates that
mates from the observed data. A positive value Pbias rep- the GCMs of both CMIPs overestimated monthly precipi-
resents overvaluation by GCMs and vice-versa. tation levels. The intervals between the first and third
quantiles of CMIP6 were wider than those of CMIP5,
except for GISS, MIROC, and Nor. Overall, the results
3.7 | REA method showed inconsistency in GCM performance outcomes in
terms of different metrics. However, the improvement in
REA, developed by Giorgi and Mearns (2002), evaluates most CMIP6 GCMs in simulating monthly precipitation
the similarity of GCM outputs in the historical period was evident in two evaluation metrics. The most signifi-
and the differences in GCM projections for the future cant improvement was found in CanESM5.
period. The REA outcomes of CMIP5 and CMIP6 GCMs Figure 3 shows the lowest performances of MPI-ESM-
can be calculated using Equations (13) and (14). LR in replicating historical monthly average tempera-
tures. The medians of NRMSE were similar for all GCMs
h i
1 and close to the optimal value, except for MPI-ESM-LR.
m n ðm× nÞ
wi = ½ðwB,i Þ × ðwD,i Þ  ð13Þ Pbias showed an overestimation of the monthly average
h i temperature by all GCMs. Overall, the improvement in
m n 1 most CMIP6 GCMs in simulating the historical monthly
ε ε ðm× nÞ
= ð14Þ average temperature appeared to be very clear. MPI-
absðBi Þ absðDi Þ ESM1-2-LR showed the highest enhancement in simulat-
ing the monthly average temperature.
Here, wi the weight of the GCM simulations, wB,i is the The Taylor diagram (Taylor, 2001) presented in
error term between the GCM outputs to the station data Figure 4 shows the correspondence between the raw and
for the historical period, and wD,i is the difference term in bias-corrected CMIP5 and CMIP6 GCMs' simulations
the GCM projection for the future period. The initial cal- against the observed data according to the RMSE, Pear-
culation
 of the difference
 term is calculated via son correlation coefficient, and standard deviation. As a
Di = ΔGCM i −ΔGCM ], where ΔGCM is the difference result, the bias-corrected GCMs were found much closer
between the averages of the GCM outputs for the histori- to the observed data. The bias-corrected precipitation and
cal period. Equation (14) is an equation detailing Equa- temperature indicate their usability for drought analysis.
tion (13). Here, ε is the difference between the maximum Figure 5 shows the PDF of the bias-corrected GCM
and minimum values between the moving averages of the simulations and observed data. Overall, the PDF of
observed data over the 10 years. m and n denote the weight GCMs climate factors were similar to observed data.
of error and difference terms. A higher reliability value Especially, the PDF of temperature showed a better
means lower uncertainty (Giorgi and Mearns, 2002). match with the observed PDF. For precipitation, the PDF
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2695

F I G U R E 2 Evaluation results
of the bias-corrected precipitation
of CMIP5 and CMIP6 against
observation at 22 stations [Colour
figure can be viewed at
wileyonlinelibrary.com]

F I G U R E 3 Evaluation results
of the bias-corrected temperature
of CMIP5 and CMIP6 against
observation at 22 stations [Colour
figure can be viewed at
wileyonlinelibrary.com]

of CanESM5 of CMIP5 was different from other GCMs. among the CMIP5 GCMs. For the CMIP6 GCMs, MRI-
These results confirmed that the precipitation perfor- ESM2-0 showed the highest entropy weights for precipi-
mance of CanESM2 had the lowest performance, tation whereas GISS-E2-1-G was the highest entropy
observed using evaluation indices. Overall, the PDFs of weights for temperature.
CMIP6 GCMs were very close to the observed PDF. For The projections of annual precipitation and annual
temperature, GCMs of both CMIPs showed very close to average temperatures for the RCP4.5 & SSP2-4.5 and
the observation. RCP8.5 & SSP5-8.5 scenarios are presented in
Figure 6. The projections consider using GCM MME
based on the entropy weights, averaged for 22 stations.
4.2 | Development of MME Figure 6a shows an increase in the average annual
precipitation for all four emission scenarios of 12.1%
4.2.1 | Projection using entropy- for 2025–2060 and 17.8% for 2065–2,100 compared to
weighted MME the base period (1970–2000). The relatively low-level
emission scenarios show a gentler rising slope,
This study estimated the entropy weights based on the whereas high-level ones show a more steeply rising
GCMs' performance outcomes presented in Section 4.1, slope.
as shown in Table 3. CanESM2 and MPI-ESM-LR had Figure 6b shows the projection of the annual average
the lowest, while MIROC5 and GISS-E2-R had the temperature based on the entropy-weighted MME.
highest entropy weight for precipitation and temperature RCP4.5 and SSP2-4.5 showed corresponding average
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2696 SONG ET AL.

F I G U R E 4 Comparison between the raw and bias-corrected historical estimates by CMIP5 and CMIP6 GCMs against observations at
22 stations using Taylor diagram [Colour figure can be viewed at wileyonlinelibrary.com]

F I G U R E 5 Probability density functions of observed and bias-corrected monthly precipitation and temperature of CMIP5 and CMIP6
GCMs at the 22 stations [Colour figure can be viewed at wileyonlinelibrary.com]
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2697

T A B L E 3 Entropy weights estimated based on the historical performance of GCMs (Yellow bold: minimum value, Orange bold:
maximum value) [Colour table can be viewed at wileyonlinelibrary.com]

GCMs

CMIP Clim. Factor ACCESS CanESM GISS INM GFDL IPSL MIROC MPI MRI NOR SUM
CMIP5 PR 0.099 0.085 0.086 0.088 0.097 0.110 0.117 0.110 0.105 0.105 1.00
TEM 0.087 0.083 0.114 0.097 0.111 0.111 0.110 0.076 0.107 0.105 1.00
CMIP6 PR 0.090 0.079 0.082 0.086 0.117 0.093 0.107 0.113 0.126 0.107 1.00
TEM 0.017 0.087 0.149 0.075 0.101 0.131 0.138 0.082 0.073 0.147 1.00

F I G U R E 6 Projected precipitation
and temperature using MMEs based on
entropy weights at 22 stations. (a).
Precipitation (b). Temperature [Colour
figure can be viewed at
wileyonlinelibrary.com]

increases in the temperature of 1.21 and 1.20 C in the 4.2.2 | Projection using TOPSIS-
near future and 0.71 and 0.85 C in the far future. The weighted MME
lower increase in the temperature in the far future than
in the near future would reduce drought severity in the TOPSIS weights of the GCMs based on their performance
far future compared to the near future. outcomes in terms of statistical metrics were calculated
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2698 SONG ET AL.

as shown in Table 4. For the CMIP5 GCMs, the weight in 2025–2060 and 18.2% for 2065–2100 compared to the base
precipitation was the highest for INM-CM4 and the low- period (1970–2005). The precipitation trend showed a
est for CanESM5. For temperature, ACCESS1-3 had the steady increase over the future period for the low-
highest and MPI-ESM-LR had the lowest weight. For the emission scenarios. The trend was steeper for the high-
CMIP6 GCMs, INM-CM4, and ACCESS-ESM1-5 had emission scenarios than for the low-emission scenarios.
the highest weights for precipitation and temperature, The projections also revealed that annual precipitation in
respectively. The weight of INM-CM4-8 and MIROC6 South Korea may exceed 1,700 mm for the scenarios with
had the lowest for precipitation and temperature, respec- higher emissions.
tively. These results revealed a small difference in GCM The TOPSIS-weighted MME showed an average
weights using entropy and TOPSIS methods. increase in temperature of 1.15 and 1.03  C in the near
TOPSIS assesses and compares the performance of future and 0.68 and 0.73 C in the far future for
GCMs relative to each other. However, TOPSIS is sensi- RCP4.5 and SSP2-4.5, respectively. The TOPSIS-
tive to performance differences between factors. There- weighted MME also showed a lower increase in tem-
fore, it assigns a higher weight to a GCMs that shows a perature in the far future than in the near future. The
relatively high performance than the GCM that shows high-emission scenarios showed an average increase
low performance. This causes a larger difference in in temperature of more than 1.5 C in both the near
TOPSIS weights for the GCMs. Besides, it assigns zero and far futures. Overall, the entropy- and TOPSIS-
weight to the GCM that shows the lowest performance. weighted MME precipitation and temperature projec-
Therefore, the precipitation and temperature weight for tions showed good consistency.
CanESM2 and MPI-ESM-LR, respectively, was zero.
Nevertheless, the two ranks are mostly identical. The
MPI-ESM2-0 outcomes of CMI6 showed objectively high 4.3 | Historical drought analysis
performance capabilities.
Figure 7 shows the projections of the annual precipi- The SPIs and SPEIs for 3, 6, 9, and 12 months were calcu-
tation and temperature based on the TOPSIS-weighted lated using observed data for the base period (1970–2005)
MMEs. The precipitation projected, averaged for four for drought characterization in the study area. Figure 8
emission scenarios, revealed an increase of 10.7% for shows the spatiotemporal distributions of extreme

T A B L E 4 TOPSIS weights
GCMs
estimated based on the historical
CMIP Clim. Factor ACCESS CanESM GISS INM GFDL performance of the GCMs (Yellow bold:
CMIP5 PR Value 0.790 0 0.872 0.995 0.560 minimum value, Orange bold:
maximum value) [Colour table can be
Weight 0.123 0 0.136 0.155 0.087
viewed at wileyonlinelibrary.com]
TEM Value 0.953 0.959 0.749 0.926 0.990
Weight 0.113 0.113 0.088 0.109 0.117
CMIP6 PR Value 0.431 0.146 0.439 0.019 0.668
Weight 0.099 0.034 0.101 0.004 0.153
TEM Value 0.995 0.894 0.429 0.915 0.832
Weight 0.137 0.123 0.059 0.126 0.115
CMIP Clim. Factor IPSL MIROC MPI MRI NOR
CMIP5 PR Value 0.580 0.773 0.472 0.662 0.711
Weight 0.090 0.120 0.074 0.103 0.111
TEM Value 0.995 0.983 0 0.957 0.962
Weight 0.117 0.116 0 0.113 0.113
CMIP6 PR Value 0.192 0.486 0.492 1 0.486
Weight 0.044 0.111 0.113 0.229 0.111
TEM Value 0.715 0.200 0.905 0.930 0.439
Weight 0.099 0.028 0.125 0.128 0.061
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2699

F I G U R E 7 Projected precipitation
and temperature using MMEs based on
the TOPSIS weights at 22 stations. (a).
Precipitation (b). Temperature [Colour
figure can be viewed at
wileyonlinelibrary.com]

droughts for different durations over South Korea. 4.4 | Spatiotemporal projections of
Extreme droughts provide a more generalized view of the droughts
drought characteristics of a region; therefore, the results
of extreme droughts are presented in this study as shown 4.4.1 | SPI projection using the entropy-
in Figure 8. Three- and six-month extreme droughts weighted MME
occur more in the SW and relatively less in the central
region. Nine-month extreme droughts are limited to cer- The maximum SPI droughts projected for RCPs and SSPs
tain regions, such as the NW, NE, SE, and Jeju Island, using the entropy-weighted MME are spatially summarized
while 12-month extreme droughts occur more widely in Table 5. Overall, the highest drought events based on
than the 9-month in the NW and SE. Overall, the esti- SPI were projected from all over the country for most sce-
mated regions in which extreme droughts occur using narios for the near future and in the SE for the far future.
SPEI and SPI were similar, but SPEI showed a higher The maximum droughts in the near future were much
drought severity level than SPI. Therefore, SPEI esti- more severe than in the far future. The projected maxi-
mated more extreme droughts. mum SPI for a 3-month duration for RCP4.5 in the near
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2700 SONG ET AL.

F I G U R E 8 Spatial
distributions of the frequency of
extreme droughts for the historical
period (1970–2005) estimated using
SPI and SPEI [Colour figure can be
viewed at wileyonlinelibrary.com]

future was less than −3.0 in the west, while it was more particular, The projected drought in the west was higher
severe in most of the regions in the near future than in the for the near future than for the far future. The maximum
far future. SPI-9 for RCP4.5 showed a drought with an SPI drought severity levels in the near and far futures were
value of less than −3.0 in most regions. However, the sever- spatially drawn for the four durations and are presented
ity in the far future for SPI-9 was lower than in the near in Figure 9. SPI-3 showed extreme drought (SPI below
future. The differences in the maximum drought intensity −2.0) over the entire country for all scenarios, except in
between the near and the far futures was the largest to the NW. SPI-6 projected higher drought severity in the
RCP4.5 and SSP5-8.5, which showed higher intensity near future for the SSP scenarios. SPI-9 and SPI-12
droughts in the NW in the near future than in the far future. showed extreme droughts in all regions in the near
The maximum SPI-12 value was higher in the west future, where SPI-9 showed moderate droughts and nor-
than in the east in both future periods. The differences in mal conditions in the far future.
drought severity levels between the near and far futures The area showing the maximum percentile SPI
were greater for RCP4.5 and SSP5-8.5 than the other sce- drought level as estimated from the entropy-weighted
narios. The largest difference was for RCP4.5. In MME for all durations and RCP and SSP scenarios was
SONG ET AL.

TABLE 5 Maximum projected drought severities for different scenarios based on SPI and entropy MME

Near future Far future Differences

Duration Region RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5 RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5 RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5
SPI-3 NW −4.70 −3.20 −3.17 −3.05 −3.28 −2.66 −2.81 −1.89 1.43 0.53 0.36 1.16
NE −3.83 −3.02 −3.04 −3.16 −2.55 −2.63 −2.59 −2.02 1.28 0.40 0.45 1.13
SW −3.72 −3.32 −3.29 −3.12 −2.85 −2.62 −2.97 −2.93 0.87 0.70 0.32 0.19
SE −3.12 −2.97 −3.16 −2.92 −2.75 −2.32 −3.11 −3.06 0.37 0.65 0.05 −0.14
SPI-6 NW −4.65 −2.80 −2.94 −2.99 −3.19 −1.71 −1.71 −1.77 1.47 1.09 1.23 1.21
NE −3.84 −2.88 −3.33 −3.18 −2.73 −2.37 −2.37 −2.02 1.11 0.50 0.96 1.17
SW −3.79 −2.87 −3.29 −2.80 −2.66 −2.53 −2.53 −2.38 1.12 0.34 0.76 0.42
SE −3.22 −3.10 −3.34 −2.77 −3.17 −2.37 −2.37 −2.29 0.05 0.73 0.97 0.48
SPI-9 NW −4.67 −2.70 −2.60 −2.76 −1.98 −1.58 −3.08 −1.58 2.69 1.11 −0.48 1.18
NE −3.78 −2.79 −2.93 −2.95 −2.02 −1.95 −2.54 −1.66 1.76 0.83 0.39 1.29
SW −3.60 −2.62 −3.10 −2.73 −2.58 −2.26 −2.66 −2.17 1.02 0.37 0.44 0.57
SE −3.22 −2.34 −3.09 −2.66 −2.97 −2.29 −2.97 −2.10 0.26 0.05 0.12 0.56
SPI-12 NW −4.55 −2.97 −2.74 −2.75 −1.61 −1.46 −2.37 −1.14 2.94 1.51 0.37 1.61
NE −3.93 −2.64 −2.94 −2.87 −1.72 −2.15 −2.24 −1.33 2.21 0.49 0.70 1.54
SW −3.46 −2.41 −2.87 −2.65 −2.32 −2.13 −2.39 −1.70 1.15 0.28 0.48 0.95
SE −3.25 −2.25 −2.83 −2.79 −2.48 −2.05 −2.84 −1.71 0.77 0.21 0.00 1.08
2701

10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2702 SONG ET AL.

calculated, as shown in Figure S1. Areas experiencing summarized in Table 6. Overall, the highest levels of sever-
extreme SPI-3 and SPI-6 drought conditions in the near ity of droughts were projected in the near future for most
future were projected to become areas of severe drought scenarios and durations for which the highest droughts
in the far future. For SPI-9 and SPI-12, reducing drought were projected in the NE. The drought severity levels of all
areas from the near future to the far future was much scenarios and durations were lower in the SE. Overall, the
higher than for SPI-3 and SPI-6. TOPSIS-weight MME showed less severe droughts than
those estimated using the entropy-weight MME. However,
both MMEs projected lower severity levels of longer-
4.4.2 | SPI projection using the TOPSIS- lasting droughts compared to short-duration droughts..
weighted MME The spatial distribution of the maximum drought
severity levels estimated using the TOPSIS-weighted
The maximum SPI droughts projected for RCPs and SSPs MME for all scenarios and durations was presented in
using the TOPSIS-weighted MME are spatially Figure 10. Overall, all scenarios projected extreme

F I G U R E 9 Spatial distribution
of maximum drought severity
levels estimated using SPI and
entropy MME [Colour figure can
be viewed at
wileyonlinelibrary.com]
SONG ET AL.

TABLE 6 Maximum projected drought severities for different scenarios based on SPI and TOPSIS MME

Near future Far future Differences

Duration Region RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5 RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5 RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5
SPI-3 NW −4.27 −3.11 −2.97 −2.80 −2.46 −1.97 −2.69 −1.92 1.81 1.14 0.28 0.88
NE −3.75 −3.11 −3.18 −3.30 −2.13 −2.12 −3.00 −2.12 1.62 0.99 0.18 1.18
SW −3.47 −2.89 −3.11 −3.05 −2.35 −2.39 −2.52 −2.76 1.12 0.5 0.59 0.29
SE −3.37 −2.72 −3.00 −2.88 −2.63 −2.58 −3.11 −2.78 0.74 0.14 −0.11 0.1
SPI-6 NW −4.26 −2.98 −2.90 −2.73 −2.30 −1.58 −2.48 −1.67 1.96 1.4 0.42 1.06
NE −3.84 −3.11 −3.24 −3.30 −2.11 −2.02 −2.55 −1.83 1.73 1.09 0.69 1.47
SW −3.25 −2.42 −3.19 −2.84 −2.44 −2.19 −2.47 −2.16 0.81 0.23 0.72 0.68
SE −3.08 −2.53 −3.22 −2.80 −2.85 −2.52 −2.86 −2.12 0.23 0.01 0.36 0.68
SPI-9 NW −4.28 −2.97 −2.44 −2.72 −2.55 −1.47 −2.80 −1.44 1.73 1.5 −0.36 1.28
NE −3.68 −3.08 −2.87 −2.93 −2.11 −1.71 −2.28 −1.57 1.57 1.37 0.59 1.36
SW −3.10 −2.55 −3.09 −2.69 −2.49 −2.01 −2.29 −1.72 0.61 0.54 0.8 0.97
SE −3.19 −2.38 −2.98 −2.80 −2.71 −2.40 −2.92 −2.06 0.48 −0.02 0.06 0.74
SPI-12 NW −4.05 −3.10 −2.45 −2.63 −1.98 −1.43 −2.23 −1.07 2.07 1.67 0.22 1.56
NE −3.86 −2.98 −2.83 −2.76 −2.13 −1.56 −2.36 −1.24 1.73 1.42 0.47 1.52
SW −3.19 −2.21 −2.71 −2.60 −2.25 −1.85 −2.33 −1.49 0.94 0.36 0.38 1.11
SE −3.44 −2.28 −2.76 −2.85 −2.54 −2.09 −2.82 −1.72 0.9 0.19 −0.06 1.13
2703

10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2704 SONG ET AL.

droughts in the near and far futures for shorter durations experience extreme 3- and 6-month duration of drought in
and moderately dry conditions in the far future for longer the near future will transform into severe drought areas in
durations. SPI-6 projected extreme droughts in the most the far future. In the SPI-9 and SPI-12 cases, areas experienc-
region for all scenarios and severe droughts over the ing severe drought for low-emission scenarios will become a
entire country for low-emission scenarios. Severe moderate drought areas for the high-emission scenarios.
droughts with 9-month durations were projected for the
entire country except for NW for RCP8.5 and the moder-
ate droughts projected in the NW. SPI-12 showed moder- 4.4.3 | SPEI projection using the entropy-
ate droughts over most of the country in all scenarios in weighted MME
the far future, except for low-emission scenarios.
The drought coverage area (%) for different durations and For entropy-weighted MME, the maximum SPEI
scenarios, projected using the TOPSIS-weighted MME, was droughts projected for RCPs and SSPs are spatially sum-
also calculated, as shown in Figure S2. The areas projected to marized in Table 7. Overall, most scenarios projected

F I G U R E 1 0 Spatial
distribution of maximum drought
severity levels estimated using SPI
and TOPSIS MME [Colour figure
can be viewed at
wileyonlinelibrary.com]
SONG ET AL.

TABLE 7 Maximum projected drought severities for different scenarios based on SPEI and entropy MME

Near future Far future Differences

Duration Region RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5 RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5 RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5
SPEI-3 NW −2.73 −2.39 −2.32 −2.32 −2.47 −2.77 −2.46 −2.36 0.25 −0.38 −0.15 −0.04
NE −2.55 −2.36 −2.44 −2.43 −2.67 −2.59 −2.39 −2.43 −0.13 −0.23 0.05 0
SW −2.58 −2.35 −2.63 −2.49 −2.37 −2.69 −2.39 −2.59 0.21 −0.34 0.24 −0.10
SE −2.34 −2.32 −2.36 −2.51 −2.38 −2.56 −2.53 −2.64 −0.04 −0.24 −0.18 −0.13
SPEI-6 NW −3.01 −2.40 −2.22 −2.48 −2.98 −2.42 −2.27 −2.28 0.03 −0.02 −0.05 0.19
NE −2.71 −2.42 −2.51 −2.35 −2.75 −2.41 −2.31 −2.09 −0.04 0.01 0.20 0.26
SW −2.57 −2.15 −2.36 −2.28 −2.55 −2.77 −2.35 −2.47 0.02 −0.62 0.01 −0.18
SE −2.25 −2.24 −2.18 −2.35 −2.71 −2.67 −2.42 −2.56 −0.46 −0.43 −0.24 −0.21
SPEI-9 NW −3.07 −2.48 −2.16 −2.44 −2.19 −2.52 −2.51 −2.28 0.88 −0.04 −0.36 0.16
NE −2.88 −2.40 −2.37 −2.45 −2.13 −2.43 −2.63 −2.08 0.75 −0.03 −0.26 0.37
SW −2.47 −2.26 −2.43 −2.34 −2.32 −2.77 −2.37 −2.47 0.14 −0.51 0.06 −0.12
SE −2.26 −2.11 −2.30 −2.44 −2.37 −2.70 −2.46 −2.31 −0.11 −0.59 −0.16 0.13
SPEI-12 NW −2.99 −2.38 −2.21 −2.32 −1.92 −2.36 −2.22 −2.12 1.06 0.02 −0.01 0.20
NE −2.92 −2.26 −2.38 −2.18 −2.07 −2.35 −2.29 −1.88 0.85 −0.09 0.10 0.30
SW −2.37 −2.14 −2.17 −2.20 −2.21 −2.64 −2.34 −2.44 0.16 −0.50 −0.17 −0.24
SE −2.26 −2.21 −2.06 −2.14 −2.30 −2.60 −2.42 −2.18 −0.04 −0.40 −0.37 −0.03
2705

10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2706 SONG ET AL.

SPEI values of less than −2 in all regions for all durations was severe in the far future for all RCPs and SSPs. For
in the near and far futures. The drought severity in the SPEI-6 and SPEI-9, drought in the near future for low-level
NW for SPEI-6 was the highest compared to the other emission scenarios was extreme for the entire country and
regions. The difference in the SPEI-9 drought severity severe drought of SPEI-9 for the far future in some areas for
level between the near and far futures was small com- the RCP4.5 and SSP5-8.5. The projected SPEI-12 outcome
pared to those in the other durations. SPEI-12 in the near showed severe drought in the NW for RCP4.5 and extreme
future showed extreme drought in the north, whereas and severe drought for SSP2-4.5 in the far future. Extreme
severe drought in the south in the far future. drought conditions based on SPEI for the four scenarios were
Figure 11 presents the spatial distribution of the maxi- estimated to be more widespread over the country. In con-
mum SPEI drought severity level for four emission scenarios trast, severe drought was projected for the RCP4.5 and
over South Korea. An extreme 3-month duration drought SSP5-8.5. Areas experiencing severe drought for RCP8.5 were
was projected over the entire country for all four scenarios larger than expected to experience extreme drought in the
and future periods. The projected SPEI-3 drought in the NW far future.

F I G U R E 1 1 Spatial
distribution of maximum drought
severity levels estimated using
SPEI of entropy MME [Colour
figure can be viewed at
wileyonlinelibrary.com]
SONG ET AL.

TABLE 8 Maximum projected drought severities for different scenarios based on SPEI and TOPSIS MME

Near future Far future Differences

Duration Region RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5 RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5 RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5
SPEI-3 NW −4.27 −3.11 −2.97 −2.80 −2.46 −1.97 −2.69 −1.92 1.81 1.15 0.28 0.88
NE −3.75 −3.11 −3.18 −3.30 −2.13 −2.12 −3.00 −2.12 1.62 0.99 0.18 1.18
SW −3.47 −2.89 −3.11 −3.05 −2.35 −2.39 −2.52 −2.76 1.11 0.50 0.59 0.30
SE −3.32 −2.73 −3.02 −2.89 −2.57 −2.52 −3.08 −2.69 0.75 0.20 −0.06 0.19
SPEI-6 NW −4.26 −2.98 −2.90 −2.73 −2.30 −1.58 −2.48 −1.67 1.96 1.40 0.42 1.06
NE −3.84 −3.11 −3.24 −3.30 −2.11 −2.02 −2.55 −1.83 1.73 1.09 0.69 1.47
SW −3.25 −2.62 −3.19 −2.84 −2.44 −2.19 −2.47 −2.16 0.81 0.43 0.73 0.68
SE −3.08 −2.42 −3.22 −2.81 −2.75 −2.43 −2.80 −2.05 0.33 −0.01 0.43 0.76
SPEI-9 NW −4.28 −2.97 −2.44 −2.72 −2.55 −1.47 −2.80 −1.44 1.73 1.51 −0.36 1.29
NE −3.68 −3.08 −2.87 −2.93 −2.11 −1.71 −2.28 −1.57 1.57 1.37 0.59 1.36
SW −3.10 −2.55 −3.09 −2.75 −2.49 −2.01 −2.29 −1.72 0.61 0.54 0.80 1.03
SE −3.17 −2.37 −2.97 −2.78 −2.63 −2.30 −2.85 −2.00 0.54 0.07 0.12 0.78
SPIE12 NW −4.05 −3.10 −2.45 −2.63 −1.98 −1.43 −2.23 −1.07 2.07 1.68 0.22 1.55
NE −3.86 −2.98 −2.83 −2.76 −2.13 −1.56 −2.36 −1.24 1.74 1.42 0.48 1.52
SW −3.19 −2.23 −2.71 −2.60 −2.25 −1.85 −2.33 −1.49 0.94 0.38 0.38 1.10
SE −3.42 −2.30 −2.75 −2.80 −2.44 −2.05 −2.76 −1.68 0.98 0.25 −0.01 1.12
2707

10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2708 SONG ET AL.

F I G U R E 1 2 Spatial
distribution of maximum drought
severity levels estimated using
SPEI and TOPSIS MME [Colour
figure can be viewed at
wileyonlinelibrary.com]

The drought coverage area (%) for the different dura- and SSPs are spatially summarized in Table 8. Overall,
tions and scenarios estimated using SPEI was calculated, as the highest level of drought severity of SPEI was projec-
shown in Figure S3. The results showed an expansion of ted to be less than −2.0 in all durations, with more robust
the areas projected to undergo severe SPEI-3 and SPEI-6 drought estimated in the near than far futures. The
drought in the far future for high-emission scenarios. For drought severity for SPEI-3 was the highest in SW,
SPEI-9 and SPEI-12, certain areas experiencing extreme whereas, in the SE, it was the lowest. Both emission sce-
drought would meet severe drought in the far future. narios clearly have high drought severity in the near
future. For SPEI-6, the maximum drought severity in the
NW for the near future would be less than −4 in RCP4.5.
4.4.4 | SPEI projection using the TOPSIS- However, the drought severity in the far future would be
weighted MME less than −3 in the same region and scenario for the near
future. In the SPEI-9, the drought severity level in the far
The maximum drought outcomes based on the TOPSIS future in most scenarios was lower than in the near
MME of the projected SPEI for each station in the RCPs future in all regions except for NW in the SSP2-4.5, and
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2709

the drought severity of SSP2-4.5 was higher in the near compared to the drought conditions from the entropy
future than in the far. For SPEI-12, the drought severity weights, and more severe droughts were estimated.
levels in all regions for most scenarios were clearly higher The maximum percent of the drought area for each
in the near future than in the far future, though the SE of duration as projected by SPEI based on the TOPSIS MME
SSP2-4.5 showed the opposite outcome. in four future scenarios was calculated, as shown in
Figure 12 presents the severity level of the maximum Figure S4. Overall, the drought severity level of this per-
drought based on the TOPSIS MME from SPEI for the centage of area for all durations does not show a severity
four emission scenarios over South Korea. Overall, the type below severe dry conditions in the near and far
drought severity for all scenarios was very extremely dry futures. Most area is distributed with drought severity
and extremely dry in the near future, and the projections levels that exceed severely dry conditions. Furthermore,
did not capture moderately dry conditions for all dura- the drought severity level of TOPSIS weight is higher
tions. Despite the fact that the far future was estimated than that of the entropy weight, and the area showing
severely dry in some regions, the primary weather condi- low severity is small.
tion remains extremely dry. The severely dry conditions
of the four scenarios for SPEI-3 are estimated to be com-
mon in the most region, but not in the near future for 4.5 | Uncertainty in future drought
RCP4.5. In the far future, high-level emission scenarios projection
were expanded to extremely dry all over the country,
whereas the drought patterns in the low-level scenarios 4.5.1 | Uncertainty in SPI
were erratic, making it challenging to identify the
droughts pattern. For SPEI-6, the extremely dry condi- In this study, SPIs from 10 pairs of CMIP5 and CMIP6
tions of the near future were in the northern and east GCMs were used to estimate the uncertainty of future
regions, whereas the south region of most scenarios drought projections in the near and far future. Uncertainties
showed areas of robust drought with scores of less than in projected drought outcomes were quantified using the
−2.5 in the far future. The near future in the RCP scenar- REA method for all durations and scenarios at the 22 sta-
ios for SPEI-9 estimates severely dry conditions in some tions. Table 9 presents the average REA outcomes in SPI for
regions. On the other hand, severely dry conditions in the four scenarios. The SPI-3 values for RCP4.5 and
high-level emission scenarios were predicted for the far SSP2-4.5 were lower in the near future than in the far
future. In the SPEI-12, all scenarios were estimated future, whereas the other durations showed the opposite
higher than extremely dry in the near future, whereas the results. The reliability in the temporal variation for RCP8.5
far future was projected severely dry conditions in all sce- was different from those for SSP5-8.5. The reliability for
narios except for SSP2-4.5. Droughts from the TOPSIS RCP8.5 was higher in the near future, whereas for SSP5-8.5,
weights were very dry and extremely dry conditions it was higher in the far future than in the near for all dura-
tions. The difference between the near and far futures was

TABLE 9 Temporal average REA of SPI estimated for RCPs and SSPs compared to the base period of 1970–2000

Period Indices RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5


Near SPI-3 0.065 0.063 0.065 0.062
Far 0.061 0.059 0.059 0.082
Difference −0.004 −0.004 −0.006 0.020
Near SPI-6 0.058 0.067 0.076 0.067
Far 0.059 0.064 0.105 0.098
Difference 0.001 −0.003 0.029 0.031
Near SPI-9 0.054 0.064 0.073 0.071
Far 0.071 0.06 0.108 0.078
Difference 0.017 −0.004 0.035 0.007
Near SPI-12 0.056 0.119 0.068 0.06
Far 0.067 0.057 0.077 0.099
Difference 0.011 −0.062 0.009 0.039
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2710 SONG ET AL.

F I G U R E 1 3 Uncertainty in
SPI projected droughts over
South Korea for the SSP and RCP
scenarios [Colour figure can be
viewed at wileyonlinelibrary.com]

the largest for SSP5-8.5 and the lowest for RCP4.5. Overall, for most of the regions, except for SSP5-8.5 in the far
the reliabilities in the near and far futures for the RCP sce- future, which showed reliability higher than 0.30 in the
narios were different at four durations. On the other hand, south. For SPI-6, the reliability in the south for the SSP
the reliability for the SSP scenarios was higher in the far scenarios exceeded 0.3. For SPI-9, the reliability outcomes
future than in the near for most drought durations, mean- for RCP4.5 and SSP2-4.5 were higher than 0.3. The reli-
ing that the projected droughts in the far future for SSP sce- ability for the low-emission scenarios was higher than
narios have low uncertainty levels compared to those of that of the high-emission scenarios. For SPI-12, the reli-
RCP scenarios. ability was higher for the low-emission, and the highest
The spatial distribution of the uncertainties in SPI- was in the SE. Overall, the reliability of drought for
projected droughts is presented in Figure 13. The reliabil- higher durations was higher and greater for all four sce-
ity levels in SPI-3 for the four scenarios were below 0.15 narios. RCP8.5 showed the highest and the widest range
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2711

TABLE 10 Temporal average REA of SPEI estimated for RCPs and SSPs compared to the base period of 1970–2000

Period Indices RCP4.5 RCP8.5 SSP2-4.5 SSP5-8.5


Near SPEI-3 0.058 0.057 0.067 0.057
Far 0.073 0.056 0.065 0.064
Difference 0.015 −0.001 −0.002 0.007
Near SPEI-6 0.052 0.054 0.066 0.077
Far 0.074 0.05 0.061 0.064
Difference 0.022 −0.004 −0.005 −0.013
Near SPEI-9 0.051 0.041 0.059 0.074
Far 0.053 0.044 0.055 0.098
Difference 0.002 0.003 −0.004 0.024
Near SPEI-12 0.059 0.072 0.064 0.078
Far 0.063 0.095 0.059 0.067
Median 0.002 0.003 −0.004 0.024

F I G U R E 1 4 Uncertainty in SPEI
projected droughts over South Korea for the
SSP and RCP scenarios [Colour figure can
be viewed at wileyonlinelibrary.com]
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2712 SONG ET AL.

of reliability for SPI-12. The reliability levels for the SSP (3, 6, 9, and 12 months), estimated using the SPI and
scenarios exceeded 0.15 for all durations, and the highest SPEI, for two future periods (near future: 2025–2060 and
values were for SPI-9 and SPI-12. These results indicate far future: 2065–2100) and four emission scenarios were
that the projected droughts during SPI for SSPs have evaluated. The performances of CMIP5 and CMIP6
lower uncertainty levels than those for RCPs. GCMs relative to observed data for the historical period
were compared using two evaluation metrics and PDF.
The weights of the GCMs estimated using the entropy
4.5.2 | Uncertainty in SPEI and TOPSIS were used to generate the MME outcome.
Overall, the performances of bias-corrected CMIP6
The averages of REA in SPEI were calculated for the four GCMs were improved compared to CMIP5 in simulations
scenarios and are presented in Table 10. The low emis- of the historical temperature and precipitation levels. The
sion scenarios showed consistent results for REA in the weights for precipitation and temperature were similar
near and far futures. The average REA for all durations for CMIP5 and CMIP6 GCMs, respectively, except for
of droughts for RCP4.5 was higher in the far future than CanESM5 & INM and MPI & MIROC. It means that
in the near, whereas it was the opposite for SSP2-4.5. The except for the low-performing GCMs, such as CanESM5,
SPEI-3 and SPEI-6 outcomes for RCP8.5 showed higher MPI, GISS, and MIROC, others are less sensitive to each
REA values in the near future than in the far. The oppo- other, which means that the configured GCMs are robust.
site was derived in SPEI-9 and SPEI-12. The average The GCM subsets can be selected based on evaluation
REAs for SPEI-3 and SPEI-9 for SSP5-8.5 were higher in results obtained in this study. The MME projections
the far future than in the near. However, the opposite showed a gradual increase in precipitation and tempera-
was for SPEI-6 and SPEI-12. The difference between the ture in the near and far futures. The increase was higher
near and far futures was the highest for SSP5-8.5 and the for high-emission scenarios than for low-emission scenar-
lowest for SSP2-4.5. Therefore, the temporal variation ios. Future drought estimated using SPI and SPEI with
was the highest in SSP5-8.5 and the lowest in SSP2-4.5. TOPSIS weighted GCMs was more robust than that with
This indicates that the reliability levels of the projected the entropy weighted GCMs. In general, the drought
droughts for SSPs are higher than the reliability in RCPs. scale and severity estimated using SPI were lower than
Figure 14 presents the interpolated average REA of that obtained using SPEI. SPI only considers precipita-
the SPEI outcomes. The reliability levels for SPEI-3 of tion, while SPEI considers both precipitation and evapo-
RCP8.5, SSP2-4.5 and SSP5-8.5 were less than 0.13 transpiration for drought estimation; therefore, it is
throughout the country. On the other hand, the reliabil- important to use various drought indices to analyse the
ity in RCP4.5 exceeded 0.2 in the central region. The reli- drought scale and severity. The estimated drought sever-
ability of SPEI-6 was greater compared to that of SPEI-3. ity was lower for the far future than for the near future.
Most scenarios have reliability levels of more than 0.16 in This may be due to more precipitation in the far future
the near and far futures, except for RCP8.5. SSP5-8.5 has than in the near. Regional changes in drought are gener-
the highest reliability in the near future. For SPEI-9, ally active in NW, and SPI detects more regional changes
SSP5-8.5 estimated overwhelmingly high reliability in the than SPEI. The drought reliability of SPEI was higher
near and far futures and high reliability in the than that of SPI, which means that it is more reliable if
NE. RCP8.5 and SSP5-8.5 showed reliability higher than used to consider evapotranspiration for a drought analy-
0.20 in SPEI-12. Overall, the reliability of SSP5-8.5 was sis. SPI showed higher confidence levels in the southwest
much higher than those for the other scenarios for most and central regions, and SPEI showed higher reliability
drought durations. Except for SSP5-8.5, lower drought in the central and eastern regions. The 9- and 12-month
durations were linked to greater reliability for low- drought outcomes showed higher reliability in the far
emission scenarios. On the other hand, drought projec- future than in the near, whereas the 3- and 6-month out-
tions for high-emission scenarios are more reliable for comes showed higher reliability in the near future than
higher durations. in the far. SPEI of the RCP scenarios is less reliable than
in the SSP scenario in most durations except for SPEI-12
for the far future. Likewise, the reliability of SPI in the
5 | C ON C L U S I ON SSP scenario is higher than that of the RCP scenario.
Therefore, the SPI and SPEI projected droughts for the
This study estimated the differences in projected SSP scenarios had lower uncertainty levels than those for
droughts in South Korea from 10 pairs of CMIP5 and the RCP scenarios. This study can help clarify future
CMIP6 GCMs for SSP and RCP scenarios. For this pur- drought conditions on the Korean peninsula using SSPs
pose, projected drought severity levels for four durations and RCPs.
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2713

A C K N O WL E D G E M E N T Agriculture Organization of the United Nations (FAO) http://


This research has been performed as Project No Open www.fao.org/3/I8656EN/i8656en.pdf.
Innovation R&D (D-W-004) and supported by K-water. Cook, B.I., Markin, J.S., Williams, A.P., Smerdon, J.E. and
Anchukaitis, K.J. (2020) Twenty-first century drought projec-
tions in the CMIP6 forcing scenarios. Earth's Future, 8(6),
A U T H O R C ON T R I B U T I O NS e2019EF001461. https://doi.org/10.1029/2019EF001461.
Young Hoon Song: Data curation; formal analysis; Dai, A. (2010) Drought under global warming: a review. Wiley Inter-
investigation; visualization; writing – original draft. disciplinary Reviews Climate Change, 2, 45–65. https://doi.org/
SHAMSUDDIN SHAHID: Methodology; validation; 10.1002/wcc.81.
writing – original draft; writing – review and editing. Enayati, M., Bozorg-Haddad, O., Bazrafshan, J., Hejabi, S. and
Eun-Sung Chung: Conceptualization; funding acquisi- Chu, X. (2021) Bias correction capabilities of quantile mapping
methods for rainfall and temperature variables. Journal of
tion; methodology; supervision; writing – original draft;
Water and Climate Change, 12(2), 401–419. https://doi.org/10.
writing – review and editing.
2166/wcc.2020.261.
Eyring, V., Bony, S., Meehl, G.A., Senior, C.A., Stevens, B.,
ORCID Stouffer, R.J. and Taylor, K.E. (2016) Overview of the Coupled
Shamsuddin Shahid https://orcid.org/0000-0001-9621- Model Intercomparison Project Phase 6 (CMIP6) experimental
6452 design and organization. Geoscientific Model Development, 9(5),
Eun-Sung Chung https://orcid.org/0000-0002-4329- 1937–1958. https://doi.org/10.5194/gmd-9-1937-2016.
1800 Fan, X., Miao, C., Duan, Q., Shen, C. and Wu, Y. (2020) The perfor-
mance of CMIP6 versus CMIP5 in simulating temperature
extremes over the global land surface. Journal of Geophysical
R EF E RE N C E S Research: Atmospheres, 125(18), e2020JD033031. https://doi.
Ahmed, K., Sachindra, D.A., Shahid, S., Demirel, M.C. and org/10.1029/2020JD033031.
Chung, E.S. (2019) Selection of multi-model ensemble of gen- Ghosh, S. and Mujumdar, P.P. (2007) Nonparametric methods for
eral circulation models for the simulation of precipitation and modeling GCM and scenario uncertainty in drought assess-
maximum and minimum temperature based on spatial assess- ment. Water Resources Research, 43, W07405. https://doi.org/
ment metrics. Hydrology and Earth System Sciences, 23(11), 10.1029/2006WR005351.
4803–4824. https://doi.org/10.5194/hess-23-4803-2019. Gibson, P.B., Waliser, D.E., Lee, H., Tian, B. and Massoud, E.
Alamgir, M., Mohsenipour, M., Homsi, R., Wang, X., Shahid, S., (2019) Climate model evaluation in the presence of observa-
Shiru, M.S., Alias, N.E. and Yuzir, A. (2019) Parametric assess- tional uncertainty: precipitation indices over the contiguous
ment of seasonal drought risk to crop production in Bangladesh. United States. Journal of Hydrometeorology, 20, 1339–1357.
Sustainability, 11(5), 1442. https://doi.org/10.3390/su11051442. https://doi.org/10.1175/JHM-D-18-0230.1.
Almazroui, M., Saeed, F., Saeed, S., Islam, M.N., Ismail, M., Giorgi, F. and Mearns, L.O. (2002) Calculation of average, uncer-
Klutse, N.A.B., and Siddiqui, M.H. (2020) Projected change in tainty range and reliability of regional climate changes from
temperature and precipitation over Africa from CMIP6. Envi- AOGCM simulations via the "Reliability Ensemble Averaging"
ronmental Earth Sciences, 4, 155–175. https://doi.org/10.1007/ (REA) method. Journal of Climate, 5, 1141–1158. https://doi.
s41748-020-00161-x. org/10.1175/1520-0442(2002)015<1141:COAURA>2.0.CO;2.
Asdak, C., Supian, S. and Subiyanto. (2018) Watershed manage- Gudmundsson, L., Bremnes, J.B., Haugen, J.E. and Engen-
ment strategies for flood mitigation: a case study of Jakarta's Skaugen, T. (2012) Technical Note: Downscaling RCM precipi-
flooding. Weather and Climate Extremes, 21, 117–122. https:// tation to the station scale using statistical transformations – a
doi.org/10.1016/j.wace.2018.08.002. comparison of methods. Hydrology and Earth System Sciences,
Ayugi, B., Tan, G., Rouyun, N., Zeyao, D., Ojara, M., Mumo, L., 16, 3383–3390. https://doi.org/10.5194/hess-16-3383-2012.
Babaousmail, H. and Ongoma, V. (2020) Evaluation of meteo- Hernandez, E.A. and Uddameri, V. (2014) Standardized precipita-
rological drought and flood scenarios over Kenya, East Africa. tion evaporation index (SPEI)-based drought assessment in
Atmosphere, 11(3), 307. https://doi.org/10.3390/atmos11030307. semi-arid south Texas. Environmental Earth Sciences, 71, 2491–
Berg, A., Sheffield, J. and Milly, P.C.D. (2017) Divergent surface 2501. https://doi.org/10.1007/s12665-013-2897-7.
and total soil moisture projections under global warming. Geo- Hwang, C.L. and Yoon, K. (1981) Multiple Attribute Decision Mak-
physical Research Letters, 44(1), 236–244. https://doi.org/10. ing: Methods and Applications. New York: Springer-Verlag.
1002/2016GL071921. Karavitis, C.A., Alexandris, S., Tsesmelis, D.E. and
Cannon, A.J., Sobie, S.R. and Murdock, T.Q. (2015) Bias correction Athanasopoulos, G. (2011) Application of the Standardized Pre-
of GCM precipitation by Quantile mapping: how well do cipitation Index (SPI) in Greece. Water, 3, 787–805. https://doi.
methods preserve changes in quantiles and extremes? Journal org/10.3390/w3030787.
of Climate, 28, 6938–6959. https://doi.org/10.1175/JCLI-D-14- Khan, N., Sachindra, D.A., Shahid, S., Ahmed, K., Shiru, M.S. and
00754.1. Nawaz, N. (2020) Prediction of droughts over Pakistan using
Chen, S.J. and Hwang, C.L. (1992) Fuzzy Multiple Attribute Decision machine learning algorithms. Advances in Water Resources,
Making: Methods and Applications. Berlin: Springer. 139, 103562. https://doi.org/10.1016/j.advwatres.2020.103562.
Conforti, P., Ahmed, S. and Markova, G. (2018) Impact of Disasters Keyantash, J. and Dracup, J.A. (2002) The quantification of
and Crises on Agriculture and Food Security. Rome: Food and drought: an evaluation of drought indices. Bulletin of the
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2714 SONG ET AL.

American Meteorological Society, 83, 1167–1180. https://doi.org/ Conference on Applied Climatology. Anaheim, CA: American
10.1175/1520-0477-83.8.1167. Meteorological Society, 174–184.
Kim, J.H., Sung, J.H., Chung, E.S., Kim, S.U., Son, M.W. and McKee, T.B., Doesken, N.J. and Kleist, J. (1995) Drought Monitoring
Shiru, M.S. (2021) Comparison of projection in meteorological with Multiple Time Scales. Boston, MA: American Meteorologi-
and hydrological droughts in the Cheongmicheon Watershed cal Society, pp. 233–236.
for RCP4.5 and SSP2-4.5. Sustainability, 13(4), 2066. https://doi. Mo, K.C. (2008) Model-based drought indices over the
org/10.3390/su13042066. United States. Journal of Hydrometeorology, 9, 1212–1230.
Kim, Y.H., Min, S.K., Zhang, X., Sillmann, J. and Sandstad, M. https://doi.org/10.1175/2008JHM1002.1.
(2020) Evaluation of the CMIP6 multi-model ensemble for cli- Oh, S.B., Byun, H.R. and Kim, D.W. (2014) Spatiotemporal charac-
mate extreme indices. Weather and Climate Extremes, 29, teristics of regional drought occurrence in East Asia. Theoretical
100269. https://doi.org/10.1016/j.wace.2020.100269. and Applied Climatology, 117(1), 89–101. https://doi.org/10.
Kim, W., Iizumi, T. and Nishimori, M. (2019) Global patterns of 1007/s00704-013-0980-3.
crop production losses associated with droughts from 1983 to O'Neill, B.C., Tebaldi, C., Van Vuuren, D.P., Eyring, V.,
2009. Journal of Applied Meteorology and Climatology., 58(6), Friedlingstein, P., Hurtt, G., Knutti, R., Kriegler, E.,
1233–1244. https://doi.org/10.1175/JAMC-D-18-0174.1. Lamarque, J.F. and Lowe, J. (2016) The scenario model
Knutti, R., Furrer, R., Tebaldi, C., Cermak, J. and Meehl, G.A. intercomparison project (ScenarioMIP) for CMIP6. Geoscientific
(2010) Challenges in combining projections from multiple cli- Model Development, 9, 3461–3482. https://doi.org/10.5194/gmd-
mate models. Journal of Climate, 23(10), 2739–2758. https:// 9-3461-2016.
doi.org/10.1175/2009JCLI3361.1. O'Neill, B.C., Kriegler, E., Ebi, K.L., Kemp-Benedict, E., Riahi, K.,
Kouhestani, S., Eslamian, S.S., Abedi-Koupai, J. and Rothman, D.S., van Ruijven, B.J., van Vuuren, D.P.,
Besalatpour, A.A. (2016) Projection of climate change impacts Birkmann, J., Kok, K., Levy, M. and Solecki, W. (2017) The
on precipitation using soft-computing techniques: a case study roads ahead: narratives for shared socioeconomic pathways
in Zayandeh-rud Basin, Iran. Global and Planetary Change, describing world futures in the 21st century. Global Environ-
144, 158–170. https://doi.org/10.1016/j.gloplacha.2016.07.013. mental Change, 42, 169–180. https://doi.org/10.1016/j.
Kwon, H.H., Lall, U., and Kim, S.J. (2016) The unusual 2013–2015 gloenvcha.2015.
drought in South Korea in the context of a multicentury precip- Park, C.K., Ho, C.H., Park, R.J., Woo, J.H., Lim, C., Park, D.S.R.,
itation record: Inferences from a nonstationary, multivariate, Park, H., Kim, M.J., Kim, Y., Hong, S.C. and Kim, J. (2020) Pro-
Bayesian copula model. Geophysical research letters, 43, 8534– jections of future drought intensity associated with various
8544. https://doi.org/10.1016/j.wace.2020.100269. local greenhouse gas emission scenarios in East Asia. Terrestrial
Labudova, L., Labuda, M. and Takač, J. (2017) Comparison of SPI Atmospheric and Oceanic Sciences, 31, 9–19. https://doi.org/10.
and SPEI applicability for drought impact assessment on crop 3319/TAO.2019.06.16.01.
production in the Danubian Lowland and the East Slovakian Pomerol, J.C. and Romero, S.B. (2000) Multicriterion Decision
Lowland. Theoretical and Applied Climatology, 128, 491–506. in Management: Principles and Practice. Netherlands:
https://doi.org/10.1007/s00704-016-1870-2. Kluwer.
Lemordant, L., Gentine, P., Swann, A.S., Cook, B.I. and Scheff, J. Pokhrel, Y., Felfelani, F., Satoh, Y., Boulange, J., Burek, P.,
(2018) Critical impact of vegetation physiology on the continen- Gädeke, A., Gerten, D., Gosling, S.N., Grillakis, M.,
tal hydrologic cycle in response to increasing CO2. Proceedings Gudmundsson, L., Hanasaki, N., Kim, H.J., Koutroulis, A.,
of the National Academy of Sciences of the United States of Liu, J., Papadimitriou, L., Schewe, J., Schmied, H.M.,
America, 115(16), 4093–4098. https://doi.org/10.1073/pnas. Stacke, T., Telteu, C.E., Thiery, W., Veldkamp, T., Zhao, F.
1720712115. and Wada, Y. (2021) Global terrestrial water storage and
Liu, C., Yang, C., Yang, Q. and Wang, J. (2021) Spatiotemporal drought severity under climate change. Nature Climate
drought analysis by the standardized precipitation index (SPI) Change, 11(3), 226–233. https://doi.org/10.1038/s41558-020-
and standardized precipitation evapotranspiration index (SPEI) 00972-w.
in Sichuan Province, China. Scientific Reports, 11, 1280. https:// Raju, K.S. and Kumar, N.D. (2010) Multicriterion Analysis in Engi-
doi.org/10.1038/s41598-020-80527-3. neering and Management. New Delhi: Prentice Hall of India.
Liu, Y. and Chen, J. (2021) Future global socioeconomic risk to Riahi, K., Kriegler, E., Johnson, N., Bertram, C., den Elzen, M.,
droughts based on estimates of hazard, exposure, and vulnera- Eom, J., Schaeffer, M., Edmonds, J., Isaac, M., Krey, V.,
bility in a changing climate. Science of the Total Environment, Longden, T., Luderer, G., Méjean, A., McCollum, D.L.,
751, 142159. https://doi.org/10.1038/s41598-020-80527-3. Mima, S., Turton, H., van Vuuren, D.P., Wada, K., Bosetti, V.,
Longley, P.A., Goodchild, M.F., Maguire, D.J. and Rhind, D.W. Capros, P., Criqui, P., Hamdi-Cherif, M., Kainuma, M. and
(2005) Geographic Information Systems and Science. New York: Edenhofer, O. (2015) Locked into Copenhagen pledges—
Wiley. implications of short-term emission targets for the cost and fea-
Mankin, J.S., Viviroli, D., Mekonnen, M.M., Hoekstra, A.Y., sibility of long-term climate goals. Technological Forecasting
Horton, R.M., Smerdon, J.E. and Diffenbaugh, N.S. (2017) and Social Change, 90, 8–23. https://doi.org/10.1016/j.techfore.
Influence of internal variability on population exposure to 2013.09.016.
hydroclimatic changes. Environmental Research Letters, 12(4), Ringard, J., Seyler, F. and Linguet, L. (2017) A quantile mapping
044007. https://doi.org/10.1088/1748-9326/aa5efc. bias correction method based on hydroclimatic classification of
McKee, T.B., Doesken, N.J., Kleist. J. (1993). The relationship of the Guiana shield. Sensors, 17, 1413. https://doi.org/10.3390/
drought frequency and duration to time scales. Eighth s17061413.
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SONG ET AL. 2715

Salman, S.A., Shahid, S., Afan, H.A., Shiru, M.S., Al-Ansari, N. and over contiguous US regions. Weather and Climate Extremes, 29,
Yaseen, Z.M. (2020) Changes in climatic water availability and 100268. https://doi.org/10.1016/j.wace.2020.100268.
crop water demand for Iraq region. Sustainability, 12(8), 3437. Stevens, B. and Bony, S. (2013) What are climate models missing? Sci-
https://doi.org/10.3390/su12083437. ence, 340, 1053–1054. https://doi.org/10.1126/science.1237554.
Shiau, J.T. (2020) Effects of gamma-distribution variations on SPI- Su, B., Huang, J., Mondal, S.K., Zhai, J., Wang, Y., Wen, S.,
based stationary and nonstationary drought analyses. Water Gao, M., Lv, Y., Jiang, S., Jiang, T. and Li, A. (2021) Insight
Resources Management, 34, 2081–2095. https://doi.org/10.1007/ from CMIP6 SSP-RCP scenarios for future drought characteris-
s11269-020-02548-x. tics in China. Atmospheric Research, 250, 105375. https://doi.
Shannon, C.E. (1948) A mathematical theory of communication. org/10.1016/j.atmosres.2020.105375.
Bell Labs Technical Journal, 27(3), 379–423. https://doi.org/10. Tobler, W. (1970) A computer movie simulating urban growth in
1002/j.1538-7305.1948.tb01338.x. the Detroit region. Economic Geograpy, 46, 234–240.
Sheffield, J. and Wood, E.F. (2007) Projected changes in drought Taylor, K.E. (2001) Summarizing multiple aspects of model per-
occurrence under future global warming from multi-model, formance in a single diagram. Journal of Geophysical
multi-scenario, IPCC AR4 simulations. Climate Dynamics, 31, Research Atmospheres, 106, 7183–7192. https://doi.org/10.
79–105. https://doi.org/10.1007/s00382-007-0340-z. 1029/2000JD900719.
Shiru, M.S., Shahid, S., Alias, N. and Chung, E.S. (2018) Trend Taylor, K.E., Stouffer, R.J. and Meehl, G.A. (2012) An overview of
analysis of droughts during crop growing seasons of CMIP5 and the experiment design. Bulletin of the American
Nigeria. Sustainability, 10(3), 871. https://doi.org/10.3390/ Meteorological Society., 93, 485–498.
su10030871. Teutschbein, C. and Seibert, J. (2012) Bias correction of regional cli-
Shiru, M.S., Shahid, S., Chung, E.S., Alias, N. and Scherer, L. (2019) mate model simulations for hydrological climate-change
A MCDM-based framework for selection of general circulation impact studies: review and evaluation of different methods.
models and projection of spatio-temporal rainfall changes: a Journal of Hydrology, 456, 12–29. https://doi.org/10.1016/j.
case study of Nigeria. Atmospheric Research, 225, 1–16. https:// jhydrol.2012.05.052.
doi.org/10.1016/j.atmosres.2019.03.033. Thornthwaite, C. (1948) An approach toward a rational classifica-
Shiru, M.S., Shahid, S., Dewan, A., Chung, E.S., Alias, N., tion of climate. Geographical Review, 38, 55–94. https://doi.org/
Ahmed, K. and Hassan, Q.K. (2020) Projection of meteorologi- 10.2307/210739.
cal droughts in Nigeria during growing seasons under climate Trenberth, K.E. Dai, A., G. Schrier, V.D. Jones, P.D., Barichivich, J.,
change scenarios. Scientific Reports, 10(1), 1–18. https://doi.org/ Briffa, K.R., Sheffield, J. (2014) Global warming and changes in
10.1038/s41598-020-67146-8. drought. Nature Climate Change, 4, 17–22. https://doi.org/10.
Sillmann, J., Kharin, V.V., Zhang, X., Zwiers, F.W. and 1038/nclimate2067
Bronaugh, D. (2013) Climate extremes indices in the CMIP5 Ukkola, A.M., Kauwe, M.G., Roderick, M.L., Abramowitz, G. and
multimodel ensemble: part 1. Model evaluation in the present Pitman, A.J. (2020) Robust future changes in meteorological
climate. Journal of Geophysical Research: Atmospheres, 118, drought in CMIP6 projections despite uncertainty in precipita-
1716–1733. https://doi.org/10.1002/jgrd.50203. tion. Geophysical Research Letters, 47(11), 1–14. https://doi.org/
Spinoni, J., Barbosa, P., Bucchignani, E., Cassano, J., Cavazos, T., 10.1002/essoar.10502465.1.
Christensen, J.H., Christensen, O.B., Coppola, E., Evans, J., Vicente-Serrano, S.M., Beguería, S. and L opez-Moreno, J.I. (2010) A
Geyer, B., Giorgi, F., Hadjinicolaou, P., Jacob, D., Katzfey, J., multi-scalar drought index sensitive to global warming: the stan-
Koenigk, T., Laprise, R., Lennard, C.J., Kurnaz, M.L., Li, D., dardized precipitation evapotranspiration index. Journal of Cli-
Llopart, M., McCormick, N., Naumann, G., Nikulin, G., mate, 23(7), 1696–1718. https://doi.org/10.1175/2009JCLI2909.1.
Ozturk, T., Panitz, H.J., Rocha, R.P., Rockel, B., Solman, S.A., Wang, C., Zhang, L., Lee, S.K., Wu, L. and Mechoso, C.R. (2014) A
Syktus, J., Tangang, F., Teichmann, C., Vautard, R., Vogt, J.V., global perspective on CMIP5 climate model biases. Nature Cli-
Winger, K., Zittis, G. and Dosio, A. (2020) Future global meteo- mate Change, 4, 201–205. https://doi.org/10.1038/nclimate2118.
rological drought hot spots: a study based on CORDEX data. Wang, H., Pan, Y. and Chen, Y. (2017) Comparison of three
Journal of Climate, 33(9), 3635–3661. https://doi.org/10.1175/ drought indices and their evolutionary characteristics in the
JCLI-D-19-0084.1. arid region of northwestern China. Atmospheric Science Letters,
Song, Y.H., Chung, E.S. and Shiru, M.S. (2020) GCMs using multi- 18(3), 132–139. https://doi.org/10.1002/asl.735.
ple bias correction methods under different RCPs. Sustainabil- Wang, T., Tu, X., Singh, V.P., Chen, X. and Lin, K. (2021) Global
ity, 12, 7508. https://doi.org/10.3390/su12187508. data assessment and analysis of drought characteristics based
Song, Y.H., Nashwan, M.S., Chung, E.S. and Shahid, S. (2021a) on CMIP6. Journal of Hydrology, 596, 126091. https://doi.org/
Advances in CMIP6 INM-CM5 over CMIP5 INM-CM4 for pre- 10.1016/j.jhydrol.2021.126091.
cipitation simulation in South Korea. Atmospheric Research, Wang, L. and Chen, W. (2014) A CMIP5 multimodel projection of
247, 105261. https://doi.org/10.1016/j.atmosres.2020.105261. future temperature, precipitation, and climatological drought
Song, Y.H., Chung, E.S. and Shahid, S. (2021b) Spatiotemporal dif- in China. International Journal of Climatology, 34, 2059–2078.
ferences and uncertainties in projections of precipitation and https://doi.org/10.1002/joc.3822.
temperature in South Korea from CMIP6 and CMIP5 GCMs. Wang, M., Zhang, D.Q., Su, J., Dong, J.W. and Tan, S.K. (2018)
International Journal of Climatology, 1–21. https://doi.org/10. Assessing hydrological effects and performance of low impact
1002/joc.7159. development practices based on future scenarios modeling.
Srivastava, A., Grotjahn, R. and Ullrich, P.A. (2020) Evaluation of Journal of Cleaner Production, 179, 12–23. https://doi.org/10.
historical CMIP6 model simulations of extreme precipitation 1016/j.jclepro.2018.01.
10970088, 2022, 5, Downloaded from https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.7386 by UNIVERSIDADE FEDERAL DO PARA CAMPUS UNIVERSITARIO DO GUAMA, Wiley Online Library on [23/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2716 SONG ET AL.

Wehner, M.F., Reed, K.A., Li, F., Prabhat, Bacmeister, J., Chen, C. Zamani, Y., Monfared, S.A.H. and Hamidianpour, M.H. (2020) A
T., Paciorek, C., Gleckler, P.J., Sperber, K.R., Collins, W.D., comparison CMIP6 and CMIP5 projections for precipitation to
Gettelman, A., Jablonowski, C. (2014) The effect of horizontal observational data: the case of Northeastern Iran. Theoretical
resolution on simulation quality in the Community Atmo- and Applied Climatology, 142, 1613–1623. https://doi.org/10.
spheric Model, CAM5.1. Journal of Advances in Modeling Earth 1007/s00704-020-03406-x.
Systems, 6, 980–997. https://doi.org/10.1002/2013MS000276
Woldemeskel, F.M., Sharma, A., Sivakumar, B. and Mehrotra, R. SU PP O R TI N G I N F O RMA TI O N
(2014) A framework to quantify GCM uncertainties for use in Additional supporting information may be found in the
impact assessment studies. Journal of Hydrology, 519, 1453– online version of the article at the publisher's website.
1465. https://doi.org/10.1016/j.jhydrol.2014.09.025.
Wu, T., Lu, Y., Fang, Y., Xin, X., Li, L., Li, W., Jie, W., Zhang, J.,
Liu, Y., Zhang, L., Zhang, F., Zhang, Y., Wu, F., Li, J., Chu, M., How to cite this article: Song, Y. H., Shahid, S.,
Wang, Z., Shi, X., Liu, X., Wei, M., Huang, A., Zhang, Y. and & Chung, E.-S. (2022). Differences in multi-model
Liu, X. (2019) The Beijing Climate Center Climate System ensembles of CMIP5 and CMIP6 projections for
Model (BCC- CSM): the main progress from CMIP5 to CMIP6.
future droughts in South Korea. International
Geoscientific Model Development, 12, 1573–1600. https://doi.
Journal of Climatology, 42(5), 2688–2716. https://
org/10.5194/gmd-12-1573-2.
doi.org/10.1002/joc.7386

You might also like