Download Comparison of Monthly Temperature Extremes Simulated by CMIP3

7692
JOURNAL OF CLIMATE
VOLUME 26
Comparison of Monthly Temperature Extremes Simulated by CMIP3
and CMIP5 Models
YAO YAO, YONG LUO, JIANBIN HUANG, AND ZONGCI ZHAO
Center for Earth System Science, and Ministry of Education Key Laboratory for Earth System Modeling,
Tsinghua University, Beijing, China
(Manuscript received 30 July 2012, in final form 20 March 2013)
ABSTRACT
The extreme monthly-mean temperatures simulated by 28 models in the fifth phase of the Coupled Model
Intercomparison Project (CMIP5) are evaluated and compared with those from 24 models in the third phase
of the Coupled Model Intercomparison Project (CMIP3). Comparisons with observations and reanalyses
indicate that the models from both CMIP3 and CMIP5 perform well in simulating temperature extremes,
which are expressed as 20-yr return values. When the climatological annual cycle is removed, the ensemble
spread in CMIP5 is smaller than that in CMIP3. Benefitting from a higher resolution, the CMIP5 models
perform better at simulating extreme temperatures on the local gridcell scale. The CMIP5 representative
concentration pathway (RCP4.5) and CMIP3 B1 experiments project a similar change pattern in the near
future for both warm and cold extremes, and the pattern is in agreement with that of the seasonal extremes. By
the late twenty-first century, the changes in monthly temperature extremes projected under the three CMIP3
(B1, A1B, and A2) and two CMIP5 (RCP4.5 and RCP8.5) scenarios generally follow the changes in climatological annual cycles, which is consistent with previous studies on daily extremes. Compared with the
CMIP3 ensemble, the CMIP5 ensemble shows a larger intermodel uncertainty with regard to the change in
cold extremes in snow-covered regions. Enhanced changes in extreme temperatures that exceed the global
mean warming are found in regions where the retreat of snow (or the soil moisture feedback effect) plays an
important role, confirming the findings for daily temperature extremes.
1. Introduction
A growing number of extreme climate events on
various spatial–temporal scales have been observed in
the past, and these events have significantly affected
human society and the natural environment (Field et al.
2012). The reason for the increase in the frequency and
intensity of extreme climate events and whether these
events are related to anthropogenic global warming has
attracted much attention from scientists. Currently, accumulating evidence suggests that the increase in temperature extremes is closely related to human activities
(Rahmstorf and Coumou 2011; Stott et al. 2011).
Climate models have been widely used to study anthropogenic effects on climate extremes, as these models
can, to some extent, reproduce these climate extremes
(Tebaldi et al. 2006). The third phase of the Coupled
Corresponding author address: Yao Yao, Room 315, Weiqing
Building, Tsinghua University, Haidian, Beijing 100084, China.
E-mail: [email protected]
DOI: 10.1175/JCLI-D-12-00560.1
Ó 2013 American Meteorological Society
Model Intercomparison Project (CMIP3) has opened up
countless investigations, including studies on extreme
temperatures, based on CMIP3 model outputs (Meehl
et al. 2007). Because of the perceptible change in the
shape of the tail of the daily temperature distribution
(Donat and Alexander 2012), most studies are performed using the extreme daily temperatures based on
commonly used climate indices (Meehl and Tebaldi
2004; Kodra et al. 2011; Zhang et al. 2011) and extremes
with long return periods (Kharin et al. 2007; Zwiers et al.
2011) calculated from the daily outputs of models. In
contrast, the extreme temperature on the monthly scale
has seldom been examined in previous studies. There is
no doubt that anomalously cold or warm months have
a profound impact on human society, particularly in
relation to health (Goklany 2008), agriculture (Dreesen
et al. 2012), and economics (Halsnaes et al. 2007). As an
extreme example, consecutive daily high-temperature
events in Moscow resulted in the hottest month on record in July 2010 (Grumm 2011). As a result of both the
magnitude and duration of the Russian heatwave, more
1 OCTOBER 2013
YAO ET AL.
7693
than 50 000 people died, and Russian grain production
was reduced by 20%–30% relative to the 2009 levels,
causing an economic loss of more than $15 billion (U.S.
dollars). Thus, there is an urgent social need to learn
more about the risks of extreme months. Accordingly,
the present study focuses on the extreme monthly temperatures and attempts to assess their simulation in climate models.
Compared with previous studies of model-simulated
monthly temperature extremes (e.g., R€
ais€
anen and
Ruokolainen 2008; R€
ais€
anen and Ylh€
aisi 2011), we take
a more comprehensive approach and analyze the extremes on multispatial scales. Indeed, the evaluation of
simulations across diverse scales will help us better understand the performance of these models. Because of
the lack of homogeneous time series of global daily
temperatures, reanalysis data are typically used when
evaluating the daily extremes simulated by models;
however, uncertainties in the daily temperature extremes
also exist in such reanalyses (Kharin et al. 2007). Therefore, in this study we validate the model simulations with
two station-based datasets and two reanalyses to better
estimate the observational uncertainty and strengthen
the confidence in validations for model simulations.
The latest climate models participating in the fifth
phase of the Coupled Model Intercomparison Project
(CMIP5) are now available, and components such as
terrestrial and marine carbon cycles, dynamic vegetation, and indirect effects of aerosols are included in most
of the models for the first time (Taylor et al. 2012).
Representative concentration pathways (RCPs) have
replaced the Special Report on Emission Scenarios
(SRESs) as the latest generation of scenarios for longterm experiments. The comparison between the two
generations of climate models is important not only for
researchers from various fields who will use the model
outputs but also for modelers by facilitating further
model development. An objective of the present paper
is to identify whether there are improvements in the new
models in simulating the monthly temperature extremes
in the recent past climate; the projected future changes
under the new scenarios are also addressed. The historical simulations of warm and cold months in the 24
CMIP3 models and 28 CMIP5 models are compared,
and the projected future changes under the SRES and
RCP scenarios are evaluated.
Climate Network (GHCN) 1 Climate Anomaly Monitoring System (CAMS; Fan and van den Dool 2008) from
the Climate Prediction Center (CPC). Both datasets are
gridded monthly-mean surface temperature datasets with
a spatial resolution of 0.58 3 0.58, and they cover the
global land area. However, only the land area north of
608S is considered for the observations and model output in this study because the CRU data in Antarctica are
missing.
On the monthly scale, the temperatures in the reanalyses agree well with those from station observation
data and can provide added value in poorly observed
regions, such as high latitudes and parts of Africa (Dee
et al. 2011a). In this study, the monthly-mean 2-m temperatures from the European Centre for Medium-Range
Weather Forecasts (ECMWF) Interim Re-Analysis
(ERA-Interim; Dee et al. 2011b) and NCEP Climate
Forecast System Reanalysis (CFSR) (Saha et al. 2010) are
also used for validation. Note that because the 2-m temperature is not assimilated in CFSR the monthly-mean
value of the 6-hourly forecast is used instead.
The monthly-mean surface air temperatures from
the 24 CMIP3 models are provided by the Program for
Climate Model Diagnosis and Intercomparison. The
twentieth-century climate simulations (20C3M) and
the SRES B1, A1B, and A2 experiments are used for the
present-day study and for future projections. However,
the results from the SRES B1 and A2 experiments are
only available for 21 and 19 CMIP3 models, respectively.
GISS-EH, INGV-SXG, and HadGEM1 lack the SRES
B1 simulations, whereas CGCM3.1-T63, FGOALS-g1.0,
GISS-AOM, GISS-EH, and MIROC3.2 (hires) lack the
SRES A2 simulations. The 28 CMIP5 model results are
obtained through data portals of the Earth System Grid
Federation. The historical RCP4.5 and RCP8.5 experiments from the 28 CMIP5 models are used in this study.
Both the 20C3M and the historical experiments are
driven by anthropogenic and natural forcings, and the
simulation is also forced by a time-evolving land cover
for most of the CMIP5 models (Taylor et al. 2012). Only
one realization from each model is used here so that all
models are weighted equally in the considered multimodel statistics. The institution and resolution information of each model and full model expansions are
listed in Table 1, showing a generally higher horizontal
resolution for the CMIP5 models.
2. Datasets
3. Methods
Two station-based datasets are used to evaluate the
performance of the models in this study: University of
East Anglia’s Climate Research Unit (CRU) CRU TS3.1
(Mitchell and Jones 2005) and the Global Historical
There are many approaches to describing climate
extremes. Thus, differing conclusions may be drawn. For
extremes on the daily scale, such climate indices as warm
nights (TN90P) and the diurnal temperature range (DTR)
—
Beijing Climate Center (BCC), China
—
CMIP3 model expansion
—
Istituto Nazionale di Geofisica e
Vulcanologia, SINTEX-G
—
Australian Community Climate and
Earth-System Simulator, version 1.3
ACCESS1.3 (192 3 145)
MPI-ESM-MR (192 3 96)
MPI-ESM-LR (192 3 96)
Max Planck Institute Earth System Model,
low resolution
Max Planck Institute Earth System Model,
medium resolution
CSIRO-Mk3.6.0 (192 3 96) Commonwealth Scientific and Industrial
Research Organisation Mark, version 3.6.0
—
—
Australian Community Climate and
Earth-System Simulator, version 1.0
ACCESS1.0 (192 3 145)
Centre National de Recherches
Meteorologiques Coupled Global Climate
Model, version 5
NorESM1-M (144 3 96)
Norwegian Earth System Model, version
1 (intermediate resolution)
HadGEM2-AO (192 3 145) Hadley Centre Global Environment
Model, version 2 - Atmosphere and Ocean
CCSM4 (288 3 192)
Community Climate System Model,
version 4
—
—
INM-CM4 (180 3 120)
Institute of Numerical Mathematics Coupled
Model, version 4.0
CNRM-CM5 (256 3 128)
—
CanESM2 (128 3 64)
—
Beijing Normal University - Earth System
Model
Second Generation Canadian Earth System
Model
BNU-ESM (128 3 64)
—
Beijing Climate Center, Climate System
Model, version 1.1
—
CMIP5 model expansion
BCC_CSM1.1 (128 3 64)
CMIP5 models
JOURNAL OF CLIMATE
—
Istituto Nazionale di Geofisica e
INGV-SXG (320 3 160)
Vulcanologia (INGV), Italy
Max Planck Institute (MPI), Germany ECHAM5 (192 3 96)
Bjerknes Centre for Climate Research BCCR-BCM2.0 (128 3 64) Bjerknes Centre for Climate
(BCCR), Norway
Research Bergen Climate Model,
version 2.0
Beijing Norml University
—
—
(BNU), China
CGCM3.1-T47 (96 3 48) Canadian Centre for Climate
Canadian Centre for Climate
Modelling and Analysis (CCCMA),
Modelling and Analysis
Canada
(CCCma) Coupled Global Cli
mate Model, version 3.1-T47
CGCM3.1-T63 (128 3 64) Canadian Centre for Climate
Modelling and Analysis
(CCCma) Coupled Global
Climate Model, version 3.1-T63
Centre National de Recherches
CNRM-CM3 (128 3 64)
Centre National de Recherches
Meteorologiques (CNRM), France
Meteorologiques Coupled
Global Climate Model, version 3
Norwegian Climate Center (NCC),
—
—
Norway
National Institute of Meteorological
—
—
Research (NIMR), Korea
National Center for Atmospheric
CCSM3 (256 3 128)
Community Climate System
Research (NCAR), United States
Model, version 3
PCM (128 3 64)
Parallel Climate Model
Institute of Numerical Mathematics
INM-CM3.0 (72 3 45)
Institute of Numerical
(INM), Russia
Mathematics Coupled Model,
version 3.0
Commonwealth Scientific and
CSIRO-Mk3.0 (192 3 96) Commonwealth Scientific and
Industrial Research Organisation
Industrial Research
(CSIRO), Australia
Organisation Mark, version 3.0
CSIRO-Mk3.5 (192 3 96) Commonwealth Scientific
and Industrial Research
Organisation Mark, version 3.5
—
—
CMIP3 models
Sponsor, country
TABLE 1. List of the CMIP3 and CMIP5 models used in this study with full expansions. Horizontal resolution of the atmosphere is given after each model identification.
7694
VOLUME 26
CMIP3 models
L’Institut Pierre-Simon Laplace
(IPSL), France
Goddard Institute for Space
Studies (GISS), United States
—
—
IPSL-CM4 (96 3 72)
GISS-ER (72 3 46)
GISS-EH (72 3 46)
GISS-AOM (90 3 60)
—
GFDL-CM2.1 (144 3 90)
—
Geophysical Fluid Dynamics
GFDL-CM2.0 (144 3 90)
Laboratory (GFDL), United States
ECHO-G (96 3 48)
Meteorological Institute of the
University of Bonn (MIUB)/Korea
Meteorological Administration
(KMA), Germany/Korea
IAP, China
FGOALS-g1.0 (128 3 60)
Sponsor, country
GFDL-ESM2G (144 3 90)
FGOALS-s2 (128 3 108)
GFDL-CM3 (144 3 90)
FGOALS-g2 (128 3 60)
—
CMIP5 models
Flexible Global Ocean–Atmosphere–Land
System Model gridpoint, second spectral
version
—
Geophysical Fluid Dynamics Laboratory
Climate Model, version 3
—
CMIP5 model expansion
Geophysical Fluid Dynamics Laboratory
Earth System Model with Generalized
Ocean Layer Dynamics (GOLD)
component (ESM2G)
—
GFDL-ESM2M (144 3 90) Geophysical Fluid Dynamics Laboratory
Earth System Model with Modular
Ocean Model 4 (MOM4) component
(ESM2M)
Goddard Institute for Space Studies, GISS-E2-R (144 3 90)
Goddard Institute for Space Studies
Atmosphere–Ocean Model
Model E, coupled with the Russell
ocean model
Goddard Institute for Space Studies
—
—
Model E, coupled with the
HYCOM ocean model
Goddard Institute for Space Studies
—
—
Model E-R
L’Institut Pierre-Simon Laplace
IPSL-CM5A-LR (96 3 96) L’Institut Pierre-Simon Laplace Coupled
Coupled Model, version 4
Model, version 5, coupled with NEMO,
low resolution
—
IPSL-CM5A-MR
L’Institut Pierre-Simon Laplace Coupled
(144 3 143)
Model, version 5, coupled with NEMO,
mid resolution
—
IPSL-CM5B-LR (96 3 96) L’Institut Pierre-Simon Laplace Coupled
Model, version 5B, coupled with
NEMO, low resolution
Flexible Global Ocean–Atmosphere–
Land System Model gridpoint,
version 1.0
—
Geophysical Fluid Dynamics
Laboratory Climate Model,
version 2.0
Geophysical Fluid Dynamics
Laboratory Climate Model,
version 2.1
ECHAM and the global Hamburg
Ocean Primitive Equation
CMIP3 model expansion
TABLE 1. (Continued)
1 OCTOBER 2013
YAO ET AL.
7695
7696
Model for Interdisciplinary Research
on Climate, Earth
System Model, Chemistry Coupled
MIROC5 (256 3 128)
Model for Interdisciplinary Research
on Climate, version 5
MRI-CGCM3 (320 3 160) Meteorological Research Institute
Coupled Atmosphere–Ocean General
Circulation Model,
version 3
HadGEM2-CC (192 3 145) Hadley Centre Global Environment
Model, version 2 - Carbon Cycle
HadGEM2-ES (192 3 145) Hadley Centre Global Environment
Model, version 2 - Earth System
HadCM3 (96 3 73)
Met Office (UKMO), UK
HadGEM1 (192 3 145)
MRI-CGCM2.3.2
(128 3 64)
Meteorological Research
Institute (MRI), Japan
—
MIROC-3.2 (medres)
(128 3 64)
Meteorological Research
Institute Coupled
Atmosphere–Ocean General
Circulation Model, version 2.3.2
Hadley Centre Coupled
Model, version 3
Hadley Centre Global
Environment Model, version 1
MIROC-ESM-CHEM
(128 3 64)
Model for Interdisciplinary Research
on Climate, Earth System Model
MIROC-ESM (128 3 64)
Model for Interdisciplinary
Research on Climate,
version 3.2 (high resolution)
Model for Interdisciplinary
Research on Climate,
version 3.2 (medium resolution)
—
MIROC-3.2 (hires)
(320 3 160)
Model for Interdisciplinary
Research on Climate
(MIROC), Japan
CMIP3 model expansion
CMIP3 models
Sponsor, country
TABLE 1. (Continued)
CMIP5 models
CMIP5 model expansion
JOURNAL OF CLIMATE
VOLUME 26
have been widely used to indicate ‘‘moderate extremes’’
(Zhang et al. 2011). For a rare climate event, such as
a 1-in-20-yr daily maximum temperature, the extreme
value theory (EVT) has been applied to estimate the
event’s return value or return level (Kharin et al. 2007).
Given a rather small sample size (one value per year),
the monthly-mean temperature extremes favored in this
study are difficult to quantify nonparametrically. Because we define the extreme in terms of 20-yr return
values of annual temperature extremes, the extreme can
be estimated from the tail of a fitted parametric distribution. For warm months (the warmest month in each
year), a generalized extreme value (GEV) distribution is
fitted, and the 95th percentile is considered the warm
extreme. For cold months (the coldest month in each
year), the temperatures are multiplied by 21 before
fitting a distribution; the sign of the 95th percentile of the
fitted GEV distribution is then reversed again, thereby
generating the cold extreme. The method of L-moments
is applied to estimate the parameters of the GEV distribution (Hosking 1990), which is a feasible approach
that can be universally performed for all the grid boxes
in this study. An alternative to the application of EVT,
and an easier way to define the extreme, is the mean plus
(or minus) 1.645 standard deviation for the annual
maxima (or minima) of the monthly temperatures (e.g.,
Becker et al. 2013), which is approximately equal to the
20-yr return value if the monthly-mean temperature for
each grid box is assumed to be from a Gaussian distribution. However, such a definition can cause a biased
estimation for extremes in regions where the monthlymean temperatures are markedly nonnormal (von
Storch and Zwiers 2002). To evaluate whether the two
hypotheses in distribution are appropriate here, the
Kolmogorov–Smirnov tests are performed on the null
hypothesis that the annual extremes of the monthly
means in a grid box are drawn from GEV or from
a Gaussian distribution, with the parameters estimated
by the L-moment method. The tests show that there is
no need to reject either of the two distributions. At the
0.05 level, 5.18% and 5.67% of the global land fails the
test for warm and cold extremes, respectively, when
fitted using the GEV distribution. Bootstrapping, generated by resampling the global fields to preserve the
spatial dependencies, is employed to quantify the sampling error, which may be large when the distribution is
estimated from a small sample (Efron 1987). The variance of sampling errors generated from fitting the GEV
distribution is slightly larger than from fitting the Gaussian
distribution due to the estimation of more parameters in
the former.
The time period 1980–99 is considered for the current
climate simulation because it is covered by both the
1 OCTOBER 2013
YAO ET AL.
7697
FIG. 1. Zonal bands and regions considered in this study. Giorgi and Francisco (2000) provide
the coordinates of the regions.
reanalyses and CMIP model simulations. The time periods 2016–35 and 2080–99 are chosen for near-future and
long-term projections, respectively. All the temperature
data are preprocessed as anomalies to the annual mean
climatology in 1980–99 to eliminate the possible influence caused by the systematic bias of model climatology because the extreme, rather than the mean
climate, is considered here. The observation and model
data are remapped onto a Gaussian N64 grid (256 3
128) using bilinear interpolation prior to the analysis. To
study the extreme temperatures at different latitudes,
the global land area is divided into four zonal bands in
this study: the north frigid zone (NFZ; 66.58–908N), the
north temperate zone (NTZ; 23.58–66.58N), the torrid zone
(TZ; 23.58S–23.58N), and the south temperate zone (STZ;
66.58–23.58S). For regional study purposes, the temperature extremes are spatially averaged over 21 regions,
similar to the approach employed by Giorgi and Francisco
(2000; Fig. 1). This region definition has been widely
adopted in previous evaluations based on CMIP3 models
(e.g., Weisheimer and Palmer 2005; Stott et al. 2011).
4. Historical simulation of monthly temperature
extremes
The spatially averaged 20-yr return values of monthlymean temperatures are evaluated and presented in the
left panel of Fig. 2. The red and blue boxes represent the
CMIP3 and CMIP5 models, respectively, and the five
bands in each box indicate the lower and upper bounds,
the median, and the 25th and 75th percentiles of the
multimodel ensemble (MME). The corresponding statistics for the observations and reanalyses are calculated
from the bootstrap samples to illustrate the sampling
error. Globally (land only and excluding Antarctica)
and zonally averaged, the model-simulated that extreme monthly temperatures agree well with both the
observations and reanalyses. The CFSR cold extremes
in the NFZ are warmer than those in the observations
and ERA-Interim; however, further investigation is required to determine whether this is because the 2-m
temperatures are not analyzed in CFSR.
The good performance of the models for simulation
on a large spatial scale is unsurprising because previous
studies have indicated that globally averaged daily
(Kharin et al. 2007) and seasonal (Anderson 2011) temperature extremes can be well simulated using CMIP3
models. Moreover, spatial smoothing reduces the sampling error at the expense of the loss of local information
(Masson and Knutti 2011). Regardless, most models tend
to simulate too high warm and too low cold extremes of
monthly-mean temperature, and t tests performed on the
global land mean extremes show that such bias is significant at the 5% level for both the CMIP3 and CMIP5
ensembles. However, the bias is much lower when the
annual cycle is removed (Fig. 2, right). As a first-order
approximation, the simulation of an overly strong annual cycle may account for most of the bias in extreme
monthly-mean temperatures.
The sampling error is small on global and zonal scales
(Kharin et al. 2007; Stott et al. 2011) and can be negligible when compared with the intermodel uncertainty
interval. The difference between the two station-based
datasets is larger than the sampling error, particularly
for the TZ, a region for which the station spatial coverage is poor. The intermodel differences are larger at
high-latitude regions, where the interannual temperature variability is greater. F tests are performed to determine whether the CMIP5 ensemble has a smaller
intermodel variance than the CMIP3 ensemble; at the
5% level, this difference is only significant for the cold
extremes in the TZ. As the first-order approximation,
the uncertainty in the simulated seasonal cycle contributes
to most of the intermodel variance. In a snow-covered
7698
JOURNAL OF CLIMATE
VOLUME 26
FIG. 2. Boxplots of spatially averaged (left) 20-yr return values for (top) warm and (bottom) cold extremes in 1980–99
and (right) those with climatological annual cycles removed.
region such as the NFZ, the strength of seasonal cycle is
correlated with the snow–albedo feedback (Hall and Qu
2006), and the large intermodel spread in the snow–albedo feedback found in the CMIP3 models (Qu and
Hall 2007) partly explains the relatively greater intermodel difference in the temperature extremes in the
NFZ. Removing the annual cycles, the extreme monthlymean temperature mainly depends on the magnitude of
the interannual variability. If we eliminate the bias on
the seasonal cycle, the intermodel differences become
smaller but are still larger than the sampling error.
Consistent with previous research on the interannual
variability (Scherrer 2011), the intermodel differences
are greater over high latitudes. F tests are again performed between the CMIP3 and CMIP5 ensembles, with the
results showing that CMIP5 has a significantly smaller
intermodel spread than CMIP3 at a 5% level for the
global and zonal mean extreme temperatures, with the
exception of the warm extreme in the NFZ. It is worth
mentioning that recent studies also suggest that improvements in the CMIP5 models can be found in the
simulation of ENSO (Kim and Yu 2012; Zhang and Jin
2012), which greatly influences the global climate interannual variability.
To evaluate the overall performance in simulating
extreme monthly-mean temperatures on the gridbox
scale, the root-mean-square error (RMSE) between
the global fields (land only and excluding Antarctica) is
calculated with reference to the observations and
reanalyses. The uncentered RMSE used here is defined
as follows:
1
W
å å wij (Xij 2 Yij )2 .
i
j
The simulated and observed extremes correspond to
X and Y, with the longitude and latitude dimensions
denoted by i and j, respectively. The term W is the sum
of the weights (wij) proportional to the gridbox area. In
the interest of brevity, the individual outcomes obtained
from the two station-based datasets and two reanalyses
are not shown here (the differences among them are in
fact small). Instead, the average of the four RMSEs
with respect to the different station-based datasets and
reanalyses is provided to represent the overall competence (Fig. 3). Note that the results in this study
are obtained from temperatures on a Gaussian N64
grid (approximately 1.48 3 1.48) and that the choice of
spatial scale would have an impact on the outcome (e.g.,
Sakaguchi et al. 2012). On larger scales, the models tend
to show better performance, and their RMSEs become
smaller.
1 OCTOBER 2013
YAO ET AL.
7699
FIG. 3. The mean of the RMSEs of model-simulated (top) warm and (bottom) cold extremes
compared with the two station-based datasets and two reanalyses in 1980–99. The horizontal
solid lines represent the spatially averaged sampling errors in CRU observation.
The RMSEs of the warm extremes of most of the
models are approximately 28–38C, which is greater than
the global average of sampling errors at each grid box. In
our study, the global averages of the warm and cold
extreme sampling errors in CRU are 0.318 and 0.598C,
respectively, similar to those (0.348 and 0.548C) estimated from the daily temperatures (Kharin et al. 2007).
The RMSEs of the cold extremes are generally larger
than those of the warm extremes, as is the sampling error. The MME means have smaller RMSEs than any of
the individual models, as some biases may be canceled
out in the ensemble (R€
ais€
anen 2007). The RMSE of the
CMIP5 ensemble mean (blue) is smaller than for the
CMIP3 (red), although the difference is minor. Some
improvements can also be found in the simulation from
the individual models. As shown in Fig. 3, of the 13
7700
JOURNAL OF CLIMATE
modeling groups employed in both CMIP3 and CMIP5,
10 modeling groups have smaller RMSEs for the cold
extreme simulations in their latest versions compared
with the older versions. Chi-squared tests show that such
a difference is significant at the 10% level. For the warm
extreme simulations, 8 of the 13 modeling groups show
improvements. Because of the generally higher horizontal resolutions in the CMIP5 models, the topography is
more detailed and thus benefits the temperature simulation on the gridbox scale (Masson and Knutti 2011). For
example, the latest models from the Institute of Atmospheric Physics (IAP; FGOALS-g2 and FGOALS-s2)
perform markedly better than their previous version
(FGOALS-g1.0), according to the decrease in RMSE.
We also evaluate the simulations on the gridbox scale
after the climatological annual cycles are removed from
the extreme monthly-mean temperatures (not shown).
Consistent with the above results, the RMSEs become
smaller but are still larger than the sampling errors.
Improvements in the CMIP5 ensemble mean and individual models are also detectable, implying a better
representation of the interannual variability in the latest
models.
To evaluate the simulations of the extreme monthlymean temperatures in different regions, the relative error defined by Gleckler et al. (2008) is applied as a metric
in this study. The relative error is a normalized RMSE,
and thus a lower value indicates a better performance.
Note that to identify the differences between the two
ensembles, the normalization for RMSEs of both the
CMIP3 and CMIP5 models are with respect to the
multimodel ensemble median errors in the CMIP3 ensemble. According to this definition, a relative error less
than zero would indicate a better performance than the
median of the CMIP3 ensemble. Figure 4 shows the relative error of the extreme monthly-mean temperatures in
the 24 CMIP3 and 28 CMIP5 models for the 21 regions
compared using both observations and reanalyses. Each
grid cell that represents a single model for a certain region is separated into four smaller boxes. The top corners
represent the outcome of the CRU- and CPC-gridded
station observations, and the bottom corners represent
those for the ERA-Interim and CFSR reanalyses. In
general, the differences between the observations and
reanalyses play a minor role here, and the biases are
systematic compared with the different reference data.
Consistent with the findings above, the ensemble mean
and ensemble median outperform most of the individual
models in almost every region: the errors can be partly
canceled through the multimodel mean or median, resulting in a better performance (Tebaldi and Knutti 2007).
Because the observational uncertainty discussed above
does not exert much influence on our results, the relative
VOLUME 26
errors with respect to the different station-based datasets
and reanalyses are then averaged for the purposes of
comparison. The improvements in the CMIP5 models on
the regional scale can be found in some regions, particularly for cold extremes. The relative errors of the CMIP5
ensemble mean are smaller than those of the CMIP3
ensemble mean in 9 regions for warm extremes and 16
regions for cold extremes. The individual model simulations also indicate greater improvements for cold extremes than warm extremes. Chi-squared tests are
performed on the number of 28 CMIP5 models having
relative errors less than zero at each region, with significant improvements at the 5% level found in 4 and 6
regions for warm and cold extremes, respectively. Consistent with the above findings, the improvements still
exist after the climatological annual cycles are removed.
5. Future projection of monthly temperature
extremes
In this section, we begin by examining the projected
future changes of the extreme monthly- mean temperatures in the near future (2016–35) under two scenarios
(SRES B1 and RCP4.5 in the CMIP3 and CMIP5 experiments, respectively). Considering that scenario uncertainty does not play an important role in the next
one to two decades (Hawkins and Sutton 2009), interscenario differences can be largely ignored when comparing the CMIP3 and CMIP5 ensembles. The change
relative to that in the recent (1980–99) climate is often
not significant due to the low signal-to-noise ratio. Thus,
a method similar to that suggested by Tebaldi et al.
(2011) is adopted here to determine the agreement of
the models. For each model and its bootstrap samples,
a Student’s t test (or Wilcoxon’s signed rank test) is performed for the derived 20-yr return value (or waiting
period) grid point by grid point. The corresponding region is masked out in white if less than half the models
show a significant change at the 10% level through
t tests; otherwise, the corresponding region is stippled if
90% of the models agree on the direction of the change.
Figure 5 shows the MME mean change in the temperature extremes simulated by the CMIP3 (top) and CMIP5
(bottom) models with the intermodel agreements measured from the method discussed above. Except for several grid boxes, nearly no value in the global land area is
masked out, as over half the models project a significant
change in temperature extremes. Additionally, a positive change is projected by the ensemble mean for almost all the land areas around the globe, although the
direction of the change lacks sufficient confidence at
locations over the global land, as indicated by the unstippled grid boxes in Fig. 5. The most substantial
1 OCTOBER 2013
YAO ET AL.
7701
FIG. 4. Relative errors for (top) warm and (bottom) cold extremes (1980–99) simulated
in (left) CMIP3 (labeled in red) and CMIP5 (labeled in blue) models compared with those
in (top-left corner in each grid cell) CRU, (top-right corner in each grid cell) CPC,
(bottom-left corner in each grid cell) ERA-Interim and (bottom-right corner in each grid
cell) CFSR.
change can be found in the snow-covered regions at the
high latitudes of the Northern Hemisphere during cold
months, a result that has also been found for extreme
daily minimum temperatures and is attributed to snow
retreat (Kharin et al. 2007). As a result, the global (land
only and excluding Antarctica) mean cold extremes will
warm faster than the warm extremes in the near future.
Despite the different scenarios applied in CMIP3 and
CMIP5, the patterns of the warm and cold extreme
changes are similar between the two ensemble means.
The spatial correlations between the two fields are 0.63
for warm extremes and 0.89 for cold extremes.
Figure 6 shows the MME median of the near-future
(2016–35) return periods for the temperature extremes
under the current climate (1980–99). The waiting period
for warm extremes is significantly shortened, whereas
7702
JOURNAL OF CLIMATE
VOLUME 26
FIG. 5. Change in 20-yr return values for (left) warm and (right) cold extremes simulated in (top) CMIP3 B1 and (bottom) CMIP5
RCP4.5 experiments in 2016–35 relative to 1980–99. Black stippling indicates areas where more than 90% of the models agree on the sign
of the change.
the cold extremes show a tendency to occur much more
rarely. In the tropics and southern Eurasia, regions
where the signal-to-noise ratio of warming is generally
higher, many areas are projected to experience 20-yr
return values of warm extremes in less than 2 years, indicating that the present-day extremes will become the
norm in those regions. This result is in accordance with
previous studies on extreme warm seasons from CMIP3
outputs (Anderson 2011; Diffenbaugh and Scherer 2011).
The change is less significant at high latitudes, where the
magnitude of interannual variability is larger, but it still
shows a reduction in the waiting period by a factor of
2 to 4.
For cold extremes, the changes are most significant in
tropical South America, Africa, and South Asia. Consistent with the findings of R€
ais€
anen and Ylhaisi (2011),
the cold months in areas where the interannual variability
is small become much less frequent given a modest
warming. However, in parts of the Northern Hemisphere midlatitudes, the probability projected by the
ensemble medians for the occurrence of cold extremes is
only slightly less than under the current climate and not
all the models agree on the direction of the change,
which is in accordance with the discussion regarding
Fig. 5. Yang and Christensen (2012) reported on the
European cold winter months simulated by 13 CMIP5
models and suggest that the chance of cold winters
occurring in Europe is only slightly reduced in the near
future. In their study, the pattern of cold Europe/warm
Arctic is detected, potentially indicating a connection
between cold European winters and the decrease in
Arctic sea ice. Note that cold winters differ from the cold
extremes defined in our study, and further investigation
is required to determine whether the proposed mechanism also applies here.
The average spatial change in extreme monthly-mean
temperatures by the end of the twenty-first century
(2080–99) under different scenarios in the CMIP3 and
CMIP5 ensembles is shown in the left panel of Fig. 7.
The change is greater under higher greenhouse gas emission scenarios for both the warm and cold global (land
only and excluding Antarctica) and zonal mean (land
only) extreme temperatures, with the greatest warming
projected under the RCP8.5 scenario and the least under
the B1 scenario. The CMIP5 RCP4.5 simulations project
a lesser warming compared with A1B and A2, but it is
slightly greater than B1. Globally averaged, the cold
extremes will warm faster than the warm extremes under all five scenarios, and t tests show that the difference
is significant at the 10% level under the B1, RCP4.5, and
A1B scenarios. Changes in the snow cover and Arctic
sea ice are mainly responsible for these differences, as
the cold extremes in the NFZ warm much faster than in
other zonal bands.
The projected changes in the extreme monthly-mean
temperatures are similar to those for the extreme daily
1 OCTOBER 2013
YAO ET AL.
7703
FIG. 6. Waiting time for present time (1980–99) (left) warm and (right) cold extremes simulated in (top) CMIP3 B1 and (bottom)
CMIP5 RCP4.5 experiments in 2016–35. Black stippling indicates areas where more than 90% of the models agree on the sign of the
change.
temperatures (e.g., Kharin et al. 2007), as the contribution
is mostly attributed to the change in the climatological
maxima and minima of the annual cycle. The results after
removing the changes in the annual cycles are shown in
the right panel of Fig. 7. Roughly speaking, the changes in
extremes with the climatological annual cycles removed
are not significant, indicating that the interannual variability will most likely remain unchanged. This is also the
case for the daily temperatures, with the exception of cold
temperatures at high latitudes where the change in cold
extremes exceeds the warming of the mean temperatures
in the climatologically coldest months due to changes in
the short-term temperature variability.
Consistent with Hawkins and Sutton (2009), the scenario uncertainty becomes dominant with time. As illustrated in Fig. 7, there is little overlap between the
intermodel uncertainty ranges under scenarios B1 and
RCP8.5 during 2080–99, and the intermodel differences
are larger for the cold extremes in the NFZ than elsewhere, which may be related to the uncertainties in snow
cover feedback. The uncertainty range of cold extremes in
the NFZ is larger in the CMIP5 experiments than in the
CMIP3 experiments, and F tests performed on the CMIP5
RCP8.5 and CMIP3 simulations show that the difference is
significant at the 5% level. As the interannual variability
changes little, the intermodel variability in the changes in
extremes with the annual cycles removed is small.
By the late twenty-first century, the simulated changes
in the regionally averaged extreme monthly-mean
temperatures follow the mean changes in climatology, as
discussed above, and the uncertainties are mainly derived from the scenarios. Thus, the change in extremes in
proportion to the global mean warming may make additional sense. Figure 8 shows the change in the regionalaveraged extremes scaled by the change in the global
mean temperature. To indicate the intermodel differences,
each grid cell is divided into five parts, representing the
5%, 25%, 50%, 75%, and 95% quantiles of MME. Scaled
by the global mean temperature change, the changes in
both the warm and cold extremes vary little under different scenarios, rather the geographical features become
notable. The ratio is close to one for most of the regions at
low latitudes [e.g., eastern Africa (EAF), western Africa
(WAF), and Southeast Asia (SEA)], indicating that the
changes in the warm and cold extremes in these regions
approximate the magnitude of the global mean warming.
Moreover, this result does not contradict previous studies
on extreme seasonal mean temperatures that used similar
regional definitions (Weisheimer and Palmer 2005; Stott
et al. 2011). These studies suggested that low-latitude
regions are expected to experience the largest increase in
the frequency of extreme seasonal temperatures. Consistent with the above findings, the interannual variability
is lower in these regions, and a small shift in the mean
climatology would make the extreme become the norm.
The change is generally greater for cold extremes than
warm extremes at the moderate to high latitudes. In
snow-covered regions ([North Asia (NAS), Northern
7704
JOURNAL OF CLIMATE
VOLUME 26
FIG. 7. Boxplots of changes in spatially averaged (top) warm and (bottom) cold temperature (left) extremes simulated by the two MMEs
under five scenarios in 2080–99 relative to 1980–99 and (right) those with changes in climatological annual cycles removed.
Europe (NEU), Greenland (GRL), and Alaska (ALA)]),
the large changes in cold extremes are most likely due to
snow retreat, which is also the case for the extreme daily
temperatures (e.g., Orlowsky and Seneviratne 2012). The
change in warm extremes is greater in regions that become generally drier and is largest in Mediterranean
(MED), a region where the heat stress risks are also
projected to increase dramatically (Diffenbaugh et al.
2007), perhaps due to soil moisture effects (Hirschi et al.
2011).
6. Summary and discussion
The simulations of extreme temperatures in 24 CMIP3
models and 28 CMIP5 models are evaluated in this paper.
Different from previous studies on extreme daily or
seasonal temperatures, our study focuses on the annual
extremes of monthly means, as extreme events on this
time scale may exert a profound impact on both the
natural environment and human society.
Validated through observations (CRU and CPC) and
reanalyses (ERA-Interim and CFSR), the performance
of the two MMEs in simulating the monthly temperature
extremes under the current climate (1980–99) is evaluated. In general, both the CMIP3 and CMIP5 MMEs can
simulate warm and cold extremes reasonably well. Note
that the temperature extremes are calculated from the
anomalies of the annual mean climatology of the individual
models. Globally and zonally averaged, the models tend to
overestimate the magnitude of temperature extremes; as
first-order approximations, the biases in the climatological
annual cycles contribute to most of this overestimation.
When the annual cycles are removed, improvements in the
CMIP5 models can be found, with a narrower ensemble
spread in the CMIP5 ensemble than in CMIP3. This result
may imply a better representation in interannual variability by the CMIP5 models. Improvements in the simulations
of extreme temperatures on a gridbox scale are discernible,
possibly due to the generally higher resolution of the
CMIP5 models. Improvements in simulating the regional
temperature extremes can also be found, based on the
overall performance, compared with the two stationbased datasets and the two reanalyses.
The CMIP5 MME simulates a similar pattern in the
change in the monthly extreme temperatures in the near
future (2016–35), as does the CMIP3 ensemble. The
patterns are also in agreement with the CMIP3 simulations for seasonal extremes (e.g., Diffenbaugh and
Scherer 2011). The projections of temperature extremes
in the late twenty-first century under the three SRES
1 OCTOBER 2013
YAO ET AL.
FIG. 8. Regional-averaged change in (top) warm and (bottom) cold extremes scaled by global
mean temperature change simulated by the CMIP3 and CMIP5 models. Within each scenario,
the five grid cells from left to right represent the 5%, 25%, 50%, 75%, and 95% quantiles of the
MME.
7705
7706
JOURNAL OF CLIMATE
(B1, A1B, and A2) and the two RCP scenarios (RCP4.5
and RCP8.5) are compared. Consistent with previous
studies on daily extremes (Kharin et al. 2007), the changes
in the extreme monthly temperatures mainly follow the
changes in the mean temperatures of the climatologically
warmest (or coldest) months. However, the uncertainty
range is still large for the change in cold extremes in snowcovered regions. The change in the extreme monthly
mean temperatures scaled by global mean temperature
changes is examined to determine the response in regional temperature extremes, regardless of the scenario
uncertainty. Robust regional changes are projected in
regions possibly affected by snow retreat or soil moisture feedback, which is in accordance with studies on
daily extremes (e.g., Orlowsky and Seneviratne 2012).
Although the sampling errors for the spatial means of
temperature extremes are negligible compared with the
intermodel differences and scenario uncertainty, the
local sampling errors are large for some regions, thus
reducing confidence in the near-future predictions. In
addition, the signal-to-noise ratio is lower for the nearfuture change in temperature extremes than for the
long-term projections due to the weaker signal. As a new
endeavor for CMIP5, the decadal experiment assimilates the initial conditions and may provide the possibility for a more skillful prediction of the mean (Fyfe
et al. 2011) and extreme temperatures (Eade et al. 2012).
Our future studies will be conducted on the initialized
decadal predictions simulated by the CMIP5 models.
Acknowledgments. This work is supported by the
National Basic Research Program of China (2010CB95050103 and 2009CB421407) and the National Natural Science
Foundation of China (41275076 and 40905046). We acknowledge the World Climate Research Programme’s
Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling
groups (listed in Table 1 of this paper) for producing
and making available their model output. For CMIP,
the U.S. Department of Energy’s Program for Climate
Model Diagnosis and Intercomparison provides coordinating support and led the development of software
infrastructure in partnership with the Global Organization for Earth System Science Portals. We thank the three
anonymous reviewers for constructive suggestions.
REFERENCES
Anderson, B. T., 2011: Near-term increase in frequency of seasonal
temperature extremes prior to the 28C global warming target.
Climatic Change, 108, 581–589.
Becker, E. J., H. van den Dool, and M. Pe~
na, 2013: Short-term
climate extremes: Prediction skill and predictability. J. Climate, 26, 512–531.
VOLUME 26
Dee, D. P., K. Kallen, A. J. Simmons, and L. Haimberger, 2011a:
Comments on ‘‘Reanalyses suitable for characterizing longterm trends.’’ Bull. Amer. Meteor. Soc., 92, 65–70.
——, and Coauthors, 2011b: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system.
Quart. J. Roy. Meteor. Soc., 137, 553–597, doi:10.1002/qj.828.
Diffenbaugh, N. S., and M. Scherer, 2011: Observational and model
evidence of global emergence of permanent, unprecedented
heat in the 20th and 21st centuries. Climatic Change, 107, 615–
624, doi:10.1007/s10584-011-0112-y.
——, J. S. Pal, F. Giorgi, and X. Gao, 2007: Heat stress intensification in the Mediterranean climate change hotspot.
Geophys. Res. Lett., 34, L11706, doi:10.1029/2007GL030000.
Donat, M. G., and L. V. Alexander, 2012: The shifting probability
distribution of global daytime and night-time temperatures.
Geophys. Res. Lett., 39, L14707, doi:10.1029/2012GL052459.
Dreesen, F. E., H. J. De Boeck, I. A. Janssens, and I. Nijs, 2012:
Summer heat and drought extremes trigger unexpected
changes in productivity of a temperate annual/biannual plant
community. Environ. Exp. Bot., 79, 21–30, doi:10.1016/
j.envexpbot.2012.01.005.
Eade, R., E. Hamilton, D. M. Smith, R. J. Graham, and A. A.
Scaife, 2012: Forecasting the number of extreme daily events out
to a decade ahead. J. Geophys. Res., 117, D21110, doi:10.1029/
2012JD018015.
Efron, B., 1987: Better bootstrap confidence intervals. J. Amer.
Stat. Assoc., 82, 171–185.
Fan, Y., and H. van den Dool, 2008: A global monthly land surface
air temperature analysis for 1948–present. J. Geophys. Res.,
113, D01103, doi:10.1029/2007JD008470.
Field, C. B., and Coauthors, Eds., 2012: Managing the Risks of
Extreme Events and Disasters to Advance Climate Change
Adaptation. Cambridge University Press, 582 pp.
Fyfe, J. C., W. J. Merryfield, V. Kharin, G. J. Boer, W.-S. Lee, and
K. von Salzen, 2011: Skillful predictions of decadal trends in
global mean surface temperature. Geophys. Res. Lett., 38,
L22801, doi:10.1029/2011GL049508.
Giorgi, F., and R. Francisco, 2000: Uncertainties in regional climate
change prediction: A regional analysis of ensemble simulations with the HADCM2 coupled AOGCM. Climate Dyn., 16,
169–182.
Gleckler, P. J., K. E. Taylor, and C. Doutriaux, 2008: Performance
metrics for climate models. J. Geophys. Res., 113, D06104,
doi:10.1029/2007JD008972.
Goklany, I., 2008: Deaths and death rates from extreme weather
events: 1900-2008. J. Amer. Physicians Surg., 14, 102–109.
Grumm, R. H., 2011: The central European and Russian heat event
of July–August 2010. Bull. Amer. Meteor. Soc., 92, 1285–1296.
Hall, A., and X. Qu, 2006: Using the current seasonal cycle to
constrain snow albedo feedback in future climate change.
Geophys. Res. Lett., 33, L03502, doi:10.1029/2005GL025127.
Halsnaes, K., J. Kuhl, and J. E. Olesen, 2007: Turning climate
change information into economic and health impacts. Climate Change, 81, 145–162.
Hawkins, E., and R. Sutton, 2009: The potential to narrow uncertainty in regional climate predictions. Bull. Amer. Meteor.
Soc., 90, 1095–1107.
Hirschi, M., and Coauthors, 2011: Observational evidence for soilmoisture impact on hot extremes in southeastern Europe. Nat.
Geosci., 4, 17–21, doi:10.1038/NGEO1032.
Hosking, J. R. M., 1990: L-moments: Analysis and estimation of
distributions using linear combinations of order statistics.
J. Roy. Stat. Soc., B52, 105–124.
1 OCTOBER 2013
YAO ET AL.
Kharin, V. V., F. W. Zwiers, X. Zhang, and G. C. Hegerl, 2007:
Changes in temperature and precipitation extremes in the
IPCC ensemble of global coupled model simulations. J. Climate, 20, 1419–1444.
Kim, S. T., and J.-Y. Yu, 2012: The two types of ENSO in CMIP5
models. Geophys. Res. Lett., 39, L11704, doi:10.1029/
2012GL052006.
Kodra, E., K. Steinhaeuser, and A. R. Ganguly, 2011: Persisting
cold extremes under 21st-century warming scenarios. Geophys. Res. Lett., 38, L08705, doi:10.1029/2011GL047103.
Masson, D., and R. Knutti, 2011: Spatial-scale dependence of climate model performance in the CMIP3 ensemble. J. Climate,
24, 2680–2692.
Meehl, G. A., and C. Tebaldi, 2004: More intense, more frequent,
and longer lasting heat waves in the 21st century. Science, 305,
994–997.
——, C. Covey, T. Delworth, M. Latif, B. McAvaney, J. F. B.
Mitchell, R. J. Stouffer, and K. E. Taylor, 2007: The WCRP
CMIP3 multi-model datasets: A new era in climate change
research. Bull. Amer. Meteor. Soc., 88, 1383–1394.
Mitchell, T. D., and P. D. Jones, 2005: An improved method of
constructing a database of monthly climate observations and
associated high-resolution grids. Int. J. Climatol., 25, 693–712,
doi:10.1002/joc.1181.
Orlowsky, B., and S. I. Seneviratne, 2012: Global changes in extreme events: Regional and seasonal dimension. Climatic
Change, 110, 669–696, doi:10.1007/s10584-011-0122-9.
Qu, X., and A. Hall, 2007: What controls the strength of snowalbedo feedback? J. Climate, 20, 3971–3981.
Rahmstorf, S., and D. Coumou, 2011: Increase of extreme events in
a warming world. Proc. Natl. Acad. Sci. USA, 108, 17 905–
17 909, doi:10.1073/pnas.1101766108.
R€
ais€
anen, J., 2007: How reliable are climate models? Tellus, 59A,
2–29, doi:10.1111/j.1600-0870.2006.00211.x.
——, and L. Ruokolainen, 2008: Estimating present climate in
a warming world: A model-based approach. Climate Dyn., 31,
573–585, doi:10.1007/s00382-007-0361-7.
——, and J. S. Ylh€
aisi, 2011: Cold months in a warming climate.
Geophys. Res. Lett., 38, L22704, doi:10.1029/2011GL049758.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System
reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1057.
7707
Sakaguchi, K., X. Zeng, and M. A. Brunke, 2012: The hindcast skill
of the CMIP ensembles for the surface air temperature trend.
J. Geophys. Res., 117, D16113, doi:10.1029/2012JD017765.
Scherrer, S. C., 2011: Present-day interannual variability of surface
climate in CMIP3 models and its relation to future warming.
Int. J. Climatol., 31, 1518–1529, doi:10.1002/joc.2170.
Stott, P. A., G. S. Jones, N. Christidis, F. W. Zwiers, G. Hegerl, and
H. Shiogama, 2011: Single-step attribution of increasing frequencies of very warm regional temperatures to human influence. Atmos. Sci. Lett., 12, 220–227, doi:10.1002/asl.315.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of
CMIP5 and the experiment design. Bull. Amer. Meteor. Soc.,
93, 485–498.
Tebaldi, C., and R. Knutti, 2007: The use of the multi-model ensemble in probabilistic climate projections. Philos. Trans. Roy.
Soc., 365A, 2053–2075, doi:10.1098/rsta.2007.2076.
——, K. Hayhoe, J. M. Arblaster, and G. A. Meehl, 2006: An intercomparison of model-simulated historical and future
changes in extreme events. Climatic Change, 79, 185–211.
——, J. M. Arblaster, and R. Knutti, 2011: Mapping model
agreement on future climate projections. Geophys. Res. Lett.,
38, L23701, doi:10.1029/2011GL049863.
von Storch, H., and F. W. Zwiers, 2002: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.
Weisheimer, A., and T. N. Palmer, 2005: Changing frequency of occurrence of extreme seasonal temperatures under global warming. Geophys. Res. Lett., 32, L20721, doi:10.1029/2005GL023365.
Yang, S., and J. H. Christensen, 2012: Arctic sea ice reduction and
European cold winters in CMIP5 climate change experiments.
Geophys. Res. Lett., 39, L20707, doi:10.1029/2012GL053338.
Zhang, W., and F.-F. Jin, 2012: Improvements in the CMIP5 simulations of ENSO-SSTA meridional width. Geophys. Res.
Lett., 39, L23704, doi:10.1029/2012GL053588.
Zhang, X., L. Alexander, G. C. Hegerl, P. Jones, A. K. Tank, T. C.
Peterson, B. Trewin, and F. W. Zwiers, 2011: Indices for
monitoring changes in extremes based on daily temperature
and precipitation data. Wiley Interdiscip. Rev. Climate Change,
2, 851–871, doi:10.1002/wcc.147.
Zwiers, F. W., X. Zhang, and Y. Feng, 2011: Anthropogenic influence on long return period daily temperature extremes at
regional scales. J. Climate, 24, 881–892.