15 NOVEMBER 2006 ZHOU AND YU 5843

Twentieth-Century Surface Air Temperature over China and the Globe Simulated by
Coupled Climate Models
TIANJUN ZHOU AND RUCONG YU
State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric
Physics, Chinese Academy of Sciences, Beijing, China

(Manuscript received 26 May 2005, in final form 27 February 2006)

ABSTRACT

This paper examines variations of the surface air temperature (SAT) over China and the globe in the
twentieth century simulated by 19 coupled climate models driven by historical natural and anthropogenic
forcings. Most models perform well in simulating both the global and the Northern Hemispheric mean SAT
evolutions of the twentieth century. The inclusion of natural forcings improves the simulation, in particular
for the first half of the century. The reproducibility of the SAT averaged over China is lower than that of
the global and hemispheric averages, but it is still acceptable. The contribution of natural forcings to the
SAT over China in the first half of the century is not as robust as that to the global and hemispheric
averages. No model could successfully produce the reconstructed warming over China in the 1920s. The
prescribed natural and anthropogenic forcings in the coupled climate models mainly produce the warming
trends and the decadal- to interdecadal-scale SAT variations with poor performances at shorter time scales.
The prominent warming trend in the last half of the century over China and its acceleration in recent
decades are weakly simulated. There are discrepancies between the simulated and observed regional
features of the SAT trend over China. Few models could produce the summertime cooling over the middle
part of eastern China (2736N), while two models acceptably produce the meridional gradients of the
wintertime warming trends, with north China experiencing larger warming. Limitations of the current
state-of-the-art coupled climate models in simulating spatial patterns of the twentieth-century SAT over
China cast a shadow upon their capability toward projecting credible geographical distributions of future
climate change through Intergovernmental Panel on Climate Change (IPCC) scenario simulations.

1. Introduction volcanic aerosols has improved the simulated SAT

variations and changes in several coupled climate mod-
Partly due to the promotion of Intergovernmental
els (Houghton et al. 2001, 496502). Some studies
Panel on Climate Change (IPCC) coupled model evalu-
aimed at the reproduction of twentieth-century climate
ation activities, the twentieth-century coupled model
have suggested that changes in solar irradiance are im-
integrations including major greenhouse gases and sul-
portant for reproducing the warming in the early part of
phate aerosols have been conducted worldwide at a
the century (Tett et al. 1999; Stott et al. 2000; Broccoli
number of research centers (Haywood et al. 1997;
et al. 2003; Meehl et al. 2003). However, another study
Emori et al. 1999; Boer et al. 2000; Dai et al. 2001a,b;
has suggested that this early warming can be solely ex-
Stott et al. 2001; Tett et al. 2002; Broccoli et al. 2003;
plained as a consequence of natural internal climate
Johns et al. 2003; Meehl et al. 2004). Many coupled
variability (Delworth and Knutson 2000). The IPCC
climate models are able to reproduce the major trends
Third Assessment Report (TAR) suggested that there
in twentieth-century global surface air temperature
was an urgent need for a systematic twentieth-century
(SAT) when driven by historical radiative forcing sce-
climate intercomparison project with a standard set of
narios. The inclusion of changes in solar irradiance and
forcings, including volcanic aerosols and changes in so-
lar irradiance. This appeal has been partly met through
the coordination of the Joint Scientific Committee/
Corresponding author address: Dr. Tianjun Zhou, LASG, In-
stitute of Atmospheric Physics, Chinese Academy of Sciences, Climate Variability and Predictability of the World Cli-
P.O. Box 9804, Beijing 100029, China. mate Research Program Working Group on Coupled
E-mail: zhoutj@lasg.iap.ac.cn Modeling for the IPCC Fourth Assessment Report

2006 American Meteorological Society

JCLI3952
5844 JOURNAL OF CLIMATE
(AR4). By the end of 2005, nearly 20 coupled climate
models worldwide have finished twentieth-century cli-
mate simulations. While great effort should still be de-
voted to the comparison of the observed and simulated
global mean temperatures, stricter standards should be
used in the model validations, including the twentieth-
century time evolution of regional average tempera-
ture. Acceptable performances of climate models in re-
producing the current state of regional climate are the
bases for developing credible geographical distributions
of future climate change through IPCC scenario simu-
lations (e.g., Kimoto 2005). Previous results based on
coupled climate models show considerable differences
in regional details (Giorgi et al. 2001). Observational
analyses indicated that the distinctive climate variation
over China could be attributed to multicontributions,
for example, a unique cloudradiative feedback (Yu et
al. 2004a), teleconnections (Yu and Zhou 2004; Li et al.
2005; Zhao et al. 2005; Xin et al. 2006), and the strato-
spheretroposphere exchange (Yu et al. 2004b). The
main motivation of this paper is to examine the extent
to which the twentieth-century coupled model integra-
tions can reproduce the observed major variations and
warming trend of SAT over China. The ability of the
models to reproduce the twentieth-century Chinese
temperature will serve as the basis for deriving uncer-
tainty measures for future climate change over East
Asia.
The outline of the paper is as follows. Section 2 in-
troduces the data and methods used in the analyses.
Section 3 starts with a description of the observational
variation of Chinese SAT, and follows with detailed
analyses of the twentieth-century SAT over China and
the globe simulated by coupled climate models. The
time-scale dependence of the simulated SAT evolution
and its recent warming trend are also addressed in sec-
tion 3. Section 4 presents a discussion. A summary is
given in section 5.
2. Data and analysis method
The published observational time series of annual
mean SAT of China covers the period from 1880 to
2002 (Wang et al. 1998; Wang and Gong 2000; Wang et
al. 2001). Because the modern large networks of
weather observing stations have started in 1951 in
China, in order to estimate annual mean temperature
series of China from 1880 to 1950, historical documen-
tary data together with several proxy data were used as
a supplement to instrumental records (Wang et al.
2001). Continental China was divided into nine regions
according to spatial correlation among the 1 1 grid- Li 1999). Assume that x is the time series of the global
ded annual mean SAT data from 1951 to 2002 obtained
VOLUME 19
from the Chinese Meteorological Administration.
Within each region, temperatures of different grid
points were correlated at the 5% significant level
(Wang et al. 1998). An area-weighted averaging was
used to derive the mean temperature of the whole of
continental China. The uncertainty range in annual
mean temperature of China is 0.27C for 18801910
and 0.16C for 191150 on average (Wang et al.
2001). The observational time series of global and
Northern Hemispheric mean temperatures are from
Jones et al. (1999) and Jones and Moberg (2003). The
anomalies were calculated relative to the 196190
mean. The monthly SAT data from 160 stations for
195199 compiled by the Chinese Meteorological Ad-
ministration are also used. These station SAT data have
been widely used in studies of the East Asian climate
(e.g., Hu et al. 2003; Zhou and Yu 2005).
The twentieth-century climate simulations by coupled
climate models used for the IPCC AR4 are often called
by the project name: The Twentieth-Century Climate
in Coupled Models (20C3M). The twentieth-century
climate simulations were made with various combina-
tions of forcings including greenhouse gases (GHGs),
sulfate aerosols, ozone, volcanic aerosols, and solar
variability. The focus of this paper is the SAT during
18801999, over which 19 models provided data. The
names of the models analyzed are listed in Table 1,
together with the natural and anthropogenic forcings
used by the different modeling groups in their IPCC
20C3M simulations. Table 1 is compiled using informa-
tion that the participating modeling centers provided to
the U.S. Department of Energys Program for Climate
Model Diagnosis and Intercomparison (PCMDI) (see
http://www-pcmdi.llnl.gov/ipcc/about_ipcc.php for
more information about the models). Since these mod-
els employ different horizontal resolutions, the Chi-
nese-domain-averaged SAT in the simulations is de-
fined as the area-weighted average of three rectangular
boxes: (2850N, 8097.5E), (22.543N, 97.5
122.5E), and (4354N, 117.5130E). To further dis-
cuss the regional difference, we consider east China as
two parts: north China (3543N, 102.5122.5E), and
south China (22.535N, 102.5122.5E). Regional
time series is calculated as the area-weighted average
for each box.
In the following analyses, linear regression is used in
the detrending and the estimation of trends. To quan-
titatively reveal the spread and convergence of the cli-
mate models in reproducing the twentieth-century
SAT, one widely used technique, namely, analysis of
variance, is employed (Harzallah and Sadourny 1995;
mean SAT, x(n, t). Here x(n, t) represents the nth mem-
15 NOVEMBER 2006 ZHOU AND YU 5845

TABLE 1. Forcings used in IPCC simulations of twentieth-century climate: SO solar irradiance, LU land use change, VL
volcanic aerosols, BC black carbon, SD Sulfate aerosol direct effects, OC organic carbon, SI sulfate aerosol indirect effects,
MD mineral dust, GHG well-mixed greenhouse gases, SS sea salt, and O tropospheric and stratospheric ozone. A letter Y
denotes inclusion of a specific time-varying forcing, with changes on interannual and longer time scales. Forcings that were varied over
the seasonal cycle only, or not at all, are identified with a dash. A question mark indicates a case where there is uncertainty regarding
inclusion of the forcing.

Model SO VL SD SI GHG O LU BC OC MD SS
BCC-CM1a Y Y Y
BCCR-BCM2 Y Y
CGCM3.1(T47) Y Y Y
CGCM3.1(T63) Y Y Y
CCSM3 Y Y Y Y Y Y
CNRM-CM3 Y Y Y Y ? Y Y
CSIRO-Mk3.0 Y Y ? ? ?
ECHAM5/MPI-OM Y Y Y Y
FGOALS-g1.0 Y Y Y
GFDL-CM2.0 Y Y Y Y Y Y Y
GFDL-CM2.1 Y Y Y Y Y Y Y
GISS_AOM Y Y Y
GISS_ER Y Y Y Y Y Y Y Y Y Y
INM-CM3.0b Y Y Y Y Y
IPSL-CM4 Y Y Y
MRI-CGCM2.3.2 Y Y Y Y
MIROC3.2(medres)c Y Y Y ? Y Y Y Y Y Y Y
UKMO-HadCM3 Y Y Y Y
UKMO-HadGEM1d Y Y Y Y Y Y Y Y Y
a
BCC-CM1 Beijing Climate Center Coupled Model version 1.
b
INM-CM3.0 Institute of Numerical Mathematics Coupled Model version 3.0.
c
MIROC3.2(medres) Model for Interdisciplinary Research on Climate 3.2, medium-resolution version.
d
UKMO-HadGEM1 Met OfficeHadley Centre Global Environmental Model version 1.

ber among an ensemble size of N for the tth year of a and

simulation of length T. Two averages are defined: en- T N

semble mean and climatological mean. The multimodel 1 1
i2 xn, t xet2 . 4
ensemble mean is taken with respect to the simulation T t1 N n1
count N:
It is easy to show that the total variance, defined by
N the squared total standard deviation,

1
xet xn, t, 1
N n1 N T

xn, t x ,
1
t2 c
2
5
and the climatological mean is taken with respect to the NT n1 t1
numbers of years T and simulations N: is the sum of the external and internal variance:
N T T

xn, t T x t.
1 1 t2 2e i2. 6
xc e 2
NT n1 t1 t1 The signal-to-noise ratio is defined as
The interannual variability can be represented by the e
year-to-year variation of the ensemble means xe(t). The . 7
i
standard deviation e of the xe(t) is used to measure the
external, or forced, signal. The dispersion of the simu- To estimate the difference of the SAT trends among
lation [measured by the standard deviation i of x(n, t)] single-model ensemble runs, the spread is defined as

indicates the intermodel variability, which is noise for
the twentieth-century climate reproduction. From the N

1
definition of the two averages, we have yn yc2, 8
N n1
T

x t x
1 where y is the trend of JuneAugust (JJA) or Decem-
2e e c
2
3
T t1 berFebruary (DJF)] mean SAT. Here y(n) represents
5846 JOURNAL OF CLIMATE
FIG. 1. Observed variations from 1880 to 1999 of the annual
mean China-wide (solid line), global (long dashed line), and
Northern Hemispheric (short dashed line) average surface air
temperature anomalies in units of C.
the nth realization among an ensemble size of N. The
ensemble mean yc is taken with respect to the simula-
tion count N.
3. The evolution of twentieth-century global and
Chinese SAT
a. Global annual mean SAT in coupled climate
models
Figure 1 shows the history of the Chinese tempera-
ture index (CTI), the global and Northern Hemispheric
mean SAT for 18801999. The warming in China gen-
erally parallels the global average with a correlation
between them of 0.60 for 18801999. From 1880 to
1910, the CTI is below that of the later decades. There
is a broad maximum from the 1920s through the 1940s,
a decrease from the late 1940s to late 1970s, and an-
other rising trend from the late 1970s to the present. Of
particular interest are the two warming peaks in the
1920s and 1940s. The warming of the 1920s is not as
apparent as in the time series of both the global and the
Northern Hemispheric average. Further comparison re-
vealed that north China has experienced a similar evo-
lution as south China except that north China is stron-
ger in amplitude (figure not shown).
To be consistent with the observational index, the
model anomalies here are calculated relative to the
196190 mean. The upward trends of both the multi-
FIG. 2. The (a) global, (b) Northern Hemispheric, and (c) Chinese average surface air temperature anomalies simulated by the 20C3M
models. Each thin color line corresponds to run1 of each model. CSIRO-Mk3.0 and FGOALS-g1.0 are excluded from the ensemble
mean of global and Northern Hemispheric average. CSIRO-Mk3.0, UKMO-HadCM3, and FGOALS-g1.0 are excluded from the
ensemble mean of Chinese average.
VOLUME 19
model ensemble global and the Northern Hemispheric
average SAT (Figs. 2a,b) are broadly comparable to the
observations, but with smaller interannual to decadal
variations. There is also a large spread among the
model results, especially during the early part of the
period. Correlation coefficients between the observed
and the simulated SAT time series for different models
are listed in Table 2. The correlations for the global
average are generally higher than those for the North-
ern Hemispheric average. Nearly all models have high
correlations with the observations except for the Com-
monwealth Scientific and Industrial Research Organi-
sation Mark version 3.0 (CSIRO-Mk3.0) model and the
Flexible Global OceanAtmosphereLand System
Model gridpoint version 1.0 (FGOALS-g1.0). If these
two models were eliminated from the statistics, the en-
semble mean of the other 17 models would have a high
correlation of 0.87 (0.82) with the observed global
(Northern Hemispheric) mean SAT.
The 20C3M simulations started from different points
of preindustrial control runs. A one-by-one check of
the models indicates that climate drifts in the control
run have negligible impacts on the results of most mod-
els except for FGOALS-g1.0, which suffers from a
strong cooling trend in the first 60 yr. The reason for the
drift is that the 20C3M integration of FGOALS-g1.0
began without the preindustrial spinup process. Since it
started from a much higher initial sea ice cover, the
20C3M run shows a rapid increase of sea ice area from
1850 to 1920 (Zhang and Walsh 2006). The Northern
Hemispheric mean SAT of CSIRO-Mk3.0 is dominated
by large multidecadal oscillations with minima in the
1920s and 1950s. Since no natural forcings were added
during the 20C3M integration in this model, the mul-
tidecadal oscillation around the 1920s and 1950s can be
regarded as internal variability in CSIRO-Mk3.0. These
two models are excluded in the following construction
of the multimodel ensembles.
Analyses of variance for the time series of globally
averaged annual mean SAT from 1880 to 1999 show
that the dispersion (external forcing) accounts for
39.5% (60.5%) of the total variance. The forced signal
is larger than the intermodel noise, with a signal-to-
noise ratio of 1.2. Estimation of the time series of
Northern Hemispheric average SAT indicates that the
external forcing (dispersion) accounts for 55.4%
15 NOVEMBER 2006 ZHOU AND YU 5847
5848 JOURNAL OF CLIMATE VOLUME 19

TABLE 2. Correlation coefficients between the simulated and the observed annual SAT time series from 1880 to 1999. Models without
inclusion of time-varying natural forcings are identified with italic fonts. Bold numbers are statistically significant at the 5% level. Only
run1 of each model is analyzed. CSIRO-Mk3.0 and FGOALS-g1.0 are not included in the global and Northern Hemispheric ensemble
means. CSIRO-Mk3.0, FGOALS-g1.0, and UKMO-HadCM3 are not included in the whole, north, and south China ensemble means.

Global Northern Hemispheric Whole of China North China South China

Model average average average average average
BCC-CM1 0.85 0.75 0.35 0.20 0.11
BCCR-BCM2 0.49 0.57 0.35 0.07 0.05
CGCM3.1(T47) 0.81 0.71 0.34 0.22 0.12
CGCM3.1(T63) 0.81 0.73 0.24 0.18 0.06
CCSM3 0.82 0.78 0.36 0.29 0.20
CNRM-CM3 0.77 0.67 0.42 0.29 0.15
CSIRO-Mk3.0 0.32 0.21 0.06 0.13 0.10
ECHAM5/MPI-OM 0.62 0.53 0.33 0.09 0.23
FGOALS-g1.0 0.23 0.18 0.14 0.12 0.22
GFDL-CM2.0 0.80 0.80 0.39 0.08 0.18
GFDL-CM2.1 0.70 0.68 0.48 0.31 0.33
GISS_AOM 0.82 0.71 0.40 0.16 0.14
GISS_ER 0.80 0.66 0.32 0.23 0.07
INM-CM3.0 0.75 0.65 0.30 0.07 0.14
IPSL-CM4 0.75 0.65 0.29 0.19 0.19
MRI-CGCM2.3.2 0.79 0.74 0.44 0.26 0.17
MIROC3.2(medres) 0.77 0.65 0.27 0.10 0.06
UKMO-HadCM3 0.59 0.46 0.13 0.05 0.13
UKMO-HadGEM1 0.77 0.63 0.26 0.29 0.22
Ensemble mean 0.87 0.82 0.55 0.38 0.33

(44.5%) of the total variance, with a signal-to-noise ra-
tio of 1.1.
The dispersion among the models partly comes from
differences of forcing agents used in driving simulations
of the twentieth-century climate. The natural forcings
(time-varying solar irradiance and volcanic aerosols)
were not included in several models (Table 1). Large,
different forcing agents prescribed in the 20C3M simu-
lations make direct comparisons among the simulations
difficult. The inclusion of changes in solar irradiance
and volcanic aerosols, however, clearly has improved
the simulations (Table 2).
Previous climate model experiments have suggested
that the early twentieth-century warming could have
been primarily due to natural forcings, while the late
twentieth-century warming was likely mainly due to in-
creases of greenhouse gases (e.g., Stott et al. 2001;
Meehl et al. 2003, 2004; Nozawa et al. 2005). We sepa-
rate 18801999 into two periods: 18801940 and 1941
99. Statistics corresponding to Table 2 are listed sepa-
rately for these two periods in Tables 3 and 4 . For the
early part of the twentieth century, without the inclu-
sion of natural forcings, barely any model can reason-
ably simulate the observed variations [see the statistics
for Bjerknes Centre for Climate Research Bergen Cli-
mate Model version 2 (BCCR-BCM2), CSIRO-Mk3.0,
ECHAM5/Max Planck Ocean Model (MPI-OM),
LInstitut Pierre-Simon Laplace Coupled Model ver-
sion 4 (IPSL-CM4), and the Met Office Third Hadley
Centre Coupled OceanAtmosphere General Circula-
tion Model (UKMO-HadCM3) in Table 3]. Although
the only yearly changes of the forcings in the Goddard
Institute for Space Studies AtmosphereOcean Model
(GISS_AOM) were PCMDIs greenhouse gases and
the linear interpolation of the Boucher tropospheric
sulfate burden multiplied by a single global constant to
convert it to an optical depth (G. L. Russell 2005, per-
sonal communication), this model still has an accept-
able result. Nevertheless, this does not mean that inter-
nal variability plays dominant roles in regulating the
SAT before 1940, because the acceptable correlation
comes from the warming trends. After linear detrend-
ing, no significant correlation is found (Table 3), which
demonstrates the important contribution of natural
forcings from another aspect. Similar conditions are
found for the Canadian Centre for Climate Modelling
and Analysis (CCCma) Coupled General Circulation
Model version 3.1 [CGCM3.1(T47) and CGCM3.1(T63)],
and Centre National de Recherches Mtorologiques
Coupled Global Climate Model version 3 (CNRM-CM3)
models.
For the late twentieth-century warming, all models
have high significant correlations with the observations
for both the global and Northern Hemispheric averages
(Table 4). By comparing runs with or without the GHG
forcings, Stott et al. (2000) showed the important con-
15 NOVEMBER 2006 ZHOU AND YU 5849

TABLE 3. Correlation coefficients between the simulated and the observed annual SAT time series from 1880 to 1940. Models without
inclusion of time-varying natural forcings are identified with italic fonts. Values enclosed in parentheses correspond to the correlation
after linear detrending. Bold numbers are statistically significant at the 5% level. Only run1 of each model is analyzed. Models
without time-varying natural forcings, and CSIRO-Mk3.0 and FGOALS-g1.0, are not included in the ensemble mean.

Model Global average Northern Hemispheric average Whole of China average

BCC-CM1 0.79 (0.74) 0.64 (0.57) 0.23 (0.19)
BCCR-BCM2 0.05 (0.1) 0.19 (0.13) 0.48 (0.21)
CGCM3.1(T47) 0.45 (0.17) 0.47 (0.20) 0.17 (0.04)
CGCM3.1(T63) 0.35 (0.08) 0.24 (0.13) 0.20 (0.11)
CCSM3 0.47 (0.35) 0.47 (0.43) 0.31 (0.22)
CNRM-CM3 0.45 (0.24) 0.35 (0.24) 0.41 (0.23)
CSIRO-MK3.0 0.10 (0.15) 0.03 (0.06) 0.06 (0.09)
ECHAM5/MPI-OM 0.14 (0.05) 0.13 (0.01) 0.12 (0.07)
FGOALS-g1.0 0.33 (0.08) 0.47 (0.19) 0.34 (0.06)
GFDL-CM2.0 0.57 (0.39) 0.62 (0.40) 0.32 (0.22)
GFDL-CM2.1 0.40 (0.12) 0.45 (0.11) 0.55 (0.03)
GISS_AOM 0.40 (0.07) 0.24 (0.03) 0.24 (0.06)
GISS_ER 0.52 (0.32) 0.47 (0.26) 0.23 (0.12)
INM-CM3.0 0.30 (0.09) 0.25 (0.05) 0.14 (0.15)
IPSL-CM4 0.10 (0.22) 0.03 (0.19) 0.06 (0.07)
MRI-CGCM2.3.2 0.34 (0.03) 0.41 (0.06) 0.37 (0.19)
MIROC3.2(medres) 0.48 (0.34) 0.27 (0.30) 0.17 (0.14)
UKMO-HadCM3 0.05 (0.10) 0.09 (0.16) 0.14 (0.12)
UKMO-HadGEM1 0.43 (0.30) 0.31 (0.21) 0.08 (0.05)
Ensemble mean 0.64 (0.51) 0.64 (0.45) 0.61 (0.17)

tribution of GHG to the warming over recent decades. radiative forcing in simulating the observed warming
Using a simple climate/ocean model with prescribed during recent decades. The high correlations in Table 4
climate sensitivity, Andronova and Schlesinger (2000) might result from the inclusion of GHG forcings. Since
also demonstrated the importance of the anthropogenic we only have results that show improved correlations

TABLE 4. Correlation coefficients between the simulated and the observed SAT time series for 194199. Models without inclusion of
time-varying natural forcings are identified with italic fonts. Values enclosed in parentheses correspond to the correlation after
detrending. Bold numbers are statistically significant at the 5% level. Only run1 of each model is analyzed.

Global Northern Hemispheric Whole of China North China South China

Model average average average average average
BCC-CM1 0.74 (0.73) 0.69 (0.68) 0.27 (0.36) 0.18 (0.17) 0.03 (0.14)
BCCR-BCM2 0.62 (0.31) 0.51 (0.29) 0.06 (0.08) 0.05 (0.05) 0.04 (0.02)
CGCM3.1(T47) 0.69 (0.39) 0.52 (0.30) 0.25 (0.30) 0.23 (0.26) 0.11 (0.27)
CGCM3.1(T63) 0.72 (0.55) 0.61 (0.58) 0.05 (0.14) 0.07 (0.04) 0.08 (0.06)
CCSM3 0.67 (0.44) 0.63 (0.46) 0.21 (0.13) 0.19 (0.18) 0.19 (0.25)
CNRM-CM3 0.52 (0.0) 0.44 (0.13) 0.27 (0.25) 0.18 (0.18) 0.08 (0.01)
CSIRO-Mk3.0 0.53 (0.02) 0.38 (0.02) 0.10 (0.21) 0.15 (0.19) 0.11 (0.05)
ECHAM5/MPI-OM 0.62 (0.36) 0.44 (0.33) 0.32 (0.27) 0.05 (0.04) 0.21 (0.29)
FGOALS-g1.0 0.54 (0.20) 0.41 (0.21) 0.07 (0.07) 0.05 (0.03) 0.15 (0.08)
GFDL-CM2.0 0.56 (0.38) 0.60 (0.35) 0.27 (0.24) 0.11 (0.12) 0.01 (0.02)
GFDL-CM2.1 0.55 (0.37) 0.60 (0.39) 0.28 (0.22) 0.25 (0.25) 0.25 (0.26)
GISS_AOM 0.72 (0.52) 0.60 (0.43) 0.40 (0.22) 0.19 (0.18) 0.02 (0.14)
GISS_ER 0.64 (0.42) 0.45 (0.28) 0.34 (0.28) 0.18 (0.19) 0.08 (0.05)
INM-CM3.0 0.47 (0.23) 0.36 (0.18) 0.21 (0.16) 0.03 (0.04) 0.05 (0.07)
IPSL-CM4 0.67 (0.35) 0.56 (0.32) 0.41 (0.33) 0.38 (0.38) 0.43 (0.46)
MRI-CGCM2.3.2 0.66 (0.33) 0.58 (0.41) 0.36 (0.29) 0.26 (0.26) 0.12 (0.19)
MIROC3.2(medres) 0.65 (0.41) 0.63 (0.50) 0.17 (0.40) 0.09 (0.07) 0.07 (0.06)
UKMO-HadCM3 0.66 (0.30) 0.55 (0.38) 0.08 (0.00) 0.02 (0.03) 0.02 (0.04)
UKMO-HadGEM1 0.77 (0.58) 0.73 (0.59) 0.48 (0.39) 0.42 (0.43) 0.37 (0.41)
Ensemble mean 0.78 (0.72) 0.69 (0.70) 0.42 (0.64) 0.36 (0.46) 0.17 (0.45)
5850 JOURNAL OF CLIMATE
with observed SAT when GHG forcings are included in
models, these results are suggestive, but do not prove
that these forcings are important for reproducing the
observed SAT time series. Further studies need to re-
peat Stott et al. (2000) and compare with runs without
these forcings.
b. SAT averaged over China in coupled climate
models
The SAT anomalies averaged over China from
20C3M are shown in Fig. 2c. Large spreads are found
prior to the late 1970s, and the warming during the
1920s1930s is not simulated well by the models. After
the late 1970s, the multimodel ensemble mean con-
verges toward the observations. Most models have sig-
nificant correlations with the observations except
for the UKMO-HadCM3, FGOALS-g1.0, and CSIRO-
Mk3.0 models (Table 2). The reasons for the shortcom-
ings of the CSIRO-Mk3.0 and FGOALS-g1.0 models
have been discussed, and that for the UKMO-HadCM3
model appears to be a lack of time-varying natural forc-
ings. The ensemble mean of the other 16 models has a
correlation of 0.55 with the observations from 1880 to
1999, which is lower than those for the global and
Northern Hemispheric averages, but it is still statisti-
cally significant at the 5% level. The correlation for
north China is slightly higher than south China (0.38
versus 0.33). The external forcing (dispersion) accounts
for 32.5% (67.5%) of the total variance of the Chinese
SAT during 18801999, with a signal-to-noise ratio of
0.69. The spread of Chinese SAT evolution among the
models is, as shown by Figs. 2ac, larger than that for
the global and Northern Hemispheric averages.
Paralleling the global and Northern Hemispheric av-
erages, the Chinese SAT in the first half of the twenti-
eth-century is also significantly modulated by GHG
forcings. The contribution of natural forcings seems to
be not as robust as that for the global and hemispheric
averages. After removing the linear warming trend, no
statistically significant correlations at the 5% level is
found (Table 3). In the last half of the century, both the
GHG forcing and the other forcings such as the anthro-
pogenic aerosols affect the Chinese SAT. Even after
removing the warming trends, significant correlations
can still be found between the multimodel ensemble
mean and the observations, and 8 of the 19 models have
significant correlations with the observations (Table 4).
Six models have low correlations with the observations
in the period of 194199. Among them, the BCCR-
BCM2, CGCM3.1(T63), CSIRO-Mk3.0, and UKMO-
HadCM3 models did not include time-varying natural
forcings. Since the ECHAM5/MPI-OM and IPSL-CM4
models, which did not include natural forcings, have
VOLUME 19
FIG. 3. Same as Fig. 2 except for the ensemble mean of CNRM-
CM3, GFDL-CM2.1, GISS_AOM, and MRI-CGCM2.3.2 models
(long-dashed line), and the single GFDL-CM2.1 model (short-
dashed line). The observation is shown as a solid line.
acceptable performances in this period (Table 4), we
cannot draw a conclusion that the natural forcings play
significant roles in modulating the recent half-century
variation of the SAT over China. Andronova and
Schlesinger (2000) showed that solar radiation contrib-
utes a small warming in recent decades. The multimo-
del model ensemble mean of north China SAT from
1941 to 1999 has a higher correlation with the observa-
tions than that of south China (0.36 versus 0.17). After
removing the linear trends, however, the results are
comparable (0.46 versus 0.45). The UKMO-HadGEM1
and IPSL-CM4 models have good performances in
simulating the SAT variation over both north and south
China. The result of Geophysical Fluid Dynamics
Laboratory Climate Model version 2.1 (GFDL-CM2.1)
is also acceptable.
Four models [CNRM-CM3, GFDL-CM2.1, GISS_
AOM, and Meteorological Research Institute Coupled
Global Climate Model version 2.3.2 (MRI_CGCM2.3.2)]
have correlations exceeding 0.40 from 1880 to 1999 for
the SAT averaged over China (Table 2). The GFDL-
CM2.1 model has the highest correlation. Figure 3
shows the evolution of annual SAT averaged over
China for the four-model ensemble and the GFDL-
CM2.1 together with the observations. The observed
peak warming in the 1940s is not seen in the ensemble
mean, although the general warming is captured from
the 1910s to the 1930s. The GFDL-CM2.1 model shows
a peak in the mid-1930s, which lies in the center of the
two observational peaks. Further model-by-model ex-
aminations also reveal that no model can reasonably
simulate the warming of China in the 1920s.
Multimodel ensemble clearly improves the simula-
tion. The improvement of using ensembles is also clear
15 NOVEMBER 2006 ZHOU AND YU
in the single-model simulations. Since the Community
Climate System Model version 3 (CCSM3) model has
an ensemble member number of eight, correlation co-
efficients with the observations for each single realiza-
tion from 1880 to 1999 are calculated. Among the eight
realizations, correlations for the global (Northern
Hemispheric) mean SAT range from 0.79 to 0.87 (0.69
to 0.82), and those for the SAT averaged over China
range from 0.23 to 0.44. The ensemble mean is better
than any single realization. The correlation between the
observation and the ensemble mean from 1880 to 1999
is 0.87 for global, 0.82 for Northern Hemispheric, and
0.49 for Chinese averages. Analysis of variance indi-
cates that the external forcing (internal noise) accounts
for 89.5% (10.5%) of the total variance for the global
mean SAT from 1880 to 1999, with a signal-to-noise
ratio of 2.9. The corresponding statistics for the Chinese
SAT are 51.3%, 48.7%, and 1.03. Analyses using data
after detrending get similar results, indicating the im-
proved correlation does not come purely from the im-
proved correlation between the trends.
Using ensembles improves the correlation as random
variations are reduced during the averaging. Note
higher correlation only indicates a better phase rela-
tionship, not necessarily a good simulation of the SAT
values. Working with ensembles increases the correla-
tion, but decreases the amplitude. As shown in Figs.
2ac, the amplitudes for multimodel ensemble means
are too weak.
c. SAT averaged over China in SST-forced
simulations
As a comparison, we force the National Center for
Atmospheric Research (NCAR) Community Atmo-
sphere Model version 2 (CAM2; Collins et al. 2003)
with the monthly historical sea surface temperature
(SST) and sea ice from 1870 to 2003. An ensemble run
with 12 members was carried out. The observational
SST and sea ice data used to force the model are the
Global Ocean Surface Temperature climatologies
(GISST, version 2.3b), which are a set of SST data in
monthly 1 1 area grids for 1870 to 2003, provided
by the Met Office of the United Kingdom (Rayner et al.
2003). The time evolutions of the SAT averaged over
China for 18801999 simulated by the CAM2 are shown
in Fig. 4. The ensemble mean resembles the observa-
tions in the cooling of the early twentieth century, the
warming of the middle portion, and the rising trend
from the late 1970s except with weaker amplitudes. The
correlations between each single realization and the ob-
servations for 18801999 range from 0.25 to 0.44, which
are statistically significant at the 5% level. The corre-
lation for the ensemble mean is 0.50, which is compa-
5851
FIG. 4. Ensemble simulation of the annual mean Chinese sur-
face air temperature anomalies using the CAM2 model forced by
historical SST. Each thin line corresponds to one single realiza-
tion, and the thick black line corresponds to the ensemble mean.
rable to the results of the coupled climate models. A
great part of the observed SAT variation over China
can be reproduced by the SST forcing. Since both the
SST and the SAT changes ultimately result from the
GHG and other forcings, the CAM2 results indicate the
role of the ocean in remembering the forced signals.
The external forcing (dispersion) accounts for 37.5%
(62.5%) of the total variance, with a signal-to-noise ra-
tio of 0.77. The external forcing here comes from the
SST. The reconstructed peak warming of the 1920s is
not evident in any of the single realizations. Consider-
ing the failure of coupled climate models in this regard,
the explanation of the warming over China in the 1920s
remains an open problem.
The reliability of the reconstructed SAT over China
before 1950 needs validation. One might argue the pos-
sibility that the reconstructed 1920s warming is artifact.
Zhao et al. (2005) compared four reconstructed time
series of the annual mean SAT in China for the twen-
tieth century and found that the 1920s warming is ro-
bust in three of them. A recent study showed that the
time series of Wang et al. (2001) has a high correlation
of 0.84 from 1900 to 2002 with the time series based on
Climatic Research Unit (CRU) data, while the 1920s
warming also exists in the CRU data-based time series,
it is weaker than that of Wang et al. (2001). The stron-
ger amplitude of 1920s warming in Wang et al. (2001)
resulted from the inclusion of ice core data over the
Tibetan Plateau (validation of the CRU data over
China; S. W. Wang and X. Y. Wen 2006, personal com-
munication).
The warming in the 1920s is not a unique feature
observed in China. Rather, it is a local manifestation of
a much larger hemispheric-scale SAT variation. The
occurrence of the rapid global warming in the 1920s has
already been documented (Diaz and Quayle 1980; Rog-
5852 JOURNAL OF CLIMATE
FIG. 5. The local wavelet power spectrum of (a) the observed
and (b) the simulated global mean SAT anomalies using the Mor-
let wavelet. The left axis is the Fourier period (in years). The
bottom axis is time (years). The simulation is an ensemble mean
of nine models. The SAT time series were normalized prior to the
wavelet analyses.
ers 1985). Fu et al. (1999) found that both middle and
high latitudes of the Northern Hemisphere had experi-
enced a rapid warming in the 1920s. Andronova and
Schlesinger (2000) stated that the sun played only a
minor role in this process. Schlesinger and Ramankutty
(1994) suggested that the 1920s warming was the result
of a temperature oscillation with a period of 6570 yr.
d. Time-scale dependence of the simulated SAT
evolution
Figures 14 exhibit prominent linear trends with
large decadal variations. To reveal the time scale de-
pendence of the simulated SAT, we performed a Mor-
let wavelet analysis (Torrence and Compo 1998). The
wavelet spectra for the global mean SATs are shown in
Fig. 5. In the observations, one can find the powers
concentrating within the bands of interannual, decadal,
or even longer interdecadal time scales (Fig. 5a).
Mainly powers concentrated within interdecadal time
scales can be found in the simulations (Fig. 5b). Note
the construction of the multimodel ensemble mean
used only nine models with time-varying natural forc-
ings.
The wavelet spectra for the twentieth-century SATs
averaged over China are shown in Fig. 6. Strong inter-
annual-, decadal-, and interdecadal-scale fluctuations
are seen in the power spectrum of the CTI index (Fig.
6a). In the 20C3M multimodel ensemble, mainly the
fluctuations on interdecadal time scales are reasonably
VOLUME 19
FIG. 6. Same as Fig. 5 except for the SAT averaged over China.
(a) Observation, (b) ensemble mean of nine coupled climate mod-
els, and (c) CAM2.
reproduced (Fig. 6b). The power concentrated within
the band less than 10 yr, which it is more prominent
from 1930 to 1999, is observed in the observational
SST-forced simulation (Fig. 6c).
To examine the correlation at different time scales,
we band-filtered the SAT time series into three parts:
the linear trend, variations longer than 10 yr, and varia-
tions shorter than 10 yr. For the global average, the
unfiltered time series of the nine-model ensemble mean
has a correlation of 0.88 with the observations from
1880 to 1999; after linear detrending, the correlation
decreases to 0.59; at time scales longer (shorter) than 10
yr, the correlation is 0.80 (0.30). For the SAT averaged
over China, the corresponding statistics are 0.56, 0.41,
and 0.54 (0.26), respectively. All of these values are
statistically significant at the 5% level. These results
suggest that compared with the interannual fluctua-
tions, the decadal and interdecadal SAT variations are
more closely linked to the prescribed external forcings.
Note in above analyses, all trends are calculated from
linear regressions. Since neither the SAT trends nor the
GHG trends are strictly linear, if one want to remove
the GHG-induced warming, one should use the GHG
forcing times series to detrend the SAT. In the model
15 NOVEMBER 2006 ZHOU AND YU
TABLE 5. Linear trends of annual mean SAT in the observations and the simulations. Run1 of nineteen 20C3M models are used
in the construction of multimodel ensembles.
OBS [C (100 yr)1]
Global (18801999) 0.53
Global (191045) 1.47
Global (198599) 2.35
Northern Hemisphere (18801999) 0.54
Northern Hemisphere (191045) 1.66
Northern Hemisphere (198599) 3.34
China (18801999) 0.52
China (191045) 2.81
China (198599) 7.01
simulations, however, the GHG forcing is a combina-
tion of several greenhouse gases including at least the
carbon dioxide, methane, nitrous oxide, and chloroflu-
orocarbon. These forcing data are not available in the
IPCC data archive. Approximately, we make detrend-
ing by subtracting the third-order polynomial from the
original time series of both observed and simulated
SATs. The residuals are mainly interannual and dec-
adal to interdecadal variabilities. Compared with the
linear detrending, the correlation after the nonlinear
detrending between the observation and the multimo-
del ensemble for the global mean SAT from 1880 to
1999 decreased from 0.59 to 0.55, and for Chinese SAT
it decreased from 0.41 to 0.39. On the decadal time
scale, the correlation for detrended global (Chinese)
SAT decreased from 0.80 (0.54) to 0.78 (0.53). On the
interannual time scale, the corresponding results for
global (Chinese) SAT are from 0.30 to 0.29 (0.26 to
0.24). All of these values are still statistically significant
at the 5% level, indicating that the detrending method
does not change our conclusion.
e. SAT trends over different periods
The most prominent feature of the twentieth-century
SAT variation is its warming trend (Fig. 1). To quanti-
tatively evaluate the models performances in this re-
gard, the trends of the SAT in the observations and the
simulations for three different time periods are esti-
mated by using linear regression (Table 5). For the
warming trends starting from 1880 to 1999, the 20C3M
simulations are 0.14, 0.14, and 0.24C per century
higher than the observations in the global, Northern
Hemispheric, and Chinese averages, respectively. For
the warming trends in the early part of the century
(191045), however, the 20C3M simulations are 0.92,
1.08, and 2.19C per century lower than the observa-
tions. The observed accelerations of the warming trend
from 1985 to 1999 are also weakly simulated in the
20C3M. Three models reproduced warming rates com-
5853
20C3M [C (100 yr)1] CAM2 [C (100 yr)1]
0.67
0.55
1.75
0.68
0.58
2.04
0.76 0.26
0.62 1.36
2.51 1.40
parable to the observations for the SAT over China
from 1985 to 1999: the GFDL-CM2.1 has a warming
rate of 7.6C per century, and the GISS_ER and
UKMO-HadGEM1 have rates of 6.4 and 7.8C per
century, respectively. The observed trends over China
for the analyzed three time periods are weakly repro-
duced in the SST-forced CAM2 simulation.
Since there is a better coverage of observational sta-
tions in eastern China (east of 100E), this domain is
selected for evaluating model-simulated spatial pat-
terns. The trends of annual mean SAT for 195199
from the observations and the simulations are esti-
mated by using linear regression (Fig. 7). In the obser-
vations, there are cooling trends of 0.31.5C (50
yr)1 in southwestern China (downstream of the Ti-
betan Plateau) and warming trends elsewhere, in par-
ticular north China (Fig. 7a). The maximum warming
signal in northeast China is above 2.0. Nine of the
models show warming trends over the whole eastern
China. Four of the models incorrectly show cooling
trends over north China. The cooling over southwest-
ern China is only weakly simulated in three models.
While most models have the ability to approximately
simulate the observed evolution of Chinese mean SAT,
most of them failed to simulate the spatial patterns, in
particular the cooling in southwestern China.
The trend of Chinese SAT varies seasonally and re-
gionally (Hu et al. 2003). We further analyze the linear
trends of SAT in summer (JJA) and winter (DJF) by
linear regressions. The zonal mean of the JJA SAT
trend for 19511999 averaged over eastern China
(102.5122.5E) from the observations and the simu-
lations are shown in Fig. 8a. The middle part of eastern
China (2736N) experienced a moderate summertime
cooling. The warming south of 27N is stronger than
that north of 36N. All models except the GFDL-
CM2.1 failed to reproduce the cooling trend, with large
spread among them. The amplitude of the cooling in
GFDL-CM2.1 is close to the observations [0.5C
5854 JOURNAL OF CLIMATE VOLUME 19
15 NOVEMBER 2006 ZHOU AND YU 5855

FIG. 8. Zonal mean of the (a) JJA and (b) DJF SAT linear
trends (195199) over east China (102.5122.5E) for run1 of
the different models. The observation is shown as thick black line
[Units: C (50 yr)1].

(50 yr)1]. Results for the DJF SAT are given in Fig. 8b.
The warming trend over eastern China increases from
24 to 40N. Several models [e.g., CGCM3.1(T47) and
HadCM3] capture the meridional gradient qualita-
tively. However, many models underestimate the DJF FIG. 9. The spread of (a) JJA and (b) DJF SAT linear trends
warming trend, with cooling in a number of models (195199) among eight realizations of CCSM3 [Units: C (50
north of 35N. The spread is also much larger north of yr)1]. Values greater than 0.5 are shaded.
30N than south of it.
In above discussions, only the first realization of each lation, spreads of this model in simulating the trends of
model is analyzed. The spread among single-model en- JJA and DJF SAT are calculated according to the Eq.
semble runs should be estimated. Since the CCSM3 (8). The results are shown in Fig. 9. The spread of the
model has eight ensemble members in the 20C3M simu- SAT trends in north China is larger than that in south

FIG. 7. The distribution of the linear trend of annual mean SAT for 195199 in a single realization of the 20CM3 simulation by
different models and the observation [Units: C (50 yr)1]. Negative trends are shaded.
5856 JOURNAL OF CLIMATE
China. This condition exists not only in summer but
also in winter. North China is more strongly disturbed
by the internal noise, which is consistent with the large
spread in north China shown in Fig. 8.
4. Discussion
When driven by historical natural and anthropogenic
forcings, while current state-of-the-art coupled climate
models generally perform acceptably in simulating the
major variations of the SAT averaged over China, there
are still large deficiencies in simulating the spatial dis-
tributions. This is particularly true for the southwestern
China, namely, downstream of the Tibetan Plateau
(Figs. 7 and 8). The difficulty may arise from the dis-
tinctive formation of the regional climate downstream
of the Tibetan Plateau.
Cloudradiation feedback is instrumental in under-
standing the long-term climatic trends over central east-
ern China (Yu et al. 2004a). Li et al. (2005) showed the
importance of the cloudradiation feedback in ampli-
fying the cooling trend downstream of the Tibetan Pla-
teau. Model-based attribution of regional SAT varia-
tion includes large uncertainty due to the uncertainty of
cloud feedback processes. Our analyses have identified
the discrepancies of climate models in simulating the
observed regional SAT variation. The physics govern-
ing these responses, in particular the cloudradiation
process will be the subject of a subsequent study.
Besides the cloudradiation feedback, teleconnec-
tions also contribute to the SAT variation over China.
In contrast with global warming trends, a strong cooling
trend occurs in early spring and late summer in the lee
side of the Tibetan Plateau. This cooling is associated
with an eastward extension of a cooling signal originat-
ing from North Africa, which is related to the positive
trend of the North Atlantic Oscillation (NAO; Yu and
Zhou 2004). Further analyses indicate that barely any
model of 20C3M could reasonably simulate the ob-
served positive trend of the NAO in recent decades.
Both IPSL-CM4 and UKMO-HadGEM1 moderately
exhibit the linear trends of the NAO; the NAO-related
SAT anomalies are, however, distorted by the bias in
the simulation of the NAO pattern (figures not shown).
Explanation of the cause of the cooling downstream
of the Tibetan Plateau has been extremely controver-
sial. A few works have established the links between
the SAT variations and the changes in aerosols (Qian
and Giorgi 1999; Qian et al. 2001; Xu 2001). Among the
analyzed models, only the GFDL-CM2.1 partly re-
sembles the observations in the cooling pattern (figure
not shown). Many models that are forced with time-
varying forcings of anthropogenic aerosols including
VOLUME 19
black carbon still failed to simulate the cooling trend
(figures not shown). The role of black carbon in forcing
the SAT variation over China might be a possibility.
Decadal variations of the large-scale circulation over
East Asia should also have contributions (Yu et al.
2004b).
In addition, there are large differences among the
20C3M models in the prescribed radiative forcings.
Further work is required to develop improved consen-
taneous estimates of various radiative forcings, with the
aim toward forming a standard set of forcings for the
twentieth-century climate intercomparison project.
Finally, our multimodel analyses of the 20C3M re-
veal an encouraging level of performance of the current
state-of-the-art climate models in simulating the SAT
variation averaged over China and a poor performance
in simulating the spatial patterns. This indicates that
while the future climate change scenario simulations of
IPCC AR4 might be valuable in the context of the Chi-
nese domain average, the projected spatial distributions
might be highly suspect. Considering the shortcomings
of the climate models in simulating the twentieth-
century climate, any plans to produce a credible geo-
graphical distribution of future climate change over
China will be beyond the ability of current state-of-the-
art climate models. This casts a shadow upon the effort
toward developing credible geographical distributions
of future climate change through IPCC scenario simu-
lations.
5. Summary
We have analyzed the variations of the twentieth-
century SAT over China and the globe from observa-
tions and 19 coupled climate models included in IPCC
AR4. The major results are summarized as follows.
Similar to global and Northern Hemispheric mean
SAT, most coupled climate models are able to simulate
the major variations and upward of the SAT averaged
over China in the twentieth-century climate simula-
tions. The correlation between the observed and simu-
lated Chinese SAT for 18801999 (r 0.55) is, how-
ever, lower than that for either the global (r 0.87) or
the Northern Hemispheric (r 0.82) average. For the
global mean SAT, the forced signal is approximately 2
times larger than the internal noise. For the SAT aver-
aged over China, the internal noise is larger than the
forced signal.
Both GHG and natural external forcings contribute
to the global and hemispheric average SAT variations
in the early twentieth century, as shown by improved
correlation with the GHG forcing and still significant
correlation after detrending the SAT. The SAT over
15 NOVEMBER 2006 ZHOU AND YU
China before 1940 in the coupled climate models cor-
relates with observations mainly because of the warm-
ing trend. After removing this trend, there is no signifi-
cant correlation between the observed and the simu-
lated SAT. Both the GHG and other forcings such as
anthropogenic aerosols contribute to the SAT variation
over China in the last half of the twentieth century.
Chinese SAT from 1941 to 1999 in multimodel en-
semble mean estimate is significantly correlated with
the observations, even after removing the warming
trend. Note these correlation results with or without
detrending are not sufficient to make conclusions about
GHGs role since the detrended signal is not necessarily
the GHG-induced warming; further runs without the
GHG are encouraged for the international modeling
groups.
There is strong time-scale dependence in the simu-
lated SAT variations. The prescribed natural and an-
thropogenic external forcings in the coupled climate
models mainly simulated the observed warming trends
and the decadal to interdecadal SAT variations, with
poor performances on shorter scales. Compared with
the coupled climate models, the observational-SST-
forced AGCM simulation has a comparable (slightly
better) result in simulating the decadal to interdecadal
(interannual) scale SAT variations.
Although nearly all models can produce a fairly rea-
sonable variation of the SAT averaged over China in
the last 50 yr, there are still large discrepancies between
the observed and simulated regional features. The most
prominent warming trend in the last half of the century
and its acceleration in recent decades over China are
only weakly reproduced in the coupled climate models.
The performance of the coupled climate models in pro-
ducing the Chinese SAT trend is seasonally dependent.
Few models could reproduce the moderate summer-
time cooling over central eastern China (2736N),
while two of the models acceptably produced the me-
ridional gradients of the wintertime warming trends,
with north China experiencing a stronger warming. The
spread of SAT trends over north China is larger than
that over south China in the model simulations.
The warming in the 1920s over China cannot be pro-
duced in any of the coupled climate models, leaving the
mechanism governing this warming an open problem.
Future efforts should also be devoted to the reconstruc-
tion and validation of Chinese SAT before 1950.
Acknowledgments. We acknowledge the interna-
tional modeling groups for providing their data for
analysis, the Program for Climate Model Diagnosis and
Intercomparison (PCMDI) for collecting and archiving
the model data, the JSC/CLIVAR Working Group on
5857
Coupled Modelling (WGCM) and their Coupled Model
Intercomparison Project (CMIP) and Climate Simula-
tion Panel for organizing the model data analysis activ-
ity, and the IPCC WG1 TSU for technical support. The
IPCC Data Archive at Lawrence Livermore National
Laboratory is supported by the Office of Science, U.S.
Department of Energy. This work was jointly sup-
ported by the Chinese Academy of Sciences through its
Knowledge Innovation Program (KZCX3-SW-221),
the Major State Basic Research Development Program
of China (973 Program) under Grant 2005CB321703,
the International Partnership Creative Group entitled
The Climate System Model Development and Appli-
cation Studies, and the National Natural Science
Foundation of China (40375029, 40233031). We thank
Mr. Xinyu Wen for performing the SST-forced CAM2
integration on the Lenovo DeepComp 6800 supercom-
puter at the Supercomputing Center of the Chinese
Academy of Sciences.
REFERENCES
Andronova, N. G., and M. E. Schlesinger, 2000: Causes of global
temperature changes during the 19th and 20th centuries.
Geophys. Res. Lett., 27, 21372140.
Boer, G. J., G. Flato, M. C. Reader, and D. Ramsden, 2000: A
transient climate change simulation with greenhouse gas and
aerosol forcing: Experimental design and comparision with
the instrumental record for the 20th century. Climate Dyn.,
16, 405426.
Broccoli, A. J., K. W. Dixon, T. D. Delworth, T. R. Knutson, R. J.
Stouffer, and F. Zeng, 2003: Twentieth-century temperature
and precipitation trends in ensemble climate simulations in-
cluding natural and anthropogenic forcing. J. Geophys. Res.,
108, 4798, doi:10.1029/2003JD003812.
Collins, W. D., and Coauthors, 2003: Description of the NCAR
Community Atmosphere Model (CAM3). National Center
for Atmospheric Research Tech. Note NCAR/TN-464STR,
Boulder, CO, 171 pp. [Available online at http://www.ccsm.
ucar.edu/models/atm-cam/docs/description/index.html.]
Dai, A., T. M. L. Wigley, B. A. Boville, J. T. Kiehl, and L. E. Buja,
2001a: Climates of the 20th and 21st centuries simulated by
the NCAR Climate System Model. J. Climate, 14, 485519.
, , G. A. Meehl, and W. M. Washington, 2001b: Effects of
stabilizing atmospheric CO2 on global climate in the next two
centuries. Geophys. Res. Lett., 28, 45114514.
Delworth, T. L., and T. R. Knutson, 2000: Simulation of early 20th
century global warming. Science, 287, 22462250.
Diaz, H. F., and R. G. Quayle, 1980: The climate of the United
States since 1895: Spatial and temporal changes. Mon. Wea.
Rev., 108, 149226.
Emori, S., T. Nozawa, A. Abe-Ouchi, A. Numaguti, M. Kimoto,
and T. Nakajima, 1999: Coupled oceanatmosphere model
experiments of future climate change with an explicit repre-
sentation of sulfate aerosol scattering. J. Meteor. Soc. Japan,
77, 12991307.
Fu, C. B., H. Diaz, D. F. Dong, and J. Fletcher, 1999: The changes
of atmospheric circulation over Northern Hemispheric
5858 JOURNAL OF CLIMATE
oceans associated with the global rapid warming of the 1920s.
Int. J. Climatol., 19, 581606.
Giorgi, F., P. Whetton, R. Jones, J. H. Christensen, L. O. Mearns,
B. Hewitson, and H. Von Storch, 2001: Emerging patterns of
simulated regional climatic changes for the 21st century due
to anthropogenic forcings. Geophys. Res. Lett., 28, 3317
3327.
Harzallah, A., and R. Sadourny, 1995: Internal versus SST-forced
atmospheric variability as simulated by an atmospheric gen-
eral circulation model. J. Climate, 8, 474495.
Haywood, J. M., R. J. Stouffer, R. T. Wetherald, S. Manabe, and
V. Ramaswamy, 1997: Transient response of a coupled model
to estimated changes in greenhouse gas and sulfate concen-
trations. Geophys. Res. Lett., 24, 13351338.
Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der
Linden, X. Dai, K. Maskell, and C. A. Johnson, Eds., 2001:
Climate Change 2001: The Scientific Basis. Cambridge Uni-
versity Press, 892 pp.
Hu, Z. Z., S. Yang, and R. Wu, 2003: Long-term climate variations
in China and global warming signals. J. Geophys. Res., 108,
4614, doi:10.1029/2003JD003651.
Johns, T. C., and Coauthors, 2003: Anthropogenic climate change
for 1860 to 2100 simulated with the HadCM3 model under
updated emissions scenarios. Climate Dyn., 20, 583612.
Jones, P. D., and A. Moberg, 2003: Hemispheric and large-scale
surface air temperature variations: An extensive revision and
an update to 2001. J. Climate, 16, 206223.
, M. New, D. E. Parker, S. Martin, and I. G. Rigor, 1999:
Surface air temperature and its changes over the past 150
years. Rev. Geophys., 37, 173199.
Kimoto, M., 2005: Simulated change of the East Asian circulation
under global warming scenario. Geophys. Res. Lett., 32,
L16701, doi:10.1029/2005GL023383.
Li, J., R. C. Yu, T. J. Zhou, and B. Wang, 2005: Why is there an
early spring cooling shift downstream of the Tibetan Plateau?
J. Climate, 18, 46604668.
Li, Z. X., 1999: Ensemble atmospheric GCM simulation of climate
interannual variability from 1979 to 1994. J. Climate, 12, 986
1001.
Meehl, G. A., W. M. Washington, T. M. L. Wigley, J. M. Ar-
blaster, and A. Dai, 2003: Solar and greenhouse gas forcing
and climate response in the twentieth century. J. Climate, 16,
426444.
, , C. Ammann, J. M. Arblaster, T. M. L. Wigley, and C.
Tebaldi, 2004: Combination of natural and anthropogenic
forcings and 20th century climate. J. Climate, 17, 37213727.
Nozawa, T., T. Nagashima, H. Shiogama, and S. A. Crooks, 2005:
Detecting natural influence on surface air temperature
change in the early twentieth century. Geophys. Res. Lett., 32,
L20719, doi:10.1029/2005GL023540.
Qian, Y., and F. Giorgi, 1999: Interactive coupling of regional
climate and sulfate aerosol models over eastern Asia. J. Geo-
phys. Res., 104, 64776499.
, , Y. Huang, W. L. Chameides, and C. Luo, 2001: Re-
gional simulation of anthropogenic sulfur over East Asia and
its sensitivity to model parameters. Tellus, 53B, 171191.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V.
Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003:
VOLUME 19
Global analyses of SST, sea ice and night marine air tem-
perature since the late nineteenth century. J. Geophys. Res.,
108, 4407, doi:10.1029/2002JD002670.
Rogers, J. C., 1985: Atmospheric circulation changes associated
with warming over the northern Atlantic in the 1920s. J. Cli-
mate Appl. Meteor., 24, 13031310.
Schlesinger, M. E., and N. Ramankutty, 1994: An oscillation in the
global climate system of period 6570 years. Nature, 367, 723
726.
Stott, P. A., S. F. B. Tett, G. S. Jones, M. R. Allen, J. F. B. Mitch-
ell, and G. J. Jenkins, 2000: External control of 20th century
temperature by natural and anthropogenic forcings. Science,
290, 21332137.
, and Coauthors, 2001: Attribution of twentieth century tem-
perature change to natural and anthropogenic causes. Cli-
mate Dyn., 17, 122.
Tett, S. F. B., and Coauthors, 1999: Causes of twentieth century
temperature change. Nature, 399, 569572.
, and Coauthors, 2002: Estimation of natural and anthropo-
genic contributions to twentieth century temperature change.
J. Geophys. Res., 107, 4306, doi:10.1029/2000JD000028.
Torrence, C., and G. P. Compo, 1998: A practical guide to wavelet
analysis. Bull. Amer. Meteor. Soc., 79, 6178.
Wang, S. W., and D. Gong, 2000: Enhancement of the warming
trend in China. Geophys. Res. Lett., 27, 25812584.
, J. Ye, D. Gong, J. Zhu, and T. Yao, 1998: Reconstruction of
the temperature series in China for the last one hundred
years (in Chinese). Quart. J. Appl. Meteor., 9, 392401.
, D. Gong, and J. Zhu, 2001: Twentieth-century climate
warming in China in the context of the Holocene. Holocene,
11, 313321.
Xin, X. G., R. C. Yu, T. J. Zhou, and B. Wang, 2006: Drought in
late spring of south China in recent decades. J. Climate, 19,
31973206.
Xu, Q., 2001: Abrupt change of the mid-summer climate in central
east China by the influence of atmospheric pollution. Atmos.
Environ., 35, 50295040.
Yu, R. C., and T. J. Zhou, 2004: Impacts of winter-NAO on March
cooling trends over subtropical Eurasian continent in the re-
cent half century. Geophys. Res. Lett., 31, L12204,
doi:10.1029/2004GL019814.
, B. Wang, and T. J. Zhou, 2004a: Climate effects of the deep
continental stratus clouds generated by Tibetan Plateau. J.
Climate, 17, 27022713.
, , and , 2004b: Tropospheric cooling and summer
monsoon weakening trend over East Asia. Geophys. Res.
Lett., 31, L22212, doi:10.1029/2004GL021270.
Zhang, X., and J. E. Walsh, 2006: Toward a seasonally ice-covered
Arctic Ocean: Scenarios from the IPCC AR4 model simula-
tions. J. Climate, 19, 17301747.
Zhao, Z., Y. Ding, Y. Luo, and S. Wang, 2005: Recent studies on
attributions of climate change in China. Acta Meteor. Sin., 19,
389400.
Zhou, T. J., and R. C. Yu, 2005: Atmospheric water vapor trans-
port associated with typical anomalous summer rainfall pat-
terns in China. J. Geophys. Res., 110, D08104, doi:10.1029/
2004JD005413.