Comparative Study of Performance of CMIP 3 GCMs in Simulating the East Asian Monsoon Variability

This study evaluates variability of East Asian Monsoon simulated by 24 coupled general circulation models (GCMs) participating in Coupled Model Inter-comparison Project 3 (CMIP3). Fifty years (1950 1999) of each model’s twentiethcentury climate simulation are analyzed and compared with observed data. Both East Asian Summer Monsoon (EASM) and East Asian Winter Monsoon (EAWM) are considered. Suitable indices are selected to analyze EASM and EAWM. The Wang-Fan index, associated with 850 hPa circulation pattern and tripole rainfall pattern are considered to analyze EASM. ICHENW, associated with 850 hPa circulation pattern, IGONGDY, associated with Siberian High, and ISUNBM, associated with 500 hPa East Asian trough are considered for analysis of EAWM. Inter-decadal and inter-annual variability of the EASM and EAWM are major focus of this study. The results indicate that, amplitude of inter-annual component EASM becomes larger after 1980’s. The decadal component shows, weakening trend and switching from positive monsoon phase to negative monsoon phase in mid-80s for both EASM and EAWM. The simulated composite differences between weak and strong monsoon decades by models with above mentioned characteristics are compared with observation for further evaluation. The EASM weakening is associated with weaker southwesterly and stronger anti-cyclonic pattern over Western North Pacific and results in more rainfall in south and north China, Korea, Japan and less rainfall in central China. The EAWM weakening is associated with weakening of Siberian High and temperature rise over East Asia. EASM and EAWM matrices are constructed according to the relative merits of GCMs.


IntroduCtIon
In preparation for the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4), more than a dozen international climate modeling centers have conducted a comprehensive set of long-term simulations of climate during the twentieth century and various climate scenarios in the twenty-first century (IPCC 2007) under the coordination of the Coupled Model Inter-comparison Project 3 (CMIP3).As the need for research on climate change has grown and with the advances in computing power, GCMs have become more and more complex.Because of their complexity, they often lead to greater uncertainty than desired (Raick et al. 2006).One of the primary factors related to uncertainty in CGMs is that different models can simulate quite different regional changes under the same anthropogenic forcings (Whetton et al. 1995;Kittel et al. 1998;Giorgi and Francisco 2000) and it is difficult to ascertain which GCMs are most reliable.Often, projections from GCMs are characterized by a low level of confidence and high level of uncertainty (Giorgi and Francisco 2000;Visser et al. 2000).Furthermore, previous studies have shown that there can be considerable differences in the accuracy of predictions from different models (Lorenz 1982;Delworth and Knutson 2000).It is difficult to identify the most suitable model simulations because of the varying performance of AOGCMs (Atmosphere Ocean Coupled General Circulation Model).CMIP3 has released 24 state-of-art GCMs from institutes all over the world and it is a common practice to project future climate based on the results of CMIP3 GCMs.Before projecting future changes in climate, evaluating these models is important.
The year-to-year variation in the Asian monsoon is one of the most important signals of Earth's climate variability.How to best simulate and predict annual variation in the East Asian Monsoon is an important task in climate prediction.In order to quantify the variability in the East Asian monsoon, using a representative variable (or variables) as an objective measure (or measures) is one approach (Sun and Li 1997;Wang and Fan 1999;Chen et al. 2000;Gong et al. 2001).Using concise and meaningful indices to characterize monsoon variability can facilitate empirical studies on the relationship between monsoon variability and lower boundary forcing (Wang et al. 2008b;Wang and Chen 2010).In order to project the future characteristics of the East Asian monsoon, it is necessary to validate the capability of CMIP3 GCMs in quantifying the variability in the East Asian monsoon.This issue has been relatively unexplored in previous studies.The present study includes a comprehensive assessment of the relative merits of CMIP3 models in simulating the East Asian summer and winter monsoon (EASM and EAWM) variability.A large number of summer and winter monsoon indices have been proposed.Some of these indices are adopted in this study for model validation.

dAtA And MEthodoloGy
This study is based on the 20th century simulations (20C3M) of 24 GCMs provided by CMIP3, which is the main basis for the AR4 (IPCC 2007).Table 1 provides a list of 24 AOGCMs.Our analysis compares the performance of different GCM simulations in simulating the East Asian monsoon variability in the domain 0 -50°N to 90 -150°E.The data used for evaluation are: (1) CRU TS 2.1 rainfall (Mitchell and Jones 2005), (2) Hadley Centre mean sea level pressure (Basnett and Parker 1997), and (3) the NCEP Reanalysis (Kalnay et al. 1996).
The performances of GCMs in simulating the interannual variability of the EASM and EAWM variability are evaluated using existing monsoon indices.A survey of previous studies indicated that there are at least 25 existing indices for the EASM (Wang et al. 2008a) and 18 indices for the EAWM (Wang and Chen 2010).Thus, it is difficult to select the most suitable indices for comparative studies.
The EASM is characterized by a complex spatial structure and temporal fluctuations that encompass both the tropics and the extra-tropics.An east-west elongated rain-belt is one of the prominent features of the EASM.The intense rains during summer in different regions and times are referred to by different names, such as Mei-yu in China, Baiu in Japan, and Changma in Korea.It is difficult to quantify the variability in the EASM with averaged rainfall over East Asia because of the large spatial and temporal variation.Therefore, most of the investigators searching for a simple index for the EASM strength use circulation parameters instead of rainfall (Webster and Yang 1992;Guo 1994;Wang and Fan 1999;Zhang et al. 2002).Wang et al. (2008a) analyzed all 25 indices for the EASM and recommended the Wang-Fan (WF) index as the most suitable (Wang and Fan 1999).Thus we used the WF index as one of our indices for the study of summer monsoon intensity over East Asia.The WF index is defined as the U 850 in 5 -15°N to 90 -130°E minus U 850 in 22.5 -32.5°N to 110 -140°E, where U 850 denotes the zonal wind at 850 hPa.
In contrast to the circulation, most studies characterize the rainfall variability in terms of the dominant spatial pattern, e.g., the leading EOF (Empirical Orthogonal Function).Hsu and Lin (2007) studied the tripole pattern, which is the leading EOF of rainfall at both inter-annual and decadal time scales in East Asia, and the associated circulation during the northern summer.The tripole pattern is closely associated with the wave-like Pacific Japan pattern, which is the most recurrent dominant circulation pattern associated with the EASM (Nitta 1987;Lau 1992;Nitta and Hu 1996).In this study we consider the tripole pattern as another index of the summer monsoon index.The following procedures were carried out to isolate the inter-annual variability of the tripole pattern.The linear trend at each grid point is derived using linear regression and then subtracted.We analyzed the inter-annual and decadal components.The 9-year running means were then computed and subtracted from the detrended data to exclude the decadal and interdecadal signals, retaining the inter-annual component.Sun and Li (1997)] for the lower-tropospheric temperature variability.
The justification for selecting these indices is as follows.ICHENW is a circulation index which is related to the El Niño and Southern Oscillation (ENSO) and is suitable for looking at the tropical Pacific's influence on the EAWM.The monsoon index IGONGDY is related to the intensity of Siberian High and is therefore suitable for evaluating the effect of high latitudes on the EASM.The monsoon index ISUNBM represents the fluctuation of the 500 hPa geo-potential height over East Asia, where cold air usually penetrates southward.The ISUNBM index is used in this study to reflect the fluctuation of air temperature in the lower troposphere along the East Asian coast.
The above indices are calculated based on the output from the GCMs and compared with the observed data to evaluate the ability of 24 GCMs in simulating the EAM climate variability.The time period analyzed for this study is from 1950 to 1999.Seasonal means are considered throughout this study.We use the means for June-July-August (JJA) for summer and December-January-February (DJF) for winter.The difference between strong decadal monsoon and weak decadal monsoon in the time series for the decadal component (9 years average mean) was compared between the GCM output and the observed data.We consider this a composite difference between a strong and weak monsoon period for different parameters (temperature, rainfall, wind speed etc.) which reflect the main features of EASM and EAWM.Strong and weak monsoon decades are identified when the normalized indices of the decadal component exceeded ±.5.

SuMMEr MonSoon VArIAbIlIty
Here the simulated seasonal average (JJA) rainfall and variability for the last 50 years (1950 -1999) extracted from the 20C3M simulations for each of the 24 models are compared and evaluated against the observed data.Many models simulated a reasonable spatial distribution.Correlation coefficients (pattern correlation) between the observed and simulated mean rainfall (Table 2, column 1) are above 0.6 for all but 5 GCMs (i.e., FGOALS-g1.0,IPSL-CM4, MRI-CGCM2.3.2,PCM, and CNRM-CM3).However, a high correlation does not necessarily warrant a model's ability to simulate the major characteristics of the rainfall distribution.As seen in Fig. 1, the measured rainfall is characterized by a northwestward decrease from the coastal area to deep area in the continent, i.e., more rainfall in the Indochina peninsula, Philippines, southeastern China, Japan and Korea, and less rainfall in northwestern China.Ten [CSIRO, GFDL, GISS-AOM, INGV-SXG, MIROC3.2(hires),UKMO, CNRM-CM3] out of twenty four GCMs reasonably simulated this overall pattern.It is interesting to note that all of these GCMs have high spatial resolution.On the other hand,  higher resolution does not guarantee better simulations.The ECHAM5/MPI-OM model with relatively high resolution is an example of this.Almost all GCMs simulated anomalously high rainfall in central China near Sichuan Basin where lee cyclogenesis is active.Whether these GCMs overestimated lee cyclogenesis in this region is a worthwhile focus for further investigation.
The simulated spatial distribution of inter-annual variation in rainfall is also examined.As shown in column 2 of Table 2 (figure not shown), the models did not accurately simulate the inter-annual variation.Eleven out of 24 GCMs had correlation coefficients less than 0.6.A comparison between column 1 and 2 also reveals that the models that are simulating mean annual rainfall also provide better simulations of inter-annual variability.
The temporal/spatial performance and prediction potential for 1950 -1999 for different GCMs are evaluated using the WF Index and the tripole pattern.The temporal variability in normalized inter-annual and decadal component of the WF index simulated by different GCMs and the observed data are compared in Fig. 2. The observed WF index (derived from NCEP reanalysis data) is dominated by inter-annual variability.The WF index also exhibits clear inter-decadal fluctuations characterized by the change from an upward trend to a downward trend around the late 1980s.This shift reflects a weakening trend of the EASM starting from 1980s, as pointed out in previous studies (Yu et al. 2004).It is also worth noting that the amplitude of the interannual variability appears larger after the late 1970s.Even though the 20C3M simulations include precribed external forcings such as greenhouse gas aerosol concentration, and the solar constant, etc., atmospheric and oceanic variability are still affected by models' natural variability.Due to the stochastic nature of the forcings, the temporal fluctuation of the simulated EASM may differ markedly from the observed data even if the variability is essentially the same as the variation in the observed data in a statistical sense.It is therefore not meaningful to compare the simulated interannual and inter-decadal variation with the observed data in detail.On the other hand, it would be interesting to know if any of these GCMs have simulated a similar transition from a strong monsoon to a weak monsoon during the 50-year simulation period even though the occurrence time does not match the observed record.If so, the next question would be whether the observed changes in circulation characteristics are properly simulated.
A comparison between the observed and simulated time series does not yield an impression that any of the 24 GCMs simulate the amplification of inter-annual variability (not shown).This might suggest that the observed amplification of inter-annual variability is not forced by an increasing anthropogenic greenhouse effect.On the other hand, it may be argued that current GCMs are far from perfect to simulate this phenomenon even if it is forced by the increasing anthropogenic greenhouse effect.Time series of thirteen GCMs that simulated a weakening trend anytime in the 50 year record are presented in Fig. 2. A further comparison between the observed and simulated time series indicates that only seven GCMs (i.e., GISS-EH, GISS-ER, MRI-CGCM2.3.2,UKMO-HadGEM1, CSIRO-Mk3.0,IPSL-CM4, and CCSM3) simulated the weakening of EASM in the latter part of the record, although the timing did not perfectly match the observations.This means only seven out of 24 GCMs are capable of simulating the observed weakening EASM.
Despite this, it would be interesting to know whether the GCMs accurately simulate the contrasting rainfall and circulation characteristics between the strong and weak EASM.This study examined the 14 GCMs shown in Fig. 2 to search for one strong and one weak EASM decade.The selected decade for each model is presented in Table 3.Only those GCMs with the normalized anomaly in an entire decade larger than (±) 0.5 standard deviation are chosen.The selected decades for the NCEP are 1975 -1984 and 1990 -1999 for a strong and weak EASM.Because UKMO-HadGEM1 and CCSM3 are the two GCMs whose selected decades best matched the observed data, we include them in the analysis even though the amplitude anomaly is relatively small.BCCR-BCM2.0and IPSL-CM4 are the two GCMs excluded from further analysis because of small amplitudes.
The summertime circulation in East Asia and the Western North Pacific consists of three major components: thermal cyclonic circulation over land, the monsoon trough extending from the Eurasian continent to the Philippine Sea, and the Pacific subtropical anticyclonic ridge north of Table 3.Strong and weak monsoon decades for NCEP and the selected GCMs that shows weakening of summer monsoon according to IWF.

GCM
Strong monsoon period (10 yrs) the monsoon trough.The relative movement and amplitude variation of these three components have significant influence on the regional weather and climate.The WF index is designed to identify the strength of the monsoon trough.
A larger (smaller) WF index indicates a stronger (weaker) monsoon trough and is interpreted as a strong (weak) EASM.
A composite of the weak monsoon decades (i.e., later decades) are subtracted from a composite of strong monsoon decades (earlier decades) to examine the capability of the GCMs to simulate the observed contrasts between the strong and weak EASM.The composite difference of 850 hPa wind field between the negative and positive phase monsoon decades (Fig. 3, upper left corner) shows a weaker westerly near the equator, which is the southern branch of an anticyclonic circulation anomaly covering the entire subtropical western North Pacific, and is accompanied by the northeasterly anomaly over Japan, the East China Sea, and eastern China.This means that during the latter decades when the monsoon is weak, the anomaly is anti-cyclonic and the monsoon trough is weaker when the anticyclone is stronger.
The corresponding rainfall anomaly shown in the upper left corner of Fig. 4 indicates a positive anomaly in southeastern China, northeastern Indochina peninsula, Korea, and northwestern China, and a negative anomaly in central China.The weaker monsoon trough is accompanied by stronger southwesterly flow to coastal East Asia.This circulation configuration would lead to more moisture transport and convergence, and therefore more rainfall in the area.
Simulated circulation and rainfall anomalies for the GCMs weakening of monsoon are shown in Figs. 3 and 4, and the corresponding pattern correlations are shown in columns 6 and 4 of Table 2, respectively.The stream function west of 120°E is not included in the computation of pattern correlation because of the lack of data in the mountainous area and the WF index is defined based on the circulation over ocean.Six GCMs [i.e., CGCM3.1(T63),CSIRO-Mk3.0,INGV-SXG, MIROC3.2(medres),MRI-CGCM2.3.2, and UKMO-HadGEM1] reasonably simulate the circulation pattern with pattern correlation larger than 0.6.All of these six GCMs simulate an anticyclonic circulation somewhere in the Western North Pacific.However, the location of the anticyclonic circulation differs between GCMs.Qualitatively it can be seen that all the GCMs (Table 2, columns 6, 7) showing this weakening and can simulate an anti-cyclonic pattern somewhere over the western north Pacific although pattern correlation with observation may not be good.
The correlation for rainfall is much lower with only three GCMs exceeding 0.2 (Table 2, column 4).Qualitatively, the observed data show a negative rainfall anomaly (i.e., decrease of rainfall) over central China, which is reflected in some of the models, including CGCM3.1(T63),CSIRO-Mk3.0,MRI-CGCM2.3.2,CCSM3, UKMO-HadGEM1 (Table 2, column 5).It is interesting to note that the GCMs with the best correlation for rainfall are those models with the best performance in wind circulation simulation.On the other hand, GCMs that performed well in simulating wind circulation [e.g., INGV-SXG, MIROC3.2(medres), and UKMO-HadGEM1] do not warrant a reasonable rainfall simulation.Also, none of the GCMs which simulate circulation poorly produce a reasonable simulation for rainfall.This indicates that the rainfall is so sensitive that a slight displacement of circulation results in a significantly different rainfall pattern.For example, the MIROC3.2(medres)model produces a circulation pattern correlation as high as 0.76 but with the correlation for rainfall is -.05 (Table 2, column 6).It is obvious by comparing Figs. 3 and 4 that the northward-displacement of the anticyclonic anomaly results in an easterly anomaly in southern China and negative rainfall anomaly in the northern Indochina Peninsula and southwestern China, instead of the southwesterly anomaly and positive rainfall anomaly as shown in the observed data.On the other hand, a seemingly good precipitation simulation does not necessarily result from a correct circulation pattern.An interesting example is MRI-CGCM2.3.2 that had correlations of 0.84 and 0.35 for stream function and rainfall, respectively.A close inspection reveals that the model simulated northerly anomaly, instead of southerly, over southeastern China.The northerly anomaly is part of the weak trough along the southeastern coast of China.The trough is evidently the cause for the positive rainfall anomaly.This indicates that a reasonable simulation of rainfall fluctuation is still a challenge for these GCMs even at decadal time scales.
As in previous studies and discussed above (Hsu and Lin 2007), the tripole pattern is a dominant pattern in both inter-annual and inter-decadal variability of EASM.Because it makes no sense to compare annual data, we evaluate the model performance in simulating the inter-decadal and as well as inter-annual variability by comparing the leading mode of the observed and simulated rainfall.The empirical orthogonal analysis is applied to the CRU data to retrieve the most recurrent pattern in East Asia.As noted before, both the trend and decadal fluctuations are removed before applying EOF to isolate the inter-annual variability.The first EOF of CRU data explains 23% of total variance and exhibits the well known tripole structure (Hsu and Lin 2007) with negative polarity over southern China, Taiwan, and northern China and positive polarity over central China, Japan, and Korea (Fig. 5).The positive and negative anomalies are elongated in the east-west direction and are relatively narrow in meridional direction.Corresponding time series show a distinct widening of amplitude of inter-annual variability in two recent decades, CSIRO-Mk3.0,CSIRO-Mk3.5,GFDL-CM2.0,GFDL-CM2.1,GISS-AOM, GISS-EH, INM-CM3.0,FGOALS-g1.0,INGV-SXG, MIROC-3.2(hires),MIROC3.2(medres),PCM, UKMO-HadCM3, UKMO-HadGEM1 show a distinct tripole pattern in first or second EOF of their inter-annual variability with reasonably well spatial pattern correlation (Table 2, columns 10, 11).The corresponding time series are also presented in Fig. 5 for reference, although we do not expect the similarity to the observed time series.GISS-EH was the only one GCM showing the amplified amplitude after 1980s.Some models (e.g., BCCR-BCM2.0,IPSL-CM4, ECHO-G, CNRM-CM3) do not simulate any EOF similar to the tripole pattern.For the case of decadal component, CRU data explains the first EOF as the tripole pattern of rainfall with 22.2% of total variance (Fig. 6).The tripole pattern exhibits positive polarity over southern China, Taiwan, and northern China and negative polarity in central part of China, Japan, and Korea.The corresponding time series shows decadal variability with switching of phase in the late 1970s, indicating that northern and southern China become drier while central China became wetter after the late 1970s.Some models, including BCCR-BCM2.0,CSIRO-Mk3.5,GFDL-CM2.1,GISS-AOM, FGOALS-g1.0,INGV-SXG, IPSL-CM4, MIROC3.2-(hires),MIROC3.2(med),UKMO-HadCM3, UKMO-Had-GEM1 show the tripole pattern in their decadal component with significant pattern correlation with the observed data (Table 2, columns 8, 9).Although some models simulate the sign of switching, other models (e.g., BCCR-BCM2.0,FGOALS-g1.0,UKMO) indicate a drier central China after the phase change.MIROC3.2(medres)seems to be the only GCM simulated correct phase shift, although the timing does not seem correct.Thus the CMIP3 GCMs are capable of simulating the tripole pattern as leading EOFs but are not able to simulate the temporal characteristics.

WIntEr MonSoon VArIAbIlIty
The East Asian winter monsoon directly influences the winter climate of East Asia.This is the most important circulation system of boreal winter over East Asia.The year to year variability of intensity of East Asian winter monsoon affects the wintertime temperature and rainfall variations over this region.Therefore, it is a necessity to quantify the strength of the East Asian winter monsoon and its variability.The simulated and observed seasonal average (DJF) temperature and MSLP climatology over East Asia are compared.All models reasonably simulate the average surface air temperature pattern during 1950 -1999 and the correlations with the observed data are also very high (Table 4, column 1).Besides, the simulated variability pattern of surface air temperature is also quite close to the observation.However, overestimates for the northern region result from some models such as INM-CM3.0,PCM, UKMO-Had-GEM1, UKMO-HadCM3, GFDL-CM2.1,GFDL-CM2.0(not shown).All of models perform well in case of mean MSLP pattern (Table 4, column 3).The MSLP variance pattern simulated by GCMs is close to the observations in most of the case except for GCMs GISS-AOM, GISS-EH, INM-CM3.0,CCSM3.
Different investigators tried to define the large scale East Asian winter monsoon index with circulation parameters.Some of the distinct characteristics, like 500-hpa East Asian trough, east-west pressure contrast or subtropical surface meridional wind are used to define the East Asian winter monsoon strength (Guo 1994;Shi 1996;Sun and Li 1997;Cui and Sun 1999;Chen et al. 2000;Wang et al. 2009a).Thus far, 18 existing East Asian winter monsoon indices have been proposed and all are concerned with these characteristics.Three indices, suggested by Wang and Chen (2010) namely ICHENW (Chen et al. 2000), IGONGDY (Gong et al. 2001), and ISUNBM (Sun and Lin 1997) are considered here.These indices are considered to produce skillful simulations of annual variation in the EAWM (Wang and Chen 2010).
We also compared temporal variability in the East Asian winter monsoon between the GCMs and the observations, as for the summer monsoon.The computed time series of ISUNBM for the observed and GCM data show strong inter-annual variability (Fig. 7).The positive (negative) index corresponds to positive (negative) phase of winter monsoon.The 9-year running mean produces the inter-decadal component which exhibits a continuous weakening trend 1970 onwards and a clear switching from positive monsoon phase to negative monsoon phase in mid-1980s (Fig. 7).This continuous weakening in decadal time scale and associated warming over East Asia is an important characteristic of EAWM (Chang et al. 2006;Wang et al. 2009b).Therefore, it is very important to evaluate GCMs in this respect.The decadal component of most of the GCMs like GISS-AOM, INM-CM3.0,IPSL-CM4, MRI-CGCM2.3.2,PCM, UKMO-HadCM3, UKMO-HadGEM1, MIROC3.2(hires),MIROC3.2(medres),CNRM-CM3, BCCR-BCM2.0,CG-CM3.1(T47),CGCM3.1(T63),GISS-ER, CSIRO-Mk3.0,GISS-EH, INGV-SXG, ECHAM5/MPI-OM exhibit the weakening.But among these GCMs, GISS-EH, INGV-SXG, MIROC3.2(medres)fail to catch the continuous   weakening as indicated by the strengthening of monsoon in the late 1990s.The transition time, i.e., the time for switching of phase, varies for different GCMs compare to observation.The positive and negative monsoon decades are identified by decadal average of the decadal component of monsoon index more than ±.5 (Table 5).Although CSIRO-Mk3.0 and CNRM do not satisfy this criteria, but still they are considered due to continuously weakening since 1960s for CNRM, and a sharp weakening in recent two decades for CSIRO-MK3.4, column 8), and the spatial correlation coefficients with the observation are also reasonably high for most of the GCMs (Table 4, column 7).Similar types of study are performed using the index IGONGDY, which is associated with the intensity of Siberian High.The continuous weakening in decadal scale after 1970s is a distinctive feature of East Asian Winter Monsoon.This is pronounced in the observed data and some of the GCM simulations like CGCM3.1(T63),CGCM3.1(T47),GISS-AOM, GISS-ER, IPSL-CM4, ECHO-G, PCM, CN-RM-CM3 (Fig. 10), although the transition time as well as the chosen positive and negative monsoon decades according to the above mentioned criteria imposed are not the same for the GCMs and the observations (Table 5).The negative phase of winter monsoon is characterized by weakening of Siberian High exhibited by the composite difference of mean sea level pressure between the weak and strong winter monsoon for observation (Fig. 11, upper left corner).Significant weakening of the Siberian High over the northern area of the domain as well as high pressure system over China are evident (Fig. 11), and are correlated with the GCM (we considered only GCMs with pronounced weak/strong decades according to our assumption)   4, columns 13, 14).Another index also illustrated here is ICHENW, which is associated with low level meridional wind component.The observed ICHENW index shows a clear weakening trend from 1980 with a switching of positive to negative monsoon phase in the late 1980s.It is interesting to note that the two previous indices show two weakening signs in the time series, the first one occurs in the late 1960s which shows a gradual weakening and the second one occurs in the late 1980s which shows a sharp change from positive to negative.The weakening northeasterly that occurred in the late 1980s coincides with the second weakening.The reasons for this are not clear.It is conjecture that the ICHENW reflects the northeasterly in the subtropical East Asia, i.e., in the lower latitudes than the previous two indices.The weakening of EAWM occurred earlier in higher latitudes and started later in the subtropical region when a more significant weakening occurred in a much larger area.Some GCMs like BCCR-BCM2.0,CGCM3.1(T63),CGCM3.1(T47),GISS-AOM, GISS-ER, INM-CM3.0,IPSL-CM4, MIROC3.2(hires),ECHAM5/MPI-OM, PCM exhibit this weakening feature clearly (Fig. 12).However, the transition period as well as the positive and negative monsoon decades varies in case of different GCMs with respect to observation (Table 5).The GCMs having pronounced strong/weak monsoon are considered only for further study (Table 5).The composite difference between negative and positive monsoon for observed temperature characterizes negative monsoon with significant rise of temperature over north China, Korea, Japan, Taiwan and significant weakening of the Siberian High, which is pronounced in GCMs like INM-CM3.0,IPSL-CM4, UKMO-HadCM3, with reasonable spatial pattern correlations (Table 4, column 20).The weaker northerly and northeasterly are simulated for the low-level wind component (925 hPa) in negative monsoon phase is another distinctive feature in the observed data (Fig. 13) and is simulated by some GCMs, including BC-CR-BCM2.0, INM-CM3.0,IPSL-CM4, MIROC3.2(hires),MIROC3.2(medres),CNRM-CM3 (Table 4, column 19), with reasonable spatial correlations for the 925 hPa stream function for some of the GCMs (Table 4, column 18).

ConStruCtIon oF A MonSoon MAtrIX
Here we assess the capability of the CMIP3 models to simulate the EAM variability.Because a large number of variables show a diverse number of outcomes, we develop   winter monsoon.The constructed matrices provide a concise and complete representation of relative merits of CMIP3 GCMs in simulating the East Asian Monsoon Variability.The summer monsoon matrix shows a complicated result.Although some of the GCMs like, CSIRO-Mk3.0,MRI-CGCM2.3.2,UKMO-hadGEM1 show better performance than other in most of aspect still they are poor in few aspect also.Based on this result, it is difficult to choose any GCM with good performance in every aspect.This kind of situation is not surprising considering the complicated variability of the EASM involving convection, which is one of the major deficiencies of the current GCMs.On the contrary, the winter monsoon matrix provides a clear picture about the relative performances of the GCMs.The qualitative and quantitative assessment shows very good performance of CGCM3.1(T63),GISS-AOM, INGV-SXG, INM-CM3.0,MIROC3.2(medres),MRI-CGCM2.3.2,UKMO-HadCM3, UKMO-HadGEM1, and CNRM-CM3 almost in every respect.
Wang and Chen (2010) have elaborated the interpretation of the 18 existing EAWM indices and classified them into four categories: low level wind indices, upper zonal wind shear indices, east-west pressure contrast indices, and East Asian trough indices.According to their results, three indices described below are chosen for this study.(1) ICHENW [average v wind component over East China Sea and South China Sea (10 -25°N, 110 -130°E and 25 -40°N, 120 -140°E)] [Chen et al. (2000)] for the northeasterly variability.(2) IGONGDY [mean SLP over the centre of the Siberian High (40 -60°N, 70 -120°E)] [Gong et al. (2001)] for the Siberian High variability.(3) ISUNBM [average Normalized 500 hpa Geo-potential Height over (35 -40°N, 125 -145°E)] [ between observation and the GCMs for mean rainfall pattern; 2 = Correlation between observation and the GCMs for rainfall variability pattern; 3 = Showing weakening according to IWF*; 4 = Spatial correlation between model observation according to IWF* for rainfall; 5 = Decrease of rainfall over Central China; 6 = Spatial correlation between model observation according to IWF* for stream function; 7 = Showing anti-cyclonic circulation over western north Pacific; 8 = Tripole pattern of rainfall for decadal component; 9 = Spatial correlation between observation and GCM; 10 = Tripole pattern of rainfall for inter-annual component; 11 = Spatial correlation between observation and GCM.* IWF= Wang-Fag Index.

Fig. 2 .
Fig. 2. Time series of monsoon index (IWF) of the GCMs showing weakening and observation (black line shows the inter-annual component and blue line shows the decadal component of monsoon index in each figure).

Fig. 3 .
Fig. 3. Comparison of decadal difference between the negative and positive monsoon decades for wind field in between different GCMs and observation.

Fig. 4 .
Fig. 4. Comparison of decadal difference between the negative and positive monsoon decades for precipitation in between different GCMs and observation.

Fig. 5 .
Fig. 5. Comparison of tripole pattern of inter-annual component of precipitation in between GCMs and observation.

Fig. 6 .
Fig. 6.Comparison of tripole pattern of decadal component of precipitation in between GCMs and observation.
correlation of average temperature; 2 = spatial correlation of temperature variability; 3 = spatial correlation of average MSLP; 4 = spatial correlation of MSLP variability; 5 = showing weakening according to ISUNBM; 6 = spatial correlation between model observation according to ISUNBM for geopotential height; 7 = spatial correlation between model observation according to ISUNBM for temperature; 8 = rise of temperature ISUNBM; 9 = spatial correlation between model observation according to ISUNBM for MSLP; 10 = weakening of Siberian High; 11 = showing weakening according to IGONGDY; 12 = spatial correlation between model observation according to IGONGDY for MSLP; 13 = spatial correlation between model observation according to IGONGDY for temperature; 14 = rise of temperature over north EA according to IGONGDY; 15 = showing weakening according to ICHENW; 16 = spatial correlation between model observation according to ICHENW for temperature; 17 = rise of temperature ICHENW; 18 = spatial correlation between GCMs and observations according to ICHENW for strmline; 19 = showing less north-easterly flow; 20 = spatial correlation between model observation according to ICHENW for MSLP; 21 = weakening of Siberian High.

Fig. 7 .
Fig. 7. Time Series of ISUNBM of GCMs showing weakening of EAWM and Observation(black line shows the inter-annual component and blue line shows the decadal component of monsoon index in each figure).

Fig. 11 .
Fig. 11.Decadal difference between the negative and positive monsoon decades for MSLP.

Fig. 12 .
Fig. 12.Time series of ICHENW of GCMs showing weakening and observation(black line shows the inter-annual component and blue line shows the decadal component of monsoon index in each figure).

Fig. 13 .
Fig. 13.Decadal difference between the negative and positive monsoon decades for stream function at 925 hPa.