J. Meteor. Res.  2014, Vol. 28 Issue (4): 481-509   PDF    
http://dx.doi.org/10.1007/s13351-014-4001-y
The Chinese Meteorological Society
0

Article Information

ZHOU Tianjun, CHEN Xiaolong, DONG Lu, WU Bo, MAN Wenmin, ZHANG Lixia, LIN Renping, YAO Junchen, SONG Fengfei, ZHAO Chongbo. 2014.
Chinese Contribution to CMIP5:An Overview of Five Chinese Models’Performances
J. Meteor. Res., 28(4): 481-509
http://dx.doi.org/10.1007/s13351-014-4001-y

Article History

Received January 1, 2014;
in final form April 1 2014
Chinese Contribution to CMIP5:An Overview of Five Chinese Models’Performances
ZHOU Tianjun1,3 , CHEN Xiaolong1,2, DONG Lu1,2, WU Bo1, MAN Wenmin1, ZHANG Lixia1, LIN Renping1,2, YAO Junchen1,2, SONG Fengfei1,2, ZHAO Chongbo4    
1 State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029;
2 University of the Chinese Academy of Sciences, Beijing 100029;
3 Climate Change Research Center, Chinese Academy of Sciences, Beijing 100029;
4 Laboratory for Climate Studies, National Climate Center, China Meteorological Administration, Beijing 100081
ABSTRACT:An overview of Chinese contribution to Coupled Model Intercomparison Project-Phase 5 (CMIP5) is presented. The performances of five Chinese Climate/Earth System Models that participated in the CMIP5 project are assessed in the context of climate mean states, seasonal cycle, intraseasonal oscillation, interan- nual variability, interdecadal variability, global monsoon, Asian-Australian monsoon, 20th-century historical climate simulation, climate change projection, and climate sensitivity. Both the strengths and weaknesses of the models are evaluated. The models generally show reasonable performances in simulating sea surface tem- perature (SST) mean state, seasonal cycle, spatial patterns of Madden-Julian oscillation (MJO) amplitude and tropical cyclone Genesis Potential Index (GPI), global monsoon precipitation pattern, El Niño-Southern Oscillation (ENSO), and Pacific Decadal Oscillation (PDO) related SST anomalies. However, the perfor- mances of the models in simulating the time periods, amplitude, and phase locking of ENSO, PDO time periods, GPI magnitude, MJO propagation, magnitude of SST seasonal cycle, northwestern Pacific mon- soon and North American monsoon domains, as well as the skill of large-scale Asian monsoon precipitation need to be improved. The model performances in simulating the time evolution and spatial pattern of the 20th-century global warming and the future change under representative concentration pathways projection are compared to the multimodel ensemble of CMIP5 models. The model discrepancies in terms of climate sensitivity are also discussed.
KeywordsCMIP5     Chinese models     seasonal cycle     MJO     GPI     ENSO     PDO     global monsoon     Asian monsoon     global warming     climate sensitivity    
1. Introduction

The Coupled Model Intercomparison Project(CMIP)was established by the JSC/CLIVAR Working Group on Coupled Modelling(WGCM)underthe World Climate Research Programme(WCRP)in1995 as a st and ard experimental protocol for studying the outputs of coupled atmosphere-ocean generalcirculation models(AOGCMs)(Meehl et al., 2000). CMIP provides a community-based infrastructure insupport of climate model validation, intercomparison, processes diagnosis, climate change attribution, and climate change projection. CMIP enables a diversecommunity of scientists around the world from boththe developed and developing countries to analyze theclimate models in a systematic fashion. This facilitates both model improvement and our underst and -ing of climate change sciences. The CMIP multimodeldataset has provided the basis for thous and s of peer reviewed papers and played prominent roles in the pastassessment reports of Intergovernmental Panel on Climate Change(IPCC)on climate variability and cli-mate change.

In the past about 20 years, CMIP has experienced five phases. The first phase of CMIP, calledCMIP1, collected output from coupled GCM controlruns in which CO2, solar constant and other external climatic forcing are kept constant. A subsequentphase, CMIP2, collected output from both coupledmodel control runs and matching runs in which CO2increases at the rate of 1% per year(Meehl et al., 1997). In both CMIP1 and CMIP2, besides CO2, noother anthropogenic climate forcing factors, such asanthropogenic aerosols, are included. Neither the con-trol runs nor the increasing CO2 runs include naturalvariations from volcanic eruptions or changing solarbrightness in climate forcing. The natural and an-thropogenic forcing agents were included in a subse-quent pilot project called "20th Century Climate inCoupled Models"(20C3M), which is a core experi-ment of CMIP3(Meehl et al., 2005). The 20C3Msimulation has been widely used in underst and ing thedriving factors of the 20th-century historical climateevolution(Zhou and Yu, 2006). After CMIP3, addi-tional simulations were performed that could be usedto separate anthropogenic and natural influences onthe 20th-century climate during CMIP4(Meehl et al., 2007a). CMIP3 has been an unprecedented international effort to run a coordinated set of 20th- and 21st-century climate simulations, as well as several climate change commitment experiments. The CMIP3multimodel dataset has provided a new era in climatechange research(Meehl et al., 2007a).

In the beginning stage of CMIP, the preliminarypurpose of the project is simple, i. e., to provide climatescientists with a database of coupled GCM simulationsunder st and ardized boundary conditions. Thus, theinternational investigators can use the model outputto simply identify aspects of the simulations in which"consensus" in model predictions or common problematic features exist(Meehl et al., 1997). By diagnosingthe model outputs, scientists can also discover and underst and why different models give different outputs inresponse to an identical forcing. Following the implementation of CMIP, the purpose of the project hasbeen greatly enriched. In the subsequent phases ofCMIP, in addition to the fundamental experiments ofthe earlier phases of CMIP, more and more new experiments are designed. In the ongoing Coupled ModelIntercomparsion Project-Phase 5(CMIP5), the total amount of listed core experiments, and Tier-1 and Tier-2 experiments is far more than that of the pastphases of CMIP. CMIP5 aims to provide a multimodelcontext for "assessing the mechanisms responsible formodel differences in poorly understood feedbacks associated with the carbon cycle and clouds; examiningclimate predictability and exploring the predictive capabilities of forecast systems on decadal timescales; and , more generally, determining why similarly forcedmodels produce a range of responses"(Taylor et al., 2012).

Nearly the entire international climate modelingcommunity has participated in the CMIP project sinceits inception in 1995. In the ongoing CMIP5 project, more than 20 international modeling groups/centershave performed simulations using more than 40 models. The Chinese climate modeling community hasa long history of climate model development. Recognizing the central importance of climate models inclimate studies, the Sate Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophys-ical Fluid Dynamics(LASG)at the Institute of Atmospheric Physics(IAP), Chinese Academy of Sci-ences(CAS), has invested huge efforts in GCM developments since the establishment of the laboratoryin 1985. Many pioneering Chinese models have beendeveloped at LASG/IAP(Zhang et al., 2000). Modelexperiments in the context of the various phases ofCMIP from CMIP1 to CMIP3 have been run in support of all the assessment reports of the IPCC(seereviews by Zhou et al., 2007; Yu et al., 2008; Wang et al., 2009; Zhang et al., 2010).

From CMIP1 to CMIP3, the LASG/IAP modelhas been the unique model from China that participated in CMIP. This situation has changed in CMIP5. In the past years, more laboratories and research centers of China have been engaged in the efforts ofclimate model development. There are five climatemodels, i. e., BCC-CSM1-1, BNU-ESM, FGOALS-g2, FGOALS-s2, and FIO-ESM, that have participatedin CMIP5. The main motivation of this study is toassess the general performances of these five Chineseclimate models in the context of CMIP5 core experiment. This paper can serve as a useful reference and provide an overview of Chinese contribution toCMIP5.

The remainder of the paper is organized as follows. In Section 2, we present an overview of Chinese models that participated in CMIP5. In Section 3, we introduce the model data, the observational or reanalysis data used as observational evidence for modelvalidation. In Section 4, we evaluate the models'performances in the context of climate mean states, seasonal cycle, intra-seasonal oscillation, interannualvariability, interdecadal variability, global monsoon, Asian-Australian monsoon, 20th-century historical climate simulation, climate change projection, and climate sensitivity. A summary is given in Section 5. 2. Overview of Chinese contribution toCMIP5

Five Chinese Climate/Earth System Models haveparticipated in the ongoing CMIP5 projects. As listedin Table 1, these five models include: 1)BCC-CSM1-1 from Beijing Climate Center, China MeteorologicalAdministration(Wu et al., 2010, 2014); 2)BNU-ESMfrom Beijing Normal University(Wu et al., 2013); 3)two versions of FGOALS2 model, i. e., FGOALS-g2(Li et al., 2013) and FGOALS-s2(Bao et al., 2013), fromthe IAP, CAS; 4)FIO-ESM from First Institute ofOceanography, State Oceanic Administration of China(Qiao et al., 2013). More detailed model informationcan be found in Table 1.

Table 1. Institution, model designation, and horizontal and vertical resolution of the five Chinese Climate/EarthSystem Models used in this study

Although the five coupled models are different intheir components, the model resolutions are, however, generally lower than most CMIP5 models, especiallyfor the atmospheric components as evidenced in Table 2. This suggests that the resolutions of Chinese models need to be increased, and the focus of the climatemodeling community in China should be the development of atmospheric models with higher resolutions.

Table 2. Horizontal resolution of CMIP5 models

In addition, some Chinese models listed in Table 1 have different versions that employ different resolutions. For example, the st and ard version of FGOALS-g2 holds a horizontal resolution of 2. 8°× 2. 8° in itsatmospheric component, while a lower resolution version, whose atmospheric resolution is 5. 0°× 4. 0°, isused to run the past millennial simulation(Zhou et al., 2008a, 2011). The second example is the Beijing Climate Center(BCC)model. In addition to thest and ard version BCC-CSM1-1, whose horizontal resolution is 2. 8°× 2. 8° in its atmospheric component, a higher resolution version named BCC-CSM1-1(m), with a horizontal atmospheric resolution of 1. 125°×1. 125°, is also used to run parts of CMIP5 experiments(Wu et al., 2014). We should note that thehorizontal resolution of the atmosphere general circulation model(AGCM)in BCC-CSM1-1(m)is compa-rable to the multimodel mean of CMIP5 models(cf. Table 2).

Not all Chinese models have performed all CMIP5experiments. A statistical comparison of CMIP5 experiments performed by different Chinese models isgiven in Table 3. While nearly all the models havecompleted the core experiments of CMIP5, there arespreads among the models in performing the Tier-1 and Tier-2 experiments of CMIP5(Taylor et al., 2012). Both BCC and IAP models have completed majorparts of CMIP5 Tier-1 and Tier-2 experiments. TheEarth System Model(ESM)experiments were also runby using BCC, BNU models and partly by using IAPmodels. Only part of CMIP5 experiments were run byusing FIO-ESM.

Table 3. List of Chinese models' CMIP5 experiments*

The data outputs of Chinese models are huge, as other CMIP5 models. Our statistics show thatthe number of completed CMIP5 experiments is 80, 12, 16, 28, 23, and 8 for BCC-CSM1-1, BCC-CSM-1-1(m), BNU-ESM, FGOALS-g2, FGOALS-s2, and FIO-ESM, respectively. The corresponding number oftotal datasets archived in the ESG-node is 3328, 283, 241, 899, 186, and 170, respectively. It has been agreat challenge to treat and archive all these datasets.

Although the five Chinese CMIP5 models are developed by four institutions, there are intimate collaborations among the Chinese climate modeling agen-cies. A tangible example of success in interagency collaboration is FGOALS model. The development ofFGOALS2 is a team effort and involves a broad crosssection of expertise and graduate students from bothwithin and outside LASG/IAP. The two versions ofFGOALS2, i. e., FGOALS-g2 and FGOALS-s2, sharea common coupling framework and common ocean and l and components but employ different atmospheric and sea ice components. Although the developmentsof both versions are being undertaken primarily atLASG/IAP, many individuals and institutions havecontributed in substantive ways to the developmentsof the FGOALS2 model. For example, the First Institute of Oceanography of the State Oceanic Admin-istration has contributed to the setup of the couplerused in both versions of FGOALS2. The Center ofEarth System Science(CESS)at Tsinghua Universityhas contributed to optimization of the code for the atmospheric component of FGOALS-g2, and parts of theFGOALS-g2 CMIP5 experiments were run by the supercomputer at Tsinghua University. The State KeyLaboratory of Atmospheric Boundary Physics and Atmospheric Chemistry(LAPC)of IAP has contributedto the ocean carbon cycle module of FGOALS-s2, and the School of Atmospheric Sciences at Nanjing University has contributed to the assessment and improve-ment of the FGOALS models. The supercomputingcenters of the IAP and the Chinese Academy of Sciences have provided most of the computer resources re-quired for the FGOALS CMIP5 experiments. This cooperative effort between several research groups(bothwithin LASG/IAP and from research centers and institutions outside LASG/IAP)is crucial to achievingthe successful development of FGOALS2, although achallenging degree of coordination is needed, as whatwe have done in the past six years. The success ofFGOALS CMIP5 experiment should be an encouraging beginning in promoting model collaborations inChina. 3. Data and analysis method

The observational datasets used in this study arelisted as follows:

1)Monthly sea surface temperature(SST)datafrom the Met Oflce Hadley Centre's sea ice and SSTdataset(HadISST; Rayner et al., 2003);

2)Monthly and pentad precipitation data fromthe Global Precipitation Climatology Project(GPCP; Adler et al., 2003; Xie et al., 2003), and CPC MergedAnalysis of Precipitation(CMAP; Xie and Arkin, 1997);

3)850-hPa wind from the National Centers forEnvironment Prediction-Department of Energy Atmospheric Model Intercomparison Project II Reanalysis(NCEP2; Kanamitsu et al., 2002), and Japan Meteorological Agency and the Central Research Institute of Electric Power Industry Reanalysis-25(JRA-25; Onogi et al., 2007);

4)850- and 200-hPa horizontal wind, 500-hPap-velocity, 700-hPa relative humidity, air temperature and specific humidity from 1000 to 1 hPa fromERA-Interim produced by the European Centre forMedium-Range Weather Forecasts(ECMWF; Dee et al., 2011);

5)Surface air temperature from NASA GISS Surface Temperature Analysis(GISTEMP; Hansen et al., 2010).

The outputs of five Chinese Climate/Earth System Models' CMIP5 experiments are analyzed in thisstudy. Note that the BCC model has two versions and only the data of BCC-CSM1-1 are analyzed, since wefocus on large-scale features of the climate and the twomodel versions are similar in their performance basedon the metrics used in our evaluation. The analysismainly focuses on the 20th-century historical climatesimulation. We employ on the first realization of historical experiment of each model to evaluate the simulated climatology and variability.

In addition, the abrupt 4×CO2 and piControl experiments of 18 CMIP5 models are used for estimatingclimate sensitivity of the models. In the analysis of the20th-century historical climate simulation and futureclimate change projection, the multimodel ensemblemean of 35 CMIP5 models(van Oldenborgh, 2013)is also used. The details of CMIP5 experiments arereferred to Taylor et al. (2012).

In the present study, the evaluation is mainly forthe period 1986-2005. The climate mean state is defined as an average of 1986-2005. The empirical orthogonal function(EOF)for SST is used to reveal ElNi~no-Southern Oscillation(ENSO)mode(20flS-20°N, 120flE-80flW) and Pacific Decadal Oscillation(PDO)mode(20fl-60°N, 115flE-115flW). For global monsoon, the monsoon mode(i. e., solstitial mode)is derivedfrom the rainfall difference between June-September and December-March, whereas the spring-fall asymmetric mode(i. e., equinoctial asymmetric mode)is defined as the rainfall difference between April-May and October-November(Wang and Ding, 2008; Lin R. et al., 2013; Zhang and Zhou, 2014). Linear regression isused to investigate the circulation patterns related toENSO and monsoon. Filtering method is used for extracting the intraseasonal and interdecadal signals inthe assessment of Madden-Julian Oscillation(MJO) and PDO, respectively. Trend analysis is applied tothe 20th-century warming and future climate projection. Gregory-style method is used to estimate climatesensitivity of coupled models(Gregory et al., 2004). 4. Results4. 1 SST mean state and seasonal cycle

To examine the performance of the five Chinesemodels in simulating SST distribution, we firstly assessthe simulated climatological SST pattern by comparing model results to HadISST(Rayner et al., 2003). The simulated SST patterns and their bias relativeto the observation are shown in Fig. 1. The warmpool areas(surrounded by 28℃ isotherms)in the western Pacific and Indian Ocean are more confined to theequator in these five models than in observation. Inaddition, the warm pool in models, except FGOALSg2, extends eastward compared to observation. Thelarge warm biases of SST mainly are located in thePacific cold tongue in all the five Chinese models. Thebias in FGOALS-g2 might be related to the overestimation in surface shortwave radiation and the weakcoastal oceanic upwelling(Lin P. F. et al., 2013). Inthe Indo-Pacific warm pool, cold biases dominate inFGOALS-s2, FGOALS-g2, and BCC-CSM1-1, whilewarm biases dominate in BNU-ESM and FIO-ESM. Toquantitatively compare the simulation skills of thesemodels, the weighted mean SST bias over the globalocean and the root-mean-square-error(RMSE)between each simulated pattern and the observation arecalculated and shown in Fig. 1. The results show thatall the five models exhibit a colder SST, with biases of-0. 85, -0. 87, -1. 59, -0. 96, and -0. 88 K, respectively. Compared to other four models, BCC-CSM1-1 has asmaller cold bias and RMSE and is closer to observation.

Fig. 1. (a)Observed climatological SST pattern(℃)for 1986-2005 of HadISST. Simulated climatological SST(contours) and the difference(simulation minus observation)(℃)between the simulated SST and that from HadISST(shaded)in(b)BCC-CSM1-1, (c)BNU-ESM, (d)FGOALS-g2, (e)FGOALS-s2, and (f)FIO-ESM. The numbers on the top rightof each figure denote the mean of model bias and the RMSE relative to observation(in bracket), respectively. The darkthick lines represent 28℃ isotherms.

The observed SST shows a significant annual cyclein the eastern Pacific because of dominant oceanic dynamic processes, while a semiannual cycle in the western Pacific because of prevailing thermodynamic processes(Fig. 2a). This characteristic is well capturedby the five Chinese models. However, most modelsshow that the simulated cold tongue shifts westwardby about 10fl. The results are consistent with thatof Yu et al. (2013)who have compared FGOALS2with earlier version FGOALS1. Yu et al. (2013)alsoshow that the improvement of FGOALS2 relative toFGOALS1 in the simulation of SST annual cycle issignificant. The annual cycle in the eastern Pacific isweaker in FGOALS-g2, BCC-CSM1-1, and FIO-ESMthan in observation(Figs. 2b, 2d, and 2f). The intensity of the annual cycle in the eastern Pacific and thesemiannual cycle in the western Pacific in FGOALS-s2 and BNU-ESM are similar to those in observation(Figs. 2c and 2e). In conclusion, the comparison tothe observation indicates reasonable performances ofthe five Chinese models in simulating SST mean state and seasonal cycle.

Fig. 2. Seasonal cycle of tropical SST(℃)averaged between 5±N and 5±S from(a)observation(HadISST) and simulations of(b)BCC-CSM1-1, (c)BNU-ESM, (d)FGOALS-g2, (e)FGOALS-s2, and (f)FIO-ESM.
4. 2 MJO

MJO is the most prominent intraseasonal phenomenon in the tropics. The precipitation data sub-jected to a 20-100-day b and -pass filter based on harmonic decomposition represent the intraseasonal com-ponent in the following analyses. We use 20(1986-2005)winters(November-April)data to validate thesimulations and examine the biases of MJO characteristics.

Because the FIO has not submitted the daily datato the CMIP5 dataset archives up to now, only the restfour Chinese models are analyzed here. The horizontaldistribution of the intraseasonal precipitation varianceduring boreal winter presenting the intensity of MJOis shown in Fig. 3. The four models basically producethe MJO signal, and the pattern correlation coeflcientbetween models and CMAP are around 0. 7. However, BCC-CSM1-1 obviously overestimates the MJO variability, while BNU-ESM and FGOALS-g2 underestimate it compared to the observation.

Fig. 3. 20-100-day filtered precipitation variance(mm2 day-2)during boreal winter(November-April)of 1986-2005derived from(a)CMAP, (b)GPCP, (c)BCC-CSM1-1, (d)BNU-ESM, (e)FGOALS-g2, and (f)FGOALS-s2. Thecorrelation coe±cient with CMAP is labeled on the top right of each panel.

To provide a comprehensive evaluation on theperiod and propagation simulations of MJO, Fig. 4 shows the wavenumber-frequency spectrum of 20-100-day filtered precipitation anomaly during borealwinter. In the observation, the eastward propagation of MJO is dominant and the ratio between theMJO variance and the variance of westward counterpart is around 2. 1. The strongest energy spectrum appears during the period of 45-60 days. Forthe four model spectrums, eastward over westwardpower is too weak on MJO timescales. The ratiosin BNU-ESM and FGOALS-s2 being nearly equal toone show st and ing oscillations instead of the dominanteastward-propagating signals. The significant periodhas the spectral peak in 30 days in BCC-CSM1-1 and FGOALS-s2, which is shorter than that in CMAP and GPCP(Figs. 4c and 4f).

Fig. 4. Wavenumber-frequency spectra(mm2 day-2)of 20-100-day filtered precipitation averaged between 20±S and 20±N during November-April 1986-2005 in(a)CMAP, (b)GPCP, (c)BCC-CSM1-1, (d)BNU-ESM, (e)FGOALS-g2, and (f)FGOALS-s2. Ratio between the MJO variance and the variance of westward counterpart(wavenumbers 0-6, 20-100-day mode)is labeled on the top right of each panel.
4. 3 ENSO

ENSO is the strongest signal of interannual climate variability originated from air-sea interaction inthe tropical Pacific(Philander, 1990). To extractENSO signal objectively, we apply EOF analysis tomonthly SST anomaly(SSTA)in the tropical Pacific(20flS-20°N, 120flE-80flW)from 1979 to 2005 for theobservation data and each model run. The first EOFmodes are ENSO modes for all the model runs. Then, we regress SST, precipitation, and 850-hPa wind fieldsonto the normalized first principal component(PC)time series(Fig. 5).

Fig. 5. (a, b, c, d, e, f)ENSO-related SST anomalies(K)in the observation and five model runs. The SST anomaliesare derived from the EOF analysis applied to the monthly SSTA in the tropical Pacific. (g, h, i, j, k, l)Correspondingprecipitation(mm day-1) and 850-hPa wind anomalies(m s-1).

All the five models can simulate the warm SSTAin the equatorial central-eastern Pacific, though thespatial patterns of the warm SSTA are somewhat different from the observation(Figs. 5a-f). Firstly, thewarm SSTA simulated by BNU-ESM, FGOALS-g2, and FGOALS-s2 excessively extends westward relative to that in the observation. Secondly, meridionalwidths of the warm SSTA in all the five models are narrower than that in the observation. For the amplitude, ENSOs simulated by BCC-CSM1-1, FGOALS-g2, and FGOALS-s2 are close to that in the observation, whilethose simulated by BNU-ESM and FIO-ESM are muchstronger.

The warm SSTA enhances the local convection. However, because of the nonlinear response of thetropical convection to underlying SST anomalies(Hoerling et al., 1997), the positive precipitation anomalies are primarily located in the equatorial central Pacific(Fig. 5g). Meanwhile, equatorial low-level westerly anomalies extend from about 150flE to 100flW. The westerly anomalies correspond to the weakeningof the easterly trade wind, and play essential roles inboth the positive Bjerknes feedback(Bjerknes, 1969) and the negative feedback mechanisms, such as delayed oscillator theory(Suarez and Schopf, 1988), or recharge-discharge theory(Jin, 1997). All thefive models can reproduce the positive precipitationanomalies and westerly anomalies reasonably(Figs. 5g-1). The major discrepancies are the somewhatwestward extension of the precipitation and westerlyanomalies in BNU-ESM, FGOALS-g2, and FGOALSs2, as well as their eastward shifts in FIO-ESM.

In the observation, El Ni~no events generally establish in boreal summer, reach peak phases in bo-real winter and then decay fast, which is referred toas the phase locking of ENSO to the seasonal cycle(Rasmusson and Carpenter, 1982). To assess the fivemodel performances in simulating this characteristics, we select strong El Ni~no events, with normalized PCindices greater than 1, and then make composite analysis. BNU-ESM, FIO-ESM, and FGOALS-g2 can rea-sonably reproduce the phase locking of ENSO, whilethe characteristic is not significant in BCC-CSM1-1 and FGOALS-s2(figure omitted).

Except for the above basic characteristics ofENSO, many other important characteristics deservefurther evaluations, such as temporal evolution, frequency, strengths of various feedbacks, asymmetry between El Ni~no and La Ni~na, ENSO-monsoon relationship, and many others(readers may refer to Bellenger et al., 2013). The ENSO flavors simulated by differentcoupled models would impact the performance of themodels in simulating ENSO-related climate anomaliessuch as ENSO-monsoon relationship(Wu and Zhou, 2013). 4. 4 PDO

PDO is the strongest interdecadal variability ofSST in the northern Pacific Ocean and can have a"far field" influence on climate through atmosphericteleconnections(Mantua and Hare, 2002). In ourstudy, PDO index is defined as the principal component of the first leading EOF mode of 9-30-yr b and -pass filtered SST over the North Pacific region(20fl-60°N, 115flE-115flW). We obtain the PDO-relatedSST anomalies by regressing the SST onto the st and ardized PDO index(Fig. 6). The SST patterns as-sociated with positive PDO phase have a broad areaof positive SST anomalies in the tropical Pacific and along the west coast of North and South America, with negative SST anomalies in Northwest and Southwest Pacific(Fig. 6a). The five Chinese models canwell reproduce the PDO-related SST anomaly pattern, with explained variance from 31. 71% in FGOALS-s2to 37. 28% in FGOALS-g2 of the interdecadal SST variability in North Pacific, all lower than that in observation(42. 83%). The patterns of FGOALS-s2 and FIO-ESM are closer to that of observation, with correlationcoeflcient exceeding 0. 7(Figs. 6e and 6f). The pattern correlation coeflcients between each model and observation are all statistically significant at the 1%level by Student's t-test. Thus, all the five Chinesemodels can well capture the main characteristics ofPDO-related SST anomaly pattern. This providesa solid basis for the future study of PDO evolutionmechanisms as done by Dong and Zhou(2014)usingFGOALS-gl.

Fig. 6. PDO-related SST anomalies(℃)in(a)observation(HadISST) and (b)BCC-CSM1-1, (c)BNU-ESM, (d)FGOALS-g2, (e)FGOALS-s2, and (f)FIO-ESM. PDO index is defined as the principal component of the first leadingEOF of 9-30-yr b and -pass filtered SST over the North Pacific region(20±-60±N, 115±E-115±W). The numbers on thetop right of each figure denote the explained percent variance and the pattern correlation coe±cient with observation(in bracket), respectively.

Power spectral analyses of PDO index show thatthe dominant periods are about 12 and 17 yr forBCC-CSM1-1, 13 yr for BUM-ESM, 14 and 20 yr forFGOALS-g2, 12-13 and 20 yr for FGOALS-s2, and 12-14 yr for FIO-ESM(figure omitted). In the observation, the dominant time period of PDO is about 20yr during 1900-2008(Deser et al., 2010). The comparison indicates that while the Chinese models are ableto reasonably simulate the pattern of SST associatedwith PDO, weaknesses are still evident in the simulation of the time periods of PDO. The limitation ofthe models in this regard may impact its ability in thesimulation of PDO-related climate anomalies over themonsoon domain(Qian and Zhou, 2013). 4. 5 Global monsoon

The global monsoon system can be seen as apersistent global-scale overturning of atmosphere thatvaries with the time of year(Trenberth et al., 2000) and can be examined as a whole. It has been an important metric to evaluate the performance of climatemodels(Zhou et al., 2008b; Zhang et al., 2010; Lin et al., 2012, 2013; Kitoh et al., 2013). The climatologyof global monsoon is examined from three aspects proposed by Wang and Ding(2008), i. e., monsoon mode, spring-fall asymmetric mode, and global monsoon domain.

The distributions of monsoon and spring-fallasymmetric mode in observation and simulations areshown in Fig. 7. The observed monsoon mode showsan anti-symmetric pattern about the equator, whilethe spring-fall asymmetric mode shows a negative pattern over the Northern Hemispheric ocean and a positive pattern over the Asian continent and SouthernHemispheric ocean. The two modes are realisticallyreproduced by the five models, with the pattern correlation coeflcients(PCC)higher than 0. 75 in monsoonmode(Figs. 7a-f) and 0. 60 in spring-fall asymmetricmode(Figs. 7g-i). The monsoon mode over western North Pacific(WNP) and maritime continent isoverestimated by BCC-CSM1-1, but underestimatedby FGOALS-g2 and FGOALS-s2. In the spring-fallasymmetric mode, all models overestimate the magnitude over the tropical Pacific Ocean. The monsoonmode is best simulated by BNU-ESM(Fig. 7c), whichhas the highest PCC(0. 81) and lowest RMSE(1. 67), while for spring-fall asymmetric mode the highest PCCis seen in BNU-ESM(Fig. 7i) and lowest RMSE(1. 46)is seen in FGOALS-g2(Fig. 7j).

Fig. 7. Observed and simulated monsoon mode(left panels) and spring-fall asymmetric mode(right panels)of precipita-tion(shaded; mm day-1)that are derived from(a, g)GPCP, (b, h)BCC-CSM1-1, (c, i)BUN-ESM, (d, j)FGOALS-g2, (e, k)FGOALS-s2, and (f, l)FIO-ESM, respectively. Contours range from -3 to 7 by 2(no zero line). Monsoon(spring-fall asymmetric)mode is defined as precipitation in June-September(April-May)minus that in December-March(October-November).

The distributions of precipitation annual range(shaded) and monsoon domain(contours)in observation and simulations are shown in Fig. 8. In the observation, large annual range is seen in the monsoondomains(Fig. 8a). All models can well capture thedistribution of precipitation annual range and monsoon domains, with the PCC of annual range higherthan 0. 7(Figs. 8b and 8f). However, the northernwestern Pacific monsoon domain and North American monsoon are smaller than the observation in allmodels, except for BCC-CSM1-1. The weaker WNPmonsoon in FGOALS-s2 is due to its simulated colderSST over the western Pacific warm pool(Zhang and Zhou, 2014). In comparison, BNU-ESM shows thelargest PCC and smallest RMSE in simulating annualrange, and the monsoon domain in BCC-CSM1-1 isthe most similar to the observation.

Fig. 8. Observed and simulated annual ranges of precipitation(shaded; mm day-1) and global monsoon domain(contour)for(a)GPCP, (b)BCC-CSM1-1, (c)BUN-ESM, (d)FGOALS-g2, (e)FGOALS-s2, and (f)FIO-ESM. Themonsoon domain is delineated by the monsoon precipitation index, i. e., precipitation annual range normalized by theannual mean precipitation. The monsoon domains are the areas where annual range exceeding 300 mm and monsoonprecipitation index exceeding 0. 5(Wang and Ding, 2008).
4. 6 Genesis potential index(GPI)of tropicalcyclone

Tropical cyclone genesis potential index(GPI)isa useful tool for evaluating the performance of globalclimate models in the tropical cyclone(TC)genesissimulation. Previous study has used GPI to evalu-ate the LASG/IAP AGCM over western North Pacific(Tian et al., 2013). Following Murakami et al. (2011), GPI is defined as:

where η is absolute vorticity at 850 hPa, RH is relativehumidity at 700 hPa, Vmax represents the maximumpotential intensity(MPI)(Emanuel, 1988), Vs is themagnitude of vertical wind shear between 850 and 200hPa, and ω is the vertical wind velocity at 500 hPa.

The climatology of GPI during July-October of1986-2005 is shown in Fig. 9. The TC genesis positions are mainly distributed between 5fl and 20°Nover WNP. In the ERA-Interim reanalysis(Fig. 9a), the spatial structure of GPI reasonably resemblesthe climatology of TCs distribution. All models canwell reproduce the high values of GPI between 5° and 20°N over WNP, while most models overestimatethe magnitude of GPI except for BCC-CSM1-1(Fig. 9d). Note that both FGOALS-g2 and FGOALS-s2overestimate the GPI value to the south of Japan. The PCCs between models and ERA-Interim are 0. 82(BCC-CSM1-1), 0. 90(BNU-ESM), 0. 70(FGOALS-g2), 0. 75(FGOALS-s2), and 0. 82(FIO-ESM), respectively. The evaluation suggests that while the fiveChinese models can reasonably reproduce the spatial pattern of tropical cyclone genesis, weaknesses areevident in the simulation of GPI magnitude. Thisprovides a useful reference for the future development of high resolution models based on these models.

Fig. 9. Climatology of tropical cyclone genesis potential index from(a)ERA-Interim, (b)FGOALS-g2, (c)FGOALS-s2, (d)BCC-CSM1-1, (e)BNU-ESM, and (f)FIO-ESM. The black dots denote individual genesis events from 1986 to 2005.
4. 7 Asian-Austrian monsoon4. 7. 1 Mean precipitation and 850-hPa wind

The precipitation and wind are important indicators of the monsoon. The Asian-Australian monsoon is a rigorous test for climate models(Zhou et al., 2009a, b). To compare and evaluate the performances of the five Chinese Climate/Earth SystemModels on Asian-Australian monsoon(A-ASM; 20flS-50°N, 40fl-160flE), we focus on the climatology ofJune-September precipitation and 850-hPa wind during 1986-2005. The data are derived from the historical runs of CMIP5(Taylor et al., 2012).

The precipitation products derived from observations such as GPCP and CMAP used in this studyhave similar climatological patterns with a PCC of0. 93(Fig. 10a). Compared with GPCP, the five models can well capture the large-scale monsoon rainfall, with PCCs larger than 0. 7(Figs. 10b-f). BNU-ESMhas the highest skill of 0. 83, which is close to thebest CMIP5 model analyzed by Sperber et al. (2013). However, model biases are also prominent(Figs. 10h-1), larger than the uncertainty between the GPCP and CMAP(Fig. 10g). The rainfall at the south slope ofHimalayas is deficit in all the five models, which is lessevident in FGOALS-s2. Except FIO-ESM, the otherfour models show weak rainfall from southern Chinato Japan where Meiyu/Baiu/Changma front systemis located. The five models show consistent biases ofdeficit rainfall in tropical eastern Indian Ocean nearthe Sumatra, accompanied by the redundant rainfallover the Arabian Sea. The rainfall biases over theWNP have large uncertainty, with negative sign inFGOALS-g2 and FIO-ESM and positive sign in BCC-CSM1-1, BNU-ESM, and FGOALS-s2. The similarityof climatological pattern of precipitation and corresponding biases in BNU-ESM and FIO-ESM may bedominated by the similar atmospheric component usedin models though they have different oceanic components(Qiao et al., 2013; Wu et al., 2013).

Fig. 10. Climatology of June-September precipitation(mm day-1)over the Asian-Australian summer monsoon regionduring 1986-2005 for(a)GPCP and (b-f)historical run of different models, and bias pattern relative to GPCP for(g)CMAP and (h-1)historical run of different models. Contours in(g-1)range from -7 to 7 by 2(without zero line). Number on the top right of(a-f)is the PCC of CMAP and climate models relative to GPCP.

The 850-hPa wind reflects the monsoon circulation and water vapor transport at the low level. Thedifferences between NCEP2 and JRA-25 show highconsistence in reanalysis datasets, with PCC of 0. 98(Fig. 11a). In contrast to the climatological patternof rainfall, wind pattern has higher skill(Figs. 11b-f). The wind skills of all the five models are closeto other CMIP5 models(Sperber et al., 2013). Thecross-equatorial flow over the western Indian Ocean, westerly flow from Arabian Sea to the South ChinaSea, monsoon trough over the Bay of Bengal, southerlywind over East Asia are all reproduced except for somebiases. FGOALS-g2 has an evidently weak monsoontrough over the Bay of Bengal, with strong easterliesover the equatorial eastern Indian Ocean(Fig. 11j). Except BCC-CSM1-1, the other four models tend toproduce a southerly bias over China. Correspondingto the deficit rainfall, FGOALS-g2 and FIO-ESM showan anticyclone bias over the WNP(Figs. 11i and 11l). The similarity of wind bias in BNU-ESM and FIO-ESM may be caused by the similar version of atmospheric component as shown in the rainfall pattern.

Fig. 11. Climatology of June-September 850 hPa wind(m s-1)over the Asian-Australian summer monsoon regionduring 1986-2005 for(a)NCEP2 and (b-f)historical run of different models, and bias pattern relative to NCEP2 for(g)JRA-25 and (h-1)historical run of different models. Number on the top right of(a-f)is the PCC of JRA-25 and climate models relative to NCEP2.

The skills on A-ASM of the five models areshown for 850-hPa wind versus precipitation, relativeto NCEP2 and GPCP(Fig. 12). In the same picture, we show the "skill" of JRA-25 and CMAP and themulti-model ensemble in CMIP3 and CMIP5. BNU-ESM performs best among the five models, whereasthe skill of FOGOAL-g2 is the lowest due mainly tothe precipitation bias. The skills of FGOALS-s2, BCC-CSM1-1, and FIO-ESM are close to each other. From CMIP3 to CMIP5, the ensemble skills of climatology of the A-ASM have improved and are closerto the reanalysis/observation, which is mainly contributed by the improvement of precipitation.

Fig. 12. PCC of JJAS precipitation against 850-hPawind.
4. 7. 2 Interannual variability of Asian summer mon-soon

The spatial pattern of the ENSO-forced rainfall anomalies obtained from linear regression ofJJAS(June-July-August-September)rainfall anomalies onto JJAS Nino3. 4 SST anomalies is shown in Fig. 13. The GPCP rainfall data show the largest rainfalldecrease adjacent to the western Ghats and near thefoothills of the Himalayas, and a secondary rainfalldeficit over central India for El Ni~no conditions. Overnortheastern India and near the Burmese coast, abovenormal rainfall anomalies prevail in the GPCP ob-servations. The BNU-ESM and FIO-ESM simulations exhibit the deficit of rainfall adjacent to thewestern Ghats and over India, and enhanced rainfallnear Burma(Figs. 13c and 13f), but the FIO-ESMsimulation overestimates the observed deficit of rainfall. However, the BCC-CSM1-1, FGOALS-g2, and FGOALS-s2 simulations have enhanced rainfall overthe Arabian Sea, with the FGOALS-s2 simulation displaying additionally strong rainfall over the Bay ofBengal. As discussed in Sperber et al. (2013), thereare many critical factors influencing a realistic ENSO-monsoon teleconnection, including the model's ability to capture the mean monsoon rainfall distribution, the ENSO-related SST and diabatic heating anomalies along the equatorial Pacific, the correct simulation of regional SST anomalies over the tropical Indian Ocean and west Pacific, as well as indirect influences from the June-July and August-SeptemberIndian monsoon preceding boreal winter ENSO development.

Fig. 13. Interannual JJAS precipitation anomalies(mm day-1)from linear regression with JJAS Nino3. 4 SST anoma-lies:(a)GPCP rainfall data versus HadISST, (b)BCC-CSM1-1, (c)BNU-ESM, (d)FGOALS-g2, (e)FGOALS-s2, and (f)FIO-ESM. The number on the right top of(b-f)denotes the model pattern correlation with GPCP for the interannualJJAS precipitation anomalies. The GPCP rainfall data and the HadISST are for 1979-2006. The model data are for1961-1999.

The East Asian summer monsoon(EASM)is animportant component of the A-ASM. The distinc-tive topography and orography of East Asia produceunique features in the EASM. While the South Asianor Indian summer monsoon is purely a tropical mon-soon system, the EASM is composed of both tropical and subtropical systems(Zhou et al., 2009c). Thereare many indices for measuring the strength of theEASM(Wang et al., 2008), and we adopt the negativezonal wind shear index of Wang and Fan(1999)in thisstudy:

The interannual EASM JJA 850-hPa windanomalies and precipitation anomalies from linear regression with the WFN for JRA-25 reanalysis and GPCP rainfall data are shown in Fig. 14a. The pattern is characterized by enhanced precipitation alongthe East Asian subtropical front associated with interannual variations of the Meiyu/Baiu/Changma rainb and , and deficit rainfall over the western Pacific. Thepattern correlations of these anomalies with those derived from the CMAP rainfall data and the NCEP2reanalysis are 0. 94 and 0. 99, respectively(figure omitted), indicating that these features are robust characteristics of EASM variability. All of the five Chinesemodels reasonably represent the wind anomalies and the deficit rainfall anomalies adjacent to the west coastof the Philippines. However, the models show poorperformance in representing the rainfall maxima thatextends from central China to Southwest Japan except for the BNU-ESM model. The BNU-ESM modelalso has the largest rainfall pattern correlation of themodels analyzed, with a reasonable representation ofthe rainfall minima adjacent to the west coast of thePhilippines, and the maxima over Southeast China and Southwest Japan(Fig. 14f). Compared to theatmospheric model analysis(Song and Zhou, 2013), the deficient rainfall anomalies over the western Pacific are considerably improved, indicative of the roleof air-sea interaction. However, the rainfall maximafrom central China to Japan is almost missing in mostatmospheric and coupled models, suggesting that theinterannual variation of Meiyu/Baiu/Changma is stilla challenge for current models.

Fig. 14. Interannual East Asian summer monsoon JJA 850-hPa wind anomalies and precipitation anomalies from linearregression with the revised JJA Wang and Fan(1999)850-hPa zonal wind index for(a)JRA-25/GPCP, (b)BCC-CSM1-1, (c)BNU-ESM, (d)FGOALS-g2, (e)FGOALS-s2, and (f)FIO-ESM. The numbers on the middle top of(b-f)are themodel pattern correlations with JRA-25 and GPCP over the region 0±-50±N, 100±-140±E. The units for the 850-hPawind anomalies and precipitation anomalies are m s-1 and mm day-1, respectively. The JRA-25 reanalysis and theGPCP rainfall data are for 1979-2006. The model data are for 1961-1999.
4. 8 20th-century historical climate simulation and RCP projection

The time evolution of global mean surface air temperature(SAT)simulated by the five Chinese modelsis compared to other CMIP5 models in Fig. 15a. Except for FGOALS-s2, all Chinese models show comparable increasing trend of SAT to other CMIP5 modelsduring the historical period(1861-2005). The surfacewarming trend in the historical period is about 0. 68℃(145 yr)-1 in the multi-model ensemble mean(MME)of 35 CMIP5 models. However, while the result ofFGOALS-g2 is comparable to the MME with a trendof 0. 63℃(145 yr)-1, all other Chinese models showlarger warming trend than the MME. The stronger in-creasing trend of SAT in FGOALS-s2, which is 1. 58℃(145 yr)-1, is related to too strong water vapor feed-back and sea-ice albedo feedback(Zhou et al., 2013) and high climate sensitivity(Chen et al., 2014).

Under RCP4. 5 scenario, BNU-ESM and FIO-ESM show the largest(1. 94℃(95 yr)-1) and smallest(0. 88℃(95 yr)-1)warming amplitude during 2005-2099, respectively, while the result of BCC-CSM1-1(1. 44℃(95 yr)-1)is close to the MME(1. 75℃(95yr)-1). The SAT in FGOALS-s2 begins to declinearound the year 2070, in opposite to other models and the reason has been investigated in He et al. (2014).

Under RCP8. 5 scenario(Fig. 15b), the MMEshows a larger warming trend than RCP4. 5 scenario, with the magnitude of 4. 09℃(95 yr)-1. The warmingtrends in FGOALS-s2(5. 13℃(95 yr)-1) and BNU-ESM(4. 66℃(95 yr)-1)are larger than the MME, while FGOALS-g2(3. 21℃(95 yr)-1), BCC-CSM1-1(3. 73℃(95 yr)-1), and FIO-ESM(3. 90℃(95 yr)-1)underestimate the warming trend as the MME. Thedifferent warming trends in different models are closelyrelated to their climate sensitivities(see Section 3. 9). As will be shown in Fig. 17, the climate sensitivitiesin FGOALS-s2 and BNU-ESM are the highest amongChinese models.

Fig. 15. Time series of surface air temperature(℃)averaged over global domain for(a)RCP4. 5 and (b)RCP8. 5scenarios. The reference period is 1986-2005 in the historical simulation. Gray lines show all the model simulations and black line show the multi-model ensemble mean results.

The linear trends of SAT of 5 Chinese models during 1950-2005 are displayed in Fig. 16. The modelresults are compared to the observational data of GIS-TEMP(Hansen et al., 2010). In the GISTEMP, thestrongest trends appear over the high-1atitude continent(Fig. 16a). This feature is well captured by allthe Chinese models. However, the observed coolingtrends over North Pacific and North Atlantic are stilldiflcult for climate models. The North Pacific coolingis partly reproduced in BCC-CSM1-1, while the NorthAtlantic cooling is evident in BNU-ESM, FGOALS-g2, and FIO-ESM. During 1950-2005, the global meanSAT trend is 0. 55℃(56 yr)-1 in GISTEMP, this trendis well captured by FGOALS-g2 which is 0. 72℃(56yr)-1. FGOALS-s2 has the largest trend(1. 28℃(56yr)-1), while other three models show similar trends, which are 1. 08, 1. 01, and 0. 96℃(56 yr)-1 for BCC-CSM1-1, BNU-ESM, and FIO-ESM, respectively.

Fig. 16. The linear trends of surface air temperature(℃(56 yr)-1)during 1950-2005 in(a)GISTEMP, (b)BCC-CSM1-1, (c)BNU-ESM, (d)FGOALS-g2, (e)FGOALS-s2, and (f)FIO-ESM. Black dots denote where the trends arestatistically significant at the 5% level.
4. 9 Climate sensitivity to greenhouse gasses

Climate sensitivity is an important aspect of Climate/Earth System Models. It is the main source ofuncertainty for future climate projection(Meehl et al., 2007b). The climate sensitivity of a coupled model isclosely related to its ability in reproducing the pastchanges of global climate driven by both natural and anthropogenic external forcing agents(Guo and Zhou, 2013). The climate sensitivity of the five Chinesemodels is evaluated using the benchmark experimentabrupt 4×CO2 in CMIP5 based on Gregory-style regression method(Gregory et al., 2004; Taylor et al., 2012). First we show the 150-yr evolution of globallyaveraged SAT under the forcing of quadrupled CO2relative to pre-industrial period. The change of SATis relative to the piControl experiment.

Because the FIO has not submitted its abrupt4×CO2 experiment to the CMIP5 dataset archives upto now, only the rest four Chinese models are showntogether with other 14 CMIP5 models(Fig. 17a). Aprominent feature is that SAT increases dramaticallyduring the first 10-20 years, especially in BNU-ESM and FGOALS-s2. In contrast, the evolutions of SAT inBCC-CSM1-1 and FGOALS-g2 are close to the MME. In general, the four Chinese models are within therange of SAT of the 18 models.

Fig. 17. (a)Evolution of globally averaged SAT(K)of abrupt 4×CO2 relative to piControl for 4 Chinese models(colored thick line), other 14 CMIP5 models(gray thin line), and multi-model ensemble(black thick line). (b)Climatesensitivity estimated by Gregory-style method represented by the equilibrium ¢T under the forcing of quadrupled CO2.

Based on a detailed analysis of climate sensitivity for FGOALS-g2 and FGOALS-s2 by Chen etal. (2014), climate sensitivity that estimated fromGregory-style method may be affected by the responseduring the first 10-20 years. Compared with BCC-CSM1-1 and FGOALS-g2, BNU-ESM, and FGOALS-s2 have similar trend of SAT response during 30-150yr. However, because of the rapid increase of SATduring the first 20 years(Fig. 17a), the estimatedequilibrium temperature in FGOALS-s2 is 3 K higherthan BCC-CSM1-1(Fig. 17b). The rapid increase ofSAT in FGOALS-s2 is attributed to the strongly positive water vapor and sea-ice albedo feedback(Chen et al., 2014). The high climate sensitivity of FGOALS-s2causes the warming trend larger than other models inhistorical and RCP4. 5 experiments(He et al., 2014). A preliminary diagnosis shows that the excessive water vapor cloud at the North Pole may be responsiblefor the extreme warming through water vapor-cloud-radiation positive feedback(He et al., 2014). 5. Summary and concluding remarks

The contribution of China to the ongoing CMIP5project includes five ESM/CSM, which are calledBCC-CSM1-1, BNU-ESM, FGOALS-g2, FGOALS-s2, and FIO-ESM, respectively. After providing an overview of Chinese models that participated in CMIP5, the performances of these five Chinese models are evaluated based on the historical, RCP projection, abrupt4×CO2, and piControl run of CMIP5. The assessmentfocuses on the mean state of SST and its seasonal cycle, typical climate variability across intra-seasonal, interannual and interdecadal timescales(MJO, ENSO, and PDO), global monsoon, tropical cyclone, Asian-Australian monsoon in boreal summer, 20th-centuryglobal warming, future climate change projection, and climate sensitivity. A summary of performance for thefive Chinese Climate/Earth System Models is listed inTable 4. Major conclusions are summarized as follows.

Table 4. Summary of performances of the five Chinese Climate/Earth System Models

1)Great progress has been made in the development of climate models from CMIP3 to CMIP5 inChina. Five coupled Climate/Earth System Modelshave participated in CMIP5. In addition to the traditional modeling center such as LASG/IAP, new climate modeling centers are emerging in China. Interagency collaboration has promoted the development and improvement of Chinese models. In comparisonto other CMIP5 models, the resolutions of Chinesemodels are generally lower, especially for the atmospheric components.

2)All the five Chinese models show reasonableperformances in simulating SST mean state and seasonal cycle. The annual cycle in the eastern Pacific isweaker in FGOALS-g2, BCC-CSM1-1, and FIO-ESM, while that in FGOALS-s2 and BNU-ESM are similarto observation.

3)The main MJO characteristics including theintensity, significant period, and propagation are basically reproduced. The MJO variance is overestimated(underestimated)by BCC-CSM1-1(BNU-ESM and FGOALS-g2). The MJO westward propagation powerin the models is too strong. The significant period inBCC-CSM1-1 and FGOALS-s2 is shorter than that inthe observation.

4)All the five Chinese models show reasonableperformances in simulating the ENSO-related SSTA, however, those in the equatorial central-eastern Pacificsimulated by BNU-ESM, FGOALS-g2, and FGOALS-s2 extend westward excessively; ENSO amplitudesin BCC-CSM1-1, FGOALS-g2, and FGOALS-s2 areclose to that in the observation, while those in BNU-ESM and FIO-ESM are much stronger; El Ni~no eventssimulated by FGOALS-s2 and BCC-CSM1-1 do notshow strong phase locking to the seasonal cycle.

5)All the five Chinese models can well capture themain characteristics of PDO-related SSTA pattern, with the highest PCC to observation in FGOALS-s2 and FIO-ESM. But the simulated dominant time period of PDO is model-dependent.

6)The climatology and domain of global monsoon precipitation are well reproduced by five Chi-nese models. The monsoon mode over WNP in BCC-CSM1-1(FGOALS-g2 and FGOALS-s2)is stronger(weaker)than the observation. All models simulatestronger spring-fall asymmetric mode over the tropical Pacific compared with the observation. The WNPmonsoon and North American monsoon domains in allmodels are smaller than the observation, except BCC-CSM1-1.

7)The climatology of GPI distribution is wellsimulated in all models, while most models overes-timate the magnitude. The overestimation of GPIto the south of Japan is evident in FGOALS-g2 and FGOALS-s2.

8)The climatology of large-scale Asian monsoonprecipitation and 850-hPa wind can be well capturedby all the five models. Using PCC as a metric, BNU-ESM has the highest skill whereas the performanceof FGOALS-g2 needs to be improved. The rainfallskill, of which the spread is larger than that of thewind, is the major limitation to individual model performance compared with the reanalysis and CMIP ensemble mean.

9)The time evolution of SAT during the historical period(1861-2005)is well simulated by Chinese models, except for the strong warming trend inFGOALS-s2. Compared to CMIP5 MME, the warming trend in the historical period is overestimated bymost Chinese models. The warming trend in the MMEduring 2005-2099 under RCP4. 5 scenario is well captured by BCC-CSM1-1. Under RCP8. 5, all Chinesemodels are comparable to CMIP5 models. FGOALS-s2 and BNU-ESM(FGOALS-s2, BCC-CSM1-1, and FIO-ESM)overestimate(underestimate)the warmingtrend compared to the MME. The different warmingtrends in different models are closely related to theirclimate sensitivities.

10)The overall warming pattern during 1950-2005 is well captured by five Chinese models. Allmodels have the polar amplification phenomenon, which is most evident in FGOALS-s2 but weak inFGOALS-g2. The observed cooling trends over NorthPacific and North Atlantic are not evident in the simulation. The linear trends of SAT during 1950-2005are overestimated in all Chinese models.

11)Among 18 CMIP5 models, climate sensitivities of FGOALS-g2 and BCC-CSM1-1(BNU-ESM and FGOALS-s2)are equivalent or lower(higher)thanthat of the MME based on the Gregory-style method. The high sensitivities of BNU-ESM and FGOALS-s2are caused by the dramatic increase of temperatureunder CO2 forcing during the first two decades. Acknowledgments: We acknowledge Dr. XiaoXia and Mr. Chen Kangjun for helping us to collectstatistical data of CMIP5 models from the ESG-node.

References
[1] Adler, R. F., G. J. Huffman, A. Chang, et al., 2003: The version-2 global precipitation climatology project 11 (GPCP) monthly precipitation analysis (1979-present). J. Hydrometeorology, 4, 1147-1167.
[2] Bao, Q., P. F. Lin, T. J. Zhou, et al., 2013: The flex-ible global Ocean-Atmosphere-Land system model, spectral version 2: FGOALS-s2. Adv. Atmos. Sci., 30, 561-576, doi: 10.1007/s00376-012-2113-9.
[3] Bellenger, H, E. Guilyardi, J. Leloup, et al., 2013: ENSO representation in climate models: From CMIP3 to CMIP5. Climate Dyn., 42, 7-8, doi: 10.1007/s00382-013-1783-z.
[4] Bjerknes, J., 1969: Atmospheric teleconnections from the equatorial Pacific. Mon. Wea. Rev., 97, 163-172.
[5] Chen, X. L., T. J. Zhou, and Z. Guo, 2014: Climate sensitivities of two versions of FGOALS model to idealized radiative forcing. Sci. China Earth Sci., 57, 1363-1373, doi: 10.1007/s11430-013-4692-4.
[6] Dee, D. P., S. M. Uppala, A. J. Simmons, et al., 2011: The ERA-Interim reanalysis: Configuration and perfor-mance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553-597.
[7] Deser, C., M. A. Alexander, S.-P. Xie, et al., 2010: Sea surface temperature variability: Patterns and mech-anisms. An. Rev. Mar. Sci., 2, 115-143.
[8] Dong Lu and Zhou Tianjun, 2014: Contributions of nat-ural and anthropogenic forcings to the twentieth-century Pacific sea surface temperature variabil-ity simulated by a climate system model. Acta Oceanologica Sinica, 36, 48-60. (in Chinese)
[9] Emanuel, K., 1988: The maximum intensity of hurri-canes. J. Atmos. Sci., 45, 1143-1155.
[10] Gregory, J. M., W. J. Ingram, M. A. Palmer, et al., 2004: A new method for diagnosing radiative forc-ing and climate sensitivity. Geophys. Res. Lett., 31, L03205, doi: 10.1029/2003GL018747.
[11] Guo, Z., and T. J. Zhou, 2013: Why does FGOALS-gl re-produce a weak Medieval Warm Period but a reason-able Little Ice Age and 20th century warming? Adv. Atmos. Sci., 30, 1758-1770, doi: 10.1007/s00376-013-2227-8.
[12] Hansen, J., R. Ruedy, M. Sato, et al., 2010: Global surface temperature change. Rev. Geophys., 48, RG4004, doi: 10.1029/2010RG000345.
[13] He, B., Y. Liu, T. Zhou, et al., 2014: A preliminary diagnosis of high climate sensitivities simulated by FGOALS-s2 in CMIP5 Historical and RCP4. 5 sce-narios. Flexible Global Ocean-Atmosphere-Land Sys-tem Model. Zhou T. J., Y. Q. Yu, Y. M. Liu, et al., Eds., Springer Press, 225-231.
[14] Hoerling, M. P., A. Kumar, and M. Zhong, 1997: El Niño, La Niña, and the nonlinearity of their telecon-nections. J. Climate, 10, 1769-1786.
[15] Jin, F. F., 1997: An equatorial ocean recharge paradigm for ENSO. Part I: Conceptual model. J. Atmos. Sci., 54, 811-829.
[16] Kanamitsu, M., W. Ebisuzaki, J. Woollen, et al., 2002: NCEP-DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 1631-1643.
[17] Kitoh, A., H. Endo, K. K. Kumar, et al., 2013: Mon-soons in a changing world: A regional perspective in a global context. J. Geophys. Res. Atmos., 118, doi: 10.1002/jgrd.50258.
[18] Li, L., P. Lin, Y. Yu, et al., 2013: The flexible global ocean-atmosphere-land system model: Grid-point version 2: FGOALS-g2. Adv. Atmos. Sci., 30, 543-560, doi: 10.1007/s00376-012-2140-6.
[19] Lin, P. F., Y. Q. Yu, and H. L. Liu, 2013: Oceanic climatology in the coupled model FGOALS-g2: Im-provements and biases. Adv. Atmos. Sci., 30, 819-840.
[20] Lin, R., T. Zhou, and Y. Qian, 2013: Evaluation of global monsoon precipitation changes based on five reanalysis datasets. J. Climate, 27, 1271-1289, doi: 10.1175/JCLI-D-13-00215.1.
[21] Lin Renping, Zhou Tianjun, Xue Feng, et al., 2012: The global monsoon variability revealed by NCEP/NCAR reanalysis data. Chinese J. Atmos. Sci., 36, 1027-1040. (in Chinese)
[22] Mantua, N., and S. Hare, 2002: The Pacific decadal os-cillation. J. Oceanogr., 58, 35-44.
[23] Meehl, G. A., G. J. Boer, C. Covey, et al., 1997: Inter-comparison makes for a better climate model. Eos. Trans. Amer. Geophys. Union, 78, 445-451.
[24] —-, G. J. Boer, C. Covey, et al., 2000: The Cou-pled Model Intercomparison Project (CMIP). Bull. Amer. Meteor. Soc., 81, 313-318.
[25] —-, C. Covey, B. McAvaney, et al., 2005: Overview of the Coupled Model Intercomparison Project. Bull. Amer. Meteor. Soc., 86, 89-93.
[26] —-, —-, T. Delworth, et al., 2007a: The WCRP CMIP3 multimodel dataset: A new era in climate change research. Bull. Amer. Meteor. Soc., 88, 1383-1394, doi: 10.1175/BAMS-88-9-1383.
[27] —-, T. F. Stocker, W. D. Collins, et al., 2007b: Global Climate Projections. Climate Change 2007: The Physical Science Basis. Contribution of Work-ing Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Changes. Solomon, S., et al., Eds., Cambridge University Press, 747-845.
[28] Murakami, H., B. Wang, and A. Kitoh, 2011: Future change of western North Pacific typhoons: Projec-tions by a 20-km-mesh global atmospheric model. J. Climate, 24, 1154-1169.
[29] Onogi, K., J. Tsutsui, H. Koide, et al., 2007: The JRA-25 reanalysis. J. Meteor. Soc. Japan, 85, 369-432.
[30] Philander, S. G. H., 1990: El Niño, La Niña, and the Southern Oscillation. Academic Press, 293 pp.
[31] Qian, C., and T. J. Zhou, 2013: Multidecadal variability of North China aridity and its relationship to PDO during 1900-2010. J. Climate, 27, 1210-1222, doi: 10.1175/JCLI-D-13-00235.1.
[32] Qiao, F. L., Z. Y. Song, Y. Bao, et al., 2013: Devel-opment and evaluation of an Earth System Model with surface gravity waves. J. Geophys. Res., 118, 4514-4524, doi: 10.1002/jgrc.20327.
[33] Rasmusson, E., and T. Carpenter, 1982: Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño. Mon. Wea. Rev., 110, 354-384.
[34] Rayner, N. A., D. E. Parker, E. B. Horton, et al., 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late 19th century. J. Geophys. Res., 108, 4407, doi: 10.1029/2002JD002670.
[35] Song, F. F., and T. J. Zhou, 2013: Interannual variability of East Asian summer monsoon simulated by CMIP3 and CMIP5 AGCMs: Skill dependence on Indian Ocean-western Pacific anticyclone teleconnection. J. Climate, 27, 1679-1697, doi: 10.1175/JCLI-D-13-00248.1.
[36] Sperber, K. R., H. Annamalai, I.-S. Kang, et al., 2013: Asian summer monsoon: An intercompari-son of CMIP5 vs. CMIP3 simulations of the late 20th century. Climate Dyn., 41, 2711-2744, doi: 10.1007/s00382-012-1607-6.
[37] Suarez, M. J., and P. S. Schopf, 1988: A delayed action oscillator for ENSO. J. Atmos. Sci., 45, 3283-3287. Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485-498.
[38] Tian Fangxing, Zhou Tianjun, and Zhang Lixia, 2013: Tropical cyclone genesis potential index over the western North Pacific simulated by LASG/IAP AGCM. Acta Meteor. Sinica, 27, 50-62, doi: 10.1007/s13351-013-0106-y.
[39] Trenberth, K. E., D. P. Stepaniak, and J. M. Caron, 2000: The global monsoon as seen through the divergent atmospheric circulation. J. Climate, 13, 3969-3993.
[40] van Oldenborgh, G. J., M. Collins, J. M. Arblaster, et al., 2013: Annex I: Atlas of Global and Regional Climate Projections. Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Inter-governmental Panel on Climate Change. Stocker, T. F., D. Qin, G.-K. Plattner, et al., Eds., Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 1311-1393.
[41] Wang, B., and Z. Fan, 1999: Choice of South Asian sum-mer monsoon indices. Bull. Amer. Meteor. Soc., 80, 629-638.
[42] —-, and Q. Ding, 2008: Global monsoon: Dominant mode of annual variation in the tropics. Dyn. At-mos. Oceans, 44, 165-183.
[43] —-, Z. Wu, J. Li, et al., 2008: How to measure the strength of the East Asian summer monsoon. J. Climate, 21, 4449-4463.
[44] Wang Bin, Zhou Tianjun, Yu Yongqiang, et al., 2009: A view of earth system model development. Acta Meteor. Sinica, 23, 1-17.
[45] Wu, B., and T. J. Zhou, 2013: Relationships between East Asian-western North Pacific monsoon and ENSO simulated by FGOALS-s2. Adv. Atmos. Sci., 30, 713-725, doi: 10.1007/s00376-013-2103-6.
[46] Wu, T., R. Yu, F. Zhang, et al., 2010: The Beijing climate center for atmospheric general circulation model (BCC-AGCM2. 0.1): Description and its per-formance for the present-day climate. Climate Dyn., 34, 123-147.
[47] Wu Tongwen, Song Liancun, Li Weiping, et al., 2014: An overview of progress in climate system model devel-opment at the Beijing Climate Center applications for climate change studies. Acta Meteor. Sinica, 28, 34-56, doi: 10.1007/s13351-014-3041-7.
[48] Wu Qizhong, Feng Jinming, Dong Wenjie, et al., 2013: Introduction of the CMIP5 experiments carried out by BNU-ESM. Adv. Climate Change Res., 9, 291-294, doi: 10.3969/j.issn.1673-1719.2013.04.008.
[49] Xie, P., and P. A. Arkin, 1997: Global precipitation: A 17-year monthly analysis based on gauge obser-vations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2539-2558.
[50] —-, J. E. Janowiak, P. A. Arkin, et al., 2003: GPCP pen-tad precipitation analyses: An experimental dataset based on gauge observations and satellite estimates. J. Climate, 16, 2197-2214.
[50] Yu, Y., H. Zhi, B. Wang, et al., 2008: Coupled model simulations of climate changes in the 20th century and beyond. Adv. Atmos. Sci., 25, 641-654.
[51] —-, J. He, W. P. Zheng, et al., 2013: Annual cycle and interannual variability in the tropical Pacific as sim-ulated by three versions of FGOALS. Adv. Atmos. Sci., 30, 621-637.
[52] Zhang Lixia, Zhou Tianjun, Wu Bo, et al., 2010: The annual modes of tropical precipitation simulated by the LASG/IAP coupled ocean-atmosphere model FGOALS? s1. 1. Acta Meteor. Sinica, 24, 189-202.
[53] —-, and —-, 2014: An assessment of improvements in global monsoon precipitation simulation in FGOALS-s2. Adv. Atmos. Sci., 31, 165-178, doi: 10.1007/s00376-013-2164-6.
[54] Zhang, X., G. Shi, H. Liu, et al., 2000: IAP Global Ocean-Atmosphere-Land System Model. Science Press, Bei-jing, 251 pp.
[55] Zhou, T. J., and R. Yu, 2006: Twentieth century surface air temperature over China and the globe simulated by coupled climate models. J. Climate, 19, 5843-5858.
[56] —-, Y. Yu, H. Liu, et al., 2007: Progress in the develop-ment and application of climate ocean models and ocean-atmosphere coupled models in China. Adv. Atmos. Sci., 24, 729-738.
[57] —-, B. Wu, X. Wen, et al., 2008a: A fast version of LASG/IAP climate system model and its 1000-year control integration. Adv. Atmos. Sci., 25, 655-672.
[58] —-, R. C. Yu, H. M. Li, et al., 2008b: Ocean forcing to changes in global monsoon precipitation over the recent half-century. J. Climate, 21, 3833-3852.
[59] —-, B. Wu, and B. Wang, 2009a: How well do at-mospheric general circulation models capture the leading modes of the interannual variability of the Asian-Australian monsoon? J. Climate, 22, 1159-1173.
[60] —-, B. Wu, A. A. Scaife, et al., 2009b: The CLIVAR C20C Project: Which components of the Asian-Australian monsoon circulation variations are forced and reproducible? Climate Dyn. , 33, 1051-1068.
[61] —-, D. Gong, J. Li, et al., 2009c: Detecting and un-derstanding of multi-decadal variability of the East Asian summer monsoon-Recent progress and state of affairs. Meteorologische Zeitschrift, 18, 455-467.
[62] —-, B. Li, W. Man, et al., 2011: A comparison of the me-dieval warm period, little ice age and 20th-century warming simulated by the FGOALS Climate System Model. Chinese Sci. Bull., 56, 3028-3041.
[63] —-, F. Song, and X. L. Chen, 2013: Historical evolu-tions of global and regional surface air temperature simulated by FGOALS-s2 and FGOALS-g2: How reliable are the model results? Adv. Atmos. Sci., 30, 638-657, doi: 10.1007/s00376-013-2205-1.