One of the major achievements of the interna-tional climate research community since the 1970s hasbeen the identification of climate system componentsalong with an extension of this field of research fromthe atmosphere to the whole climate system. The cli-mate system is an interactive system consisting of fivemajor components: the atmosphere, the hydrosphere, the cryosphere, the l and surface, and the biosphere.Addressing the interplay among the different compo-nents or establishing the climate system's response tochanges in natural(orbital factors, solar variation, and volcanic eruptions) and anthropogenic forcing agents(emissions of greenhouse gases, aerosols, and l and use)requires a coupled climate system model(CSM)orearth system model(ESM)approach. Coupled CSMs and ESMs have proved to be useful tools for under-st and ing the mechanisms of climate variability, repro-ducing past climate change, performing seasonal cli-mate predictions, and projecting potential future cli-mate change.

The worldwide development of atmospheric gen-eral circulation models(AGCMs), CSMs, and ESMshas been driven and encouraged by the World ClimateResearch Programme(WCRP). In the past 20 years, many international climate model intercomparisonprojects, such as AMIP(Atmospheric Model Inter-comparison Project; Gates et al., 1992) and CMIP(Coupled Model Intercomparison Project; Meehl et al., 1997, 2000), have been organized and coordinatedby WCRP. The successful implementations of theseprojects have greatly enhanced international collabo-ration in the field of climate model development and climate modeling activities. Both AMIP and CMIPare among the list of successful international collab-orations in climate research and climate change sci-ence. The implementation of CMIP has also advancedthe progress of climate change modeling and climatechange projection around the world. Many peer-reviewed publications based on the outputs of CMIPmodels have been cited by the assessment reportsof the Intergovernmental Panel on Climate Change(IPCC).

The Chinese climate research community beganto develop climate models in as far back as the late1970s, and as such, has a long history in this field.The models developed by the Chinese research com-munity have been widely used in ocean-atmosphere in-teraction studies, climate variability studies, seasonalpredictions, and climate change projections. The Chi-nese Academy of Sciences(CAS)has led climate mod-eling activities in China. For example, by recognizingthe central importance of climate models in climatestudies, the development of climate models quickly be-came a focus of the research activities of the Sate KeyLaboratory of Numerical Modeling for AtmosphericSciences and Geophysical Fluid Dynamics, Instituteof Atmospheric Physics, CAS(hereafter LASG/IAP)since the establishment of the laboratory in 1985. Dif-ferent versions of its AGCM, oceanic general circu-lation model(OGCM), l and surface model, sea icemodel, and the associated fully coupled models havebeen developed inside LASG/IAP(see Zhang et al., 1999 for a review).

As overarching geophysical modeling platformsfor the Chinese climate research community, the cli-mate models developed by LASG/IAP have been serv-ing as powerful tools to enhance our underst and ing ofthe fundamental mechanisms of the climate system, as well as make seasonal predictions and scenario pro-jections of future climate change. One particularlysuccessful aspect of LASG/IAP climate model devel-opment in the past 30 years is international collabo-ration. For example, a variety of CSMs developed byLASG/IAP have participated in all the past phasesof CMIP and contributed to the assessment reports ofthe IPCC. The participation in international projectssuch as CMIP has helped LASG/IAP scientists toidentify both the strengths and weaknesses of theirmodels, forming useful references for future improve-ments. In the most recent 10 years, the developmentof climate models has been granted high priority inChina, with many research centers(e.g., the NationalClimate Center, also known as the Beijing ClimateCenter, of the China Meteorological Administration)engaged in activities to develop and improve climatemodels. The expansion of the climate modeling com-munity in China has provided a solid human-resourcebasis for developing high-performance climate models.

From CMIP3 to CMIP5, the fully coupled physi-cal CSM has been developed into the ESM, which ad-ditionally considers terrestrial carbon and nutrient cy-cling processes. The development of ESMs will be oneof the frontiers of the international climate modelingcommunity in the coming decade(Wang et al., 2004;Wang et al., 2008, 2009). The issues of global climatechange and its impacts on sustainable developmenthave been and continue to be of great concern to theChinese government. In recent years, the level of fund-ing that supports the development of earth/climatesystem models has increased rapidly. Many new re-search centers that work on climate modeling havebeen set up, which will undoubtedly enhance the na-tion's ability to tackle climate change issues. Theachievements of China in developing high-performancehardware have also provided a solid platform for ad-vancing climate modeling activities in the country.However, along with the opportunities and develop-ments such as these present, the Chinese climate mod-eling community is also facing some great challenges.For example, the current performance levels of Chi-nese climate models are still generally behind those ofthe developed countries. How to improve the performances of Chinese models based on observational met-rics posed by the international community is a chal-lenge, and one that is not restricted to model tech-niques. In addition to model improvements, anotherchallenge we face is achieving successful and effcientcooperation and coordination among the many re-search centers/universities in the field of earth/climatesystem model development. Up to now, we still do nothave a national strategy for advancing climate mod-eling in China. The path for China to move forwardinto the next generation of earth/climate system mod-els and to provide the best possible climate informa-tion for the nation remains unknown.

CMIP is among the most successful internationalprojects organized and coordinated by WCRP. It hasbeen nearly 20 years since WCRP launched the firstCMIP project in 1995. At that time, there was onlyone participating Chinese climate model; but in thelatest phase of the project(CMIP5), there are fivemodels developed in China. There will be even moreclimate models from China participating in CMIP6 inthe near future. How to coordinate the developmentof climate models in China and provide the best possi-ble climate information for the nation are challengingissues for Chinese funding agencies. The aims of thecurrent paper are to 1)summarize the contributionsof Chinese models to CMIP since its inception nearly20 years ago; 2)compare current Chinese models withother CMIP5 models in the context of technical met-rics and identify the key issues that need to be ad-dressed by the Chinese climate modeling community.We hope that this review will provide a useful refer-ence for the Chinese climate modeling community and encourage collaboration in future model developments.The paper also provides an outlook for the future de-velopment of climate models in China.

The remainder of the paper is organized as fol-lows. In Section 2, we summarize the contributionsof Chinese models to the past phases of CMIP, i.e., from CMIP1 to CMIP4. The technical features ofChinese models are compared with CMIP models. InSection 3, the major improvements of climate modelsfrom CMIP1 to CMIP4 are synthesized. In Section4, the characteristics of ongoing CMIP5 models are described, providing a reference for assessing Chinesemodels. Characteristics of the five Chinese modelsthat participated in CMIP5 are summarized in Section5. Section 6 outlines the challenges for global climatemodel development in China. Section 7 clarifies theopportunities for the Chinese climate modeling com-munity. Finally, concluding remarks are provided inSection 8, along with a list of national strategies foradvancing the climate modeling enterprise in the nexttwo decades in the United States of America based ona national strategic report published by the US Na-tional Research Council(NRC).

In 1995, the Joint Scientific Committee(JSC) and CLIVAR sponsored Working Group on Coupled Mod-els(WGCM), part of theWorld Climate Research Pro-gram, launched the Coupled Model IntercomparisonProject(CMIP). Since then, phases 2-5 have subse-quently also been conducted. The basis for the re-sults of the various CMIPs is the assessment of theperformance of climate models, simulations of cur-rent climate change, and projections of future climatechange scenarios, which are used to inform correspond-ing IPCC reports released every five to seven years.For example, the results based on CMIP1 were used toinform the IPCC's Second Assessment Report(SAR), released in 1995; the results based on CMIP2 wereused to inform the IPCC's Third Assessment Report(TAR), released in 2001; the results based on CMIP3were used to inform the IPCC's Fourth AssessmentReport(AR4), released in 2007; and the results basedon the ongoing CMIP5 are being used to inform theIPCC's Fifth Assessment Report(AR5), released in2014.

CMIP provides a community-based infrastructurein support of climate model validation, intercompari-son, process diagnosis, climate change attribution, and climate change projection. The CMIP multi-modeldataset has provided the basis for thous and s of peer-reviewed papers and played prominent roles in pastIPCC assessment reports of climate variability and climate change. Note that CMIP is not organizedsolely for the purpose of the IPCC reports, and so re-ferring to the models as "IPCC models", as is the casein many peer-reviewed papers and reports, is not ac-curate. Instead, they should be referred to as "CMIPmodels." In fact, from their independent beginnings, CMIP and the IPCC's reports have grown naturally topromote one another. Before CMIP1 was launched byWCRP in 1995, the IPCC's First Assessment Report(FAR), released in 1990, used the results derived from22 AGCMs coupled with mixed-layer ocean models, and 4 fully coupled GCMs(Table 1). FAR and thecorresponding supplementary report, released in 1992, increased the attention of the climate community toclimate model research, and partly prompted the es-tablishment of CMIP by WCRP in 1995.

Table 1. The models used for IPCC FAR

Table options

From CMIP1 to CMIP3, the model developed atIAP was the only model from China that participatedin CMIP. The supplementary report of IPCC FAR re-leased in 1992 used future climate projections madeby a two-level IAP AGCM coupled with a mixed-layerocean model(Wang et al., 1993), which representedsome of the earliest model results on climate warming(Table 1).

Ten models participated in CMIP1(Meehl et al., 1997, 2000). The Chinese model was GOALS2, devel-oped by LASG/IAP(Wu et al., 1997; Zhang et al., 2000), which consisted of a 9-level AGCM with a hor-izontal resolution of R15, a 20-layer OGCM with ahorizontal resolution of 5°× 4°, an SSiB l and surface model, and a thermodynamic sea-ice model. Aprediction-correction monthly °ux anomaly couplingscheme was used during the air-sea coupling(Yu and Zhang, 1998). Only heat fluxes and wind stresseswere included in the air-sea coupling, while the surfacesalinity was restored to the climatology, since ocean-atmospheric freshwater exchange was not included.The CMIP1 model results were used in IPCC SAR, released in 1995(Table 2).

Table 2. The models used for IPCC SAR

Table options

Eighteen models participated in CMIP2(Meehl et al., 2005). The Chinese model was GOALS4, devel-oped by LASG/IAP. Relative to the previous versionGOALS2, the daily variation of solar radiation was in-troduced in GOALS3 through collaboration with Nan-jing University(Shao et al., 1998). GOALS4 improvedthe coupling processes by including ocean-atmosphericfreshwater exchange(Zhou et al., 2000, 2001), whichis a key process for simulating and underst and ing theresponses of thermohaline circulation to global warm-ing. The CMIP2 model results were used in IPCCTAR, released in 2001(Table 3).

Table 3. The models used for IPCC TAR

Table options

The IAP's CSM was the only model from a de-veloping country among the 10 models that partici-pated in CMIP1 and the 18 models that participatedin CMIP2. Therefore, this model objectively repre-sented the participation of the developing world ininternational activities relating to climate modeling and projection coordinated by CMIP. This was one ofthe key achievements of LASG/IAP in an assessmentof state key laboratories in 2000, and ultimately was a major contributing factor to LASG/IAP receiving anexcellent score^① .

There were 23 models in CMIP3, 4 of which werefrom China(2 from BCC and 2 from LASG/IAP). TheLASG/IAP model was FGOALS-g1.0(Yu et al., 2002, 2004). The atmospheric component of this model has26 levels in the vertical direction and a horizontal res-olution of 2.8° × 2.8°. The oceanic component has 30levels in the vertical direction and a horizontal resolu-tion of 1.0° × 1.0°. The l and surface model and sea-ice model are the community l and model(CLM) and community sea-ice model(CSIM), respectively, bothderived from NCAR. These four components were cou-pled together by using the NCAR Community ClimateSystem Model(CCSM)coupler(Yu et al., 2004, 2008;Zhou et al., 2007). The CMIP3 model results wereused in IPCC AR4, released in 2007(Table 4). An-other version of FGOALS(FGOALS-s1.0)developedby LASG/IAP did not participate in CMIP3 due to itsincomplete representation of the processes of various greenhouse gases and aerosols in the atmospheric ra-diation package(Zhou et al., 2005a, b, 2007). Due tosome technical problems, the CSM developed by BCCwithdrew from CMIP3.

Table 4. The models used for IPCC AR4

Table options

CMIP3 is so far the most successful and sig-nificant international coupled model intercomparisonproject. As of December 2010, over 1 Pbyte of datahave been downloaded among the 3000+ registeredusers. Over 550 journal articles, based at least in parton the dataset, have been published. The daily peakof downloaded CMIP3 data reached 1 TB from thePCMDI(Program for Climate Model Diagnosis and Intercomparison)server during 2004-2010. To date, there are still many research results based on CMIP3data to be published.

Following CMIP3, CMIP4 was organized, inwhich the models were forced separately by natu-ral variability and anthropogenic external forcing for20th-century global climate change(Meehl et al., 2007). CMIP4 is regarded as a transition program be-tween CMIP3 and CMIP5 and has been relatively lessinfluential. The separated experiments based on nat-ural variability and anthropogenic external forcing areusually considered as "tier experiments" of CMIP3.Due to the significance of this type of experiments onthe detection and attribution of climate change, theseexperiments have also been performed in CMIP5.

There are over 40 CSMs and ESMs from over 20modeling groups worldwide participating in the ongo-ing CMIP5(Taylor et al., 2012), among which thereare 5 Chinese models(Table 5). The inclusion of morethan two models from one country in CMIP has onlybefore been achieved by the United States, France, Japan, Australia, and the United Kingdom, providinga clear indication of the rapid growth in the numberof climate model developers in China.

Table 5. The models used for IPCC AR5

Table options

The structures and physical processes in climate system models exhibited significant improve-ments from CMIP1 to CMIP4. The "climate model evaluation" chapter in every IPCC report systemati-cally summarizes the improvements of climate mod-els over each 5-yr period. Based on these chapters inIPCC FAR, SAR, TAR, and AR4, the major improve-ments can be summarized as follows.

As noted in FAR and the corresponding supple-mentary report(Gates et al., 1990, 1992), the large-scale structure of the ocean and atmosphere could besimulated with some skill in the coupled general circu-lation models(CGCM)that participated in CMIP1.In those models, an adjustment was sometimes madeto the surface heat and salinity fluxes. The effectsof clouds remained a major area of uncertainty in themodeling of climate change, although the treatment ofclouds in CGCMs was becoming more complex. Thereport also pointed out that a lack of adequate obser-vational data remained a serious impediment to cli-mate model improvement.

As noted in SAR(Gates et al., 1995), sea ice and l and surface components were introduced in CGCMs, although flux adjustment was used. L and surfaceprocesses could be modeled more realistically, and the simulated large-scale distribution of temperature, salinity, and sea ice was much improved. The majorareas of uncertainty in climate models included clouds and their radiative effects, the hydrological balanceover the l and surface, and the heat flux at the oceansurface. The comprehensive diagnosis and evaluation of both component and coupled models were essentialparts of model development, but a lack of observations still limited progress, so a comprehensive globalclimate observing system was urgently needed.

It was noted in TAR(McAvaney et al., 2001)that the simulations of clouds and humidity were muchimproved in the coupled models that participated inCMIP2. Some models that did not use °ux adjust-ment maintained good stability and exhibited reason-able performance. The warming trend in 20th-centurysurface air temperature was reproduced when drivenby radiative forcing due to increasing greenhouse gases and sulphate aerosols. Simulation of the El Nio-Southern Oscillation(ENSO)was also much improved, although the simulated strength was generally under-estimated. A realization that no single model couldever be considered "best" came to the fore, along withthe importance of utilizing results from a range of cou-pled models. Coupled models were now recognized bythe community as suitable tools for providing usefulprojections of future climates.

As noted in AR4(Randall et al., 2007), mostCGCMs that participated in CMIP3 no longer usedflux adjustments. There had been ongoing improve-ments to model resolutions, computational methods, and parameterizations, and additional processes(e.g., interactive aerosols)were beginning to be included inan increasing number of models. An explicit treat-ment of the carbon cycle had been introduced in afew climate CGCMs and some ESMs of intermedi-ate complexity. The shortwave impact of changes inboundary-layer clouds, and to a lesser extent mid-levelclouds, constituted the largest contribution to inter-model differences in global cloud feedbacks.

The CMIP5 experiments include three types(Taylor et al., 2012). The first is long-term inte-gration, in which the integration time is longer than100 yr and there are two core experiments:(1)TheAtmospheric Model Intercomparison Project(AMIP)experiment, in which the observed sea surface tem-perature(SST) and sea ice are specified for the past 100 years; and (2)the climate system model experi-ment. The second type of experiments is near-termintegrations, which mainly refer to decadal-scale pre-diction experiments. These experiments consider bothexternal forcing changes(e.g., greenhouse gases, an-thropogenic aerosols, solar variability, volcanic erup-tions) and the initial state of the ocean to perform 10- and 30-yr climate predictions by using the CSM. Thethird type is high-resolution atmospheric model exper-iments. The core experiments here include the AMIPexperiments spanning the period 1979-2008 and fu-ture climate time-slice simulations for the period 2026-2035. Although the time range is relatively short, thecomputer resources required are still huge in these ex-periments due to the higher resolution.

It is important to note that the model resolu-tion requirement is different in each of the above threetypes of experiments. Models carrying out the long-term integrations often adopt a medium-to-low res-olution because huge computer resources are neces-sary. The near-term integration times are often short, so these experiments require high-resolution designs.However, in practice it has been revealed that decadalprediction skill is not due to the model resolution butthe initial scheme, so a medium-level resolution tendsto be used in this type of experiments. The third typeof experiments is designed for high-resolution climatemodels or numerical forecast models.

Of the 35 CMIP5 models, 24 are CSMs, and 13of these include an atmospheric chemistry component.The remaining 11 are ESMs, 5 of which include boththe l and and ocean carbon cycle and an atmosphericchemistry process, a further 5 include the l and and ocean carbon cycle only, and 1 includes the ocean car-bon cycle only(Flato et al., 2013). In terms of modelresolution, the CMIP5 models present the followingcharacteristics.

(1)There are 13 models participating in bothlong-term integrations and near-term integrations.The atmospheric model horizontal resolution spansfrom 2.8° to 0.8°, with an average of 1.5°. The oceanmodel horizontal resolution spans from 2.0° to 0.5°, with an average of 1.0°.

(2)There are 19 models participating in long-term integrations only. The atmospheric model horizontalresolution spans from 4.5° to 1.1°, with an averageof 2.1°. The ocean model horizontal resolution spansfrom 2.0° to 0.2°, with an average of 0.9°.

(3)There are seven models participating in near-term integrations only. The atmospheric model hori-zontal resolution spans from 2.5° to 0.5°, with an av-erage of 1.3°. The ocean model horizontal resolutionspans from 1.3° to 0.3°, with an average of 0.8°.

(4)The horizontal resolution of the high-resolution atmospheric models participating in thetime-slice integrations spans from 0.6° to 0.2°, withan average of 0.4°.

The CMIP5 experiments performed by the fiveChinese models mainly belong to the first and secondof the aforementioned types. However, as shown inFig. 1, the models with the lowest horizontal reso-lution in these experiments are from China, indicat-ing that the resolution of Chinese models used for cli-mate change research has fallen behind the interna-tional st and ard(Table 5 and Fig. 1). Increasing modelresolution is not only a technical problem, but alsoinvolves the overall ability of the model developmentteam and cooperation with high-performance comput-ing experts. How to effectively utilize the world'sleading computing resources in climate modeling is acommon problem for the climate modeling and high-performance computing fields.

Fig. 1.(a)A comparison of AGCM horizontal resolutions of CMIP5 AGCMs and CGCMs. The abscissa is for thelatitude resolution while the ordinate is for the longitude resolution(°), and one dot corresponds to one CMIP5 model.(b)Total number of models at different horizontal resolutions. The abscissa is for the model resolution while the ordinate is the number of models at different horizontal resolutions. The Chinese models are marked.

Figure options

An important development since CMIP3 is the more widespread implementation of ESMs in CMIP5.As shown in Table 6, there are 10 ESMs includingboth l and and ocean carbon components in CMIP5(Anav et al., 2013). The parameterization of marinebiology can be classified into three types in the oceancarbon component:(1)nutrient-based models, wherethe export of carbon below the surface ocean is a func-tion of the surface nutrient concentration;(2)nutrient-restoring models, in which biological carbon fluxes areset to the rates required for maintaining observed nu-trient concentration gradients against dissipation byocean mixing; and (3)models that explicitly representthe food chain, involving nutrients, phytoplankton, zooplankton, and detritus(NPZD models). Most ofthe current ESMs in CMIP5 are NPZD models. NPZDmodels include seven different parts: nutrient(phos-phate, nitrate, and iron), phytoplankton, zooplankton, dissolved organic matter(DOM), and particulate or-ganic matter.

Table 6. Summary of names, l and and ocean carbon cycle components of CMIP5 ESMs

Table options

Process-based terrestrial models used in ESMs aredynamic global vegetation models(DGVMs)(Anav et al., 2013; Shao et al., 2013), which include threecomponents:(1)a biogeographical component, whichdescribes the climatic constraints of survival and establishment;(2)a biogeochemical component, whichsimulates the growth of vegetation, including photo-synthesis and respiration; and (3)a vegetation dynamics component, which represents changes in ecologicalcharacteristics such as phenology, physiology, morphology, and species competition.

Five models developed by Chinese institutionshave participated in CMIP5. The performances ofthese five models have been systematically assessedby using same observational metrics in Zhou et al.(2014a). The technical details of the five Chinese models participating in CMIP5(Table 5)are summarizedin this section.

Two versions of the Flexible Global Ocean-Atmosphere-L and System model(FGOALS)are usedto implement the CMIP5 experiments(see Zhou et al., 2014b for a review book edited by LASG/IAP climatesystem model development team). One is FGOALS-g2(Li et al., 2013), in which the atmospheric componentis GAMIL(Grid Atmospheric Model of LASG/IAP), agrid-point model. The horizontal resolution of GAMILis approximately 2.8° × 2.8° with 26 model levels. Theocean model resolution is also approximately 2.8° × 2.8° in the horizontal direction and it has 30 levels inthe vertical direction. The l and and ice componentsare CLM3(Community L and Model version 3) and CICE4(Los Alamos sea-ice model version 4.0). Allcomponents are coupled via the CPL6 coupler. Thetuning work and basic control experiments are accomplished at LASG/IAP. Other experiments are carriedout on the computing platform in the Department ofComputer Science and Technology, Tsinghua University. The other version of FGOALS is FGOALS-s2(Bao et al., 2013), in which the atmospheric component is SAMIL(Spectral Atmospheric Model ofLASG/IAP), a spectral model. SAMIL is truncatedby R42(approximately 2.8° × 1.4°)with 26 model levels. The coupling framework is similar to FGOALS-g2, except that the ice component is CSIM5(Community Sea-Ice Model version 5). All experiments ofFGOALS-s2 are completed at IAP. The two versionshave performed all of the core experiments and partof the tier experiments for CSMs. A terrestrial carboncycle model called "Vegas" and basic ocean carbon cycle processes are involved in FGOALS-s2 to constructan ESM version of the model, FGOALS-s2-ESM. This version has performed the core experiment for ESMsin CMIP5. In addition, to reduce the huge computing cost of the millennium simulation, FGOALS-gl, aversion with coarse resolution, is used to perform thelast millennium experiment at IAP(Zhou et al., 2008, 2011; Man and Zhou, 2011, 2014; Zhang et al., 2013).

Two versions of the BCC's model are involvedin CMIP5. One is BCC_CSM1.1, for which the atmospheric component is BCC_AGCM2.1 with T42truncation(approximately 2.8ff) and 26 model levels. The ocean component is MOM4 with a horizontal resolution of 1/3°(approximately 30 km) and 40 levels. The l and component is BCC_AVIM1.0.The other version of the BCC model is BCC_CSM1.1(m). The major advance relative to BCC_CSM1.1is that the atmospheric component has been updatedto BCC_AGCM2.2, with a higher horizontal resolution of T106(approximately 1.1°). Both of the twomodel systems include simple carbon cycle processes, belonging to the ESM group. The BCC models haveperformed all of the core experiments and part of thetier experiments for CSMs and ESMs(Wu et al., 2010, 2013a, b, 2014; Xin et al., 2012).

This model is an ESM based on CCSM2. Withthe NCAR coupler, it couples the AGCM CAM3.5, the ice model CICE4.0, the Common L and ModelCoLM3.0 developed at BNU(Beijing Normal University), and the ocean model MOM4p1 developed in theGeophysical Fluid Dynamics Laboratory, NOAA. Theatmospheric component employs a spectral dynamicalframework with T42 truncation(approximately 2.8°) and 26 model levels(Wu et al., 2013; Ji et al., 2014).

This model is also an ESM based on CCSM2. Theatmospheric component is CAM3.0 with a horizontalresolution of T42(approximately 2.8°) and 26 modellevels. The ocean component is POP2 with a horizontal resolution of 1.1°(enhanced near the equatorby 0.3°-0.5°) and 40 levels. The l and and ice components are CLM3 and CICE4 respectively. A significantfeature is that a sea-wave model is coupled in the system with a horizontal resolution of 2.0°× 2.0°. The terrestrial carob cycle model is CASA and the oceancarbon cycle model is OCMIP-2. The model preformspart of the core and tier experiments for CSMs and the control experiments for ESMs(Qiao et al., 2004, 2013; Song et al., 2011, 2012).

For the five Chinese models in CMIP5, the simulated climate mean state, intra-seasonal oscillation, interannual ENSO variability, global and East Asianmonsoons, climate evolution of the 20th century, major atmospheric teleconnections, Indian Ocean warming, and many other features show reasonable performance(Bellenger et al., 2013; Sperber et al., 2013;Wu and Zhou, 2013; Dong and Zhou, 2014; Dong et al., 2014; He and Zhou, 2014; Song and Zhou, 2014;Song et al., 2014; Zhang and Zhou, 2014; Zhou et al., 2014a). However, a distinct spread in climate sensitivity to greenhouse gases forcing is apparent across themodels(Chen et al., 2014; Zhou et al., 2014a).

The CMIP5 experimental design and data size areunprecedented. Compared with previous CMIPs, anew feature in CMIP5 for IAP/LASG is the alliances and collaborations formed with other research centers.For example, IAP/LASG joined forces with FIO/SOAto construct the coupling framework of the FGOALS2system, and it also cooperated with the Departmentof Computer Science and Technology and the Centerfor Earth System Science(CESS)of Tsinghua University to optimize the code and improve the effciencyfor GAMIL and FGOALS-g2. Most of the projectionexperiments of FGOALS-g2 have been undertaken bythe high-performance computer at Tsinghua University. LASG/IAP and LAPC/IAP jointly developedthe initial version of FGOALS-s2-ESM. "Alliance and collaboration" should be advocated as the scientificculture in the ESM developer community in China.

There are many types of metrics available for evaluating the overall level of climate model development, among which comparison with CMIP5 models is an effective approach. Taking the CMIP5 models as reference criteria, the main gaps for current Chinese climate models compared with the international st and ard are discussed in this section.

Firstly, the resolutions of nearly all Chinese models lag behind the international average. The development of high-resolution models in China falls seriously behind the level of other nations. Comparedwith non-Chinese CMIP5 models, the disadvantagesof Chinese models in the area are very clear(Table 5 and Fig. 1). At present, the average horizontalresolution of international atmospheric models, whichcan be used in long-term climate simulations, reachesapproximately 1.5°. The resolutions of some modelsdeveloped by internationally advanced modeling centers are even higher than 1.0ff. For the short-term climate simulations, high-resolution atmospheric modelswith resolutions of around 0.2°-0.6° have emerged. Incontrast, the resolutions of the atmospheric components of the Chinese models participating in CMIP5are around 2.8°, which is far lower than the international average. High-resolution models can resolvefiner physical processes on smaller spatial scales, and thus obtain higher modeling skill levels. Meanwhile, high-resolution models can simulate some basic atmospheric and oceanic phenomena that cannot be simulated by low-resolution models, such as tropical cyclones and the fine structure of the Meiyu front inthe atmosphere, and mesoscale eddies in the ocean.Therefore, resolution is an important metric for evaluating the skill of atmospheric models. Even thoughthere are five Chinese models participating in CMIP5, we should recognize that the overall level of Chinesemodels lags far behind those of developed countries. Infact, the lag is becoming increasingly larger, especiallywith respect to high-resolution models and ESMs.

Secondly, China falls behind in terms of ESMdevelopment. There are 11 ESMs participating inCMIP5(Anav et al., 2013), and for the Chinese models, although BCC, BNU, FIO, and FGOALS-s2 include the carbon cycle, they all simplify both the l and and oceanic carbon cycle processes. It is importantto improve the simulations of the l and and oceaniccarbon cycle processes in Chinese models through integrating studies on l and and oceanic biogeochemicalcycle components with original physical CSMs. In thisway, the development of ESMs in China can begin toreach the same level as that achieved elsewhere in theinternational research community.

Thirdly, China does not have a suffcient workforce operating in the development of atmospheric and oceanic circulation models, both of which are key components of ESMs. There are 10 earth/climate system models currently being developed in China(Table 7). However, only IAP/CAS and BCC are engagedin the development of atmospheric models, and onlyIAP/CAS is engaged in the development of oceaniccirculation model. Those institutions that are new tothe development of coupled climate models generallyuse atmospheric and oceanic components developedabroad, which of course has the tendency to reducesample sizes when it comes to performing multi-modelensemble simulations.

Table 7. List of current earth/climate system models developed in China

Table options

The above three aspects can be summarized intoone point; that is, China needs to enhance its capacity for innovation in the development of CSMs. Modeldevelopment involves the dynamics, thermodynamics, and physical processes of the atmosphere, ocean, seaice, l and , and their interactions. It also involves the expression of the above processes on a high-performancecomputer. For many of these aspects, China falls behind the more advanced level shown internationally.Several China's models originate from overseas models via different degrees of modification. In fact, thebigger challenge is to make improvements to the dynamic core and physical processes of a model based onthe existing structure. For the development of highresolution models, in addition to the aforementionedefforts, support from high-performance supercomputerhardware and software developments is also needed.Therefore, we still need long-term unremitting effortsto shorten the gap between the current state-of-the-artin China and that achieved elsewhere internationally.

Despite the apparent lagging behind of China compared to the international community, as mentioned in the previous section, the development of ESMs nevertheless faces unprecedented favorable opportunities in the following three aspects:

Firstly, national attention and financial supportare both suffcient. China has boosted support forthe development of climate models through variousresearch projects, including the "973" and "863"projects of the Ministry of Science and Technology, CAS Strategic Priority Research Programs, SpecialScientific Research Funds of the State Oceanic Administration and China Meteorological Administration, the National Natural Science Foundation, and otherfunding sources. According to statistics, there are several "973" projects that are directly related to the development of climate models. For example, a majorscientific project entitled "Development and Evaluation of High-Resolution Climate System Models" involves the construction of a 50-km AGCM with goodstability and physical conversion, as well as a 30-50-km OGCM. Meanwhile, the objective of a project en-titled "Research of Key Processes Associated with theCarbon Cycle and Its Coupling with the Climate System" is to achieve a three-dimensional coupling of thecarbon cycle in the atmosphere, l and , and ocean, and use it to enhance the research of the interactions between climate change and the carbon cycle. Further-more, another project entitled "Development and Improvement of Ecological and Environmental SystemModels" aims to develop our own global vegetationecosystem dynamical model, global aerosols and atmospheric chemistry model, and global l and and oceanbiological process models(mainly for carbon and nitrogen cycles), and to ultimately form a complete ecological and environmental system. The objective ofan "863" key project entitled "Research of EffcientParallel Algorithms for ESMs and the Development ofa Parallel Coupler" is to design effcient parallel algorithms for an ESM, construct a parallel applicationframework, and finally build a modular parallel coupler with intellectual property in China, which appliesinnovative, high-performance algorithms, and softwareimplementation techniques to the development of aphysical CSM. Lastly, the objective of CAS Strategic Priority Research Programs under the heading"Uncertainties of Simulation and Projection by Climate Models" is to study the key physical processesassociated with the uncertainties of model simulations, design parameterization schemes, and finally developan ESM at CAS via internal and external collaboration.

Secondly, the rapid development of highperformance computers provides a solid computingplatform to develop ESMs and facilitate their participation in international competitions. Taking thesupercomputers "Tianhe One" and "Tianhe Two" assymbols of success, China has become one of the countries able to develop a peta°op supercomputer. Taking IAP as an example, LASG/IAP used thous and s ofcores to conduct quasi-global ocean circulation simulations with a 10-km resolution on "Tianhe One" in2012. It has also used approximately 1000 core computing resources to test its global atmospheric circulation model with 12.5- and 6-km resolutions. This has greatly promoted research and development of high-resolution climate models.

Thirdly, the research and development(R&D)community in the climate model field is growingstronger, with an increasing number of young scientists rapidly emerging to the fore in this area of workin China. In recent years, teaching and research institutions able to train young scientists has graduallyincreased. In addition to IAP/CAS, Nanjing University, and Beijing University, with their long histories inmodel development, other units such as Beijing Normal University, Tsinghua University, Nanjing University of information Science & Technology, and the Chinese Academy of Meteorological Sciences/China Meteorological Administration have also begun to establish departments or research centers to develop climate models and train young scientists in recent years. Therefore, our R&D community of climate modelersis growing year on year. At a national level, once climate model development teams are able to maintain acertain volume, the overall level of climate model development in China will be improved through positivecompetition and collaboration.

CSMs have evolved from physical CSMs to ESMs.The purpose of developing physical CSMs is to underst and the physics of interactions among variousspheres, whereas the purpose of developing ESMs isto underst and the roles of energetic, ecological, and metabolic processes of the earth by investigating theexchange of energy, momentum, and mass among theatmosphere, l and surface, and ocean, and to unravelthe climate responses to changes of l and surface cover, l and use, and greenhouse gas emissions through theseprocesses. Therefore, the development of ESMs shouldbe open and collaborative because of the multidisciplinary characteristics involved. Such an approach hasunderpinned the successes around developing ESMs inthe developed world.

As many as 10 earth/climate system models arecurrently being developed in China(Table 7). Howto achieve a structured development of earth/climate system models at a national level is now an urgentproblem for funding agencies. To achieve this, Chinacan turn to many strategies from elsewhere around theworld. Here, we list nine such strategies as concludingremarks to this paper, which were recommended inthe book entitled "A National Strategy for Advancing Climate Modeling", released by the US NationalAcademy of Sciences. The strategic framework wasproduced by the National Research Council(NRC)to guide the process of the US's climate modelingenterprises over the next 10-20 years. In response, the NRC appointed the Committee for the NationalStrategy for Advancing Climate Modeling. The committee recommends a national strategy for advancingclimate modeling enterprises in the next two decades, consisting of four main new components and five supporting elements as follows:

1)Evolve a common national software infrastructure that supports a diverse hierarchy of differentmodels for different purposes, and which supports avigorous research program aimed at improving theperformance of climate models on extreme-scale computing architectures; 2)Convene an annual climatemodeling forum that promotes tighter coordination and more consistent evaluation of the US regional and global models, and helps knit together model development and user communities; 3)Nurture a unifiedweather-climate modeling effort that better exploitsthe synergies between weather forecasting, data assimilation, and climate modeling; and 4)Develop training, accreditation, and continuing education for "climateinterpreters" who will act as a two-way interface between modeling advances and diverse user needs. Atthe same time, the nation should nurture and enhanceongoing efforts to 5)Sustain the availability of stateof-the-art computing systems for climate modeling;6)Continue to contribute to a strong internationalclimate observing system capable of comprehensivelycharacterizing long-term climate trends and climatevariability; 7)Develop a training and reward systemthat entices the most talented computer and climatescientists into climate model development; 8)Enhancethe national and international information technologyinfrastructure that supports climate modeling data sharing and distribution; and 9)Pursue advances inclimate science and uncertainty research.

The elements of this strategic report should bea useful reference to Chinese funding agencies in theprocess of decision-making. The key scientific issuesidentified by the report have also provided guidancefor climate change and variability studies in China.