Long-term surface air temperature trend and the possible impact on historical warming in CMIP5 models

2016-11-23LINPengFeiFENGXioLiZHIHiLIUHiLongnWANGLi

Atmospheric and Oceanic Science Letters 2016年3期

LIN Peng-Fei， FENG Xio-Li， ZHI Hi， LIU Hi-Longn WANG Li

aState Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics （LASG）， Institute of Atmospheric Physics（IAP）， Chinese Academy of Sciences， Beijing， China；bMeteorological Observatory， Huangnan Tibetan Autonomous Prefecture， China；cMinistry of Education Key Laboratory of Meteorological Disaster of Cooperation of Ministries and Provincial Governments and College of Atmospheric Sciences， Nanjing University of Information Science and Technology， Nanjing， China；dEmergency and Disaster Reduction Section， Meteorological Bureau of Qinghai Province， Xining， China

LIN Peng-Feia， FENG Xiao-Lib， ZHI Haic， LIU Hai-Longaand WANG Lid

Whether large trends exist in pre-industrial control （PICTL） runs is critically important for evaluating simulations of present climate change. This study examined the long-term trends in PICTL surface air temperature （SAT） in CMIP5 models. Small trends （＜0.06 °C/100 year） in the globally averaged SAT （GASAT） exist in most CMIP5 models. Of these， positive （negative） trends result from positive（negative） net radiation fuxes at the TOA. This conclusion was further confrmed by the signifcant positive correlations between the TOA and the SAT tendency. The PICTL GASAT trends constitute less than 10% of the historical trends， indicating that such trends are of negligible importance in estimates of historical global warming in most models. Spatially， relative to the historical trends， the PICTL trends comprise a nontrivial fraction （＞20%） in the Southern Ocean between 50°S and 65°S and in the northern Atlantic and Pacifc oceans north of 40°N， with large inter-model spread. The long-term behavior of SAT is signifcantly related to ocean circulation adjustment in the mid—high latitudes.

ARTICLE HISTORY

Revised 22 January 2016

Accepted 15 February 2016

Climate drift； CMIP5；

模式对历史气候变化的模拟往往延用工业革命前控制试验（PICTL）的物理过程及设置，同时使用PICTL的模拟结果作为初值。因此，控制试验（PICTL）中是否存在气候漂移将直接影响模式对全球变暖的模拟。本文评估了27个CMIP5模式模拟的气温长期漂移趋势及其对全球变暖模拟的影响。结果表明，大部分模式模拟的全球平均气温的长期漂移趋势都很小（＜0.06 °C/100 a），但个别模式对于全球变暖的模拟存在较大影响。在50°S～65°S纬度带内的南大洋，40°N以北的北大西洋和北太平洋区域，大部分模式模拟的气温漂移趋势对局地变暖的模拟影响较大（10%～20%）。大气层顶向上长波辐射和向上短波辐射共同影响全球气温的长期变化。另外，积分时间越长，中高纬深层海洋得到有效调整，气温的长期漂移趋势越小。本文量化了气候漂移，检验了模式的稳定性，这有助于评估模式对全球变暖的模拟能力。

1. Introduction

Coupled climate system models feature large uncertainty in their simulations and projections of global warming （e.g. Deser et al. 2012）. One particular type of uncertainty arises from the infuence of climate drift in historical simulations and projections， since climate drift will also be present in pre-industrial control （PICTL） simulations. Designed as part of CMIP5， these PICTL simulations are used for the initial values of historical simulations （Taylor， Stoufer， and Meehl 2012）. The simulated statistical climatic states and the rates of global warming diverge from their initial values due to inaccurate variability caused by obvious climate drift （e.g. Stoufer， Weaver， and Eby 2004）. In this sense， analyzing the long-term behavior of PICTL simulations in coupled models is important for estimating the rates of global warming in the past and future.

Climate drift refers to spurious trends simulated in coupled models （Sen Gupta et al. 2012）， and is a phenomenon that， generally speaking， has existed in coupled models since the frst attempts at coupling atmospheric and oceanic models in the late 1960s and early 1970s （Manabe and Bryan 1969； Bryan， Manabe， and Pacanowski 1975）. To begin with， the fux-adjustment approach was applied in coupled models to reduce climate drift （Sausen， Barthel，and Hasselmann 1988）； however， over the past 40 years，the need for fux-adjustment has gradually reduced because of the extensive improvements made to coupled models， and has not been widely used in CMIP5 （Meehl 1995； Houghton， Ding， and Griggs 2001； Covey et al. 2006；Sen Gupta et al. 2012， 2013）. Nevertheless， since the 1990s，such studies have noted that obvious climate drift exists in the polar regions， in the marginal sea ice areas near Antarctica， and at the periphery of the Arctic Ocean， based on analyzing the drift in individual models （Bryan 1998； Cai and Gordon 1999； Stoufer， Hegerl， and Tett 2000）. Covey et al. （2006） showed in their study of CMIP2 + models that drift was still signifcant， although the models had been improved over earlier versions. More recently， Sen Gupta et al. （2012） quantifed the climate drift in CMIP3 models， and found the drift was better controlled compared with previous coupled models. However， obvious trends still existed in some CMIP3 models. Do these obviously spurious trends associated with climate drift persist in the current CMIP5 models， and， if so， how does this problem afect the warming simulated by CMIP5 models？ These questions have been partially answered by studies that evaluated certain individual CMIP5 models （Lin et al. 2013）and selected variables （excluding surface air temperature（SAT）） in multiple CMIP5 models （Sen Gupta et al. 2013）. In the present study， we chose SAT， an important indicator of global warming， to examine the climate drift in CMIP5 models. Furthermore， based on the results， we discuss what causes the drift in coupled models， which has not yet been fully investigated.

Table 1.Details of the CMIP5 models used in this study are listed with the modeling center， country， horizontal resolution， integration time length in the PICTL run submitted to CMIP5. Each model is labeled with a model number.

The purpose of this work is to demonstrate how stateof-the-art coupled models control climate drift in SAT in their PICTL simulations， as well as to estimate their possible contribution in quantifying global warming. We also present the relationship between the SAT tendency and the heat fuxes at the TOA， and that between the SAT and ocean temperature （OT， layer-defned temperature）， to explore the causes and efects of long-term SAT trends. An outline of the remainder of the paper is as follows: An overview of the CMIP5 models and experiments used for analysis is presented in Section 2. The stabilities in the PICTL runs and their efects on estimates of global warming are provided in Section 3. Section 4 presents a summary and discussion.

2. Models and methods

One of the main purposes of CMIP5 was to provide results that could be used in the production IPCC AR5 （Taylor，Stoufer， and Meehl 2012）. We selected 27 available CMIP5 models and their outputs from two types of experiments: their PICTL runs， which involve external forcing and greenhouse gases prescribed at pre-industrial levels （ca. 1860）；and their historical runs forced by historical greenhouse gases， aerosols， ozone， volcanic activity， and solar activity. A maximum of two PICTL runs had been submitted to CMIP5 for each model， whereas at least one historical runs had been submitted for each model. For the PICTL simulations， we only used the run prescribing the CO2concentration in each model. For comparing the PICTL trendwith the historical trend， the frst ensemble member of the historical runs （i.e. ‘r1i1p1'） was used for calculation. Table 1 summarizes the key information regarding the CMIP5 models used in this study.

For each model， we remapped the data to a standard 1° × 1° grid and quantifed the climate drift using two approaches， least-squares linear ftting （LSLF） and ensemble empirical mode decomposition （EEMD）， in the PICTL runs of CMIP5. The EEMD approach emerged from the empirical mode decomposition （EMD） approach （Huang et al. 1998； Huang and Wu 2008）， which obtains the diference between the data and the mean of the two envelopes individually constructed from local maxima and minima. However， the decomposition determined by EMD may be sensitive to white noise. Therefore， to counter this possible issue， the EEMD approach was developed by adding the white noise when fnding the local maxima and minima （Wu and Huang 2009； Wu et al. 2011）. In the EEMD approach， the original time series can be split into several ‘intrinsic mode functions' that independently refect diferent-period oscillation components from high to low frequency. The EEMD trend line is obtained after removing these oscillation components from the original time series. The EEMD trend line can be a time-varying curl line. The approach has been used previously to obtain the SAT trend in historical runs （Lin， Feng， and Liu 2015）. Following that study， we added the noise with an amplitude of 0.1 of the standard deviation of the time series and selected the 500-time ensemble size to compute the ensemble mean. To quantify the diferences in the linear and EEMD trends lines using these two approaches， the EEMD trend lines were linearized using the LSLF approach. Only the frst ensemble member of a historical run （i.e. ‘r1i1p1'） was used to compute the linear trend， and was compared with the PICTL trend for each given CMIP5 model. Based on the assumption that climate drift exists in the unforced（PICTL） and forced simulations， the raw historical trends were corrected by subtracting the PICTL trend value from the historical trend value directly for each CMIP5 model. The observed SATs from the HadCRUT4 data-set （Morice et al. 2012） were used for comparison in this study. The SATs were available as a set of anomalies relative to the period 1961—90.

To understand the possible causes of the long-term SAT trends， we also incorporated the net radiation fuxes at the TOA （including the incoming shortwave radiation （ISR），refected shortwave radiation （RSW）， and OLR）， and the OT. We only used the heat fuxes at the TOA from 25 available CMIP5 models （except CMCC-CM and EC-EARTH）. To derive the net heat fuxes at the TOA， the ISR （positive downward）was subtracted from the OLR （positive upward） and RSW（positive upward）.

3. Results

3.1. Long-term trends in SAT in the PICTL runs

The trend lines （as a proxy for the drift） in the PICTL run of CMIP5 obtained from LSLF and EEMD are presented in Figures 1（a） and （b）， respectively. In the PICTL runs， most models have relatively small linear trends （Figure 1（a）） in global averaged SAT （GASAT）， except for models 6， 11，12， 17， and 22. Compared with the linear trend lines， the EEMD trend lines of GASAT undergo linear evolutions for most of models. This indicates the trends in most models are almost linear. For some models （e.g. models 5 and 17），the trend lines have an infection point， which implies these models possess lower frequency variability that the shorter time series （＜500 year） provided for CMIP5 cannot distinguish.

In the PICTL runs， most models have relatively small trends （＜0.06 °C/100 year） in global averaged SAT （GASAT），except for models 6， 11， 12， 17， and 22 （Figure 1（c））， whose trend magnitudes exceed one RMSD （root-mean-square deviation） relative to the ensemble mean trend in all models （Figure 1（c））. The trend in GASAT for models 12 and 22 is particularly large and the value in both cases is approximately 0.11 °C/100 year （Figure 1（c））. To quantify the EEMD trend magnitudes in each model， linear ftting of each EEMD trend line was carried out. The trend magnitudes are presented in Figure 1（d）. The linear ftting EEMD trends are quite similar to those derived by LSLF. The trend diference according to the EEMD and LSLF approaches is no more than 0.01 °C/100 year in 16 models. This indicates most of the models feature linear trend behavior as the integration length becomes sufciently long （＞500 year），and that LSLF is a good approach to detect the trend for these long time series.

Small trends in GASAT derived by these two approaches indicate that most models （24/27） have reached relatively stable integration states. Meanwhile， the magnitudes of the GASAT trends in CMIP5 are smaller than those （with a range from -0.32 to 0.14 °C/100 year） in CMIP3 （Sen Gupta et al. 2012）.

Figure 1.The （a） LSLF and （b） EEMD trend lines based on GASAT （°C） annual mean anomalies （relative to mean values during the whole period） in the PICTL runs from CMIP5 models. （c） The linear trend （°C/100 year） of GASAT along with its standard error （SE， error bars）during the whole period of the PICTL run submitted to CMIP5. （d） The linear ftting EEMD trends of GASAT during the whole period of the PICTL run. （e） Raw historical （1901—2005） trends of GASAT along with their SEs （error bars） and drift-corrected historical trends. （f）Ratios （%） of PICTL linear and linear ftting EEMD trends to the historical trends for GASAT during the whole period of the PICTL runs. The numbers on the right side of the lines in （a， b） are the model numbers as listed in Table 1. The x-axes in （c—f） are labeled with the 27 model numbers， their ensemble mean value （28th column）， and the observed trend from HadCRUT4 （Morice et al. 2012）（29th column）. The black dashed lines indicate the positions where the ensemble mean and ensemble mean ± 1 RMSD are located in （c， d）， the positions where the observed trend and ensemble mean ± 1 RMSD are located in （e）， and where the PICTL trends are 10% and 30% of the historical trends in （f）.

Spatially， the ensemble mean trends and their spread errors （RMSD relative to ensemble mean） in the PICTL runs are given in Figure 2（a）. A number of common long-term trend features from the models are clearly presented. In the tropics， the long-term trends are less than 0.05 °C/100 year，especially around the western Pacifc， eastern Indian Ocean and western Atlantic Ocean. Meanwhile， the differences between these models are also relatively small in these regions due to the small RMSD （contour in Figure 2（a））. In the middle to high latitudes， the trends are relatively large （＞0.1 °C/100 year） with relatively large RMSD（＞0.05 °C/100 year）. The largest PICTL trends are located in the Southern Ocean （SO） between 55°S and 75°S， with the largest inter-model spread. In addition to this region，relatively large PICTL trends are also located south of 40°S（including the Antarctic continent） and north of 40°N（including the Bering Strait， the North Atlantic around Greenland and the Labrador Sea， and the Arctic Ocean）. These regions with large trends have large model diferences， which imply that large spreads of PICTL trends exist in these models. Trends with large magnitude appear in the middle to high latitudes. In these regions， there is often a deep mixing layer or deep convective zone due to strong wind， especially in local winter （fgures not shown）. The corresponding relationship between the large trend and deep mixing layer implies the mixing or deep convection plays an important role in modulating the climate drift because of their efect on the ocean meridional overturning circulation and ocean adjustment. The ocean circulation will slowly afect the ocean density and ocean surface temperature， and lead to the large trend. Therefore， mixing or deep convection may afect the long-term integration behavior in a coupled model or stand-alone ocean model. Meanwhile， the RMSD of the linear trend is large， whichmay on the one hand be due to the diferent positions of convective zones in diferent coupled models （Geofroy，Saint-Martin， and Ribes 2012）； while on the other hand， the diferent schemes used to simulate eddy efects （Collins et al. 2007； Brierley， Collins， and Thorpe 2010）. Because the oceanic horizontal resolution of most climate models is around 100 km， these climate models cannot simulate the efect of oceanic mesoscale eddies， and use diferent eddy parameterization schemes.

Figure 2.（a） Ensemble mean （shaded） of the absolute values of the SAT linear trends and RMSD （contours） of the absolute values of the SAT linear trends during the whole period in the 27 PICTL runs. （b） Ensemble mean （shaded） of the historical trends and RMSD （contours）of the historical trends during the period 1901—2005 in the 27 historical runs. （c） The ratios （%） of the ensemble mean of the absolute values of the SAT linear trends during the whole period in the PICTL runs to the ensemble mean of the historical （1901—2005） trends from 27 models. （d） Ensemble mean of the ratios （%） of the PICTL trends during the whole period to the historical trends （1901—2005） from each model. The contour intervals are 0.05 °C/100 year and 0.25 °C/100 year in （a） and （b）， respectively.

3.2. Potential importance for historical warming

To determine whether the long-term trends have potential efects on estimates of historical global warming， we compared the long-term PICTL trends with the historical linear trends. The observed linear trend of GASAT is 0.72 °C/100 year during the period 1901—2005， as estimated from HadCRUT4. Compared with the observed trend， the simulated historical trends show considerable spread in the available CMIP5 models （Figure 1（e））. Thirteen （fourteen） models overestimate （underestimate）the warming.

Because the initial values of historical runs are chosen from the PICTL runs， the existence of climate drift in the PICTL runs will incur a warming/cooling trend in the historical simulations， due to the long-term oceanic memory. The corrected historical trends are not particularly diferent from the raw historical trends， except in three models（12， 17， and 22）（Figure 1（e））. To further quantify the infuence of the PICTL trends on the historical warming， the ratios （or percentages） of the PICTL trends to the historical trends during the period 1901—2005 for each model are presented in Figure 1（f）. As expected， the PICTL trends for most models （23/27） account for less than 10% of the simulated historical trends， which implies that small PICTL trends do not greatly afect estimates of historical warming trends. The PICTL trends do， however， contribute 10—30% to the historical trends during 1901—2005 in four models （12， 17， 18， and 22）. Using these models （with large ratio ＞10%） to estimate the simulated historical trends，corrections must be performed due to climate drift in the PICTL runs.

As shown in Figure 2（b）， the simulated historical warming （1901—2005） has a distinct spatial pattern. The warming in the NH is larger than the warming in the SH， as was also observed （fgure not shown）. The historical warming over land is larger than that over the ocean between 40°S and 40°N. Larger historical trends （＞1.4 °C/100 year） are located in the regions north of 60°N （including the land and ocean regions of the Arctic）. The trends in the high latitudes of the SO are less than half of those in the Arctic Ocean. The smallest historical trend occurs over the North Atlantic，SO， and North Pacifc， where the ocean has strong internal variability that can eliminate the SAT warming response to external forcing. Larger inter-model diferences in the historical trends exist across models in the southern Arctic Ocean， around Greenland and the Labrador Sea， around the Bering Strait， the Barents Sea， and the SO.

The distribution of the ratios （percentages） of the ensemble mean PICTL trends to the ensemble mean historical trends is further given in Figure 2（c）. The large percentages （＞10%） are located in the SO （the latitudinal bands of 50°S and 65°S）， especially in the Pacifc sector. In the northern Atlantic Ocean and northern Pacifc Ocean（north of 40°N）， the percentages are also relatively large（10%—20%）. To retain the unique features of the percentages of PICTL trends to historical trends in each model， the ensemble mean of the percentages in all models was computed （Figure 2（d））. Compared with the pattern in Figure 2（c）， the pattern of large percentages （with larger values）is similar in Figure 2（d）. This suggests that there are common features among these models. The large values are mainly associated with a small trend in the historical run and non-negligible PICTL trend as a result of strong ocean internal variability. The large drift in these regions may be related to the ocean not fully adjusting in the PICTL run，since the deep ocean （e.g. Antarctic bottom water） needs thousands of years to become stable. This pattern in Figure 2（d） also indicates that the long-term PICTL trends cannot be neglected locally and are potentially important in estimating the historical warming in the regions with large percentages. However， a number of special areas （large values located in very limited and small regions） exist in the distribution （Figure 2（d）） due to the very large percentages in individual models. These areas imply that PICTL longterm trends should be considered in analyses of detection/ attribution studies， and in estimating the uncertainties at regional scales using individual models.

3.3. Possible causes of long-term trends

The importance of the TOA to surface temperature has long been recognized （Cess et al. 1985； Ma et al. 1994； Forster and Ramaswamy 2007）. The relationships between globally averaged net TOA radiation and SAT trends in the 25 CMIP5 PICTL runs are shown in Figure 3（a）. Some of the models exhibit a certain imbalance in their net TOA radiation. The values range from -0.79 to 3.63 W m-2. Although the correlations between the net TOA and SAT trends are small （only -0.1） and insignifcant， the positive （negative）net TOA radiation fuxes can result in positive （negative）SAT trends in most models （16/25）， as noted by Miller and Tegen （1999）.

To diagnose the temporal variation of SAT and the TOA over long （＞10 year） timescales in the PICTL runs， the correlations between SAT tendencies and net TOA radiation fuxes globally on such scales （＞10 year） were calculated（Figure 3（b））. Their correlations are signifcantly positive （at the 0.05 level） for all models. The signifcant positive correlations indicate that the SAT trends can be attributed to the globally averaged net TOA radiation fux imbalances in the PICTL runs. Because the downward shortwave radiation at the TOA is given and unchanged in each annual cycle，only OLR and RSW have a temporal change， excluding the annual cycle. The correlations between the SAT tendency and OLR and RSW at the TOA on long timescales （＞10 year）were also calculated （Figure 3（b））. The correlations are negative in most models （25/27）. Relatively insignifcant correlations （including negative and small positive correlations） exist in nine models and twelve models for the correlations between the SAT tendency and OLR and RSW，respectively. The signifcantly negative correlations indicate that a decrease in OLR and RSW at the TOA can induce an increase in SAT. The magnitudes of the correlations for OLR or RSW are smaller than the correlations of net TOA radiation， implying that there is a combined efect of OLR and RSW on SAT. Interestingly， the relationships between the net TOA fuxes and SAT tendency in some models are controlled by the efect of OLR （e.g. 4， 21， 22， 24， 25，27）， while others are controlled by the efect of RSW （e.g. 1， 11， 19， 20）. This diference implies that diferent processes play diferent roles because OLR is mainly afected by the presence of clouds， while RSW is also mainly afected by cloud and surface albedo （e.g. Aumann， Ruzmaikin， and Behrangi 2012）. Meanwhile， there are connections and feedback interactions between the clouds， surface albedo，and SAT. As can be seen in Figure 2（a）， the largest SAT drift in the polar regions is related to those in the TOA heat fux via the albedo—temperature—sea-ice feedback. As sea ice increases in the polar regions， surface albedo increases. The RSW increases （decreases）， reducing （enhancing）the absorption of solar radiation. The SAT then decreases（increases）， leading to a further increase （decrease） in sea ice. Besides， SAT drift in the polar regions may be related to the slow change in ocean circulation， whose deep branch takes thousands of years to become stable. Thus， the SATtrends may be closely related to diferent feedback interactions in the climate system.

Figure 3.（a） Scatter plot of the PICTL LSLF and linear ftting EEMD trends of GASAT （°C/100 year） versus the net TOA radiation （W m-2）during the whole period in the 25 PICTL runs， except for modes 6 and 9 due to a lack of TOA data. （b） Correlations between the GASAT tendency and net TOA radiation， OLR at the TOA， and RSW at the TOA. （c） Correlations between the GASAT and averaged OT in the upper 700 m. （d） Correlations between the GASAT and OT in the lower 700 m. Before calculating the correlations， the net radiation， OLR，and RSW at the TOA， OT， and GASAT were determined by the 10-year running mean in the PICTL runs. The magnitudes of signifcant correlation at the 0.05 level are indicated by the black dashed lines in （b）. The white regions in （c， d） indicate the correlations are not signifcant at the 0.05 level.

The correlation coefcients between GASAT and OT in the upper 700 m （the depth of the main thermocline） are signifcant and positive in the eastern tropical Pacifc， in the western Indian Ocean and the SO between 40°S and 60°S，and the regions located between 60°N and 80°N （Figure 3（c））. In contrast to the upper 700 m， the correlations show signifcant positive correlations in most regions except the Arctic Ocean and the SO south of 60°S， close to the Antarctic continent. In the SO between 40°S and 60°S， in the central and eastern Pacifc， and in the eastern Indian Ocean，the positive correlations are much more signifcant （Figure 3（d））. Thus， the change in GASAT is not only closely connected to upper OT， but also to deeper OT. The signifcant correlations with OT indicate that the long-term behavior of SAT is also related to the variation in the thermohaline circulation in the North Atlantic and the areas of deep to intermediate water formation processes in the SO. The thermohaline circulation can redistribute the meridional heat transport slowly. If the thermohaline circulation still has some long-term trends， it will afect the SAT trend locally via ocean dynamics， especially in the high-latitude ocean. Slow adjustments of ocean circulation in the middle—high latitudes may be an important factor afecting the longterm behavior of the climate system. A changed pattern or magnitude of oceanic convection is potentially important for these slow adjustments （e.g. Cai and Gordon 1999）.

4. Conclusion

In this study， we estimated the long-term trends of the GASATs and the distributions of SAT trends in the PICTL runs from the results of 27 CMIP5 models and compared the trends with the historical trends. In most CMIP5 models， the PICTL trends contribute to a small （＜10%） part of the forced historical trends， indicating that estimated historical global warming is unlikely afected， to a signifcant extent， by the trends in the PICTL runs in most models. However， the spatial SAT trends constitute a nontrivial fraction of the estimation of historical warming in the SO between 50°S and 65°S， in the northern Atlantic Ocean，and in the northern Pacifc Ocean north of 40°N. When an individual model is used for detection/attribution at certain regional scales， or even global scales， the long-term PICTL trends should be evaluated frst. Meanwhile， spatially， there are large inter-model diferences in estimating the percentage of the contribution of the PICTL trends to the historical trends.

Positive （negative） net incoming TOA radiation fuxes can result in positive （negative） trends of GASAT in most CMIP5 models. At long （＞10 year） timescales， an increase in the net TOA radiation can result in an increase in SAT. The changed OLR and RSW have a combined efect on the changed SAT in most models. The long-term SAT change is also related to the adjustment of ocean circulation in the middle—high ocean， especially in the SO.

Before estimating historical trends， it is important to frst estimate the trends in the PICTL runs. The directly related change may be manifested in the radiation processes at the TOA and slow ocean processes （needing the relative long integration time）.

Acknowledgements

We acknowledge the modeling development groups and the PCMDI for the availability of their datasets. We are also grateful to the two anonymous reviewers for their constructive comments， which helped to improve this paper.

Disclosure statement

No potential confict of interest was reported by the authors.

Funding

This project was supported by the National Key Basic Research Program of China ［grant numbers 2010CB950502 and 2013CB956204］； the ‘Strategic Priority Research Program—Climate Change: Carbon Budget and Related Issues' of the Chinese Academy of Sciences ［grant number XDA05110302］；the National Natural Science Foundation of China ［grant numbers 41376019 and 41376039］.

References

Aumann， H. H.， A. Ruzmaikin， and A. Behrangi. 2012. “On the Surface Temperature Sensitivity of the Refected Shortwave，Outgoing Longwave， and Net Incident Radiation.” Journal of Climate 25: 6585—6593.

Brierley， C. M.， M. Collins， and A. J. Thorpe. 2010. “The Impact of Perturbations to Ocean-model Parameters on Climate and Climate Change in a Coupled Model.” Climate Dynamics 34（2—3）: 325—343.

Bryan， F. O. 1998. “Climate Drift in a Multicentury Integration of the NCAR Climate System Model.” Journal of Climate 11: 1455—1471.

Bryan， K.， S. Manabe， and R. C. Pacanowski. 1975. “A Global Ocean-atmosphere Climate Model. Part II. the Oceanic Circulation.” Journal of Physical Oceanography 5: 30—46.

Cai， W.， and H. B. Gordon. 1999. “Southern High-latitude Ocean Climate Drift in a Coupled Model.” Journal of Climate 12: 132—146. Cess， R. D.， G. L. Potter， S. J. Ghan， and W. L. Gates. 1985. “The Climatic Efects of Large Injections of Atmospheric Smoke and Dust: A Study of Climate Feedback Mechanisms with One- and Three-Dimensional Climate Models.” Journal of Geophysical Research 90: 12937—12950.

Collins， M.， C. M. Brierley， M. MacVean， B. B. B. Booth， and G. R. Harris. 2007. “The Sensitivity of the Rate of Transient Climate Change to Ocean Physics Perturbations.” Journal of Climate 20 （10）: 2315—2320.

Covey， C.， P. J. Gleckler， T. J. Phillips， and D. C. Bader. 2006. “Secular Trends and Climate Drift in Coupled Ocean—Atmosphere General Circulation Models.” Journal of Geophysical Research 111: 1—19.

Deser， C.， A. Phillips， V. Bourdette， and H. Teng. 2012. “Uncertainty in Climate Change Projections: The Role of Internal Variability.”Climate Dynamics 38: 527—546.

Forster， P.， and V. Ramaswamy. 2007. “Changes in Atmospheric Constituents and in Radiative Forcing.” Climate Change 20: 129—234.

Geofroy， O.， D. Saint-Martin， and A. Ribes. 2012. “Quantifying the Sources of Spread in Climate Change Experiments.”Geophysical Research Letters 39: （L24703） 1—5. doi: 10.1029/2012GL054172.

Houghton， J. T.， Y. Ding， D. J. Griggs， M. Noguer， P. J. Linden，X. Dat， K. Maskell， and C. A. Johnson. 2001. The Scientifc Basis；Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press. http://www.grida.no/publications/other/ ipcc_tar/.

Huang， N. E.， and Z. Wu. 2008. A Review on Hilbert-Huang Transform: Method and Its Applications to Geophysical Studies， Reviews of Geophysics 46 （2）， RG2006. doi:10.1029/2007RG000228.

Huang， N. E.， Z. Shen， S. R. Long， M. C. Wu， H. H. Shih， Q. Zheng，N. C. Yen， C. C. Tung， and H. H. Liu. 1998. “The Empirical Mode Decomposition and the Hilbert Spectrum for Nonlinear and Non-stationary Time Series Analysis.” Proceedings of the Royal Society a: Mathematical， Physical and Engineering Sciences 454（1971）: 903—995.

Lin， P. F.， H. L. Liu， Y. Q. Yu， and T. J. Zhou. 2013. “Long-term Behaviors of Two Versions of FGOALS2 in Preindustrial Control Simulations with Implications for 20th Century Simulations.”Advances in Atmospheric Sciences 30: 577—592.

Lin， P. F.， X. L. Feng， and J. J. Liu. 2015. “Historical Trends in Surface Air Temperature Estimated by Ensemble Empirical Mode Decomposition and Least Squares Linear Fitting.”Atmospheric and Oceanic Science Letters 1: 10—16.

Ma， C. C.， C. R. Mechoso， A. Arakawa， and J. D. Farrara. 1994.“Sensitivity of a Coupled Ocean—Atmosphere Model to Physical Parameterizations.” Journal of Climate 7: 1883—1896. Manabe， S.， and K. Bryan. 1969. “Climate Calculations with a Combined Ocean-atmosphere Model.” Journal of the Atmospheric Sciences 26: 786—789.

Meehl， G. A. 1995. “Global Coupled General Circulation Models.” Bulletin of the American Meteorological Society 76: 951—957.

Miller， R. L.， and I. Tegen. 1999. “Radiative Forcing of a Tropical Direct Circulation by Soil Dust Aerosols.” Journal of the Atmospheric Sciences 56: 2403—2433.

Morice， C. P.， J. J. Kennedy， N. A. Rayner， and P. D. Jones. 2012. “Quantifying Uncertainties in Global and Regional Temperature Change Using an Ensemble of Observational Estimates: The HadCRUT4 Data Set.” Journal of Geophysical Research Atmospheres 117: D08101.

Sausen， R.， K. Barthel， and K. Hasselmann. 1988. “Coupled Ocean-atmosphere Models with Flux Correction.” Climate Dynamics 2: 145—163.

Sen Gupta， A. S.， L. C. Muir， J. N. Brown， S. J. Phipps， P. J. Durack，D. Monselesan， and S. E. Wijfels. 2012. “Climate Drift in the CMIP3 Models.” Journal of Climate 25 （13） : 4621—4640. doi:10.1175/jcli-d-11-00312.1.

Sen Gupta， A. S.， N. C. Jourdain， J. N. Brown， and D. Monselesan. 2013. “Climate Drift in the CMIP5 Models.” Journal of Climate 26 （21）: 8597—8615.

Stoufer， R. J.， G. Hegerl， and S. Tett. 2000. “A Comparison of Surface Air Temperature Variability in Three 1000-Yr Coupled Ocean—Atmosphere Model Integrations.” Journal of Climate 13: 513—537.

Stoufer， R. J.， A. J. Weaver， and M. Eby. 2004. “A Method for Obtaining Pre-Twentieth Century Initial Conditions for Use in Climate Change Studies.” Climate Dynamics 23: 327—339. doi:10.1007/s00382-004-0446-5.

Taylor， K. E.， R. J. Stoufer， and G. A. Meehl. 2012. “An Overview of CMIP5 and the Experiment Design.” Bulletin of the American Meteorological Society 93: 485—498.

Wu， Z.， and N. E. Huang. 2009. “Ensemble Empirical Mode Decomposition: A Noise-assisted Data Analysis Method.”Advances in Adaptive Data Analysis 1: 1—41.

Wu， Z.， N. E. Huang， J. M. Wallace， B. V. Smoliak， and X. Chen. 2011. “On the Time-Varying Trend in Global-Mean Surface Temperature.” Climate Dynamics 37: 759—773.

historical warming； trend

气候漂移；气温； CMIP5；历史变暖

29 September 2015

CONTACT LIN Peng-Fei linpf@mail.iap.ac.cn

This is an Open Access article distributed under the terms of the Creative Commons Attribution License （http://creativecommons.org/licenses/by/4.0/）， which permits unrestricted use， distribution， and reproduction in any medium， provided the original work is properly cited.