1 3

DOI 10.1007/s00382-016-3262-9Clim Dyn

Evaluating Arctic warming mechanisms in CMIP5 models

Christian L. E. Franzke1 · Sukyoung Lee2 · Steven B. Feldstein2

Received: 19 November 2015 / Accepted: 1 July 2016 © Springer-Verlag Berlin Heidelberg 2016

more closely related to changes in the large-scale atmos-pheric circulation, whereas in Group 2, the albedo feed-back effect plays a more important role. Interestingly, while Group 1 models have a warm or weak bias in their Arctic SAT, Group 2 models show large cold biases. This stark difference in model bias leads us to hypothesize that for a given model, the dominant Arctic warming mechanism and trend may be dependent on the bias of the model mean state.

Keywords Arctic amplification · CMIP5 models · Model bias

1 Introduction

The Arctic is one of the regions most strongly impacted by anthropogenic global warming. The Arctic is warming at a faster rate than the tropics and the mid-latitudes (Manabe and Wetherald 1975; Bekryaev et al. 2010). Arctic amplifi-cation is not a new phenomenon; on geological time scales this phenomenon has occurred many times. Paleoclimate records reveal that during warm periods the equator-to-pole temperature gradient was much smaller than the cur-rent value. Budyko and Izrael (1991) proposed a universal relationship between the equator-to-pole temperature gradi-ent and the global-mean temperature which is supported by paleoclimate data (Hoffert and Covey 1992). This suggests that Arctic amplification is a fundamental process of cli-mate dynamics and the general circulation.

Arctic amplification is one of the most iconic images of global warming (Liu et al. 2012; Screen 2014). Arc-tic sea ice decline will open up northern shipping routes. Arctic warming can also lead to the thawing of perma-frost and methane clathrates, which represents a potential

Abstract Arctic warming is one of the most striking sig-nals of global warming. The Arctic is one of the fastest warming regions on Earth and constitutes, thus, a good test bed to evaluate the ability of climate models to reproduce the physics and dynamics involved in Arctic warming. Dif-ferent physical and dynamical mechanisms have been pro-posed to explain Arctic amplification. These mechanisms include the surface albedo feedback and poleward sensible and latent heat transport processes. During the winter sea-son when Arctic amplification is most pronounced, the first mechanism relies on an enhancement in upward surface heat flux, while the second mechanism does not. In these mechanisms, it has been proposed that downward infrared radiation (IR) plays a role to a varying degree. Here, we show that the current generation of CMIP5 climate models all reproduce Arctic warming and there are high pattern cor-relations—typically greater than 0.9—between the surface air temperature (SAT) trend and the downward IR trend. However, we find that there are two groups of CMIP5 mod-els: one with small pattern correlations between the Arctic SAT trend and the surface vertical heat flux trend (Group 1), and the other with large correlations (Group 2) between the same two variables. The Group 1 models exhibit higher pattern correlations between Arctic SAT and 500 hPa geo-potential height trends, than do the Group 2 models. These findings suggest that Arctic warming in Group 1 models is

* Christian L. E. Franzke christian.franzke@uni-hamburg.de

1 Meteorological Institute and Center for Earth System Research and Sustainability, University of Hamburg, Hamburg, Germany

2 Department of Meteorology, The Pennsylvania State University, State College, PA, USA

http://crossmark.crossref.org/dialog/?doi=10.1007/s00382-016-3262-9&domain=pdf

C. L. E. Franzke et al.

1 3

tipping point in the Earth system and could lead to the sudden release of methane, potentially initiating a posi-tive feedback which could accelerate the release of green-house gases and, thus, further global warming (Lenton et al. 2008). There is also a debate on whether changes in the Arctic influence the mid-latitude jet-stream and are the cause of extreme weather and climate events over the US and Europe (Francis and Vavrus 2012; Barnes et al. 2014).

Different physical mechanisms have been proposed to explain Arctic amplification. These include surface albedo feedback (Stroeve et al. 2012b provide a review), convec-tive cloud feedback (Abbot and Tziperman 2008a, b), and lapse-rate feedback (Pithan and Mauritsen 2014). The surface albedo feedback mechanism is arguably the most prominent among these theories. For this positive feedback mechanism, a change in sea ice extent alters the albedo in a manner which reinforces the change: a reduction in sea ice extent leads to a decrease of the albedo, thus, to an increase in sea surface temperature because less incoming short-wave radiation is reflected back to space. This is a local mechanism and implies that the shrinking sea ice extent is driving Arctic warming.

Using observation-based European Centre for Medium-Range Weather Forecasts Re-Analysis reanalysis data (ERA-40; Uppala et al. 2005), Graversen (2006) found that during 1979–2001 the atmospheric northward energy transport across 60 N increased significantly. Extending the idea of Cai (2005, 2006) and Lu and Cai (2009) proposed that in response to an increase in atmospheric greenhouse gas loading, the upper tropospheric temperature gradient increases, and this in turn causes baroclinic eddies to trans-port more heat to the poles. Langen and Alexeev (2007), on the other hand, proposed that since a warmer atmosphere is moister, the effective diffusivity for poleward eddy heat flux increases. While these mechanisms are distinct, both theories rely on flux-gradient relationships. In contrast, there is also a poleward heat transport theory that does not rely on a flux-gradient relationship (Lee et al. 2011a, b; Yoo et al. 2012a, b) and instead appeals to forced tapping of zonal available potential energy through convectively excited Rossby waves (Lee 2014; Baggett and Lee 2015).

Taylor et al. (2013) decomposed the feedback contribu-tions to Arctic amplification using CCSM4 and found that the surface albedo effect provides the largest contribution, followed by cloud feedback, external forcing and atmos-pheric dynamic transport. They also found that the water vapor feedback, ocean heat transport and surface turbulent flux feedbacks contribute negatively to Arctic amplifica-tion. Similar results have been reported by Yoshimori et al. (2014a, b) based on the MIROC GCM.

Regardless of the precise mechanism, the poleward transport mechanisms involve changes in the large-scale circulation [or moisture for the theory by Langen and

Alexeev (2007)] and downward infrared radiation (IR). Changes in the large-scale circulation and/or moisture transport can lead to an alteration in cloudiness and in IR. The potential impact of cloudiness and radiation on sea ice extent has been shown by Kay et al. (2008). In contrast, during the winter when Arctic amplification is most pro-nounced, the surface-albedo feedback mechanism relies on vertical heat flux from the surface (Serreze and Barry 2011; Stroeve et al. 2012b); since solar radiation is absent during the winter, it was stipulated that during summer the surface feedback process allows extra energy to be stored in the Arctic Ocean which is released during the following winter.

Because Arctic amplification is such a fundamental process of the climate system, it is important to evaluate whether the current generation of climate models accu-rately simulate it and whether the warming in these models is due to changes in the large-scale circulation, downward IR or the surface heat fluxes. The surface heat flux may also cause downward IR to increase both because of the direct warming of the atmosphere and because the surface heat flux is often accompanied by evaporation which can result in increased cloudiness and therefore greater downward IR.

Using the same data set as Graversen (2006), but with a longer time period, Lee et al. (2011b) found that in the ERA-40 data downward IR plays a central role in warm-ing the surface. This downward IR is associated with large-scale teleconnection patterns and with La-Nina-like tropical convection characterized by enhanced convection over the western Pacific warm pool and suppressed convection over the central and eastern Pacific. Furthermore, they found that the surface heat flux is actually negative over most of the ice-covered Arctic Ocean with a positive surface heat flux being limited to the Greenland, Barents and Kara Seas.

In this study, we evaluate the role of the above processes for the warming found in the CMIP5 models and compare the model results with the ERA-40 data. While there is a known deficiency in the Arctic temperature representation in the ERA-40 reanalysis (Screen and Simmonds 2010), ERA-40 covers a relatively long time period and was used by many other Arctic amplification studies. In Sect. 2 we present the data and methods used in this study. We also show that two other reanalysis products give very similar results for the surface temperature trend. This is followed in Sect. 3 by a pattern correlation analysis. In Sect. 4 we discuss the model biases, and in Sect. 5 the model trend patterns. A summary is given in Sect. 6.

2 Data and methods

In this study we analyze ERA-40 reanalysis data over the period 1958 through 2002. Here we focus on the winter season (December through February).

Evaluating Arctic warming mechanisms in CMIP5 models

1 3

For the models, we use the output from 10 historical CMIP5 simulations (Taylor et al. 2012) for the period 1958 through 2002. The names of the models and the horizon-tal resolution of their atmospheric components are given in Table 1. At the time that the CMIP5 data were downloaded, the variables that we needed for our investigation were available only for these 10 model simulations.

To evaluate the trend in the model simulations we com-pare two time periods: the winter periods of (1) 1958–1977 and (2) 1982–2001. The trend is defined as the difference between the respective mean values averaged over these two periods. We have chosen to focus on this period in order to compare our results with an earlier study (Lee et al. 2011b) and our main focus in this study is to look into the mechanisms of Arctic amplification in the CMIP5 models. Hence, the precise periods are not too important. Furthermore, the CMIP5 simulations typically end in 2005. Therefore using longer reanalysis data sets would not add much more useable data.

We compute pattern correlations within each CMIP5 model between the SAT trend and the geopotential height (ZG) and downward IR trends poleward of 60 N to cap-ture the features of the large-scale circulation and between the SAT trend and the vertical sensible heat flux (HFSS) and the vertical latent heat flux (HFLS) trends poleward of 70 N to capture the fluxes over the Arctic Ocean. This will allow us to compare the physical processes that drive Arc-tic amplification in the various CMIP5 models. We perform correlations between the ERA-40 and the CMIP5 SAT, ZG, Downward IR, horizontal temperature advection poleward of 60 N, and tropical convective precipitation between 20 S and 20 N. As indicated in the previous section, we will also calculate the bias in each CMIP5 model and compare the model trends with those in the ERA-40 data.

The Arctic is a data sparse region, so the reliability of data products in this region has to be carefully considered. Since our study is mainly based on the two time periods 1958–1977 and 1982–2001 we checked the temperature

difference over these periods for two further reanalysis products: NCEP-NCAR (Kistler et al. 2001) and JAR-55 (Kobayashi et al. 2015). As Fig. 1 indicates the three rea-nalysis products show very similar trend patterns over the Arctic region, though their magnitude differs slightly. From this we infer that ERA-40 provides a good and reliable esti-mate of past Arctic climate change for the purpose of this study.

3 Spatial field correlation analysis

We first evaluate pattern correlations between different var-iables within each CMIP5 model. An inspection of Table 2 reveals the following picture: All models show high corre-lations (>0.90) between the SAT and downward IR trends. However, there are two groups of CMIP5 models: one with small correlations (at most 0.19) between the Arctic SAT trend and the vertical sensible surface heat flux trend (Group 1; the GFDL, IPSL, MIROC, MPI and NorESM1 models), and the other (Group 2; the BCC, BNU, CMCC, HadGEM2 and MRI models) with much larger correlations between the same two variables. The vertical latent heat flux behaves in a qualitatively similar manner. Furthermore, the SAT trend and the ZG trend are much larger in Group 1 models than in Group 2 models (Table 2). This suggests that two different processes are responsible for Arctic warming in these two groups of models. Arctic warming in Group 1 models appears to be more strongly related to the large-scale circulation, whereas Group 2 is consistent with the surface albedo feedback.

The fact that surface air temperature and downward IR are highly correlated in all models (Table 2) suggests that the observed Arctic temperature trend may be mostly due to changes in downward IR. This is consistent with the recent finding that more than 50 % of the winter Arctic sea ice decline during the past few decades can be accounted for by an increase in Arctic downward IR (Park et al. 2015a).

Table 1 CMIP5 climate models

Institute Model Atmospheric horizontal resolution

Beijing Climate Center, China Meteorological Administration BCC-CSM1-1 T42

College of Global Change and Earth System Science, Beijing Normal University BNU-ESM T42

Centro Euro-Mediterraneo per Cambiamenti Climatici CMCC-CESM 3.75◦ × 3.75◦

NOAA Geophysical Fluid Dynamics Laboratory GFDL-ESM2M 2.5◦ × 2◦

Met Office Hadley Centre HadGEM2-CC 1.875◦ × 1.25◦

Institut Pierre-Simon Laplace IPSL-CM5A-MR 2.5◦ × 1.25◦

Japan Agency for Marine-Earth Science and Technology MIROC-ESM-CHEM T42

Max-Planck Institute for Meteorology MPI-ESM-MR T63

Meteorological Research Institute MRI-CGCM3 1.125◦ × 1.125◦

Norwegian Climate Centre NorESM1-M 2.5◦ × 1.875◦

C. L. E. Franzke et al.

1 3

4 CMIP5 model bias

To investigate the underlying differences between the two groups of CMIP5 models, we examine if the two groups of models have different systematic errors or biases: We dis-play this in the form of the difference between the model fields and the corresponding ERA-40 fields. In this respect we consider the ERA-40 as our best estimate of the North-ern Hemispheric state.

Figure 2 shows the SAT bias of the CMIP5 models rela-tive to the ERA-40 data over the period 1958–2001. As can be seen, the CMIP5 models have large biases in SAT. The GFDL and MIROC models are too warm in the Arc-tic region whereas many of the other models have cold

biases. These biases might be partially due to the quality of the representation of sea ice in the models (Stroeve et al. 2012a) indicating local biases, or due to remote effects such as biases in the large-scale atmospheric circulation (Perl-witz et al. 2015). Group 1 models tend to have a relatively weak warm bias in the Arctic while Group 2 models show a sizable cold bias. This different behavior of the groups is nicely seen in the group averages of Fig. 2.

The circulation bias, represented by the 250 hPa stream-function field, is displayed in Fig. 3. The most striking difference between the Group 1 and Group 2 circulation biases is that Group 2 has positive bias in the tropics and subtropics whereas Group 1 has a rather weak bias in these regions. This suggests that the tropical forcing is different

ERA-40 NCEP/NCAR JRA-55

Fig. 1 Surface air temperature trend (contour interval is 1 K) for different reanalysis products

Table 2 Pattern correlation of trend fields between CMIP5 Climate Models fields for NH and DJF

Pattern correlations are computed poleward of 60 N for SAT*Downward IR, SAT*ZG and ZG*Downward IR. Pattern correlations are computed poleward of 70N for SAT*HFSS and SAT*HFLS; these correlations have been computed over a smaller area to focus more on the Arctic ocean area which is more important for the heat flux. SAT denotes the surface air temperature, Downward IR the downward longwave radia-tion, ZG the 500 geopotential height, HFSS the sensible and HFLS the latent heat flux

Model SAT*Downward IR SAT*ZG ZG*Downward IR SAT*HFSS SAT*HFLS

ERA-40 0.91 0.21 0.11 −0.31 −0.34

Group 1

GFDL-ESM2M 0.96 0.80 0.71 0.17 0.30

IPSL-CM5A-MR 0.91 0.52 0.40 −0.11 0.27MIROC-ESM-CHEM 0.96 0.59 0.65 0.04 0.19

MPI-ESM-MR 0.82 0.50 0.41 0.19 0.25

NorESM1-M 0.94 0.67 0.50 −0.11 −0.27Group 2

BCC-CSM1-1 0.96 0.48 0.36 0.49 0.54

BNU-ESM 0.85 0.57 0.39 0.39 0.48

CMCC-CESM 0.89 0.36 0.28 0.31 0.36

HadGEM2-CC 0.89 0.31 0.26 0.28 0.39

MRI-CGCM3 0.95 0.38 0.39 0.30 0.49

Evaluating Arctic warming mechanisms in CMIP5 models

1 3

Fig. 2 Surface air temperature biases (contour interval is 3 K). ERA-40: climatological mean state of surface air tempera-ture from ERA-40 data; Other panels CMIP5 Group 1 Average model surface air temperature minus ERA-40

Group 1ERA-40

IPSLGFDL

MPIMIROC

NorESM

C. L. E. Franzke et al.

1 3

Group 2ERA-40

BNUBCC

HadGEM2CMCC

MRI

Fig. 2 continued

Evaluating Arctic warming mechanisms in CMIP5 models

1 3

ERA-40 Group 1 Group 2Streamfunction

Downward IR

Sensible Heat Flux

Latent Heat Flux

Fig. 3 CMIP5 model mean state bias for 250 hPa streamfunction (contour interval 5× 106 Wm−2), downward IR (contour interval 30Wm−2), vertical sensible heat flux (contour interval 30 Wm−2)

and vertical latent heat flux (contour interval 30Wm−2). Displayed are CMIP5 model variables minus the corresponding ERA-40 vari-able

C. L. E. Franzke et al.

1 3

between the two groups, which has implications for the northward transport of heat and moisture in the CMIP5 models.

The downward IR bias (Fig. 3) is large in all models. These biases are often negative over the oceans and posi-tive over the continents. Downward IR also shows differ-ences between the two groups. Group 1 has a positive bias over Northern Canada, Greenland, Eastern Siberia and a negative bias over the Barents sea. Group 2 has negative bias over Canada, Scandinavia, Siberia and most of the Arctic Ocean.

The vertical sensible and latent heat flux biases are also displayed in Fig. 3. All models show a positive sensible heat flux bias over the North American continent and over north and central Africa and Europe, and, a negative sensi-ble heat flux bias over much of Asia. The models show pos-itive latent heat flux biases over North America and central Africa, and negative biases over Southern and Eastern Asia. The sensible heat flux has almost no bias over the oceans while the latent heat flux has bias over the oceans; these biases are also mainly located in the subtropics.

5 CMIP5 model trends

First, we discuss the pattern correlations between the ERA-40 and the CMIP5 model trend fields in order to elucidate which models have high correlations with the reanalysis fields and, thus, faithfully reproduce the observed warming trend in the ERA-40 data. The pattern correlations between

the trends of different variables are shown in Table 3. For the SAT trend, poleward of 70 N, the pattern correlations are generally low or even negative for the HadGEM2, MIROC, MPI, MRI and NorESM1 models (Table 3). The only exception is the GFDL model which has a correla-tion value of 0.60 (Table 3). All models show rather low or even negative pattern correlations for 500 hPa geopotential height, downward IR, and 850 hPa poleward temperature advection poleward of 60 N, and tropical convective pre-cipitation between 20 S and 20 N.

Next, we examine how well the CMIP5 models repro-duce the observed trend in SAT (Fig. 4). In the ERA-40 data, the most pronounced warming occurs east of Green-land over the Greenland Sea and the Arctic Ocean. There is also wide-spread warming over Siberia, Alaska and Can-ada and strong cooling west of Greenland over Baffin Bay, Davis Strait, and the Labrador Sea. This warming pattern is consistent over the Eurasian continent with trends inferred from station data (Franzke 2012). The CMIP5 models, on the other hand, show different warming patterns. The best ones, as measured by pattern correlations (Table 3) are GFDL, CMCC and BNU. They show similar warming pat-terns to that for the ERA-40, though their warming magni-tude differs. Most models miss the cooling to the west of Greenland and over the Baffin Bay.

Figure 5 shows the surface downward IR trend. In the ERA-40 data there is a positive anomaly east of Green-land, over Alaska and Siberia. The models which capture this well are MRI, GFDL, IPSL and NorESM1 (not shown) as is also confirmed by the pattern correlations poleward of 70 N with ERA-40 (Table 3).

Figure 5 also displays the surface latent and sensi-ble heat fluxes, respectively. In the ERA-40 data most of the latent heat flux trend occurs over the Greenland, Bar-ents, Norwegian, Kara and Bering Seas. The latent heat flux trend over the continents is much smaller. The sur-face sensible heat flux trend is strong over both the oceans and the continents. In the ERA-40 data most of the Arctic and Europe have negative trends while Siberia and Green-land have positive trends. The CMIP5 models show large model-to-model variability and do not reproduce the ERA-40 heat flux trends.

To investigate if Group 1 models show evidence of a poleward wave activity flux from the tropical Pacific into the extra-tropics, we examine the trend in 250 hPa stream-function and the wave activity flux vector (Takaya and Nakaruma 2001) (Fig. 5). The horizontal components of the wave activity flux are parallel to the local Rossby wave group velocity. In ERA-40 there is a large region of diver-gent wave activity flux with a poleward component over the central North Pacific. Outside of the Arctic, the Group 1 models mostly show negative streamfunction trends while the Group 2 models exhibit more positive streamfunction

Table 3 Pattern correlation of trend fields (pattern mean subtracted) between ERA-40 and CMIP5 Climate Models for NH and DJF

Pattern correlations are computed poleward of 70N and between 20S and 20N for convective precipitation (PRC). SAT denotes the surface air temperature, PRC smoothed convective precipitation, ZG the 500 hPa geopotential height, Downward IR the downward longwave radi-ation and TVAS the poleward temperature advection at 850 hPa

Model SAT PRC ZG Downward IR TVAS

Group 1

GFDL-ESM2M 0.60 −0.00 0.36 0.22 −0.00IPSL-CM5A-MR 0.18 −0.24 0.47 0.22 0.26MIROC-ESM-

CHEM−0.36 −0.11 0.24 −0.25 −0.09

MPI-ESM-MR −0.24 0.15 −0.84 −0.37 0.02NorESM1-M −0.12 0.04 0.24 0.41 0.07Group 2

BCC-CSM1-1 0.11 0.11 −0.32 0.13 −0.07BNU-ESM 0.31 −0.08 0.31 0.02 0.16CMCC-CESM 0.39 0.05 0.02 −0.36 −0.24HadGEM2-CC −0.41 −0.11 −0.08 −0.04 −0.19MRI-CGCM3 −0.18 0.04 −0.40 0.21 −0.14

Evaluating Arctic warming mechanisms in CMIP5 models

1 3

Fig. 4 Surface air temperature differences between the periods 1958–1977 and 1982–2001 (contour interval 1 K)

Group 1ERA-40

IPSLGFDL

MPIMIROC

NorESM

C. L. E. Franzke et al.

1 3

Group 2ERA-40

BNUBCC

HadGEM2CMCC

MRI

Fig. 4 continued

Evaluating Arctic warming mechanisms in CMIP5 models

1 3

trends. Except for GFDL (and perhaps NorESM1; not shown), none of the other models show a poleward wave activity flux as in ERA-40. In the two group averages a region of divergent wave activity flux is located to the east of that in the ERA-40 and the wave activity flux has a strong equatorward component. Because of the equator-ward direction of the wave activity fluxes, this indicates that congruent spatial patterns in the large-scale circula-tion, downward IR, and SAT can occur without the tropical forcing. However, we note that the downward IR increase is greater for GFDL and NorESM than for the remaining three Group 1 models (not shown).

For the atmosphere, tropical forcing is realized through convective heating. There- fore, we next examine convec-tive precipitation. To highlight the large scale features of convective precipitation, we truncated the fields by retain-ing only zonal wavenumbers smaller than 4. Figure 5 shows this trend in the tropics. The ERA-40 data show an increase of convective precipitation over the Maritime con-tinent and the western tropical Pacific. The ERA-40 trend is consistent with other precipitation products such as the Global Precipitation Climatology Project (GPCP) and the Climate Prediction Center Merged Analysis of Precipita-tion (CMAP) (Lee et al. 2011b).

The precipitation trends in the CMIP5 models vary widely, and none of the models show the aforementioned zonal structure estimated from the observations. While ERA-40 shows largely positive trends, most CMIP5 mod-els exhibit a mixture of positive and negative trends over the tropical Pacific and Indian Oceans. Also, the magni-tude of the trend varies widely. It is well known that most coupled climate models still have many problems accu-rately representing tropical dynamics and in particular the Madden-Julian Oscillation (MJO) (Hung et al. 2013), and that the MJO has a significant influence on convective pre-cipitation. In light of the evidence provided by the grow-ing body of work that tropical convective heating plays an important role in warming the Arctic (Lee et al. 2011a, b; Yoo et al. 2012a, b; Ding et al. 2014; Krishnamurti et al. 2015; Baggett and Lee 2015; Park et al. 2015b), this finding suggests that the poor representation of tropical dynamics may inhibit an accurate simulation of Arctic warming.

6 Summary

In this study, we examined various mechanisms through which Arctic amplification occurs in CMIP5 historical sim-ulations. We find evidence that CMIP5 models are warming

the Arctic through different physical processes: (1) for some models (Group 1), the large-scale atmospheric circu-lation appears to play a relatively important role. Amongst the models in Group 1, the two models (GFDL and NorESM) that show increases in the poleward wave activ-ity flux over the tropical Pacific also have relatively large increases in the downward IR trend in the Arctic; (2) for the Group 2 models, the surface heat fluxes play a relatively more important role. We hypothesize that these differences may stem from the tuning of the model because Group 2 models have systematically a cold bias.

The relatively good performance of the GFDL model, which (1) has relatively high correlations with the ERA-40 SAT and 500 hPa geopotential height trend patterns, (2) shows a poleward wave activity flux from the tropi-cal Pacific as in ERA-40, and (3) shows a relatively large Arctic SAT trend, suggests that for a more accurate projec-tion of the SAT trend pattern in the Arctic, it may be criti-cal to correctly simulate the wave activity flux. This in turn means that tropical convection and tropical circulation pat-terns need to be more accurately simulated in the next gen-eration of climate models (CMIP6).

Our results show that most CMIP5 models have large mean biases. These biases are especially large in the trop-ics where natural modes of variability such as the MJO are still not accurately simulated (Hung et al. 2013). The theory of Hoskins and Karoly (1981) and Sardeshmukh and Hoskins (1988) show how tropical convective heat-ing can generate planetary-scale waves in the extra-tropics. The recent work on the Tropically Excited Arctic warM-ing (TEAM) mechanism (Lee et al. 2011b; Lee 2014; Bag-gett and Lee 2015; Goss et al. 2016) shows that tropically forced planetary scale waves play an important role in Arctic warming. Therefore, a bias in the tropics is likely to have large effects on the mid- and high-latitude circula-tions and to impede the accurate operation of Arctic warm-ing mechanisms such as the TEAM. This is also consist-ent with the model studies by Compo and Sardeshmukh (2009) and Shin and Sardeshmukh (2011). These two studies showed that the observed warming of the tropi-cal oceans can explain most of the observed continental warming and that the lack of accurate representation of tropical variability and tropical ocean warming in the cli-mate models causes much of the discrepancy between the model simulations and observations.

Our results show that Group 2 models have a large cold bias. This suggests that sea ice and/or snow cover and the surface-atmosphere exchange is not properly tuned. If this finding turns out to be indeed the case then our results offer the prospect of improving Arctic warming in

C. L. E. Franzke et al.

1 3

Evaluating Arctic warming mechanisms in CMIP5 models

1 3

climate models by either fine tuning the surface-atmos-phere exchange parameters or by developing more appro-priate parameterization schemes.

Acknowledgments We would like to thank two anonymous review-ers for their helpful comments. We thank the Integrated Climate Data Center at CEN for making the ERA-40 data available and the Earth System Grid Federation for making the CMIP5 data available. We thank Silke Schubert for help with the ERA-40 data. CF acknowl-edges funding by the German Research Foundation (DFG) through the cluster of excellence CliSAP (EXC177). SL acknowledges NSF Grant AGS-1455577. SBF acknowledges NSF Grant AGS-1401220.

References

Abbot DS, Tziperman E (2008a) A high-latitude convective cloud feedback and equable climates. Q J R Meteorol Soc 134:165–185

Abbot DS, Tziperman E (2008b) Sea ice, high-latitude convection, and equable climates. Geophys Res Lett 35:L03702

Baggett C, Lee S (2015) Arctic warming induced by tropically forced tapping of available potential energy and the role of the plane-tary-scale waves. J Atmos Sci 72:1562–1568

Barnes EA, Dunn-Sigouin E, Masato G, Woollings T (2014) Explor-ing recent trends in Northern Hemisphere blocking. Geophys Res Lett 41:638–644

Bekryaev RV, Polyakov IV, Alexeev VA (2010) Role of polar amplifi-cation in long-term surface air temperature variations and mod-ern Arctic warming. J Clim 23:3888–3906

Budyko MI, Izrael YA (1991) In: Budyko MI, Izrael YA (eds) Anthro-pogenic climate change, pp 277–318. Uni. Ariz. Press, Tucson

Cai M (2005) Dynamical amplification of polar warming. Geophys Res Lett 32:L22710. doi:10.1029/2005GL024481

Cai M (2006) Dynamical greenhouse-plus feedback and polar warm-ing amplification. Part I: a dry radiative-transportive climate model. Clim Dyn 26:661–675

Compo GP, Sardeshmukh PD (2009) Oceanic influences on recent continental warming. Clim Dyn 32:333–342

Ding Q, Wallace JM, Battisti DS, Steig EJ, Gallant AJE, Kim H-J, Geng L (2014) Tropical forcing of the recent rapid Arctic warm-ing in northeastern Canada and Greenland. Nature 509:209–213

Francis JA, Vavrus SJ (2012) Evidence linking Arctic amplification to extreme weather in mid-latitudes. Geophys Res Lett 39:L06801. doi:10.1029/2012GL051000

Franzke C (2012) On the statistical significance of surface air tem-peratures trends in the Eurasian Arctic region. Geophys Res Lett 39:L23705. doi:10.1029/2012GL054244

Goss M, Feldstein S, Lee S (2016) Stationary wave interference and its relation to tropical convection and Arctic warming. J Clim 29:1369–1389

Graversen RG (2006) Do changes in the midlatitude circulation have any impact on the Arctic surface air temperature trend? J Clim 19:5422–5438

Hoffert MI, Covey C (1992) Deriving global climate sensitivity from palaeoclimate reconstructions. Nature 360:573–576

Hoskins BJ, Karoly DJ (1981) The steady linear response of a spheri-cal atmosphere to thermal and orographic forcing. J Atmos Sci 38:1179–1196

Hung M-P, Lin J-L, Wang W, Kim D, Shinoda T, Weaver SJ (2013) MJO and convectively coupled equatorial waves simulated by CMIP5 climate models. J Clim 26:6185–6214

Kay JE, L’Ecuyer T, Gettelmann A, Stephens G, O’Dell C (2008) The contribution of cloud and radiation anomalies to the 2007 Arctic sea ice extent minimum. Geophys Res Lett 35:L08503. doi:10.1029/2008GL033451

Kistler R, Collins W, Saha S, White G, Woollen J, Kalnay E, Chelliah M, Ebisuzaki W, Kanamitsu M, Kousky V et al (2001) The NCEP-NCAR 50-year reanalysis: monthly means CD-ROM and documentation. Bull Am Meteorol Soc 82:247–267

Kobayashi S, Ota Y, Harada Y, Ebita A, Moriya M, Onoda H, Onogi K, Kamahori H, Kobayashi C, Endo H, Miyaoka K, Takahashi K (2015) The JRA-55 reanalysis: general specifications and basic characteristics. J Met Soc Jpn 93:5–48

Krishnamurti T, Krishnamurti R, Das S, Kumar V, Jayakumar A, Simon A (2015) A pathway connecting the monsoonal heating to the rapid arctic ice melt. J Atmos Sci 72:534

Langen PL, Alexeev VA (2007) Polar amplification as a preferred response in an idealized aquaplanet GCM. Clim Dyn 29:305–317

Lee S (2014) A theory for polar amplification from a general circula-tion perspective. Asia Pac J Atmos Sci 50:31

Lee S, Feldstein SB, Pollard D, White TS (2011a) Do planetary wave dynamics contribute to equable climates? J Clim 24:2391–2404

Lee S, Gong T, Johnson N, Feldstein SB, Pollard D (2011b) On the possible link between tropical convection and the Northern Hem-isphere Arctic surface air temperature change between 1958 and 2001. J Clim 24:4350–4367

Lenton TM, Held H, Kriegler E, Hall JW, Lucht W, Rahmstorff S, Schellnhuber HJ (2008) Tipping elements in the earth’s climate system. Proc Natl Acad Sci USA 105:1786–1793

Liu J, Curry JA, Wang H, Song M, Horton RM (2012) Impact of declining Arctic sea ice on winter snowfall. Proc Natl Acad Sci USA 109:4074–4079

Lu J, Cai M (2009) A new framework for isolating individual feed-back processes in coupled general circulation climate models. Part I: formulation. Clim Dyn 32:873–885

Manabe S, Wetherald RT (1975) The effects of doubling the CO2 con-centration on the climate of a general circulation model. J Atmos Sci 32:3–15

Park D-S, Lee S, Feldstein SB (2015a) Attribution of the recent win-ter sea-ice decline over the Atlantic sector of the Arctic Ocean. J Clim 28:4027–4033

Park H-S, Lee S, Son S-W, Feldstein SB, Kosaka Y (2015b) The impact of poleward moisture and sensible heat flux on Arctic winter sea-ice variability. J Clim 28:5030–5040

Perlwitz J, Hoerling M, Dole R (2015) Arctic tropospheric warming: causes and linkages to lower latitudes. J Clim 28:2154–2167

Pithan F, Mauritsen T (2014) Arctic amplification dominated by tem-perature feedbacks in contemporary climate models. Nat Geosci 7:181–184

Sardeshmukh PD, Hoskins BJ (1988) The generation of global rota-tional flow by steady idealized tropical divergence. J Atmos Sci 45:1228–1251

Screen JA (2014) Arctic amplification decreases temperature variance in northern mid- to high-latitudes. Nat Clim Change 4:577–582

Screen JA, Simmonds I (2010) The central role of diminishing sea ice in recent Arctic temperature amplification. Nature 464:1334–1337

Fig. 5 Differences between the periods 1958–1977 and 1982–2001 for downward IR (contour interval 5 Wm−2), vertical sensible heat flux (contour interval 1 Wm−2), vertical latent heat flux (con-tour interval 1 Wm−2), 250 hPa streamfunction (contour interval 106 Wm−2), 250 hPa Plumb vector and convective precipitation (contour interval 10−6 kgm−2 s−1). In order to smooth the fields only zonal wavenumbers smaller than 14 are retained for convective pre-cipitation

◂

http://dx.doi.org/10.1029/2005GL024481http://dx.doi.org/10.1029/2012GL051000http://dx.doi.org/10.1029/2012GL054244http://dx.doi.org/10.1029/2008GL033451http://dx.doi.org/10.1029/2008GL033451

C. L. E. Franzke et al.

1 3

Serreze MC, Barry RG (2011) Processes and impacts of Arctic ampli-fication: a research synthesis. Glob Planet Change 77:85–96

Shin S-I, Sardeshmukh PD (2011) Critical influence of the pattern of tropical ocean warming on remote climate trends. Clim Dyn 36:1577–1591

Stroeve JC, Kattsov V, Barrett A, Serreze M, Pavlova T, Holland M, Meier WN (2012a) Trends in Arctic sea ice extent from CMIP5, CMIP3 and observations. Geophys Res Lett 39:L16502

Stroeve JC, Serreze M, Holland M, Kay JE, Malanik J, Barrett AP (2012b) The Arctics rapidly shrinking sea ice cover: a research synthesis. Clim Change 110:1005–1027

Takaya K, Nakaruma H (2001) A formulation of a phase-independent wave-activity flux for stationary and migratory quasigeostrophic eddies on a zonally varying basic flow. J Atmos Sci 58:608–627

Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experimental design. Bull Am Meteorol Soc 93:485–498

Taylor PC, Cai M, Hu A, Meehl J, Washington W, Zhang GJ (2013) A decomposition of feedback contributions to polar warming amplification. J Clim 26:7023–7043

Uppala SM, Kallberg PW, Simmons AJ, Andrae U, Da Costa Bech-told V, Fiorino M, Gibson JK, Haseler J, Hernandez A, Kelly

GA, Li X, Onogi K, Saarinen S, Sokka N, Allen RP, Andersson E, Arpe K, Balmaseda MA, Beljaars ACM, Van De Berg L, Bid-lot J, Bormann N, Caires S, Chevallier F, Dethof A, Dragosavac M, Fisher M, Fuentes M, Hagemann S, Holm E, Hoskins BJ, Isaksen K, Janssen PAE, Jenne R, McNally AP, Mahfouf J-F, Morcrette J-J, Rayner NA, Saunders RW, Simon P, Sterl A, Tren-berth KE, Untch A, Vasiljevich D, Viterbo P, Woollen J (2005) The ERA-40 re-analysis. Q J R Meteorol Soc 131:2961–3012

Yoo C, Lee S, Feldstein S (2012a) Arctic response to an MJO-like tropical heating in an idealized GCM. J Atmos Sci 69:2379–2393

Yoo C, Lee S, Feldstein S (2012b) Mechanisms of extratropical sur-face air temperature change in response to the Madden–Julian oscillation. J Clim 17:5777–5790

Yoshimori M, Abe-Ouchi A, Watanabe M, Oka A, Ogura T (2014a) Robust seasonality of Arctic warming processes in two different versions of the MIROC GCM. J Clim 27:6358–6375

Yoshimori M, Watanabe M, Abe-Ouchi A, Shiogama H, Ogura T (2014b) Relative contribution of feedback processes to Arctic amplification of temperature change in MIROC GCM. Clim Dyn 42:1613–1630

Evaluating Arctic warming mechanisms in CMIP5 modelsAbstract 1 Introduction2 Data and methods3 Spatial field correlation analysis4 CMIP5 model bias5 CMIP5 model trends6 SummaryAcknowledgments References