Arctic sea ice reduction and European cold winters in CMIP5 climate change experiments



[1] European winter climate and its possible relationship with the Arctic sea ice reduction in the recent past and future as simulated by the models of the Climate Model Intercomparison Project phase 5 (CMIP5) is investigated, with focus on the cold winters. While Europe will warm overall in the future, we find that episodes of cold months will continue to occur and there remains substantial probability for the occurrence of cold winters in Europe linked with sea ice reduction in the Barents and Kara Sea sector. A pattern of cold-European warm-Arctic anomaly is typical for the cold events in the future, which is associated with the negative phase of the Arctic Oscillation. These patterns, however, differ from the corresponding patterns in the historical period, and underline the connection between European cold winter events and Arctic sea ice reduction.

1. Introduction

[2] A warming trend of about 1 K in the averaged surface air temperature (SAT) in Northern Hemisphere has been observed during the 20th century. This temperature rise is particularly noticeable in the Arctic during winter where the averaged temperature change is approximately double of the Northern Hemispheric mean changes and accompanied by a rapid reduction in sea ice cover in recent decades [Intergovernmental Panel on Climate Change, 2007; Polyakov et al., 2002; Serreze and Francis, 2006; Screen and Simmonds, 2010; Comiso et al., 2008; Arctic Monitoring and Assessment Programme, 2011; Polyakov et al., 2012].

[3] In spite of this, a number of abnormally cold winters in Europe have occurred in the recent years and brought in much public interest in particular during the winters 2005/2006, 2009/2010 and 2010/2011 [Scaife and Knight, 2008; Cohen et al., 2010]. Several modelling studies [e.g., Honda et al., 2009; Petoukhov and Semenov, 2010] revealed that the anomalous decrease of sea ice concentration (SIC) in the Barents and Kara (B-K) Seas sector could result in cold winter in Europe, implying that a series of cold European winter events by no means conflict with an overall global and even regional warming trend [Serreze and Barry, 2011; Levermann et al., 2012]. Numerous model sensitivity studies have been conducted to isolate the effects of Arctic sea ice loss on the atmosphere. Recently, Porter et al. [2012]identified both local and remote responses to reduced Arctic sea ice in a study using re-analysis data 1994 to 2008 for example, whileBudikova [2009]provides a comprehensive review of previous studies. Other local large-scale interactions that link the Arctic with more southern latitudes have been identified. A candidate that apparently influences the frequency and strength of cold winters in Europe was recently been proposed;Haigh et al. [2010] demonstrated an apparent link between the long term varying radiative forcing in the ultraviolet spectrum that has been observed using satellite instruments in the last decade. The potential mechanism isolated by Petoukhov and Semenov [2010] can therefore by no means be seen as the only driver of European winter time conditions.

[4] Although the precise mechanism controlling the linkage between B-K SIC and European winter temperatures is not well understood,Petoukhov and Semenov [2010]in their analysis suggest a conceptual model that explains the nonlinear local atmospheric response in the B-K Seas region by counter play between convection over the surface heat source and baroclinic effect due to modified temperature gradients in the vicinity of the heating area.

[5] While the understanding of the nature of this possible link remains a challenge, it is nevertheless an interesting question to pose, whether or not cold winter anomalies are likely to occur in the future, when climate projections suggest further warming and reduction in Arctic sea ice. One way to analyse this is by investigating climate change scenarios from coupled GCMs. Here we take advantage of new state-of-the-art models that have been made publically available due to the Coupled Model Intercomparison Project phase 5 (CMIP5). We examine the European winter climate in connection with Arctic (focused on the B-K region) sea ice reduction based on 13 CMIP5 simulations of the coming century, with a focus on the cold winters.

2. Data and Method

[6] This study uses data from 13 CMIP5 model outputs obtained from the CMIP5 data archive ( as listed in Table 1. (These were the models we were able to access at the web site up and until March 1, 2012.) The monthly mean data from the historical and future period using the radiative concentration pathways (RCPs) 4.5 and 8.5 experiments from these CMIP5 models [Taylor et al., 2012] are analyzed.

Table 1. Probability (in %) for CWMs in Europe for Various Time Periods in Historical and Future Experiments of the RCP4.5 and RCP8.5 Scenarios as Simulated by the CMIP5 Modelsa
ModelClimatology 1971–2000Historical 1956–2005RCP4.5RCP8.5
  • a

    Numbers in parentheses are the ratio of respective probability to that of the corresponding model climatology (in %).

BCC-CSM1.14348 (111)35 (80)17 (39)23 (52)8 (19)
CanESM24745 (97)17 (36)9 (20)22 (47)4 (9)
CCSM43944 (113)32 (82)21 (55)32 (82)8 (21)
CNRM-CM55245 (88)31 (59)7 (13)16 (30)3 (5)
EC-EARTH4847 (99)25 (52)9 (20)20 (43)2 (4)
HadGEM2-ES4652 (114)16 (35)10 (21)14 (32)4 (9)
INM-CM44745 (97)34 (73)21 (46)24 (52)15 (33)
IPSL-CM5A-LR4446 (103)23 (51)13 (30)18 (41)1 (3)
IPSL-CM5A-MR4645 (100)16 (35)10 (22)23 (52)4 (9)
MIROC54443 (98)19 (43)4 (9)15 (34)3 (8)
MPI-ESM-LR4757 (123)42 (89)30 (64)36 (78)9 (20)
MRI-CGCM34647 (102)33 (73)21 (47)35 (77)7 (16)
NorESM1-M4043 (108)23 (57)12 (30)27 (66)8 (20)
All model mean4547 (103)27 (59)14 (31)23 (52)6 (13)

[7] To assess the winter climate in Europe in the future, we use for each model as its present-day climatology a 30 year average of the variable in concern from the simulation for the baseline period of 1971 to 2000 in the historical experiment. It follows that the anomaly of a variable (e.g., surface air temperature; SAT) for a given model is the difference of its monthly mean with that of the model climatology. For convenience, we have chosen to use a European SAT index, Teu, which is the area mean monthly SAT within the central European sector defined by 10°E-30°E, 45°N-55°N. Similar to the global mean SAT trend, all models simulate an overall increasing trend of Teu in winter (i.e., December, January and February) from 1950 to the end of the 21st century for both the RCP4.5 and RCP8.5 scenario, with large interannual to decadal variability (not shown). We define a European cold winter month (CWM) in a model simulation as whenever the monthly mean Teu is below the model climatological Teucl of the same month, i.e., Teu < Teucl. This definition of a CWM is very rough, corresponding roughly to a half of the months for CWM in 30 winters in the baseline period assuming a normal distribution. But we argue that our concern in this study is the occurrence of a winter month colder than present climatology seen against a background, where European winter temperature (Teu) continues to increase along with global warming.

[8] Considering that the sea ice concentrations in the B-K Seas has a large inter-annual variability but also has substantially declined in the recent decades [Inoue et al., 2012; Petoukhov and Semenov, 2010], we define a B-K ice concentration index, SICBK, as the area mean monthly sea ice concentration within the B-K sector of 30°E-80°E, 65°N-80°N. In contrast to the increasing trend of winter Teu, the time series of wintertime SICBK(not shown) demonstrates an overall decreasing trend in the CMIP5 future RCP scenarios, also associated with large variability on interannual to decadal timescales. This is further emphasized by a large inter-model variability. For a particular model, the SICBK reduction is assessed with respect to its own baseline climatology in a relative sense as ΔSICBK = (SICBK- SICBKcl)/SICBKcl. The situation will be referred to as ‘No SICBK reductions’, if the SICBK deviates less than 10% from the climatology, or ΔSICBK is positive (i.e., ΔSICBK ≥ −10%).

3. European Cold Winter Episodes in Projections of the Future

[9] Table 1 lists the simulated probability for a European CWM in various periods as simulated by the 13 CMIP5 models. Most models simulate that Teu is below the climatological mean of the present day period of 1971–2000 in slightly less than 50% of the winter months (Table 1, second column), indicating a slightly skewed distribution of Teu towards the warm anomalies. For the last 50 years of the historical period from 1956 to 2005, this probability is about 97% to 120% of the climatology for most models (Table 1, third column). This implies similar or increased probability for the occurrence of a CWM as one may expect when the analyzed period extended toward the mid-20th century. The only exception is for the model CNRM-CM5 that demonstrates about 12% of less probability for CWMs than in the baseline climatology. As Teu continue to increase in the first half of the 21st century (i.e., the period 2006–2050), the probability for CWMs reduces for all models. However, all models still project a clear occurrence of CWMs. For scenario RCP4.5 (Table 1, fourth column), models project the CWM probability in a range from 15% to 42% , that corresponds at least to 35% of the climatology. In particular, 9 out of 13 models project a 23% or higher probability for CWMs, which is exceeding 50% of the corresponding baseline climatology. Some models (i.e., BCC-CSM1.1, CCSM4 and MPI-ESM-LR) even project a chance of CWMs with more than an 80% probability of that in the climatology. The averaged probability of CWMs by all models is 27%, which corresponds to 59% of the model mean baseline climatology. For scenario RCP8.5 (Table 1, sixth column), models project an averaged CWM probability of 23% ranging from 14% to 36%, indicating a model mean average of 52% of CWM compared to the baseline climatology. In the second half of the 21st century, the probability for the occurrence of CWMs reduces considerably. The averaged CWM probability of all models is only 14% for scenario RCP4.5 and 6% for RCP8.5, which are only 31% and 13% of the model mean baseline climatology (Table 1, fifth and seventh columns).

[10] Having seen that there is a substantial probability for European CWMs in the future, we investigate the associated temperature patterns for the historical period and the future. Figure 1ashows the multi-model ensemble of SAT anomaly composite for European cold Januaries in the historical period of 1956–2005. During the European cold January, SAT anomalies of about −2°C or colder cover most of Europe with the cold signal extending to central Siberia and eastern Asia. These cold anomalies are accompanied with weak warm anomalies over Greenland and northeast Canada – reflecting a well-known seesaw relation. To examine the robustness of this pattern, we define avery cold January month as the January Teu being colder than 1.5 standard deviation from the climatological mean, i.e., Teu < Teucl – 1.5 σ(Teucl). The similarity between the SAT anomaly patterns for the European very cold Januaries (Figure 1b) and for the European cold Januaries (Figure 1a) reveals that our rough definition for cold (and very cold) months is able to identify their characteristics. It is worth to point out that, in Figures 1a and 1b, the only significant signals of anomalies in the Arctic region are mostly associated with Greenland, indicating that, during the historical period, European CWMs commonly have little correlation with Arctic Ocean.

Figure 1.

Multi-model ensemble mean of composites of SAT anomalies for European (a) cold and (b) very cold Januaries in the historical period of 1956–2005. The monthly mean anomaly of a given model is the difference of the monthly SAT with that of the respective model climatology as defined insection 2. The crossed areas indicate where more than 75% of the models (i.e., 10 out of 13 models) agree in sign. The European area and the Barents and Kara Sea area used in defining Teu and SICBK in section 2 are marked with red boxes. Unit: °C.

[11] The SAT anomaly pattern for CWMs is very different for the near future (2006–2050) as projected in scenarios RCP4.5 and RCP8.5. As shown in Figure 2, during this period, the European cold Januaries manifest themselves as cold anomalies of below −1°C over most of Europe with a signal extending to central Siberia for both RCP4.5 (Figure 2a) and RCP8.5 (Figure 2b). These cold anomalies are accompanied with strong warm anomalies of more than 2.5°C in the Arctic. In particular, a very strong positive anomaly of more than 5.5°C is seen over the B-K Seas and the surrounding area. The pattern of cold-Europe and warm-Arctic/very-warm-B-K sector is quite similar for RCP4.5 and RCP8.5. The pattern is robust and not just limited to the near future. A similar analysis for the second half of the 21st century (i.e., from 2051 to 2100) of RCP4.5 and RCP8.5 shows consistent patterns to those seen inFigure 2 (not shown).

Figure 2.

Same as in Figure 1 but for European cold Januaries in the future period of 2006–2050 as projected in scenarios (a) RCP4.5 and (b) RCP8.5. The crossed areas indicate where more than 75% of the models (i.e., 10 out of 13 models) agree in sign. The European area and the Barents and Kara Sea area are marked with black boxes. Unit: °C.

[12] The pattern of cold-Europe warm-Arctic indicates a strong connection between the European CWMs and the reduction of sea ice in the Arctic, particularly the B-K region. The difference betweenFigures 1 and 2 evidences that this relationship seems to be connected to the Arctic warming as climate changes, implying that, as a consequence of global warming, the mechanism for the European cold winters in the future might become very different from what have been observed in the past.

4. The Link With the Reduction of Sea Ice in the B-K Area

[13] The accompanied centre of strong positive SAT anomaly over the B-K sector inFigure 2prompts a possible connection between the European CWMs and the sea ice decline in the B-K sector. We therefore examine how the CWMs coincide with the SICBK reduction. For a give winter month, we first check whether Teu is lower than the model climatology, Teucl. If yes, we then check the corresponding SICBK reduction level of the same month. We perform such an analysis through the whole future period of 2006–2100 for both the RCP4.5 and RCP8.5 scenario, as well as for the historical period of 1956–2005. The results are shown in Table 2 and Figure 3. From Table 2 it is clear that, in the historical periods, the European CWMs rarely occur simultaneously with reduced SICBK. Most models (i.e., 12 out of 13 models) simulate less than 30% of the CWMs to coincide with reduced SICBK (i.e., ΔSICBK < −10%). In some models there are only less than 5% of the CWMs coinciding with a reduced SICBK. As a contrast, the CWMs are commonly accompanied with reduced SICBK in the 21st century. In fact, 10 out of 13 models project more than 70% of the European CWMs to be occurring together with the SICBK reductions for both the RCP4.5 and RCP8.5 scenario. Among them, 4 models project 95% or more CWMs coinciding with a reduced SICBK. Even for the models that show a relative rareness of the simultaneous occurrence of European CWMs and SICBKreduction in the 21st century (i.e., BCC-CSM1.1 and IPSL-CM5A-LR for RCP4.5, and BCC-CSM1.1, IPSL-CM5A-LR and MRI-CGCM3 for RCP8.5, respectively), the fraction of such CWMs are much higher than in the historical period. One aspect of this finding is simply reflecting the fact that sea-ice is generally reduced, but the B-K sector never the less stands out as of particular important as already mentioned (e.g.,Figure 2).

Table 2. Ratio (in %) of the Number of European CWMs Coincided With More Than 10% Sea Ice Reduction in B-K Sector With Respect to the Total Number of CWMs in Various Time Periods in Historical and Future Experiments of the RCP4.5 and RCP8.5 Scenarios as Simulated by the CMIP5 Models
ModelClimatology 1971–2000Historical 1956–2005RCP4.5 2006–2100RCP8.5 2006–2100
All model mean15157677
Figure 3.

Fraction (in %) of European CWMs stratified with various categories of SICBK changes (ΔSICBK, in %) with respect to the total number of CWMs in the future period of 2006–2100 as projected in scenarios RCP4.5 (triangle) and RCP8.5 (squared) by the 13 CMIP5 Models. The SICBK changes are classified into five categories with the midpoint of each category marked by a triangle (for RCP4.5) and a squared (for RCP8.5). Category 0% indicates ‘No SICBK reduction’ as define in section 2.

[14] To further explore whether the European CWMs are connected with the SICBK reduction, Figure 3 shows the fraction of European CWMs stratified with various levels of SICBK reductions with respect to the total number of CWMs in the future period of 2006–2100. Interestingly, for the models that project the occurrence of CWMs commonly accompanied with SICBK reduction, a maximum fraction in the occurrence is found at SICBKreduction levels of around −20% or around −40%. Although one model, HadGEM2-ES, projects the maximum fraction of CWMs at the largest SICBKreduction level of more than −70%, it still shows sub-maxima at moderate SICBK reduction levels of around −20% or −40%. Evidently, the condition of SICBK reduction at moderate levels relative to present day climate favors the occurrence of cold European winter in these models.

5. The Associated Circulation Pattern

[15] We next examine how the corresponding circulation pattern of the European CWMs in the future differs from that in the historical period. For the historical period, the multi-model ensemble of the composite of 500 hPa geopotential height anomalies for the European CWMs,Figure 4a, is characterized by a height anomaly contrast between the polar region and midlatitudes. Cyclonic anomalies are centered over Europe and anticyclonic anomalies over Iceland. The anomaly contrast between the polar region and Europe becomes much stronger CWMs in the next few decades (i.e., 2006–2050) as projected in the RCP4.5 scenario shown in Figure 4b. In particular, the magnitude of the positive anomalies doubles with the maximum center shifted northward covering a large area over the Norwegian Sea, Greenland Sea and into B-K Seas. The strong positive height anomalies over the Arctic and negative anomalies over Europe indicate a weakening of the poleward gradient, consistent with previous studies that have shown this tendency [e.g.,Inoue et al., 2012]. The pattern is associated with the negative phase of the Arctic Oscillation/North Atlantic Oscillation (AO/NAO), and is also robust during the whole future period (not shown). Figure 4b resembles greatly the large scale response to reduced ice cover over the Atlantic sector in a study by Alexander et al. [2004, Figure 17], in which they found wintertime negative response between the AO/NAO and the sea ice variability over the Atlantic sector. Less sea ice in the B-K Sea sector and east of Greenland tends to cause weakening of the main branch of the North Atlantic storm track and projects strongly on the negative phase of the AO/NAO with a ridge over the poles and a trough at midlatitudes over the European region.

Figure 4.

Multi-model ensemble mean of composites of 500 hPa geopotential height (contour lines) and its anomalies (color shadings) for European cold Januaries for (a) the historical period of 1956–2005 and (b) the future period of 2006–2050 as projected in scenario RCP4.5. The monthly mean anomaly of a given model is relative to the respective model climatology as define insection 2. Unit: meters.

6. Discussion and Conclusion

[16] The occurrence of the European CWMs and their connection with the sea ice reduction in the B-K Seas has been identified in an analysis using 13 CMIP5 climate change experiments. Despite the projected increase in European winter SATs closely following a global warming trend, most models indicate that CWMs will still occur with a frequency exceeding 50% of the baseline climatology in the first half of the 21st century. Thus the chance for cold winter occurrence in the near future will remain and only somewhat reduced compared to climatology. The characteristics of the CWMs in the future differ from that in the historical period. The projected future CWMs are expressed as a pattern of cold-Europe warm-Arctic anomaly in SAT, and a circulation anomaly pattern resembling the negative phase of the AO/NAO. Along with the SAT anomaly pattern, reduction of sea ice concentration in the B-K Seas occurs coinciding with the European CWMs.

[17] Are recent wintertime extremely cold anomalies in Europe then related to the rapid decrease in the Arctic sea ice extent? The differences between the associated SAT anomaly pattern as projected for the future (Figure 2) and that as simulated for the recent past (Figure 1) prompt that the B-K sea ice among others (e.g., SST or land surface characteristics) could become a dominant factor as a mechanism controlling the occurrence of CWMs, as global warming continues leading to further decline of the sea ice extent in the next decades. A moderate reduction of SICBK (i.e., Figure 3) seems to provide favorable condition for the occurrence of cold winters in Europe. In a data analysis aiming at understanding the role of Barents Sea ice on wintertime cyclone characteristics, Inoue et al. [2012]suggest that sea ice variability over the Barents Sea is likely to control the cyclone tracks through changes in baroclinicity. During light ice years over the Barents Sea, cyclone tracks shift northward resulting from a weak gradient in sea surface temperature, with anomalous warm horizontal advection over the Barents Sea and cold advection over northern Siberia. This leads to a warm-Arctic cold-Siberian anomaly, a pattern highlighting the strong contrast between the Arctic and Eurasia as seen inFigure 2.

[18] When will the cold winters cease to exist? Table 1 shows a large decrease in the probability for the occurrence of the CWMs in the second half of the 21st century in comparison with the first half of the 21st century. In particular, most models (12 out of 13) project only less than a 10% probability for the occurrence of CWMs, which is as little as only 20% of the baseline climatology in the RCP8.5 scenario. Whether or not there could still be cold episodes occurring towards the end of the century remains to be further investigated.


[19] We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1 of this paper) for producing and making available their model output. For CMIP the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. This work was partly funded by the European Commission's 7th Framework Programme, under grant agreement 226520, COMBINE project. In addition, this study received financial support from the Danish Agency for Science, Technology and Innovation through the Centre for Regional Change in the Earth System (CRES; under contract DSF-EnMi 09-066868 and is also part of the Greenland Climate Research Centre (project 6504).

[20] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.