Extreme Arctic cyclones in CMIP5 historical simulations


  • Stephen J. Vavrus

    Corresponding author
    1. Nelson Institute Center for Climatic Research, University of Wisconsin-Madison, Madison, Wisconsin, USA
    • Corresponding author: S. J. Vavrus, Nelson Institute Center for Climatic Research, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA. (sjvavrus@wisc.edu)

    Search for more papers by this author


[1] Increasing attention is being paid to extreme weather, including recent high-profile events involving very destructive cyclones. In summer 2012, a historically powerful cyclone traversed the Arctic, a region experiencing rapid warming and dramatic loss of ice and snow cover. This study addresses whether such powerful storms are an emerging expression of anthropogenic climate change by investigating simulated extreme Arctic cyclones during the historical period (1850–2005) among global climate models in the Coupled Model Intercomparison Project 5 (CMIP5) archive. These general circulation models are able to simulate extreme pressures associated with strong polar storms without a significant dependence on model resolution. The models display realism by generating extreme Arctic storms primarily around subpolar cyclone regions (Aleutian and Icelandic) and preferentially during winter. Simulated secular trends in Arctic mean sea level pressure and extreme cyclones are equivocal; both indicate increasing storminess in some regions, but the magnitude of changes to date are modest compared with future projections.

1 Introduction

[2] Many types of extreme weather have become more common in recent decades, accentuated by high-profile events with severe societal impacts—such as Superstorm Sandy, the 2003 and 2010 European heat waves, and the 2010 Pakistan floods—that are consistent with expectations from greenhouse warming [Gleason et al., 2008; Peterson et al., 2012]. Concurrently, the Arctic has been undergoing particularly pronounced climate changes, including rapid warming, snow cover decline, and a loss of marine and terrestrial ice cover [Jeffries and Richter-Menge, 2012]. In 2012, the region was in the spotlight when sea ice extent reached a record low, and a cyclone of unprecedented summertime strength traversed the Arctic Ocean and enhanced the ice loss [Simmonds and Rudeva, 2012]. Strong storms exacerbate coastal erosion—especially when combined with retreating sea ice cover and a warming ocean—which have been increasing in the past few decades and causing considerable societal and biological stress [Overeem et al., 2011].

[3] Given this confluence of large-scale and regional trends, it is natural to wonder how ongoing global climate change is affecting extreme weather in the Arctic. One of the most robust fingerprints of greenhouse-forced simulations by climate models is a drop in high-latitude sea level pressure (SLP) accompanying polar amplification of global warming [Meehl et al., 2007]. This decline in mean SLP is related to an increasing trend in extremely deep cyclones over the Arctic expected in the future [Vavrus et al., 2012], consistent with greater penetration of cyclones into the central Arctic observed during light ice years [Inoue et al., 2012]. These anticipated trends are consistent with forcing mechanisms such as a poleward shift of storm tracks [Bengtsson et al., 2006], boundary layer heating [Deser et al., 2010], northward shifts in baroclinicity in the marginal ice zone [Inoue and Hori, 2011], and enhanced upward surface energy fluxes with greater open water coverage [Simmonds and Keay, 2009].

[4] Most prior research on the response of cyclones to greenhouse warming has considered a hemispheric domain, rather than polar regions specifically. Although past studies have generally concluded that the frequency of cyclones over the Northern Hemisphere will probably diminish in the future, there is disagreement as to changes in strength [Lambert and Fyfe, 2006; Ulbrich et al., 2009; Chang et al., 2012]. Closer to the Arctic, dynamical downscaling simulations by Zahn and von Storch [2010] found that mesoscale polar lows in the North Atlantic should become less common, although the study did not address changes in their intensity. Kolstad and Bracegirdle [2008] showed that general circulation models (GCMs) simulate stronger marine cold-air outbreaks, which are often associated with polar lows, over the region of extensive sea ice retreat in the Barents Sea. Arctic cyclone activity has shown an increasing trend in recent decades [Zhang et al., 2004; Sorteberg and Walsh, 2008] and is enhanced during the positive phase of the Arctic Oscillation [Simmonds et al., 2008].

[5] This study builds on these past findings by focusing on extreme Arctic cyclones in global climate models and their behavior to date in the midst of ongoing climate trends. Specifically, this paper addresses the following three questions: (1) What are the spatial and seasonal characteristics of extreme Arctic cyclones? (2) How well do state-of-the-art GCMs simulate them? and (3) Are extreme Arctic cyclones already showing a response to greenhouse forcing?

2 Methods

[6] This study utilizes two primary data sources: the Coupled Model Intercomparison Project 5 (CMIP5) collection of global climate models [Taylor et al., 2012] and NASA's Modern-Era Retrospective Analysis for Research and Applications (MERRA) reanalysis product [Rienecker et al., 2011]. The GCM data consist of the CMIP5 historical simulations, which were driven by observed radiative forcing during 1850–2005, from 13 climate models with one ensemble member each (Table 1). The MERRA reanalysis covers a much shorter and more recent time interval (1979–2005) and employs finer horizontal resolution (1/2° latitude × 2/3° longitude). Although the twentieth century reanalysis [Compo et al., 2011] extends back to 1871, its Arctic cyclone data early in the record appeared questionable and therefore were not chosen for evaluating the GCMs.

Table 1. Listing of the CMIP5 Models Used in This Studya
ModelCountryHorizontal ResolutionVertical Levels
  1. a

    Horizontal resolution is converted into approximate degrees for the spectral models.

CCSM4United States0.94° × 1.25°26
CNRM-CM5France1.4° × 1.4°31
BCC-CSM1.1China2.8° × 2.8°26
NorESM1-MNorway1.875° × 2.5°26
CanESM2Canada2.8° × 2.8°35
MIROC5Japan1.4° × 1.4°40
CSIRO-MK3.6Australia1.875° × 1.875°18
IPSL-CM5A-LRFrance1.875° × 3.75°39
HadGEM2-ESUnited Kingdom1.25° × 1.875°38
MPI-ESM-LRGermany1.875° × 1.875°47
MRI-CGCM3Japan1.125° × 1.125°48
INMCM4Russia1.5° × 2.0°21
GFDL-ESM2GUnited States2.0° × 2.5°24

[7] Extreme cyclones in this study are defined as occurrences of daily mean SLP at least 40 hPa below the climatological annual average SLP at a grid point. As such, no cyclone tracking algorithm is employed, because the purpose here is to identify instances of extremely strong storms, regardless of origin or propagation. The assumption is that potential damage from polar cyclones is strongly dependent on their central pressure; therefore, it is important to identify all cases when the SLP falls to extremely low values, whether that occurs from separate storms during their 1 day peak intensity or from a single storm throughout its multiday evolution. By identifying extreme cyclones in relative terms, on the basis of a deviation from the local climatological SLP, this method accounts for the large spatial variation in normal pressure patterns across high latitudes, such as the much lower (higher) background SLP around the Icelandic and Aleutian Lows (Beaufort High). Although the choice of a 40 hPa anomaly is arbitrary, it is consistent with the magnitude of SLP perturbation used to identify extreme cyclones in Chang et al. [2012], and the results of the present study are not highly sensitive to the precise value, nor to whether this anomaly threshold is expressed instead in terms of a local standard deviation of SLP. Pressure anomalies of this magnitude are found here to be associated with clearly identifiable cyclones with strong pressure gradients, as expected from such deep central SLP. While extreme cyclones in the central Arctic are of primary interest in this study, the analysis domain extends to 50°N to include the climatological domain of the subpolar Aleutian and Icelandic lows.

3 Results

[8] Figure 1 illustrates the frequency of extreme cyclones simulated by the GCMs, covering the 1979–2005 period to facilitate comparison with MERRA. Averaged among all the models, there are typically 64.4 days per year when an extreme cyclone occurs somewhere poleward of 50°N, compared with 75.8 days annually in MERRA. This 15% underestimate by the CMIP5 simulations is consistent with the simulated dynamical intensity of North Atlantic extratropical cyclones being too weak [Zappa et al., 2013] and the deepest 1% of simulated central pressures over the Arctic being too high (not shown). A noticeable feature in Figure 1 is the large variation in extreme cyclone frequency among GCMs (intermodel standard deviation = 31.2 days), ranging from only 24.1 days annually in MIROC5 to 141.0 days in the outlier MRI-CGCM3 model. Somewhat surprisingly, there is no significant correlation between simulated cyclone frequency and model resolution, either with respect to horizontal grid spacing (r = −0.25) or the number of levels (r = 0.26). This result differs from the analysis of Hodges et al. [2011], who reported that reanalysis products generally depict stronger maximum cyclone intensities with increasing resolution, although their correlation depends on the metric used to define an extreme cyclone (vorticity, wind speed, or minimum SLP). The findings here are in closer agreement with those of Mizuta [2012], who found no apparent relationship between model resolution and the frequency of extreme cyclones during winter.

Figure 1.

Number of days per year with extreme cyclones in the CMIP5 historical simulations (1979–2005). Extreme cyclones are defined as occurrences of daily mean SLP at least 40 hPa below the annual average SLP at a grid point. The multimodel average frequency (64.4) is denoted by the solid line and the MERRA average (75.8) by the dashed line.

[9] The spatial distribution of extreme Arctic cyclones during the entire historical simulation (1850–2005) reveals that these systems are closely associated with the climatological subpolar lows (Aleutian and Icelandic) (Figure 2a) (see supporting information for maps of individual models). These extreme storms are also very much marine phenomena, occurring predominantly over polar oceans and especially in ice-free regions, in agreement with other studies of Arctic cyclones that have applied different classification procedures [Hoskins and Hodges, 2002; Zhang et al., 2004; Mizuta, 2012]. The pattern during more recent years (1979–2005) is very similar (Figure 2b) and generally compares favorably with MERRA (Figure 2c) in terms of favored locations. Compared with the reanalysis, however, the GCMs show a couple of noteworthy differences. Although both models and MERRA simulate a greater annual frequency of extreme cyclones within the Icelandic Low region than the Aleutian Low area, these subpolar maxima in CMIP5 show less activity (2.5 and 2.3, respectively) than in MERRA (3.5 and 2.7). By contrast, the climate models simulate farther poleward propagation of extreme cyclones into the Arctic from both climatological storm tracks.

Figure 2.

Average number of extreme cyclones per year in (a) CMIP5 1850–2005, (b) CMIP5 1979–2005, and (c) MERRA 1979–2005.

[10] These intense storm systems exhibit a pronounced annual cycle, with maximum occurrences during winter and very little activity in summer (Figure 3). This seasonal variation agrees with the results of Zhang et al. [2004] in their analysis of overall Arctic cyclone intensity using the National Centers for Environmental Prediction–National Center for Atmospheric Research Reanalysis [Kalnay et al., 1996]. There is also close agreement between the CMIP5 models and MERRA, except that the GCMs simulate somewhat fewer extreme cyclones than the reanalysis during autumn-early winter and slightly more during late winter and spring. The three winter months account for the majority of all extreme Arctic cyclones in both GCMs (58%) and MERRA (61%). Consistent with this result, the annual cycle of extreme cyclones in the models and reanalysis is significantly correlated with that of the vertically integrated eddy flux of atmospheric moist static energy into the polar cap (70°N), which also peaks during winter [Overland and Turet, 1994] (r = 0.66 in MERRA, r = 0.61 in CMIP5, compared with the monthly energy transports reported by Serreze et al. [2007]).

Figure 3.

Annual cycle of extreme Arctic cyclones in the MERRA reanalysis (dashed) and the CMIP5 average (solid) with intermodel standard deviation depicted with whiskers. The time period of both data sets is 1979–2005.

[11] Given the robust decline in SLP over the Arctic projected by GCMs under future greenhouse gas forcing [Meehl et al., 2007], a logical question is whether such a trend is already underway. The multimodel mean trend in annual SLP since the mid-1800s indicates falling pressure over a wide region of the Arctic (Figure 4a), consistent with this expectation and suggestive of an association with the observed polar warming to date, estimated as 1.8 K since 1875 over land poleward of 59°N [Bekryaev et al., 2010]. The magnitude of the average simulated SLP trend is modest (< 1 hPa everywhere) but statistically significant over much of the Arctic domain and the subpolar climatological lows (using the intermodel standard deviation in a Student's t test). The corresponding trend in extreme cyclone frequency is similar, in that the models simulate an overall increasing trend over high latitudes (50°–90°N) but mostly small changes (Figure 4b) (see supporting information for trends in SLP and extreme cyclones in individual models). Exceptions are the significant positive trends around the Bering Sea-Aleutian Islands, southeast of Iceland, and small patches around the North Pole. Over the Arctic Ocean, the changes in the multimodel mean are generally positive but mostly not significant. However, the one model for which a future trend has been calculated (CCSM4) also produces only modest trends over the central Arctic in its historical simulation (through 2005) but simulates the emergence of more frequent extreme cyclones in this region and its periphery during the 21st century (see supporting information), in conjunction with an average SLP fall of 2–3 hPa. This behavior in CCSM4 suggests that although polar climate change has not been strong enough yet to exert a substantial influence on extreme cyclones, the much more pronounced climatic trends projected in the future may spawn more frequent (and possibly more intense) extreme storms in the interior Arctic.

Figure 4.

(a) Linear trend in mean annual SLP (hPa) from 1850 to 2005 expressed as the multimodel average (shaded) with contours of the 90% significance level, based on a two-sided Student's t test using the intermodel standard deviation of trends. (b) Same but for extreme cyclone frequency trend per decade.

4 Conclusions

[12] An important finding in this study is that GCMs are able to reproduce the basic characteristics of extreme Arctic cyclones (spatial pattern, seasonal variations, and formation of extremely low SLP), despite their relatively low resolution (horizontal grid spacing as coarse as ~2.8° in three of the models). However, such coarseness generally prevents representation of mesoscale polar lows, whose average diameter is around 400 km [Carleton, 1996]. The GCMs in the CMIP5 archive simulate cyclones in high latitudes with exceptionally deep pressures without the need for downscaling, although the frequency of such extreme storms is undersimulated by 15% in the multimodel average. The models produce a realistic spatial pattern of extreme cyclones that emanate from the climatological subpolar low regions and reach the permanent sea ice pack less often. In accord with the annual cycle of mean SLP, extreme Arctic cyclones are most common during winter and very infrequent in summer, making the “great Arctic cyclone of August 2012” [Simmonds and Rudeva, 2012] so remarkable. These major features simulated by the CMIP5 models are similar to those reported by other studies using reanalyses and climate models but adopting different variables and definitions to identify extreme cyclones [e.g., Hoskins and Hodges, 2002; Zhang et al., 2004; Mizuta, 2012].

[13] The GCMs produce an equivocal signal of trends during the historical period of Arctic warming from 1850 to 2005. Consistent with the fingerprint of falling polar SLP from greenhouse forcing, there is a significant decrease in mean annual SLP over a wide expanse of the Arctic, but the decline is small (< 1 hPa) compared with projected changes in the 21st century [Meehl et al., 2007; Vavrus et al., 2012]. The overall frequency of extreme cyclones has trended significantly upward in two regions south of the Arctic Circle (Bering Sea-Aleutian Islands, and southeast of Iceland) but smaller changes with time have occurred over the interior Arctic Ocean so far. A caveat is that trends in extreme cyclones can depend strongly on the definition of such events, such as whether absolute SLP or the Laplacian of pressure is used as the basis for identification [Ulbrich et al., 2009].

[14] Because the CMIP5 historical simulations end in 2005 and undersimulate the observed decline in Arctic sea ice cover [Stroeve et al., 2012], an important consideration is that this study does not address the pronounced changes in the Arctic in the most recent years, when sea ice loss has reached record proportions since 2007. Resembling the CMIP5 multimodel mean pattern, one of the GCMs (CCSM4) produces virtually no trend in extreme cyclones in the interior Arctic through 2005, yet it simulates a pronounced emergence of such storms over the Arctic Ocean and peripheral regions in a 21st century simulation.

[15] The primary goal of this retrospective analysis has been to identify the simulated characteristics of extreme Arctic cyclones in state-of-the-art GCMs, rather than to diagnose the underlying physical mechanisms. The immediate next steps are to identify projected changes in these events through a prospective analysis of the CMIP5 21st century simulations and to determine the physical drivers that might be responsible for altered characteristics of Arctic cyclones in a warming climate.


[16] This research was supported by the Office of Naval Research (N00014-11-1-0611). I acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and thank the climate modeling groups for producing and making available their model output. Computing resources were provided by the Computational and Information Systems Laboratory (CISL) at the National Center for Atmospheric Research, which is sponsored by the National Science Foundation. Helpful comments were provided by Jennifer Francis.

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