Impacts of global warming on Northern Hemisphere winter storm tracks in the CMIP5 model suite

Authors

  • Timothy Paul Eichler,

    Corresponding author
    1. Department of Earth and Atmospheric Sciences, Saint Louis University, St. Louis, Missouri, USA
    • Corresponding author: T. P. Eichler, Department of Earth and Atmospheric Sciences, Saint Louis University, 3642 Lindell, Blvd, O'Neil Hall 205, St. Louis, Missouri 63108, USA. (teichler@slu.edu)

    Search for more papers by this author
  • Natalie Gaggini,

    1. Department of Earth and Atmospheric Sciences, Saint Louis University, St. Louis, Missouri, USA
    Search for more papers by this author
  • Zaitao Pan

    1. Department of Earth and Atmospheric Sciences, Saint Louis University, St. Louis, Missouri, USA
    Search for more papers by this author

Abstract

[1] A key question in assessing how global warming may affect climate is how it may impact day-to-day weather. To help answer this question, we evaluate the frequency and intensity of northern hemisphere storm tracks in the National Center for Climate Prediction reanalysis I dataset, and the historical, RCP4.5, and RCP8.5 climate scenarios featured in the CMIP5 simulations. We found that a warmer climate resulted in a general decrease in storm frequency in midlatitudes, especially in RCP8.5. In contrast, frequency trends in the reanalysis data reflected an increase in the North Pacific consistent with a shift towards a positive Pacific Decadal Oscillation and more frequent El Niño events post mid-1970s. An examination of frequency and intensity trends in the active storm track regions of the North Pacific and North Atlantic showed that a significant decrease in storm track frequency was evident for RCP8.5. In contrast, intensity trends were dichotomous, with RCP8.5 exhibiting an increase in intensity in the North Atlantic active storm track region and a decrease in intensity in the North Pacific active storm track region. Poleward of these regions, a significant decrease in storm intensity in the North Atlantic and a significant increase in intensity in the North Pacific in RCP8.5 occurred. We also examined the intensity distribution of storms in the active storm track regions of the North Atlantic and North Pacific and determined that the models produced weaker storms with reduced variability relative to reanalysis data regardless of external climate forcing.

1 Introduction

[2] A key point in understanding the impacts of global warming is to assess how the statistics of weather may change. With the advent of model and reanalysis data of sufficient temporal resolution, it has become feasible to assess the role of extratropical cyclones in the current and future climate. The general consensus in the scientific community is that decreased low-level baroclinicity decreases the frequency of storms in midlatitudes during Northern Hemisphere (NH) winter (e.g., Chang et al., 2012; Catto et al., 2011; Bengtsson et al., 2009). Since a warmer climate would result in more available moisture for storms, the intensity of cyclones may also increase.

[3] Observational studies have suggested that midlatitude storm frequency has decreased, while intensity has increased. For example, McCabe et al. (2001) studied storm track trends in National Center for Climate Prediction (NCEP) reanalysis data from 1979 to 1997 and concluded that storm track frequency has decreased in midlatitudes but increased at high latitudes, while storm track intensity has increased at all latitudes. McCabe et al. (2001) suggested that the trends they were seeing may be attributed to the climate shift of the 1970s and a positive shift in the Atlantic Oscillation (AO) post-1989. Wang et al. (2006) compared the NCEP reanalysis and ERA-40 data for two periods, 1958–77 and 1982–2001, and concluded that strong-cyclone intensity increased in the later period in the high-latitude North Atlantic and midlatitude North Pacific. However, some caution is suggested in assessing trends in reanalysis datasets as Chang (2007) found that errors in ship observations in the late 1960s and early 1970s reduce storm track trends to 20%–60% and 70%–80% of what they found in reanalysis for the Atlantic and Pacific, respectively. Keim et al. (2004) studied storm track trends related to oceanic wave climatology in the North Atlantic Basin. They concluded that although storm intensity has increased in agreement with rougher oceanic conditions in the last two to three decades, stronger oceanic waves may have been due to a more positive North AO (NAO). Since Keim et al. (2004) also found that stronger oceanic waves existed in the North Atlantic Basin approximately 100 years prior, they could not conclusively attribute their results to global warming.

[4] Analyses of global warming scenarios in climate models generally support the trends seen by McCabe et al. (2001). For example, Jiang and Perrie (2007) used the Canadian Climate Center model and found that global warming caused storms to increase slightly in size and intensity during northern hemisphere autumn. Ulbrich et al. (2008), Bengtsson et al. (2009), and Yin (2005>) used several models to demonstrate that storms tend to shift their track poleward in response to a warmer climate. Bengtsson et al. (2006) evaluated storms in the European Center for Medium Range Forecasts (ECMWF) model and concluded that while the intensity of storms is unchanged in a global warming climate, the tracks are significantly altered. Although Meehl et al. (2007) summarized the general consensus of the climate community that midlatitude storms are decreasing in frequency and increasing in intensity due to global warming, not all studies support this conclusion. For example, Catto et al. (2011) analyzed storm tracks in the High-Resolution Global Environmental Model version 1.1 (HiGEM1.1) in doubled and quadrupled CO2 scenarios and found that while midlatitude storms decreased in frequency similar to studies done with the IPCC AR4 models, a slight decrease in intense cyclones was also found in response to global warming. Their results agreed with Bengtsson et al. (2009), Watterson (2006), and Geng and Sugi (2003), but disagreed with Lambert and Fyfe (2006), Pinto et al. (2007), and Pinto et al. (2009). As discussed by Catto et al. (2011), the differences may be attributed to either differences in model physics among the different climate models or the various methodologies used to evaluate storm tracks. The impact of applying various storm track methodologies is discussed by Hodges et al. (2011), who note that 850hPa vorticity becomes quite noisy at high resolution requiring a reduction in spatial resolution. However, tracking 850hPa vorticity as opposed to mean sea-level pressure (MSLP) captures more systems, especially weaker ones (e.g., Hodges et al., 2003; Hodges et al., 2011). Although tracking storms by MSLP has the disadvantage of having a dependency on the background flow (Hoskins and Hodges 2002), Wang et al. (2006) found that using the local Laplacian of the pressure field resolves this issue.

[5] In addition to evaluating changes in storm track frequency and trends, several studies have also examined how extreme events linked to storms may have responded to global warming. Salathe (2006) studied storm tracks in the IPCC AR4 models and found that in the North Pacific there is a northward shift and increased intensity in storm tracks in agreement with Yin (2005). However, there was not a corresponding increase in model precipitation. Champion et al. (2011) used the fifth-generation atmospheric general circulation model (ECHAM5) at resolutions of T213 (60 km) and T319 (40 km) and found an increase in extreme precipitation linked to storms in all seasons in the higher resolution model, similar to Bengtsson et al. (2009). Finally, Gastineau and Soden (2009) evaluated a multimodel ensemble of coupled models and found that heavier rainfall occurred in response to global warming despite a reduction in upward vertical motion. Gastineau and Soden (2009) also found a weakening of extreme winds in the tropics and an increased frequency of extreme winds at higher latitudes; the latter due to a poleward shift in the storm track associated with amplified baroclinicity.

[6] Given the results above, it is particularly useful to examine storm tracks in the new CMIP5 models. Our paper is organized as follows: Section 2 presents our methodology for generating storm tracks and evaluating impacts due to global warming. Section 3 discusses our results, and section 4 gives our conclusions.

2 Methodology

[7] We generate storm tracks for the historical and representative concentration pathway (RCP) scenario climates from the CMIP5 climate models for which 6 hourly sea-level pressure (SLP) data was available (Table 1a). Storm tracks are generated using the storm track program described by Serreze (1995) and Serreze et al. (1997) which tracks storms using an absolute minimum in SLP relative to surrounding grid points. We use a one-hPa criterion for identifying storms similar to Eichler and Higgins (2006). We also choose a maximum propagation distance of 800 km between timesteps, which allows for the possibility of center jumps. For the generation of storm track frequency and intensity climatologies for January through March (JFM), we followed the procedure of Eichler and Higgins (2006), which is summarized as follows: Frequency climatologies are computed by binning the storms into 5×5° (lat/lon) boxes. Although arbitrary, this bin size has been found to produce an appropriate climatology (e.g., see Eichler and Higgins, 2006). To compute storm intensity, storm MSLP is binned in a similar fashion as storm track frequency. To eliminate latitudinal dependence, we subtracted the mean JFM SLP climatology from the storm track mean intensity climatology. To accomplish this, we compute a JFM SLP climatology utilizing monthly data. Once complete, we detrend the SLP climatology. Finally, the storm track intensity dataset is created by subtracting the detrended seasonal SLP climatology from the mean storm track intensity climatology.

Table 1. Models and Scenarios Used With Abbreviations
a. Models
Model nameModel CenterAtm. Res. (lon. × lat.)Number of atm. layersReference
CNRM-CM5.1National Center for Meteorological Research, France1.4 × 1.431Voldoire et al., 2011
GFDL-ESM2MGeophysical Fluid Dynamics Laboratory, United States2.5 × 2.048

Donner et al. (2011)

HadGEM2-ESMet Office Hadley Center, United Kingdom1.875 × 1.2538

Jones et al. (2011)

MIROC5Atmos and Ocean Res. Inst., Agency for Marine-Earth Sci & Tech, Japan1.4 × 1.444Watanabe et al., 2010
MRI-CGCM3Meteorological Research Institute, Japan1.1 × 1.148

Yukimoto et al., 2012

b. Scenarios with abbreviations
Re1: Storm tracks generated from Reanalysis I from 1950 to 2004.
Hist: Storm tracks generated from Historical runs for five models from 1950 to 2004.
RCP4.5: Storm tracks generated from RCP4.5 runs for five models from 2006 to 2098.
RCP4.5_trend: Storm tracks generated from RCP4.5 for five models from 2006 to 2058.
RCP4.5_equil: Storm tracks generated from RCP4.5 for five models from 2059 to 2098 when CO2 was in equilibrium.
RCP8.5: Storm tracks generated from RCP8.5 for 5 models from 2006 to 2098.

[8] The design of CMIP5 includes the new short-term decadal experiments hindcasting the interannual variability, emission(versus concentration) driven Earth system model (ESM) simulations exploring the sensitivity of the carbon cycle feedback, and time-evolving land use runs allowing for the dynamic vegetation feedback (Taylor et al., 2012). The core long-term CMIP5 simulations include historical and projection experiments. The projection experiments consist of four new RCP emission scenarios RCP2.6, 4.5, 6, and 8.5, representing anthropogenic radiative forcing stabilizing at 2.6–8.5 W m−2 by 2100 (Moss et al., 2010; Meinshausen et al., 2011). In this study, we focus on the all-forcing historical, RCP4.5, and RCP8.5 experiments. The historical simulations are forced by observed atmospheric composition changes (reflecting both anthropogenic and natural sources). The temporal span of the historical experiment covers 1851–2005, and thus is sometimes referred to as “20th century” simulations (Taylor et al., 2012). The RCP4.5 scenario assumes the anthropogenic forcing will essentially level off at 4.5 Wm−2 around the mid-21st century and represents the intermediate range of the four scenarios. RCP4.5 runs cover 2005–2100. The RCP8.5 scenario increases to 8.5 Wm−2 by the year 2100, with equilibration occurring at 12 Wm−2 by 2250 (some model groups extend it to 2300, see Meinshausen et al., 2011 for details).

[9] To ensure consistency, the ensemble member used in all experiments is the r1i1p1. The realization number (r) is used to distinguish between the different members of the model ensemble and identifies that the run was initialized from a specific point of time in the control run. The “i” variable represents the initialization method indicator, which is used to distinguish between different initializations from observations for the decadal-prediction experiments. The climate models in this study are initialized from the control run and not from observations. The last portion of the ensemble member (p) indicates if the model parameters have been changed. The perturbed physics number changes when a different forcing combination is prescribed by the experiment. The models chosen for this study do not have different prescribed forcings. Therefore, the ensemble number indicates that the experiments will have been initialized from exactly the same initial conditions using the same method and parameters.

[10] Several comparisons are made to assess storm track response to global warming. Since storms occur most frequently in winter, we show results only for JFM. Our results are presented as a mean of all the different models for each climate scenario as shown in Table 1a. The models chosen were based on the availability of 6 hourly SLP data. Although there were some local differences between the various models (e.g., off of southern Greenland), the models were able to capture general features such as the active North Pacific and North Atlantic storm tracks, as well as maximum intensity in the vicinity of the Aleutian and Icelandic lows. All storm track datasets generated are summarized in Table 1b. Our first comparison is between the historical run from 1950–2004, and storm tracks created from NCEP's reanalysis (Re1) data (Kalnay et al., 1996) for the same period. This is done to assess how well the models reproduce storm track climatologies in the current climate.

[11] Although we could have utilized other reanalysis datasets, we chose Re1 because it had the same time period as the historical run. Other studies have compared storm tracks in various reanalysis datasets and have found excellent agreement. For example, Eichler and Higgins (2006) compared storm track intensity and frequency climatology between ERA40 and NCEP reanalysis in the northern hemisphere and found that they were consistent, with insignificant differences between the various datasets. Wang et al. (2006) also compared storm track climatology and trends between the ERA40 and NCEP and found that while they were in general agreement along the U.S. east coast and northern Europe, storm tracks from ERA40 produced stronger storms over the ocean. Trigio (2006) tracked 1000 hPa geopotential height centers for ERA40 and NCEP reanalysis data for the Euro-Atlantic sector and found that they produced similar results, except that ERA40 produced a greater number of storms due to its tendency to better capture subsynoptic scale systems. Finally, Hodges et al. (2003) compared storm tracks from ECMWF, NCEP-NCAR, NCEP-Department of Energy, and NASA reanalysis data, and found excellent agreement in the NH when tracking lower tropospheric features such as 850 hPa vorticity and MSLP. However, there was less agreement in the SH, likely due to a lack of availability of observations relative to the NH. When comparing storm tracks using higher-resolution reanalysis data including Japanese Reanalysis (JRA), NASA MERRA, ERA Interim, and CFS reanalysis, Hodges et al. (2011) found better agreement in the SH.

[12] We also compare RCP4.5 and RCP8.5 from 2006–2098 with the historical run (Hist) to evaluate the effects of increased levels of CO2 on model-simulated storm tracks. For the RCP4.5, we break up the climatologies into two periods. For the first period, we examine the first 53 years of the run in which CO2 is increasing (RCP4.5_trend). The second period is taken for the last 40 years of the run in which CO2 is invariant (RCP4.5_equil). The RCP8.5 experiment is examined from 2006–2098 as CO2 is continually increasing throughout this time period. To evaluate significance, we perform a two-tailed t-test at 90% for all of the difference plots.

[13] Since it is feasible that any response of storm track frequency and intensity to global warming may be due to changes in low-level baroclinicity, we compute the five-model mean of global surface air temperature for each scenario. The global surface air temperature difference of each RCP scenario relative to Hist is then computed to assess the response of low-level baroclinicity to global warming. For the sake of brevity, results are only shown for the RCP8.5-Hist differences.

[14] To assess trends, we performed a linear regression analysis on storm track frequency in each grid box for each of our cases to determine the spatial variability of frequency change. For this analysis, we did not include intensity, as data are temporally inhomogeneous where there is a low frequency of storms. However, we did generate time series of storm track frequency and intensity in the active storm track regions of the midlatitude North Atlantic and North Pacific (see Table 2 for locations). A regional analysis was completed for two reasons: First, to assess intensity trends in the active storm track region, where the data is continuous; second, the active storm track regions represent areas where the upper-level jetstream is strongest, so changes in storm tracks in these areas are also a proxy for jetstream response to climate change. Since the literature suggests a deepening and poleward shift of storms, especially in the North Pacific (e.g., Yin, 2005; Salathe, 2006), we also did this analysis for areas poleward in the vicinity of the Aleutian and Icelandic lows (Table 2). A linear regression analysis is also included to indicate trends in intensity and frequency. To assess the significance of the trends, a t-test was performed at the 95% significance level.

Table 2. Storm Track Regions Used in Frequency/Intensity Analyses
Active Atlantic storm track area: 35°N–45°N, 75°W–55°W
Active Pacific storm track area: 35°N–45°N, 140°E–175°E
Icelandic Low storm track region: 50°N–65°N, 55°W–15°W
Aleutian Low storm track region: 45°N–55°N, 160°E–170°W

[15] To evaluate the intensity distribution of storms, we generated histograms using a 5 hPa bin size for the active storm track regions of the North Atlantic and North Pacific for all model scenarios and reanalysis I data. This was done to determine how well the models represent the storm intensity distribution from reanalysis I data, and if global warming alters the distribution.

3 Discussion

[16] Figure 1 shows the storm track frequency for each model scenario, and for Re1. The various model simulations are consistent with Re1, with the most frequent storms in the North Pacific east of Japan and in the North Atlantic (Figure 1a–e).

Figure 1.

Winter storm track frequency for JFM for a: Re1, b: Hist, c: rcp4.5_trend, d: RCP4.5_equil e: RCP8.5 Units: Average number of storms crossing a 5×5 lat/lon box per season.

[17] Although the frequency distributions agree qualitatively, there are significant differences between the various scenarios relative to historical (Hist, Figure 2). For example, our results suggest that the storm track responds to a poleward shift in the meridional temperature gradient in response to global warming as suggested by a decrease in frequency in RCP8.5 relative to Hist from the Gulf of Alaska eastward to, southern Canada (Figure 2a). A reduction in storm frequency is also noted east of Japan and from the southern U.S. northeastward to Greenland (Figure 2a). Increased frequency is located in the upper-Midwest United States and in Mongolia/China, which may be due to topographical influences (e.g., lee-side cyclogenesis). An increase in frequency is also found in northern Hudson Bay. For RCP4.5_trend (Figure 2b), the pattern is similar to the RCP8.5/Hist differences, although their areal coverage and magnitude are less (compare Figure 2a with Figure 2b). For RCP4.5_equil (Figure 2c), the results are quite similar to RCP4.5_trend although there is a larger region of significant decrease east of Japan, more similar to the RCP8.5/Hist differences (compare Figures 2a and 2b with Figure 2c). Interestingly, the largest differences are seen when comparing Re1 with Hist (Figure 2d). Overall, Hist produces significantly less storms when compared with Re1, although significant increases are seen in the North Pacific from Japan northeastward to the southern tip of the Aleutians and south of the Gulf of Alaska. Significant increases are also seen generally along 65°N. Since the Hist/reanalysis frequency magnitude differences are similar to or greater than the RCP8.5/Hist differences, caution is advised in detecting global warming impacts. However, the evolution of significant storm reduction we see as one proceeds from RCP45_trend through RCP45_equil to RCP85 suggests a response of storm track frequency to global warming, especially when considered in the context of other studies (e.g., Meehl et al., 2007; Catto et al., 2011).

Figure 2.

JFM Storm track frequency difference for a: RCP8.5-Hist, b: RCP4.5_trend-Hist, c: RCP4.5_equil-Hist, d: Hist-Re1 Hatched areas indicate significant at the 90% level for a two-tailed test.

[18] The storm track intensity winter climatologies are broadly similar, with the Aleutian and Icelandic lows clearly depicted (Figure 3). Unlike the frequency differences, the intensity differences show an increase in intensity in RCP8.5 relative to Hist from: 1) the Commonwealth of Independent States (formally USSR) eastward across the North Pacific 2) The central and eastern U.S. northeastward into Canada and 3) off the U.S. east coast (Figure 4a). Interestingly, weaker intensity is located in the North Atlantic to the south of Greenland at the same latitude as the increase in intensity in the North Pacific (Figure 4a).

Figure 3.

JFM storm track intensity relative to regressed JFM SLP climatology (hPa) for a: Re1, b: Hist, c: RCP4.5_trend, d: RCP4.5_equil, e: RCP8.5.

Figure 4.

JFM Storm SLP difference (hPa) for a: RCP8.5-Hist, b: RCP4.5_trend-Hist, c: RCP4.5_equil-Hist, d: Hist-Re1 Hatched areas indicate significance at 90% for a two-tailed test.

[19] The RCP4.5/Hist differences show similar features as RCP8.5/Hist, but once again less in areal coverage and magnitude (Figures 4b and 4c). RCP4.5_equil shows more similarity to RCP8.5 than RCP4.5_trend, which is not surprising considering that RCP4.5_equil is an equilibrium climate state and RCP4.5_trend is a climate state evolving towards RCP4.5_trend. As was the case for the frequency differences, the persistence and evolution of the storm track intensity differences as one proceeds from RCP4.5_trend through RCP4.5_equil to RCP8.5 suggests a climate change response. However, caution is still warranted, since Hist storm track intensity exhibits a weak (strong) bias in mid (high) latitudes, respectively, when compared with Re1 (Figure 4d). As discussed below, differences in low-level baroclinicity between the North Pacific and North Atlantic likely play a role in these differences.

[20] Many of the frequency and intensity changes in the global warming scenarios are likely due to alterations in low-level baroclinicity. To investigate this possibility, we computed the five-model ensemble mean global surface air temperature and generated differences between RCP8.5 and Hist (Figure 5). For example, excessive warming from Alaska extending westward and northward relative to the North Pacific (Figure 5a) suggests that a reduction of baroclinicity played a major role in the decrease in storm frequency in the Gulf of Alaska (e.g., refer to Figure 2a). Excessive warming is also seen over Hudson Bay (Figure 5a), which is consistent with a reduction in baroclinicity and hence, less storms across Southern Canada (refer to Figure 2a). Conversely, baroclinicity is increased from northern Hudson Bay northward (Figure 5a), which agrees with the increase in storm frequency we see in this region (refer to Figure 2a). While a decrease in storms east of Japan may be caused in part by a decreased coastal temperature gradient (Figure 5b), where considerable warming over the Sea of Okhotsk likely plays a major role (Figure 5b). Finally, a NW/SE gradient in warming is found across the eastern U.S. (Figure 5c), which supports the reduction in storm frequency found east of the U.S. east coast (refer to Figure 2a).

Figure 5.

JFM 5-model mean surface air temperature difference (degrees K) for RCP8.5-Hist for a: NH, b: Japan, c: U.S. east coast, d: North Pacific, e: North Atlantic.

[21] Regarding intensity changes, the dichotomy between the North Pacific and North Atlantic are due to different responses in low-level baroclinicity to global warming. For example, although excessive warming over Alaska relative to the North Pacific (Figure 5d) suggests a reduction in storm frequency due to decreased baroclinicity, a warming of the North Pacific of approximately 1 to 3°K (Figure 5d) supports more intense storms due to an increase in available latent heat (LH). In contrast, while warming occurs over Greenland, cooling of SSTs south of Greenland is likely due to melting of the Greenland icecap (Figure 5e). While the reduction in baroclinicity in the North Atlantic suggests a decrease in storm track frequency (similar to the North Pacific), the decreased SSTs south of Greenland also suggests less intense storms due to a decrease in available LH. The zone of more intense storms from the northern U.S. to east of Hudson Bay (refer to Figure 4a) is more difficult to explain; however, it does coincide with an area of increased frequency (refer to Figure 2a) consistent with increased lee cyclogenesis. More open water over the Great Lakes and Hudson Bay would also support stronger storms in this area.

[22] To provide further perspective on the response of storm tracks to global warming, we performed a linear regression on storm track frequency (Figure 6). When comparing Re1 (Figure 6a) with Hist (Figure 6b), we see large differences. For example, Re1 shows an increase in storm frequency in the North Pacific of at least 30% (i.e., increase of storms by at least 2 in an area that averages five to six storms), which is absent in the historical run (compare Figure 6a with Figure 6b). The trend for the North Pacific matches well to the tropospheric response due to the increased frequency of El Nino seen since the 1970s and agrees with the results of Wang et al. (2006). An increase in storms is also noted in Re1 in the high-latitude North Atlantic, with a decrease in the midlatitude North Atlantic, which also agrees with the findings of Wang et al. (2006) and is consistent with the positive trend seen in the NAO during the 1980s and 1990s. The lack of these trends in the historical run relative to Re1 is likely due to the historical run creating its own SST field rather than being forced by observed SSTs.

Figure 6.

JFM Storm track frequency trends (# of storms over entire temporal period) for a: Re1, b: Hist, c: RCP8.5, d: RCP4.5_trend e: RCP4.5_equil. Hatched areas are significant at 90% for a two-tailed test.

[23] In contrast, the RCP8.5 scenario shows a decrease in storm frequency of greater than 30% in the main storm track regions in the North Pacific and North Atlantic (Figure 6c). Smaller decreases (increase) are evident across Canada (northern Hudson Bay). The RCP8.5 trends are consistent with a reduction (increase) in baroclinicity at midlatitudes (high latitudes) in response to global warming. A decrease in storms is also seen in the Mediterranean Sea in RCP8.5 (Figure 6c), which is consistent with the findings of Brayshaw (2005), Bengtsson et al. (2006), Pinto et al. (2007), and Trigo et al. (2000). For RCP4.5_trend (Figure 6d), while there are decreases in storm frequency in the North Atlantic and North Pacific, they are less evident than in RCP8.5 (compare Figure 6c with Figure 6d). A curious feature is the small increase in trends of storms near the southern end of Hudson Bay in RCP4.5_trend, which is of opposite sign to RCP8.5 (compare Figures 6c and 6d). An examination of the five-model mean surface air temperature differences between RCP4.5_equil and Hist (not shown) reveals that while the Great Lakes warmed, the land area just to the north showed little change in temperature, which may have provided a local baroclinic source for this feature. Finally, RCP4.5_equil (Figure 6e) showed some continuation of the trends found in RCP4.5_trend (e.g., decrease in frequency east of Japan and an increase in frequency south of Alaska, suggesting that the storm track response to global warming continued to some degree even after CO2 concentrations equilibrated.

[24] To evaluate storm track frequency trends for active storm track regions of the North Atlantic and North Pacific, time series were generated for the areas shown in Table 2. For the active storm track region in the Atlantic (Figure 7a), storms decrease in Re1 but not in Hist, although neither trend is significant. Re1 also shows greater variability than Hist likely due to Hist being composed of a multimodel mean, which reduces variability. For the RCP4.5 and RCP8.5 scenarios (Figure 7b), a slight decrease in north Atlantic storm tracks is seen for RCP4.5, with a significant decrease in RCP8.5, consistent with the notion that storm track frequency decreases due to a reduction of the meridional temperature gradient. For the North Pacific storm track areas, a nonsignificant increase occurs in Re1 (Figure 7c). Interestingly, a significant decrease occurs in Hist, which is difficult to explain. Similar to the North Atlantic, a decrease in storm frequency is also evident for RCP4.5 and RCP8.5 in the North Pacific, with RCP8.5 once again being significant (Figure 7d).

Figure 7.

JFM Storm track frequency (# of storms) for active storm track regions for a: North Atlantic b: North Atlantic c: North Pacific d: North Pacific. Straight lines show trend (change in storm track frequency/winter). Dashed lines indicate trends are not significant at 95%, while bold lines indicate trends are significant at 95%. Years plotted are 1950–2004 for Re1and Hist and 2006–2098 for RCP45 and RCP85.

[25] Intensity trends generated from the storm tracks program for the North Atlantic and North Pacific active storm track regions are shown in Figure 8. Hist shows little trend in the North Atlantic, while Re1 shows a decreasing (though nonsignificant) trend in intensity (Figure 8a). As was the case for the frequency, the amplitude is much less for Hist than in Re1 (Figure 8a). RCP4.5 shows a slight weakening trend (Figure 8b). However, a significant increase in intensity occurs in RCP8.5 in the North Atlantic active storm track region (Figure 8b). For the North Pacific, a small decrease in intensity occurs in Hist, with a significant increase in intensity in Re1 (Figure 8c). The significant increase in Re1 is consistent with the climate shift in the North Pacific towards a positive Pacific Decadal Oscillation (PDO) and more frequent ENSO events since the 1970s (e.g., Xie et al. 2009 and Mantua et al. 1997). These results are also consistent with the trends found by McCabe et al. (2001) and Wang et al. (2006). For Wang et al. (2006), the trend was confirmed in both NCEP's reanalysis data and ERA40 data. Interestingly, Allen et al. (2010) found no significant trend in either NH or SH explosive cyclogenesis when evaluating annual mean storm tracks from the second NCEP reanalysis (NCEP2), the 25 year JRA (JRA-25), and the ECMWF reanalyses (ERA40 and ERA-INTERIM) over the period 1979–2008. However, Allen et al. (2010) also suggested that caution be used when interpreting potential trends from the ERA-INTERIM data considering its short temporal availability (1989–2008).

Figure 8.

Like Figure 7 but for storm-track generated SLP (hPa). Trend lines show changes in hPa/winter.

[26] The RCP4.5 and RCP8.5 runs both show a decrease in intensity in the North Pacific in contrast to the North Atlantic for RCP8.5 (compare Figure 8b with Figure 8d). The decrease in intensity in the North Pacific in the RCP scenarios is likely due to a reduction in baroclinicity due to a poleward shift of the storm track.

[27] In addition to examining trends in areas of greatest storm track frequency, we also looked at areas of greatest storm track intensity associated with the Icelandic and Aleutian lows.

[28] For storm tracks in the vicinity of the Icelandic low, little trend is seen in storm track frequency for Hist (Figure 9a), while there is a significant positive trend in Re1 (Figure 9a), which agrees well with the positive trend for the NAO seen in the 1980s and 1990s. For RCP4.5 and RCP8.5, a decreasing trend is seen, with the trend being significant for RCP8.5 (Figure 9b). For the area of the Aleutian low, there is a decreasing trend in both Hist and Re1, with Hist being significant. (Figure 9c). Interestingly, a significant positive trend in storm track frequency is noted for RCP4.5 (Figure 9d), which may be due to an increase in baroclinicity along 50°N in the North Pacific, as baroclinicity shifts poleward in response to global warming (e.g., several small areas of increased frequency are seen in RCP4.5_equil from Figure 6d). The absence of this trend in RCP8.5 (Figure 9d) suggests that this effect was temporary as baroclinicity subsequently declined. However, the areal coverage of frequency increase in RCP4.5_equil in the North Pacific is small, which limits our ability to make a firm conclusion.

Figure 9.

Like Figure 7 but for Aleutian and Icelandic low regions.

[29] The intensity of storms in the vicinity of the Icelandic low shows little change in Hist or Re1 (Figure 10a). However, a weakening trend for the Icelandic low is seen in RCP4.5 (Figure 10b), with the trend being significant in RCP8.5 (Figure 10b). A significant deepening of storm intensity in the vicinity of the Aleutian low also occurs in Re1 but not in Hist (Figure 10c), likely due to the tendency towards a positive PDO/more frequent El Niño events in the 1980s and 1990s. RCP4.5 and RCP8.5 also produced a deepening trend for storms in the vicinity of the Aleutian low, with the trend being significant in RCP8.5 (Figure 10d). This is in opposition to the active storm track area equatorward (e.g., compare Figure 10d with Figure 8d). A possible reason for the difference is that while baroclinicity likely decreases in the active storm track areas (as indicated by negative frequency trends in Figure 7d), sufficient baroclinicity (as suggested by the nonsignificant trend in storm track frequency in Figure 9d) combined with warmer SSTs poleward results in more intense storms in the vicinity of the Aleutian low. It is also possible that El Niño frequency may have been changed in the global warming scenarios resulting in more intense storms in the region of the Aleutian low. However, it is beyond the scope of this paper to investigate that possibility. It is also interesting to note the different responses between the North Pacific and North Atlantic with the RCP4.5 and RCP8.5 scenarios showing a deepening (weakening) of the Aleutian (Icelandic) lows. Possible explanations for the weakening trend of the Icelandic low in the RCP scenarios could be melting of the Greenland icecap reducing local baroclinicity between Greenland and the adjacent ocean to the south. A freshening of the North Atlantic due to ice melt may have also reduced North Atlantic SSTs limiting the amount of LH available for storm intensification (refer to Figure 5e).

Figure 10.

Like Figure 8 but for Aleutian and Icelandic low regions.

[30] To demonstrate if global warming causes a change in storm intensity distribution, histograms for the active North Atlantic and North Pacific storm tracks are shown in Figure 11. For the North Pacific, differences are seen between Re1 and Hist, with Hist showing a much more narrow peak centered between 995 hPa and 1000 hPa, as opposed to Re1's broader distribution with a peak between 990 hPa and 995 hPa (Figure 11a). The distributions for RCP8.5 and RCP4.5 are quite similar, with narrow peaks centered between 995 hPa and 1000 hPa similar to Hist (e.g., compare Figure 11a with Figure 11b). For the North Atlantic, the pattern for Hist and Re1 are similar to the North Pacific, with Hist once again showing a narrow peak centered between 995 hPa and 1000 hPa, with Re1 having a broader peak between 990 hPa and 995 hPa (compare Figure 11c with Figure 11a). For RCP4.5 and RCP8.5 (Figure 11d), the spectrum is shifted towards more shallow systems, with a peak between 995 hPa and 1000 hPa for both scenarios. It is apparent from the above analysis that the model simulations produce slightly weaker storms within a more narrow distribution relative to Re1 regardless of any affects from global warming. It would be interesting to see if this model bias also occurs in other simulations (e.g., “colder” simulations and simulations driven by observed SST).

Figure 11.

Histograms for storm intensity in active storm track for a and b: Pacific and c and d: Atlantic.

4 Conclusions

[31] We have presented an analysis of storm tracks in several climate scenarios for CMIP5 models for which 6 hourly SLP data were available. Overall, the CMIP5 suite of models were able to qualitatively reproduce the winter storm track climatology of Re1, featuring active North Pacific and North Atlantic storm tracks and well-defined Aleutian and Icelandic lows. However, when compared with Re1, Hist generally produced significantly less frequent storms, although significant increases occurred in portions of the North Pacific and in the vicinity of 65°N. Intensity differences also occurred, with Hist featuring less (more) intense storms in mid (high) latitudes relative to Re1. Given these differences, caution is warranted when assessing the response of storm tracks to global warming.

[32] When compared with Hist, RCP8.5 showed the largest decrease in storm track frequency, relative to other climate scenarios, especially south of 65°N. Reduced baroclinicity in RCP8.5 likely played a significant role. When comparing Re1 with Hist, the differences were greater than comparing climate change scenarios, although the magnitudes were small. As with the frequency, the changes in intensity were larger as a function of greater external forcing. Relative to Hist, RCP8.5 produced more intense storms in the North Pacific and weaker storms in the North Atlantic.

[33] In response to global warming, frequency trends were mostly negative, with the greatest magnitude occurring in RCP8.5. Decreases in active midlatitude storm track areas (e.g., east of Japan) suggest a reduction in baroclinicity as a likely cause. An increase (decrease) of storms in northern (southern) Hudson Bay in the RCP scenarios may be due to a poleward shift in baroclinicity due to a stronger meridional temperature gradient between a more open Hudson Bay and relatively cold land in the Polar Regions to the north. In Re1, well-defined trends are seen in the North Pacific and North Atlantic. The former may be due to the climate shift in the North Pacific in the 1970's resulting in a positive phase of the PDO/more frequent ENSO events, and the latter due to a positive trend in the NAO in the 1980s and 1990s. The lack of these trends in Hist is likely due to Hist using a free ocean to simulate SSTs as opposed to being forced by observed SSTs.

[34] For Re1, a decrease in storms is noted in the North Atlantic, which when combined with an increase in storms in the vicinity of Iceland, reflects the positive trend in the NAO seen in the 1980s and 1990s. An increase in storm frequency for Re1 in the North Pacific is consistent with a positive shift in the PDO and an increased frequency of ENSO events. Greater year-to-year variability in Re1 relative to Hist is likely due to Hist being composed of a multimodel mean, while Re1 is composed of only one analysis,. Negative frequency trends occurred for RCP8.5 and RCP4.5, with the trends being greater for RCP8.5 indicating a reduction in baroclinicity in these regions.

[35] For the RCP scenarios, the intensity of storms increases in the active storm track regions of the North Atlantic, with decreases in the active North Pacific storm track areas. In the latter, the decrease in intensity may have been due to decreased baroclinicity being a greater influence than increased moisture availability for intense cyclones due to warmer SSTs. This is supported by an increase in intensity poleward towards the Aleutians, where baroclinicity is maintained (as suggested by a lack of frequency trend), working synergistically with warmer SSTs. In the vicinity of the Icelandic low, a weakening trend is noted in RCP8.5 and RCP4.5, possible due to cooler SSTs southeast of Greenland resulting in a reduction of LH available for storm intensification and reduced baroclinicity between Greenland and the North Atlantic.

[36] Finally, the intensity distributions in the active storm track areas showed a smaller range of intensities in all of the model runs relative to Re1 for the North Atlantic and North Pacific. The shape of the distribution from RCP4.5, RCP8.5, and Hist. were narrower than Re1 indicating that the models produce somewhat weaker storms with a smaller distribution of storm intensities relative to Re1 regardless of climate scenario.

[37] Our results agree with the consensus given by Meehl et al. (2007), which illustrated a decrease (or latitudinal shift) in storm track frequency and an increase in intensity in response to global warming. However, regional differences between the North Pacific and North Atlantic appear to be related to different baroclinic responses in the North Atlantic relative to the North Pacific. Future work will focus on changes in oceanic heat content, the magnitude and position of the polar jetstream, and an investigation of storm structure. We intend to generate comparisons with higher-resolution reanalysis datasets such as NASA's MERRA and CFS reanalysis datasets to provide further validation of our results.

Acknowledgments

[38] We acknowledge the international modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP's Working Group on Coupled Modelling (WGCM) for their efforts in making available the WCRP CMIP5 multi-model dataset. This study is partly supported by the NOAA/MAPP grant NA11OAR4310094. We would also like to thank Mr. Frank Alvarez for his assistance in acquiring and processing data used in this manuscript.

Ancillary