Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
1
1
2
3
Arctic summer storm track in CMIP3/5 climate models 4
5
Kazuaki Nishii* ([email protected]) 6
Hisashi Nakamura ([email protected]) 7
Research Center for Advanced Science and Technology, University of Tokyo 8
4-6-1 Komaba, Meguro-ku, Tokyo, 153-8904, Japan 9
10
Yvan J. Orsolini ([email protected]) 11
NILU - Norwegian Institute for Air Research 12
Instituttveien 18, 2027 Kjeller, Norway 13
14
Corresponding author address: Kazuaki Nishii, Research Center for Advanced Science and Technology, 15
University of Tokyo, 4-6-1 Komaba, Meguro-ku, Tokyo, 153-8904, Japan. Tel: +81-3-5452-5147, Fax: 16
+81-3-5452-5148 17
Keywords (4-6): Arctic; cyclone; storm track; climate change; climate model 18
19 First submission to Climate Dynamics on 8 October 2013 20
Revised on 1 May 2014, Accepted on 22 June 2014 21 22 23
2
Abstract 24
Model performance and future projection of Arctic summertime storm-track activity and associated 25
background states are assessed on the basis of Coupled Model Intercomparison Project Phase 3 (CMIP3) / 26
5 (CMIP5) climate models. Despite some improvement in the CMIP5 models relative to the CMIP3 27
models, most of the climate models underestimate summertime storm-track activity over the Arctic Ocean 28
compared to six reanalysis data sets as measured locally as the variance of subweekly fluctuations of sea 29
level pressure. Its large inter-model spread (i.e., model-to-model differences) is correlated with that of the 30
intensity of the Beaufort Sea High and the lower-tropospheric westerlies in the Arctic region. Most of the 31
CMIP3/5 models project the enhancement of storm-track activity over the Arctic Ocean off the eastern 32
Siberian and Alaskan coasts, the region called the Arctic Ocean Cyclone Maximum (AOCM), in 33
association with the strengthening of the westerlies in the warmed climate. A model with stronger 34
enhancement of the storm-track activity tends to accompany stronger land-sea contrast in surface air 35
temperature across the Siberian coast, which reflects greater surface warming over the continent and 36
slower warming over the Arctic Ocean. Other processes, however, may also be likely to contribute to the 37
future changes of the storm-track activity, which gives uncertainty in the projection by multiple climate 38
models. Our analysis suggests that further clarification of those processes that influence storm-track 39
activity over the Arctic is necessary for more reliable future projections of the Arctic climate. 40
41
42
43
44
45
3
1. Introduction 46
As a key component of the Arctic climate system, cyclones transport heat (Jungclaus and Koenigk 47
2010) and moisture (Oshima and Yamazaki 2006; Sorteberg and Walsh 2008) into the Arctic from the 48
lower latitudes. The transported moisture influences the Arctic Ocean through fresh water supply with 49
precipitation (Oshima and Yamazaki 2004; Zhang et al. 2012) and through radiation budget with cloud 50
formation (Sorteberg et al. 2007). In summer and autumn, when the Arctic sea ice is the thinnest in the 51
year, intense cyclones can act to reduce the ice (Simmonds and Keay 2009), as exemplified by the impact 52
of an intense cyclone observed in August 2012 (Simmonds and Rudeva 2012). Transient cyclone activity 53
is also related to the primary mode of circulation variability over the northern mid and high latitudes, 54
called the summertime Northern Annular Mode (SNAM; Ogi et al. 2004), which is also one of factors 55
that modulate the Beaufort Sea High (Serreze and Barrett 2011). The positive phase of SNAM, in 56
association with enhanced cyclone activity over the Arctic Ocean, accompanies a negative pressure 57
anomaly over the Arctic Ocean and a positive pressure anomaly in midlatitudes, while its negative phase 58
tends to accompany a reduction of the Arctic sea ice in September (Ogi and Wallace 2007). Screen et al. 59
(2011) have demonstrated that fewer than usual cyclones are observed over the Arctic Ocean during late 60
spring and early summer concomitantly with both an intensification of the Beaufort Sea High and a 61
reduction of the perennial Arctic sea ice. Realistic representation of cyclone activity over the Arctic 62
Ocean in climate models is thus crucial for deeper understanding of the Arctic climate and its better 63
future projection. 64
Atmospheric disturbances of migratory cyclones and anticyclones tend to organize themselves into 65
zonally elongated domains, called “storm tracks”. Future projection of the storm track activity has been 66
4
investigated (Yin 2005; Lambert and Fyfe 2006; Ulbrich et al. 2008; O’Gorman 2010; Lang and Waugh 67
2011; Woollings et al. 2012; Chang et al. 2013), based on multiple climate models that participated in the 68
World Climate Research Programme’s (WCRP’s) Coupled Model Intercomparison Project Phase 3 69
(CMIP3; Meehl et al. 2007). O’Gorman (2010) found linear scaling between future changes in 70
climatological-mean available potential energy and transient eddy kinetic energy. Woollings et al. (2012) 71
have demonstrated that uncertainties in the future projections for the North Atlantic wintertime 72
storm-track activity by the CMIP3 models are linked to those in the Atlantic Meridional Overturning 73
Circulation that changes lower-tropospheric baroclinicity. Chang et al. (2012), Mizuta (2012), Harvey et 74
al. (2013), and Zappa et al. (2013b) have also investigated the future projection of storm-track activity 75
based on climate models that participate in the phase 5 of CMIP (CMIP5; Taylor et al. 2012). 76
Thus far, studies of storm tracks under the changing climate have focused mostly on midlatitudes. 77
Bengtsson et al. (2006) and Orsolini and Sorteberg (2009) examined a particular storm track that forms 78
only in summer over Northern Eurasia and the Beaufort Sea by applying Lagrangian cyclone tracking to 79
lower-tropospheric (850-hPa) vorticity obtained from the ECHAM and BCM climate models, 80
respectively. Bengtsson et al. (2006) found a future increase in summertime storm activity over the Arctic. 81
Likewise, Orsolini and Sorteberg (2009) found a future increase in the number of storms over the Arctic 82
and along the Eurasian Arctic coast in particular. They pointed out that the enhanced storminess is 83
associated with locally enhanced meridional temperature gradient between the Arctic Ocean and the 84
warmed Eurasian continent and with the enhanced subpolar westerlies as well. Figure 7 of Lang and 85
Waugh (2011) hints a slight increasing tendency in summertime intense cyclones over the Arctic Ocean 86
into future as the multi-model ensemble mean (MEM) among the CMIP3 models. Figure 2 of Harvey et 87
5
al. (2013) also shows future enhancement of the Arctic summer storm-track activity as the MEM among 88
the CMIP5 models. 89
Nevertheless, no such systematic multi-model comparison as Woollings et al. (2012) and Harvey et al. 90
(2013) has been carried out yet to assess model performance and future projection of the climatological 91
activity of the Arctic summer storm track, focusing on its linkage with the background thermal structure 92
and mean atmospheric circulation that can yield inter-model spread in the storm-track activity. This study 93
presents such an assessment based on multi-model data sets for CMIP3 and CMIP5, including 94
benchmarking against atmospheric reanalysis data sets. Although climate models are by no means perfect, 95
they still represent many aspects of the climate system in nature reasonably well. Understanding what 96
yields the inter-model spread in storm-track activity changes into future among climate models should 97
give us some insight into the nature of Arctic cyclones in the current and future climate and also hints for 98
improving climate models for better future projection. 99
The rest of this paper is organized as follows. The data sets and methods used in this study are 100
introduced in section 2. Section 3 presents storm-track activity in the current and future climates in 101
climate models. Section 4 includes discussions on future changes in temperatures, sea ice, and surface 102
heat fluxes, along with uncertainty in reanalysis data sets and comparisons among measures of 103
storm-track activity. A summary is given in section 5. 104
2. Data sets and methodology 105
The Japanese 25-year Reanalysis (JRA-25; Onogi et al. 2007) is used for the benchmarking against the 106
CMIP3 and CMIP5 climate models. We also use the NCEP/NCAR (Kalney et al. 1996), ERA-40 (Uppala 107
6
et al. 2005), NCEP-CFSR (Saha et al. 2010), ERA-Interim (Dee et al. 2011), and JRA-55 (Ebita et al. 108
2011) reanalysis data sets. As we found that uncertainty among those reanalysis data sets is much smaller 109
than that among the climate models, we only show results based on JRA-25. See section 4.4 for details. 110
We analyze outputs of the 17 CMIP3 and 17 CMIP5 climate models, as listed in Tables 1 and 2, 111
respectively. Only one ensemble member is used for each model. For these models, daily-mean outputs of 112
sea-level pressure (SLP), temperature and meridional wind velocity at selected pressure levels are 113
available at the web site of the Program for Climate Model Diagnosis and Intercomparison (PCMDI), 114
which are used to evaluate storm-track activity as noted below. Monthly-mean outputs are used for the 115
other fields. Experiments under the current climatic condition are called 20C3M for CMIP3 and 116
HISTORICAL for CMIP5 (hereafter referred to as 20C experiment), and those under the future climatic 117
scenarios investigated in this study are SRES-A1B for CMIP3 and RCP4.5 for CMIP5 (hereafter referred 118
to as 21C experiment). For all the reanalysis data sets and model outputs of the 20C experiment used in 119
the present analysis, the climatologies for the summer season (June, July and August) have been defined 120
for the 18-year period from 1981 to 1998. The corresponding 18-year climatology for the 21C experiment 121
has been defined for the period from 2081 to 2098 for all the models. We consider the differences in 122
climatology between the 20C and 21C experiments as future changes in the climatological fields. The 123
future changes in globally-averaged JJA-mean surface air temperature (SAT) are 1.8~4.2K (2.5K as 124
MEM) among the 17 CMIP3 models, while they are 1.1~2.6K (1.9K as MEM) among the 17 CMIP5 125
models. This difference can be attributed partly to the difference in the greenhouse gas concentrations in 126
the SRES-A1B and RCP4.5 scenarios. Thus the projections of CMIP3 and CMIP5 are not directly 127
comparable. 128
7
As in Ulbrich et al. (2008) and Woollings et al. (2012), the variance of sub-weekly fluctuations in SLP 129
obtained through 8-day high-pass filtering applied to daily-mean SLP (SLP’2) is evaluated as a local 130
measure of storm-track activity. This measure is qualitatively consistent with 850-hPa poleward eddy heat 131
flux associated with sub-weekly disturbances (V’T’850) as used, for example, by Nakamura et al. (2002) 132
for reanalysis data. The calculation of SLP’2 and V’T’850 is based on daily-mean SLP, meridional wind, 133
and temperature fields. V’T’850 emphasizes baroclinic development of transient disturbances, while 134
SLP’2 measures the local strength of pressure variability associated with the transient disturbances. Note 135
that SLP’2 has been evaluated separately for individual models before taking MEM. The usage of these 136
Eulerian measures enable us to evaluate storm-track activity in the CMIP3 model data, whose 6-hourly 137
outputs necessary for the Lagrangian cyclone tracking are not available (Ulbrich et al. 2008). Although 138
our analysis based on these Eulerian measures cannot treat the intensity and number of individual 139
cyclones, we can show that SLP’2 is a good measure of intense cyclones. See further discussion in section 140
4.4. 141
As a local measure of baroclinicity of the background state in which the transient disturbances are 142
embedded, we focus mainly on climatological-mean meridional SAT gradient, following recent studies 143
that show a crucial role of temperature gradient at the lower boundary or lower-most atmosphere in the 144
maintenance and variability of storm tracks (Nakamura and Shimpo 2004; Brayshow et al. 2008; Ogawa 145
et al. 2012; Woollings et al. 2012). This is consistent with the potential vorticity (PV) thinking (Hoskins 146
et al. 1985), where lower-most temperature anomalies are considered as PV anomalies that interact with 147
upper-atmospheric PV anomalies, leading to development of baroclinic disturbances. 148
All the reanalysis and model fields have first been interpolated onto a regular 2.5°x2.5° 149
8
longitude-latitude grid. The effect of the interpolation is very small. In fact, climatological SLP’2 based 150
on the JRA-25 data interpolated onto a 2.5°x2.5° longitude-latitude grid is smaller only by about 4% than 151
that based on the same data but on a 1.25°x1.25° grid. 152
3. Results 153
3.1 Current climatology 154
In summertime climatology, the Arctic Ocean and its surroundings are characterized by the deep 155
westerlies in both the upper and lower troposphere (Figs. 1a-b). Unlike for the major midlatitude storm 156
tracks, collocation is not necessarily obvious between the local axes of the low-level westerlies and a 157
storm track as measured by SLP’2 (Fig. 1c) and V’T’850 (Fig. 1d). In fact, the SLP’2 and V’T’850 are 158
strongest along the western Siberian coast and slightly to its south, respectively, while the mean low-level 159
westerlies are relatively weak in these regions. To the downstream of this primary storm track, a 160
well-defined band of local maxima of SLP’2 extends from the Siberian coast to the maritime domain off 161
eastern Siberia and Alaska (Figs. 1c-d). This extended storm track forms along a band of local maxima of 162
the westerly wind speed (Figs. 1a-b). This storm track identified through the Eulerian measure is 163
consistent with Serreze and Barry (1988), who found that cyclones most commonly enter the Arctic from 164
Siberia particularly along the Kara and Laptev Sea coast in summer, as the major cyclone of August 2012 165
did (Simmonds and Rudeva 2012). It is also consistent with bands of maxima of track density and 166
intensity of migratory cyclones identified through the Lagrangian tracking method, as in Figs. 1a and 2a 167
of Orsolini and Sorteberg (2009). The storm track over the Arctic Ocean around the date line roughly 168
9
corresponds to a region where the number of cyclones locally maximizes in summer1 (Serreze and 169
Barrett 2008). Hereafter we refer to this region [75-87.5°N, 150-210°E] as the Arctic Ocean Cyclone 170
Maximum (AOCM), marked with red lines in Fig. 1c. Cyclones entering the Arctic from the lower 171
latitudes in summer tend to collect over the AOCM, particularly over its eastern portion (Serreze and 172
Barry 1988). 173
Zappa et al. (2013a) have demonstrated that, in spite of some improvement in the CMIP5 models, the 174
CMIP3/5 models generally underestimate the summertime cyclone intensity. This can be confirmed 175
through a comparison between the climatological SLP’2 and V’T’850 fields based on the JRA-25 data 176
(Figs. 1c-d) and the corresponding MEM fields among the CMIP5 models (Figs. 1g-h) as well as those 177
among the CMIP3 models (not shown). In particular, SLP’2 and V’T’850 as the MEM exhibit no 178
well-defined bands of their maxima from the western Siberian coast to AOCM. Note that this 179
underestimation is not an artifact of the MEM. In fact, among the CMIP5 models, only model P exhibit 180
well-defined bands of their maxima (figures in supplementary material). The underestimation of the 181
Arctic storm-track activity in other models is dynamically consistent with negative MEM biases in speed 182
of the upper- and lower-tropospheric westerlies (Figs. 1e-f). 183
Performance of each of the CMIP3/5 models in reproducing the climatological fields of SLP’2, 184
850-hPa and 300-hPa westerlies over the Arctic is assessed from the viewpoint of pattern similarity with 185
those of the JRA-25 reanalysis data (Fig. 2). In doing so, pattern (or spatial) correlation is evaluated over 186
the region poleward of 60°N between the JRA-25 reanalysis and each of the models, with 187
latitudinally-dependent area weighting in the calculation of spatial variance and covariance. On the whole, 188
1 A recent study found a maximum also in winter around the same domain based on the Arctic System Reanalysis (ASR) interim, which are not clear in global reanalysis data (Tilinina et al. 2013).
10
the CMIP5 models tend to show higher correlations than the CMIP3 models (Fig. 2). In fact, the 189
correlations averaged among the CMIP3 and CMIP5 models are 0.40 and 0.48 for SLP’2, 0.43 and 0.53 190
for the 850-hPa westerlies, and 0.68 and 0.71 for 300-hPa westerlies, respectively. The figure also 191
indicates that models that have higher spatial correlation in a particular field tend to do so in the other 192
fields. In fact, inter-model correlations of any pairs of those fields are 0.57~0.64 (significant at the 5% 193
confidence level). The above analysis suggests that performance in reproducing spatial pattern of 194
storm-track activity over the Arctic is intimately connected to that of the climatological circulation fields. 195
Note that the pattern correlation analysis as above measures only pattern similarity but not intensity of 196
storm-track activity. Still, we also find the latter to be linked to the intensity of the mean westerlies. A 197
scatter plot in Fig. 3a for the CMIP3/5 models between climatological SLP’2 and 850-hPa westerlies both 198
averaged over the AOCM indicates that (i) these two variables are underestimated in most of the models 199
compared to reanalysis data; (ii) the underestimation tends to be less in the CMIP5 models than in the 200
CMIP3 models; (iii) the inter-model spreads among the CMIP5 models (standard deviations are 1.8 hPa2 201
for SLP’2 and 0.63 m/s for 850-hPa westerlies) are slightly smaller than those of the CMIP3 models (2.1 202
hPa2 for SLP’2 and 0.75 m/s for 850-hPa westerlies); (iv) the spread among the models (Fig. 3a) is much 203
larger than that among the 6 reanalysis data sets (Fig. 3b); (v) those two variables exhibit positive 204
inter-model correlation that exceeds the 5% significance level (0.59 among all the models, 0.62 among 205
the CMIP3 models, and 0.59 among the CMIP5 models). To investigate the relationship in 206
model-to-model differences between the storm-track activity averaged over the AOCM and the 207
storm-track activity or 850-hPa westerlies at each grid point over the high and middle latitudes, we 208
calculate inter-model regression between them, together with the corresponding correlation to evaluate 209
11
statistical significance (Fig. 4). Among the CMIP3 models, the local storm-track activity within the 210
AOCM is found to show positive inter-model correlations with that particular activity (Fig. 4a) and the 211
lower-tropospheric westerlies (Fig. 4b) both over the entire polar/subpolar regions and even in the 212
mid-latitudes. Thus the bias in the storm-track activity over the Arctic in a CMIP3 climate model tends to 213
be connected to the circulation bias over the extratropical Northern Hemisphere. Their causal-relationship 214
is, however, hard to clarify because stronger storm-track activity can reinforce the westerlies further, 215
while stronger westerlies are more favorable for cyclone development. In contrast to the CMIP3 models, 216
the corresponding correlations among the CMIP5 models are significant only over the polar and subpolar 217
regions (Figs. 4c-d), which suggest that the bias in the storm-track activity over the Arctic in a CMIP5 218
climate model may be attributable to local processes. 219
In the JRA-25 reanalysis (Fig. 5a), the interannual variability of the storm-track activity observed 220
within the AOCM accompanies cyclonic SLP anomaly over the Arctic Ocean, which resembles the 221
SNAM, as is consistent with Ogi et al. (2004). This SLP anomaly pattern is observed typically when the 222
storm-track activity within the AOCM is enhanced (weakened) by its one standard deviation from the 223
climatology, corresponding to the weakening (enhancement) of the Beaufort Sea High (Fig. 5b (c)). A 224
similar argument can be made for inter-model spreads among the CMIP5 models in the climatological 225
storm-track activity and SLP fields. Models with stronger climatological storm-track activity within the 226
AOCM than its MEM tend to accompany lower SLP over the Arctic Ocean (Fig. 5d), consistently with 227
stronger westerlies (Fig. 4b). Thus, the climatological Beaufort Sea High in such models (Fig. 5e) tends 228
to be weaker than those models with weaker storm-track activity within the AOCM (Fig. 5f). The 229
corresponding SLP difference becomes greater if all the 34 CMIP3/5 models are used for the regression 230
12
(not shown). 231
3.2 Future projection 232
Figure 3c indicates that more than 85% of the CMIP3/5 models (30 out of 34) project future 233
enhancement of the storm-track activity around the AOCM as measured by SLP’2, which is consistent 234
with previous studies based on individual models (Bengtsson et al. 2006; Orsolini and Sorteberg 2009). 235
This enhancement is also evident in the MEM projection among the CMIP5 models (Fig. 6a) and CMIP3 236
models (not shown), as is dynamically consistent with the projected strengthening of both the 237
lower-tropospheric westerlies and southward gradient in SAT across the Siberian coast as the MEM 238
projection (Figs. 6b and 6c, respectively), since migratory cyclones and anticyclones generally develop to 239
relax the meridional temperature gradient while translating westerly momentum downward. In fact, 240
enhancement of the heat transport due to sub-weekly fluctuations (V’T’850) is also projected along that 241
coast in more than half of the models (Fig. 6d). As discussed in detail later, the strengthening of the 242
meridional SAT gradient can be attributed to the projected inhomogeneous surface warming that is 243
greater over the continent than over the Arctic Ocean. While these projected enhancements in the 244
storm-track activity, westerlies, and land-sea thermal contrast are simulated coherently in most of the 245
CMIP3/5 models, their inter-model spreads are pronounced as evident in scatter diagrams in Figs. 3c-d. In 246
a dynamically consistent manner, those models that project stronger enhancement of the storm-track 247
activity (SLP’2) over the AOCM tend to project the stronger low-level westerlies over the AOCM as 248
shown in Fig. 3c (their correlation is +0.70 for the CMIP3/5 models) and the stronger southward SAT 249
gradient averaged over the Siberian coastal region (65-75°N, 60-180°E) shown in Fig. 3d (with their 250
correlation +0.49). We also confirmed that both the southward gradient of air temperature and the Eady 251
13
growth rate evaluated at the 850-hPa level show positive correlation with the storm-track activity, but 252
their correlations are weaker (+0.36 and +0.43) than the southward SAT gradient. The overall tendency 253
can be confirmed in positive inter-model regression and correlation among CMIP5 models along the 254
coasts of the Northeastern Siberia and Northwestern America of the projected changes in the southward 255
SAT gradient with SLP’2 change averaged over the AOCM (Fig. 7a), in a region that roughly corresponds 256
to the Arctic frontal zone (e.g., Serreze et al. 2001). The corresponding positive regression and correlation 257
of the low-level westerlies are also found over the Arctic Ocean (not shown). Positive regression maxima 258
in the future change in V’T’850 along the Siberian coast with that in SLP’2 over the AOCM (Fig. 7b) 259
suggest that development of cyclones along the Siberian coast, which then travel into the AOCM, tends to 260
be more efficient in a model that projects stronger SAT gradient across the coast. This can be confirmed 261
by evaluating inter-model regression/correlation of the V’T’850 change with the corresponding change in 262
meridional SAT gradient averaged along the Siberian coast (Fig. 7c). Specifically, models that yield 263
larger changes in meridional SAT gradient tend to project larger V’T’850 changes over the Siberian coast. 264
We can also show that models with greater changes in the SAT gradient tend to project greater SLP2 265
changes over the Arctic Ocean, and that the inclusion of the Alaskan coast to the averaged domain for the 266
meridional SAT gradient does not change the result significantly (not shown). Such inter-model 267
variability in the storm-track activity change over the AOCM tends to accompany a negative SLP 268
anomaly over the Arctic Ocean (Fig. 7d), which resembles the SNAM (Fig. 5a). Note that those 269
correlations in Figs. 7a-d become more significant if all the 34 CMIP3/5 models are used for evaluating 270
them (not shown). 271
The meridional SAT gradient across the Siberian coast is pronounced in summer between the warmer 272
14
Siberian continent heated by insolation and the cooler Arctic Ocean to the north. The future changes in 273
the coastal SAT gradient projected by the CMIP3/5 models are in significant positive correlation (+0.48) 274
with the corresponding SAT changes averaged over Siberia (55-65°N, 60-180°E) (Fig. 3e), while 275
correlated negatively but insignificantly with the SAT changes averaged over the Arctic Ocean off Siberia 276
(75-85°N, 60-180°E) (–0.30; not shown). If the two CMIP3 models labeled 12 (INGV-SXG) and 14 277
(MIROC3.2(hires)) are excluded, the range of future SAT changes projected over the Arctic Ocean 278
among the CMIP3/5 models is relatively small (0.4~2.8°K) (not shown). The corresponding inter-model 279
spread in the future projection of the SAT gradient is therefore explained primarily by that of SAT change 280
projected over Siberia (0.8~5.5°K) (Fig. 3e). Those two exceptional CMIP3 models also project greatest 281
warming over Siberia (+3.4 and +5.3K for the models 12 and 14, respectively), which is comparable to 282
the warming over the Arctic Ocean (+3.8 and +5.2K) and thus results in relatively small changes in SAT 283
gradient (0.1 and 0.4K/1000km) across the coast (Fig. 3e). 284
We have shown that the future enhancement of storm-track activity over the AOCM projected by most 285
of the 34 CMIP3/5 climate models is accompanied by future enhancement in land-sea thermal contrast 286
across the Siberian and Alaskan coasts. However, 4 out of the 34 models do not project enhancement of 287
the AOCM storm-track activity, although all the models except only two models (E and P) project 288
strengthening of the thermal contrast (Fig. 3d). Figure 7e shows the same map of inter-model 289
regression/correlation of V’T’850 as Fig. 7b does but without the contribution from the SAT gradient 290
change averaged along the Siberian coast through partial regression/correlation technique (See Appendix 291
for details). The effect of the removal of the particular contribution is overall very limited, and thus Figs. 292
7b and 7e look almost identical to one another. This similarity between the two figures suggests that some 293
15
processes other than the land-sea thermal contrast across the Siberian coast can also influence the future 294
projection of the storm-track activity over the Arctic Ocean. Comparison between Figs. 7d and 7f reveals 295
that the contribution of this thermal contrast seems also negligible in the inter-model 296
regression/correlation of climatological SLP changes in correlation with the storm-track activity change 297
over the AOCM. The signature common in these two figures is a SLP anomaly pattern that resembles the 298
SNAM (Fig. 5a), suggesting its important contribution to inter-model spread in the AOCM storm-track 299
activity. In fact, models L and P project greatest weakening of the storm-track activity over the AOCM 300
(Fig. 3c) with positive SLP changes in the Arctic (contours in Fig. 8). These patterns bear certain 301
similarity to the corresponding anomalies for the negative phase of the SNAM simulated in the respective 302
models (shading in Fig. 8 represents the positive SNAM). The pattern correlations between the SLP 303
changes and SNAM-associated SLP anomalies in the Arctic (poleward of 70°N) for the two models are 304
the largest negative among the CMIP3/5 models (–0.41 and –0.67, respectively), indicating that these 305
models project the negative phase of the SNAM under the warmed climate in association with weakening 306
of the storm-track activity in the Arctic. As the SNAM has characteristics of an atmospheric internal 307
mode that can be triggered even without any external forcing to vary the mean westerlies around the 308
Arctic Ocean, it is reasonable to hypothesize that the SNAM can yield uncertainties in the background 309
state for the Arctic storm track simulated under both the current and future climatic conditions. Another 310
possibility is that global warming may trigger the negative SNAM in those two models, but detailed 311
examination how the SNAM responds to the global warming in each of the models are beyond the scope 312
of this study. 313
4. Discussions 314
16
4.1 Projected temperature changes 315
We have shown that the most of CMIP3/5 climate models project future enhancement of the land-sea 316
thermal contrast across the Siberian coast (Fig. 6c), which is contributed to largely by warming over 317
Siberia (Fig. 3e). Model uncertainties in future projection of summertime SAT changes over land can 318
arise from various processes, including soil-moisture-temperature feedback (e.g., Seneviratne et al. 2006), 319
large-scale atmospheric circulation and cloud-related radiation bias (e.g., Cattiaux et al. 2013). It is 320
noteworthy that models projecting larger future changes in SAT over the Arctic Ocean tend to project a 321
larger decrease in Arctic sea ice (correlation is +0.61). The causal relationship is, however, not obvious 322
due to a positive feedback between them. 323
For more detailed discussion of temperature changes, we show future changes in JJA-mean zonal-mean 324
temperatures projected as MEM by the 17 CMIP5 models (Fig. 9a). The warming in the lower-most 325
troposphere over the Arctic Ocean, roughly poleward of 70°N and below the 850-hPa level, is less than in 326
its surroundings. At the same time, the maximum warming in the lower-most troposphere is simulated 327
around 70°N, yielding enhancement of low-level baroclinicity poleward. This is dynamically consistent 328
with enhancement of lower-tropospheric storm-track activity over the Arctic, as represented by the 329
zonal-mean heat flux associated with subweekly disturbances (V’T’) (Fig. 9b). In mid-latitudes, by 330
contrast, meridional temperature gradient weakens in the lower and mid-troposphere due to the rapid 331
warming in the subpolar domain accompanies reduction of mid-latitude storm-track activity. Such mid- 332
and high-latitude differences in projected baroclinicity changes are already pointed out by Oroslini and 333
Sortberg (2009). In contrast to summer, the projected wintertime warming into future is the greatest just 334
above the Arctic Ocean, and this Arctic warming extends up to the mid-troposphere (Fig. 9c). This deep 335
17
warming leads to the reduction of mean baroclinicity, acting to lower the future storm-track activity in 336
both the lower and mid-troposphere (Fig. 9d) over the polar and subpolar regions. The slow warming in 337
summer and rapid warming in winter in the lower-most troposphere over the Arctic Ocean seem to be 338
consistent with the observed temperature trend in the reanalysis data in association with the polar 339
amplification (Screen and Simmonds 2010). Anomalous downward heat flux at the surface (shown in 340
section 4.2) may be related to this slower warming in summer, although further analysis on the 341
near-surface heat budget is necessary. 342
Harvey et al. (2013) found that the inter-model spread among the CMIP5 models in future summertime 343
projection of midlatitude storm-track activity is positively correlated with that of zonal-mean 344
equator-to-pole 850-hPa temperature differences in the lower troposphere (ΔT850NH), evaluated as the 345
difference between the tropics (30°S-30°N) and high latitudes (60°N-90°N). However, they also found 346
negative correlation between ΔT850NH and storm-track activity over the Arctic Ocean, especially around 347
the AOCM. We confirmed that ΔT850NH is negatively correlated with future changes of the storm-track 348
activity over the AOCM (–0.41) and southward gradient in SAT across the Siberian coast (–0.60) both 349
among the 17 CMIP5 models used in this study. Harvey et al. (2013) also noted that the inter-model 350
spread of the future changes of ΔT850NH is dominated by SAT changes at the high latitudes. We have 351
also confirmed this with the corresponding correlation of +0.88 between the two variables, and the 352
correlation is even higher if the averaged domain is limited to 60°N-75°N (+0.91). In fact, the inter-model 353
standard deviation of zonal-mean SAT changes over 60°N-75°N (1.1K), which is mostly over continents, 354
is larger than that over 75°N-90°N (0.59K), mostly over the Arctic Ocean. The inter-model spread of the 355
high latitudes (60°N-90°N) is dominated by the former, and so is the meridional SAT gradient. Our 356
18
analysis suggests that the projected future changes in the summertime storm-track activity over the Arctic 357
are constrained not by hemispheric changes of the equator-to-pole temperature gradient, but mostly by 358
local changes of the lower-most tropospheric temperatures, at least among the 17 CMIP5 models. 359
4.2 Projected changes of sea ice and surface turbulent heat fluxes 360
Compared to ice-covered ocean, ice-free ocean can release more sensible and latent heat fluxes to the 361
atmosphere, which may contribute to the enhancement of cyclone activity. In fact, slight intensification of 362
a cyclone in association with enhanced surface heat fluxes from the ice-free ocean has been demonstrated 363
in numerical experiments by Long and Perrie (2012). As shown in Figs. 10a-b, the CMIP5 models2 as 364
MEM underestimate JJA-mean sea ice concentration over the Arctic in the late 20th century compared to 365
the observations based on HadISST (Rayner et al. 2003). All the CMIP5 models project reduction of sea 366
ice in the future over the Arctic (Fig. 10d). In particular, almost complete loss of sea ice is projected over 367
the Barents Sea (Fig. 10c), and this is the only region where the models project enhancement of the 368
upward turbulent heat fluxes (Fig. 10h). In the rest of the Arctic Ocean, in contrast, models project 369
weakening of the upward heat fluxes in association with sea ice reduction, due to the greater warming in 370
SAT than in SST. We have checked that each of the sensible and latent fluxes shows the same tendency. 371
In some coastal areas, even downward heat fluxes are projected (encircled by white contours in Fig. 10g). 372
The above analysis suggests that future changes in turbulent heat fluxes in association with sea ice 373
reduction may not contribute to the enhancement of the storm-track activity over the AOCM (Figs. 6a and 374
d). The above discussion is, however, based on climatological-mean states, and detailed analysis of the 375
surface heat fluxes in individual cyclone events is required. 376
2We only use CMIP5 models because some of data are not available for some of CMIP3 models.
19
4.3 Underestimation of Arctic storm-track activity in climate models 377
As shown in Figs. 1g-h and 2a, most of the CMIP3/5 models underestimate storm-track activity around 378
the AOCM in the 20C experiment, with large inter-model spread among them. In contrast to the projected 379
future changes, model biases in the storm-track activity in the 20C experiment exhibit no significant 380
correlation with those of the meridional SAT gradient across the Siberian coast (not shown), despite the 381
fact that the standard deviation of the JJA-mean climatology of meridional SAT gradient among the 382
CMIP3/5 models in the 20C experiment is larger than that of the future changes (1.5 and 0.59 K/1000km, 383
respectively). Further analysis is required to determine what factor in the background state controls the 384
summertime storm-track activity and its model biases in the Arctic. Although the physical processes 385
behind it are not necessarily obvious, one possible factor may be model resolution. Model P 386
(MRI-CGCM3) is one of those CMIP5 models that have relatively high horizontal resolution, and the 387
particular model reproduces magnitudes of storm-track activity and low-level westerlies over the AOCM 388
in a fairly realistic manner (Fig. 3a). Unlike the majority of the models, this model projects future 389
weakening of storm-track activity (Fig. 3c) in the AOCM in association with no enhancement of thermal 390
contrast across the Siberian coast (Fig. 3d). There is positive inter-model correlation (+0.41) in the 20C 391
experiment between the number of longitudinal grid points (as a proxy for horizontal resolution of a 392
model) and the climatological-mean SLP’2 over the AOCM. This result is consistent with Chang et al. 393
(2013), who found a similar tendency for cyclone activity over the Northern Hemisphere among the 394
CMIP3 models. Interestingly, however, the future change in the storm-track activity is correlated 395
negatively (–0.41) with the model grid point number. These positive and negative correlations may be 396
related to Chang et al. (2012, 2013), who found a tendency for a model with weaker climatological 397
20
cyclone activity over the Northern Hemisphere to project a greater fractional augmentation in its intensity. 398
While it is understandable that models with higher horizontal resolution can better represent 399
synoptic-scale eddies, there is no simple reasoning for the negative correlation between model resolution 400
and future projection of the storm-track activity. 401
4.4 Comparison among measures of storm-track activity 402
The relationship between the variance of sub-weekly SLP fluctuation (SLP’2) and the number and 403
intensity of cyclones is discussed here, with particular focus on cyclones over the summertime AOCM. 404
We detected a cyclone as a local minimum of daily-mean SLP at a grid point within the AOCM. At that 405
particular grid point, lowering of the local SLP value from the previous day was also evaluated. The local 406
SLP dropping thus detected in the Eulerian sense is largely a manifestation of a migratory synoptic-scale 407
cyclone, which should thus be well compared with SLP’2. On the basis of the 20C experiment outputs 408
from the individual models and the JRA-25 data, inter-model scatter diagrams are plotted in Fig. 11 409
between climatological SLP’2 averaged over the AOCM and the climatological number of cyclones 410
within the AOCM (i) whose central SLP values are below 990hPa (Fig. 11a) or (ii) undergo local SLP 411
dropping by more than 10hPa per day (Fig. 11b). Consistently with the underestimation of SLP’2 in most 412
of the CMIP3/5 models, the criteria (i) and (ii) both lead to the detection of fewer cyclones in most of the 413
climate models than in the reanalysis data. As evident in those diagrams, SLP’2 and the numbers of 414
intense cyclones from the viewpoint of the criteria (i) and (ii) exhibits strong linear relationship with their 415
significant correlations of +0.81 in Fig. 11a and +0.75 in Fig. 11b, respectively, among the CMIP3/5 416
models. Taking it into consideration that inter-model correlations of SLP’2 are insignificant both with the 417
number of local SLP minima whose values are above 990hPa and with the number of local SLP minima 418
21
that undergo local SLP dropping by less than 10hPa per day (not shown), we conclude that SLP’2 is a 419
good measure of the local activity of intense cyclones. It is noteworthy that the underestimation of the 420
number of intense cyclones in most of the CMIP5 models is also found over the North Atlantic (Zappa et 421
al. 2013a). 422
Positive inter-model correlations between the two measures of intense cyclones based on the 20C 423
experiment as discussed above are also found in the future changes (Figs. 11c-d), although no inter-model 424
correlation is found among the CMIP3 models between SLP’2 and the number of cyclones whose central 425
pressures are below 990hPa (Fig. 11c). However, if model 12 (INGV-SXG) is excluded as an outlier, the 426
positive correlation (+0.40) becomes significant at the 5% significance level. It is noteworthy that 25 (24) 427
out of the 34 CMIP3/5 models project future increase of intense cyclones whose central pressures are 428
below 990hPa (with local SLP dropping greater than 10hPa a day). Meanwhile, more than half of the 429
CMIP3 models (10 out of 17) project future increase in the total number of cyclones within the AOCM, 430
although most of the CMIP5 models (14 out of 17) project its future decrease. We need further study on 431
whether the discrepancy in the future projection of the total number of cyclones between the CMIP3 and 432
CMIP5 models is due to development of the models or to the difference in the global warming scenarios 433
(SRES-A1B for CMIP3 versus RCP4.5 for CMIP5). 434
4.5 Uncertainty among reanalysis data sets 435
Due to sparseness of in-situ observations, uncertainty in atmospheric reanalysis data over the Arctic 436
Ocean is larger than in the surrounding regions (Inoue et al. 2009, 2013). Here we assess the uncertainty 437
among the 6 reanalysis data sets with special focus on the storm-track activity and westerlies averaged 438
over the AOCM. As already shown in Fig. 3b, spreads of the two quantities among those reanalysis data 439
22
sets are much smaller than those among the climate models shown in Fig. 3a. In addition, despite slight 440
mutual differences, horizontal distributions of storm-track activity and 850-hPa zonal wind in these 6 441
reanalysis data sets are similar to each other (Fig. 12), which suggests that our results shown above do not 442
depend on a particular choice of a reanalysis data set. It is noteworthy that storm-track activity and 443
westerlies represented in the newer generation of reanalysis data sets (JRA-55, NCEP-CFSR, and ERA-I) 444
are not necessarily stronger than those in the first generation (JRA-25, NCEP/NCAR, and ERA-40). 445
Tilinina et al (2014) reported that new regional reanalysis data (ASR Interim) represents more 446
synoptic-scale cyclones than other modern-era global reanalysis data sets do over the Arctic, which 447
suggests even greater underestimation of storm-track activity in climate models compared to the real 448
climate. 449
5. Concluding remarks 450
In the present study, we have found that most of the CMIP3/5 models have negative biases (i.e., 451
underestimation) in summertime storm-track activity and westerly wind speed around the Arctic Ocean 452
compared to reanalysis data, and spreads of these two variables are mutually correlated among the models. 453
We have also found that future enhancement of summertime storm-track activity over the AOCM 454
projected by the CMIP3/5 models tends to be linked to that of the land-sea meridional SAT gradient 455
across the Siberian coast, the latter of which is accounted for mainly by greater surface warming over 456
Siberia than over the Arctic Ocean. We have further found fairly large inter-model spread in the projected 457
storm-track activity over the AOCM to be correlated with that of the meridional SAT gradient associated 458
with the surface differential warming. Our results suggest that more reliable climate-model projection of 459
the summertime storm-track activity in the Arctic requires deeper understanding of the origin of the 460
23
SNAM variability in the current and future climates, and of processes influencing the inter-model spread 461
in the future changes of land temperatures over Eurasia. 462
463
24
Appendix. Partial correlation and regression 464
Consider three variables of Xi, Yi, and Zi, and linear regression equations among them: 465
𝑋! = 𝑎! 𝑍! + 𝑏! + 𝜖!" ,
𝑌! = 𝑎!𝑍! + 𝑏! + 𝜖!" ,
where ax, bx, ay, and by are constant. Then residuals εxi and εyi are not correlated with Zi. Correlation 466
between εxi and εyi is the partial correlation between Xi and Yi without influence of Zi. Regression 467
coefficient of εyi onto εxi is partial regression of Yi onto Xi without influence of Zi. 468
Acknowledgments 469
KN and HN are supported in part by the Japanese Ministry of Environment through the Environment 470
Research and Technology Development Fund A-1201 and by Japanese Ministry of Education, Culture, 471
Sports, Science and Technology (MEXT) through a Grant-in-Aid for Scientific Research in Innovative 472
Areas 2205. KN is supported by MEXT also through the GRENE Arctic Climate Change Research 473
Project. YO is supported by the Norwegian Research Council East Asian DecCen Project (193690). We 474
acknowledge the World Climate Research Program's Working Group on Coupled Modeling, which is 475
responsible for CMIP, and we thank all contributing climate modeling groups. The U.S. Department of 476
Energy's Program for Climate Model Diagnosis and Intercomparison provided coordination and support 477
for CMIP, and led the development of software infrastructure in partnership with the Global Organization 478
for Earth System Science Portals. We also acknowledge the "Data Integration and Analysis System" Fund 479
(DIAS) for National Key Technology and the Innovative Program of Climate Change Projection for the 480
21st Century ("Kakushin" program) from MEXT. 481
482
25
References 483
Bengtsson L, Hodges KI, Roeckner E (2006) Storm tracks and climate change. J Climate 19:3518−3543 484
Brayshaw D, Hoskins BJ, Blackburn MJ (2008) The storm track response to idealized SST perturbations 485
in an aquaplanet GCM. J Atmos Sci 65: 2842–2860, doi:10.1175/2008JAS2657.1 486
Cattiaux J, Douville H, Peings Y (2013) European temperatures in CMIP5: origins of present-day biases 487
and future uncertainties. Clim Dyn doi 10.1007/s00382-013-1731-y 488
Chang EKM, Guo Y, Xia X (2012) CMIP5 multimodel ensemble projection of storm track change under 489
global warming. J Geophys Res 117: D23118. doi:10.1029/2012JD018578 490
Chang EKM, Guo Y, Xia X, Zheng M (2013) Storm track activity in IPCC AR4/CMIP3 model 491
simulations. J Clim 26: 246-258. doi:10.1175/JCLID-11-00707.1 492
Dee DP and coauthors (2011) The ERA-Interim reanalysis: configuration and performance of the data 493
assimilation system. Q J R Meteorol Soc 137: 553-597 494
Ebita A and coauthors (2011) The Japanese 55-year Reanalysis “JRA-55”: An interim report. SOLA 7: 495
149-152 496
Harvey BJ, Shaffrey LC, Woollings TJ (2013) Equator-to-pole temperature differences and the 497
extra-tropical storm track responses of the CMIP5 climate models. Clim Dyn DOI 498
10.1007/s00382-013-1883-9 499
Hoskins BJ, McIntyre ME, Robertson AW (1985) On the use and significance of isentropic potential 500
vorticity maps. Quart J R Met Soc 111, 877-946 501
Inoue J, Enomoto T, Miyoshi T, Yamane S (2009) Impact of observations from Arctic drifting buoys on 502
the reanalysis of surface fields, 36, Geophys Res Lett L08501, doi:10.1029/2009GL037380 503
26
Inoue J, Enomoto T, Hori ME (2013) The impact of radiosonde data over the ice-free Arctic Ocean on the 504
atmospheric circulation in the Northern Hemisphere, Geophys Res Lett 40, 864-869, doi: 505
10.1002/grl.50207 506
Jungclaus JH, Koenigk T (2010) Low-frequency variability of the arctic climate: the role of oceanic and 507
atmospheric heat transport variations. Clim Dyn 34:265-279. doi:10.1007/s00382-009-0569-9 508
Kalnay E and coauthors (1996) The NCEP/NCAR 40-Year Reanalysis Project. Bull Amer Meteor Soc 509
77(3):437–471 510
Lambert SJ, Fyfe JC (2006) Changes in winter cyclone frequencies and strengths simulated in enhanced 511
greenhouse warming experiments: results from the models participating in the IPCC diagnostic 512
exercise. Clim Dyn 26: 713-728. Doi 10.1007/s00382-006-0110-3 513
Lang C, Waugh DW (2011) Impact of climate change on the frequency of Northern Hemisphere summer 514
cyclones. J Geophys Res 116:D041103. doi:10.1029/2010JD014300 515
Long Z, Perrie W. (2012) Air-sea interactions during an Arctic storm. J Geophys Res 117:D15103, 516
doi:10.1029/2011JD016985 517
Meehl GA, Covey C, Delworth T, Latif M, McAvaney B, Mitchell JFB, Stouffer RJ, Taylor KE (2007) 518
The WCRP CMIP3 multi-model dataset: A new era in climate change research. Bull Amer Meteor Soc 519
88:1383-1394 520
Mizuta R (2012) Intensification of extratropical cyclones associated with the polar jet change in the 521
CMIP5 global warming projections. Geophys Res Lett 39: L19707, doi:10.1029/2012GL053032 522
Nakamura H, Izumi T, Sampe T (2002) Interannual and decadal modulations recently observed in the 523
Pacific storm track activity and East Asian winter monsoon. J Clim 15:1855-1874 524
27
Nakamura H, Shimpo A (2004) Seasonal variations in the Southern Hemisphere storm tracks and jet 525
streams as revealed in a reanalysis dataset. J. Climate, 17, 1828–1844 526
Ogawa F, Nakamura H, Nishii K, Miyasaka T, Kuwano-Yoshida A (2012) Dependence of the 527
climatological axial latitudes of the tropospheric westerlies and storm tracks on the latitude of an 528
extratropical oceanic front. Geophys. Res. Lett. 39: L05804 529
Ogi M, Yamazaki K, Tachibana Y (2004) The summertime annular mode in the Northern Hemisphere 530
and its linkage to the winter mode. J Geophys Res 109:D20114, doi:10.1029/2004JD004514 531
Ogi M, and Wallace JM (2007) Summer minimum Arctic sea ice extent and the associated summer 532
atmospheric circulation. Geophys Res Lett 34: L12705, doi:10.1029/2007GL029897 533
O’Gorman PA (2010) Understanding the varied response of the extratropical storm tracks to climate 534
change. Proc Natl Acad Sci USA 107:19176–19180 535
Onogi K, Tsutsui J, Koide H, Sakamoto M, Kobayashi S, Hatsushika H, Matsumoto T, Yamazaki N, 536
Kamahori H, Takahashi K, Kadokura S, Wada K, Kato K, Oyama R, Ose T, Mannoji N, Taira R 537
(2007) The JRA-25 reanalysis. J Meteo Soc Japan 85:369-432 538
Orsolini YJ, Sorteberg A (2009) Projected changes in Eurasian and Arctic summer cyclones under global 539
warming in the Bergen climate model. Atmos Ocean Sci Lett 2:62-67 540
Oshima K, Yamazaki K (2004) Seasonal variation of moisture transport in the polar regions and the 541
relation with annular modes. Polar Meteorol Glaciology 18:30–53 542
Oshima K, Yamazaki K (2006) Difference in seasonal variation of net precipitation between the Arctic 543
and Antarctic regions. Geophys Res Lett 33:L18501. doi:10.1029/2006GL027389 544
Rayner, NA, Parker DE, Horton EB, Folland CK, Alexander LV, Rowell DP, Kent EC, Kaplan A, 545
28
(2003) Global analyses of sea surface temperature, sea ice, and night marine air temperature since the 546
late nineteenth century. J Geophys Res 108:D14, 4407. 10.1029/2002JD002670 547
Saha S and coauthors (2010) The NCEP Climate Forecast System. J Clim 19:3483–3517 548
Screen JA, Simmonds I (2010) The central role of diminishing sea ice in recent Arctic temperature 549
amplification. Nature 464:1334-1337. doi: 10.1038/nature09051 550
Screen JA, Simmonds I, Keay K (2011) Dramatic interannual changes of perennial Arctic sea ice linked 551
to abnormal summer storm activity. J Geophys Res 116:D15105. doi:10.1029/2011JD015847 552
Seneviratne SI, Lüthi D, Litschi M, Schär C (2006) Land-atmosphere coupling and climate change in 553
Europe. Nature 443:205-209. doi:10.1038/nature05095 554
Serreze MC, Barry RG (1988) Synoptic activity in the Arctic basin, 1979–85. J Clim 1:1276–1295 555
Serreze MC, Lynch AH, Clark MP (2001) The Arctic Frontal Zone as Seen in the NCEP–NCAR 556
Reanalysis. J Clim 14:1550-1567 557
Serreze MC, Barrett AP (2008) The summer cyclone maximum over the central Arctic Ocean. J Clim 558
21:1048−1065 559
Serreze MC, Barrett AP (2011) Characteristics of the Beaufort Sea High. J Clim 24:159-182 560
Simmonds I, Keay K, (2009) Extraordinary September Arctic sea ice reductions and their relationships 561
with storm behavior over 1979–2008. Geophys Res Lett 36: L19715. doi:10.1029/2009GL039810 562
Simmonds I, Rudeva I (2012) The great Arctic cyclone of August 2012. Geophys Res Lett 39:L23709. 563
doi:10.1029/2012GL054259 564
Sorteberg A, Kattosov V, Walsh JE, Pavlova T (2007) The Arctic surface energy budget as simulated 565
with the IPCC AR4 AOGCMs. Clim Dyn 29:131-156. doi:10.1007/s00382-006-0222-9 566
29
Sorteberg A, Walsh J (2008) Seasonal cyclone variability at 70°N and its impact on moisture transport 567
into Arctic. Tellus 60A:570-586 568
Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experiment design. Bull Amer 569
Meteor Soc 93:485-498 570
Tilinina N, Gulev SK, Bromwich DH (2014) New view of Arctic cyclone activity from the Arctic system 571
reanalysis, Geophys Res Lett 41, doi:10.1002/2013GL058924. 572
Ulbrich U, Pinto JG, Kupfer H, Leckebusch GC, Spangehl T, Reyers M (2008) Changing Northern 573
Hemisphere storm tracks in an ensemble of IPCC climate change simulations. J Clim 21:1669−1679 574
Uppala SM and coauthors (2005) The ERA-40 Re-Analysis. Quart J Roy Meteor Soc 131:2961–3012. 575
doi:10.1256/qj.04.176 576
Woollings T, Gregory JM, Pinto JG, Reyers M, Brayshaw DJ (2012) Response of the North Atlantic 577
storm track to climate change shaped by ocean–atmosphere coupling. Nature Geoscience 5:313–317. 578
doi:10.1038/ngeo1438 579
Yin JH, A consistent poleward shift of the storm tracks in simulations of 21st century climate. Geophys 580
Res Lett 32, L18701. doi:10.1029/2005GL023684 581
Zappa G, Shaffrey LC, Hodges KI (2013a) The ability of CMIP5 models to simulate North Atlantic 582
extratropical cyclones. J Clim 26:5379-5396. doi:10.1175/JCLI-D-12-00501.1 583
Zappa G, Shaffrey LC, Hodges KI, Sansom PG, Stephenson DB (2013b) A multimodel assessment of 584
future projections of North Atlantic and European extratropical cyclones in the CMIP5 climate models. 585
J Clim 26:5846-5862. doi:10.1175/JCLID-12-00573.1 586
30
Zhang X, He J, Zhang J, Polyakov IV, Gerdes R, Inoue J, Wu P (2012) Enhanced poleward moisture 587 transport and amplified northern high-latitude wetting trend. Nature Climatic Change, 3: 47-51, doi: 588 10.1038/nclimate1631. 589
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618
31
Table 1. A list of the CMIP3 models used in this study. The third column represents horizontal resolution 619
of atmospheric data provided. 620
Model name Atm. Res. Institution(s)
1 BCCR-BCM2.0 128x64 Bjerknes Centre for Climate Research
2 CGCM3.1(T47) 96x48 Canadian Centre for Climate Modelling & Analysis
3 CGCM3.1(T63) 128x64
4 CNRM-CM3 128x64 Météo-France / Centre National de Recherches Météorologiques
5 CSIRO-Mk3.0 192x96 CSIRO Atmospheric Research
6 CSIRO-Mk3.5 192x96
7 GFDL-CM2.0 144x90 US Dept. of Commerce / NOAA / Geophysical Fluid Dynamics Laboratory
8 GFDL-CM2.1 144x90
9 GISS-AOM 90x60 NASA / Goddard Institute for Space Studies
10 GISS-ER 72x45
11 FGOALS-g1.0 128x60 LASG / Institute of Atmospheric Physics
12 INGV-SXG 320x160 Instituto Nazionale di Geofisica e Vulcanologia
13 INM-CM3.0 72x45 Institute for Numerical Mathematics
14 MIROC3.2(hires) 320x160 Center for Climate System Research (The University of Tokyo), National Institute for Environmental Studies, and Frontier Research Center for Global Change (JAMSTEC) 15 MIROC3.2(medres) 128x64
16 ECHAM5/MPI-OM 192x96 Max Planck Institute for Meteorology
17 MRI-CGCM2.3.2 128x64 Meteorological Research Institute
621
622
32
Table 2. As in Table 1 but for the CMIP5 models used in this study. 623
Model name Atm. Res. Institution(s)
A BCC-CSM1.1 128x64 Beijing Climate Center, China Meteorological Administration
B BNU-ESM 128x64 College of Global Change and Earth System Science, Beijing Normal University
C CMCC-CM 480x240 Centro Euro-Mediterraneo per I Cambiamenti Climatici
D CMCC-CMS 192x96
E CNRM-CM5 256x128 Centre National de Recherches Meteorologiques / Centre Europeen de Recherche et Formation Avancees en Calcul Scientifique
F CSIRO-Mk3.6.0 192x96 Commonwealth Scientific and Industrial Research Organisation in collaboration with the Queensland Climate Change Centre of Excellence
G CanESM2 128x64 Canadian Centre for Climate Modelling and Analysis
H FGOALS-s2 128x60 LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences
I GFDL-ESM2G 144x90 Geophysical Fluid Dynamics Laboratory
J GFDL-ESM2M 144x90
K HadGEM2-CC 192x145 Met Office Hadley Centre
L INMCM4 180x120 Institute for Numerical Mathematics
M MIROC-ESM-CHEM 128x64 Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), and National Institute for Environmental Studies
N MIROC5 256x128 Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology
O MPI-ESM-LR 192x96 Max Planck Institute for Meteorology (MPI-M)
P MRI-CGCM3 320x160 Meteorological Research Institute
Q NorESM1-M 144x96 Norwegian Climate Centre
624
625
626
627
628 629 630 631 632 633 634 635
33
636
Fig. 1 a JJA-mean climatology of 300-hPa zonal wind velocity (m/s) for the period of 1981-1998 based 637
on JRA-25 reanalysis data. b As in a, but for 850-hPa zonal wind velocity. White contours indicate zero. 638
c As in a, but for SLP’2 (hPa2). No data is plotted where climatological JJA-mean surface pressure is 639
under 950hPa. Red lines mark the boundary of AOCM (75°N-87.5°N, 150°E-210°E). d As in a but for 640
V’T’850 (Km/s). e-h As in a-d, respectively, but for multi-model ensemble mean of the HISTORICAL 641
experiment with the 17 CMIP5 models. 642
643
644
645
646
647
648
649
(e) U300 HIST (f) U850 HIST (g) SLP’^2 HIST
(a) U300 JRA-25 (b) U850 JRA-25 (c) SLP’^2 JRA-25
-1 0 1 2 3 4 5 6 7 8 9 10 -0.5 0 1 20.5 1.5 2.5 3 3.5 4 5 6 7 8 9 10 11 12 13
-1 0 1 2 3 4 5 6 7 8 9 10 -0.5 0 1 20.5 1.5 2.5 3 3.5 4 5 6 7 8 9 10 11 12 13
(d) V’T’850 JRA-25
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
(h) V’T’850 HIST
34
650 Fig. 2 Pattern correlations of JJA-mean climatological fields poleward of 60°N between the JRA-25 data 651
and each of the CMIP3/CMIP5 models. The averaged period is 1981 to 1998, and the CMIP3/5 models 652
are based on the 20C experiment. Blue line with squares designates SLP’2, red line with diamonds 653
850-hPa westerlies, and yellow line with triangles 300-hPa westerlies. Labels on the horizontal axis 654
correspond to the individual models listed in Tables 1 and 2. 655
656
657
658
659
660
661
662
663
664
665
666
35
667 Fig. 3 a Scatter plot showing inter-model correlation between JJA-mean SLP’2 (hPa2) and 850-hPa 668
westerlies (m/s) both averaged over the AOCM (75°N-87.5°N, 150°E-210°E) as the climatological 669
statistics based on the 20C experiment (1981-1998). Red numerals and blue alphabets designate the 17 670
CMIP3 and 17 CMIP5 models, respectively, as listed in Tables 1 and 2. A black label “j” signifies the 671
JRA-25 reanalysis data. b As in a, but for 6 reanalysis data sets. The ranges of both axes are different 672
from a. c As in a, but for correlation between future changes in the same variables among the CMIP3/5 673
models. d As in a, but for correlation between future changes in the southward SAT gradient (K/1000km) 674
across the Siberian coast (65°N-75°N, 60°E-180°) and those in SLP’2 over the AOCM. The future change 675
is evaluated by the difference between the 2081-2098 mean of the 21C experiment and 1981-1998 mean 676
of the 20C experiment. e As in d, but for correlation between future changes in the southward SAT 677
gradient (K/1000km) across the Siberian coast and those in the SAT (K) over the Siberian continent 678
(55°N-65°N, 60°E-180°). Note that correlation of 0.48 and 0.34 corresponds to the 5% significance level 679
for 17 and 34 independent samples, respectively. 680
681
682
36
683
Fig. 4 a A map of local inter-model regression of JJA-mean climatological SLP’2 (contour: hPa2) against 684
the same variance but averaged locally within the AOCM (marked with red lines) among CMIP3 models 685
of the 20C3M experiment, plotted for the extratropical Northern Hemisphere (north of 40°N). Shading is 686
for the 10% and 5% significance levels estimated by the t-statistic (negative significance levels 687
corresponding to the negative t-values) based on correlations. b As in a, but for regression of 850-hPa 688
westerlies (m/s). c and d As in a and b, respectively, but among CMIP5 models of the HISTORICAL 689
experiment. The climatological mean for each of the models is evaluated for the period of 1981-1998. 690
691
692
693
694
695
696
697
698
699
700
(a) SLP^2 - SLP^2 C3 (b) U850-SLP^2 C3
5 (%)10-10-5 5 (%)10-10-5
(c) SLP^2 - SLP^2 C5 (d) U850-SLP^2 C5
5 (%)10-10-5 5 (%)10-10-5
37
701
Fig. 5 a A map of local interannual regression of JJA-mean SLP (contour: hPa) against JJA-mean SLP’2 702
averaged locally within the AOCM (marked with red lines) in JRA-25 reanalysis (north of 60°N). 703
Shading is for the 10% and 5% significance levels estimated by the t-statistic. b Regression field shown 704
in a added to the JJA climatological SLP field (shading: hPa). c As in b but with the regression field 705
subtracted (shading: hPa). d As in Fig. 6a, but for inter-model regression of JJA climatological-mean SLP 706
(hPa) (north of 60°N) among CMIP5 models based on the HISTORICAL experiment. e As in b but for 707
the regression field shown in d added to multi-model mean JJA climatological SLP field of CMIP5 708
models (shading: hPa). f As in b but for multi-model-mean JJA climatological SLP field with the 709
regression field subtracted (shading: hPa). The period used for calculating interannual regression and 710
climatology is 1981-1998 for all the panels. 711
712
713
714
5 (%)10-10-5
(d) SLP - SLP^2 C5 (e) SLP CLM + Reg. C5 (f) SLP CLM - Reg. C5
5 (%)10-10-5
(a) SLP - SLP^2 JRA (b) SLP CLM + Reg. JRA (c) SLP CLM - Reg. JRA
100810041000 1012 1016 1020 100810041000 1012 1016 1020
100810041000 1012 1016 1020 100810041000 1012 1016 1020
38
715
Fig. 6 a Future multi-model ensemble mean change in JJA climatology of SLP’2 (contoured for every 716
±0.2 hPa2), projected as the difference between two periods of 2081-2098 and 1981-1998 of 17 CMIP5 717
models. For the statistics for the former and latter periods, outputs of the RCP4.5 and HISTORICAL 718
experiments, respectively, are used. Positive (negative) values indicate enhancement (weakening). 719
Shading indicates percentage of the CMIP5 models that project its positive change at the grid point. The 720
AOCM is indicated with red lines. b-d As in a, but for b 850hPa westerlies (m/s), c southward SAT 721
gradient (K/1000km), and d V’T’850 (Km/s). 722
723
724
725
726
727
728
729
730
731
732
(a) SLP^2 Change (b) U850 Change (c) -dSAT/dy Change (d) V’T’850 Change
1 20 40 8060 99 (%) 1 20 40 8060 99 (%) 1 20 40 8060 99 (%) 1 20 40 8060 99 (%)
39
733 Fig. 7 a A map of local inter-model regression among 17 CMIP5 models of the future change of 734
JJA-mean southward SAT gradient (contour: every 0.4 K/1000km) over the domain north of 60°N against 735
the future change in JJA-mean SLP’2 averaged over the AOCM (marked with red lines). Shading 736
indicates significance levels estimated by the t-statistic. b As in a, but for the corresponding map of the 737
future change in V’T’850 (contour: every 0.05 K m/s). c As in b, but for the corresponding map against 738
the future change in JJA-mean southward SAT gradient averaged over the Siberian coast (marked with 739
red lines: 65-75°N, 60-180°E). d as in a, but for the corresponding map of the future SLP change 740
(contour: every 0.2 hPa). e As in b, but for partial regression/correlation for the future change in V’T’850 741
from which the effect of meridional SAT gradient change averaged over the Siberian coast has been 742
removed. f As in e, but for the corresponding map for the future change in JJA-mean SLP (contour: every 743
0.2hPa). The future change is estimated by the difference between two periods of 2081-2098 (RCP4.5) 744
and 1981-1998 (HISTORICAL). 745
746
747
(a) -dSAT/dy - SLP^2 Ch
5 (%)10-10-5
(b) V’T’850 - SLP^2 Ch
(d) SLP - SLP^2 Ch5 (%)10-10-5
5 (%)10-10-5
(c) V’T’850 - -dSAT/dy Ch
(e) V’T’850 - SLP^2 Ch
5 (%)10-10-5
(f) SLP - SLP^2 Ch5 (%)10-10-5
5 (%)10-10-5
40
748 Fig. 8 a SNAM as interannual variability in the HISTORICAL experiment from 1981 through 1998 by 749
model L (INMCM4), which is defined as the 1st EOF of interannual variability of JJA-mean SLP north of 750
70°N (shading). The pattern is obtained by regressing the 1st principal component time series onto 751
JJA-mean SLP (hPa). The future change of JJA-mean SLP climatology (hPa) projected in the model L 752
under the RCP4.5 scenario is superimposed with contour lines (dashed for negative change). b Same as in 753
a, but for model P (MRI-CGCM3). 754
755
756
757
758
759
760
761
762
763
764
765
766
(a) L:INMCM4 (b) P:MRI-CGCM3
41
767
Fig. 9 a Meridional section of future change in JJA-mean zonal-mean temperature (K: shading) and JJA 768
climatological temperature in the current climate (K: contour), as the MEM of 17 CMIP5 models. b As in 769
a, but for poleward heat flux by transient eddies (V’T’; Km/s). c and d, As in a and b, but for the DJF 770
climatologies, respectively. The future and current climatologies are evaluated for the two periods of 771
2081-2098 (RCP4.5) and 1981-1998 (HISTORICAL), respectively. 772
773
774
775
776
777
778
779
780
781
782
783
784
(a) T JJA500
600
700
800
900
1000EQ 30 60 NP
1 1.5 2 2.5 3 3.5 4 4.5 5 (K) -0.30 -0.18 -0.06 0.06 0.18 0.30 (Km/s)
500
600
700
800
900
1000EQ 30 60 NP
1 1.5 2 2.5 3 3.5 4 4.5 5 (K)
500
600
700
800
900
1000EQ 30 60 NP
-0.30 -0.18 -0.06 0.06 0.18 0.30 (Km/s)
500
600
700
800
900
1000EQ 30 60 NP
(b) V’T’ JJA (c) T DJF (d) V’T’ DJF
42
785 Fig. 10 a JJA climatological-mean sea ice concentration (%) for the period of 1981-1998 based on 786
HadISST (Rayner et al. 2003). b As in a, but for MEM of the HISTORICAL experiment of the 17 787
CMIP5 models. c As in b, but of the RCP4.5 experiment for the period of 2081-2098. d The difference 788
between c and b (contour). Shading indicates percentage of the CMIP5 models that project its negative 789
change at the grid point. e As in a, but for the sum of surface sensible and latent turbulent heat fluxes 790
based on JRA-25 (W/m2). Positive value indicates upward flux. f-h As in b-d, but for the sum of the 791
surface turbulent sensible and latent heat fluxes, respectively. In h, Shading indicates percentage of the 792
models that project its positive change at the grid point. 793
794
795
796
797
798
799
(a) SIC HadISST
15 30 45 7560 90 (%) 15 30 45 7560 90 (%) 15 30 45 7560 90 (%) 1 20 40 8060 99 (%)
-90 -60 -30 300 60 90 (W/m^2) 1 20 40 8060 99 (%)-90 -60 -30 300 60 90 (W/m^2) -90 -60 -30 300 60 90 (W/m^2)
(b) SIC HIST (c) SIC RCP45 (d) SIC Change
(e) STHF JRA25 (f) STHF HIST (g) STHF RCP45 (h) STHF Change
43
800
Fig. 11 a As in Fig. 3a, but for inter-model correlation between JJA climatological-mean SLP’2 (hPa2) 801
averaged over the AOCM and the number of cyclones (per summer) whose central pressures are below 802
990hPa identified within the AOCM. b As in a, but for correlation between the SLP’2 (hPa2) and the 803
number of cyclones (per summer) which accompanies SLP dropping more than 10hPa/day from the 804
previous day. c and d, as in a and b, respectively, but for future changes by the CMIP3/5 models. The 805
future change is estimated by the difference between the 2081-2098 mean of the 21C experiment and 806
1981-1998 mean of the 20C experiment. 807
808
809
810
811
812
44
813 Fig. 12 JJA-mean climatology of SLP’2 (shading; hPa2) and 850-hPa zonal wind velocity (contour; m/s) 814
for the period of 1981-1998. a JRA-25, b ERA-40, c NCEP/NCAR, d JRA-55, e ERA-Interim, and f 815
NCEP-CFSR reanalysis data sets. 816
5 6 7 8 9 10 11 12 13 5 6 7 8 9 10 11 12 13 5 6 7 8 9 10 11 12 13
5 6 7 8 9 10 11 12 13 5 6 7 8 9 10 11 12 13 5 6 7 8 9 10 11 12 13
(a) JRA-25 (b) ERA-40 (c) NCEP/NCAR
(f) NCEP-CFSR(e) ERA-Interim(d) JRA-55