Mechanisms of Decadal Arctic Climate Variability in the Community Climate System Model, Version 2 (CCSM2)

Transcription

1 3552 J O U R N A L O F C L I M A T E VOLUME 18 Mechanisms of Decadal Arctic Climate Variability in the Community Climate System Model, Version 2 (CCSM2) HUGUES GOOSSE Institut d Astronomie et de Géophysique G. Lemaître, Université Catholique de Louvain, Louvain-la-Neuve, Belgium MARIKA M. HOLLAND National Center for Atmospheric Research, Boulder, Colorado (Manuscript received 1 September 2004, in final form 9 February 2005) ABSTRACT Several mechanisms have been proposed to explain natural climate variability in the Arctic. These include processes related to the influence of the North Atlantic Oscillation/Arctic Oscillation (NAO/AO), anticyclonic/cyclonic regimes, changes in the oceanic and atmospheric North Atlantic Arctic exchange, and changes in the Atlantic meridional overturning circulation. After a brief critical review, the influence and interrelation of the above processes in a long climate integration of the Community Climate System Model, version 2 (CCSM2) are examined. The analysis is based on the time series of surface air temperature integrated northward of 70 N, which serves as a useful proxy for general Arctic climate conditions. This gives a large-scale view of the evolution of Arctic climate. It is found that changes in oceanic exchange and heat transport in the Barents Sea dominate in forcing the Arctic surface air temperature variability in CCSM2. Changes in atmospheric circulation are consistent with a wind forcing of this variability, while changes in the deep overturning circulation in the Atlantic are more weakly related in CCSM2. Over some time periods, the NAO/AO is significantly related to these changes in Arctic climate conditions. However, this is not robust over longer time scales. 1. Introduction Marked climate variations have occurred in the high latitudes of the Northern Hemisphere during the last decades as summarized in recent reviews (Serreze et al. 2000; Moritz et al. 2003; Overland et al. 2004). These changes include a large increase in surface air temperature with a trend of as much as 2 C per decade in spring in some regions (Chapman and Walsh 1993; Rigor et al. 2000) and a decrease in Arctic sea level pressure (Walsh et al. 1996). Furthermore, the Arctic ice extent decreased by km 2 from 1972 through 2002 (Cavalieri et al. 2003). Observations made using submarine-based sonars suggest a significant decrease of the sea ice thickness during the same period, although the magnitude of the inferred change depends on the Corresponding author address: Dr. Hugues Goosse, Institut d Astronomie et de Géophysique G. Lemaître, Université Catholique de Louvain, Chemin du Cyclotron, 2, B-1348 Louvainla-Neuve, Belgium. hgs@astr.ucl.ac.be timing and location of the cruises (Rothrock et al. 1999; Wadhams and Davis 2000; Tucker et al. 2001; Holloway and Sou 2002). In the ocean, observations performed during the 1990s indicate a retreat of the halocline in the Eurasian basin (Steele and Boyd 1998) as well as a subsurface warming in the Arctic (Carmack et al. 1995; Grotefendt et al. 1998) caused by a larger-than-usual inflow of water originating in the North Atlantic. In addition to these recent trends, large decadal to multidecadal climate variations have also been observed earlier in the Arctic. For instance, long reconstructions of the ice conditions in the Nordic Seas (Vinje 2001a), in the Barents Sea (Vinje 1999), and around Iceland (e.g., Ogilvie 1992) display significant decadal variability superimposed on variability on longer time scales. Based on a compilation of various paleoclimate records covering the last four centuries, Overpeck et al. (1997) obtain a reconstruction of surface temperature in the Arctic indicating centennialscale variations with the recent warming marking the end of a relatively cold period in the whole Arctic that started before Furthermore, a well-known ex American Meteorological Society

2 1 SEPTEMBER 2005 G O O S S E A N D H O L L A N D 3553 FIG. 1. (top) The time series of the NAO index [dashed, right axis; updated from Hurrel; (1995)] and temperature anomaly in the Arctic in C (solid left axis; Polyakov et al. 2002). A 5-yr running mean has been applied to the time series. (bottom) Running correlation between the NAO index and the temperature anomaly in the Arctic using a 25-yr time window. ample of multidecadal change in the Arctic occurs during the twentieth century with high surface temperature in the early 1940s and in the 1990s, while the 1960s where characterized by significantly colder conditions (e.g., Polyakov et al. 2002; Johannessen et al. 2004; Fig. 1). In addition, Russian observations of ice thickness, as well as model simulations of the evolution of the ice cover in the Arctic during the last 50 yr, clearly display large variations on decadal time scales (e.g., Johannessen et al. 2004; Hilmer and Lemke 2000; Holloway and Sou 2002; Polyakov et al. 2003a; Goosse et al. 2004). After the oceanic warming in the 1990s, the subsurface temperatures display a cooling and a partial recovery to past conditions, and other Arctic Ocean warming events have been documented in the 1930s, 1960s, and 1970s (e.g., Saloranta and Haugan 2001; Polyakov et al. 2003b; Gerdes et al. 2003). It is nearly certain now that at least a part of the recent changes in the Arctic are related to human activities (e.g., Vinnikov et al. 1999; IPCC 2001; Johannessen et al. 2004). Furthermore, coupled atmosphere ocean general circulation models (CGCMs) agree that large changes are expected in the Arctic as a consequence of human activities (e.g., Vinnikov et al. 1999; IPCC 2001; Holland and Bitz 2003; Johannessen et al. 2004). To improve our projection of future Arctic climate change, it is necessary to more clearly understand the relative role of anthropogenic contributions and natural variations in forcing the recent changes discussed above. To do this, a better understanding of the mechanisms that govern the large-scale natural climate variability of the Arctic is needed. In the last few years, numerous studies have analyzed the observed climate variability of the Arctic. However, they have used different datasets and different time periods for analysis. This makes it difficult to identify the links or the differences between the various processes described. Additionally, from observational

3 3554 J O U R N A L O F C L I M A T E VOLUME 18 studies, it is difficult to ascertain variations in the Arctic system that are naturally forced from those that are anthropogenically forced. The goal of the present study is to provide additional insight into the mechanisms recently proposed to explain natural variability in the Arctic. To do so, we will first present a critical overview of previous studies and will discuss if recent analyses support or invalidate some of the hypotheses that have been invoked to explain Arctic climate variability (section 2). These various hypotheses will then be examined in a 1000-yr-long control experiment performed with the Community Climate System Model, version 2 (CCSM2). This control integration has no changes in anthropogenic forcing, allowing us to isolate the natural variations in the simulated Arctic climate system. Additionally, the model simulation provides a complete, self-consistent dataset for analysis. It is thus possible to look at the respective influence and the links between the different processes reviewed in section 2. The CCSM2 model is presented in section 3. In section 4, the mechanisms driving the simulated large-scale Arctic climate variability are analyzed, before some concluding remarks are given (section 5). 2. A review of mechanisms driving Arctic variability a. Processes related to North Atlantic Oscillation/Arctic Oscillation The most widely studied mode of variability in the Arctic is associated with the North Atlantic Oscillation or the closely related Arctic Oscillation [also referred to as the Northern Annular Mode (NAM)], the two of which are the leading modes of atmospheric variability in the North Atlantic and in the extratropical Northern Hemisphere, respectively (e.g., Hurrell 1995; Thompson and Wallace 1998; Deser 2000). The North Atlantic Oscillation (NAO) is characterized by a dipole of surface pressure between mid- and high latitudes, resulting in changes in the strength of the westerly winds in midlatitudes and large winter temperature variations, in particular over northern Europe and Asia. A positive index corresponds to stronger westerlies, a decrease in surface pressure at high latitudes, and warmer conditions in Europe. The changes at high latitudes related to the NAO/Arctic Oscillation (AO) are thus imbedded in climate variations at nearly hemispheric scale. As discussed in recent studies, a large fraction of the climate changes observed during the last decades in the Arctic could be related to the positive trend in the NAO/AO index during this period (e.g., Grotefendt et 1998; Dickson et al. 2000; Rigor et al. 2000; Rigor et al. 2002; Moritz et al. 2003). Years with a high NAO index are characterized by a cooling and an increased ice cover in the Labrador Sea as well as a warming and a decreased ice cover in the Greenland and Barents Sea (Walsh and Johnson 1979; Slonosky et al. 1997; Deser et al. 2000). The NAO also influences the ice velocity (e.g., Kwok 2000; Rigor et al. 2002; Krahmann and Visbeck 2003; Zhang et al. 2004) and is associated with a reorganization of the sea ice mass between the Siberian Arctic and the North American Arctic (e.g., Zhang et al. 2000; Rigor et al. 2002; Zhang et al. 2004). As discussed by Holland (2003), the CCSM2 model accurately captures many of these relationships between the NAO/AO and Arctic climate. In addition to the response of the ocean ice system to the large-scale atmospheric conditions associated with the NAO/AO, Mysak and Venegas (1998) have proposed that modifications of ice extent could have a significant influence on the atmospheric circulation. This hypothesis is part of a feedback loop in which large positive ice anomalies are created in the Beaufort Sea during the positive phase of the NAO index. These anomalies are then transported out of the Arctic resulting in a higher-than-normal ice concentration in the Greenland Sea after 3 4 yr. This results in a reduced winter heat flux to the atmosphere in this region that could lead to a change in the intensity of the low pressure systems in the Atlantic and modify the NAO polarity. This provides the negative feedback necessary to sustain a feedback loop with a time scale of about 10 yr. The formation of ice anomalies associated with the NAO and their propagation in the Arctic have been confirmed by other studies (e.g., Arfeuille et al. 2000). However, recent analyses do not display a dominant influence of ice concentration anomalies in the Nordic Seas on the large-scale atmospheric circulation (e.g., Alexander et al. 2004; Magnusdottir et al. 2004). As a consequence, these studies fail to confirm the strong feedback between atmospheric circulation and surface conditions needed for the Mysak and Venegas (1998) loop, or for the conceptual models of Ikeda et al. (2001) and Dukhovskoy et al. (2004). It should be stressed that the majority of the data used to derive the relationship between NAO/AO and Arctic climate were obtained during the last 40 yr, a period characterized by a large-scale warming in the Arctic and large positive trends of NAO/AO indexes. For the period , an episode with a persistent positive winter NAO index, a similar Arctic warming is mentioned by Rogers et al. (2004). However, the warm conditions of the period were not related to the NAO, which was weakening at this time (Bengtsson et al. 2004). As a consequence, the correlation between

4 1 SEPTEMBER 2005 G O O S S E A N D H O L L A N D 3555 the NAO index and Arctic temperature appears relatively high for the period but is negative for the period (Fig. 1). This suggests that the relationship between the NAO/AO and some aspects of Arctic climate may not be robust for longer time scales. This could be due to small shifts in the spatial patterns associated with NAO/AO, shifts that would have a particularly large influence over the Arctic. For instance, the Fram Strait ice transport is well correlated with the NAO for the last 25 yr. But, because of a small modification of the pressure changes associated with the NAO, the correlation is very weak for the previous 25-yr period (Kwok and Rothrock 1999; Hilmer and Jung 2000; Vinje 2001b; Schmith and Hansen 2003). This is also supported by coupled climate modeling studies (Holland 2003), which show no robust relationship between the NAO and Fram Strait ice export over a long integration. b. Processes related to cyclonic/anticyclonic regimes In contrast to the changes related to the NAO/AO that are part of a large-scale pattern, the cyclonic anticyclonic regimes of the oceanic circulation described by Proshutinsky and Johnson (1997) are clearly focused on the high latitudes. According to their analysis, the wind-driven oceanic motion in the Central Arctic alternates every 5 to 7 yr between anticyclonic and cyclonic circulation, resulting in an oscillatory behavior that has been called the Arctic Ocean Oscillation (AOO; Proshutinsky et al. 1999). Proshutinsky and Johnson (1997) proposed an index for the AOO based on the gradient of sea surface height in the central Arctic simulated by a barotropic model of the Arctic Ocean. During the cyclonic regime of the AOO, in addition to the reduction of atmospheric pressure and stronger winds that have an impact on sea ice and ocean circulation, the surface air temperatures in the Arctic tend to be higher and the ice thinner. This regime is also characterized by higher freshwater and ice transport through Fram Strait into the Greenland Sea (Proshutinsky et al. 1999; Polyakov et al. 1999). Determining the links and differences between the AAO and NAO/AO is not straightforward as the NAO/AO is also related to changes in the pressure and wind driven oceanic and sea ice circulation in the Arctic (e.g., Kwok 2000; Rigor et al. 2002). Furthermore, the years 1987 and 1992 that were chosen by Polyakov et al. (1999) as typical anticyclonic and cyclonic regimes, respectively, are also characterized by large variations of the NAO index (Fig. 2). The response of the Arctic Ocean to the NAO/AO could thus be partly represented as a change in AOO index, but this does not seem to be robust over time (Johnson et al. 1999). FIG. 2. Time series of the NAO (solid) and the AOO (dashed) indexes. The AOO index is based on EOF analysis of sea level variation in a 2D barotropic ice ocean model driven by winds derived from (National Centers for Environmental Prediction) NCEP NCAR reanalysis. Max depth in the model for this experiment is 200 m in order to concentrate wind energy in the upperocean layer and to obtain sea level gradients that indicate regime of circulation. Positive values correspond to a cyclonic regime and negative values correspond to an anticyclonic regime (A. Proshutinsky 2004, personal communication). When analyzing the time series of both the NAO and AOO indices, the agreement is indeed good during some periods such as the 1990s but weak during other periods (Fig. 2). In any case, although they are likely related, the NAO/AO and AOO should be considered as different modes of variability since the first one is based on large-scale atmospheric dynamics while the second one is based on changes in oceanic circulation in a relatively restricted area. Recently, Proshutinsky et al. (2002) proposed that freshwater and ice accumulate in the Beaufort Gyre during the anticyclonic regime. During the cyclonic regime, this freshwater is released and transported to the North Atlantic where it could have an influence on vertical mixing and deep water formation. When testing this hypothesis using an ice-ocean model, Häkkinen and Proshutinsky (2004) found relatively small changes in freshwater accumulation in the Beaufort Gyre in response to these atmospheric variations. Instead, in their model, the variations in the exchanges between the Arctic and the North Atlantic explain almost all the simulated changes in the freshwater content in the Arctic basin. c. Influence of the Atlantic meridional overturning circulation The Atlantic meridional overturning circulation (AMOC) transports large amounts of heat northward in the Atlantic until the Norwegian Sea. This heat is partly released at high latitudes, increasing air tempera-

5 3556 J O U R N A L O F C L I M A T E VOLUME 18 ture in these regions. Variations of the intensity of the AMOC could thus have a large impact on Arctic climate. However, observations are too scarce to precisely determine the impact of potential changes in the AMOC on the high-latitude climate. As a consequence, the majority of hypotheses on the influence of the AMOC on Arctic climate are based on model results. Using a coupled atmosphere ocean general circulation model, Delworth et al. (1993) found a multidecadal mode of variability of the oceanic thermohaline circulation described by the following mechanism. During a reduced thermohaline circulation, lower oceanic heat transport results in a cold anomaly throughout the top 1 km in the North Atlantic. This cold and dense anomaly enhances the horizontal circulation that transports more salt to the regions of deep-water formation. This leads to an increase in surface density there and a strengthening of the oceanic thermohaline circulation, closing half a cycle in about 30 yr. This mode of variability appears associated with fluctuations in the intensity of the East Greenland current and with large-scale changes in near-surface salinity in the Arctic with a potential impact on the Arctic climate (Delworth et al. 1997). Delworth and Mann (2000) have argued that the multidecadal variability observed by various authors in the North Atlantic (e.g., Mann and Park 1994; Tourre et al. 1999) could be related to this mechanism of variability of the thermohaline circulation. This mode of variability could also have an influence on the Arctic climate system (e.g., Polyakov and Johnson 2000; Polyakov et al. 2002), but strong observational evidence is still lacking because of the short time series available. The role of Arctic processes in the variability of the AMOC has been underlined in other studies (e.g., Wohlleben and Weaver 1995; Yang and Neelin 1997; Holland et al. 2001). For instance, in the simulations of Holland et al. (2001), fluctuations of ice export from the Arctic to the North Atlantic excite a damped oscillatory mode of the AMOC that induces large changes in ice coverage in the northern North Atlantic and variations in the inflow of warm water under the ice pack. Delworth et al. (1993, 1997) as well as Holland et al. (2001) suggest that the mode of variability obtained in their simulations can be considered primarily as an oceanic feature, coupling with the atmosphere being of secondary importance. On the other hand, Weaver and Valcke (1998) suggest that the mode analyzed by Delworth et al. (1993) is a mode of the fully coupled system. Timmermann et al. (1998) also present the mode of thermohaline circulation variability in their model as a coupled atmosphere ocean one. Furthermore, Wohlleben and Weaver (1995) on the basis of observations, made the hypothesis that the response of the atmosphere to changes in surface conditions in the Labrador Sea, caused by variations of deep-water formation there, could play a large role in decadal variability in the Arctic. According to their hypothesis, this atmospheric response could have an influence on ice and freshwater transport trough Fram Strait, finally influencing convection in the Labrador Sea. These few examples show that work is still needed to clarify the exact role of atmosphere ocean interactions on the variability of the AMOC and in turn to quantify the impact of this variability on the Arctic climate system. d. Role of exchanges between the Arctic and the North Atlantic The processes influencing Arctic climate variability presented above all include, more or less explicitly, modifications in the meridional transport of heat between the Arctic and the North Atlantic. Depending on the process, these modifications take place in the ocean, the atmosphere, or in both media. Nevertheless, the main focus of these processes is elsewhere, that is, the intensity of the midlatitude westerlies for the NAO, the oceanic circulation in the Arctic for the AOO, or the ocean circulation in the Atlantic for the AMOC. A different way of looking at these processes is to consider that the Arctic North Atlantic exchanges are crucial for Arctic climate variability. Modifications of these exchanges could be associated with large-scale features during some periods, as appears to be the case for the NAO over the last 20 yr. However, these associations may not be robust over long time scales, while the association between North Atlantic Arctic exchange and Arctic climate is. In particular, it has long been recognized that a large amount of heat is transported northward in the Barents Sea because of the inflow of warm Atlantic waters, with a significant influence on sea ice and oceanic conditions there [for recent observations in this region, see, e.g., Furevik (2001), Schauer et al. (2002), and Ingvaldsen et al. (2004a,b)]. Atmospheric conditions can influence this oceanic heat transport. This includes changes in wind stress that modify ocean circulation and the northward mass transport of the Atlantic water; changes in the heat flux that generate temperature anomalies, which are then advected northward by the mean flow; or a combination of these two effects (e.g., Adlandsvik and Loeng 1991; Furevik 2001; Orvik and Skagseth 2003; Ingvaldsen et al. 2004a,b). Analyzing long simulations and sensitivity experiments performed with a coarse-resolution atmosphere ocean sea ice model, Goosse et al. (2002) have argued that the northward heat transport, by the ocean in the Nordic Seas and above them by the atmosphere, have a

6 1 SEPTEMBER 2005 G O O S S E A N D H O L L A N D 3557 dominant influence on Arctic ice volume changes. Indeed, an increase in the sea ice volume is driven by a regional atmospheric pattern characterized by a positive surface pressure anomaly over the Norwegian Barents Kara central Arctic area and a negative one over the Greenland ice sheet. This pattern induces an atmospheric cooling in the Arctic since it implies a smaller northward inflow of warm air from the Atlantic sector. In this coupled simulation, the increase in ice volume is also associated with a reduction of the intensity of oceanic mixing in the Nordic Seas and with a smaller inflow of Atlantic waters in the Barents Sea. In addition to this large role of anomalous transport from the Atlantic to the Arctic, Goosse et al. (2002) proposed that the increase of sea ice volume in the Arctic induces an increase in salinity there. This salinity anomaly is transported to the Greenland Sea where it promotes convective activity. This warms up the surface oceanic layer and the atmosphere in winter and induces a decrease of the ice volume, providing a negative feedback that results in a mode of variability with a preferred time scale of about 18 yr. On the basis of data covering the whole twentieth century and of numerical model results, Semenov and Bengtsson (2003) and Bengtsson et al. (2004) suggest that the warming during the last 30 yr in the Arctic is linked with the NAO. However, the temperature variations averaged over the Arctic prior to the 1970s were not directly linked to any standard large-scale atmospheric circulation pattern. In particular, the large Arctic warming during the 1930s 40s appears to be related to an enhanced inflow of warm water from the Atlantic to the Barents Sea, inducing in that area a large sea ice retreat and a surface temperature increase. The anomalous water inflow and temperature changes in the Arctic are related to a decrease of sea level pressure above the Barents and Kara Seas, implying modifications of the strength of the westerlies into the Barents Sea. In addition to the importance of the air and water exchanges between the North Atlantic and the Arctic underlined in both studies, Goosse et al. (2002) and Bengtsson et al. (2004) both suggest that the response of the atmosphere to anomalous surface conditions (i.e., sea surface temperature and ice cover) plays a role in the variability of the Arctic climate. In Goosse et al. (2002), this atmospheric response is mainly thermodynamic. The retreat of the ice in the Barents Sea induces a warming of the air there, and this positive temperature anomaly is then transported by the mean atmospheric circulation to other regions of the Arctic. Changes in atmospheric circulation appear to be less important, except in the case of very large anomalies in which local modifications of the atmospheric circulation over the Barents Sea provide a positive feedback to the temperature anomaly (Goosse et al. 2003). Bengtsson et al. (2004) also suggest such a positive feedback, in which the atmospheric response to the anomalous heat flux in the Barents Sea reinforces the initial temperature anomaly related to the anomalous oceanic transport. It should be underlined that the studies mentioned in section 2b and 2c (Wohlleben and Weaver 1995; Mysak and Venegas 1998; Ikeda et al. 2001; Dukhovskoy et al. 2004) require a large-scale atmospheric response to changes in the Arctic surface conditions. In contrast, the atmospheric response proposed by Goosse et al. (2002) and Bengtsson et al. (2004) is a much weaker local one, restricted to an area close to the largest surface anomalies. The Barents Oscillation (BO), identified by Skeie (2000) as the second EOF of monthly winter sea level pressure (the first one is the AO; see above), has its stronger center of action in the Barents Sea and is related to the variability of meridional atmospheric flow in the Nordic Seas as well as to surface heat losses in this area. Thus the BO bears interesting similarities with the atmospheric modes described by Goosse et al. (2002) and Bengtsson et al. (2004). It has also been suggested that the BO and AO cannot be considered independently and that the BO could be related to the apparent nonstationarity of the atmospheric pattern associated with the AO (Tremblay 2001). Unfortunately, a thorough examination of the BO and its implications for the Arctic climate has not yet been performed. e. Other processes influencing Arctic climate While it is not feasible to present a complete inventory of possible mechanisms driving Arctic climate variability, a number of other notable mechanisms have been proposed to explain decadal variability at high latitudes in the Northern Hemisphere. A few of these are discussed here. Venegas and Mysak (2000) have performed a systematic analysis of the coupled changes in atmospheric circulation and sea ice concentration in the North Atlantic sector. They found enhanced variability on time scales of 6 7, 9 10, 16 20, and yr. Of relevance to this study, they suggested, in agreement with the arguments presented in section 2d, that the large variability in the Barents Sea ice cover is influenced by the inflow of warm North Atlantic waters. These variations in the inflow are related to changes in both the strength of the current (on a 9 10-yr time scale) and the temperature of the current (on a yr time scale). Using a different method (i.e., a zonal Fourier analysis of monthly averaged sea level pressure), Cavalieri

7 3558 J O U R N A L O F C L I M A T E VOLUME 18 and Häkkinen (2001) suggest that the changes in the phases of the planetary waves are the main driver of the low-frequency variability of the ocean and sea ice in the Arctic. Shifts in these waves would also be associated with a modification of the meridional atmospheric flow in the Nordic seas, presenting some similarities with the processes invoked in section 2d. Another type of intriguing variability affecting the high northern latitudes is the occurrence of Great Salinity Anomalies (GSAs) in the Nordic Seas and the North Atlantic (e.g., Dickson et al. 1988; Belkin et al. 1998; Mysak 1999; Belkin 2004). The most famous GSA (also called the GSA 70s; see Dickson et al for a review) started in 1968 in the Greenland Sea, probably because of a large inflow of freshwater and sea ice from the Arctic (Aagaard and Carmack 1989; Häkkinen 1993). This salinity anomaly propagated around the subpolar gyre and returned to the Greenland Sea, significantly attenuated, in 1981/82. The negative salinity anomaly induced a decrease in oceanic convection, both in the Greenland Sea and in the Labrador Sea and was associated with an increase in winter ice concentration (e.g., Mysak and Manak 1989). Other GSAs have propagated in the subpolar gyre in the 1980s and 1990s (e.g., Belkin et al. 1998; Mysak 1999; Belkin 2004), but they might have a different origin. Indeed, the GSA 70s appears to have its source in the Greenland Sea while other GSA 80s and GSA 90s seem to originate in the Baffin Bay/Labrador Sea area. Furthermore, the GSA 70s is not related to the NAO while the one of the 1990s may have been caused by a freshening associated with the high NAO values at that time (Belkin et al. 1998; Dickson et al. 2000; Belkin 2004). As a consequence, it is difficult to state that all the GSAs are driven by the same processes. 3. Model description The studies discussed in section 2 used a variety of observations and model results to elucidate mechanisms driving Arctic variability. This diversity of data and method makes it difficult to identify the relative importance and links among the various mechanisms. Additionally, these studies focus on a time period during which anthropogenic forcing of the system cannot be neglected. This contributes to the difficulty of examining mechanisms that naturally force the Arctic system. To address these issues, we examine a long selfconsistent dataset obtained from a control simulation of a coupled general circulation model, namely the CCSM2 (Kiehl and Gent 2004). CCSM2 is a state-ofthe-art CGCM that includes atmosphere, ocean, land, and sea ice components. No flux corrections are used in this model. The integration used here is run under present-day conditions with no changes in anthropogenic forcing. Six hundred and fifty years of model integration are analyzed (years ). This time period was chosen because many of the initial climate drifts in the ice and ocean are considerably reduced by year 350, and a small change was made at year 350 in the integration allowing for consistent constants across different component models. The community land model (Bonan et al. 2002) is the land surface component used within the CCSM2. The model includes a subgrid mosaic of land cover types and plant functional types derived from satellite observations, a 10-layer soil model that explicitly treats liquid water and ice, a multilayer snowpack model, and a river routing scheme. The ocean component of the CCSM2 uses the parallel ocean program (POP) with a number of improvements (Smith and Gent 2002). The horizontal resolution averages less than one degree, and there are 40 vertical levels. The model grid smoothly displaces the North Pole into Greenland, avoiding the problem of converging meridians within the Arctic Ocean. The community sea ice model incorporated into CCSM2 is a new dynamic thermodynamic scheme that includes a subgrid-scale ice thickness distribution (Bitz et al. 2001; Lipscomb 2001). The model uses the energy conserving thermodynamics of Bitz and Lipscomb (1999), which has multiple vertical layers and accounts for the thermodynamic influences of brine pockets within the ice cover. The ice dynamics uses the elastic viscous plastic rheology of Hunke and Dukowicz (1997) with a number of updates (Hunke 2001; Hunke and Dukowicz 2002). The ice model uses the same horizontal grid as the ocean. The atmospheric component of the CCSM2 is the community atmosphere model (CAM2). It builds on the National Center for Atmospheric Research (NCAR) atmospheric general circulation model, Community Climate Model, version 3 (CCM3; Kiehl et al. 1996), with a number of improvements and updates. The model is run at T42 ( ) resolution with 26 vertical levels. The Arctic climate and its variability does appear to be relatively well simulated in this model integration (Holland 2003; Briegleb et al. 2004), leading us to believe it can provide useful information on the mechanisms driving high-latitude variability. For example, the annual average Arctic ice concentration shown in Fig. 3 is well simulated, although somewhat too extensive in the Labrador Sea. Nevertheless, as with all climate models, the CCSM2 shows biases in its climate com-

8 1 SEPTEMBER 2005 G O O S S E A N D H O L L A N D 3559 most representative of general Arctic climate conditions. The time series and spectrum of the annual mean N SAT is presented in Fig. 4. It has a standard deviation of 0.5 C, which is comparable to the observations compiled by Polyakov et al. (2002), which show a standard deviation of 0.6 C over the last 125 yr. The simulated time series exhibits a red spectrum, with enhanced power at low frequencies. As expected, the changes in simulated N SAT are highly correlated with the temperature evolution in the Central Arctic with correlation higher than 0.7 close to the North Pole (Fig. 5a). The correlation decreases with decreasing latitudes and becomes slightly negative in the Labrador Sea, reaching values of the order of 0.1. In contrast to the nearly zonally symmetric spatial FIG. 3. Annual mean ice concentration simulated in CCSM2 averaged over the last 650 yr of the simulation. Contour interval is 0.1. The observed position of the ice edge defined as the 15% concentration in annual mean is also shown as a thick solid line. Latitude circles at 50, 60, and 70 N and longitudes of 0, 90 E, and 180, 270 E are represented as dashed lines. pared to observations. As discussed by Briegleb et al. (2004), the simulated Arctic sea ice is relatively thin. Furthermore, because of its resolution, the model is not able to precisely reproduce small-scale currents like the coastal and boundary currents in the Arctic. These limitations must be kept in mind when discussing the model results. More information about the polar climate produced in this model simulation is given by Briegleb et al. (2004). 4. Arctic variability in the CCSM2 a. Surface air temperature as a proxy for Arctic climate As discussed in section 3, we examine here 650 yr of model output. We are interested in the evolution of the general Arctic climate state. However, it is not possible to analyze the relation between all the high-latitude ocean atmosphere sea ice state variables, which together characterize the Arctic climate. As a consequence, we focus our attention to the annual mean, N averaged surface air temperature (SAT), which represents a reasonable proxy of the integrated Arctic climate in CCSM2. A number of other candidate Arctic climate indices (e.g., Arctic ice volume) were examined, and it was determined that the SAT was the FIG. 4. The time series of (top) the annual mean surface air temperature averaged over the region N with a 10-yr running mean (bold) and (bottom) the spectral analysis of the N SAT. The smooth solid line represents the theoretical red spectrum, and the dashed line is the 95% significance level.

9 3560 J O U R N A L O F C L I M A T E VOLUME 18 FIG. 5. (top) Correlation and (bottom) regression between annual mean SAT and the time series of the SAT averaged over N. The contour interval is (top) 0.1 and (bottom) 0.25 K. Negative values are dashed. correlation, the regression map (Fig. 5b) has a clear maximum in the Barents/Kara Sea, the region that exhibits the largest variance in SAT. In this area, changes per standard deviation of SAT averaged over the area northward of 70 N can be higher than 2 C, that is, more than 2 times the response in the central Arctic. Sea ice conditions are also related to the annual mean N SAT as shown in Fig. 6. When the averaged Arctic air temperature is high, the annual average ice concentration is low everywhere. In winter, changes in ice concentration occur close to the ice edge while a modification over a wider area is noticed in summer as reflected on the annual mean pattern. During both seasons, the response appears large in the eastern sector of the Arctic where the maximum in the regression is found. These changes in the ice cover in the Barents Sea imply a local amplification of the SAT anomalies because of classical feedbacks such as the FIG. 6. (top) Correlation and (bottom) regression between annual mean ice concentration and the time series of N SAT. Data were low-pass filtered with a 5-yr cutoff before performing the analysis. The contour interval is (top) 0.1 and (bottom) per standard deviation of the SAT. one between the temperature and the albedo. Similar conclusions are drawn from the analysis of ice thickness (not shown). When the averaged Arctic temperature is high, the sea ice is thin everywhere in the Arctic, with the largest signal in the Barents Kara Seas sector. b. Mechanisms influencing simulated Arctic climate variability 1) ATMOSPHERIC PROCESSES Figure 7 shows the correlation between sea level pressure (SLP) and N SAT. Before performing this analysis, the time series of N SAT and SLP were low-pass filtered. This results in slightly higher correlations but does not qualitatively change the results (which is also the case for ice concentration correlations shown in Fig. 6). Because of the long length of

10 1 SEPTEMBER 2005 G O O S S E A N D H O L L A N D 3561 FIG. 7. Same as in Fig. 6, but for annual mean SLP. Contour interval is (top) 0.1 and (bottom) 0.2 hpa. the unfiltered time series, correlations as low as 0.1 are significant at the 95% level, taking into account the autocorrelation of the time series. For the filtered results, correlations higher than 0.22 in absolute value are significant at the 95% level. The correlation between SLP and N SAT reaches its maximum at zero lag (Fig. 7). This shows that, on the annual average, the interaction between the two variables can be considered as immediate with no process providing a clear lag between them. Although the correlation is quite low, the pattern of SLP associated with the N SAT clearly displays a low pressure centered over Spitzbergen, Norway, that covers a large fraction of the Arctic basin. The correlation and regression analysis between N SAT and SLP was also performed for the winter period only [December January February (DJF)]. The same dipole as for the annual mean time series was obtained with correlation of the order of 0.2 in the Nordic Seas for both the filtered and unfiltered data. However, the regression for winter was higher than the one displayed in Fig. 7b by roughly a factor 2, the standard deviation of SLP being much higher in winter. The SLP correlations shown in Fig. 7 are associated with northerly winds along the Greenland coast and southerly winds in the Barents Sea. This induces an increased inflow of warm air to the Barents Sea/Kara Sea region. This result is consistent with the correlation of the N SAT and atmospheric heat transport (not shown) that displays its highest value (larger than 0.2) northward of 60 N in the area between 10 and 40 E, that is, roughly between the longitudes of Spitzbergen and Novaya Zemlya, Russia. In addition to the changes in the meridional heat transport by the atmosphere, the anomalous atmospheric circulation influences oceanic circulation in the Nordic and Barents Seas, as discussed further in section 4b(4). This oceanic influence is also suggested from the correlation between the turbulent heat flux at the ocean/ice interface and N SAT (not shown). In the Nordic and Barents Seas, the correlation is positive with values up to 0.40, meaning that the ocean is warming the atmosphere. The northward advection of this warm atmospheric anomaly, together with the anomalous atmospheric circulation, result in the positive anomaly of the northward heat transport by the atmosphere discussed above. This atmospheric heat is then released in the Kara, Laptev, and East Siberian Seas, warming the surface there and resulting in a significant negative correlation of the surface turbulent heat flux and the N SAT in these regions. The changes due to the anomalous meridional heat transport are also amplified locally in the Arctic since the reduction of the ice cover associated with the warming induces a lower surface albedo in summer, leading to the well-known temperature albedo feedback. Indeed, the correlation between the net downward shortwave flux and N SAT is positive in the whole Arctic and reaches its maximum value in the Barents Sea with a value higher than 0.5 (not shown). The net longwave flux also induces a positive feedback in the Barents Sea, while the correlation is negative over the majority of the central Arctic. The warmer SATs are associated with a generally warmer and wetter atmospheric column, higher surface longwave emission, and increased surface downwelling longwave radiation. Over the central Arctic, changes in water vapor content are modest, and the higher surface longwave emission dominates in the net surface longwave response. In the Barents Sea region, larger water vapor content anomalies are present, and higher downwelling longwave radiation dominates, reinforcing the initial SAT anomaly.

11 3562 J O U R N A L O F C L I M A T E VOLUME 18 FIG. 8. Running correlation between the NAO index in the model and the time series of the N SAT. A 50- (25) yr time window has been used for the black (dashed) curves. The NAO index series is defined as the time series of the principal component of the first EOF (EOF1) of SLP (DJF) in the Arctic North Atlantic region. 2) LINKS WITH THE NAO In addition to the decrease centered over Spitzbergen, the regression of SLP on the N SAT time series displays a weak dipole between high and midlatitudes in the Atlantic sector. Furthermore, a weak dipole is also present in the temperature response (Fig. 5) and the sea ice response (Fig. 6) between the Labrador Sea and the Barents Sea. The correlations in the Labrador Sea for temperature and ice concentration and for SLP at midlatitudes are low. Nevertheless, as these features are reminiscent of conditions associated with the NAO/AO, it seems appropriate at this stage to analyze the link between NAO/AO and Arctic SAT. The correlation between the simulated NAO and the N SAT computed over the whole time series is weak. Nevertheless, during some periods, it can reach values of 0.5, while negative values are found during other periods of the experiments (Fig. 8). An analysis of the patterns associated with the NAO during periods when this correlation is high and during periods when this correlation is low has been performed. This shows, as expected, that during the whole time series the basic characteristics of the NAO are present. For example, the presence of an SLP dipole between the mid- and high latitudes and large changes in the magnitude of the westerly winds at midlatitudes are always associated with the NAO. However, during periods when the correlation between the NAO and Arctic SAT is low, the trough in the northern North Atlantic is slightly displaced. This results in anomalous winds from the northeast in the Barents Sea that strongly reduce the heat transport from the Atlantic sector to the Arctic (Fig. 9). It appears that the nonstationary link between the NAO and Arctic climate deduced from the observations during the twentieth century (section 2a) is a common feature in the model results. Furthermore, such a shift in the link is not necessarily caused by external forcing since it occurs in the CCSM2 control run with constant forcing. FIG. 9. The regression between winter mean SLP and the NAO index for years (corresponding to a high correlation between the NAO and Arctic SAT) minus the regression between the NAO and winter SLP for years (corresponding to a low correlation between the NAO and Arctic SAT).

12 1 SEPTEMBER 2005 G O O S S E A N D H O L L A N D ) OCEANIC PROCESSES As compared to atmospheric variables, a stronger relationship is present between Arctic SAT and oceanic circulation anomalies (Fig. 10). High temperatures in the Arctic are associated with a stronger northward transport of Atlantic waters off the Norwegian coast and a stronger inflow into the Barents and Kara Seas. A stronger southward export is also noticed at Fram Strait, resulting in a stronger east Greenland current. The majority of these changes occur in the Nordic Seas and in the Atlantic sector of the Arctic. More cyclonic conditions in the Canadian basin are simulated during the periods when the temperature is high in the Arctic with some similarities to the AOO. However, the correlations are much lower there than in the Nordic and Barents Seas. This is confirmed by the correlation between N SAT and sea surface height in the model that displays a maximum value of 0.14 in the Beaufort Gyre at zero lag (not shown). The correlation between the northward advective heat transport in the Barents Sea and SAT in the Arctic nearly reaches 0.6 when a 10-yr running mean is applied to the time series (Fig. 11). When analyzing the advective heat transport for each model layer, the maximum correlation between N SAT and oceanic heat transport in the Barents Sea is found under the surface, with a peak correlation at a depth at about 25 m (not shown). Figure 11 is consistent with the large role of the ocean suggested by Fig. 10, and with the large response FIG. 10. Regression between annual mean surface ocean currents in CCSM2 (depth range: 0 10 m) and the time series of the SAT averaged over N. The arrow on the bottom left corner gives the scaling of the vectors (in cm s 1 ). FIG. 11. The correlation between simulated Barents Sea ocean heat transport into the Arctic N SAT as a function of lag. The solid line displays the correlation for the raw times series, and the dashed line displays the correlation for the time series that have been smoothed with a 10-yr running mean. The Barents Sea oceanic heat transport is computed as the integral of the oceanic velocity times the ocean temperature multiplied by density and heat capacity of seawater, using a reference temperature of 0 C. of the ice concentration and SAT in the Barents Sea. In addition, the correlation is maximum at zero lag, showing that the influence of the Barents Sea inflow is not due to a slow propagation of the Atlantic waters at depth followed by a slow upwelling. The heat transported by the currents is released locally in the Barents Sea and then influences a large fraction of the Arctic by anomalous heat transport in the atmosphere as discussed in section 4b(1). The correlation between SAT in the Arctic and heat transport in the Barents Sea is about 35% larger than that for the mass transport in the Barents Sea. This indicates that modifications in both the currents and the oceanic temperatures within the Barents Sea play a significant role in Arctic SAT changes. The influence of the temperature anomalies is further illustrated by the correlation between the temperature at 50 m and heat transport in the Barents Sea for various lags (Fig. 12). This suggests the propagation of a positive temperature anomaly from the passage between Iceland and Norway to the Barents Sea and into the Arctic in about 2 yr. This temperature anomaly contributes to the anomalous heat transport. The variations of oceanic heat transport in the Barents Sea are mainly related to changes in ocean circulation within the upper 500 m of the Nordic and Arctic Seas. They do not appear to be strongly linked to the large-scale AMOC (Fig. 13). Indeed, the correlation between the Barents Sea heat transport and the maximum of the overturning streamfunction in the North Atlantic between 30 and 60 N is always smaller than

13 3564 J O U R N A L O F C L I M A T E VOLUME 18 FIG. 13. Correlation of the meridional overturing streamfunction in the North Atlantic and in the Arctic with the Barents Sea ocean heat transport at zero lag. The contour interval is 0.1. Density changes in the surface layer could also play a role in driving ocean transport in the Barents Sea, in particular changes in the salinity distribution in the North Atlantic and Arctic. To illustrate this point, the correlation between oceanic heat transport in the Barents Sea and the sea surface salinity has been performed for different lags (Fig. 15). This figure suggests a propagation of salinity anomalies in the North Atlantic for lags 4 and 2. However, the connection between those anomalies and the Barents Sea (lag 0) is not FIG. 12. Correlation of the ocean temperature at 50 m with the Barents Sea ocean heat transport at lags 2, 1, and 0 (heat transport is lagging the ocean temperature). The contour interval is 0.1. Negative values are dashed for all lags between 20 and 20 yr. Furthermore, the correlation between the maximum of the overturning streamfunction in the North Atlantic and the SAT is quite low in CCSM2 (Fig. 14). FIG. 14. Correlation between the N SAT time series and the time series of the max of the overturning streamfunction in the Atlantic between 30 and 60 N in the depth range m. The solid line displays the correlation for the raw times series, while the dashed line displays the correlation for time series that have been smoothed with a 10-yr running mean. A positive lag corresponds to the SAT lagging the overturning streamfunction.

14 1 SEPTEMBER 2005 G O O S S E A N D H O L L A N D 3565 FIG. 15. Correlation between the annual mean surface salinity and the time series of the Barents Sea ocean heat transport at lags 4, 2, 0, and 2 (with the ocean heat transport lagging the salinity). The contour interval is 0.1. straightforward compared to the propagation discussed above for temperature anomalies. In the Arctic, the salinity correlations appear quite weak for lags 4 and 2. At zero lag, a clear maximum is seen in the Barents/ Kara Sea sector because of a stronger advection of salty Atlantic waters. In the central Arctic, the salinity tends to decrease at zero lag because of warming of the Arctic and melting of the ice. These salinity anomalies in the Arctic vanish quite quickly with no clear sign of propagation at lag 2, in contrast to the results of Goosse et al. (2002). A qualitatively similar picture is obtained when performing the correlation between N SAT and surface salinity. As a consequence, the changes in salinity in the Arctic appear to be due to the modification of atmospheric and oceanic conditions rather than a driver of those changes. 4) LINKS BETWEEN OCEANIC AND ATMOSPHERIC PROCESSES The above analysis suggests that ocean circulation changes are important for modifying the heat transported into the Arctic through the Barents Sea. The atmospheric forcing associated with these changes in heat transport (Fig. 16) exhibits a pattern that is remarkably similar to the one associated with the changes in SAT in the Arctic (Fig. 7), but with considerably higher correlation. This suggests that this pattern could have an influence on the SAT in the Arctic directly through anomalous atmospheric heat transport as well as through its impact on the oceanic circulation, with the latter connection having larger correlations. It is not possible to determine from our analysis if the oceanic heat transport in the Barents Sea and the subsequent warming in that area are able to have a significant influence on the SLP around Spitzbergen. This could lead to a positive feedback that would amplify the anomalous heat transport. Nevertheless, if such a feedback occurs in the model, it must be local. Any largescale response involving modes of variability like the NAO would certainly have a modest amplitude, as the correlation displayed in Fig. 16 reaches values only on the order of 0.1 at midlatitudes. No sensitivity experiment was performed here to clearly prove this point, but this is in good agreement with the studies of Magnusdottir et al. (2004) and Alexander et al. (2004), who analyzed the impact of changes in surface conditions (sea ice extent and SST) on the atmospheric circulation

15 3566 J O U R N A L O F C L I M A T E VOLUME 18 FIG. 16. (top) Correlation and (bottom) regression between the annual mean SLP and the time series of the Barents Sea ocean heat transport. The contour interval is (top) 0.1 and (bottom) 0.2 hpa. using the same atmospheric model as the one used in the present study. 5. Summary and conclusions A number of mechanisms, supported by both observational and modeling studies, have been proposed to drive decadal/interdecadal variability at high northern latitudes. These include the influence of changes in the oceanic and atmospheric North Atlantic Arctic exchange, changes in AMOC, the NAO/AO, and anticyclonic/cyclonic regimes (the AOO). Many of these mechanisms are likely interrelated. However, the links between the different processes and their relative importance are not easy to assess on the basis of the original studies, which use a variety of observational and modeling evidence over different time periods. Additionally, since observational studies are limited by relatively short time series during which anthropogenic forcing is likely important, it is difficult to assess the robustness of different mechanisms for the natural variability of the system over long time scales. In an effort to address the relative role of these different mechanisms in driving Arctic climate variability, a long control (present day) integration of the Community Climate System Model, version 2 (CCSM2) has been examined. This model simulation represents Arctic climate conditions reasonably well and has the advantage of providing a long self-consistent dataset for analysis. Surface air temperature integrated from 70 to 90 N has been used as a proxy for the large-scale Arctic climate conditions. Our analysis of CCSM2 results are in agreement with earlier findings that the meridional exchanges between the Arctic and the North Atlantic sector play a dominant role in the variability of the Arctic climate. This occurs because of the anomalous heat transport by both the atmosphere and the ocean. More specifically, variability in both the temperature and velocity of the inflow of Atlantic waters in the Barents Sea appears to drive changes in the Arctic SAT. These changes in the oceanic heat transport are related to the atmospheric circulation that has thus both a direct effect of heat transport trough the northward advection of warm air and an indirect one because of its influence on the ocean. Furthermore, the changes due to the anomalous meridional heat transport are amplified locally in the Arctic since a reduction of the ice cover induces larger turbulent heat release to the atmosphere in winter and a lower surface albedo in summer. In the CCSM2, the variability of the AMOC plays only a modest role in the changes in SAT in the Arctic. The influence of the NAO is more ambiguous since the correlation between the NAO and SAT is high for some periods and low during others. In addition, during years with high Arctic SAT, both atmospheric and oceanic circulations are more cyclonic in the Arctic, in qualitative agreement with the description of the processes associated with the AOO. However, the correlation between air temperature at high latitudes and ocean currents, sea surface height, or salinity in the central Arctic are low, advocating against an important role of the central Arctic Oceanin the SAT variability in CCSM2. On the other hand, propagation of anomalies in the subpolar gyre might play a role in the variability of the oceanic heat transport to the Arctic, and thus on Arctic climate, but this deserves further investigation. Our results strongly emphasize the influence of the Nordic Seas. Focusing interest on a different area, such as the central Arctic (for the AAO) or a large fraction