Arctic Sea Ice and Freshwater Changes Driven by the Atmospheric Leading Mode in a Coupled Sea Ice Ocean Model

Transcription

1 2159 Arctic Sea Ice and Freshwater Changes Driven by the Atmospheric Leading Mode in a Coupled Sea Ice Ocean Model XIANGDONG ZHANG Frontier Research System for Global Change, International Arctic Research Center, University of Alaska, Fairbanks, Fairbanks, Alaska MOTO IKEDA Frontier Research System for Global Change, International Arctic Research Center, University of Alaska, Fairbanks, Fairbanks, Alaska, and Graduate School of Environmental Earth Science, Hokkaido University, Sapporo, Japan JOHN E. WALSH International Arctic Research Center, University of Alaska, Fairbanks, Fairbanks, Alaska (Manuscript received 1 August 2002, in final form 2 January 2003) ABSTRACT Observational and modeling studies have indicated recent large changes of sea ice and hydrographic properties in the Arctic Ocean. However, the observational database is sufficiently sparse that the mechanisms responsible for the recent changes are not fully understood. A coupled Arctic ocean sea ice model forced by output from the NCEP NCAR reanalysis is employed to investigate the role that the leading atmospheric mode has played in the recent changes of the Arctic Ocean. A modified Arctic Oscillation (AO) index is derived for the region poleward of 62.5 N in order to avoid ambiguities in the distinction between the conventional AO and the North Atlantic Oscillation index. The model results indicate that the AO is the driver of many of the changes manifested in the recent observations. The model shows reductions of Arctic sea ice area and volume by 3.2% and 8.8%, respectively, when the AO changes from its negative to its positive phase. Concurrently, freshwater storage decreases by about 2%, while the sea ice and freshwater exports via Fram Strait increase substantially. The changes of sea ice and freshwater storage are strikingly asymmetric between the east and the west Arctic. Notable new findings include 1) the interaction of the dynamic and thermodynamic responses in the sense that changes of sea ice growth and melt are driven by, and feed back negatively to, the dynamically (transport) driven changes of sea ice volume; and 2) the compatibility of the associated freshwater changes with recently observed changes in the salinity of the upper Arctic Ocean, thereby explaining the observed salinity variations by a mechanism that is distinct from, but complementary to, the altered circulation of Siberian river water. In addition, the enhanced freshwater export could be a contributing factor to the increased salinity in the Arctic Ocean. The results of the simulations indicate that Arctic sea ice and freshwater distributions change substantially if one phase of the AO predominates over a decadal timescale. However, such results are based on an idealization of the real-world situation, in which the pattern of forcing varies interannually and the number of positive-ao years varies among decades. 1. Introduction Recent observational studies have demonstrated that the Arctic climate is undergoing dramatic changes. The Arctic anticyclone has weakened and cyclonic activity has strengthened over the central Arctic Ocean since 1988 (Walsh et al. 1996). The large-scale atmospheric leading mode, the Arctic Oscillation (AO; Thompson and Wallace 1998) and the North Atlantic Oscillation (NAO; Hurrell 1995), have shown pronounced fluctu- Corresponding author address: Xiangdong Zhang, Frontier Research System for Global Change, International Arctic Research Center, University of Alaska, Fairbanks, Fairbanks, AK xdz@iarc.uaf.edu ation. The positive phase of AO or NAO has predominated since the late 1980s, and its large amplitude has been remarkable. Analyses of Scanning Multichannel Microwave Radiometer (SMMR) and Special Sensor Microwave Imager (SSM/I) data from 1978 to 1998 have revealed a reduction of about 3% decade 1 in sea ice area and extent. The largest regional decreases, about 10.5% decade 1, have occurred over the Kara and Barents Seas (Cavalieri et al. 1997; Johannessen et al. 1999; Parkinson et al. 1999). Correspondingly, Rothrock et al. (1999) reported a mean decrease of 1.3 m in sea ice thickness, about 42%, at the end of the melt season in the deepwater portion of the Arctic Ocean between and the 1990s. This finding was supported by independent 2003 American Meteorological Society

2 2160 JOURNAL OF CLIMATE VOLUME 16 British data (Wadhams and Davis 2000). Tucker et al. (2001) examined the recent spring sea ice draft from offshore Alaska to 89 N, finding a reduction of 1.5 m and attributing it to ice dynamics associated with atmospheric variability. However, the limited coverage of the submarine upward-looking sonar data leaves open the possibility that some of the decrease may be offset by the advective redistribution of sea ice to areas not sampled by submarines. Noteworthy changes have also been detected in the ocean. Measurements of potential temperature along the perimeter of the southern Canadian basin north of the East Siberian Sea in 1993 show that the Atlantic layer is about 1 C warmer than earlier climatologies (Carmack et al. 1995; Morison et al. 1998). Steele and Boyd (1998) compared hydrographic data of the 1990s with those from previous decades, suggesting a retreat of the cold halocline layer from the Amundsen to the Makarov basin, such that the Eurasian basin became warmer and saltier. Measurements during the Surface Heat Budget of the Arctic (SHEBA) experiment in 1997 indicated a freshening of the Beaufort Sea relative to 1975 (McPhee et al. 1998). In an attempt to diagnose the changes, Hilmer and Lemke (2000) used a stand-alone sea ice model to deduce a trend of sea ice volume at the rate of 4% decade 1, with the largest thinning in the eastern Arctic. Two coupled sea ice ocean modeling studies investigated variability in the recent 20 yr. Using a coupled Arctic sea ice ocean model and forcing data originating from the International Arctic Buoy Program (IABP), Zhang et al. (1998, 2000) obtained warming and salinification of the Arctic Ocean due to a strengthened Atlantic inflow, a sea ice thickness decrease of 25% in the eastern Arctic during (lower NAO), and a thickness increase of 16% in the western Arctic during (higher NAO). Zhang and Hunke (2001) employed a different Arctic coupled model forced by the European Centre for Medium-Range Weather Forecasts (ECMWF) data, capturing generally similar features of sea ice and ocean change. A 52-yr simulation suggested that the sea ice change noted above is a reoccurring pattern whereby sea ice shifts between the central Arctic and peripheral regions (Holloway and Sou 2002). While the occurrence of local and regional changes in the Arctic during recent decades is undeniable on the basis of the studies cited here, several aspects of these changes have not been fully explored. These aspects provide the foci for this study. First, the Arctic Oscillation is clearly a contributing factor in the changes of sea ice, but its role in the forcing of sea ice and ocean changes has not been isolated in model experiments. We have designed and performed model experiments to systematically assess the impacts of different phases of the Arctic Oscillation. Second, variations of wind stress, ice dynamics, and ice transport must in turn affect the thermodynamics of sea ice through changes of ice concentration and thickness. We evaluate and compare the dynamic and thermodynamic consequences, for sea ice, of variations in the Arctic Oscillation. Third, the dynamic and thermodynamic response of sea ice to the Arctic Oscillation must inevitably affect the freshwater (FW) budget of the Arctic Ocean. We evaluate the FW impacts and show that there are large regional dependencies of the FW consequences of the variable forcing of sea ice. These consequences are especially important when they extend to the FW exchanges between the Arctic and the subpolar seas, which are marginally stratified with respect to convective processes. Fourth, we document the seasonal variation of the impacts of the Arctic Oscillation on sea ice and Arctic Ocean hydrography. The strong seasonal cycle of sea ice and upper-ocean variables in the Arctic introduces considerable complexity into the response to forcing anomalies. Yet this seasonality is essential to an understanding of the Arctic Ocean s sensitivities to forcing variations. Finally, we base our evaluation of the Arctic Ocean s response to the Arctic Oscillation on a sample of years extending back to 1958, which is considerably earlier than the initial years of the study periods of most of the investigations cited earlier. 2. Model and forcing data The Arctic coupled ocean sea ice model used here was developed for climate studies by Zhang and Zhang (2001). We briefly describe the model here. More detailed information about model and data, as well as the model s performance, can be found in Zhang and Zhang (2001). The ocean component is based on Geophysical Fluid Dynamics Laboratory (GFDL) Modular Ocean Model version 2.0 (MOM2.0; Pacanowski 1995) and characterized by the flux-corrected-transport (FCT) algorithm (Gerdes et al. 1991). The FCT algorithm is used to eliminate numerical dispersion near large gradients, nonphysical oscillations, and negative concentrations, by providing the minimum mixing that is consistent with the thermodynamic constraint. It acts to constrain the pathway and properties of Atlantic water advection in the Arctic Ocean in our model. In view of the stable stratification of the Arctic Ocean, ocean horizontal and vertical viscosity coefficients are taken as and 8.0 cm 2 s 1, respectively. The horizontal diffusivity coefficient is set to be zero due to the FCT algorithm, in which implicit diffusion has been taken into account, and the vertical diffusivity coefficient is taken as 0.6 cm 2 s 1. The sea ice model uses the Hibler (1979) dynamics and Parkinson and Washington (1979) thermodynamics with modifications of snow treatments based on Oberhuber et al. (1993) and Fichefet and Maqueda (1997). Zhang and Zhang s (2001) synchronous coupling scheme between ocean and sea ice to help ensure heat and salt conservation is also used here. A simple parameterization of brine rejection and more sophisticated parameterizations of heat and freshwater fluxes are im-

3 2161 FIG. 1. Bathymetry of the model domain. The lines A J represent vertical sections separating the entire domain into four regions for analyses in this study: the Arctic Ocean, the Barents Sea, the Kara Sea, and the GIN Sea. The Arctic Ocean is further divided into the east and west Arctic by the dot dashed line. plemented. Atmospheric stability-dependent sensible and latent heat fluxes are applied. Downwelling shortwave and longwave radiation are calculated based on the parameterization schemes by Shine (1984), Efimova (1961), and Jacobs (1978), which were verified to have the best performance for the Arctic region among various parameterizations (Key et al. 1996). Penetration of shortwave radiation through bare sea ice is included (Grenfell and Maykut 1977). The FW fluxes include sea ice growth/melt, snowmelt, precipitation, and river runoff, which are conventionally treated as negative salt fluxes for this rigid-lid model. Relatively weak surface restoring of ocean properties with a time constant of 50 days is included as a correction flux term in addition to real physical fluxes to prevent model drift. The model domain extends from the Greenland Iceland Norwegian (GIN) Sea to the Bering Strait. The horizontal resolution is 55 km. There are 29 levels vertically, ranging from the ocean surface down to 4350 m at the bottom with thickness from 10 m near the surface to 290 m in the deep ocean. Open boundaries in the Bering Strait, the Baffin Bay, and the GIN Sea allow water exchanges between the Arctic Ocean and the North Pacific and the North Atlantic, respectively. We divide this domain into four parts to investigate changes of sea ice and ocean properties: the Arctic Ocean, the Kara Sea, the Barents Sea, and the GIN Sea. Further, the Arctic Ocean is separated into the east and west Arctic (Fig. 1). The forcing data, constructed from the National Centers for Environmental Prediction National Center for Atmospheric Research (NCEP NCAR) reanalysis from 1958 to 1998 (Kalnay et al. 1996), include surface wind stress, 2-m air temperature, 2-m specific humidity, and surface pressure. Climatological precipitation and river runoff are taken from climate observations [Legates and Willmott 1990; the National Snow and Ice Data Center (NSIDC); Becker 1995]. An inflow of 0.85 Sv (Sv 10 6 m 2 s 1 ) through the Bering Strait and an outflow of 1.7 Sv through the Canadian Archipelago are prescribed, based on estimates by Coachman and Aagaard (1988) and Fissel et al. (1988). For the water mass balance in this rigid-lid model, a net inflow of 0.85 Sv through the GIN Seas is specified.

4 2162 JOURNAL OF CLIMATE VOLUME 16 We should note that our model is different from the Zhang et al. (2000) and Zhang and Hunke (2001) models, in its inclusion of FCT for advection and its low diffusivity of temperature and salinity, for example. The data and parameterization we use are also different from those used in the previous modeling studies, in which the geostrophic wind and 2-m air temperature derived from the IABP for the period (Zhang et al. 2000) and ECMWF data from 1983 to 1997 (Zhang and Hunke 2001) were used. 3. The Arctic atmospheric leading mode: An alternative AO Thompson and Wallace (1998) performed an EOF diagnosis on the monthly mean SLP north of 20 N and obtained an anomaly mode, the AO, which has three centers of action: in the Greenland Sea Denmark Strait, the North Atlantic, and the North Pacific. The AO and the NAO are major sources of atmospheric variability in the Northern Hemisphere and are closely linked in the North Atlantic sector. However, Deser (2000) argues that the annular character of the AO is more a reflection of the dominance of its Arctic center of action than any coordinated behavior of the Atlantic and Pacific centers of action in the SLP field. While the NAO is known to be strongly coupled to regional fluctuations of sea ice (Mysak and Venegas 1998), the NAO index explains only about 25% of the perennial multiyear sea ice variability (Johannessen et al. 1999). In order to concentrate on the dominant modes of climate variability over the Arctic Ocean and avoid having to distinguish between the AO and NAO (Wallace 2000), we have performed an EOF analysis on the monthly mean SLP north of 62.5 N. This analysis used an equal-area weighting to avoid problems associated with the convergence of the meridian. The first EOF explains 51.4% of total variance. We identify this EOF and its corresponding principal component as EOF62 and PC62, respectively. The overall spatial pattern of EOF62 is similar to the AO pattern but the Arctic action center is shifted northward to the Amundsen basin (Fig. 2a). Figure 2b depicts the normalized PC62, the AO index, and the NAO index. All three indices capture major SLP variations, which are characterized by lower SLP over the Arctic Ocean since the late 1980s. However, there are clearly differences among them. While the PC62 and AO indices generally vary similarly, the NAO index differs greatly from them in some particular years. The correlations coefficients of the PC62 with the AO and the NAO indices are about 0.96 and 0.55, respectively. Hereafter in this study, we use the EOF62 as the AO mode, instead of the conventional AO of Thompson and Wallace (1998). Correspondingly, PC62 is used as the AO index to investigate Arctic ocean sea ice response to the atmospheric leading mode. The relationship of the NAO to the Arctic sea ice ocean changes was explored by Zhang et al. (2000) and Zhang and Hunke (2001). These previous studies were limited to the decades of the 1980s and the early to mid- 1990s, representing the high and low phases of NAO with other variations superimposed. However, it has been shown by Hilmer and Jung (2000) that the NAO and its association with sea ice are very different before and after In the following sections, we aim to isolate the changes of sea ice and ocean driven by the AO mode. We do so by a regression analysis of the forcing data onto the AO index to produce climate anomalies corresponding to the positive and negative phases of AO. We first create a monthly climatology of the variables for the entire period of Then monthly anomalies obtained by regression onto the AO index are added to and subtracted from the climatic monthly mean data, to produce the forcing data in the positive and negative phases of the AO. The climatic annual mean of 2-m air temperature and wind stress in the positive and negative phases of the AO, and their differences are shown in Fig. 3. A well defined anticyclonic wind stress cell is centered in the Canada Basin and covers most of the Arctic Ocean in the negative phase, reflecting the Beaufort high. The wind stress in the Makarov and Eurasian Basins breaks into two branches, one forcing the convergence of sea ice against the Canadian Archipelago and the other forcing an export of sea ice through Fram Strait. In the positive phase, the anticyclonic cell contracts toward the Beaufort Sea. Flow from Eurasia to the Canadian Archipelago and Greenland occupies a large part of the Arctic Ocean. Changes of atmospheric conditions from the negative to positive phase of the AO are characterized by a cyclonic anomalous wind stress prevailing over the entire Arctic domain, with its center located from the Eurasian basin to the northern Barents Sea (Fig. 3c). Positive anomalies of air temperature occupy the Arctic Ocean except the Bering Strait, the Canadian Archipelago, and part of the GIN Sea. The maximum anomalies occur around the coast of the Barents, Kara, and Laptev Seas, and extend to the Canada Basin and the Beaufort Sea. It should be noted that the regressed 2-m air temperature anomalies in Fig. 3c are significantly different from those between successive subperiods of the ECMWF reanalysis (e.g., Plate 3b in Zhang and Hunke 2001), which shows a temperature increase over the area from the Bering Sea Chukchi Sea to the Beaufort Sea Canada Basin, with a maximum along the Alaska coast. The differences are attributable to the mix of AO index values in the subperiods of the ECMWF reanalysis, while our analysis is keyed to opposite phases of the AO. The model was integrated for 120 yr forced by monthly data from initial temperature and salinity from Levitus (1982) without initial sea ice. After about 100 yr, an approximate equilibrium state is reached by the sea ice and upper-ocean properties (Zhang and Zhang 2001). To explore the changes of the Arctic Ocean caused by

5 2163 FIG. 2. (a) Spatial pattern of first EOF of SLP north of 62.5 N (i.e., EOF62) during the period from 1958 to 1998; (b) PC of the first EOF shown in (a) (i.e., PC62, black solid line), AO index (red dashed line), and NAO index (yellow dot dashed line).

6 2164 JOURNAL OF CLIMATE VOLUME 16 new equilibria in the ocean, the fact that the AO displays decadal variability, particularly in the recent 20 yr, implies that 10-yr integration should reflect changes driven by AO in the real world. If climate changes of the next century indeed include systematic phase shift, or phase locking, of the AO, then the simulations performed here will have prognostic implications in addition to the diagnostic potential exploited in this study. FIG. 3. Annual mean 2-m air temperature ( C) and wind stress (dyn cm 2 ) in (a) positive and (b) negative phase of the AO mode in this study. Their differences are displayed in (c). AO, two additional 10-yr simulations are then carried out with the monthly forcing data in the positive and the negative phase of AO, respectively. The changes of sea ice and ocean, and the associated underlying physical mechanisms, are analyzed by using the modeling results in the last year of each simulation. Although the 10-yr integrations are not sufficiently long to produce 4. Changes of Arctic sea ice driven by the AO mode a. Anomalous pattern of sea ice The general signals of sea ice changes on the basin scale have been captured in the previous modeling results (e.g., Zhang et al. 2000; Zhang and Hunke 2001). But significant regional differences occur among various models. Moreover, noticeable regional discrepancies emerge in existing modeling results, compared with observations. For example, Zhang et al. (2000) stated that their model exaggerates ice compaction off the Alaskan and Canadian coasts in the Beaufort Sea and predicts decompacting in the Fram Strait area, whereas observations show compacting. They also claimed that the large difference in the simulated ice thickness is unrealistic. The results of our modeled changes merit examination in this regard. According to our modeling, the motion and distribution of sea ice vary substantially with the atmospheric forcing. As shown in Fig. 4, a broader Beaufort gyre appears in the negative phase than in the positive phase of the AO (Figs. 4a,c,e). With the phase change to positive AO, the Beaufort gyre contracts and weakens, and the Transpolar Drift Stream (TDS) shifts toward the Beaufort Sea and the Canada Basin, consistent with the change of wind stress. Consequently, sea ice exhibits a cyclonic anomaly pattern relative to its climatology. Sea ice motion through Fram Strait is intensified. Accompanying the changes of motion are decreases of sea ice concentration over the eastern Arctic Ocean and increases from the Beaufort Sea to the Canada Basin and Fram Strait during the positive AO (Fig. 4e). Sea ice thickness shows a similar pattern of change (Fig. 4f), ranging from a decrease of 2 m in the East Siberian Sea to increases offshore of the Canadian Archipelago, in Fram Strait and around the New Siberian Islands and Novaya Zemlya. While our results show unambiguously that the AO pattern of forcing leads to an out-of-phase response of sea ice, regional differences between ours and previous modeling results are apparent. In our simulations, the maximum increase of ice concentration occurs in the Beaufort Sea. The change of sea ice thickness is positive in the Beaufort Sea, reaches its maximum off the Canadian Archipelago, and then decreases toward Fram Strait (Fig. 4f). By contrast, Zhang et al. (2000) obtain maximum changes of ice concentration and thickness

7 2165 FIG. 4. Modeled annual mean (a) sea ice velocity (cm s 1 ) and concentration (%) and (b) thickness (m) in the positive phase; annual mean (c) sea ice velocity (cm s 1 ) and concentration (%) and (d) thickness (m) in the negative phase. Their differences are shown in (e) and (f). in the center of the Canada Basin and from the Beaufort Sea to the Alaska shelf sea, respectively (see their Fig. 4 and 8). In the results of Zhang and Hunke (2001), ice concentration decreases in the Beaufort Sea and along the Alaskan coast, while thickness increases in the Beaufort Sea. In particular, their model produces a core of sea ice thinning off the Canadian Archipelago as shown in their Plate 3. Since the different combinations of years in the two studies resulted in different changes of winds and air temperatures, differences in the response of sea ice are not surprising. For example, Zhang and Hunke s warming is considerably larger than ours offshore of the Canadian Archipelago. In the context of the long-term climate variability point of view, this warming event

8 2166 JOURNAL OF CLIMATE VOLUME 16 may not be associated with the dominant mode, which in turn may have predictive implications. b. Comparison with observations Since 1979, satellites and IABP have provided more accurate and reliable sea ice data for the Arctic Ocean. The satellite remote sensing data have been incorporated in the new Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST 1.1; Parker et al. 1995). IABP supplies sea ice velocities at selected grid points. According to Fig. 2b, an abrupt jump of the AO index apparently happens in Therefore, we compute annual means of sea ice concentration and velocity for two decadal periods of and , as shown in Figs. 5a,b. The differences between these periods are displayed in Fig. 5c. The Beaufort gyre contracts and the TDS shifts toward the Beaufort Sea in Fig. 5a compared with Fig. 5b, which represent major changes of sea ice motion from (more negative AO) to (more positive AO). The differences are further characterized by an anomalous cyclonic pattern (Fig. 5c). Contemporarily, the observation shows that sea ice concentration decreases substantially over the east Arctic, from the Laptev and the East Siberian Seas to the Chukchi Sea and the Barents Sea. Conversely, the concentration significantly increases over the west Arctic. The largest increases appear north of Banks Island in the Beaufort Sea, in Fram Strait, and west of the Franz Josef Land in the Nansen Basin. A comparison between Figs. 4 and 5 indicates that sea ice change driven by the AO mode in the model is qualitatively very close to the observational change from to Thus we may largely attribute the recently manifested sea ice changes to the AO mode. The major loss of sea ice concentration in the East Siberian Sea is not captured very well by the model. This apparent discrepancy is most likely associated with the model s overestimate of sea ice thickness in this area. The overestimated sea ice thickness is resistant to a concentration reduction. However, the model s largest decreases of sea ice thickness indeed occur in the East Siberian Sea, consistent with the sea ice change in the real world. c. Overall sea ice change and its seasonality Figures 6a,b illustrate the seasonal cycle of the differences of sea ice area and volume between two AO phases over the entire Arctic Ocean, including the east and west Arctic Oceans, the Kara Sea, and the Barents Sea. The overall sea ice loss is dramatic from the negative to positive phase but it is not uniform seasonally. The largest decrease of sea ice area occurs from spring to early fall, with the maximum reduction in September. Sea ice area changes little from fall to winter, because the ocean surface temperature over most of our domain FIG. 5. Annual mean sea ice velocity (cm s 1 ) and concentration (%) during (a) and (b) , as well as (c) their differences. Sea ice velocity is from IABP and concentration is from HadISST 1.1. always drops below the freezing point during freezing season and results in some sea ice growth, no matter how strong the air temperature anomalies are. The out-of-phase changes of sea ice between the east and west Arctic as suggested in the previous section are also reflected in Figs. 6c f. Comparison between Figs.

9 2167 FIG. 6. Annual cycle of (a) sea ice area and (b) volume over the entire Arctic Ocean; (c) sea ice area and (d) volume over the east Arctic; and (e) sea ice area and (f) volume in the west Arctic in both of phases of AO. Unit of area is 10 6 km 2 ; unit of volume is 10 3 km 3. 6d and 6f indicates a stunning opposition of changes of sea ice volume in the east and west Arctic. The former shows decreases of volume, while the latter shows increases. By contrast, sea ice area decreases over both parts of the Arctic Ocean. The sea ice area decreases dramatically in the east Arctic Ocean but by much less in the west Arctic Ocean, where a small decrease occurs only during summer, but is close to zero in the annual mean as listed in Table 1. Tables 1 and 2 provide a quantitative summary of the annual mean sea ice area and volume in both AO phases over the east and west Arctic, the Kara Sea, the Barents Sea, and the entire Arctic Ocean. The decadal-scale model simulations for the positive and negative phases of AO show differences comparable to the decadal-scale observational change reported by Cavalieri et al. (1997), Parkinson et al. (1999), and Johannessen et al. (1999). The model-derived estimate of overall sea ice volume decrease over the entire Arctic Ocean is around 8.8%. There are no corresponding observational estimates of Arctic-wide changes of sea ice volume. The sea ice area reductions vary widely among the West Arctic TABLE 1. Annual mean sea ice area (10 6 km 2 ). East Arctic Barents Sea Kara Sea Entire Arctic Positive phase Negative phase Difference Percentage % % % % %

10 2168 JOURNAL OF CLIMATE VOLUME 16 TABLE 2. Annual mean sea ice volume (10 3 km 3 ). West Arctic East Arctic Barents Sea Kara Sea Entire Arctic Positive phase Negative phase Difference Percentage % % % % % different regions of the Arctic Ocean. When the AO changes from the negative to positive phase, the percentage decrease is largest over the Barents Sea and smallest over the west Arctic (Tables 1 and 2). The contrast between the changes of sea ice over the east and west Arctic is significant. Sea ice area decreases by about 5.9% in the east Arctic Ocean, a greater decrease than for the entire Arctic Ocean, while there is essentially no change in the west Arctic. A decrease of 40.2% of sea ice volume over the east Arctic and an increase of 20.8% in the west Arctic highlight the phase opposition of the changes of sea ice in these two parts of the Arctic Ocean. d. Interpretation of the anomalous sea ice pattern in terms of thermodynamics and dynamics Dynamic and thermodynamic processes are inherently coupled in sea ice and their interactions determine sea ice redistribution. In order to elucidate the thermodynamic and dynamic roles, we have performed a diagnostic analysis of sea ice growth/melt and transport, respectively. The sea ice growth/melt here is governed by thermodynamic processes, not including changes resulting directly from deformation although deformation ultimately affects the growth/melt rates. The transport is determined by dynamics. Furthermore, sea ice growth/melt has been separated into two parts: sea ice growth from open water and sea ice growth/melt for existing sea ice. Sea ice growth from open water occurs from fall to the following spring, and the maximum sea ice growth rates appear around September and October in both phases of AO (Figs. 7a,b). However, the change with the phase of the AO differs between the east and west Arctic. In the east Arctic, sea ice growth increases from the negative to the positive phase, while in the west Arctic the change is opposite. Similar differences are found in the growth/melt of existing sea ice (Figs. 7c,d). Moreover, summer sea ice melts less in the east Arctic in the positive phase compared with the negative phase (Fig. 7c). Summer sea ice melt changes little in the west Arctic Ocean between both phases. The changes of sea ice melt in our model are completely different from those in Zhang et al. (2000), which shows a noticeable increase of sea ice melt in the east Arctic Ocean and a modest decrease in the west Arctic Ocean. As a result, thermodynamics causes more sea ice production in the east Arctic but less in the west Arctic during the positive phase of the AO. A quantitative comparison of annual mean sea ice growth/melt between different phases and among various regions is presented in Table 3. Net sea ice production increases sharply in the east Arctic but decreases in the west Arctic in the positive phase. The noticeable decrease of sea ice melt in the east Arctic Ocean and the modest increase of sea ice melt in the west Arctic in the positive phase contrast with the results of Zhang et al. (2000). Obviously, the sea ice production caused by the thermodynamic processes cannot account for all the changes of sea ice in the both parts of Arctic Ocean. The role of dynamics must also be considered. The annual mean sea ice transports are presented in Table 4. In the longterm climatology, two branches of the Beaufort gyre play critical roles in redistributing sea ice thickness, one carrying sea ice from the Beaufort Sea and the Canada Basin to the Chukchi Sea and the East Siberian Sea, while the other results in a buildup of sea ice along the Canadian Archipelago. In order to diagnose the exchanges of ice mass between the east and west Arctic, we consider the respective roles that the two branches of the Beaufort gyre play by separating the transport into two parts in Table 4. The first column in the table represents transport from the west Arctic to the east Arctic; thus transport occurs primarily from the Beaufort Sea to the Chukchi and the East Siberian Seas. The second column represents the opposite transport and represents primarily the TDS from the east Arctic to the west Arctic. The sea ice transport from the west to the east Arctic decreases when the AO shifts from the negative to positive phase, reflecting the weakened and contracted Beaufort high. Although the actual transport from the east to west Arctic also decreases, the net flow of sea ice from the east to west Arctic is significantly larger in the positive phase of the AO. Thus, there is a net loss of sea ice by transport from the east Arctic Ocean in the positive phase. The sea ice transports also indicate that an anomalously large amount of sea ice stays in the Beaufort Sea and the Canada Basin, while a considerable amount leaves the east Arctic, during the positive phase. From the individual values of transport, it is apparent that the transport from the west to the east Arctic Ocean changes tremendously, compared with the transport

11 2169 FIG. 7. Annual cycle of sea ice growth in open water in (a) the east and (b) west Arctic; sea ice growth/melt for existing sea ice in (c) the east and (d) west Arctic; and total sea ice growth/melt in (e) the east and (f) west Arctic in both of phases of AO. Positive (negative) sign denotes sea ice growth (melt). Unit is 10 3 km 3 yr 1. from the east to the west Arctic. This is mainly due to the contraction of the Beaufort gyre. At this time, wider flows from the east to the west Arctic occur in spite of its weakening. Zhang et al. (2000) deduced an overall transport budget of the east and the west Arctic Ocean based on their sea ice distribution, which did not show a clear picture of the sea ice exchange between the east and west Arctic Ocean. In particular, both the east and west Arctic Ocean defined in their paper include Fram Strait, through which large outflow occurs. Accordingly, thermodynamics and dynamics interact critically in sea ice production, loss, and redistribution. In the west Arctic, the thermodynamics results in a decrease of sea ice growth, but the dynamic mechanism limits the westward outflow of sea ice from this region and supplies sea ice from the east Arctic. As for the east Arctic, sea ice production strengthens but the substantial reduction of sea ice input from the west Arctic and a larger amount of sea ice outflow deplete sea ice volume there during the positive phase. The fact that TABLE 3. Annual mean sea ice growth/melt in the east and west Arctic Oceans (positive is sea ice growth; unit is km 3 yr 1 ). The values in the second rows of the positive and the negative phase, as well as the change, represent the overall net growths and their changes. Growth (open water) East Arctic Growth (existing ice) Melt Growth (open water) West Arctic Growth (existing ice) Positive Negative Change Melt

12 2170 JOURNAL OF CLIMATE VOLUME 16 TABLE 4. Annual mean sea ice transport between the west and east Arctic Oceans (positive is toward the east Arctic Ocean; unit is km 3 yr 1 ). Positive phase Negative phase Change Toward east Arctic Transports Toward west Arctic Net the dynamical effects predominate is responsible for the spatial pattern of model-derived anomalies described earlier. Nevertheless, a key finding for the mass budget is that the thermodynamic processes resulting from anomalous transport tend to oppose and offset (partially) the dynamic-induced changes. It might be deserving to emphasize that our thermodynamic analyses above have all thermodynamic factors involved, including the ice albedo feedback. Zhang et al. (2000) separated the sea ice growth and melt. They attribute the sea ice melt to the dynamics and ice albedo feedback through correlation coefficient analysis. We allow the ice albedo feedback and would not rule out its role in amplifying the sea ice melt. However, the ice albedo feedback is difficult to isolate because radiation absorbed by the ocean definitely depends on sea ice concentration as well as on the albedo of the sea ice surface. Less sea ice cover certainly results in more radiation into the ocean. But this does not mean that the absorbed radiation goes into the melting of sea ice because many other processes affect heat distribution; for example, outgoing longwave radiation increases with a reduction of sea ice cover. For example, even if the absorbed radiation melts sea ice, it is still hard to determine its importance because of increased oceanic heat flux by the warmer ocean during the same period (Steele and Boyd 1998; Zhang et al. 1998). Changes of heat budgets are beyond the scope of this paper and will be explored in the future. TABLE 5. Annual mean sea ice export via Fram Strait (10 3 km 3 yr 1 ), area (10 6 km 2 ), and volume over the GIN Sea (10 3 km 3 ). Positive phase Negative phase Difference Percentage Export Area Volume % % % e. Sea ice export via Fram Strait As has already been noted in the previous section, the anomalous cyclonic wind stress and the resulting sea ice velocity in the positive phase of the AO intensify sea ice outflow through Fram Strait. Table 5 quantifies this intensification and shows that sea ice export through Fram Strait increases dramatically, by more than 50%, in the positive phase compared with the negative phase. Our simulated outflow of 2033 km 3 yr 1 in the positive phase of the AO is comparable to Kwok and Rothrock s (1999) satellite-derived estimate of 2366 km 3 yr 1 for the period, when the AO was generally positive. Our results show that the GIN Sea gains more sea ice, which is apparent from the values of sea ice area and volume in Table 5, during the positive AO. However, Hilmer and Jung (2000) have shown that the relationship of the NAO to ice export through Fram Strait is not robust when subtle shifts of the NAO s subpolar center of action occur over multidecadal timescales. The major driver of the Fram Strait sea ice export is the horizontal pressure gradient across Fram Strait. Considering the continuous and coherent characteristics and the fixed spatial pattern of the AO mode, the relationship of the AO and the sea ice export applies to the any time period when the AO predominates over the decadal scale. Our model shows that the strengthened outflow and the thickened sea ice in Fram Strait combine to increase the sea ice export. However, as mentioned above, Zhang et al. (2000) indicated that their model had unrealistically thinning ice. So, their increased ice export through Fram Strait is evidently caused by the velocity increase. It is noteworthy that the mean Fram Strait sea ice export over the past 50 yr was estimated to be about 2900 km 2 yr 1 based on observations (Vinje 2001). Our model appears to underestimate the long-term mean Fram Strait sea ice export. This seems to be characteristic of other models, as summarized by Zhang and Zhang (2001), and needs to be improved in future work. 5. Changes of FW driven by the AO mode a. Anomalous pattern of FW storage It is intuitive that detectable changes of ocean hydrographic properties should accompany the changes of overlying atmosphere and sea ice. Here we examine the AO-induced changes of FW storage in the upper ocean. The FW was defined as salinity deficit relative to the salty Atlantic water. We choose a reference salinity of S 34.8 parts per thousand (ppt), consistent with previous studies (Aagaard and Carmack 1989; Steele et al. 1996; Zhang and Zhang 2001). The salty and warm Atlantic water invades the Arctic Ocean through Fram Strait and the Barents Sea. The Bering Strait throughflow also carries mildly salty and warm water into the Arctic. Both waters sink when they redistribute in the Arctic Ocean. So, the cold and fresh Arctic water resides predominantly in the upper mixed layer and halocline. From the model output, we have calculated the storage in the upper 210 m, which represents an average halocline base depth (Steele et al. 1996). In the upper layer, cold- and freshwater moves predominantly in an anticyclonic sense in both phases of the AO (Figs. 8a,b). A major amount of FW storage is found in the Beaufort

13 2171 carries FW from the Arctic Ocean into the GIN Sea and the North Atlantic through Fram Strait. Like sea ice, FW is also subject to significant changes driven by the AO. Figure 8c illustrates the cyclonic anomalous upper-ocean circulation of the positive AO phase relative to the negative AO. This cyclonic feature dominates the upper layer of most of Arctic Ocean, reflecting the degeneration and weakening of the Beaufort high. Figure 8c s differences of FW storage (positive AO less negative AO) show more FW accumulated in the west Arctic, from the Beaufort Sea to the Canada Basin, the Canadian Archipelago, and Fram Strait. The FW storage anomaly is negative in much of the east Arctic, especially in the Eurasian Basin, in the positive phase. The general out-of-phase relationship of FW storage anomalies between the east and west Arctic Ocean is consistent with the corresponding differences of sea ice growth/melt and dynamic transport. Superimposed on the broader patterns in Fig. 8c are centers of action in the western Eurasian Basin and Canada Basin, contributing to the more localized variations of FW storage. The salinification of the Eurasian Basin in the positive phase agrees with the observationally derived finding by Steele and Boyd (1998), who inferred a retreat of the cold halocline layer from the Amundsen basin back to the Makarov Basin during recent decades. The freshening of the Beaufort Sea has also been detected by observations. When comparing the hydrographic data during the SHEBA deployment in October 1997 with those during the Arctic Ice Dynamics Joint Experiment (AIDJEX) in 1975, McPhee et al. (1998) found that the upper ocean in the Beaufort Sea is less saline than in previous years. The FW storage estimate in SHEBA is about 2.5 times that in AIDJEX. Hydrographic surveys of the southern Canada Basin have also been conducted in winter almost annually since Melling (1998) analyzed data from this area for the 17-yr period ending in 1997, and found evidence for salinity decreases of about ppt at the lower 0 C isotherm and ppt at the temperature maximum in the early 1990s, compared to the earlier years. The results imply recent increases of FW storage in the Beaufort Sea and the Canada Basin, consistent with our model-derived changes from the negative to positive phase of the AO. FIG. 8. Annual mean freshwater storage (m) and vertically integrated velocity (cm s 1 ) in upper 210 m in the (a) positive and (b) negative phases, as well as (c) their differences. Sea and Canada Basin, and extending into the Eurasian basin. Another large amount of FW storage appears in the shelf seas from the Kara to the East Siberian Sea, which is attributable to Russian river runoff. Since the vertically averaged velocity vectors in Fig. 8 represent transport in the upper layer, the anticyclonic circulation b. Overall changes of FW storage and their seasonality For the entire Arctic Ocean, the area-integrated FW storage varies seasonally in a sinusoidal way in both phases of AO, attaining minimum in spring and maximum in fall (Fig. 9a). Sea ice development and decay, as well as riverine water input, contribute to such a seasonal cycle. The overall FW storage decreases yearround when the phase of the AO shifts from negative to positive. The largest difference occurs in October. Similar to the changes of sea ice, the changes of broad FW storage in the east and west Arctic are out of phase,

14 2172 JOURNAL OF CLIMATE VOLUME 16 FIG. 9. Annual cycle of freshwater storage in (a) the entire Arctic Ocean, (b) the east Arctic, and (c) the west Arctic. Unit is km 3. as seen in Figs. 9b,c. During a positive phase FW significantly decreases in the east Arctic, while it increases in the west Arctic throughout the whole year. However, a larger increase of FW storage in the west Arctic occurs in spring than in any other season. The annual mean area-integrated FW storages are given quantitatively in Table 6. The overall FW storage for the entire Arctic Ocean shrinks during the positive phase of the AO. The contrast of FW changes between the east and west Arctic Oceans reinforces the out-of-phase relationship between these two regions. Additionally, FW storage in the Kara and the Barents Seas decreases during the positive AO. c. Interpretation of FW change in terms of sea ice growth/melt and oceanic transport Freshwater fluxes resulting from sea ice growth/melt change the FW storage. In the negative phase of the AO, net annual sea ice production occurs in the Beaufort Sea, the Canada Basin, the Laptev Sea, and the Kara Sea. As a result, the FW fluxes from the surface are negative in these regions (Fig. 10b). Contemporarily, there is a net melt of sea ice in the Bering Strait, the Chukchi Sea, the East Siberian Sea, and the Greenland Sea, supplying FW to the ocean. In particular, the Bering Strait throughflow carries warm Pacific water into the Chukchi Sea, causing strong sea ice melt and supplying FW to the ocean. During the positive phase of the AO, sea ice growth weakens in the west Arctic, while it intensifies in the east Arctic (Fig. 10a). The former reduces the FW loss and the latter increases the FW loss in respective regions. Meanwhile, the FW flux from sea ice melt decreases substantially in the Bering Strait, the Chukchi Sea, and the east Siberian Sea, resulting in anomalous negative FW fluxes. However, Fram Strait and the Greenland Sea receive enhanced FW fluxes due to greater sea ice melt. The broad similarity of Figs. 10c and 8c suggests that the change of sea ice development between the two phases of the AO is a major contributor to FW storage anomalies. Throughout the year, FW fluxes caused by sea ice growth/melt vary regionally and seasonally (Fig. 11). From September to April, the loss of FW intensifies in the east Arctic but weakens in the west Arctic during the positive phase, reflecting enhanced and reduced sea ice formations in these two regions, respectively. During the summer, the FW flux diminishes in the east Arctic while it changes little in the west Arctic. A quantitative summary of the annual means can be found in Table 7. On an annual basis, there is a net loss of FW in the east Arctic and a net gain in the west Arctic when the AO phase shifts from negative to positive. Changes of upper-ocean circulation lead to FW redistribution. We consider two parts of the overall pattern of the circulation. One is the portion in the Beaufort Sea and Canada Basin, and the other is the portion located primarily in the Makarov and Eurasian Basins. In the climatology, the former transports FW from the west TABLE 6. Annual mean freshwater storage (10 3 km 3 ). West Arctic East Arctic Barents Sea Kara Sea Entire Arctic Positive phase Negative phase Difference Percentage 0.93% 4.89% 5.71% 4.37% 2.02%

15 2173 FIG. 11. Annual cycle of freshwater fluxes from sea ice growth and melt in the (a) east and (b) west Arctic Ocean in both of the phases of the AO. Positive (negative) sign denotes freshwater entering (leaving) ocean. Unit is 10 3 km 3 yr 1. FIG. 10. Annual mean freshwater flux from sea ice growth/melt in the (a) positive, (b) negative phases, and (c) their differences. Positive (negative) sign denotes freshwater entering (leaving) ocean. Unit is mm month 1. Arctic to the east Arctic, while the latter brings FW toward the west Arctic and out of the Arctic Ocean through Fram Strait. The circulation anomalies change these FW transports. The FW transport from the west to east Arctic is smaller by over one-fifth during the positive AO (Table 8). Hence, a considerable amount of FW remains in the Beaufort Sea and the Canada Basin in the positive phase, compared with the negative phase. Simultaneously, the FW transport from the east to west Arctic is also reduced. The net change is a very small increase of FW supply into the west Arctic Ocean during the positive AO. With respect to the recent freshening of the Beaufort Sea and the Canada Basin, McPhee et al. (1998) suggested that the change was caused by increased sea ice melt. However, our model shows that the dramatically decreased sea ice growth also plays an important role. Further, when comparing contributions of sea ice growth/melt and ocean transport to FW anomalies on the basin scale, sea ice growth/melt plays a dominant role in FW buildup in the west Arctic Ocean. However, when one focuses on the largest FW accumulation in the Beaufort Sea and the Canadian Archipelago, dynamic transport anomalies appear to makes important contributions through the reduced export of FW to the east Arctic (Table 8). This is because the freshwater transports from the west to the east and from the east TABLE 7. Annual mean freshwater flux from sea ice growth/melt in the east and west Arctic Oceans (positive is freshwater into ocean; unit is 10 3 km 3 yr 1 ). Positive phase Negative phase Difference East Arctic West Arctic