3.1. Freshwater flux from AISM

It is difficult to estimate the circumpolar freshwater flux from AISM. Our knowledge of the melt rate of ice shelves is mostly from estimations based on glaciological and oceanographic observations or numerical simulations, since direct measurements are still very limited due to difficulties in accessing the sub-ice-shelf cavity. However, measurements or numerical simulations have not been done for all Antarctic ice shelves. Different approaches may yield large differences in estimated ISM (e.g. due to the assumption of a steady-state ice sheet (Reference Shepherd, Wingham and RignotShepherd and others, 2004) and the selection of outflow depth (Reference Payne, Holland, Shepherd, Rutt, Jenkins and JoughinPayne and others, 2007)).

The simulated total annual mean circumpolar freshwater flux from AISM, 106 mSv, is considerably larger than previous estimates (e.g. 26.14 mSv (Reference Jacobs, Hellmer and JenkinsJacobs and others, 1996), 31.6 mSv (Reference Beckmann, Hellmer and TimmermannBeckmann and others, 1999) and 28.42 mSv (Reference HellmerHellmer, 2004)). However, closer inspection of the individual ISM values reveals that many of these estimates are comparable to other studies, particularly for the major ice shelves (Table 1). For example, the freshwater flux from the Filchner-Ronne Ice Shelf (FRIS) is 6.52 mSv in this study, in good agreement with an earlier glaciological shelf-wide estimate (Reference Jacobs, Hellmer and JenkinsJacobs and others, 1996) and the numerical results from cavity-resolving models (e.g. Reference Timmermann, Hellmer and BeckmannTimmermann and others, 2002). The freshwater fluxes from the other large ice shelves (e.g. Ross (8.36mSv) and Amery (5.44mSv), as well as eastern Weddell (4.89 mSv)) in this study are also comparable to the other studies from regional cavity-resolving models (e.g. Reference Beckmann, Hellmer and TimmermannBeckmann and others, 1999; Reference Timmermann, Hellmer and BeckmannTimmermann and others, 2002; Reference HellmerHellmer, 2004). The freshwater flux from the Fimbul Ice Shelf is ~19.80 mSv in this study, overestimated by about a factor of two compared with other studies (e.g. Reference Beckmann, Hellmer and TimmermannBeckmann and others, 1999; Reference Beckmann and TimmermannBeckmann and Timmermann, 2001; Reference Timmermann, Hellmer and BeckmannTimmermann and others, 2002; Reference HellmerHellmer, 2004; Reference Walkden, Heywood, Nicholls and AbrahamsenWalkden and others, 2009), due to the warm bias in our simulation because of the coarse model resolution (Reference WangWang, 2011). Recent in situ observations indicate that the water temperature under the Fimbul Ice Shelf is generally <0.1 °C above the surface freezing point (Reference Hattermann, NØst, Lilly and SmedsrudHattermann and others, 2012), suggesting that basal melting here cannot simply be parameterized by coastal deep ocean temperatures. The modeled freshwater flux from the Getz Ice Shelf (16.4 mSv) is more than twice that estimated from glaciological measurements (Reference Shepherd, Wingham and RignotShepherd and others, 2004). The reason for this discrepancy requires further investigation. The high melt rate in the Amundsen and Bellingshausen Seas is generally consistent with available estimates. For example, the freshwater flux from the George VI Ice Shelf is 5.35 mSv in this study, corresponding to a melt rate of 3 m a^-1, which is close to recent measurements of 3–5 m a^-1 (Reference Jenkins and JacobsJenkins and Jacobs, 2008) and 2.49 m a^-1 (Reference Holland, Jenkins and HollandHolland and others, 2010), although higher than previous estimates of 1.9 m a^-1 (Reference Jacobs, Hellmer and JenkinsJacobs and others, 1996) and 0.43 m a^-1 (Reference HellmerHellmer, 2004). Pine Island Glacier Ice Shelf (PIG) is experiencing extremely high melt rate due to the warm circumpolar deep water (CDW) flowing onto the continental shelf. Estimates of melt rates for PIG are in a wide range, from 10 m a^-1 (Reference Jacobs, Hellmer and JenkinsJacobs and others, 1996) to 24 ± 4ma^-1 (Reference RignotRignot, 1998). The freshwater flux from our simulation is 1.62 mSv, corresponding to a melt rate of 21 m a^-1, close to the simulation of 20.6 m a^-1 by Reference Payne, Holland, Shepherd, Rutt, Jenkins and JoughinPayne and others (2007) with a plume model, and approximating the simulation of 28 m a^-1 using NASA IceBridge data by Reference Schodlok, Menemenlis, Rignot and StudingerSchodlok and others (2012). The freshwater fluxes from the minor ice shelves (Lazarev, Voyeykov, Cook and Dotson) need further verification, since there have been very few studies of these ice shelves.

Table 1. Freshwater flux (mSv) from ISM in different studies

The modeled total circumpolar freshwater flux from the AISM during the integration period 1958–2000 (Fig. 2) shows significant interannual variability, with a clear increase since the mid-1960s, consistent with the observed increase in basal melting during recent decades (Reference Rignot and JacobsRignot and Jacobs, 2002; Reference Shepherd, Wingham and RignotShepherd and others, 2004). An increase in basal melting has been suggested to be a major cause of the thinning of ice shelves in the Amundsen Sea (Reference Shepherd, Wingham and RignotShepherd and others, 2004), owing to the ocean warming (Reference Shepherd, Wingham and RignotShepherd and others, 2004) and changes in wind patterns in the Southern Hemisphere (Reference Pritchard, Ligtenberg, Fricker, Vaughan, Van den Broeke and PadmanPritchard and others, 2012; Reference Steig, Ding, Battisti and JenkinsSteig and others, 2012).

3.2. Impact of AISM on temperature

Figure 3 shows climatological mean temperature differences ΔT (with minus without AISM) at different depths. The impact of including AISM in temperature is very complex. Water in the vicinity of the Antarctic ice shelves generally becomes colder above 216 m (Fig. 3a-c), due to the cold fresh water from ISM (Reference HellmerHellmer, 2004; Reference Wang and BeckmannWang and Beckmann, 2007). The largest cooling is found at 216m (Fig. 3c), the base of most ice shelves’ edges in the model (except the FRIS and the Ross Ice Shelf (RIS)), where the heat and freshwater fluxes due to AISM are applied. Substantial cooling at 216 m is found in the coastal areas of the Amundsen, Weddell and Ross seas, with the most extreme values of ΔT in each region being about –2.2°C, –1.5°C and –1.25°C, respectively. The ISM of the smaller ice shelves in East Antarctica tends to cause much smaller ΔT in the vicinity of the continent, generally <0.2°C; the exception is the region around Amery Ice Shelf.

Fig. 3. Climatology mean temperature differences ΔT (°C; with minus without AISM) at depths of (a) 5 m, (b) 55 m, (c) 216 m, (d) 512 m, (e) 1033 m and (f) 3752 m. Positive temperature differences represent temperature increase (shaded red); negative temperature differences indicate temperature decrease (shaded blue).

The coastal cooling at 512 m (Fig. 3d) is weak, only occurring near the much thicker ice shelves FRIS and RIS. Close to the surface, cooling affects a much larger area (Fig. 3a and b). At 55 m (Fig. 3b), the cooling becomes weaker but broader next to the continent, as a result of the rising cold buoyancy fresh water from the ice shelves and offshore advection by mean wind stress. At the surface (Fig. 3a), ΔT is smaller than at mid-depths, implying a more extensive mixing between the colder ice-shelf water and the warmer CDW, and a fast exchange with the atmosphere. The most pronounced cooling at the surface is not next to the continent, but in the Antarctic Circumpolar Current (ACC) region north of the Weddell Sea (down to –1.0°C). The surface cooling next to the continent is mostly <0.4°C except in two regions: one is outside the Amery Ice Shelf, and the other is close to Cape Adare in the western Ross Sea, both being about –0.8°C.

Most of the Southern Ocean is undergoing significant warming when AISM is included (Fig. 3). Above 1000 m, ΔT is generally not >1.0°C, except in the Amundsen and Bellingshausen Seas and the eastern Weddell Sea. The largest warming occurs at 216 m (Fig. 3c), up to 1.2°C. Large warming is also seen in the Atlantic and Pacific sectors of the ACC regions. Further down, the magnitude of warming tends to decrease but the warming area tends to increase. Below ~1000m, almost all the Southern Ocean water south of 50° S is warmer with AISM (Fig. 3e and f). However, the magnitude of warming is spatially inhomogeneous. For example, at 1033 m (Fig. 3e), the largest ΔT is found in the Weddell Sea, up to 1.2°C, and near the continent, generally >0.6°C. In other areas, warming is much smaller, generally <0.4°C. Down to the bottom, ΔT decreases. At the same time, the warming pattern also changes. At 3752 m (Fig. 3f), the largest warming (>0.6°C) takes place in the Australia- Antarctic Basin in the Indian Ocean, not next to the continent as in the shallower layers.

Figure 2 suggests that total AISM increased ~10% between the mid-1960s (102 mSv) and the mid-1980s, after which it stayed fairly constant between 110 and 112 mSv. If the effect of AISM on Southern Ocean hydrography is correctly captured by the model, we should see similar trends in data although with much smaller total change.

Reference GilleGille (2002) reported a warming of 0.17°C in the Southern Ocean waters at 700–1100 m depth between the 1950s and 1980s. Most observed warming occurred in the ACC region between 45°S and 60°S, especially in the Atlantic and Indian sectors. This is almost identical to the warming pattern shown above as a result of increasing AISM (Fig. 3d and e). Reference Aoki, Yoritaka and MasuyamaAoki and others (2003) showed another example at depths of 200–900 m around 30–150° E in the ACC region, where warming over three decades (1966–98) suggested that the southern ACC front had shifted southward. A similar warming is reproduced in our model (Fig. 3c-e). Reference Levitus, Antonov and BoyerLevitus and others (2005) reported changes in ocean heat content vs time for 0–300, 0–700 and 0–3000 m integrations, finding that about half of the increased heat content was from the Atlantic Ocean. They also noted the cooling above ~600m and between 1100 and 1450 m at high southern latitudes. The cooling above ~600m is reproduced near the coastal regions (Fig. 3), whereas the cooling below ~ 1100 m is not reproduced in our simulation. This is possibly caused by the dynamic responses of ISM to changing wind (Reference Pritchard, Ligtenberg, Fricker, Vaughan, Van den Broeke and PadmanPritchard and others, 2012; Reference Steig, Ding, Battisti and JenkinsSteig and others, 2012), or by the Antarctic Surface Water flushing and melting the ice base (Reference Hattermann, NØst, Lilly and SmedsrudHattermann and others, 2012), which are not captured by the model.

The warming of the deep and bottom waters in the Southern Ocean over the past half-century has been reported by a number of studies. Reference Fahrbach, Meyer, Rohardt, Schröder and WoodgateFahrbach and others (1998) observed that the bottom water in the central Weddell Sea became warmer during the interval 1989–95. This warming trend was shown to affect most of the Weddell Sea when the data were extended to the early 2000s (Reference Fahrbach, Hoppema, Rohardt, Schröder and WisotzkiFahrbach and others, 2004). Observations dating back to 1912 indicated that the warming of the Weddell Sea Warm Deep Water had occurred since the 1970s (Reference Robertson, Visbeck, Gordon and FahrbachRobertson and others, 2002). The warming of the bottom water was also observed in the Indian sector (Reference WhitworthWhitworth, 2002; Reference Johnson, Purkey and BullisterJohnson and others, 2008) and Pacific sector (Reference Ozaki, Obata, Naganobu and GamoOzaki and others, 2009) of the Southern Ocean. In particular, Reference Ozaki, Obata, Naganobu and GamoOzaki and others (2009) revealed a long-term warming of the bottom layer of the Ross Sea from 1969 to 2004. Reference Johnson, Purkey and BullisterJohnson and others (2008) reported warming of the abyssal waters in the eastern Indian Ocean between 1994/95 and 2007. These warming trends are consistent with the effects of including AISM (Fig. 3e and f).

3.3. Impact of AISM on salinity

The impact of AISM on the Southern Ocean salinity at different levels is shown in Figure 4. AISM inputs fresh water into the adjacent ocean. As for temperature, the strongest freshening is seen close to 216 m (Fig. 4c), but with large and extensive freshening at shallower depths as well (Fig. 4a and b). At 216 m, the largest freshening is in the vicinity of the continent (down to –1), where most of the ice-shelf/ocean interaction was applied in the model. In addition, freshening is dominant over the whole Southern Ocean at this depth, except for patches of slightly higher salinity (up to 0.1) in the central Kerguelen, Weddell and Ross gyres.

Fig. 4. Same as Figure 3, but for salinity.

The substantial freshening at the surface (Fig. 4a and b) arises because the fresh subsurface water rises to the surface due to its low density. The freshening at the surface is particularly remarkable in the Ross and Weddell Seas (down to –0.5), the Amundsen and Bellingshausen Seas (down to –0.4) and Prydz Bay (down to –0.3). In the Indian Ocean and western Weddell Sea, the freshening is generally much weaker. In the region east of 45° E, the water even becomes more saline, up to 0.2. At 55 m, almost all of the Southern Ocean becomes fresher (Fig. 4b). Unlike at the surface, the freshening is more pronounced next to the continent at this depth. Aside from the general freshening, small areas of more saline water up to 0.1 are found around the 45° E meridian and in the southern Ross Sea.

At depths of 512 and 1033 m, inclusion of AISM generally causes higher salinity in the Southern Ocean (Fig. 4d and e). Freshening is only found in the Weddell and Ross Seas close to the continent, where the decrease in salinity is up to 0.3. At the bottom (Fig. 4f), most of the Southern Ocean south of ~55° S shows a trend of reduced salinity when AISM is included, although the changes are small (<0.02).

As with temperature (Section 3.2), we can use these model results to predict the change in the Southern Ocean salinity associated with the modeled 10% increase in AISM between the 1960s and 1980s (Fig. 2). Reference Wong, Bindoff and ChurchWong and others (1999) found a large-scale freshening of intermediate water in the Pacific and Indian Oceans between the periods 1930–80 and 1985–94, and partly attributed this to surface water freshening in the source region of the Southern Ocean. They estimated a net increase in freshwater flux of ~31 mm a^-1 (~22 mSv) for the Southern Ocean between 55° S and 65° S. The increasing freshwater input is 8–12.5 mm a^-1 (~6–9 mSv) from coupled atmosphere-ocean models, with increasing atmospheric greenhouse-gas concentration for the same period (Reference Manabe, Bryan and SpelmanManabe and others, 1990; Reference Gordon and O’FarrellGordon and O’Farrell, 1997), which could explain about one-third of the net increase in freshwater flux. The increase in the simulated freshwater flux from AISM between the mid-1960s and mid-1980s is ~10 mSv (Fig. 2). It is likely a significant contributor to the freshwater flux variability. In addition, the simulated large salinity decrease in the vicinity of ice shelves is not inconsistent with summer observations near the RIS by Reference Jacobs, Giulivi and MeleJacobs and others (2002), who found a rapid freshening in the surface water in the Ross Gyre and much deeper close to the ice-shelf edge. Reference Jacobs, Giulivi and MeleJacobs and others (2002) attributed this freshening to the increasing AISM in the Bellingshausen and Amundsen Seas, which is also found in later in situ observations (e.g. Reference Shepherd, Wingham and RignotShepherd and others, 2004; Reference WåhlinWåhlin and others, in press).

The modeled spatial extent of surface freshening down to 500 m in the Pacific, Atlantic and Indian sectors is generally consistent with the observed freshening (Reference Boyer, Levitus, Antonov, Locarnini and GarciaBoyer and others, 2005). It is seen from the observations that strong freshening occurs in the Pacific and Atlantic sectors of the high-latitude Southern Ocean, with the most significant salinity decrease in the top surface waters. In the Indian sector of the Southern Ocean, however, the observed overall salinity is slightly increasing (Reference Boyer, Levitus, Antonov, Locarnini and GarciaBoyer and others, 2005). This pattern is qualitatively depicted in the present study (Fig. 4), where the freshening in the Indian sector is much weaker than in the Pacific and Atlantic sectors. Consistent with recent observations by Reference RintoulRintoul (2007), Reference Johnson, Purkey and BullisterJohnson and others (2008) and Reference Ozaki, Obata, Naganobu and GamoOzaki and others (2009), increasing AISM would cause freshening of the bottom water in the Southern Ocean (Fig. 4). Such overall consistency presented in this study has not been simulated in any previous study.

3.4. Impact of AISM on density

Potential density is determined by both temperature and salinity. However, in most regions of the cold Southern Ocean, density is mainly controlled by salinity. Therefore, the general patterns of density change (Fig. 5) are similar to those of salinity at most of the vertical levels, where density generally decreases. The largest decrease in density (>0.2 kg m^-3) is found at the surface (Fig. 5a), particularly in the Weddell, Bellingshausen, Amundsen and Ross seas. Below the surface layer, a large decrease in density is mainly confined to the continental margins down to 216 m. At 1033 m, the water decreases slightly in density in a fairly large area (Fig. 5e), where both temperature (Fig. 3e) and salinity (Fig. 4e) increase due to the inclusion of AISM. Here the increase in temperature plays a dominant role in the density change. At 3752 m the decrease is generally <0.1 kgm^-3.

Fig. 5. Same as Figure 3, but for density (kgm^-3).

The simulated density decrease in the bottom water is consistent with observations (Reference WhitworthWhitworth, 2002; Reference Aoki, Yoritaka and MasuyamaAoki and others, 2003; Reference RintoulRintoul, 2007; Reference Johnson, Purkey and BullisterJohnson and others, 2008; Reference Ozaki, Obata, Naganobu and GamoOzaki and others, 2009). Due to the rapid freshening of the bottom water, the decrease in density is also significant, resulting in loss of the densest water classes (e.g. Reference Johnson, Purkey and BullisterJohnson and others, 2008). Additionally, warming of the intermediate water would also cause a decrease in density. Such changes may reduce the stability of the water column, enhancing eddy formation due to baroclinic instability (e.g. Reference Meredith and HoggMeredith and Hogg, 2006; Reference Screen, Gillett, Stevens, Marshall and RoscoeScreen and others, 2009).