INTRODUCTION
In the 21st century, changes in the mass balance of Antarctic ice shelves are expected to be driven largely by basal melting (Joughin and others, Reference Joughin, Alley and Holland2012; Pritchard and others, Reference Pritchard2012). Because ice shelves are hundreds of meters thick, these changes must be driven by the circulation or hydrographic properties of the deepest water on the continental shelf (Jacobs and others, Reference Jacobs, Hellmer, Doake, Jenkins and Frolich1992; Pritchard and others, Reference Pritchard2012). The properties of this water mass, termed Antarctic continental shelf bottom water (ASBW) by Schmidtko and others (Reference Schmidtko, Heywood, Thompson and Aoki2014), vary widely around Antarctica, but can be roughly segregated into two regimes. In the first regime, epitomized by the larger continental shelves in the Ross and Weddell Seas (RS and WS, respectively), bottom waters are formed locally and influenced strongly by sea ice processes (Loose and others, Reference Loose, Schlosser, Smethie and Jacobs2009; Nicholls and others, Reference Nicholls, Østerhus, Makinson, Gammelsrød and Fahrbach2009), resulting in a water mass (high-salinity shelf water, or HSSW) that is very close to the surface freezing point. In contrast, bottom water in the Amundsen Sea (AS) is largely composed of relatively unmodified Circumpolar Deep Water (CDW), with temperatures up to 1.8°C (Jacobs and others, Reference Jacobs, Hellmer and Jenkins1996, Reference Jacobs, Jenkins, Giulivi and Dutrieux2011; Dutrieux and others, Reference Dutrieux2014; Schmidtko and others, Reference Schmidtko, Heywood, Thompson and Aoki2014). The origins and physical processes involved in the formation and modification of these water masses might suggest that different regimes are unlikely to respond similarly to climate change; indeed, while the AS is suspected to have warmed over the past few decades and exhibits strong interannual variability over the (limited) observational record (Dutrieux and others, Reference Dutrieux2014), cold-regime shelf seas have not (although there is a significant freshening trend in many locations (Jacobs and others, Reference Jacobs, Giulivi and Mele2002; Jullion and others, Reference Jullion2013; Schmidtko and others, Reference Schmidtko, Heywood, Thompson and Aoki2014)).
Atmosphere-ocean general circulation models (AOGCMs) constitute the primary tool for projections and are required to relate global changes to local bottom water temperature, however, they cannot currently resolve all the small-scale processes driving ASBW properties (e.g. Arthun and others, Reference Arthun, Holland, Nicholls and Feltham2013; St-Laurent and others, Reference St-Laurent, Klinck and Dinniman2013; Dinniman and others, Reference Dinniman2015; Kjellsson and others, Reference Kjellsson2015; Nakayama and others, Reference Nakayama, Timmermann, Schröder and Hellmer2014). Understanding biases and spread in AOGCMs is thus critical, both for developing projections that account for uncertainty as well as informing downscaling techniques that relate large-scale changes in ocean hydrography to basal melting.
Other studies have assessed AOGCM ocean warming projections of relevance to Antarctica. Yin and others (Reference Yin2011) employed 19 CMIP3 AOGCMs, finding a mean circum-Antarctic warming of ~0.5°C at 200–500 m depth. Levermann and others (Reference Levermann2014) use CMIP5 output to calculate regional 21st century warming over four oceanic sectors; the full AOGCM range is then used with additional models to relate ocean temperature to ice mass balance. Neither of these studies investigates biases or meridional variability in warming. Sallée and others (Reference Sallée2013) use a potential-density based water mass definition to examine CMIP5 southern ocean biases and warming. Although the comparison with a depth-based analysis is indirect, their CDW class shows ensemble mean warming (<0.5°C) that is comparable with the CMIP3 ensemble (Yin and others, Reference Yin2011). They also note the presence of a warm bias (~ 0.4°C) in the ensemble mean CDW. However, it is unclear whether this bias or warming is coherent near the coast. Others have analyzed sea floor temperature (Heuze and others, Reference Heuze, Heywood, Stevens and Ridley2013, Reference Heuze, Heywood, Stevens and Ridley2015), but bathymetry varies across AOGCMs and continental shelves in some regions may not be resolved, making the relevance of these projections to ice shelves unclear.
In this paper, we: (1) develop a AOGCM-based metric of coastal subsurface ocean temperature that is relevant to ice shelves and allows comparison with observations, (2) quantify the spread in bias and warming in the CMIP5 ensemble using that metric; and (3) examine the relationship of regional and large-scale warming within AOGCMs and across the ensemble. The results provide a baseline for assessing the utility of coarse-resolution AOGCMs for ice sheet mass balance projections and revisit some of the implicit assumptions in previous work (e.g. averaging regions and spatial correlation).
METHODS
AOGCM output
We analyze temperature biases and future warming using an ensemble of 19 CMIP5 AOGCMs (Table 1; see also http://cmip-pcmdi.llnl.gov/cmip5/). Each of these AOGCMs simulates the mid-19th century to 2005 climate forced by a common evolution of historical atmospheric composition. Here, we analyze potential temperature (‘thetao’) fields over 1986–2005 (the baseline period) and anomalies from this baseline. We examine two future (2006–2100) warming pathways forced with: RCP 2.6, a pathway, which implies dramatic emissions reductions in the 21st century; and RCP 8.5, a high-end business-as-usual emissions pathway. For each RCP, we use a single realization (‘r1i1p1’) from each AOGCM. CMIP5 ocean models use a variety of curvilinear grids, which we interpolate to a common 1° × 1° grid for gridded analyses shown here. To analyse biases and warming at the coast, we use the native grid, as discussed in the following section.
List of model simulations, and the circum-Antarctic root-mean-square ASBW bias and warming, for each AOGCM included in the ensemble. All use the ‘r1i1p1’ realization. Warming for each simulation is 2080–2099 mean relative to the 1986–2005 mean. All values in °C
Biases over the baseline period are presented relative to the World Ocean Atlas, 2013 edition (WOA13; Locarnini and others (Reference Locarnini2013)) on a 1° × 1° grid. Although there are apparent differences in the WOA13 and recent cruise-based and moored measurements (e.g. in the AS), it is unclear whether this result is due to missing data or the interpolation scheme. Our limited analysis of supplementary observational datasets is supported by our results showing AOGCM bias is large relative to potential errors in the observational dataset.
ASBW temperature metric
We select 273 equivalently spaced points along the Antarctic coastline as defined in the BEDMAP2 project (Fretwell and others, Reference Fretwell2013) (filled circles in Fig. 1a). The ASBW temperature of each CMIP5 AOGCM (and the WOA13) is the temperature of the deepest grid cell at the location – on each AOGCM's native grid – that is closest to each coastal point, subject to two conditions (required due to the widely varying coastline and bathymetry at the AOGCM resolution (Fig. 1b)). These conditions are: (1) a 400 m ‘ceiling’; and (2) a 600 m ‘floor’, based on the typical depth range of the continental shelf break. If the depth of the closest grid point is shallower than 400 m, the next closest grid point is used; if the closest grid point exceeds 600 m depth, ASBW is defined to be the mean potential temperature between 400 and 600 m. Our analysis focuses on spatial averages of ASBW along coastal points in the AS (red points in Fig. 1a), Weddell Sea (WS, green points) and Ross Sea (RS, blue points).
(a) Antarctic coastline from the BEDMAP2 dataset, with ice shelves shown in light grey. 273 coastal points are shown with circles; red, green, and blue points correspond to the AS, WS and RS sectors, respectively. Orange star is the point corresponding to x = 0 in subsequent plots. Distance increases clockwise along the coast. (b) Depth of sea floor at the closest native grid point from coastal locations. Black dashed line is from the WOA13 observations. Grey lines correspond to each of the 19 CMIP5 models. Solid black line is the CMIP5 ensemble mean.
RESULTS
Ensemble results
The CMIP5 ensemble mean Southern Ocean – south of 60°S and between 400 and 600 m depth – shows a cold bias (Fig. 2a; area-average = −0.31°C). Assuming this depth range represents CDW, this result might be viewed as contrasting with Sallée and others (Reference Sallée2013), who find a slight warm bias. However, Sallee and others' definition of CDW is based on water mass properties at 30°S, and biases are assessed with respect to the entire CDW volume. Although there are alternate explanations (e.g. the set of AOGCMs included in the ensemble), we suggest that our results are not inconsistent, given the meridional variability in the ensemble mean bias as shown in Figure 2a.
(a) Ensemble mean temperature bias (1986–2005) relative to WOA13 depth-averaged over 400–600 m. Contours show the ensemble standard deviation. (b) ASBW temperature along the Antarctic coast. Black dashed line is from the WOA13 observations. Grey lines correspond to each of the 19 CMIP5 models. Solid black line is the CMIP5 ensemble mean.
The zonally consistent cold bias does not persist at the coast (Fig. 2b), which is characterized by alternating warm and cold ASBW biases. AOGCM biases vary within the regions, which we analyze later in the paper, particularly the AS; indicating that water mass biases (and warming) are likely to be moderated by regional averaging.
As noted by Yin and others (Reference Yin2011) and Sallée and others (Reference Sallée2013), projected Southern Ocean warming is muted closer to Antarctica (Fig. 3). In the ensemble mean, for both RCPs, offshore warming is enhanced in the Weddell Gyre relative to the Pacific Ocean and Ross Gyre. The ensemble spread, however (shown with the contours), is highest in the RS. The greater warming in the 400–600 m depth range in the Weddell Gyre is not evident in ASBW (Figs 3c, d), suggesting coastal processes, perhaps relating to sea ice, isolate and shield the coast. Along the coast, ensemble mean 21st century ASBW warming projections are quite uniform: approximately ~0.25°C for RCP 2.6 for all three regions, increasing to ~0.75°C in RCP 8.5. The ensemble mean warming pattern obscures, however, AOGCM-specific differences in the ASBW warming pattern evident in Figures 3c, d and discussed in the following section.
(a) RCP 2.6 and (b) RCP 8.5 warming depth-averaged over 400–600 m (2080–2099 minus 1986–2005 baseline). Contours show the ensemble standard deviation. (c) RCP 2.6 and (d) RCP 8.5 ASBW warming along the Antarctic coast. Grey lines correspond to each of the 19 CMIP5 models. Solid black line is the CMIP5 ensemble mean.
AOGCM-by-AOGCM results
In Figure 4, biases and warming for the three coastal seas are analyzed by AOGCM. When averaged across the coastline, ASBW biases are far larger in the AS (17th–83rd percentile range of −1.2–0.56°C) and RS (0.13–1.17°C) than in the WS (−0.11–0.38°C). The ensemble distribution is positively skewed in the RS and WS, driven by the floor imposed by the surface freezing point. All regions are characterized by outliers, which exhibit much larger biases than other AOGCMs; the MRI-CGCM and CNRM-CM5 set the high end of bias for the RS and AS (MRI-CGCM is also high in the WS); while the GISS-E2-R is the coldest model in all three seas.
ASBW temperature bias and warming, by model, averaged over the three sectors indicated in Figure 1. Stars indicate the ensemble median, bars correspond to the 17th–83rd percentile range.
For RCP 2.6, the warming in all three seas shows a similar distribution across AOGCMs, with a median of ~0.25°C and no obvious outliers. The mean, median and inter-model variance in ASBW warming increase for RCP 8.5 (17th–83rd percentile range of 0.35–0.98°C in the AS, 0.06–1.94°C in the RS, and 0.02–1.18°C in the WS). The projections are positively skewed, with the median projections lower than the mean as shown in Figure 3. Despite the lower RS median warming of only ~0.2°C, the RS is more vulnerable to high rates of 21st century warming than other coastal seas, with an 83rd percentile warming of almost ~2.0°C (compared with ~1°C in the AS and WS). In the AS and WS, models are approximately uniformly distributed within the central range, except for a few outliers, whereas the RS has three clusters of models.
Models with a large present-day bias tend to be outliers in the magnitude of coastal warming for RCP 8.5. MRI-CGCM, in particular, is at the low end of projections for the RS and AS; CNRM-CM5 has a significant positive bias in both and shows a high ASBW warming in the RS; GISS-E2-R has a significant negative bias and the highest warming in the AS.
The relationship of warming in different coastal seas across models is more evident in Figure 5. The MRI-CGCM and CNRM-CM5, indicated with circles, are not included in the regressions due to their significant biases, which we assume are related to the circulation and are a dominant influence on their climate response. When these outliers are removed, there is very little correlation between temperature biases and warming (not shown).
Pairwise scatterplots of warming (2080–2099 minus 1986–2005 baseline) over different regions for each of 19 models in the ensemble. ASBW corresponds to the values in Figure 4. Southern Ocean and global mean warming is a 0–700 m average. RCP 2.6 is in blue; RCP 8.5 is in red. Numbers show the r 2 for a linear fit that excludes outliers (shown with open circles and discussed in the text).
Regional correlations in subsurface warming around Antarctica are dependent upon the regions considered and the imposed climate forcing. Across AOGCMs, warming in the Ross and Weddell Seas is strongly correlated for both RCPs (r 2 >0.65). While RS and AS warming is correlated (r 2 = 0.54) in RCP 2.6, the linkage is weak for RCP 8.5; RS ASBW warming is much larger (for four AOGCMs) relative to the AS under the stronger forcing. The strongest relationship between large-scale warming south of 60°S and ASBW warming is in the AS, but this is only evident for RCP 8.5 (r 2 = 0.61). ‘Cold regime’ ice shelves exhibit weaker relationships with the Southern Ocean (r 2 < 0.26). There is limited evidence for ensemble-wide correlations between regional ASBW warming and large-scale warming, either in the global mean or south of 60°.
DISCUSSION
By highlighting biases and warming in Antarctic shelf water masses, and investigating the correlation between regional and large-scale warming, these results can guide hypotheses regarding physical mechanisms underlying AOGCM warming patterns and their use in ensemble projections. These considerations are discussed separately below.
Warming mechanisms
Many processes, both local and non-local, have been found to control the subsurface heat balance of the Antarctic continental shelf. The focus has principally been on the winds: easterlies near the coast, which regulate mixing, transport and the depth of the pycnocline near the shelfbreak (Dinniman and others, Reference Dinniman, Klinck and Hofmann2012; Stewart and Thompson, Reference Stewart and Thompson2015); and the strength and position of midlatitude westerlies, which regulate the large-scale water mass properties and position (horizontal and vertical) of relevant water masses (Sallée and others, Reference Sallée2013). CMIP5 models exhibit a robust strengthening and southward shift in westerlies (Bracegirdle and others, Reference Bracegirdle2013). This mechanism might be particularly relevant to warming in the AS, which is in close proximity to the ACC. However, Southern Hemisphere winds, and the position of the AS Low (Raphael and others, Reference Raphael2015), are strongly governed by coherent modes of atmospheric variability, in particular, the Southern Annual Mode (SAM). The strong, robust, signal of an increasing SAM index in climate models (Gillett and Fyfe, Reference Gillett and Fyfe2013), particularly with stronger (RCP 8.5) forcing, may underlie subsurface warming, evident here across all regions. Consistent with this large-scale linkage, Spence and others (Reference Spence2014) have shown that a southward shift of the midlatitude westerlies weakens Antarctic easterlies and enhances on-shelf transport (Stewart and Thompson, Reference Stewart and Thompson2015).
We look for signatures of these processes in vertical profiles of upper ocean warming in Figure 6. MRI-CGCM and HadGEM2 (shown with dashed grey lines) show surface-intensified warming, which differs greatly from the ensemble in the RS and AS. Ignoring these models, the vertical distribution of warming – but not its magnitude – is fairly consistent across the ensemble, with a limited degree of surface warming in the mixed layer, a subsurface minimum and a relatively monotonic increase below ~100 m depth. Temperature profiles suggest the persistence of strong stratification (at least in the annual mean); warming, with the exception of the AS, is strongly confined to the deeper water masses. In addition to supporting our choice of the ASBW metric, the bottom-intensified warming, at rates that are often greater than at comparable depths off-shelf, suggests an increase (although model-specific) of on-shelf transport. Such a circumpolar response is consistent with a large-scale change in surface stress. However, it does not exclude a role for sea ice or related processes on the RS or WS, which may moderate or govern their distinct response.
Vertical profiles of warming in the (a) AS, (b) RS and (c) WS. Ensemble mean RCP 8.5 warming is shown with the red line, individual AOGCMs are shown in grey, and dashed lines show outlier models, described in the text. Vertical profiles of potential temperature for the AOGCM ensemble mean for 1986–2005 (black solid line), 2080–2099 (red solid line), and WOA13 (dashed black line) in the (d) AS, (e) RS and (f) WS.
A more detailed mechanistic understanding of individual AOGCM warming is difficult to extract from this broad analysis. As a starting point, subsequent analyses might use changes in sea ice formation rates, salinity and/or seasonality to reveal whether changes in temperature are forced by changes in HSSW properties or CDW flux. Because atmospheric, sea ice and oceanic processes are tightly coupled in these regions (see, for example, the explanations of coastal change in Timmermann and Hellmer (Reference Timmermann and Hellmer2013), de Lavergne and others (Reference de Lavergne, Palter, Galbraith, Bernardello and Marinov2014) and Bintanja and others (Reference Bintanja, van Oldenborgh, Drijfhout, Wouters and Katsman2013), which invoke complicated feedbacks between atmospheric modes, freshening, sea ice, convection, stratification and the ice sheet), there is a need for a comprehensive assessment of the coupled system described by AOGCMs (Bracegirdle and others, Reference Bracegirdle2015).
Implications for ensemble projections
Our results highlight three priorities for ensemble projections of ocean temperature and/or melting designed to provide climatic boundary conditions for ice-sheet models, ensembles, or intercomparisons (e.g. Bindschadler and others, Reference Bindschadler2013; Nowicki and others, Reference Nowicki2013; Levermann and others, Reference Levermann2014) – capturing coastal warming, including regional and global linkages, and assessing high-end AOGCMs in the RS.
Capturing coastal warming
As noted above, there are substantial challenges involved in the representation of Antarctic coastal processes in AOGCMs. However, capturing ‘far-field’ warming or averaging over large coastal and/or offshore regions (Yin and others, Reference Yin2011; Sallée and others, Reference Sallée2013; Levermann and others, Reference Levermann2014) is also problematic: there are continental shelf processes (e.g. buoyancy fluxes, winds and coastal currents) that are at least partially represented in AOGCMs. Here, these processes, or their representation in large-scale models, drive significant differences in biases and warming from that occuring in offshore regions. Furthermore, the meridionally tilt of isopyncals in the Southern Ocean, the (possibly related) meridional gradient in biases and warming evident in Figures 2a, 3a, c, and the large differences in regional coastal warming rates, implies that a single depth range (even one that varies as a function of ice shelf or continental shelf depth) may not be indicative of water masses relevant to ice shelves. Finally, we note that AOGCMs' coarse resolution may influence model solutions well offshore – for example, via the parameterization of eddies and wind stress in regulating the thermocline depth and heat and volume transport (Screen and others, Reference Screen, Gillett, Stevens, Marshall and Roscoe2009; Farneti and others, Reference Farneti, Delworth, Rosati, Griffies and Zeng2010; Spence and others, Reference Spence2014; Griffies and others, Reference Griffies2015).
Although continental shelves are narrow and subject to high along-coast variability, these issues are mitigated when averaging over large areas of the continental shelf. We thus suggest that it is important to restrict the northern boundary over which warming is assessed, and to use the native grid and bathymetry, so as to capture an AOGCM-specific representation of continental shelf. Such a strategy does, however, subject projections to the representation of coastal processes, which deserves further investigation.
Including regional and global linkages
In both RCPs, the RS and WS are characterized by a (correlated) model-dependent response that is largely unrelated to the global climate. We suspect these seas are influenced more by common physical processes, as discussed in the previous section (e.g. sea-ice, easterly wind changes and/or freshwater flux) than by regional circulation or global climate changes. An important question is whether the correlated warming across these large continental shelves is due to: (1) a common climate process (e.g. the important role of sea ice in their hydrography (Nicholls and others, Reference Nicholls, Østerhus, Makinson, Gammelsrød and Fahrbach2009)) or (2) the treatment of these processes in models (Kjellsson and others, Reference Kjellsson2015).
In contrast, the warming of ASBW around coastal Antarctic seas is weakly correlated between the AS and the other two seas. As might be expected from observations, in which open ocean water masses are present on this shelf, CMIP5 models indicate that the strongest relationship between ASBW and the large-scale ocean temperature is in the AS, although the relationship decays substantially north of 60°S. The theoretical basis for this is clear, with less extensive sea-ice cover and a likely connection to large-scale wind position and strength (Bracegirdle and others, Reference Bracegirdle2013; Raphael and others, Reference Raphael2015) affecting both the temperature of the water masses and on-shelf transport.
Figure 5 shows almost no correlation between global and coastal ocean warming across the ensemble, which might suggest that we cannot downscale Antarctic coastal warming from global climate properties. However, such an analysis blurs the relationship between an AOGCM's global heat uptake and the regional warming associated with a given heat uptake. In Figure 7a, we compare the full time series (1860–2100) of global and regional warming for each AOGCM for the three seas. Within most models, a linear function of a model's global ocean warming seems reasonably predictive of its regional warming, with internal variability about these linear relationships (most evident in the AS, but also evident amid the high RS warming in CNRM-CM5 and HadGEM2-ES). However, the slope of the linear relationship varies widely across models, as summarized in Figure 7b. Many of the models have weak warming in both the AS and the RS relative to their global ocean warming (low slopes for both seas), but there are also a number of models that have low AS warming but a wide range of higher RS warming (between 0.5 × and 2 × the rate of global ocean warming). Despite the wide spread among models, the linear relationships between the regional and global ocean warming suggest that a global energy balance model could predict the regional warming with reasonable skill by linearly downscaling its ocean heat projections, with some natural variability superimposed. There will be a different linear relationship for each AOGCM, which introduces multi-model uncertainty in the downscaling procedure.
(a) Pairwise scatterplots of annual mean warming over the 1860–2100 period (relative to 1986–2005 baseline) in each region (y-axis) relative to the global mean 0–700 m warming (x-axis) for each of the 19 AOGCMs (RCP 8.5 simulation). Red, green, and blue points correspond to the AS, WS and RS sectors. (b) Scatter plots of the slope of the linear fit to the timeseries shown in (a) for each model across the three sectors. Numbers above the histograms are the mean/median/standard deviation of the linear fit slopes across the ensemble.
Assessing high-end AOGCMs
This ensemble analysis indicates that ASBW in the RS is more vulnerable to high rates of warming than either the AS or WS. This contrasts with Hellmer and others (Reference Hellmer, Kauker, Timmermann, Determann and Rae2012), which indicates that the RS is less vulnerable to dramatic warming than the WS (note, however, that six AOGCMs included in this analysis exhibit WS warming of >1°C, and that Timmermann and Hellmer (Reference Timmermann and Hellmer2013) suggest evidence for rapid RS warming as well). While this finding merits further study of these individual models, it also may arise from general limitations of coarse resolution models near sharp subsurface temperature gradients. In the Southeast Pacific, the Antarctic Circumpolar Current intrudes relatively close to the coastline of Antarctica, with the AS exposed to relatively unmodified CDW. In contrast, the nearby RS exhibits only a mid depth intrusion of highly modified CDW (Loose and others, Reference Loose, Schlosser, Smethie and Jacobs2009), but is separated from warmer subsurface water masses by only a O(10 km) shelfbreak front (Whitworth and Orsi, Reference Whitworth and Orsi2006). In contrast, the WS is shielded by the Antarctic Peninsula and the expansive Weddell Gyre, and its thermal structure appears to be better represented at the coast (excluding the MRI-CGCM and GFDL-CM3) (Fig. 6f).
Of the AOGCMs which show high rates of warming in the RS two – the MRI-CGCM and CNRM-CM5 – show severe warm biases (Fig. 8). In the MRI-CGCM, the bias is largest in the RS, potentially due to a poor representation of bathymetry or wind stress curl (Bracegirdle and others, Reference Bracegirdle2013; Hosking and others, Reference Hosking, Orr, Marshall, Turner and Phillips2013), which allows offshore water masses to intrude onto the continental shelf; in the CNRM-CM5, the bias appears to be widespread in the Southern Ocean. These biases appear to drive to a muted (in the case of the MRI-CGCM) or dramatic (in the case of CNRM-CM5) warming in RS ASBW.
Temperature biases, depth-averaged over 400–600 m, in individual models with anomalous 21st century RS warming (cooling in the MRI-CGCM). Top 2 are ‘outliers’ in RS temperature bias; bottom 3 lie closer to the observations, and the ensemble mean bias.
The mechanism underlying high RS ASBW warming in the other three models – CCSM4, GFDL-ESM2G, and HadGEM2-ES – is less clear. One possible explanation is a fresh (and warm) bias on the continental shelf. In the RS (Fig. 6e), the AOGCM ensemble mean does not reflect the layered structure and HSSW evident in observations. The lack of HSSW may result from limited resolution, via its effects on coastal winds and sea ice production, or problems with sea ice physics (Timmermann and Hellmer, Reference Timmermann and Hellmer2013; Turner and others, Reference Turner, Bracegirdle, Phillips, Marshall and Hosking2013; de Lavergne and others, Reference de Lavergne, Palter, Galbraith, Bernardello and Marinov2014; Dinniman and others, Reference Dinniman2015; Kjellsson and others, Reference Kjellsson2015; Shu and others, Reference Shu, Song and Qiao2015). A light bias, independently or combined with changes in coastal buoyancy fluxes driven by freshening and/or sea ice trends, might predispose an AOGCM to the erosion of density fronts over the 21st century (Hellmer and others, Reference Hellmer, Kauker, Timmermann, Determann and Rae2012; Timmermann and Hellmer, Reference Timmermann and Hellmer2013; Kjellsson and others, Reference Kjellsson2015).
There remain other biases in AOGCMs that may influence ASBW properties, in the RS and elsewhere, over the century timescale. Regional freshwater fluxes from ice shelf and iceberg melting are systematically underestimated and may be poorly distributed, the circulation is influenced by small-scale dynamic barriers, and eddies and associated transports are not resolved (Arthun and others, Reference Arthun, Holland, Nicholls and Feltham2013; St-Laurent and others, Reference St-Laurent, Klinck and Dinniman2013; Cullather and others, Reference Cullather, Nowicki, Zhao and Suarez2014; Meijers, Reference Meijers2014; Nakayama and others, Reference Nakayama, Timmermann, Schröder and Hellmer2014; Shu and others, Reference Shu, Song and Qiao2015). Furthermore, on-shelf transport is sensitive to the representation of the bathymetry. Figure 1 indicates significant spread and a shallow bias on much of the Antarctic continental shelf. We suggest, consistent with Nakayama and others (Reference Nakayama, Timmermann, Schröder and Hellmer2014) and Hellmer and others (Reference Hellmer, Kauker, Timmermann, Determann and Rae2012) that inadequate representation of bathymetric details is likely to underlie much of the systematic ASBW biases – both warm and cold – revealed in this analysis.
CONCLUSIONS
We have used a 19-member CMIP5 ensemble to calculate biases and 21st century warming of Antarctic Shelf Bottom Water, defined in a manner that accounts for coarse AOGCM resolution. We find ensemble mean warming by 2080–2099 in the AS, RS and WS of 0.66, 0.74 and 0.58°C, respectively, for high-emissions RCP 8.5. RCP 2.6 ensemble mean projections are substantially lower: 0.21, 0.26 and 0.19°C. Across the ensemble, there is a strong, scenario-independent correlation between warming in the WS and RS. As forcing increases, the AS warming becomes less strongly correlated with the other two seas and more correlated with the Southern Ocean south of 60°S. The RS has the largest spread in its 21st century warming projections, with a 1σ range of ~ 0–2°C. In almost all AOGCMs, warming is bottom-intensified, suggesting a change in the large-scale winds induces a change in on-shelf transport that may be moderated locally by processes related to sea ice production, or their representation in AOGCMs.
For fully-coupled simulations, large temperature biases may make it difficult to achieve a realistic initial ice-sheet state for future simulations. When constructing uncoupled, or ‘offline’, forcing scenarios, these results suggest that: (1) obvious biases should be used to exclude and/or deemphasize outlier models; (2) differences in near-coastal warming rates should be addressed, either using a sampling strategy (as employed here), or a separate parameterization; (3) RS and WS warming should be treated as correlated in ensemble projections; and (4) linear global/regional relationships suggest some predictability using a global energy balance model if it accounts for inter-AOGCM uncertainty in the linear relationship.
With the exception of a few models, there is no obvious relationship between the pattern of warming and biases. However, we suggest a deeper investigation is warranted, particularly in AOGCMs with enhanced RS subsurface warming. For future projections, it is worth considering exclusion, weighting, and/or bias correction of obvious outlier models; these AOGCMs have a meaningful quantitative impact on results if they are considered in projections. However, in keeping with the suggestions of Bracegirdle and others (Reference Bracegirdle2015), if metrics are to be employed for new intercomparsions and projections (e.g. CMIP6 and ISMIP6), they should be examined with a coupled, large-scale, perspective.
ACKNOWLEDGEMENTS
C.M.L is grateful for discussions with Laurie Padman, and financial support from NSF-PLR Award #1513396. This research was also supported by the Regional and Global Climate Modeling program of the US Department of Energy Office of Science, as a contribution to the HiLAT project. The authors would like to thank the BEDMAP2 project and the developers of the Bedmap2 Toolbox for Matlab, and the NOAA Geophysical Fluid Dynamics Laboratory for data and analysis tools. We acknowledge the World Climate Research Programme's Working Group on Coupled Modeling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1) for producing and making available their model output. The U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support for CMIP and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals.
Open access
