By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
 Observations indicate that dynamic mass loss from the Antarctic ice sheet has been accelerating over recent decades, leading to a freshening of the Southern Ocean. Here we quantify the effect of several rates and spatial distributions of freshwater forcing on simulated sea ice area trends. In contrast to a previous study, our simulations show that the freshwater effect on sea ice trends over the historical period is small and fails to reproduce the observed regional pattern of trends, when using observationally consistent rates of freshwater forcing. The Coupled Model Intercomparison Project Phase 5 (CMIP5) models do not represent this dynamic ice sheet mass loss, and it has been suggested that this deficiency may significantly influence the simulated sea ice trends. We show that when accounting for internal variability, the average CMIP5 sea ice area trend is statistically consistent with the observed trend, and accounting for ice sheet derived freshwater forcing has little influence.
 Various mechanisms have been proposed to explain the observed increase in Antarctic sea ice area since 1979 (Figure 1a). Wind changes have been identified as a particularly important driver of regional Antarctic sea ice trends [Holland and Kwok, 2012], though the respective roles of internal variability and anthropogenic forcing in driving these wind-induced trends is unclear [Bitz and Polvani, 2012; Sigmond and Fyfe, 2010; Turner et al., 2009].
 Freshwater forcing of the surface Southern Ocean is also known to drive sea ice growth by increasing the oceanic stratification and reducing the vertical heat flux [Swingedouw et al., 2008; Aiken and England, 2008; Stouffer et al., 2007; Hellmer, 2004]. Both precipitation [Liu and Curry, 2010] and ocean-sea ice feedbacks [Zhang, 2007] have previously been implicated in driving historical sea ice trends. Recently, the effect of observed dynamic mass loss from the Antarctic Ice Sheet (AIS) [Rignot et al., 2011; King et al., 2012; Shepherd et al., 2012] has also been suggested as a significant driver of historical Southern Ocean freshening and sea ice trends [Bintanja et al., 2013; Jacobs and Giulivi, 2010].
 In contrast to the observations, the CMIP5 multimodel ensemble mean shows a significant decline in Antarctic sea ice (Figure 1a) given time-evolving changes in ozone (tropospheric and stratospheric), greenhouse gases, aerosols (sulfate, black carbon, and organic carbon), land use (e.g., deforestation), solar variability, and volcanic activity. The CMIP5 models simulate wind changes [Swart and Fyfe, 2012], ocean-sea ice feedbacks, and increased Southern Ocean precipitation [Fyfe et al., 2012]. However, the coupled ice sheet dynamics required to simulate the retreat of marine ice sheets responsible for most of the recent AIS mass loss [Joughin et al., 2012; Pritchard et al., 2012] are ubiquitously absent from the climate model simulations contributing to CMIP5, even though the models may prognostically compute surface mass balance and meltwater runoff.
 It has been proposed that the resulting absence of dynamic ice sheet derived freshwater forcing in the CMIP5 models could be at least partly responsible for the discrepancy between the simulated and observed historical sea ice area trends [Bintanja et al., 2013]. To test this hypothesis, we begin by comparing the CMIP5 sea ice trends with observations and then we employ the UVic Earth System Climate Model (ESCM) to quantify the impact of observationally constrained ice sheet derived freshwater forcing on simulated Antarctic sea ice area trends since 1979. Finally, we investigate the role of internal variability in explaining the discrepancy between simulated and observed trends.
2 Sea Ice Trends in the CMIP5 Multimodel Ensemble and Observations
 The CMIP5 multimodel ensemble mean shows a large negative trend in annual mean sea ice area of −3.0×1011 m2/decade over the historical period (Figures 1a and 1b). By contrast, the observations show a statistically significant positive trend of 1.39±0.82×1011m2/decade (95% confidence interval accounting for serial correlation).
 In the observations, the positive sea ice area trend represents a near cancelation between regions of sea ice gain in the Ross and Weddell Seas and regions of sea ice loss in the Amundsen and Bellinghausen Seas (Figure 2a). The CMIP5 ensemble mean does not capture the observed regionality of sea ice trends but rather exhibits a near-uniform rate of sea ice loss around Antarctica (Figure 2b). The CMIP5 sea ice trends could be significantly influenced by including the effects of dynamic ice sheet derived freshwater forcing, which we investigate next.
3 Sea Ice Response to Ice Sheet Mass Loss
 Several recent studies using surface mass balance and gravimetry have concluded that the Antarctic ice sheet as a whole has been losing mass since the observations began in 1992 [Rignot et al., 2011; King et al., 2012]. The mass loss has predominantly occurred from the West Antarctic Ice Sheet (WAIS) and Antarctic Peninsula (APIS), which are estimated to have been losing mass regionally since about the early 1970s [Rignot et al., 2008]. However, the rate of mass loss was disputed until the reconciled estimates of Shepherd et al.  who reported mass loss for the combined WAIS and APIS as 46±49 Gt yr−1 over 1992 to 2000, increasing to 138±28 Gt yr−1over 2005 to 2010 (Figure 3a).
 In order to span the range of the significant observational uncertainties in the absolute magnitude of dynamic ice sheet mass loss (Figure 3a), as well as in the timing of mass loss initiation, we conducted two ensembles of experiments using the UVic ESCM: (1) the ensemble FW74, with five different rates of mass loss acceleration in which forcing began in 1974 and (2) the ensemble FW92 with ten different rates of acceleration and with forcing initiation in 1992 (see supporting information). To account for uncertainties in the geographical location of the freshwater input, each ensemble was run for two different distributions of freshwater forcing. In the first, the fresh water was evenly distributed to the surface grid cells adjacent to the Antarctic shelf (FW circumpolar), while in the second, all the fresh water was added in the surface coastal grid cells corresponding to Amundsen Bay (FW Amundsen) near 105°W (see Figure S1 in the supporting information). The experiment without such additional dynamic freshwater forcing is used as a control. Note that the prescribed FW forcing in our experiments represents the additional forcing from ice sheet dynamics not simulated by the UVic model (or CMIP5 models). The control and all ensemble members principally experience freshwater input around Antarctica from river runoff prognostically computed by the land surface scheme with a rate of between 2200 and 2400 Gt yr−1.
 The circumpolar distribution of ice sheet derived FW forcing reduced simulated sea ice loss over the historical period, relative to the control (Figures 3b and 3c). The freshwater forcing acts to increase the Southern Ocean stratification, thereby reducing the vertical oceanic heat flux and sea ice basal melt which leads to sea ice growth, as previously documented [Bintanja et al., 2013; Aiken and England, 2008; Stouffer et al., 2007; Hellmer, 2004] (see supporting information).
 However, even the strongest rates of FW forcing we applied could not reverse the negative sea ice trend induced by the radiative forcing. Sea ice loss was reduced by 64% by 2020 relative to the control in the strongest case of the FW92 ensemble with circumpolar FW distribution (Figure 3c). But for the FW scenario closest to observed rates of mass loss (FW92 #3), sea ice loss was only reduced by 11% relative to the FW control in 2020. For the FW74 ensemble with an earlier onset, sea ice loss was reduced by between 6% and 50%, with the lower end forcing scenarios being the most observationally consistent.
 If we consider linear least squared trends, the FW forcing reduced the negative sea ice trend relative to the control by 0.09–0.84×1011m2/decade for the FW74 ensemble and 0.08–0.78×1011m2/decade for the FW92 ensemble over the period 1979–2010. For the observationally realistic low-end cases, the effect is small relative to observed sea ice area trend of 1.39×1011m2/decade. Furthermore, the effects are even smaller when the FW forcing is applied in Amundsen Bay, rather than in the circumpolar distribution (see Table S2).
 Our UVic model control run did not reproduce the observed spatial pattern of sea ice trends but showed a near-uniform loss around the circumpolar region (Figure 4a), as did the ensemble mean of the original CMIP5 models. The strongest FW forcing experiment (FW74 #5) did produce areas of positive sea ice concentration trends. However, the FW forcing fails to produce the strong positive trends observed in the Ross Sea region for either the FW circumpolar or the FW Amundsen distribution of release (Figures 4b and 4c). Rather, the FW forcing tends to drive sea ice growth in the Amundsen Sea, where observations show significant sea ice loss. Not only are the positive regional trends driven by the FW forcing in our experiments in the wrong location but also they are significantly smaller than the observed trends, especially in the observationally realistic low-end scenarios (not shown).
 Thus, the FW forcing effect that we have quantified using the UVic ESCM is very small (<0.21×1011 m2/decade) relative to the CMIP5 ensemble mean trend (−3.0×1011 m2/decade). We conclude that accounting for FW forcing is unlikely to influence the CMIP5 sea ice trend distribution. To explain the discrepancy between observed and simulated sea ice trends, we now examine the role of internal climate variability.
4 Internal Variability in CMIP5
 The histogram of CMIP5 sea ice area trends shows that there is an extremely large spread across the individual members of the CMIP5 ensemble (Figure 1b). The trends span an order of magnitude from −16.0 to 7×1011 m2/decade. The spread of trends reflects a difference in the response to historical forcing between models, and the influence of internal variability within individual models runs. This internal variability must be accounted for when comparing simulated and observed trends.
 We can use the distribution of sea ice trends across the individual members of the CMIP5 ensemble to quantify the likelihood of simulating the observed positive sea ice trend. Internal variability is taken into account using multiple realizations from each model, where available. A test of the null hypothesis that the CMIP5 ensemble mean and observed sea ice area trends are equal can be done by making the assumption that the model trends are exchangeable and by using a Monte Carlo based resampling technique to construct an empirical distribution for the model trends [Fyfe et al., 2013] (see also the supporting information for details on the technique). Then, this empirical distribution can be used to determine whether the observed trend falls within the 5% to 95% range of individual model trends, where internal variability in the models has been taken into account.
 The null hypothesis that the model mean and observed trends are equal is just accepted at the 10% significance level (p=0.051). Thus, it appears in this test using all available models, and accounting for internal variability, that the average CMIP5 model sea ice area trend is marginally statistically consistent with the observed trend. This applies to the trend in annual mean sea ice area and is consistent with other recent studies [Polvani and Smith, 2013; Zunz et al., 2013]. We have not tested the role of internal variability (or intermodel variability) in explaining discrepancies in regional sea ice trends or trend seasonality between the models and observations.
5 Discussion and Conclusions
 We have quantified the effect of freshwater forcing of the surface Southern Ocean from dynamic mass loss from the Antarctic ice sheet. Bintanja et al.  have recently suggested that this freshwater forcing could significantly influence Antarctic sea ice area trends. However, they only considered a large and constant rate of freshwater forcing of 250 Gt yr−1applied for 40 years. In our experiments, which have an accelerating rate of freshwater forcing consistent with reconciled observational estimates [Shepherd et al., 2012], there is only a small influence on simulated historical sea ice trends, with a reduction in ice area loss of about 10%.
 The ice sheet derived freshwater forcing has a strong regional pattern, with the major source of forcing in the West Antarctic [Rignot et al., 2008], but the ice sheet derived freshwater flux could be injected somewhat northward into the Southern Ocean and distributed by icebergs [Silva et al., 2006]. We tested the influence of two different spatial distributions of freshwater input, one with a distribution focused in Amundsen Bay and the other with a uniform circumpolar distribution of freshwater seaward of the Antarctic shelf. In our simulations, neither distribution of freshwater forcing was able to reproduce the observed positive sea ice trends in the regions of the Ross and Weddell Seas. Our results suggest that accounting for ice sheet derived FW forcing is unlikely to significantly improve the simulated spatial pattern of sea ice trends.
 Our freshwater forcing was applied at the ocean surface, but in reality, the fresh water is likely input at depths of up to several hundred meters where basal ice sheet melt occurs [Joughin et al., 2012]. Wind forcing is also known to strongly influence the spatial distribution of sea ice trends [Holland and Kwok, 2012; Simpkins et al., 2012; Lefebvre and Goosse, 2008; Liu et al., 2004], but in this investigation, we held the winds fixed to isolate the FW forcing effect. The influence of these additional factors needs to be further quantified, in particular, because the CMIP5 models are known to have systematic wind biases in the Southern Ocean [Swart and Fyfe, 2012].
 The absence of freshwater forcing from dynamic ice sheet mass loss is a feature of the CMIP5 ensemble. The CMIP5 multimodel ensemble mean produces a negative sea ice area trend in contrast to the observed positive trend and also does not reproduce the observed regional pattern of trends. We have shown that the effect of dynamic ice sheet derived freshwater forcing, with a magnitude constrained by observations, has little effect on Antarctic sea ice trends simulated by the CMIP5 models over the historical period. However, our statistical test which accounts for internal variability and uses all available CMIP5 runs suggests that the CMIP5 sea ice area trends are consistent with the observed trend at the 10% significance level.
 N.C.S. was supported by the NSERC CREATE training programme in interdisciplinary climate science and the University of Victoria. Compute Canada/Westgrid is acknowledged for providing computing resources. We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups for producing and making available their model output. We thank Jeremy Fyke, Bill Merryfield, Michael Sigmond, and a reviewer for their helpful comments on the manuscript.
 The Editor thanks Matthew England and an anonymous reviewer for their assistance in evaluating this paper.