1. Introduction

Ice shelves and ice sheets evolve in response to changes in oceanic and atmospheric boundary conditions. Recent dramatic losses of long-lived ice shelves in both hemispheres highlight the uniqueness of recent climate change and have focused attention on causal factors that promote ice shelf retreat. Ice shelves respond dynamically and thermodynamically to changes in underlying ocean temperature (Holland et al. 2008a) and surface air temperature (SAT). Several authors (e.g., Mercer 1978; Vaughan and Doake 1996) have linked the observed poleward migration of annual or monthly SAT isotherms to episodes of ice shelf retreat, indicating a strong correlation between SAT and the sudden breakup of existing ice shelves. While the Larsen B ice shelf was preweakened by at least a decade of thinning largely resulting from ocean-induced basal melting (Shepherd et al. 2003) prior to the final collapse [an event that Domack et al. (2005) suggested was unprecedented for Larsen B during the Holocene], the final sudden disintegration event (along with that of Larsen A) has been attributed to warming SAT, which led to development of surface melt ponds and ice fracturing of existing crevasses (van den Broeke 2005; Scambos et al. 2000, 2003). The Wilkins ice shelf, which currently undergoes extensive summer melting, also experienced a breakup event associated with the 2008 melt season that Scambos et al. (2009) suggested was triggered by meltwater-induced crevasse deepening. This breakup differs from large wintertime breakup loss events over the Wilkins ice shelf, which Braun et al. (2009) attributed to basal melt-induced bending stresses. In the Northern Hemisphere, ice shelf breakup events associated with the generation of significant summer surface melt have been observed on northern Ellesmere Island, where the Ward Hunt and Ayles ice shelves retreated (2002–09 and 2005, respectively) (Mueller et al. 2003; Copland et al. 2007). These examples indicate that increased SAT over an ice shelf that is structurally vulnerable because of internal glacial discontinuities provides one important mechanism for rapid ice shelf retreat or loss.

The stability of remaining Antarctic ice shelves has implications for West Antarctic Ice Sheet (WAIS) stability, sea level rise, and bottom-water production. The loss of the Larsen B ice shelf resulted in an increased flow velocity of several grounded inland source glaciers (Scambos et al. 2004). Similarly, present-day thinning of the grounded Pine Island Glacier has been attributed to changes in the rate of ice shelf basal melting (Payne et al. 2004). It has been suggested that loss of the Ross and/or Ronne-Filchner ice shelves would also accelerate flow of the marine-based WAIS (Mercer 1978), potentially resulting in large-scale WAIS retreat. Naish et al. (2009) and Pollard and DeConto (2009) interpreted sediment core results and an associated ice sheet–modeling simulation to suggest that periodic loss of the Ross ice shelf occurred during the warm, early Pliocene ∼5−3 × 10⁶ yr before present, indicating that it is sensitive to changes in environmental conditions that are similar to those projected for the coming centuries. Arctic ice shelves, in contrast to their southern counterparts, do not buttress significant grounded ice.

Recent Arctic and Antarctic air temperature changes have been attributed directly to human influence (Gillett et al. 2008). Continuing evolution of high-latitude climate in response to anthropogenic forcing may significantly affect the stability of remaining ice shelves in both hemispheres. We utilized a climate model forced by anthropogenic emissions of carbon dioxide (CO₂) to simulate large-scale surface warming trends. Model output was analyzed for periods of extended summer warmth to infer the potential for melting of the magnitude that is currently associated with observed ice shelf retreat. Although we focused on regions of the earth where ice shelves occur, the global model domain also allowed us to analyze melting trends over present-day grounded ice sheets.

The model output was first compared with observed SAT, an ensemble of atmosphere–ocean general circulation model (AOGCM) simulations of the twenty-first century and the timing of major ice shelf retreats and was then integrated for several hundred years into the future to explore the potential for increased melting over ice surfaces in both hemispheres.

2. Model and methods

Version 2.8 of the University of Victoria Earth System Climate Model (UVic ESCM) was used. This intermediate-complexity climate model couples atmospheric, oceanic, and land surface components, a thermodynamic/dynamic sea ice model, and a closed and coupled carbon cycle that includes carbon transfer between inorganic and organic carbon reservoirs. The ocean component is a full 3D ocean general circulation model [Modular Ocean Model, version 2 (MOM2); Pacanowski 1995]. The atmospheric component consists of a vertically integrated energy–moisture balance model in which the radiative forcing associated with atmospheric CO₂ is applied through a decrease in outgoing longwave radiation, parameterized as

View Expanded

where C(t) is the atmospheric CO₂ concentration (ppmv) at time t and F₀ = 5.35 W m⁻² corresponds to a specified radiative forcing of 3.7 W m⁻² for a doubling of atmospheric CO₂ (Houghton et al. 2001). Surface air temperature is prognostically calculated as a function of the surface energy balance and elevation (via a constant lapse rate). Ice sheets are static, noninteractive components of the model and are reflected in global elevation and albedo fields. Surface albedo fields are a function of surface type such as plant functional type, ice, and snow cover. While potential melt ponding plays an important role in the energy balance over ice (via a lowered surface albedo), we could not resolve this effect because of a lack of a detailed surface hydrological scheme. This made our subsequent determination of SAT change conservative by removing a positive feedback on SAT.

Heat and moisture fluxes between the ice shelves and the ocean were neglected in this version of the model. Heat flux from the ocean to ice shelves is proportional to the temperature difference between the pressure-dependent melt temperature of ice and the in situ temperature of seawater at the ice–ocean interface and is expected to increase in a quadratic fashion as the ocean warms (Holland et al. 2008b). However, UVic ESCM simulations indicated that the contribution of this flux to the total Southern Ocean heat budget is still small relative to heat fluxes to (from) the ocean resulting from sea ice formation (melting) and open-ocean heat exchange with the atmosphere. We therefore assume that ice shelf–to-ocean heat flux plays a small role in the large-scale SAT, which is the field in which we are primarily interested. The resolution of all of the components within the model is 3.6° longitude by 1.8° latitude.

The UVic ESCM reproduces the observed climate, including the zonal-mean amplitude of seasonal SAT change (Randall et al. 2007), and falls within the midrange of the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report model ensemble predictions of the twenty-first century (Meehl et al. 2007). Because of computational efficiency, the model is able to simulate climate over millennial time scales, making it ideal for exploring long-term effects of anthropogenic emissions that cannot be addressed using full AOGCMs. A detailed description of the model and its validation is described in Weaver et al. (2001).

Two simulations were done. Both were initialized from a preindustrial equilibrium at year 1850 in which atmospheric CO₂ was stable at approximately 280 ppmv. They were then integrated forward for 1000 yr until year 2850. In the first simulation, termed the “high-emission scenario,” observed emissions were applied until year 2000 (Marland et al. 2008). From 2000 to 2100, emissions followed the IPCC A2 scenario (Meehl et al. 2007). From 2100 to 2300, emissions decreased linearly back to zero. This emission profile added 5134 Gt of carbon to the climate system and corresponds closely to combustion of all of the present-day oil, natural gas, and coal resource bases [including nonconventional sources such as tar sands (Rogner 1997)]. The second simulation, in which CO₂ was capped at 390 ppmv from 2010 onward, was undertaken to simulate warming to which the climate system is already committed, because Weaver et al. (2007) found that reduction below present-day CO₂ concentrations within the next millenium is unrealistic if emissions are reduced by anything less than 90% of present-day levels by 2050. This simulation is termed the “commitment scenario.”

An anomaly approach was used to obtain SAT fields. This method superimposed the temperature trend simulated by the UVic ESCM on the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) climatology (Kalnay et al. 1996). This removed potential systematic SAT biases within the raw simulated UVic ESCM SAT field. Daily average SAT for the period of 1970–2001 was generated for NCEP–NCAR data and the UVic ESCM simulation. The final SAT was then calculated as

View Expanded

where SAT_raw is the raw simulated time-dependent SAT and SAT_UVic and SAT_NCEP are the 1970–2001 daily average UVic ESCM and NCEP–NCAR SAT fields, respectively. The 30-yr-average period removed high-frequency variability from the NCEP–NCAR and UVic ESCM climate that would otherwise obscure the long-term evolution of SAT trends. Bromwich et al. (2007) analyzed NCEP–NCAR data over high-latitude locations. They found large biases in presatellite NCEP–NCAR data over the winter periods at high latitudes, particularly over Antarctica. However, biases were much reduced during the summer when melting occurs.

To infer regions that experience summer melt periods of the same magnitude as those observed in conjunction with observed ice shelf retreat, the annual sum of positive degree-days (PDD), which is assumed to be proportional to the annual melt, was recorded throughout the simulation for each grid cell following the method of Vaughan (2006). Based on observations, Copland et al. (2007) estimated that the contour marking 200 PDD yr⁻¹ defines the limit of ice shelf stability for the northern Ellesmere Island ice shelves. Vaughan (2006) analyzed the distribution of PDDs over the Antarctic Peninsula. The 200 PDD yr⁻¹ contour of his analysis lies across the Larsen A/B ice shelves in 2000, which approximates the period during which both shelves underwent persistent large-scale surface melt and retreat. A limit of 200 PDD yr⁻¹ was therefore utilized as the primary threshold, with a reasonable range of values explored to determine the sensitivity of the ice shelf stability zone to this value.

3. Results

The simulated preindustrial steady-state climate exhibited two broad regions in each hemisphere in which less than 200 PDD yr⁻¹ occurred. In the Arctic, this zone encompassed northern Ellesmere Island (the location of the Ward Hunt and Ayles ice shelves), the Greenland Ice Sheet (GIS), and a region of the offshore Arctic Ocean (Fig. 1). In the Antarctic, the entire continent and nearby Southern Ocean fell within the low summer melt zone. This agrees with established ice shelves here prior to significant human-derived emissions.

In the Arctic, by 2011–21, the criterion of 200 PDD yr⁻¹ was surpassed in the ocean off Ellesmere Island (Fig. 2). When the threshold was reduced to 160 PDD yr⁻¹ the ocean off Ellesmere Island surpassed the threshold in 2001–11. This result bracketed the timing of ongoing ice shelf loss in this region remarkably well within the framework of a 1000-yr simulation and reinforced the 200-PDD limit estimate of Copland et al. (2007).

On the Antarctic Peninsula, high summer melting (based on the 200 PDD yr⁻¹ criterion) migrated over the Larsen A/B ice shelf region in the 2061–71 period in the high-emission scenario. This lies in contrast to observations, which indicate an excess of 200 PDD yr⁻¹ on the Antarctic Peninsula next to the Larsen A/B shelves by year 2000. This discrepancy was due to an underestimation of the observed northern Antarctic Peninsula warming trend in 1960–2000 (Chapman and Walsh 2007) by the UVic ESCM. Elsewhere, the PDD trend bias was smaller. Grid cells immediately next to the Wilkins ice shelf experienced warmth in excess of 200 PDD yr⁻¹ in 2011–21. Tedesco (2008) determined an average annual Antarctic melt extent of 8.6 × 10⁵ km² for 1987–2008. This melt extent was not surpassed in the high-emission scenario simulation until 2063 (although by 1997, the simulated melt extent of 3.0 × 10⁵ km² was within the measured range of melt extents). This delay was largely attributable to the underestimated warming over the Antarctic Peninsula, although an additional important cause of the low simulated melt extent was the coarseness of the UVic ESCM grid, which simply did not resolve narrow coastal regions of melt that contributed to the melt extent of Tedesco (2008). A comparison of simulated rates of PDD increase to observed rates is given on Table 1.

To gauge further the accuracy of high-latitude warming simulated by the UVic ESCM in the near future, the rate of warming for 2000–2100 under the high-emission scenario was compared with a five-model ensemble of AOGCMs [Canadian Centre for Climate Modelling and Analysis (CCCma) Coupled General Circulation Model, version 3 (CGCM3), Geophysical Fluid Dynamics Laboratory Climate Model, version 2 (GDFL CM2), ECHAM5-Max Planck Institute Ocean Model (MPI-OM), Community Climate System Model, version 3 (CCSM3), and third climate configuration of the Met Office Unified Model (HadCM3)] that was also forced with the IPCC A2 scenario for the 2000–2100 period (Meehl et al. 2007). The UVic ESCM rate of warming over the Antarctic Peninsula and the Antarctic coastal warming trend corresponded well to the ensemble average, but the rate of warming over the interior of Antarctica (including the WAIS) and the GIS was slightly lower [∼0.1°C (10 yr)⁻¹] than the ensemble average. This indicated that polar amplification of warming in the UVic ESCM, while present, was less pronounced than in the IPCC ensemble.

Overall, the simulations captured both preindustrial regions of ice shelf stability and the introduction of significant summer melting over regions characterized by recent ice shelf losses (with a particular delay in southward migration of the high-melt zone based on the 200 PDD yr⁻¹ criterion across the western Antarctic Peninsula). The underestimated rate of warming over the Antarctic Peninsula, combined with the slightly lower average rate of warming at each pole (relative to the IPCC ensemble), suggested that melt extent and PDD-per-year values obtained by the UVic ESCM simulations represented lower bounds on the rate of poleward migration of high-melt zones over the coming centuries.

Further integration of the high-emission scenario to 2850 resulted in a global annual average SAT increase of ∼7°C. Temperatures were still slowly increasing at the end of the simulation, despite a total cessation of emissions at 2300. Polar amplification of SAT changes resulted in Arctic temperature increases of up to 12°C, whereas in the Antarctic several coastal regions warmed by up to 10°C. In the Arctic, warming occurred rapidly; in the Antarctic, major warming was delayed until approximately 2100–2300.

In the Arctic, by 2100, widespread surface melting was simulated over all regions that presently contain ice shelves (Fig. 1). Further expansion of the high-melt zone resulted in marked summer melt over most of the GIS by 2850, with only a small central core remaining below the 200 PDD yr⁻¹ threshold. In Antarctica, by 2100, several isolated regions of the Antarctic Peninsula began to experience summer melting above the 200 PDD yr⁻¹ criterion, while the WAIS and EAIS remained below the 200 PDD yr⁻¹ threshold. However, between ∼2050 and ∼2500, regions encompassing the Ronne-Filchner, Ross, and Amery ice shelves experienced a rise in PDDs, such that large portions of these regions surpassed the 200 PDD yr⁻¹ threshold by year 2500 (for representative example locations, see Fig. 3). By the year 2850 the entire Ronne-Filchner and Ross ice shelves and the northern Amery ice shelf were all well above the 200 PDD yr⁻¹ threshold (Fig. 1). In addition, significant parts of the WAIS poleward of the Ronne-Filchner and Ross ice shelves and the Antarctic Peninsula moved into the high-melt region (Fig. 1). In contrast, the EAIS did not experience significant summer melting of the ice sheet interior, with the exception of sporadic coastal cells and a small area around Prydz Bay. The total simulated extent of Antarctic melt had expanded by roughly a factor of 10 from the simulated and observed present-day extent, to 3.6 × 10⁶ km².

In contrast to the high-emission scenario, the commitment scenario simulation exhibited less encroachment of melting at high latitudes. By 2500, global average SAT had equilibrated at ∼0.6°C above 2010 levels. Although the Arctic Ocean and surrounding regions still migrated almost completely into the high-summer-melt zone for all criteria, the major Antarctic ice shelves and the WAIS, all of which experienced melting in excess of 200 PDD yr⁻¹ in the high-emission scenario, now remained well below the 200 PDD yr⁻¹ criterion (Figs. 2 –4) for the commitment scenario. Although the Wilkins region experienced melting slightly in excess of 200 PDD yr⁻¹ after 2020, simulated PDD values over the Larsen region remained low. This indicated that warming in this region was too modest for the commitment scenario simulation and was consistent with the anomalously late warming experienced in the high-emission scenario: under modest CO₂ increases, the UVic ESCM (along with the IPCC ensemble) underestimated the warming over the western Antarctic Peninsula.

4. Discussion

The empirical climate threshold of 200 PDD yr⁻¹ used here to delineate regions of high summer melting was chosen to reflect SAT conditions closely associated with observed melt-induced fracturing of ice shelves and was meant to extrapolate the SAT–ice shelf loss correlation of Vaughan and Doake (1996) into the future using climate model output instead of observations. However, it is important to note that oceanic heat flux is an alternate or additional thermodynamic mechanism for driving ice shelf loss (Holland et al. 2008b). Several ice shelf retreats have been attributed at least in part to subshelf melting. Furthermore, future Antarctic ice shelf loss resulting from undershelf warming and/or increased injection of warm Circumpolar Deep Water could result in ice shelf loss that was not captured by the SAT-derived ice shelf stability inference used here.

Several first-order trends in the simulated future evolution of summer melt zones at each pole, as represented by the 200 PDD yr⁻¹ threshold, were notable. Warming trends resulted in high-summer-melt periods over the entire Arctic Ocean by 2100, even if CO₂ concentrations were capped at 2010 levels. In addition, when the model was forced by emissions equivalent to all of the present-day fossil-fuel resource base, most of the GIS experienced periods of extended summer surface melting by 2500, in agreement with other modeling studies (Ridley et al. 2005; Charbit et al. 2008). Ongoing work incorporating an integrated mass balance/dynamic ice sheet model into the coupled climate model indicates that this intensity of melting results in annually averaged net ice loss from the GIS. Increased flow of meltwater into the GIS via moulins would warm interior ice through both advection of heat and latent heat release during refreezing and would contribute toward lubrication of the ice sheet bed. However, explicit analysis of this effect was beyond the scope of this study.

The high-emission scenario resulted in extended summer surface melt periods over the Ross, Ronne-Filchner, and Amery ice shelves (as well as lower-elevation portions of the WAIS) by year 2850, with extensive summer melt beginning at ∼2100–2300. This result suggested that, in the absence of a mechanism that reinforces ice shelf stability (i.e., reduced subshelf melting), summer melting could begin to degrade the integrity of these large shelves where zones of existing surface glacial discontinuities exist. It is unlikely that ocean temperatures under the ice shelves will decrease, thereby counteracting the potential effect of melt-induced weakening. On the contrary, the simulated ocean immediately offshore of the Ross and Ronne-Filchner ice shelves between 300- and 1400-m water depth warmed by 2.2° and 2.6°C, respectively, over the course of the high-emission scenario simulation, providing an additional source of heat for basal shelf melting. Introduction of significant simulated summer melting over the WAIS implied that in the future the WAIS surface may resemble that of the GIS, with a similar accumulation of summer melt ponds, moulins, and associated meltwater transfer to the ice sheet bed.

Recent analysis of sediment cores obtained from beneath the Ross ice shelf combined with ice sheet modeling (Pollard and DeConto 2009) revealed cyclical retreats or even collapses of the Ross ice shelf and the WAIS during the warm early Pliocene, resulting in an open Ross Sea (Naish et al. 2009), despite CO₂ concentrations of only ∼400 ppmv. This implies that the WAIS was sensitive to retreat under climatic conditions that are expected to occur in coming decades. Vaughan and Doake (1996) suggested that a SAT warming of 10°C above present-day temperatures could threaten the Ross and Ronne-Filchner ice shelves. This magnitude of warming was simulated in response to combustion of all of the present day fossil-fuel resource base but was notably absent over the large Antarctic ice shelves and the WAIS if CO₂ was capped at 2010 concentrations. This indicates that at least two important climatic thresholds—that of large-scale Antarctic ice shelf degradation and significant WAIS surface melting—could be crossed if human-derived CO₂ emissions are not reduced.