The Greenland Ice Sheet's surface mass balance in a seasonally sea ice-free Arctic

Authors


Abstract

[1] General circulation models predict a rapid decrease in sea ice extent with concurrent increases in near-surface air temperature and precipitation in the Arctic over the 21st century. This has led to suggestions that some Arctic land ice masses may experience an increase in accumulation due to enhanced evaporation from a seasonally sea ice-free Arctic Ocean. To investigate the impact of this phenomenon on Greenland Ice Sheet climate and surface mass balance (SMB), a regional climate model, HadRM3, was used to force an insolation-temperature melt SMB model. A set of experiments designed to investigate the role of sea ice independently from sea surface temperature (SST) forcing are described. In the warmer and wetter SI + SST simulation, Greenland experiences a 23% increase in winter SMB but 65% reduced summer SMB, resulting in a net decrease in the annual value. This study shows that sea ice decline contributes to the increased winter balance, causing 25% of the increase in winter accumulation; this is largest in eastern Greenland as the result of increased evaporation in the Greenland Sea. These results indicate that the seasonal cycle of Greenland's SMB will increase dramatically as global temperatures increase, with the largest changes in temperature and precipitation occurring in winter. This demonstrates that the accurate prediction of changes in sea ice cover is important for predicting Greenland SMB and ice sheet evolution.

1 Introduction

[2] Mass loss from the Greenland Ice Sheet (GrIS) has been accelerating over the last two decades [Rignot et al., 2011; Shepherd et al., 2012]. The ice sheet is currently contributing nearly 1 mm yr−1 to sea level rise. Mass balance estimates, using the Gravity Recovery and Climate Experiment and the RACMO/GR regional climate model (RCM), found that the increase in mass loss for the period 2003–2008 was equally split between surface processes (such as runoff and precipitation) and ice dynamics, indicating the importance of both components in the evolution of the ice sheet [van den Broeke et al., 2009].

[3] This period of mass loss from the GrIS is congruent with a dramatic retreat of Arctic sea ice extent, the rate of which has accelerated since the 1990s [Serreze et al., 2007; Stroeve et al., 2011]. Arctic sea ice retreat is thought to provide a major contribution to the positive trend in lower tropospheric Arctic temperature, which is amplified with respect to the global mean [e.g., Serreze et al., 2009] and is associated with the GrIS mass loss, and has been linked to increased autumn and winter precipitation in Europe and the northeastern and midwestern United States [Liu et al., 2012]. Interestingly, sea ice extent achieved a record minimum in 2012, which coincided with an extreme melt event over Greenland extending as far as Summit Camp [Nghiem et al., 2012]. The decline in end of summer sea ice extent is expected to continue until the Arctic Ocean is seasonally practically ice free, potentially within years to decades, but winter sea ice decline is not as dramatic and is likely to persist well into the future [Boe et al., 2009; Wang and Overland, 2009]. A small amount of summer ice is expected to persist in the Canadian Archipelago well into the 21st century, so in the rest of this document, the term “ice free” is used to refer to states where sea ice extents of less than 1 × 106 km2 exist in the Arctic.

[4] Simulations of future mass balance of the ice sheet have predominantly been conducted with general circulation models (GCMs) [e.g., Ridley et al., 2005; Gregory and Huybrechts, 2006; P. Fitzgerald, University of Bristol, unpublished data, 2012]. Over the 21st century, the GCMs predict an increase in both temperature and precipitation over Greenland, and it is expected that the impact of increasing temperatures will dominate [e.g., Gregory and Huybrechts, 2006]. The GCM projections used in these studies, however, do not successfully capture the rapid decline in summer sea ice extent seen in the observations [Stroeve et al., 2007], nor do many possess the observed sensitivity of Arctic sea ice to changes in global temperature [Winton, 2011] and do not, therefore, adequately account for the impact of this on Greenland surface mass balance (SMB). This indicates that some GCMs, which simulate the correct change in global temperature, will simulate less sea ice loss and models simulating the correct sea ice loss may overestimate the global temperature change. It is unclear, as a consequence, what the impact on Greenland SMB will be from a rapidly declining sea ice cover. This study will investigate if moisture exchange is likely to dominate over heat exchange in terms of its impact on SMB and determine the importance of the seasonality of the changes. The simulations of the late 20th and early 21st centuries included as part of the Coupled Model Intercomparison Project (CMIP) 5 capture the loss of ice better than the CMIP3 ensemble [Stroeve et al., 2012] but, to the authors' knowledge, have not yet been used to assess future ice sheet mass balance.

[5] Investigating the climatic impact of a seasonally ice-free Arctic Ocean and the influence to the GrIS from observations alone is difficult for the following reasons: (a) the transition to a seasonally ice-free state is not complete, (b) there is confounding presence of internal variability, and (c) the observational record contains signals caused by forcing other than that by sea ice [Deser et al., 2010]. To avoid these difficulties, we use an atmosphere-only general circulation model (AGCM) to simulate the impact of reduced sea ice cover on the GrIS climate. Various previous studies have used global AGCMs to investigate the impact of prescribed changes in sea ice cover on atmospheric conditions [Budikova, 2009]. Deser et al. [2010] forced an AGCM with 1980–1999 sea surface temperatures (SSTs) and 2080–2099 sea ice conditions, and found that the surface energy flux response was largest in winter and smallest in summer since the temperature gradient across the sea surface is largest (smallest) in winter (summer) and accounts for most of the seasonal, spatial, and vertical structure of high-latitude warming present in the equivalent coupled simulations. They also predict that a significant increase in Arctic winter snow and precipitation accumulation will be associated with reduced Arctic sea ice cover. These findings are, broadly speaking, commensurate with the observed trends in reanalysis products which link the decrease in sea ice to enhanced moisture flux (evaporation) from the Arctic Ocean and increased Arctic lower troposphere moisture content and temperature [Serreze et al., 2009; Screen and Simmonds, 2010]. However, such a comparison is limited, since the observed record still has significant summer ice coverage.

[6] Singarayer et al. [2006] predict a similar increase in Arctic winter accumulation and further suggest that some areas of Arctic land ice may even undergo a net increase in SMB in the short term due to sea ice decline. This is consistent with the findings of Day et al. [2012] who simulated such conditions for the Svalbard Archipelago, by prescribing reduced sea ice conditions and holding SSTs at present-day values. Such an increase in SMB can occur when moderate increases in winter surface air temperature cause increased precipitation, but do not exceed the melting point of ice, and there is limited increase in summer surface air temperature. A similar response may occur for Greenland, but this is a significantly larger and higher elevation land mass and may be more isolated from changes in regional sea surface climate.

[7] Though there has been little investigation of the role of sea ice in the future evolution of Greenland's SMB, the importance of sea ice in Arctic climate more generally has received much attention [ACIA, 2005]. However, the multi-GCM study presented in the Intergovernmental Panel on Climate Change Fourth Assessment Report predicts an increase in annual mean precipitation over the Arctic of between 10% and 28% during the 21st century [Christensen et al., 2007, p. 857], which is likely to impact Greenland's SMB. This is in part due to warming global and Arctic temperatures throughout the atmospheric column, leading to an increase in moisture content and precipitation at high latitudes [Meehl and Stocker, 2007]. The other component is due to the increase in evaporative flux from the Arctic Ocean, particularly in winter, when the areas of open ocean are warm relative to the overlying atmosphere [Singarayer et al., 2006; Deser et al., 2010]. These changes are likely to be accompanied by a poleward shift in midlatitude storm tracks [Yin, 2005], which are an important, though highly uncertain, aspect of the 21st century climate change [e.g., Woollings et al., 2012].

[8] This study investigates what impact a seasonally ice-free Arctic Ocean will have on the seasonal temperature, precipitation, and GrIS SMB using an RCM to force an insolation-temperature melt (ITM) SMB model [e.g., Oerlemans, 2001; Robinson et al., 2010]. Here, we drive the surface boundary of the RCM with present-day SSTs and late 21st century sea ice concentrations to isolate separate factors of change [e.g., Stein and Alpert, 1993]. This factor separation method is used to attribute which changes in Greenland's climate and SMB are a result of changes in sea ice and which are the combined effect of increased global sea surface temperature. The results of this study may be used to interpret and, where appropriate, attribute changes in ice sheet surface climate and SMB to reduced sea ice cover.

2 Methods and Models

2.1 Methodology

[9] To investigate the role of SST and sea ice forcing on Greenland Ice Sheet SMB, it is necessary to simulate Greenland's surface climate under the three different SST and sea ice climatologies and use these surface climates to force an ice sheet SMB model. To this end, three sets of simulations were run; each set is composed of an integration with an AGCM, which is used to force an integration with an RCM, the surface climate variables from which were used to force the ITM SMB model. This chain of models is required in order to provide boundary conditions to each model in turn and simulate the regional climate and SMB of Greenland under each set of SST and sea ice conditions (see Figure 1).

Figure 1.

Flow diagram describing how each model was forced and in what direction the forcing happened. The arrows show the direction of forcing and the variables passed between models.

[10] A 50 km (0.44°) version of the UK Met Office's (UKMO's) RCM, HadRM3, was used for this study [Jones et al., 1995, 2004]. This model is used to downscale global climate from the equivalent global model, HadAM3, as it has been in numerous studies including the European Union-funded ENSEMBLES and ICE2SEA projects [e.g., Déqué et al., 2011; Rae et al., 2012]. The use of RCMs, coupled to a SMB model, in Greenland is increasingly common for the purposes of simulating SMB. The increased resolution provides more realistic representation of temperature and precipitation compared to GCMs which typically have a resolution of 100–400 km [e.g., Ettema et al., 2009; Mernild et al., 2010].

[11] The three sets of SST and sea ice conditions were derived and applied as follows. Both HadAM3 and HadRM3 were forced at their surface boundary by monthly mean sea surface temperature (SST) and sea ice fields, which were made by averaging over 30 year periods from two simulations conducted with the UKMO HadGEM1 global climate model [Johns et al., 2006]. These simulations were performed as part of the third Climate Model Intercomparison Project (CMIP3) and are available from the World Climate Research Program's (WCRP's) Coupled Model Intercomparison Project phase three (CMIP3) multimodel database [Meehl et al., 2007]. From this set, a historical run for the 20th century, forced with observed greenhouse gas (GHG) concentrations (20C3M) and the A1B future scenario, was used in this study. Nakienovic et al. [2000] set out scenarios of 21st century GHG emissions, of which A1B is the “business as usual” scenario (similar to RCP6.0 at the end of the 21st century [Knutti and Sedláček, 2012]), which causes a large change in sea ice extent, in HadGEM1, during the 21st century.

[12] The HadGEM1 output was used to create three sea ice concentration and SST climatologies: a present-day control set (20C3M), a set with A1B seasonally ice free but 20C3M SST (SI), and a set with A1B SST and sea ice (SST + SI). The time slices used to create the surface forcing for the models are as follows:

  1. [13] 20C3M: present-day control SST and sea ice (1961–1990),

  2. [14] SI + SST: A1B SST and sea ice (2061–2090), and

  3. [15] SI: A1B sea ice and 20C3M SSTs.

[16] In SI simulations the RCM and AGCM are forced with inconsistent fields, namely 20C3M SST and A1B sea ice. For points which are ice covered in the 20C3M sea ice field, but not in the A1B field, the SST value is undefined; we set these points to the freezing point of sea water (−1.8°C) similar to Deser et al. [2010]. The comparison between this and the SI + SST experiment, forced with A1B SST and sea ice, allows us to isolate the component of the climate change signal, which is attributed to sea ice alone from the coupled SST and sea ice signal simulated in the SI + SST experiment. In all integrations the HadAM3 and HadRM3 integrations had present-day greenhouse gas values (CO2 = 355 ppmv, CH4 = 1700 ppbv). These were held constant in the HadAM3 and HadRM3 integrations so that only changes in surface forcing differ between simulations.

[17] To calculate ice sheet surface mass balance, 30 years of surface temperature, precipitation, and surface shortwave radiation from the RCM were interpolated onto the grid of a 20 km resolution digital elevation model (DEM) derived from Bamber et al. [2001] and were averaged from 6-hourly values and updated daily in the ITM model. In addition, temperatures were lapse rate-adjusted from the RCM grid elevation to the DEM elevation using a constant lapse rate of −6.8°C km−1 [see Fausto et al., 2009]. The model was initialized with surface height from the original DEM and then spun up to equilibrium for 110 years by repeating the 30 year surface forcing before running for a further 30 years. The subsequent SMB analysis was performed on the last 30 years of each simulation.

[18] In section 3.1 the discussion of changes in climate seasonal mass balance uses the convention that winter and summer refer to the boreal winter and summer months, December–February and June–August, respectively. In section 3.2 winter and summer values for the SMB values and their components refer to the seasons of net accumulation (October–June) and ablation (July–September), respectively, as is more common in investigations of SMB.

2.2 HadAM3 and HadRM3

[19] Both HadAM3 and HadRM3 include the same model physics and only differ in some physical parameters [Pope et al., 2000]. Due to the physical similarity between these models, the large-scale climate on the HadRM3 grid closely follows the parent HadAM3 simulation (J. J. Day, University of Bristol, unpublished data, 2011). The models have time steps of 30 and 5 min for HadAM3 and HadRM3, respectively. HadRM3 is forced at its lateral boundary by potential temperature, total moisture, horizontal velocity, sea level pressure, and relative humidity by the HadAM3 driving simulation (see Figure 1). These are output from HadAM3 every 6 h and are temporally interpolated and updated at each HadRM3 time step. The surface boundary conditions for the RCM and AGCM are interpolated in time from the monthly input data and are updated every 5 days. For further details of the nesting technique, see Jones et al. [1997].

[20] Surface moisture and energy fluxes in HadRM3 (and HadAM3) are calculated with the Met Office Surface Exchange Scheme (MOSES). This is a surface hydrology model, which calculates moisture and energy fluxes between the atmosphere and four subsurface-surface levels. Wherever snow or ice is present, it is assumed to lie uniformly, changing the thermal conductivity of the surface and reducing surface roughness. Albedo varies between the snow-free value (as prescribed from the Wilson and Henderson-Sellers [1985] data set) and the maximum snow covered value, which is 0.8 for temperatures below −2°C. A complete description of this model subcomponent may be found in Cox et al. [1999]. However, the hydrological fluxes calculated by MOSES are not used as SMB estimates; rather, we use near-surface (1.5 m) air temperature, precipitation, and net downward shortwave radiation from the RCM to force the ITM SMB model (see Figure 1) as described in section 2.3.

2.3 Insolation-Temperature Melt Surface Mass Balance Model

[21] The annual SMB is calculated using a simple snowpack model as described by Robinson et al. [2010]. In this scheme, snow (hs) and ice (hi) thicknesses in meters water equivalent (m w.e.) are calculated on a daily basis by

display math(1)
display math(2)

where P is the total precipitation, Ms is the surface melt rate, and rf is the refreezing fraction. Snow cover thickness is limited to a maximum height of hs,max = 5 m, and any snow above this limit is added to the underlying ice height, hi, and hs is reset to 5 m.

[22] The refreezing fraction, rf, is zero in the absence of snow. When 0 < hs < 1 m, the refreezing fraction is defined according to Janssens and Huybrechts [2000] as

display math(3)

where f(T) is the snow fraction of the total precipitation and rmax (=0.6) is the maximum fraction of snow that is able to refreeze [Reeh, 1991]. For 1 < hs < 2 m, rf increases linearly reaching a value of 1 for snow thicknesses over 2 m.

[23] Surface melt, Ms, is calculated using an insolation-temperature method [e.g., Pellicciotti et al., 2005; van den Berg et al., 2008]. In this scheme daily melt is calculated as a function of daily temperature and incident shortwave radiation according to

display math(4)

where Δt is the time step (1 day); ρw is the density of ice; Lm is the latent heat of ice melt; αs is the surface albedo; c + λT is the sum of the longwave radiation balance and turbulent heat exchange, linearized around the melting point; λ is set to 10 W m−2 K−1 as derived by Oerlemans [2001]; and c (=−85 W m−2) is a free parameter and was chosen to lie within the range of values given by P. Fitzgerald, University of Bristol, unpublished data, 2012 and tuned to give a change in surface mass balance for the period of interest in line with previous studies. Inputs to the model are the incident shortwave radiation at the surface, SW, and 1.5 m temperature, T, both provided by interpolating output from HadRM3 onto the 20 km DEM grid (see Figure 1).

3 Results

3.1 Moisture and Energy Flux in the Arctic Ocean With a Reduced Sea Ice Cover

[24] The annual cycle of sea ice extent in the 20C3M simulation slightly overestimates the 1961–1990 mean from the HadISST observational record by ~8 × 105 km2 (Figure 2a) [Rayner et al., 2003]. In the future seasonally ice-free A1B state, the amplitude of the sea ice extent seasonal cycle is much larger, since the March (maximum) extent is reduced by ~3 × 106 km2 compared to 20C3M and the September (minimum) extent is reduced by ~8 × 106 km2, and both changes are significant at the 99% level. This is because during the 2061–2090 period of the HadGEM1 A1B simulation, conditions are warm enough to melt through almost all of the winter ice during the melt season, but in winter it is still cold enough for ice to form across the Arctic basin and beyond. This is consistent with observations, which have a strongly negative trend in September sea ice extent but a much smaller trend in March.

Figure 2.

Seasonal cycles of monthly mean (a) sea ice extent (from the HadGEM simulations and HadISST observations), (b) THF (latent + sensible), (c) longwave radiative flux, and (d) shortwave radiative flux. All energy fluxes are averaged over the Arctic Ocean (66.5–90°N) and are positive upward.

[25] In the following description of surface energy flux anomalies between simulations, we use the convention that all surface energy fluxes are positive upward. Analysis of the seasonal cycle of surface fluxes in the Arctic Ocean was performed for the area north of 66.5°N. The impact of changes to surface boundary forcing on the turbulent heat flux (THF = specific + latent heat) is similar across both the SI + SST and SI experiments, with the SI + SST fluxes increasing by 48% and SI by 55% in winter, with little (<7%) change from the 20C3M climate in summer (see Figure 2b). The winter anomalies are larger than those in summer since the atmosphere is at its coldest relative to the ocean at this time of year despite relatively moderate changes in sea ice cover during this season.

[26] Similar to Deser et al. [2010], we find that for both experiments the winter THF anomaly has a dipole pattern with positive anomalies of as much as 368 W m−2 for SI and 294 W m−2 for SI + SST, in areas of sea ice loss and with negative anomalies of a similar magnitude directly to the seaward side of the 20C3M ice edge (see Figure 3). In winter the area of open water seaward of the sea ice edge is a local maximum of the THF field, due to a high ocean-atmosphere thermal gradient in this region. Thus, sea ice retreat is coupled with a northward shift of this area of maximum flux [e.g., Alexander et al., 2004; Deser et al., 2010]. This pattern is evident in both SI + SST and SI simulations, but the SI value is larger because the overlying section of the troposphere does not warm by as much as the SI + SST simulation (see Figure 3); hence, the atmosphere in SI has a stronger thermal-height gradient than that in SI + SST in these regions (see Figure 3). This represents a significant northward shift of a major source of Arctic atmospheric moisture in winter. In the SI + SST and SI simulations, the sea ice has receded from the east Greenland coast, increasing the turbulent flux along a strip adjacent to the coast. The SI + SST simulation shows additional negative and positive THF anomalies in the mid-Atlantic, not present in SI, which are associated with the changes in SST in the region.

Figure 3.

Winter anomalies of sea ice concentration, THF, precipitation, and 1.5 m air temperature for SI + SST-20C3M and SI-20C3M. Anomalies which are not significant at the 95% level (using t test) are set to zero.

[27] These changes in the THF from the surface have a direct impact on winter precipitation in the Arctic. The anomaly is broadly similar across both simulations which both experience large increases in precipitation across the Arctic basin (see Figure 3). As one might expect, areas of increased precipitation correlate well with areas of increased THF, with precipitation more than doubling over the Barents and Kara Seas. Though the increase in precipitation is largest in the SI + SST simulation, both the SI + SST and SI anomalies have a similar pattern. Both simulations have increased precipitation over Greenland compared to 20C3M. In SI this increase is restricted to the central and east portions of the ice sheet and is caused by enhanced upwind evaporation in areas where sea ice cover is reduced but averages 11% across the ice sheet. The SI + SST experiment has increased precipitation everywhere on the ice sheet by an average of 39%. In this simulation warmer global temperatures and reduced katabatic winds lead to increased moisture transport to the ice sheet and hence increased precipitation; this is enhanced by the same sea ice-related increase in the eastern half of the ice sheet seen in SI. This causes the anomaly to be largest in east Greenland, particularly in the northeast.

[28] The winter near-surface air temperature anomaly of both experiments is larger than 24°C, over the areas of sea ice retreat in the Barents and Kara Seas. The largest temperature anomalies are in regions where the THF response is large, indicating that much of these changes in both experiments are the result of changes in surface energy flux (see Figure 2). In SI, the temperature anomaly mirrors that of sea ice. The winter temperature response over Greenland is small over most of the ice sheet in the SI simulation, but there are some large increases of more than 8°C near the northeast coast. In the SI + SST simulation most points on the ice sheet experience a temperature increase of between 4°C and 8°C with temperatures exceeding this in the northwest ice sheet in coastal areas adjacent to areas of sea ice loss.

[29] In summer, changes to the THF as a result of reduced sea ice in the SI simulation are small (<33 W m−2) (see Figures 2b and 4). However, larger changes in flux are present when SST increases are included in the SI + SST simulation, resulting in anomalies as large as (+)88 W m−2, in areas of sea ice retreat (e.g., Barents and Kara Seas). Anomalies are negative over areas of partial sea ice cover in both simulations due to a combination of reduced sea ice concentration and increased thermal gradient through the ice.

Figure 4.

Summer anomalies of sea ice concentration, THF, precipitation, and 1.5 m air temperature for SI + SST-20C3M and SI-20C3M. Anomalies which are not significant at the 95% level (using t test) are set to zero.

[30] In summer, changes in precipitation across the domain are significantly less than those in winter, indicating a strong seasonal dependence. The SI forcing results in reduced precipitation over a large southeast section of the ice sheet, with a maximum of 41% reduction near the coastline (Figure 4). Much of the precipitation in this region is the result of cyclone activity [Serreze and Barry, 2009], and the simulated reduction is driven by a reduction in cyclonic activity associated with a general decrease in the intensity of the North Atlantic storm track (from mean sea level pressure variance with a total Hoskins filter [Hoskins and Hodges, 2002], not shown). The western and northeast ice sheets experience an increase in precipitation with a maximum of 49% in the northeast. The increase in the northeast is the direct result of increases in turbulent heat flux (THF) associated with sea ice decline in the northeast Greenland coast, while the increase in west Greenland is associated with a southwesterly increase in the wind direction. In the SI + SST simulation, Greenland experiences a general increase in precipitation caused by the increase in temperature (via the Clausius-Clapeyron relation). This increase in precipitation is largest (178%) in northwest Greenland. This is associated with changes in large-scale moisture transport including an increase in poleward moisture transport (not present in SI). However, the footprint of the sea ice-driven changes is apparent in the SI + SST anomaly, which also has a significant reduction in precipitation (as large as 24%) over southeast Greenland associated with reduced cyclonic activity. The increase adjacent to the northeast Greenland coast is also significant and partly sea ice forced.

[31] The response of summer near-surface air temperature to the SI + SST forcing is as much as 8°C in the Barents and Kara Seas, smaller compared to that in winter because the temperature above sea ice is much warmer in summer than in winter, but the temperature of the underlying ocean does not vary seasonally by as much (Figure 2). The temperature anomaly of the SI simulation is much smaller than that of SI + SST, negative and isolated to areas where there is a change in sea ice cover. This is because the surface of sea ice when it is melting fixes the surface temperature to the melting point temperature (~0°C) in the 20C3M experiment, but in the SI experiment, wherever sea ice was present in the 20C3M sea ice field but not in SI + SST, the SST was set to the freezing temperature of sea water (−1.8°C). Again, the near-surface air temperature response mirrors that of the THF, indicating that changes in surface flux are more dominant than changes in global circulation. The near-surface air temperature response over Greenland is between 2°C and 4°C over most of the ice sheet in SI + SST and ~0°C in SI (see Figure 4).

3.2 Changes in the Surface Mass Balance of the Greenland Ice Sheet

[32] Before discussing the changes in SMB associated with the changes in climate discussed, it is important to assess the present-day SMB as simulated by the SMB model (described in section 2.2) when forced with the control (20C3M) experiment. The total precipitation simulated by the RCM over the ice sheet is 399 Gt yr−1, where 1 Gt is equivalent to 0.6 mm w.e. averaged over the ice sheet. This is smaller than other modeled and observation-based estimates probably due to the model's relatively low resolution (compared to other RCMs, e.g., RACMO2, 11 km) and biases in HadAM3 (see Table 1) [e.g., Ettema et al., 2009; Rae et al., 2012; Vernon et al., 2013]. However, the spatial pattern is in qualitative agreement with the in situ observations of Bales et al. [2009] and the modeled accumulation of Ettema et al. [2009] (see Figure 5). Annual melt and refreezing are within the published estimates, but with lateral compensation leading to significant spatial biases, especially in north Greenland, where melt is overestimated (e.g., compared to RACMO2, Rae et al. [2012]). Runoff is larger than other estimates, probably as a result of lateral biases in melt (see Table 1). The SMB is somewhat lower than other estimates due to the low accumulation rates simulated by the RCM, but both the spatial pattern of melt and SMB are qualitatively consistent with previous reconstructions [e.g., Ettema et al., 2009]. Though we are interested in the change in SMB rather than its absolute value, regional overestimates of melt and runoff are likely to result in high SMB sensitivity because of the melt-albedo feedback.

Table 1. Ranges of Observed and Estimated SMB Components From the Recent Studies [Ettema et al., 2009; SWIPA, 2009] and Values From the Present-Day Run in This Study in Gt yr−1
SourcePrecipitationMeltRunoffRefreezeSMB
Various literature550–743228–532213–28248–250264–469
This study3994243766722
Figure 5.

(top) Precipitation, (middle row) melt, and (bottom row) SMB for the (left) 20C3M simulation absolute values and anomalies of the (middle column) SI + SST (SI + SST-20C3M) and (right column) SI (SI-20C3M). Points which are stippled and encircled by a black line indicate areas where the anomaly is not statistically significant at the 95% level using a Student's t-test.

[33] Forcing the SMB model with the SI + SST climate, described in section 3.1, leads to a 119 Gt yr−1 reduction in SMB. This is composed of an increase in winter balance and a larger decrease in summer balance (see Figure 6). This change in SMB is of a similar order of magnitude to that of Fitzgerald et al. [2012], who use the same ITM model. The SI simulation leads to an increase of 63 Gt yr−1, with increased SMB in both winter and summer. Due to the artificial nature of the SI experiment, the meaning of the absolute value of the simulated SMB is unclear; however, comparing the SMB components in this simulation with the control and SI + SST simulations enables us to understand the impact of a reduced sea ice cover on future ice sheet SMB.

Figure 6.

Seasonal cycle of ice sheet-averaged SMB components. Black error bars show ±1 standard deviation about the 20C3M control.

[34] The SI + SST experiment's increase in winter balance is caused by a 52 Gt yr−1 (16%) increase in winter accumulation across the ice sheet. This is not mirrored in the SI simulation, where only a small increase in accumulation is simulated between February and March and a significant decrease from March to July (see Figure 6). Summer accumulation also increases in SI + SST, but by much less than that in winter, which results in increased runoff and refreezing during the ablation season. These changes culminate in an annual accumulation increase of 28% and 2% for SI + SST and SI, respectively (see Table 2). The increase in annual total accumulation in the SI + SST simulation is significant across the entire ice sheet, with increases of as much as 410 mm w.e. in the southern ice sheet and an increase of more than 250 mm w.e near the northeast coast on the Flade Isblink Ice Cap (see Figure 5). This increase in the northeast is also a feature of the SI simulation and is associated with reduced sea ice cover, whereas the increase over the ice sheet interior is due to increased temperature and humidity in the atmospheric column, permitting greater moisture advection to high elevations (Figure 5). In Greenland, precipitation rates are largest in the southeast, where much of the precipitation is driven by the orographic uplift of southeasterly air flow associated with cyclones [Serreze and Barry, 2009]. A reduction in summer cyclonic activity in the southern Greenland Sea in SI causes a reduction in precipitation in this region compared to 20C3M (see Figures 4 and 5). The SI + SST simulation has a smaller reduction in cyclone activity in this region and any change in associated precipitation locally seems to be balanced by other factors resulting in an increase in annual accumulation.

Table 2. Annual Mean Surface Mass Balance Components From the ITM-SMB Model for the 20C3M, SI + SST, and SI Experiments in Gt yr1 Where SMB = Precipitation − Runoffa, b
 20C3MSI + SSTSISI + SST%SI%
  1. aAll values are the average over the 30 simulation years.
  2. bN/A, not applicable.
Precipitation399512406128102
Melt42468937016387
Runoff37660832116285
Refreeze6713269197103
SMB22−9785N/AN/A

[35] Broadly speaking, the increase in accumulation in SI + SST is a direct effect of the increases in temperature and hence specific humidity in the atmospheric column above Greenland. As in the rest of the Arctic, this allows an increase in warm moist air advection onto the ice sheet. Comparing the SI + SST and SI simulations to 20C3M, the mean fractional change in precipitation versus change in temperature is 4.9% °C−1; this is similar to values in the coupled model comparison of Gregory and Huybrechts [2006].

[36] The SI + SST simulation has an ice sheet-averaged increase in summer temperature of 2.6°C compared to 20C3M, leading to a 265 Gt yr−1 (63%) increase in surface melt. The increase is statistically significant across the whole ice sheet, but the increase in melt is largest in the marginal areas (see Figure 5). This leads to increased runoff and contributes to a near doubling of refreezing within the snowpack. This increase in runoff is not balanced by the increase in accumulation, leading to a reduction in the SMB. The reduction in SMB is large in the marginal areas, with reductions of more than 800 mm w.e. in some grid boxes in the western ice sheet. This is contrasted with a pronounced increase of over 30 mm w.e. in the ice sheet interior. This leads to an increase in the equilibrium line altitude from 1550 to 1800 m. An increase of SMB greater than 150 mm w.e. is also predicted for the Flade Isblink Ice Cap in the northeast (see Figure 5).

[37] In the SI simulation, the increase in winter precipitation (discussed in the previous section) results in increased albedo throughout the year (see Figure 6). Coupled with a reduction in incident shortwave radiation at the surface, this results in a shorter ablation season and an ice sheet-averaged 13% reduction in surface melt and 15% reduction in runoff (Table 2). This reduction in melt is statistically significant across most of the ice sheet, but is largest in the ice sheet's low-elevation marginal areas particularly in the northeast where some grid cells experience an increase in precipitation of more than 300 mm w.e due to reduced sea ice in the region. It is interesting then to see that this results in a large increase in SMB in the northeast ice sheet where reduced sea ice causes increased accumulation. However, in the southeastern ice sheet, the reduced precipitation causes a significant reduction in SMB of as much as 130 mm w.e.

[38] The area-averaged SMB in SI is 62 Gt yr−1 larger than that in the 20C3M simulation, and changes in SMB are significant at almost all gridpoints. It is interesting to note that the high-elevation increases in SMB present in the SI + SST simulation are not present in SI. This indicates that any changes in sea ice forcing are isolated to the lower elevations of the ice sheet.

4 Conclusions

[39] By forcing a regional climate model with a reduced (A1B) sea ice cover and 20C3M SSTs, and also with A1B SST and sea ice, we were able to divide the impact of predicted 21st century climate change on the GrIS SMB into (1) the component due to global change and (2) the component due to sea ice decline alone. This revealed that the impact of the latter on the future of the GrIS SMB is significant and will act to increase SMB, but the regions of influence will be limited to near-coastal locations. This is in broad agreement with Hanna et al. [2009], who found that perturbing SST in the Modèle Atmosphérique Régional RCM only had a strong impact on precipitation and melt in low-lying coastal locations.

[40] In the artificial experiment where temperatures in the model were held fixed and seasonally sea ice-free climatology was prescribed, an increase in SMB is simulated. This is largely due to an increase in winter (December, January, February, March) accumulation in the vicinity of sea ice loss. Our results show that these sea ice-driven changes in accumulation are important in Greenland, indicating that some parts of the ice sheet may experience an increased SMB in the future, but ultimately increased summer melting dominates.

[41] The increased temperature and specific humidity over the GrIS in the simulation including A1B SST and sea ice were large and do create a significant increase in both accumulation and surface melt culminating in a decrease in the SMB. This follows the conventional wisdom, which suggests that temperature is the major driver of GrIS mass balance and that the positive accumulation response accompanying warming is lower in magnitude than the related melt, as indicated by proxy records [e.g., Alley et al., 1993, 2010]. However, it is interesting to note that this decrease is not present in all seasons; rather, the winter balance is higher, but the annual balance is dominated by a strongly negative summer balance. The SMB anomaly is also split into areas of thinning in the ice sheet's marginal areas and thickening in the interior, consistent with Krabill et al. [2000].

[42] The major areas of impact due to sea ice decline were located in the northeast and southeast. The simulations performed with the ITM SMB model indicate that in the northeast, increased precipitation associated with reduced sea ice will drive an increase in SMB of more than 0.15 m w.e. over the Flade Isblink Ice Cap. There is observational evidence to suggest such future behavior; in the present-day climate, this ice cap is in close vicinity to the North Water Polynya (NWP), an area of open water surrounded by sea ice. A 0.5 m yr−1 increase in elevation between 1994 and 1999 is thought to have been related to a local increase in snowfall driven by an increase in the size of the NWP [Krabill et al., 2000].

[43] These experiments indicate that the southeast ice sheet SMB may also be impacted by reduced sea ice cover. While most of the ice sheet will experience increased annual mean accumulation in a warming world via the Clausius-Clapeyron relation, this is not the case everywhere. In the SI simulation there is a significant accumulation decrease in southeast Greenland since a reduction in summer cyclonic activity in the southern Greenland Sea balances the increase in large-scale precipitation, resulting in reduced SMB in the region. Since we are only using one model, it is difficult to assess the robustness of the changes in storm tracks discussed; however, our simulations are in qualitative agreement with both the modeling study of Seierstad and Bader [2009], who predict a reduction in extratropical storminess associated with sea ice decline and observations which show unprecedented high pressure over Greenland during the 2007–2012 period of sea ice minima [Hanna et al., 2012; Overland et al., 2012].

[44] Although the decoupling of sea ice from the SSTs driving the sea ice decline as done in the SI simulation is not physically realistic, it is a way to isolate this effect of reduced sea ice without the influence of the driving global SST changes. We have used this to interpret the changes in both SMB and its components in a future seasonally sea ice-free A1B 2061–2090 scenario, thus providing useful insight into the contribution of sea ice decline to the future evolution of the Greenland Ice Sheet mass balance.

[45] These results indicate that the seasonal cycle of SMB in Greenland will increase dramatically as global temperatures increase, with the largest changes in temperature and precipitation occurring in winter. It also indicates that the sensitivity of sea ice in GCMs should be considered a significant source of uncertainty in GCM-derived future projections of Greenland Ice Sheet mass balance and sea level rise.

Acknowledgments

[46] This research was funded by a Natural Environment Research Council (NERC) studentship and NERC grant NE/I029447/1. This work was carried out using the computational facilities of the Advanced Computing Research Centre, University of Bristol—http://www.bris.ac.uk/acrc/. Assistance in using HadRM3 was provided by the National Center for Atmospheric Science. We acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and the WCRP's Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multimodel data set.

Ancillary