Intense Winter Surface Melt on an Antarctic Ice Shelf

The occurrence of surface melt in Antarctica has hitherto been associated with the austral summer season, when the dominant source of melt energy is provided by solar radiation. We use in situ and satellite observations from a previously unsurveyed region to show that events of intense surface melt on Larsen C Ice Shelf occur frequently throughout the dark Antarctic winter, with peak intensities sometimes exceeding summertime values. A regional atmospheric model confirms that in the absence of solar radiation, these multiday melt events are driven by outbreaks of warm and dry föhn wind descending down the leeside of the Antarctic Peninsula mountain range, resulting in downward turbulent fluxes of sensible heat that drive sustained surface melt fluxes in excess of 200 W/m2. From 2015 to 2017 (including the extreme melt winter of 2016), ∼23% of the annual melt flux was produced in winter, and spaceborne observations of surface melt since 2000 show that wintertime melt is widespread in some years. Winter melt heats the firn layer to the melting point up to a depth of ∼3 m, thereby facilitating the formation of impenetrable ice layers and retarding or reversing autumn and winter cooling of the firn. While the absence of a trend in winter melt is consistent with insignificant changes in the observed Southern Hemisphere atmospheric circulation during winter, we anticipate an increase in winter melt as a response to increasing greenhouse gas concentration.


Surface Melt in Antarctica
Current mass loss of the Antarctic Ice Sheet is made up almost entirely of ice shelf basal melting and iceberg calving (Depoorter et al., 2013). Although supraglacial and englacial runoff has been widely observed, especially in regions of low albedo such as blue ice and bare rock Kingslake et al., 2017;Lenaerts et al., 2016), models suggest that only a small fraction (<1%) of the ∼115 Gt (1 Gt = 10 12 kg) of surface meltwater produced annually Van Wessem et al., 2017) runs off directly into the ocean. Instead, it is refrozen within underlying snow and firn layers (Kuipers Munneke, Picard, et al., 2012). The indirect impact of meltwater is profound, however, as an important role for meltwater-induced fracturing is implicated in the collapse of coastal ice shelves (Banwell et al., 2013;Scambos et al., 2000). Observed collapse following atmospheric warming (Abram et al., 2013) led to a manifold acceleration of grounded-ice flow (De Angelis & Skvarça, 2003;Rott et al., 2011), explaining part of the increased dynamic ice loss witnessed in the Antarctic Peninsula in recent decades (Harig & Simons, 2015). Further loss of ice shelves may induce rapid retreat by mechanical failure of grounded-ice cliffs (Bassis & Walker, 2012;DeConto & Pollard, 2016). Ice shelves may become more vulnerable to breakup due to sustained high rates of surface meltwater refreezing, which warms and softens the ice englacially (Hubbard et al., 2016;Phillips et al., 2010) and ultimately removes the layer of snow and firn .
Model-and satellite-derived surface melt rates range from less than 20 mm w.e./year on ice shelves in Dronning Maud Land and the Amundsen Sea sector to ∼250 mm w.e./year on average over Larsen C Ice Shelf, with certain sectors of Larsen C peaking at 400 mm w.e./year . In summer, the bulk of the energy for snowmelt in Antarctica is provided by solar radiation, which is only partly offset by turbulent fluxes of sensible and latent heat directed away from the surface ( Van den Broeke et al., 2005). In winter, the sensible heat flux is directed toward the surface, to compensate for the absence of solar radiation and consequent surface cooling. Winter temperatures above the melting point of snow have been reported in the Antarctic Peninsula (e.g., Cape et al., 2015;Kuipers Munneke, Picard, et al., 2012;Leeson et al., 2017), in conjunction with warm and dry downslope winds known as föhn. In particular, Leeson et al. (2017) discuss that strong, likely föhn-related autumnal melting led to high surface melt fluxes 2 years before collapse of the Larsen B Ice Shelf in 2002. Föhn-induced melt is also observed in spring (King et al., 2017). Overall, however, the inventory of winter surface melt has been very sparse. Here we report peak annual melt fluxes in the austral winter, derived from measurements from an automatic weather station (AWS) located in Cabinet Inlet, a climatologically unsurveyed area of Larsen C Ice Shelf in the Antarctic Peninsula (Figure 1 and Appendix A), and for the first time discuss the occurrence, significance, context, and impact of wintertime surface melt in Antarctica.

In Situ Observation of Winter Melt
Almost 3 years of AWS observations (25 November 2014 to 13 November 2017) were used to drive a model of the snow surface energy budget (Appendix B; Kuipers Munneke, . It simulates a cumulative melt of 1,040 mm w.e. in Cabinet Inlet (Figure 2), equating to a melt rate of 350 mm w.e./year. We find that only 77% of the cumulative melt (800 mm w.e.) occurs in the austral summer season, which we define here to last from 1 November to 31 March. This summer melt mostly occurs in prolonged episodes of days to weeks, with peak daily melt fluxes of 40-60 W/m 2 . Melt energy in these conditions is supplied by absorbed solar radiation (Kuipers Munneke, . Thus, 23% of the surface melt in the period under consideration (240 mm w.e.) is generated in the winter season, here defined from 1 April to 31 October. Most of the observed wintertime melt occurred in the austral winter of 2016 (190 mm w.e.), with smaller fractions in 2015 (50 mm w.e.) and 2017 (10 mm w.e.). Over the three years recorded, wintertime surface melt took place in all months except July. In contrast to summer melt, the winter melt episodes are usually shorter (at most a few days) and more intense, with daily mean melt fluxes ranging from 25 to over 120 W/m 2 (equating to a melt rate of 6-31 mm w.e./day).
The strongest wintertime melt episode in the record occurred on 25-30 May 2016 ( Figure 3) and featured observed melt fluxes that greatly exceeded those seen during summer. A combination of high observed wind speed (5-18 m/s) and warm air (5-13 ∘ C at 2 m above the surface) resulted in a large turbulent flux of sensible heat downward to the surface. Negative fluxes of longwave radiation (longwave cooling and under clear skies) and latent heat (sublimation) offset some of this sensible heat flux. Still, the resulting melt flux is dominated by sensible heat transfer, which frequently reaches up to 200 W/m 2 with sustained extremes of >300 W/m 2 lasting for up to 30 min (the recording resolution of the AWS data). In total, 71 mm of meltwater was produced as a consequence of this föhn event. During melt, the strong winds advect air that is dried by adiabatic warming, with relative humidity between 35% and 65%.

Cause and Consequence
The combination of strong wind, high temperature, and low relative humidity is common to all wintertime melt events. These are fingerprints for föhn winds that sometimes occur over the Antarctic Peninsula mountain range. Such föhn winds are caused by flow across a topographic barrier, whereby the downslope winds on the leeward side are heated adiabatically (Kuipers Munneke, Luckman et al., 2014). Additional heating can occur due to the drawdown of potentially warm and dry air from aloft when the flow is blocked at lower levels on the windward side and due to entrainment of potentially warm and dry air from upper levels into the flow over the mountains (Orr et al., 2008). The vertical cross section (Figure 3d) over the Antarctic mountain range through Cabinet Inlet from a high-resolution regional atmospheric model (the UK MetOffice Unified Model, see Appendix C) confirms the occurrence of föhn during 25-30 May 2016, with moist air rising on the windward side of the mountains and relatively dry, adiabatically warmed air descending on its leeside. In the lee of the mountains, a hydraulic jump is apparent in the vertical wind component (at about 70 km on the horizontal axis in Figure 3d), characteristic of föhn (Elvidge & Renfrew, 2016).
The regional atmospheric model further shows that during this event, temperature at 10 m above the surface was above the melting point for most of the ice shelf, indicative of widespread surface melt not restricted to Cabinet Inlet (see Figure 3c). A collection of synchronous Sentinel-1A synthetic aperture radar images of a part of Larsen C Ice Shelf (Figure 4), taken near the end of the May 2016 föhn event, confirms that surface melt was occurring beyond Cabinet Inlet. It shows extensive snow melt across the northwestern and western inlets of the shelf (apparent in the satellite images as dark-hued areas), along with a large number of meltwater lakes collected in local depressions of the ice shelf (shown as black elongated features in the images of Figure 4). The elongated depressions express channels in the basal topography, possibly arising from buoyant basal meltwater generated at the grounding line (Sergienko, 2013) or from the grounded ice being extruded over a strongly undulating grounding line (Gladish et al., 2012).
Wintertime meltwater is able to percolate deeply into the snowpack, bringing snow temperatures to the melting point up to a depth of about 3 m (Figure 2b) when latent heat is released as the meltwater refreezes. Observations from thermistor-instrumented boreholes (Appendix D) suggest that effective pathways for this major source of latent energy are available year-round, as even during winter melt, the warming of near-surface layers is almost instantaneous (e.g., in two episodes of melt during end of March and April 2015, apparent in Figure 2b, percolation to 3 m depth occurred in about 12 hr). As a consequence, winter melt warms the snowpack, allowing for an earlier start of the main melt season in spring and summer and heating of the deeper ice layers. Also, it forms relatively impermeable infiltration ice (Hubbard et al., 2016) that can act as a runoff surface for meltwater .

Past, Present, and Future Winter Melt
To put the occurrence of wintertime surface melt into a longer-term perspective, we use satellite-borne Quik Scatterometer (QuikSCAT) (2000-2009) and Advanced Scatterometer (ASCAT) (2009-2016) scatterometer sensors to estimate the number of melt days over the Antarctic Peninsula for each winter season ( Figure 5 and Appendix E). Radar scatterometry is an active remote sensing technique that is sensitive to the presence of liquid water in snow or firn. While the presence of liquid water does not necessarily imply the occurrence of surface melt, the two are highly correlated (Van den Broeke et al., 2010). In some years, little or no meltwater is present during winter, whereas in other years, the number of days with liquid water present on Larsen C approaches 10. In all years, we see enhanced melt over the Larsen C Ice Shelf near the base of the eastern slopes of the Antarctic Peninsula mountain range, which is consistent with föhn-driven warming (Luckman et al., 2014). Near the calving front of Larsen C, wintertime melt occurred in the extreme melt winter of 2006 and to a lesser extent in 2016. These stand out in our records as the most intense winter melt seasons since 2000.
No trend in the occurrence of winter melt over Larsen C is evident over the 17-year record of satellite scatterometry, which implies that there have been no contemporaneous changes in the frequency or duration of westerly föhn events in winter over the Antarctic Peninsula. This is consistent with the absence of any statistically significant winter trends in the Southern Annular Mode (SAM), which is the principal mode of Southern Hemisphere extratropical climate variability, during the scattermometry period considered or even since 1958 (Marshall, 2003). By contrast, the SAM has shifted toward a more positive phase in summer in recent decades (Marshall, 2003), resulting in increased westerly flow over the Antarctic Peninsula and associated föhn events (Marshall et al., 2006;Orr et al., 2008). These led to anomalous warming and melt due to föhn (Cape et al., 2015). As the winter SAM is projected to trend toward its positive phase during the 21st century because of increased greenhouse gas concentrations (Simpkins & Karpechko, 2012), we can thus expect an enhancement of winter melt in this century, from atmospheric circulation changes alone. The magnitude of this effect increases with the atmospheric concentration of greenhouse gases (Zheng et al., 2013).
Ice shelf collapse by hydrofracturing is implicated in future scenarios of rapid Antarctic ice discharge and sea level rise (DeConto & Pollard, 2016). For these processes to be understood and predicted, a reliable estimate of future surface melt and its impact on the state of the firn layer on ice shelves is required. Such estimates can only be achieved if subtle processes like the formation of impermeable ice layers, winter warming, and densification of firn, due to meltwater percolation, ponding, and refreezing, are appropriately taken into account.

Appendix A: Automatic Weather Station
An AWS was installed in Cabinet Inlet (66 ∘ 24.1 ′ S, 63 ∘ 22.3 ′ W) in November 2014. Data were recorded at half-hourly resolution. Instrument height was monitored and usually between 1.7 and 2.4 m above the surface. Reported quantities are at a nominal levels of 2 m for temperature and humidity and 10 m for wind speed. Shortwave radiation was tilt corrected using a Moderate Resolution Imaging Spectroradiometer 10.1029/2018GL077899 satellite-guided procedure (Wang et al., 2016). Air temperature observations were unventilated, leading to overestimation during calm, sunny days. A correction function was derived from concurrent thermocouple observations during November 2014 through January 2015. Observations of relative humidity were corrected for solar heating of the housing of the humidity sensor and for hysteresis effects due to a long response time of the sensor.

Appendix B: Surface Energy Balance Model
The surface energy balance was computed using a model that includes a multilayer snowpack. The model forces the energy budget to close by iterating to a surface temperature for which all the terms balance (Kuipers Munneke, Picard, et al., 2012). If that temperature is above the freezing point, all excess energy is used for melting, and the surface temperature (and associated outgoing longwave radiation) is set to the melting point. The model is evaluated by comparing computed surface temperature with observed values (computed from the outgoing longwave radiation with the Stefan-Boltzmann law). The difference between these is 0.21 K on average (RMS = 1.85 K; RMS = root-mean-square). Further, the timing of melt is corroborated with surface height lowering observed by a sonic height ranger attached to the AWS mast and by observations of outgoing longwave radiation indicating the surface temperature to be at the melting point.