The influence of föhn winds on annual and seasonal surface melt on the Larsen C Ice Shelf, Antarctica

Warm, dry föhn winds are observed over the Larsen C Ice shelf year-round and are thought to contribute to the continuing weakening and collapse of ice shelves on the eastern Antarctic Peninsula. We use a surface energy balance (SEB) 10 model, driven by observations from two locations on the Larsen C ice shelf and one on the remnants of Larsen B, in combination with output from the Antarctic Mesoscale Prediction System (AMPS), to investigate the year-round impact of föhn winds on the SEB and melt from 2009-2012. Föhn winds have an impact on the individual components of the surface energy balance in all seasons, and lead to an increase in surface melt in spring, summer and autumn up to 100km away from the foot of the AP. When föhn winds occur in spring they increase surface melt, extend the melt season and increase the number 15 of melt days within a year. Whilst AMPS is able to simulate the percentage of melt days associated with föhn with high skill, it overestimates the total amount of melting during föhn events and non-föhn events. This study extends previous attempts at quantifying the impact of föhn on the Larsen C ice shelf by including a four-year study period and a wider area of interest and provides evidence for föhn-related melting on both Larsen C and Larsen B ice shelves.

: The metadata for the AWS observations. In the current paper, the AWS locations are numbers, following Turton et al (2018). We have also provided the name used in other studies in brackets.

AWS number (name)
Coordinates ( where SW↓ TOP is the incident shortwave radiation at the top of the atmosphere. This allows the impact of föhn conditions on the SW↓ to be assessed without seasonal bias due to the extreme changes in potentially available sunlight. SW↓ TOP is an output 115 from the SEB model of Kuipers Munneke et al. (2009) outlined in Sect. 2.2.

SEB Model
A previously published and validated SEB model was used to compute the surface energy balance at AWS2 and 3 using the AWS data as input (Kuipers Munneke et al 2009, 2012. Hourly output is available from the SEB model, from which, daily averages are calculated and used throughout the manuscript. Only a brief overview is of the SEB model provided here, but a 120 detailed description is given in Kuipers Munneke et al (2012). The SEB model is required, as not all components of the SEB (Eq.2) are measured directly by the AWSs. The SEB is here defined as: SW ↓ +SW ↑ +LW ↓ +LW ↑ +Hsen + Hlat + G + Q = E (2) 125 where SW↓↑ are the incoming and outgoing shortwave radiation, LW↓↑ are the incoming and outgoing longwave radiation, Hsen is the sensible heat flux, Hlat is the latent heat flux, G is the ground heat flux, and Q is the amount of shortwave radiation absorbed by the subsurface due to the penetration of the radiation into the snowpack. E is the net energy flux, taking into account the surface and subsurface melting, that is available to heat, melt or cool the ice surface (King et al 2017). We use the sign convention that all fluxes are positive when directed towards the surface. Therefore a positive E means that the surface is 130 warming and/or melting. To define periods where melt is possible, the following condition is followed: The additional term Emelt states that melting is possible and is equal to the residual of the SEB calculation (E), when the skin temperature (TSK) is at the melting point. Otherwise, the additional energy is not used for melting.
The sensible and latent heat fluxes are calculated using the bulk flux method. The ground heat flux is calculated using a multilayer snowpack model, which allows for multiple layers of melting, percolation and refreezing of melt water (Kuipers Munneke et al. 2012). Within the multi-layer snowpack module, the vertically-integrated change in heat content is calculated to compute the ground heat flux (G). The temperature of the snowpack is initialised using the subsurface temperatures measured by the AWS (at depths of 0.2, 0.3, 0.5, 0.75 and 1.0 m below the surface). Penetration of radiation into the snowpack and the amount of absorbed shortwave radiation (Q) are calculated by a separate module based on Brandt and Warren (1993) and van den Broeke et al. (2008). 145 The skin temperature is calculated iteratively, until the SEB is closed (Kuipers Munneke et al., 2012). By Stefan-Boltzmann's law, this skin temperature provides a value of outgoing longwave radiation, which can be compared to the observed flux of longwave radiation for model validation. As the SEB components are derived from a SEB model but based on measurements by the AWSs, they are referred to as 'observationally-derived' in this paper, to avoid confusing the output with the AMPS 150 data.

Antarctic Mesoscale Prediction System (AMPS)
AMPS is a numerical weather prediction tool that is operationally run by the National Center for Atmospheric Research (NCAR), USA (Powers et al 2012). AMPS is based on the polar version of the Weather Research and Forecasting (WRF) model and is initiated by Global Forecast System (GFS) data. For the AP domain (domain 6), AMPS output is used here at 5 155 km horizontal resolution, and 44 vertical terrain-following levels, and at a temporal resolution of 6 hours. Archived output from AMPS is available at various locations, and horizontal-resolutions around the Antarctic (http://www2.mmm.ucar.edu/rt/amps/, last accessed: July 20 2019). For more information on the set-up of AMPS, and how well AMPS resolves near-surface meteorological conditions and föhn winds over the LCIS, see King et al (2015), Turton et al (2018) and Kirchgaessner et al (2019). 160 To calculate melt from AMPS data, the Eq. (4) was used: where E is the net energy flux available for heating and, potentially, melting the surface when positive. To define periods when melting may occur (Emelt), the condition outlined in Eq. (3) is used. Following King et al (2015King et al ( , 2017, G and Q are omitted from Eq. 4, as these are not available in the AMPS output. In AMPS, Q = 0 because no subsurface absorption of solar radiation is taken into account. During melt, G is zero because the temperature gradient near the surface vanishes.

Föhn-Identification 170
Previously identified föhn winds published in Turton et al (2018) are used. We provide only a brief overview of the method used to identify föhn winds here. Föhn winds were identified from both AWS near-surface observations and upper-air model output from AMPS, with two different criteria. The method for identifying föhn at the near-surface was based on exceeding thresholds of specific relative humidity values. The absolute relative humidity value used as the threshold depended on the exact location. In the case of slightly more humid conditions, an associated increase in air temperature was also included into 175 the method. The method used to identify föhn from AMPS was based on the height change of a particular isentrope from the windward to leeside of the AP Mountains, in order to isolate the isentropic drawdown which is characteristic of föhn winds over the AP. Only when föhn winds were simultaneously identified in both datasets, was a specific period categorised as Föhn.
See Turton et al. (2018) for a discussion on the comparison of Föhn identified by AWS and AMPS.
The term 'föhn conditions' refers to a six-hour averaged period in which föhn winds have been identified. 'Föhn days' 180 are days on which föhn conditions have been identified for at least one six-hour averaged period (but föhn conditions could have been present for the full 24 hours).

Föhn Conditions
Föhn winds have been observed over the whole Larsen C ice shelf and are most frequent at the foot of the AP Mountains (Elvidge et al 2015, Turton et al, 2018, Wiesenekker, 2018. In particular seasons, föhn conditions can be observed up to 12 190 % of the time (Table 2). However, it is not uncommon to have a whole season without the occurrence of föhn conditions. They are most frequently observed in spring over the Larsen C ice shelf (AWS2 and AWS3), but closer to the foot of the mountains (AWS1), they are most frequently identified in summer (Table 2). At all three locations in this study, the average six-hourly temperature change associated with föhn events exceeds 11 K, and the relative humidity decreases by at least 19 % (Turton et al, 2018). Table 2 summarises the percentage of time that föhn conditions were identified at AWS1, AWS2 and AWS3, in each 195 season from 2009-2012.    In both the observations and AMPS, over 30 % of föhn days observed at AWS2 lead to surface melt (Table 3). AMPS slightly overestimates the percentage of föhn days which experience melting, but only by 2.3 % (2 days). However, AMPS overestimates the number of melt days coinciding with non-föhn days more considerably (260 non-föhn days experience melting in AMPS compared to 187 in observations). AMPS is therefore able to better represent the occurrence of melting on föhn days as opposed to melting on non-föhn days. 220 Similarly, AMPS overestimates the average Emelt during both föhn and non-föhn days (Table 3 and Figure 2). The largest overestimation occurs during non-föhn days, when AMPS simulates a mean Emelt of 4.7 W m -2 compared to 1.6 W m -2 in observationally-derived values at AWS2. However during föhn conditions, Emelt is better represented by AMPS, with a smaller positive bias of 0.9 W m -2 on average. This can also be seen in Figure 2. AMPS simulates the largest Emelt values best, and overestimates Emelt most during non-föhn days. When separating by season (Figure 3

), it is clear that AMPS overestimated 230
Emelt in all seasons (except winter), but more often during summer. Some of the largest amounts of daily melting often coincide with föhn days during spring and summer (Figure 3b,c), and AMPS is better able to represent events with large melt amounts than those with low melt rates. These results agree with findings by Kirchgaessner et al. (2019), who found that AMPS underestimates the air temperature during föhn days but overestimates the temperature the rest of the time, which leads to a higher surface temperature and a higher likelihood of melting on non-föhn days (Kirchgaessner et al. 2019). Figure 3c also 235 highlights that the majority of melting during spring is associated with föhn days, which AMPS represents well.
The downwelling shortwave radiation is overestimated by AMPS for this location (King et al 2015(King et al , 2017. Combined with the low albedo value and the poor representation of clouds in AMPS, the ice surface is much warmer in AMPS than in reality (mean bias of 1.8 K). Therefore, on days in late spring and summer when the skin temperature is close to the melting point in 240 observations, AMPS will simulate that it is already at 0 °C, which leads to the overestimation in the total number of days with melting ( Figure 3a,b). The overestimation of melt days was also found in other studies using AMPS and the polar WRF model over ice shelves (Grosvenor et al. 2014;King et al., 2008King et al., , 2015 (Table 4). AMPS simulated Hlat values of -24.4, -10.6 and -10.9 W m -2 for AWS1, AWS2 and AWS3 locations respectively, during föhn conditions. AMPS also overestimates the net longwave flux (LWnet) during föhn conditions, but with much smaller biases than for Hlat. During föhn conditions, AMPS 250 simulates an average LWnet of -36.1 W m -2 for AWS1, -43.0 W m -2 for AWS2 and -43.7 W m -2 for AWS3. From the observation-derived SEB values, AWS1, AWS2 and AWS3 have Lwnet values of -34.8, -40.6 and -41.1 W m -2 during föhn conditions respectively. A combination of lower (more negative) LWnet and Hlat during föhn days in the simulations acts to cool the surface more than is observed, which could be responsible for the better representation of Emelt during föhn days compared to non-föhn days in AMPS. 255 Regardless of the overestimation of non-föhn melt in AMPS, it is evident from the observations and AMPS, that melting during föhn conditions is significantly higher (at the 99 % confidence level) during föhn conditions than during non-föhn conditions, even more than 100 km away from the foot of the AP mountains. As AMPS is able to reproduce föhn-related melting, we have used it to assess the spatial distribution of föhn-induced melting for the entire ice shelf (Figure 4). Föhn-induced melting is 260 most frequent in the north of the ice shelf, largely mirroring the spatial distribution of föhn conditions and the near-surface air temperature (Turton et al. 2018). During summer, föhn-induced melting is simulated over the whole area from Scar Inlet in the north across the entire Larsen C ice shelf to 70°S. Outside of summer, föhn-induced melting is still prevalent over the ice shelf, with over 40 melt events (six-hourly) simulated on Scar Inlet during spring (Figure 4b). There are more föhn-induced melt events during spring than in autumn, likely related to the higher occurrence of föhn during spring. 265 The following analysis, separated into annual, interannual and seasonal impact of föhn, uses the observations and derived SEB components at the three locations to quantify the impact of föhn on the ice shelf. Table 4 presents the annual-averaged differences between föhn and non-föhn conditions in some of the observed SEB 270 components. In the SEB data, 14 % of non-föhn days at AWS2 from 2009-2012 were melt days. These are largely confined to summer months, when melting occurs annually. The frequency of melt days more than doubles when assessing föhn days, with 31 % of föhn days at AWS2 coinciding with melt days. A similar magnitude of increase was observed at AWS3, where melt-day occurrence increased from 12 % during non-föhn conditions to 20 % during föhn days. Therefore, even at a distance of 100 km from the foot of the mountains, föhn conditions are able to increase the number of melt days per year. The largest 275 increase in melt-day occurrence was at AWS1, where the percentage of days observing melt increased from 14 % during nonföhn days, to 43 % during föhn days in 2011 and 2012.

Annually-averaged impact of föhn
During föhn conditions incoming shortwave radiation (SW↓) is hypothesized to be larger due to the clearance of clouds, which can occur in the lee of the AP mountains (Grosvenor et al 2014, Elvidge andRenfrew, 2015). However, due to the large annual 280 cycle in SW↓ in the Polar Regions, this could bias the difference between föhn and non-föhn days if they are not evenly distributed throughout the year. From Table 2 and Turton et al (2018), it is evident that the föhn days are not evenly distributed.
Therefore, the shortwave transmissivity (tau) (see Data and Methods) is a more reliable indication of the impact of föhn winds on the downwelling shortwave radiation. Data show an increase in shortwave transmissivity at all three locations during föhn conditions, indicating an increase in the incoming shortwave radiation due to cloud clearance (Table 4 and Figure 5d). 285 However, the differences in tau between föhn and non-föhn conditions are small and are not statistically significant. The sensible heat flux (Hsen) was much larger (and positive) during föhn conditions than during non-föhn conditions (Table 4, Figure 5). This is due to the increased air temperature and higher wind speed, both leading to an increased supply of heat to the surface. The annual average sensible heat values observationally derived during föhn days for AWS2 and AWS3 are very similar (23.0 W m -2 and 24.2 W m -2 respectively) (Table 4). However, at AWS1 it is slightly smaller (19.7 W m -2 ) due to the 300 slightly lower wind speeds under föhn conditions at this location compared to the other locations. In most locations along the foot of the AP, the wind speeds are higher than further downstream. However, for the two years of available data at AWS1, this is not the case here. Despite the smaller Hsen value at AWS1, the increase in Hsen between non-föhn and föhn days is similar to the other locations; between 23.1 W m -2 at AWS1 and 23.5 W m -2 at AWS2. Figure   The latent heat flux becomes more negative during föhn days at AWS1 and AWS2, which we can attribute to the increase in sublimation of surface snow increases due to the dry air over the ice shelf ( Figure 5). At AWS3, there is an increase in Hlat (1.3 315 W m -2 ) during föhn conditions but this difference to non-föhn conditions is not statistically significant. At AWS1, the difference in Hlat between föhn and non-föhn conditions is larger than at the other locations and is significant. This is likely due to the drier föhn conditions observed close to the AP, whereas at about 100 km distance, the air has become more moist due to mixing with pre-existing non-föhn air, and therefore less sublimation occurs. This is evident in the AMPS simulation ( Figure 6), with more negative values of Hlat closer to the AP during föhn days.
LWnet becomes significantly more negative during föhn events (Figure 5), which is likely related to the 'föhn clearance', whereby there are fewer clouds during föhn conditions which reduces the downwelling flux of longwave radiation . Simultaneously, the warmer surface increases the outgoing longwave radiation flux, which both contribute to a more negative LWnet than during non-föhn conditions. Despite the larger negative fluxes of LWnet and Hlat 325 during föhn conditions, the positive increase in SWnet and Hsen contributes to more energy being available for melt (Emelt) during föhn conditions. Emelt more than tripled during föhn conditions compared to non-föhn conditions, from 1.6 to 7.6 W m -2 at AWS2, and from 2.4 to 10.0 W m -2 at AWS1 (Table 4). AWS1 is closer to the foot of the AP Mountains and therefore experiences warmer and 330 more frequent föhn conditions, which contributed to the significantly higher amount of energy available for melt during föhn periods (Table 4). Figure 6 presents a spatial distribution of the energy flux during föhn periods in AMPS. A higher energy flux is present over Larsen B than Larsen C during föhn winds. Therefore, as well as additional melt days, the amount of melting on those days also increases in association with föhn winds. The combination of these generates föhn-induced melting of the ice shelf. 335

Interannual melt
Due to the poor availability of data at AWS1 and AWS3 locations, this section will focus solely on AWS2 data. The annual average amount of melt from 2009-2012 (föhn and non-föhn conditions combined) was 180 mm w.e. yr -1 at AWS2. Excluding föhn days, the annual average melt amount at AWS2 reduced significantly to 146 mm w.e. yr -1 . In other words, föhn increases the melt volume at AWS2 by 34 mm w.e. yr -1 (18 %). The majority of non-föhn melting is restricted to the summer months. 345 Therefore, the potential for föhn winds to create additional melting, outside of the usual melt season is of interest, as they could extend the melt season. In contrast to 2010, only 11 föhn conditions were identified at AWS2 from September to December in 2012. The annual melt amount in 2012 was 83 mm (significantly less than in 2010 at the 95% level). When föhn days were removed from the 2012 360 analysis, the annual total melt only decreased by 0.1 mm w.e. During spring, little melt is observed except on föhn days. Hence, the frequency of föhn winds in spring has an impact on the melt.
The annual number of melt days, energy available for melt, annual melt amount and length of the melt season all increased due to the occurrence of föhn winds, especially in years when a large number of föhn conditions were identified during the 365 extended summer period (Oct-Mar). We will now present the impact of föhn winds in separate seasons, but with a particular focus on spring.

Spring
The largest increase in surface melting is experienced during föhn winds in spring (SON). Although not all changes in mean values between non-föhn and föhn conditions were statistically significant, the changes in the SEB indicate a large impact due 370 to föhn winds. Table 5 displays the average values for composites of föhn and non-föhn periods during spring. lower during föhn conditions than during non-föhn conditions at all locations (Table 5). Both turbulent fluxes exhibited 380 significant differences during föhn conditions compared to non-föhn conditions at AWS1 and AWS2. The sensible heat flux increased (more positive) by over 20 W m −2 at all locations (Table 5), and the largest increase was observed at AWS1 (27.5 W m -2 ). The warmer air and higher wind speeds contribute significantly to increasing the Hsen over the ice shelf. The latent heat flux is more negative at AWS1 and AWS2 during föhn conditions, indicative of sublimation and evaporation. However, at AWS3 Hlat increased, although this was not statistically significant (95 % confidence interval). 385 The cooling effect of the net longwave radiation and latent heat flux was not large enough to counteract the considerable heating processes, therefore, melt energy was available during spring föhn conditions. At AWS2 during spring, melting occurred on just 3 % of non-föhn days, due to the low temperatures in the absence of föhn winds. However, melting increases significantly when accounting for föhn conditions. In spring, 28 % of föhn days coincided with observed melting. A similar 390 increase was found during föhn days at the other locations: At AWS1 just 5 % of non-föhn days coincided with melt days, whilst 30 % of föhn-days coincided with melt days. At AWS3, 4 % of non-föhn days and 28 % of föhn days experienced melting. Föhn winds therefore contribute to an increase in the number of melting days per year at all observed locations, including those at over 100 km from the foot of the AP.
At AWS2, the average energy available for melt (Emelt) during spring föhn conditions was 7.7 W m −2 (Figure 7). This was greater than the mean daily melt energy during summer at this location (7.0 W m −2 ). The amount of melt energy associated with föhn conditions at AWS1 was lower (3.8 W m -2 ) than at AWS2, however this does not take into account the large melt amount and early melt onset associated with föhn winds in spring 2010, as data are only available from February 2011 to December 2012. When assessing the annual average melt energy for 2012 (period in which observations overlap), there is 400 considerably more daily melt energy at AWS1 (3.5 W m -2 ) than at AWS2 (1.0 W m -2 ). Therefore, during spring, föhn conditions increase both the average rate of melt production, and the number of melt days, both close to the foot of the AP and up to 130km away.

Summer 410
The energy available for melt and percentage of melt days during summer is relatively high, regardless of additional föhn induced melting, due to higher air temperatures and larger SW↓ during this season (Figure 7, Table 7). A day may already have experienced melting, and the presence of föhn winds was coincidental and did not cause the melting. However, from previous studies, it has been found that individual föhn events can increase or prolong melt when it occurs during summer Kuipers Munneke et al., 2012). 415 Table 6: Summer daily average values of SEB components and surface temperature during composites of föhn and non-föhn conditions at AWS1, 2 and 3. The * indicates statistical significant difference between föhn and non-föhn periods using the T-test at 95% level and ** is at the 99% confidence level. Emelt (W m-2) 12.1 7.0 1.6 30.8** 18.3** 6.0* Tsk (°C) -3.7 -3.1 -4.8 -2.0** -3.6 -2.9 There was a significant increase in the net shortwave radiation and a decrease in net longwave radiation during summer föhn periods in comparison to non-Föhn conditions, likely, due to the cloud clearing during föhn (Table 6) (Grosvenor et al 2014).
The sensible heat flux significantly increased at all three locations during föhn conditions, which changed the direction of energy transport from negative (away from the surface) during non-föhn to positive (downwards) during föhn conditions. 425 Negative sensible heat flux, associated with convection, is common at AWS2 during summer (Kuipers Munneke et al., 2012).
As a consequence of the higher surface temperatures, sublimation and evaporation are common in summer, leading on average to a negative latent heat flux. Under föhn conditions, the change in conditions was mixed. At AWS1, the latent heat flux became more negative during föhn conditions, indicative of enhanced sublimation (Table 6). At AWS2 and AWS3, Hlat 430 increased, however the changes from non-föhn to föhn were not significant.
The increase in Emelt during föhn conditions was statistically significant at all locations and was largest at AWS 2 (95 % confidence level), increasing from 7.0 to 18.3 W m -2 . During 78 % of föhn days the surface was melting, compared to just 54 % of non-föhn summer days (at AWS2). The number of melt days and the melt energy both increase in summer under föhn 435 conditions. Despite the already warm conditions and high melt amount, there are statistically significant changes to the SEB components and melt energy due to föhn conditions in summer.   This is far outside of the typical melt season at this location. The preceding melt day was on March 5, 2011 and was not associated with föhn winds. Therefore, föhn winds have the ability to induce melting outside of the usual summer melt period.

Autumn 450
The sensible heat flux was positive and significantly larger during föhn days in all three locations, and in AMPS output (not shown). On average, Hsen increased from 0.0 W m -2 during non-föhn periods to 20.9 W m -2 during föhn conditions at AWS2. This agrees well with the AMPS output, which simulates a mean Hsen of 24.9 W m -2 during föhn days in autumn at AWS2.
Surface warming was also significantly larger during föhn days than during non-föhn days, raising the surface temperature by 11.8 K at AWS2. 455 Figure 3d highlights the low daily Emelt values at AWS2 during autumn. The energy available for melt only increased by 0.1 W m -2 (at AWS2) between föhn and non-föhn days. Melting only occurs on 1 % of non-föhn days, whereas 8 % of föhn days experience melting at AWS2. Closer to the AP, föhn-induced melting was higher than at the stations further east. The percentage of föhn days experiencing melt at AWS1 was 43 % compared to just 1 % of non-föhn days. Daily average Emelt increased from 0 W m -2 during non-föhn days to 4.3 W m -2 during föhn days. Therefore, föhn-induced melting is possible during autumn, although it is limited in extent, and only occurs very close to foot of the AP mountains, where föhn winds are warmer.

Winter
The smallest impact from föhn conditions on surface melt was observed during winter. The radiation deficit in winter, mainly 465 due to the lack of solar radiation, is so large that increased sensible heat during föhn can almost never bring the surface to the Hsen and Hlat both experience a significant change between föhn and non-föhn days. At AWS2 the average Hsen increased from 3.4 W m -2 to 36.6 W m -2 during föhn days, and the latent heat increased from 0.6 W m -2 to 2.9 W m -2 . A similar magnitude of change in Hsen and Hlat was observationally-derived at AWS3 and AWS1. The large increase in Hsen is attributed to the considerably warmer (and often windier) conditions during föhn. There was no observation-derived melting during winter 475 föhn events.

Discussion
This study uses three locations for SEB calculation from observations, to provide a larger spatial interpretation of melting associated with föhn winds. Despite that the low number of observations is still a limitation of this study. The most reliable SEB dataset was obtained for AWS2 where a SEB model was run by Kuipers Munneke for 2009 to 2012. Unfortunately, no SEB data were available for this entire period at the foot of the AP (AWS1), where the largest melt rate and highest number 485 of melt days have been previously observed in satellite images (Luckman et al., 2014). More recently, data have become available for more locations on the LCIS, including the inlets, which observe high melt rates associated with föhn winds

500
During the period under investigation in this study there were no melt days in either föhn or non-föhn days during winter at the three AWS locations used here. There is a high spatial and temporal variability in the occurrence and strength of föhn winds over the LCIS (Turton et al. 2018), which likely explains the contradicting results to those by Kuipers Munneke et al.
(2018), who identified high rates of winter melting associated with föhn winds. The winter melting period investigated previously was not within the period investigated here. With a longer observational period at AWS1 now available, this site 505 should be investigated further, as winter melt may be specific to individual years.
The number of melt days estimated by AMPS exceeded the observationally-derived values, especially during non-föhn days.
One reason for the melt day overestimation is the positive bias in near-surface temperature during non-föhn conditions in AMPS (Kirchgaessner et al 2019). This is caused by the positive bias in incoming shortwave radiation, which results from the 510 poor representation of clouds in the model (Listowski et al. 2017), together with the low albedo value used in AMPS. This has been discussed by Grosvenor et al. (2014); King et al. (2015) and King et al. (2017). Therefore currently, the values of melt energy from AMPS cannot be trusted if used as an absolute estimate of melting specifically caused by föhn winds. However, it can be used to infer the spatial patterns of melting during föhn days, and the average melt energy when including all melt events. The poor representation of clouds in many regional climate models causes issues in the accuracy of SEB and melt 515 information. Recently, Gilbert et al. (2020) found that cloud-phase during the austral summer strongly influences the amount of melting on the LCIS and can determine whether melting is simulated or not by the MetUM model. in Bevan et al (2018). Therefore, the effects of föhn winds on the SEB could be even greater than we have highlighted in this study, if the SEB could be calculated for particularly high-melt years.

Conclusions
The discrimination between föhn and non-föhn conditions provides a robust understanding of the impact of föhn on components of the SEB and ultimately, surface melt, by assessing the more general response to föhn, as opposed to studies of 530 individual events. The limitation of assessing case studies is that the chosen event may be an anomaly, or not representative of the average föhn conditions, whereas assessing the average impact of föhn provides more confidence in the quantification of surface melt due to föhn.
This study comes to similar conclusions as studies by King et al. (2015) and King et al (2017). Especially in spring, föhn 535 conditions have the potential to prolong the melt season by initialising an early onset of the melt season. Moreover, this study concludes that the intensity of melt increases during föhn conditions, even 100km from the AP. Föhn conditions have a large impact on the sensible heat flux, which leads to an excess of energy that is available to heat and melt the snow during spring, summer and occasionally, autumn.

540
The results presented here and in previous studies (Bevan et al 2018, Luckman et al 2014, Wiesenekker et al, 2018 highlight the large interannual variability in melt amount and duration. Similarly, there is a large interannual variability in the spatial and temporal distribution of föhn winds over LCIS. The number of melt days over the majority of the LCIS has decreased between the years 2000 and 2016. However, close to the foot of the AP, where föhn winds are strongest and most frequent, the amount of melt has increased over this same time period (Bevan et al 2018). Regional climate models such as the MetUM 545 (Elvidge et al 2015), WRF model (Turton et al 2017) and RACMO2 (Weisenekker et al 2018) can now accurately represent near-surface conditions during föhn over the LCIS and these could be used to extend the study in future research. However, AMPS struggles to represent melting outside of föhn conditions, and although it performs better during föhn conditions, this is likely due to the higher negative biases in net longwave radiation and latent heat flux, and not due to a good representation of föhn warming. Therefore, to completely trust the impact of föhn winds on the SEB of the Larsen C ice shelf, the components 550 of the SEB should be improved in AMPS.

Data Availability
Subsets of the AMPS output are available online at https://www2.mmm.ucar.edu/rt/amps/. The surface energy balance values from AWS1, 2 and 3 are available upon request from Peter Kuipers Munneke at P.KuipersMunneke@uu.nl. 555