The AWS was installed on Joyce Glacier at 78∘01.245′ S, 163∘47.608′ E, roughly 400 m above sea level on 26 November 2008 and removed on 31 January 2010. The glacier tongue is located at the western end of the Garwood Valley, a coastal valley east of the Royal Society Range (about 25 km from McMurdo Sound). The station was set up in the ablation zone of Joyce Glacier, about 150 m from the north-east cliff. The glacier tongue is approximately 2 km wide, and the equilibrium line is roughly 3.5 km up from the glacier terminus at an elevation of 600 m.

Continuous measurements of air temperature, relative humidity, incoming and outgoing longwave and shortwave radiation, wind speed and direction, and air pressure were obtained. Surface temperature is calculated from observed outgoing longwave radiation using the Stefan–Boltzmann law and an emissivity of 1. Additionally, the distance to the ice surface was measured using a sonic ranger on the mast for instrument height, ablation and snowfall estimation. The instruments that were used are listed in Appendix , Table , and were the same as those described in . The measurements were taken every 300 s and stored as 15 min averages on a data logger.

Data processing was similar to that described in and includes recalculation of relative humidity measurements to correct for saturation with respect to ice instead of liquid water . Accumulated albedo was calculated from accumulated incoming and outgoing shortwave radiation over 24 h as in and filtered for outliers by removing data outside the 2nd and 98th percentiles. The surface height change record was cleaned by manually removing spikes from the data and applying a 2 h running mean through the data. A snow accumulation record was calculated from daily differences in surface height, using a lower threshold of 1 cm for precipitation to occur.

We used the energy balance model presented by and further developed as described in , which assumes the energy balance for a surface skin layer by solving the following equation for skin temperature: 1QM=Sin(1-α)+QP+Lin+Lout+SH+LH+QG, where QM is the energy used for melt, which is 0 when the surface temperature Ts<273.15 K; Sin the incoming shortwave radiation; α the surface albedo; QP the amount of penetrated shortwave radiation; Lin and Lout the incoming and outgoing longwave radiation; SH the turbulent sensible heat flux; LH the turbulent latent heat flux; and QG the subsurface conductive energy flux. All fluxes directed towards the surface are defined as positive, and the fluxes are given in W m-2. Energy supplied by rainfall is not relevant for the MDV glaciers and therefore is neglected.

Snet (Sin(1-α)) is calculated from observed Sout and accumulated albedo as suggested by . Lin is based on AWS observations, and Lout is calculated using the Stefan–Boltzmann law from the model-calculated skin layer temperature. The turbulent heat fluxes SH and LH are calculated with the bulk aerodynamic method, considering atmospheric stability based on the Monin–Obukhov similarity theory. Measurements of temperature, wind speed and humidity at the AWS measuring level are used together with the calculated skin surface temperature, zero wind speed and saturated humidity at the surface to calculate the mean gradients that are related to the turbulent heat fluxes.

Solar radiation is able to penetrate the snowpack, and attenuation of the shortwave radiation is calculated separately for 118 different wavelength bands using the model of , also used by and . The solar radiation that is able to penetrate the snowpack QP is removed from the surface layer energy balance and added to the appropriate subsurface model layers. The parameter dzrad, the surface layer thickness used in solar radiation penetration calculation, determines the partitioning of the penetrated radiation between the surface layer and subsurface energy budgets. Solar radiation can penetrate to considerable depths when the surface is ice and is reduced when it is snow-covered. A snow grain radius of 0.2 mm and an ice grain radius of 2.8 mm as measured by at Taylor Glacier are used for the solar radiation penetration routine.

The subsurface model is based on the SOMARS snow model . The vertical subsurface grid is composed of layers with increasing layer thickness with depth. The parameter dz0 sets the thickness of the upper grid cell. Subsurface melt in the snow or ice pack (hereafter called internal melt) occurs when the subsurface temperatures of a layer reach 0 ∘C and the excess energy is used for melt. Internal meltwater is added to the water content of that layer and can drain to deeper layers and refreezes if temperatures are below 0 ∘C. Refreezing in turn increases the subsurface temperature and density. Subsurface meltwater that percolates down to the firn–ice interface or subsurface melt that occurs in the ice runs off. To give an estimation of the uncertainty involved in parameter choices, we ran the model as an ensemble using different parameter combinations. Table  shows the three parameters that are considered and the set of values that is used in the ensemble.

The numerical weather prediction system AMPS is used as a real-time operational forecasting tool for Antarctica by the National Center for Atmospheric Research (NCAR), USA. AMPS is based on the polar version of the Weather Research and Forecasting (WRF) model. The boundary and initial conditions of the AMPS forecasts are taken from output of the Global Forecast System (GFS), and AMPS is initialized twice daily. The 12–24 forecast hours is used for analysis as in to discard the spin-up period in which the model is adjusting to higher-resolution topography while also avoiding model drift at later forecast hours.

AMPS is run on five or six domains over Antarctica and its subregions and archived for different domains and resolutions. In this study we use output of AMPS WRF 45 version for the year 2009 for the d05 domain. This domain is centred over Ross Island and covers the McMurdo Dry Valleys with a horizontal grid spacing of 1.67 km. AMPS is run on 44 model levels, and this study uses AMPS products that were reduced to the first 15 model levels (covering approximately the first 1.5 km above the surface) and 6 fixed pressure levels. In addition, the d02 domain is used for synoptic-scale analysis, covering the Antarctic continent with a 15 km horizontal resolution.

The AMPS forecast has been used previously in Antarctic foehn and other Antarctic meteorological studies . evaluated high-resolution AMPS forecasts with AWS observations in the MDV and concluded that the model output is suitable to study regional airflow and synoptic conditions during foehn in the MDVs. While AMPS can have difficulties representing the near-surface conditions on the valley floors, it performed well in simulating the foehn onset and cessation .

In this study, we use three different classification criteria to detect foehn events based on AWS observations of atmospheric near-surface conditions and AMPS upper-air forecasts. The three methods are combined for a final foehn detection algorithm used in this study. To test the suitability of the AMPS-forecasted near-surface conditions for these classification schemes, we evaluate the methods as well using the AMPS forecast of near-surface conditions in Sect. . For consistency all methods are tested using hourly averaged AWS observations and hourly AMPS forecast data. In the remainder of the analysis presented in this study, only the combination of classification schemes based on AWS-observed near-surface conditions is used.

We applied the three methods described in Sect.  to identify foehn conditions at Joyce Glacier using both near-surface conditions based on AWS observations and AMPS output for the 14-month study period. Figure  shows an example of foehn conditions at Joyce Glacier in April 2009 based on AWS observations and AMPS output. The chosen forecast grid point was the closest ice point in terms of location relative to the Joyce AWS. The temperature based on the AMPS forecast was corrected for the 150 m elevation difference between grid cell and AWS elevation by using a monthly average temperature lapse rate calculated from the surrounding AMPS grid points. The AMPS 2 m temperature shows a strong correlation with observed 2 m temperature (0.97 using hourly data and 0.86 for hourly anomalies from the monthly mean to remove the impact of the annual cycle). The 2 m temperature has a mean bias of -1 ∘C, with a larger negative bias in summer months (down to -3 ∘C) and a smaller bias in winter. A lower correlation is found for AMPS relative humidity (0.45), with a mean bias of -5.7 %. AMPS 10 m wind speed has a correlation coefficient of 0.5 with observations and mean bias of -1.2 m s-1. The performance is much poorer when looking at u and v components, which is reflected in the clear difference between the observed and AMPS-forecasted wind direction (Fig. a). The AWS was positioned relatively close to the valley sides, and the deviation in the wind direction through topographic effects is likely not captured in AMPS. The general wind regime agrees well between forecast and observation: both show a dominant wind direction with low wind speeds that changes abruptly to a different direction with stronger winds. On average, the strong wind events tend to have higher wind speeds in the AMPS forecast compared to observations. In contrast, the wind speeds during calm conditions are underestimated in the AMPS forecast, suggesting a poor simulation of down-glacier winds. While differences are visible in the surface conditions between AMPS and observations, the AMPS forecast performs well in representing foehn onset and cessation (Fig. ).

The Speirs method uses criteria on sudden changes in relative humidity and temperature and therefore detects mainly the onset of the foehn period (Fig. ). Foehn periods detected by the Speirs method are on average shorter compared to the other methods and do not cover the full period where the typical windy, dry and warm conditions last. There is little agreement in the Speirs detection method between using AWS-observed or AMPS-forecasted near-surface conditions. In the AWS data, 687 h was flagged with foehn conditions, and 294 h was flagged in the AMPS data, of which 122 h agreed with the AWS foehn hours (Appendix , Fig. ). The difference between the AMPS and AWS data can firstly be related to the fact that changes in the temperature and humidity forecast are typically slower than observed, indicating that other thresholds might be needed to use the Speirs method on forecast data. Adjusting the threshold to changes over a 2 h instead of a 1 h period results in almost 250 more hours being detected as foehn in the AMPS forecast and more than doubles the hours that agree between AMPS and AWS data. Secondly, an accurate representation of the near-surface conditions might be limited by the vertical resolution of the AMPS forecast.

The Wiesenekker method gives the longest foehn periods, as this method only uses periods of strong winds from one direction and is not limited by temperature and humidity criteria. In the AMPS data, 13 additional foehn events are detected by the Wiesenekker method. These are short events with stronger winds that were not observed at Joyce Glacier, such as for 4 April in Fig. . Finally, out of the methods using near-surface observations, the Turton method gives the largest number of foehn hours (Appendix , Fig. ). The method gives 934 foehn hours using AWS near-surface conditions, compared to 445 for AMPS near-surface conditions. These differences likely follow from the slower and weaker simulated humidity drops in AMPS. This confirms the findings of at the Antarctic Peninsula that AMPS underestimates the temperature increase and relative humidity decrease during foehn.

Throughout the rest of this paper a combination of the detection methods using near-surface conditions from AWS observations is used to classify foehn. The detection methods using near-surface conditions from AMPS forecast are only represented in this section for evaluation purposes. The periods that are flagged as foehn in the combined method are generally longer than indicated by the individual methods using AWS near-surface conditions, but we are confident they cover the period between foehn onset and cessation (blue shading in Fig. ). The wind speed and wind direction criteria of the Wiesenekker method play a dominant role in the detection, and either the Speirs or the Turton method confirms the drying and warming characteristic of the foehn event. In total 27 foehn periods occurred during the observation period, counting for 10 % of the study period and occurring on 15 % of the days. April and November were the months with the most foehn hours, accounting together for 50 % of the total (Appendix , Fig. ). During the melt season (December and January) six foehn events occurred, which covered 6 % of the period. This lower foehn frequency in summer compared to winter has been found in other MDV studies as well and is related to weaker cyclone activity over the Ross Sea region. The mean foehn duration is 37 h, with a minimum duration of 6 h and a maximum duration of 90 h. The highest temperature of 5.8 ∘C was measured during a foehn event on 26 December 2008.

The energy and mass balance model is evaluated against observed surface height change and surface temperature. The optimal simulation was chosen based on the smallest root-mean-square error (RMSE) and bias between observed and simulated surface temperatures (Fig. a). The ensemble spread, based on ensemble standard deviation with respect to ensemble mean, is provided to indicate uncertainty in the simulated surface height change introduced by parameter settings. The optimal surface temperature of the simulation ensemble agrees well with the observed surface temperature (RMSE 1.44 ∘C, Fig. a). The model only slightly overestimates the observed surface temperature (bias 0.02 ∘C). In summer, the simulated surface temperature reaches melt (0 ∘C) more frequently than observed (Fig. a).

The simulated and observed surface height changes agree reasonably well in December and January months when melt occurs (Fig. b). During the first melt season (melt1, Fig. b) 12 cm of ice was lost in the observations, compared to 9 cm in the simulation. In the second melt season (melt2), ablation was around 8 cm in both the observations and the simulation. The two snow accumulation events during the melt periods correspond with observations, since the surface height record was used to determine the accumulation rate.

The largest deviations between simulated and observed ablation are for the period from February until May (subl1, Fig. b). Simulated ablation is 1 cm, compared to 9 cm in the observations. The largest ablation is observed during the second half of April, during which several foehn events occurred (Appendix , Fig. ). Sublimation rates calculated by the bulk method might be underestimated during these foehn events in the simulation. Still, the SR50 surface record should be interpreted with care since the observed surface height changes can be caused by snow redistribution, which is not captured in the simulations. There is better agreement during the May–December period, where the surface height change through sublimation is 5 cm for both model and observations, although the distribution in time differs.

The ensemble spread in Fig. b indicates the uncertainty introduced by the parameter settings listed in Table . During the melt seasons the standard deviation in the simulation ensemble increases, showing the uncertainty in the simulated surface height as controlled by uncertainty in simulated melt and sublimation. Over the winter period the standard deviation stays rather constant, indicating that most of the uncertainty in surface ablation is related to the simulation of melt.

The largest uncertainty in the simulated surface height change is due to the dzrad parameter (not shown), which determines the partitioning of solar radiation to the subsurface. The grid resolution of the upper snow layers also largely impacts the simulated surface melt. The importance of solar radiation penetration and the uncertainty that is associated with the partitioning of solar radiation for MDV glaciers are also noted by . In this paper we do not focus on the amount of melt that occurs as surface or internal melt but rather on the occurrence of melt in relation to foehn events. The good agreement between the observed and simulated surface temperature gives us confidence in using the model to study the dominant energy fluxes and melt occurrence under foehn and non-foehn conditions.

A mean sea level pressure anomaly plot during all foehn hours reveals the presence of low-pressure systems over the Ross Sea during these events (Fig. a). The dark blue shading in Fig. a shows the average location of cyclones during foehn events at Joyce Glacier, rather than individual events. These results are consistent with earlier studies that link foehn events observed in the Taylor, Wright and Victoria valleys with cyclones over the Ross Sea .

The presence of the cyclone results in a southerly flow from the Ross Ice Shelf northward towards the MDV (Fig. a). Part of the observed southerly flow is blocked at low levels by the Royal Society Range and is split into a component flowing into the gap on the western side of the mountain range and another component on the eastern side (see foehn case in April 2009, Fig. a). To confirm the blocking flow regime, the Froude number is calculated as Fr=U/Nh, with h being the height of the Royal Society Range (∼ 2.5 km), U the average wind speed of the upwind flow until mountain height and N the Brunt–Väisälä frequency. We find an average Froude number of 0.2 during foehn, and it never exceeds 0.7, indicating that the low-level flow is blocked and the flow cannot ascend over the Royal Society Range but rather is diverted around it. This blocking and split of the low-level flow are also found for a clustered synoptic circulation pattern of anomalous low sea level pressure over the Ross Sea, resulting in strong synoptic southerly flow .

The low-level blocking is responsible for the along-gap pressure difference driving the gap flow and foehn signature in the northern MDV . The flow through the gap (indicated in Fig. ) is accelerated by mountain waves, leading to a foehn effect in the Taylor, Wright and Victoria valleys. The low-level blocking also results in the drawdown of warmer air from aloft on the leeward side towards Joyce Glacier (Fig. b). The strongest warming due to the isentropic drawdown is visible at the base of the north-east slope (Fig. b). This foehn warming, defined as the increase in near-surface air temperature on the leeward site compared to the temperature at the same altitude on the windward side, is on average still about 5 ∘C over Joyce Glacier. The April 2009 case in Fig. b agrees with the average potential temperature field during foehn (Fig. b) and confirms the isentropic drawdown mechanisms causing the foehn signature at Joyce Glacier.

The absence of sunlight in austral winter (April–September) ensures that the energy balance is often dominated by longwave radiative cooling (Fig. ). The average net energy loss by longwave radiation in these months is 47 W m-2, compared to a gain of energy of 40 W m-2 by sensible heating. An average QG of 6 W m-2 provides energy to the surface and cools the subsurface. The LH plays only a minor role in the winter energy balance and switches between slightly positive and negative.

March and October can be considered transition months with a decreasing or increasing contribution of solar radiation . In these months the longwave radiation remains the dominant cooling component, and ablation is limited through the small contribution of the latent heat loss and solar radiation in those months (Fig. ).

Between November and February, the largest source of energy is solar radiation. Net solar radiation (Sin(1-α)) is about 111 W m-2, of which 43 % penetrates into the subsurface (QP, Fig. ). The warmed subsurface returns energy to the surface balance through a QG of 42 W m-2. Longwave radiative cooling remains the dominant energy loss, but LH plays a larger role in summer, with 28 W m-2 on average used for sublimation. During most of summer the sign of SH is negative, indicating a relatively warm surface close to the melting point and unstable conditions. The albedo of ≈0.6 and large Snet heat the subsurface and surface and give rise to the instability in the surface layer. Although simulated surface temperatures have a positive bias in summer (Sect. ), both simulated and observed surface temperatures are often warmer than the summer air temperature, confirming the often negative sign of SH. The unstable conditions are uncommon for ablation zones of most glaciers in summer where the atmosphere is often warmer compared to the surface at the melting point. However, such unstable conditions have previously been found for blue-ice areas in Antarctica and dry-snow zones in Greenland and elsewhere . The surplus of energy available for melting is small and mostly limited to the December and January months.

The energy balance shifts from longwave radiation being dominant to sensible heat playing a large role during the 13 foehn periods in winter. The warm and windy foehn conditions almost double the contribution of sensible heating to an average flux of 70 W m-2. Longwave-equivalent cloudiness (Nϵ) is estimated from air temperature and Lin using the method of and is on average 0.38 during foehn compared to 0.15 during non-foehn conditions in winter. The cloudier conditions during winter foehn lead to longwave radiative cooling being reduced to 28 W m-2. Figure  shows this shift in the energy balance during a number of foehn events in April 2009. During the non-foehn period, radiative cooling is strong under clear-sky conditions and the subsurface cools. Down-glacier winds still retain SH values of up to 70 W m-2 during non-foehn conditions. When the foehn conditions are established, SH increases to values around 100 W m-2 (Fig. a).

During most of the winter LH is slightly positive, resulting in a small amount of deposition. The sign switches when foehn warming is initiated. On average, 8 W m-2 is used for sublimation at the surface during these windy and dry conditions. While this pattern is clearly visible in Fig.  for foehn in April 2009, the magnitude of the sublimation might be underestimated as is described in Sect. . The energy losses through sublimation and longwave radiation during winter foehn only partly compensate for the strong sensible heating, and therefore the subsurface is also warming through an average flux of -34 W m-2.

Melt at Joyce Glacier occurs almost solely in December and January, when Snet is at its peak. Melt at MDV glaciers is driven by solar radiation , and melt rarely occurs during days with clouds in the 14-month record at Joyce (Appendix , Fig. ). Of the surface melt at Joyce Glacier, 83 % occurred during clear-sky conditions (Nϵ<0.2) and only 0.5 % during overcast conditions (Nϵ>0.8). On clear-sky days Snet brings the surface temperatures to zero, allowing melt to occur. During the 2008/09 melt season, December and January had an average Snet of 137 W m-2, compared to 104 W m-2 over the 2009/10 melt season. This difference was also found for the Commonwealth, Beacon and Taylor glaciers and several valley and lake stations in the MDV . The long period of clear-sky conditions in 2008/09 and a continuous low albedo of around 0.5 resulted in a large Snet and a long period of melt occurring each day (Fig. ). During the entire 2008/09 melt season, only two foehn events close to each other are detected, showing that foehn events are not needed to generate sufficient energy for melt at Joyce Glacier. However, the foehn event appears to enhance melt rates. On 25 December 2008, energy losses from sublimation are large and ablation through sublimation during this foehn event is the highest for the entire study period. Most of December is characterized by unstable conditions, with a surface that cools through loss of SH. This changes when the foehn event sets in and the warmer, windier conditions cause a switch in the direction of SH, increasing the energy available for melt. The peak in melt energy during this season occurs just after the foehn event on 29 December 2008. At this point the surface has warmed, Snet is still large and cooling through sublimation has decreased. Importantly, the foehn event supplies additional energy to the surface and results in hourly surface temperatures being equal to zero (melting, Fig. ), leading to enhanced melt after the foehn event. Melting continues for another 2 weeks until cloudier conditions and new snowfall reduce the Snet and prohibit melt for most of the remaining season.

During the second melt season atmospheric conditions are more variable, with more clouds, and the glacier's albedo increases more frequently through snowfall events or snow redistribution (Fig. ). Net solar radiation is therefore smaller, and fewer days result in energy being available for melt (Fig. ). While most melt occurs during December and January at Joyce Glacier, a short melt event is simulated in November after a foehn event (Fig. ). The foehn warming causes a rise in the subsurface and surface temperatures. The higher surface temperatures in combination with sunny conditions following the foehn event result in the simulated melt event.

Compared to 2008/09, the 2009/10 melt season has fewer periods of large Snet, reducing the solar heating of the surface and melt occurrence. Foehn events may play a key role in the melt occurrence during such conditions. The largest melt rates in this season occur during or just after foehn events (Fig. ). Foehn conditions occur, averaged over December and January, for 6 % of the time, but surface melt during foehn conditions contributes to 23 % of the total melt. Melt occurs more frequently under foehn conditions as the foehn events provide additional energy to the surface. This is visible from the rapidly increasing surface temperatures during foehn events, bringing the surface to the melting point (Fig. b). The warming of the ice is also caused by the albedo that decreases, for example, during the foehn event on 6 January 2010 as the surface transitions from snow-covered to snow-free (grey line in Fig. a). The pattern of cloudier conditions during foehn events in winter is not evident for foehn events in December and January, where the longwave-equivalent cloudiness (Nϵ) is varying and on average ∼ 0.3 for both non-foehn and foehn conditions. Atmospheric transmission (Sin/STOA) can be used as another indicator for cloudiness and does not reveal any clear pattern in cloudiness during foehn conditions either (daily weighted average of ∼ 0.6 during both foehn and non-foehn conditions).

Not only does melt occur more frequently under foehn conditions, but the energy available for melt when the surface is at melting point is larger during time steps with foehn compared to without foehn conditions in December and January (43 W m-2 compared to 32 W m-2, respectively). The warmer foehn air results in a strong increase in the contribution of SH under melt (+31 W m-2) compared to non-foehn conditions, which is largely compensated for by an increased energy loss to sublimation (-27 W m-2). The main surplus of energy is provided by an increase in Snet (∼+12 W m-2) as the albedo is on average lower during foehn events.

Foehn events change the unique melt conditions of the MDV, where surface melt often occurs with air temperatures below zero. Figure  shows the average meteorological conditions during and outside foehn days in December and January. On days without foehn conditions, for 90 % of the hours in which the surface melts, the air temperature is below zero. Melt occurs through strong solar heating that brings the surface to melting. The net solar flux peaks just after midday, leading to diurnal variability in the stability of the surface layer over the glacier surface (Fig. ). The relatively warm surface with a colder atmosphere above leads to unstable conditions, reflected in a negative SH in Fig. c. The conditions switch to neutral during the night, when solar heating of the surface is reduced. Similarly, a diurnal cycle in atmospheric stability over soil has been found in the nearby Miers Valley, induced by topographic shading at night . During melt under foehn conditions, air temperatures are more frequently above zero (73 % of the time) and SH forms a source of energy to the surface. The distinct diurnal pattern in stability shifts to mostly stable conditions and near-neutral stability at midday during foehn days (Fig. b, d).

Foehn events also disturb the diurnal cycle in the summer sea-breeze circulation in the MDV. Under cloudless conditions, Rnet becomes negative at night and the cooled surface layer gives rise to down-glacier winds (Appendix , Fig. ). During the daytime the down-glacier south-westerly flow is replaced by stronger north-easterly onshore winds (Fig. a). A similar diurnal wind regime in the Miers Valley induced by nighttime cooling through shading of the valley floor has been observed with eddy covariance and vertical wind profiles . The diurnal wind regime is replaced by a steady south-easterly wind during foehn events (Fig. b).

The strong warm and dry wind events that characterize the MDV were previously described as katabatic winds but were later explained as foehn winds by , and . Similarly to , we found that the air mass origin of these wind events is not from the Antarctic Plateau as expected during katabatic surges but rather that the events are associated with synoptic pressure gradients due to Ross Sea cyclones. While foehn events at Joyce Glacier and the previously studied northern MDV both share this synoptic driver, we show that the forcing mechanism of the foehn events varies spatially in the complex terrain of the MDV. The different mechanism leading to the observed foehn signature in different valleys of the MDV is illustrated in Fig. . The blocking of the synoptic flow by the Royal Society Range diverts the flow at low levels (Fig. a), forcing the flow into a gap and causing foehn warming in the northern MDV, with the mechanism described by . At the same time, this low-level blocking deflects the upper cross-barrier flow and causes drawdown of potentially warmer air on the leeward side and over Joyce Glacier (Figs. b and b). Both the meteorological observations and AMPS data are used for detection of foehn events at Joyce Glacier. Foehn occurred at Joyce Glacier for 15 % of the days, which is lower than the average foehn occurrence in the Taylor, Wright and Victoria valleys. The average foehn occurrence found for the 14-month study period at Joyce Glacier might not reflect the climatological mean, but it lies within the large inter-annual variability in the northern MDV . As foehn events over the different regions of the MDV are all related to low-pressure systems, it is likely the location of these systems and the direction of the associated flow determine where foehn conditions occur within the MDV. The different driving mechanisms that occur at relatively short distances from each other in the MDV show the complex interactions of the synoptic flow with topography in the MDV.

The relative humidity dropped on average by 43 % during foehn events at Joyce Glacier, with the lowest humidity reaching 23 %. Even lower relative humidity values close to 12 % were observed by during foehn events in 2007 at Commonwealth, Canada and Howard glaciers, and values between 7 % and 17 % were observed at non-glacial MDV sites. These differences suggest that foehn drying might be weaker at Joyce Glacier. As shown for the MDV, foehn can bring air temperatures above zero, leading to temperatures of up to 5.8 ∘C at Joyce Glacier. The largest increase in air temperature during foehn at Joyce Glacier is 20.6 ∘C compared to increases of 25 ∘C found for a 2007 foehn event at Howard, Commonwealth and Canada glaciers . AWS sites in valley floors even showed temperature increases of up to 50 ∘C in the MDV during foehn events . In the Taylor, Wright and Victoria valleys, winters are characterized by cold-air pooling under stable conditions. When this cold air is eroded by foehn air, it leads to large temperature changes. Cold-pool formation is likely weaker at Joyce Glacier due to its higher location in the valley it is situated in, where down-glacier winds occur more frequently and prevent cold-air pooling. In addition, the strength of the foehn warming and drying at the surface relates to the ability of the foehn air to reach the ice surface, which is likely associated with lee-side wave activity and occurrence of hydraulic jumps. The spatial variability in foehn strength caused by different underlying foehn mechanisms should be explored in future studies.

We have presented an energy and mass balance time series from the ablation zone of a previously unstudied MDV glacier. Glaciers in the MDV have been described as challenging for SEB modelling compared to higher-melt environments. Daily variation in the energy fluxes is large, but daily mean values and melt residuals are typically small . Therefore, errors in the measured radiation components can introduce large errors into the simulated melt energy. This study is the first SEB study in the MDV to use a four-component radiometer located on a glacier. Previous energy balance modelling of MDV glaciers by and has been limited to ablation measurements and thermistor ice temperature measurements for model calibration and evaluation. The observations of Lout at Joyce Glacier allowed for model evaluation against observed surface temperatures. The good agreement between simulated and observed surface temperatures with a RMSE of 1.44 ∘C and a bias of 0.02 ∘C gives us confidence in the calculated energy fluxes.

Solar radiation plays a dominant role in the energy balance of MDV glaciers. The measured Sin and Snet at Joyce Glacier are similar to values for Taylor Glacier (Table ). While solar radiation penetration might play a small role in high-melt environments of temperate glaciers, showed the importance of including solar radiation penetration to accurately model the change in surface height of MDV glaciers. Similarly, in this study we found that solar radiation penetration needed to be included in order to reproduce the observed surface temperatures at Joyce Glacier. It was estimated that 18 % of Sin, equating to 43 % of the Snet, penetrated the subsurface for the most optimal surface temperature simulation. This value is comparable to 37 % of Snet that was found by for dry snow on top of the Greenland Ice Sheet. The percentage is smaller compared to the optimal simulations found by for Taylor Glacier using a refreezing and drainage model, where 60 % and 80 % of Snet, respectively, penetrated the subsurface. In the simulation presented in this study, density does not change when subsurface melt runs off. Including the formation of such a “weathering crust” in the surface height calculation as done by may enhance ablation and better represent observations but does not significantly affect the energy fluxes. Established methods to measure subsurface radiation, as well as internal melt rates and internal melt discharge for model calibration, are still missing, and subsurface penetration and melt remain, therefore, the largest uncertainties in MDV mass balance modelling.

The different ways to treat solar radiation penetration between the energy balance studies in the MDV are reflected in the solar and subsurface heat fluxes (Table ). When no penetration is allowed, QG remains small, while more heat is brought back to the surface for simulations with increasing rates of solar radiation penetration. The largest energy loss for all MDV glaciers was by Lnet followed by LH. LH losses to sublimation are smaller at Joyce Glacier compared to the more inland Taylor Glacier, where humidity is lower and wind speeds generally higher (Table ). The average SH at most of the MDV glaciers is close to zero (Table ), while it is more negative at Joyce Glacier, suggesting the atmosphere is often unstable. The higher elevation and lower air temperatures at Joyce Glacier ensure that energy losses through SH are larger compared to at Taylor and Canada glaciers, where the average flux is small or positive closer to the glacier terminus. The diurnal pattern in SH found in this study is comparable to a study by that showed that unstable conditions (negative SH) during the daytime lasted much longer and nighttime conditions were less stable over blue ice compared to snow surfaces.

During the two melt seasons at Joyce Glacier, melt occurred more frequently during foehn events compared to non-foehn conditions and melt energy was on average larger during foehn melt events. This agrees with findings of foehn-induced melt over the Antarctic Peninsula. There, the highest melt rates occur during foehn conditions, and foehn can contribute to 45 % of the total melt in some regions of the Larsen C shelf . The extra shortwave radiation through widespread cloud clearance during foehn events over the Antarctic Peninsula was not evident from the summer foehn events at Joyce Glacier. In fact, during winter foehn events, Joyce Glacier experienced on average cloudier conditions. The foehn warming over Joyce Glacier is dominated by isentropic drawdown, and foehn clearance might play a minor role through the strong low-level blocking by the Royal Society Range, which prohibits the warmer and more humid air from low levels from rising and condensing on the windward side . Besides, the foehn winds might not be able to penetrate fully to the surface over Joyce Glacier and might only partly erode the moist air originating from McMurdo Sound. As melt in the MDV is driven by solar radiation, future foehn studies in the MDV should investigate the cloud effects during summer foehn. There is a need for better radiation observations at the MDV AWS sites to study the sensitivity of melt with respect to clouds.

Over the Antarctic Peninsula, especially under strong foehn jets, melt can be SH driven, leading to melt even in the polar night . While this study confirms the increased SH warming and LH cooling during foehn events, melt always remains solar radiation dominated at Joyce Glacier. The strong increase in SH warming is largely compensated for by increased LH losses at Joyce Glacier, and the increased melt rates under foehn conditions more likely result from lowered surface albedo and therefore increased Snet. Distinctive for the MDV glaciers is the change from surface melt with subzero air temperatures during non-foehn conditions to melt under stable surface layer conditions during foehn events. As melt at Joyce Glacier is driven by Snet, it occurs solely around midday, when solar fluxes are peaking. The increased SH heating throughout the entire day during foehn events suggests that melt does also occur in periods with a lower solar angle, under the condition that SH is not compensated for by large LH losses. Studies covering several melt seasons are needed to confirm the impact of foehn events on melt duration and intensity in the MDV.

While melt during foehn events is still a small part of the total melt, we suspect that foehn events play an important role in melt generation in the MDV as the largest melt rates are found during or just after foehn events. At this point the ice surface has warmed and clear-sky conditions lead more frequently to melt generation compared to pre-foehn days. We hypothesize that foehn in this way “primes” the surface through warming and albedo reduction for melt generation in the MDV. Since summer temperatures are mostly below zero in the MDV, foehn occurrence likely plays a key role in the surface reaching the melting point through surface “priming”. This highlights the different role of foehn events in melt generation in the colder environment of the MDV compared to the Antarctic Peninsula, where melt is more persistent in summer. However, in a similar way foehn plays a major role in prolonging the melt season at Larsen C ice shelf and spring foehn events precondition the ice shelf for enhanced summertime melt via reduced albedo .

This study is the first to study the foehn–melt interaction in the MDV from an energy balance perspective. Further studies that include both the spatial and the temporal variability in the energy balance are needed to confirm the role of foehn occurrence in melt generation. Foehn-induced melt at the MDV might vary spatially due to interactions between synoptic flow and the local topography as well as through processes involving localized melt at topographic features such as ice cliffs . A climatological study of foehn strength and duration at the MDV glaciers is needed to better understand the impact of foehn on melt occurrence and rates. showed strong nutrient-specific responses to changes in runoff of Joyce Glacier, and therefore future studies using a “watershed” approach could provide key data on how changes to foehn events may be disproportionately impacting the wider receiving environment through enhanced hydrological and associated biogeochemical change.