2.1 Instrumentation

The study area consists of a sloping snowfield with a break in slope at nearly 210 km inland at about 2100 m a.s.l., downstream of which D47 and D17 are located (Fig. 1). The two measurement sites are 100 km apart, south-west of the permanent French station Dumont d’Urville (66.6^∘ S, 140^∘ W; 40 m a.s.l.). Because of their remote locations, access and maintenance activities are only possible in summer. At D17, a 7 m high mast is equipped with six levels, logarithmically spaced (initial heights of 0.8, 1.3, 2, 2.8, 3.9 and 5.5 m), of anemometers and thermo-hygrometers housed in naturally ventilated MET21 radiation shields (Fig. S1, left panel). The meteorological mast is oriented toward the prevailing wind direction to prevent flow distortion by the measurement structure. The wind direction is sampled at the upper level only. Site D47 is equipped with only one level of wind speed and direction measured at 2.8 m and temperature and relative humidity measured at 2.2 m (Fig. S1, right panel). The thermo-hygrometers are factory calibrated to report relative humidity with respect to liquid water. Goff and Gratch (1945) formulae are used to convert to relative humidity with respect to ice for air temperatures below 0 ^∘C, using the sensor temperature reports in the conversion. Ultrasonic depth gauges are used to monitor surface height changes at both sites, from which the elevation of the sensors above the surface is assessed throughout the year. At D17, this information is not available before December 2012, when the height ranger was deployed. The station is currently still operative, and the instruments along the profile are raised back manually to original heights at the beginning of each summer field campaign. The remoteness and the frequently harsh weather conditions of D47 allowed for limited servicing time so that summer visits were restricted to the maintenance of sensors without raising operations. As a result the measurement heights decreased from their initial values to 1.5 m for wind speed and direction and 0.9 m for temperature and relative humidity in late December 2012, when the equipment was entirely removed. The instrument types and specificities are summarized in Table S1 in the Supplement. Data were sampled at 15 s intervals and stored at a half-hourly time resolution on a Campbell CR3000 data logger.

2.2 Climate settings

The surface climate in coastal Adélie Land is dominated by intense, frequent and persistent katabatic flows originating from the continental interior, where strong temperature inversions develop. The local topography controls the drainage of the cold, dense near-surface air as it flows downslope and accelerates toward the steep coastal escarpment over an unobstructed snow-covered fetch of several hundreds of kilometres. Table 1 lists geographical settings and climate information for the two sites. Wind speed and temperature regimes at 2 m height at the two measurement locations follow an annual cycle typical of katabatic wind confluence areas (Fig. S2). Lower temperatures and higher wind speeds are observed in winter as a result of the strong radiative deficit of the surface and increased katabatic forcing. In summer, the absorption of shortwave radiation by the surface diminishes the katabatic forcing, air temperature increases and wind speed reduces. The higher incidence of drifting snow (Fig. S3) and inherent loading of air masses with moisture through sublimation (Amory and Kittel, 2019) combined with lower temperatures in winter account for an increase in near-surface relative humidity compared to summer values. Substantially lower temperatures and subsequent dampened seasonal variations in relative humidity are observed at D47 due to the higher elevation.

Even if D17 is located near the downstream end of the sloping ice terrain where stronger katabatic forcing can be expected, year-round higher wind speeds are consistently observed at D47 some 100 km inland, as already reported by Wendler et al. (1993). Although the question remains open for further study, an explanation for this feature may involve the deceleration and subsequent thickening of the atmospheric boundary-layer flow beyond the ice-sheet margins, where it is no longer sustained by the buoyancy (katabatic) force. The resulting accumulation of cold air downstream over the ocean leads to the establishment of an upslope pressure gradient force opposing the katabatic flow that is responsible for an additional slowing of the airstream when reaching the coastal area (Gallée and Pettré, 1998), possibly accounting for the lower wind speeds at D17 compared to D47.

Both measurement sites show a very high constancy in wind direction (defined as the ratio of the resultant wind speed to the mean wind speed), reflecting the quasi-unidirectional nature of the flow in coastal Adélie Land (Table 1; Fig. S1). This provides evidence that topographic channelling strongly controls the surface wind regime and indicates that cyclonic disturbances do not significantly alter the direction of the main flow.

2.3 Drifting-snow data

2.3.4 Computation of drifting-snow frequency and mass transport

To remove electronic or turbulence noise and ensure that actual occurrences are detected, drifting snow has been considered to occur when the half-hourly mean of the snow mass flux exceeds a confidence threshold of 10⁻³ kg m⁻² s⁻¹, as determined from visual observations on the field in Adélie Land (Amory et al., 2017). Note that the same confidence threshold yielded a high level of agreement (98.6 %) between the SPC-S7 and 2G-FlowCapt™ in terms of occurrence detection in the comparison study led by Trouvilliez et al. (2015) in the Alps. Since this value remains small compared to snow mass fluxes estimated during drifting-snow occurrences (see Sect. 3.4), the confidence threshold is assumed independent of the exposed length of the sensor. The sensor is considered unburied as long as at least 10 % (i.e. 0.1 m) of its initial length remain uncovered with snow.

Changes in the exposed length of the 2G-FlowCapt™ through snow accumulation and ablation affect the estimation of the snow mass flux, as it is vertically integrated over the uncovered part of the instrument. This is a matter of concern at both sites, since precipitation along the Adélie coast occurs year-round almost exclusively in the form of snowfall, with a mean accumulation amounting to 362 mm w.e. yr⁻¹ (Agosta et al., 2012), and frequent, high wind speeds induce frequent erosion and deposition of snow. As a result, the actual sampling height varied substantially and non-uniformly throughout the measurement period, preventing direct comparisons of snow-transport amounts over time. This is accounted for in a simple way by combining, when available, half-hourly snow mass fluxes from the two measurement levels to derive a standardized estimate of the drifting-snow mass flux (i.e. vertically integrated between 0 and 2 m over the snow surface) η_DR such that

$$\begin{array}{}\text{(1)}& {\mathit{\eta }}_{\mathrm{DR}}=\left\{\begin{array}{ll}{\mathit{\eta }}_{\mathrm{1}}+{\mathit{\eta }}_{\mathrm{2}},& h_{\mathrm{1}}+h_{\mathrm{2}}\mathit{⩾}{h}_{\mathrm{ref}},\\ {\mathit{\eta }}_{\mathrm{1}}+{\mathit{\eta }}_{\mathrm{2}}\cdot \frac{h_{\mathrm{ref}}}{h_{\mathrm{1}}+h_{\mathrm{2}}},& h_{\mathrm{1}}+h_{\mathrm{2}}<h_{\mathrm{ref}},\end{array}\right.\end{array}$$

where η_i (kg m⁻² s⁻¹) is the observed snow mass flux integrated over the exposed height h_i (m) of the corresponding 2G-FlowCapt™ sensor, and h_ref=2 m corresponds to the sum of two fully exposed 1 m long 2G-FlowCapt™ sensors. In other words, when $h_{\mathrm{1}}+h_{\mathrm{2}}<\mathrm{2}$ m, it is assumed that the measured snow mass flux is constant up to 2 m. To keep consistency with the confidence threshold for the detection of drifting-snow occurrences, snow mass fluxes below 10⁻³ kg m⁻² s⁻¹ were set to zero. The horizontal snow mass transport in drift conditions for a given period of time [t₀,t_n], Q_DR, can then be written as

$$\begin{array}{}\text{(2)}& Q_{\mathrm{DR}}\left(t\right)={\displaystyle }\underset{t_{\mathrm{0}}}{\overset{t_n}{\int }}{\mathit{\eta }}_{\mathrm{DR}}\left(t\right)\mathrm{d}t.\end{array}$$

Figure 2Seasonal variability in drifting-snow frequency as recovered by the 2G-FlowCapt™ instruments. Shaded areas correspond to frequencies respectively computed using a relaxed and a stricter confidence threshold of 10⁻⁴ and 10⁻² kg m⁻² s⁻¹ and are shown as a measure of uncertainty. The absence of data at D47 during May and June 2012 is due to instrument malfunction.

Download

Figure 3CRED distribution showing the increasing probability of observing drifting snow with increasing 2 m wind speed at sites D47 (red curve) and D17 (blue curve). Shaded areas correspond to CREDs respectively computed using a relaxed and a stricter confidence threshold of 10⁻⁴ and 10⁻² kg m⁻² s⁻¹ and are shown as a measure of uncertainty.

Download

3.1 Spatial and temporal variations in drifting-snow occurrences

Monthly values of drifting-snow frequency at D47 and D17 indicate that drifting snow is a regular feature of the coastal slopes of Adélie Land (Fig. 2; overall averages of 0.81 at D47 and 0.57 at D17). Frequency values were computed for each month of the observation period as the ratio between the number of half-hourly observations with a snow mass flux at the lower, unburied level η_i higher than the confidence threshold of 10⁻³ kg m⁻² s⁻¹ and the total number of observations in that month. In each panel the shaded area corresponds to the frequency respectively computed using a relaxed and a stricter threshold of 10⁻⁴ and 10⁻² kg m⁻² s⁻¹ and is shown as a measure of uncertainty. While no particular inter-annual variability is depicted (annual averages range from 0.73 to 0.85 at D47 and 0.45 to 0.68 at D17), drifting-snow frequency varies strongly within the year, with an amplitude that can differ from year to year. Both locations experience a higher incidence of drifting snow in winter (defined here as the 8-month period between 1 March and 1 November) than during the rest of the year, a pattern quite common over Antarctica (Mahesh et al., 2003; Scarchilli et al., 2010; Gossart et al., 2017; Palm et al., 2018). At the end of winter, a gradual decrease in drifting-snow frequency is observed until a minimum is reached during summer, consistent with the annual course of wind speed (see Fig. S2, upper panel). This seasonal contrast is more pronounced at D17 than at D47 due to the stronger inhibition of erosion in summer, resulting from lower wind speeds and higher air temperatures that promote the formation of cohesive bonds holding particles to the surface (e.g. Schmidt, 1980; Amory et al., 2017). Although the use of a lowered threshold does not affect the derived frequency significantly, the stronger sensitivity to the increased threshold demonstrates the important contribution of occurrences of relatively small magnitude (i.e. $<{\mathrm{10}}^{-\mathrm{2}}$ kg m⁻² s⁻¹) to the overall frequency. This demonstrates the need to specify explicitly the chosen threshold value when computing drifting-snow frequency from 2G-FlowCapt™.

Figure 4Distribution of durations of drifting-snow events at D47 (red) and D17 (blue). The minimum values of duration and drifting-snow mass transport for an event to be retained in the statistics are respectively set to 4 h and 15 kg m⁻².

Download

Figure 5Drifting-snow mass flux against 2 m wind speed recorded at D47 (a) and D17 (b). Only periods for which two 2G-FlowCapt™ sensors are installed and/or the lower sensor is not entirely covered with snow (i.e. h₁>0.1 m) are considered.

Download

Figure 6Logarithm of snow mass transport in drift conditions against duration for each drifting-snow event recorded at D47 (red circles) and D17 (blue crosses). Only periods for which two 2G-FlowCapt™ sensors were installed and/or not entirely covered with snow are considered. Linear fits for D47 (black line) and D17 (light blue line) data are also reported on the graph, in which duration is expressed in hours.

Download

Figure 7Intra-annual variability in drifting-snow mass transport (a, b) and related drifting-snow frequency (c, d) at D47 (a, c) and D17 (b, d). The relative contribution of major drifting-snow events (see text) is highlighted in dotted lines. Summed mass transport and frequency values were first determined for each month and averaged within each monthly bin to produce monthly average values. The variability (standard deviation) is not shown due to the short length of the time series.

Download

Table 2Standardized estimates of annual horizontal snow mass transport in drift conditions.

Download Print Version | Download XLSX

The monthly values of drifting-snow frequency systematically observed 100 km inland at D47 are also higher than those close to the coastline at D17. Analysis of drift conditions documented simultaneously at D17 and D47 for the 3-year period 2010–2012 shows a significant spatial variability, with almost all drifting-snow occurrences at D17 involving drifting snow at D47, while the opposite does not hold true (Table S2). Wind speeds at D47 for which drifting snow is observed at D47 only (28.3 % of occurrences) are generally lower (average of 11.5 m s⁻¹) compared to those for which the two locations experience drifting snow simultaneously (average of 13.5 m s⁻¹). This means that the largest occurrences are seen at both sites, and the higher drifting-snow frequency at D47 is mainly due to additional occurrences of lesser magnitude for which the reduced wind speed downstream at D17 is not high enough to trigger snow transport.

3.2 Frequency of occurrence

Wind speeds at 2 m height for which drifting snow is detected (averages of 13.1 m s⁻¹ for D47 and 12.6 m s⁻¹ for D17) are generally higher than those occurring without drifting snow (averages of 7 m s⁻¹ for D47 and 6.1 m s⁻¹ for D17), although a wide range of similar wind speeds coexists between both categories. Following the approach of Baggaley and Hanesiak (2005) aiming at predicting the occurrence of snow transport from a set of common meteorological parameters, a credibility index (CRED) was used in a simpler approach to provide an estimation of the frequency of occurrence of drifting snow under specific wind conditions:

where p is the number of occurrences of drifting snow for a given wind speed range and n is the number of non-occurrences within that range. CRED varies from 0 to 1 and reflects the probability of observing drifting snow for a given range of wind speeds. Here a CRED of 0 means that no occurrence of drifting snow was observed for the selected range of wind speeds, while a CRED of 1 indicates that all wind speeds in that range were associated with drifting snow. CRED was calculated from the meteorological dataset within 1 m s⁻¹ intervals of wind speed. Occurrences observed below 2 m s⁻¹ and above 22 m s⁻¹ were not considered, since their relative proportion within each wind speed interval individually accounted for less than 1 % of the observations. As in Fig. 2 the sensitivity of CREDs to the relaxed and stricter confidence thresholds used for acknowledging the occurrence of drifting snow is illustrated by the shaded areas.

The frequency of occurrence generally increases with wind speed (Fig. 3) and typically resembles a cumulative normal distribution (Baggaley and Hanesiak, 2005). As 2G-FlowCapt™ does not provide information on the source of wind-borne snow particles, CREDs in wind speed intervals lower than 5 m s⁻¹ at D17 most likely correspond to rare occurrences detected during snowfall (without necessarily involving erosion of snow) or shortly after the deposition of a loose snow layer that is easily erodible during light-wind conditions. Then, for higher intervals, small differences in wind speed involve large variations in the CRED. At both sites, the likelihood of observing drifting snow becomes important (CRED >0.5) when wind speeds rise above 8 m s⁻¹. Wind speeds above 12 m s⁻¹ almost systematically produce drifting snow (CRED >0.9), indicating that threshold (friction velocity) values for snow transport are most often exceeded in such wind conditions. Differences in local climate between the two locations could be expected to affect CRED distributions through their influence on post-depositional processes. Lower average wind speeds at D17 could be associated with lower compaction rates enabling drifting snow to be triggered by lower wind speeds than at D47. Conversely lower drift-induced compaction at D17 could be compensated by stronger interparticle bonding resulting from higher average air temperatures. Accurate investigation of these aspects would inexorably require knowledge of snow properties at the surface. However, Fig. 3 illustrates substantially similar CRED distributions, indicating that wind speed is the main driver behind the occurrence of drifting snow at these locations.

3.3 Duration of drifting-snow events

Drifting snow occurs as long as the effective shear stress exerted on the snow surface by the overlying airstream (i.e. the friction velocity) equals or exceeds the threshold value for erosion. Concurrent snowfall and advection of snow from upwind areas can also contribute to the wind-borne snow mass. The incidence of drifting snow depends on (and is affected by changes in) flow dynamics, surface roughness, cohesion of exposed surface snow particles or more generally availability of erodible snow, the combination of which determines the magnitude of snow mass fluxes and the duration of drifting-snow events (Vionnet et al., 2013; Amory et al., 2016, 2017).

Following Vionnet et al. (2013), a drifting-snow event was defined as a period over which snow transport is detected for a minimum duration of 4 h. That is, an event is considered to start and end when the half-hourly snow mass flux at the lower unburied level η_i respectively rises above and drops below the confidence threshold of 10⁻³ kg m⁻² s⁻¹. To focus on significant drifting-snow events, an additional criterion requires that a snow mass Q_DR of at least 15 kg m⁻² (resulting from a drifting-snow mass flux η_DR at the confidence threshold of 10⁻³ kg m⁻² s⁻¹ for a duration of 4 h) is transported along the event. Note that in the case of complete burial of the lower sensor (i.e. h₁<0.1 m) or in the absence of a sensor at the second level, this criterion is applied on the snow mass transport computed from the single available level without correction. By applying this selection procedure to the whole database, 1566 and 226 drifting-snow events were respectively identified at D17 and D47. Most events do not exceed 72 h at D17 and can reach 10 d at most, while a slight proportion (7 %) of events at D47 last more than 10 d, with a maximum duration of 26 d (Fig. 4). In short, drifting-snow events are on average twice as numerous but roughly 2 times shorter at D17 (yearly average number of 173 and median duration of 15 h) than at D47 (yearly average number of 95 and median duration of 27.5 h), where stronger winds can sustain longer events. Note that these statistics are not significantly altered if the length of the time series considered for D17 is reduced to that of D47.

3.4 Horizontal drifting-snow mass transport

The drifting-snow mass flux η_DR typically tends to increase with wind speed in a power-law fashion (Fig. 5). This well-known behaviour (Radok, 1977; Mann et al., 2000) is, however, depicted with significant dispersion and notable differences between the two locations; the data at D17 show that drifting-snow mass fluxes can be of greater magnitude than at D47 for similar wind speeds and exhibit a generally higher variability along the range of wind speeds. This illustrates the diversity and spatial variability in factors controlling the wind-borne snow mass, as mentioned in the previous section. While wind speed can be used to predict the occurrence of drifting snow with quite a similar probability distribution between both locations (Fig. 3), on the other hand Fig. 5 demonstrates that more caution should be taken when scaling drifting-snow mass transport with wind speed or related single parameters independent of surface snow properties (e.g. Mann et al., 2000). Such an approach would indeed involve mixtures of power laws to capture the large variability in drifting-snow mass flux within the same wind speed interval, particularly at D17, where almost the entire range of values is observed from 15 m s⁻¹. Drifting snow is highly non-linear in nature and results essentially from the competitive balance between atmospheric drag and cohesive forces acting on the snow surface. This means that concurrent documentation of turbulence and surface snow properties is required for a better assessment of drifting-snow processes and improvements of model predictability (e.g. Baggaley and Hanesiak, 2005; Vionnet et al., 2013).

Despite the non-linear behaviour of the drifting-snow mass flux illustrated in Fig. 5, Q_DR increases linearly with the event duration (Fig. 6). Values of Q_DR were computed for each drifting-snow event identified in the database along which data from the two sensors are uninterruptedly available. Linear regression fits are shown, and their respective equations are reported on the graph. A logarithm scale is preferred for readability purposes. Figure 6 also shows that Q_DR hardly exceeds 10⁵ kg m⁻² even for the longest events, which thus seems to appear as an upper bound value for the mass transported in drift conditions during a single event. This is particularly well illustrated by D47 data. High values of Q_DR for a wide range of durations involve large snow mass fluxes recorded at the two measurement levels, indicating the regular occurrence of well-developed, non-intermittent transport events in which particles are simultaneously carried out through both the saltation and suspension mechanisms. This suggests that events of small magnitude involving low values of Q_DR and/or during which transport in saltation dominates over transport in suspension must be comparatively short-lived. This, however, cannot be substantiated by studying the two levels separately because snow mass fluxes are vertically integrated over the exposed length of the sensor, which for the sensor closest to the ground almost always largely exceeds typical saltation heights (i.e. ∼0.1 m).

On an annual basis, both kinds of events combine to produce yearly values of Q_DR close to or above 2×10⁶ kg m⁻² at both locations (Table 2). Note that annual values of Q_DR decrease only very slightly (less than 5 %) when a stricter confidence threshold is applied (i.e. only snow mass fluxes ${\mathit{\eta }}_i>{\mathrm{10}}^{-\mathrm{2}}$ kg m⁻² s⁻¹ are considered), a result of large snow mass fluxes well beyond this threshold regularly occurring during drifting-snow events. Such high estimates suggest that redistribution of snow by wind and concurrent sublimation of snow particles during transport are important components of the surface mass balance in Adélie Land (Agosta et al., 2012; Amory and Kittel, 2019).

3.5 Contribution of major drifting-snow events

The linear relationship between Q_DR and event duration illustrated in Fig. 6 can be used to distinguish the contribution of the largest events to the drifting-snow mass transport from that of the residual events. Major drifting-snow events were defined as the events whose duration is higher than the 75th percentile for each site. Figure 7 shows that such major events, preferably but not exclusively grouped in winter, account for a reduced proportion of the overall events (respectively 22 % and 24 % for D47 and D17) but mainly dictate the variability in Q_DR at the monthly scale, with the largest winter events capable of transporting up to 9 % of the annual quantity alone. The average monthly frequency resulting only from the occurrence of major events in each month is reported on the graph. As mentioned above, only the periods for which the snow mass flux was measured continuously at two levels have been considered. Note that this requirement is met for distinct periods of time between both measurement locations which thus must not be compared directly. At D17 (Fig. 7b, d), major events account for about half of the observed frequency but contribute to a larger part (>70 %) of the mass transported in drifting snow. Larger monthly values of Q_DR in winter result from an increased occurrence of major events combined with stronger snow mass fluxes (Amory et al., 2017), while drifting snow in summer mainly occurs in the form of residual events of lower magnitude. The data collected at D47 (Fig. 2, left panel) indicate that major events can contribute to an even larger part (>82 %) of the annual transport and bring a different general perspective by showing that Q_DR can be as important in summer as during some winter months, depending on the occurrence of major events. Despite a high and relatively uniform incidence of drifting snow in winter, the sharp decrease in Q_DR from June to August at D47 is due to a reduced occurrence of major events during this period. This demonstrates that high monthly values of drifting-snow frequency do not directly relate to the magnitude of snow transport, since they can mainly consist of multiple but relatively brief events involving low or moderate snow mass fluxes. This also suggests that, from a modelling perspective, representing these major events rather than the complete range of drifting-snow occurrences would be sufficient to capture the bulk of the contribution of drifting-snow processes to the local surface mass balance.