Introduction

Snowdrift is a regular phenomenon on the slopes of the Antarctic ice sheet. Particularly in the steeply inclined coastal regions, the winds are often strong enough to induce drifting and blowing snow. The redistribution of snow by the wind is, in some areas, an important factor in the surface mass balance (e.g., Radok 1970). In sufficiently strong winds, snow particles become mobile but remain in periodic contact with the surface. This process is generally referred to as saltation. In even stronger winds, particles become fully afloat and enter into suspension.

In snowdrift suspension, particle paths fluctuate around a certain average height above the surface, depending on the immersed weight of the particles, as a result of the balance between the drag invoked by turbulent motions and gravity. The existence of a drag force requires a velocity difference between the snow particles and the ambient air, which continuously ventilates the suspended particles. A snow particle in an undersaturated environment is likely to be subject to significant sublimation (e.g., Schmidt 1972). In fact, there is a close balance between the sensible heat flux directed toward and a latent heat flux directed away from the particle. Because the suspended particles have been eroded originally from the surface, snowdrift sublimation essentially represents a flux of moisture from the surface to the atmosphere and thereby affects the surface mass balance of snow surfaces.

Snowdrift sublimation is potentially very effective as compared with surface sublimation, because snowdrifting particles are continuously ventilated, and their combined exposed surface area is relatively large (e.g., Schmidt 1982). Modeling and observational studies demonstrate that snowdrift sublimation can be so large that it completely saturates the near-surface atmospheric layers (e.g., Déry et al. 1998; Mann et al. 2000; Bintanja 2001a). Saturation of the air quenches snowdrift sublimation, which is proportional to the degree of undersaturation of the environment. This mechanism represents an effective negative feedback and shows the inherent self-limiting nature of the sublimation process. Nevertheless, recent modeling studies show that snowdrift sublimation may be a nonnegligible term in the surface mass balance of the Antarctic ice sheet. Bintanja (1998) calculated snowdrift sublimation from automatic weather station data and a parameterization derived from model calculations and concluded that snowdrift sublimation may represent a significant term in the surface mass budget, especially near the coast where temperatures and wind speeds are highest. His calculations indicate that snowdrift sublimation may remove as much as 20% of the annual precipitation in those regions, which, according to current estimates (e.g., van Lipzig 1999), is as high as the removal by surface sublimation. Other studies (e.g., Smith 1995; King et al. 1996; Gallée 1998) estimate that the contribution of snowdrift sublimation is more moderate. King et al. (2000) used meteorological data from Halley station to show that snowdrift and surface sublimation rates are of equal magnitude and peak in summer [see also Bintanja (1998)]. Bintanja (2001a) used the detailed Snowdrift Suspension Turbulence Interaction Research Model (SNOWSTORM) to demonstrate that snowdrift sublimation rates are significantly higher in regions in which horizontal advection or entrainment tends to dry out the boundary layer—for example, in areas where katabatic winds prevail (e.g., Gosink 1989; van den Broeke 1997). The effect of prevailing meteorological conditions on snowdrift sublimation is unfortunately still poorly understood. As a result, there are currently great uncertainties in the evaluation of snowdrift sublimation. To reduce these uncertainties, experimental verification of snowdrift sublimation calculations is imperative. Also, the effects of snowdrift sublimation on surface sublimation through the near-surface moisture budget have received very little attention so far.

This study presents the first estimate of snowdrift sublimation based on detailed meteorological data in a katabatic wind region of the Antarctic ice sheet. Directly measured latent heat fluxes will be used to validate the calculations. Mann et al. (2000) carried out analyses, similar to those presented here, for data measured at Halley station on the Brunt ice shelf. Calculations of snowdrift and surface sublimation rates from observed meteorological data will be presented, and the characteristics of the negative feedback between moisture content and snowdrift and surface sublimation rates will be investigated. The influence that snowdrift sublimation has on the total surface–atmospheric moisture flux also will be addressed, and total sublimation rates will be compared with directly measured fluxes and ablation rates.

Location and measurements

The detailed meteorological and snowdrift observations in this paper were made in the framework of an extensive meteorological and glaciological experiment that centered on the blue-ice areas (BIAs) near the Swedish Svea station (74°11′S, 10°13′W, 1250 m above sea level). Svea is located in the Heimefrontfjella, Dronning Maud Land, Antarctica, at about 350 km from the coast, close to a large BIA (3 × 6 km²) in the Scharffenbergbotnen Valley (Fig. 1). The experiment involved seven unmanned weather stations placed at different locations (three on blue-ice areas and four over snow) to quantify spatial variations in this region. In this study, only data measured at the four snow-covered sites (sites 3, 4, 6, and 7) will be used. The focus will be on the data of site 3, because the vertical profiles are determined by measurements taken at five levels (as opposed to three levels at the other sites) and because direct turbulence and snowdrift observations were carried out here. With winds blowing from mainly easterly directions, the four sites have unobstructed snow-covered fetches of at least 10 km, which guarantees that unwanted effects caused by inhomogeneous topographical conditions are negligible.

The experimental work took place in the austral summer of 1997/98 between 29 December 1997 and 7 February 1998, during which a virtually continuous dataset was obtained at all sites [see Bintanja et al. (1998) for a full account of the activities]. Wind speed, temperature, and relative humidity (with respect to ice) were measured at five levels (nominal heights 0.5, 1, 2, 4, and 9 m above the surface) at site 3 and at three heights (1, 2, and 6 m) at sites 4, 6, and 7. The temperature and humidity sensors were artificially ventilated. Table 1 shows further details of the sensors that were used, including their precision. Wind direction, subsurface temperatures, and incoming and outgoing shortwave and longwave radiative fluxes were also determined. Sampling at these profile stations was done every 2 min (every 5 min at sites 6 and 7). Extensive calibrations before, during, and after the expedition ensured that especially the vertical profiles were measured as accurately as possible.

At site 3, direct turbulence measurements were performed using eddy-correlation sonic anemometers and Lyman-α fast-response hygrometers at 2 m above the surface. Table 2 shows details of these sensors. The sampling rate of these sensors was 20 Hz, which means that most of the scales contributing to the variances and fluxes were captured. Data were averaged to hourly means. The instruments faced east, the direction of the dominant winds in this region; in all cases with snowdrift the wind is from the east, which means that the disturbance caused by the sensors themselves is minimal. The (minor) tilting of the sensors was corrected for by rotating the data around the vertical and lateral axes so that the means of the vertical velocity and one of the horizontal velocity components become zero.

Direct measurements of turbulence during snowdrift are known to be difficult to interpret because suspended snow particles can block the signal paths of the instruments, thereby distorting the signal. The data were carefully checked to determine erroneous readings. The sonic anemometer returns data concerning the quality of the measured physical quantities. These data indicated that blocking of the signal was a regularly occurring phenomenon during snowdrift events, with more than 75% of the data being affected during the strongest winds. Nevertheless, it was decided to use all data here for a number of reasons. First, rejecting the “bad” data resulted in fluxes that exhibited erratic temporal variations, probably because so many data (mainly during strong winds) are excluded, in contrast to the situation in which all data were used. Second, time series of turbulence data were checked, and no outliers could be detected. Hourly means of horizontal wind speed, temperature, and humidity agreed very well with those measured at the profile station. Third, normalized variances of the velocity components, temperature, and specific humidity were found to vary according to standard surface layer theory (Smeets et al. 1998). If the bad data were really erroneous there would most likely be a spurious dependency on wind speed or friction velocity. Last, turbulent spectra of the various components all exhibit characteristics (e.g., slopes in the inertial subrange) that agree with those predicted by standard surface layer theory. All taken together, I believe that there are very good reasons to use all available turbulence data to determine the fluxes. Errors induced by snowdrift will probably not be very large. In any case, eliminating the bad data resulted in fluxes that do not behave in a way that can be expected: a bias towards lower wind speeds is introduced.

Direct snowdrift measurements were carried out using four snowdrift sensors, placed in a vertical array at nominal heights of 10, 20, 40, and 80 cm above the surface. These instruments were placed at site 1 until 25 January, after which they were moved to site 3, at which they operated throughout the remainder of the period. Tüg (1988) and Bintanja et al. (1998, 2001) give further details about calibration, operation, and other characteristics of the snowdrift equipment. Bintanja et al. (2001) verified that the snowdrift transport rates observed at site 3 compared fairly well with empirical data found elsewhere. It is therefore assumed that standard snowdrift theory can be applied. Hence, the snowdrift-related quantities (notably drift density) were calculated over the snow surface at site 3 from wind speed data to obtain a continuous dataset. Heights of sensors and stakes were recorded every 2–3 days. Most of the instrumentation was rigorously tested and calibrated before, during, and after the field campaign, as detailed in Bintanja et al. (1998). All data were averaged to half-hourly and hourly means—a step that is required by most surface layer theories, some of which will be applied in this study.

Located in the escarpment region between the coastal and inland plateau, the experimental area is influenced by both inland and coastal weather. As a result, the local surface winds are caused by a combination of synoptic and katabatic forcing, as demonstrated by Bintanja (2000a). Table 3 shows the mean meteorological quantities for the four stations. The average wind speeds are highest at the high-altitude stations (6 and 7), which are also the coldest sites. The relative humidity exhibits only moderate spatial variations. Apart from several smaller events, one major storm occurred in the period of 8–10 January, during which snowdrift was prominently present. Because effects related to snowdrift sublimation were strongest during this storm event, special attention will be given to this period.

Calculation of snowdrift sublimation

When snow particles are airborne in an undersaturated flow, a flux of moisture from the particle to the air develops very quickly. The inherent continuous turbulent ventilation of the particles, being exposed on all sides, greatly enhances snowdrift sublimation and thereby diminishes the particles' size and mass. Meanwhile, the ambient air becomes more humid and colder, which tends to reduce the vapor pressure difference between the particle and the ambient air and hence the sublimation rates. This effect constitutes a strong negative feedback to the process, results in strongly reduced snowdrift sublimation rates, and can eventually lead to saturation of the air. The occurrence of this mechanism in snowdrift has been demonstrated in many modeling studies (e.g., Déry et al. 1998; Bintanja 2001a) and in some observational studies (Mann et al. 2000). In this section, I will investigate whether snowdrift sublimation effects can be detected in the dataset and will attempt to quantify snowdrift sublimation rates.

Calculating sublimation of snowdrifting particles involves the following well-known relation for the sublimation-induced rate of change of mass of a single spherical particle with radius r (Thorpe and Mason 1966):

View Expanded

where RH is the relative humidity of the air (with respect to ice), F(T) is a function of absolute air temperature T (Mann et al. 2000), and Nu is the Nusselt number, which represents the effects of particle ventilation. The Nu is usually written in terms of the particle Reynolds number, which in turn depends on the mean and turbulent ventilation velocities of the particle [for details, see Lee (1975), Dover (1993), and Bintanja (2000b)]. The total snowdrift sublimation rate at a given height S(z) is then given by the sum over the entire particle size spectrum:

View Expanded

where M_r is the mass of a particle with radius r, n_r is the number of particles of size r in a unit volume of air, and ρ_p is the density of the particles. The total snowdrift sublimation rate (kg m⁻² s⁻¹) then follows from integrating (2) over z:

View Expanded

where the upper limit H is taken equal to 10 m. It is acknowledged that a significant part of the total snowdrift sublimation may take place above the 10-m level in certain conditions (e.g., Mann 1998), but no meteorological data were available to estimate the contribution above this level with any reasonable accuracy. Sublimation rate Ŝ is governed mainly by wind speed (which determines the number of suspended particles and the particle ventilation rate), the degree of undersaturation of the air, ambient temperature, and particle size distribution. Hence, to determine Ŝ, one needs to know vertical profiles of all four of these variables.

The standard power-law relation (e.g., Budd 1966) is assumed to govern the vertical distribution of the suspended-particle density:

View Expanded

where η_r is the drift density of particles of radius r, V_r is the terminal fall velocity of particles with radius r (Dover 1993), η_r0 is the drift density at reference height z_ref, u∗ is the friction velocity, and κ is the von Kármán constant. Here, z_ref is taken equal to the focus height h_f, and the empirical formulas of Pomeroy and Gray (1990) will be used to evaluate the height of, and the total drift density at, h_f (Bintanja 2001b). To distribute the total saltation drift density over the 48 size classes of 10-μm width (radii of the particle spectrum range from 0 to 480 μm), the well-known two-parameter gamma distribution function (e.g., Budd 1966) is used with the shape parameter α = 2 and a mean radius of 100 μm in the saltation layer. Table 4 shows values of the various parameters used to calculate snowdrift sublimation rates. The value of the threshold friction velocity, being the friction velocity at which snowdrift starts (and ends), was determined by Bintanja et al. (2001) using actual snowdrift data measured at site 3. They also showed that snowdrift transport rates determined from the observed drift density measurements using (4) agreed well with estimates from literature.

To determine the vertical profiles of wind speed, temperature, and relative humidity, the data of the five measuring levels were fitted to the well-known log-linear profiles (Garratt 1992). The fits generally describe the various profiles well, with high correlation coefficients, as can be seen in Fig. 2. This is the case even in strong snowdrifting conditions, although the profile is in principle not logarithmic because of the moisture source from snowdrift sublimation. This apparently is not sufficient to cause significant deviations from the logarithmic form (remember that in snowdrift there is strong vertical mixing by the strong winds), perhaps because most of the sublimation takes place below the lowest level of measurement. In any case, the current dataset shows no evidence of profiles diverging substantially from their logarithmic forms during snowdrift, and therefore it is appropriate to fit the profiles using the usual log-linear forms and to extrapolate them logarithmically down to the surface.

Something interesting and important shows up in the relative humidity profile. Under strong wind conditions, the profile reaches saturation (RH = 100%) at considerable distance from the surface (Fig. 2). I will refer to the height of the top of this saturated layer as z_sat. The air below this level is assumed to be saturated down to the surface (although there are no direct measurements to confirm it). In the dataset, z_sat values larger than 1 cm occurred exclusively during the stormy period of 8–10 January (Fig. 3a). Such a saturated layer close to the surface is almost certainly caused by high snowdrift sublimation rates and the associated large particle–air moisture fluxes. Figure 3b depicts how z_sat varies with wind speed. Clearly, it increases rapidly for wind speeds higher than about 8 m s⁻¹, the rate of increase depending on the background humidity of the air. The peak values of z_sat found at winds of about 10 m s⁻¹ were caused by advection of relatively dry air, as will be discussed in the next section. A similar saturated layer caused by snowdrift sublimation was observed at Halley station (Mann et al. 2000), although the layer could be much deeper there (up to a few meters). This result may mean that the relation between z_sat and wind speed depends on the prevailing meteorological conditions and will thus vary spatially and temporally, which implies that a functional relation between the two found at a certain location cannot a priori be applied elsewhere. In the saturated layer, snowdrift sublimation ceases by virtue of (1), demonstrating the inherent self-limiting character of the snowdrift sublimation process.

A similar feature can be detected in the observed temperature profiles. In strong winds, the near-surface temperatures become equal to the surface temperature (which can be determined from the longwave radiation measurements) at heights usually a bit larger than z_sat and closely following temporal variations in z_sat. This result probably means that a layer of uniform temperature exists close to the surface during snowdrift, implying that the sensible and latent heat fluxes are very small in this layer. I therefore assume that the air temperature below z_sat is equal to the surface temperature, noting that slight errors in temperature in or just above this layer have little effect on the value of Ŝ. The specific humidity profile was then calculated from the profiles of temperature and relative humidity. The latent heat flux at the surface, representing the surface sublimation rate LH_s (W m⁻²), is defined at a height of 1 cm, as follows:

View Expanded

where ρ_a is the air density, L_υ is the latent heat of sublimation, and q is the specific humidity of the air. Note that surface sublimation ceases if z_sat is larger than 1 cm, because the vertical gradient in specific humidity is then zero. The choice of 1 cm was a compromise: the value should not be so low that the log-linear profiles become invalid, yet it must be low enough truly to represent surface sublimation. It is evident that evaluating LH_s at a higher level would increase its value, as will be shown later.

By applying the above methods, a time series of snowdrift and surface sublimation rates was calculated for the measurements made at site 3. Figure 4 shows how Ŝ varied in time during the measuring period. Clearly, there was one distinct peak in snowdrift sublimation, coinciding with the fierce storm of 8–10 January. During this event, the calculated snowdrift sublimation rates reached peak values of 250 W m⁻² or more. These moisture flux values obviously are very large when compared with, for instance, the mean surface latent heat flux of about 10 W m⁻² as determined by Bintanja (2000d) from site-3 data through use of a surface energy balance model. In strong winds, snowdrift sublimation probably dominates the surface–atmosphere moisture flux at this location. During the remainder of the measuring period, snowdrift sublimation rates of 20–40 W m⁻² occurred regularly during brief snowdrift episodes. However, most interesting phenomena occurred between 8 and 10 January. I will further focus on this period to investigate in more detail the mechanisms involved in snowdrift sublimation.

The self-limiting nature of snowdrift sublimation

Figure 5 shows the temporal variation of various sublimation components, wind speed and direction, temperature, relative humidity, and snowdrift transport rate at site 3 for the period of 7–11 January. I have included also the surface sublimation rates determined by Bintanja (2000d), which are, in principle, valid only in nondrifting conditions. The wind speed and snowdrift transport rates indicate that the snowdrift event started in the early morning of 8 January and lasted three full days, with peak values in wind speed (15 m s⁻¹) and snowdrift transport rate in the evening of 9 January. The strong easterly winds were mainly synoptically forced, resulting from a strong depression migrating eastward along the coast at unusually southern latitudes, causing strong winds in large parts of western Dronning Maud Land (Bintanja 2000a).

Particularly interesting are the temporal variations in the sublimation components. The total snowdrift sublimation gradually increased to maximum values in the evening of 9 January, interrupted only by a gap of about 12 hours of much lower values, even though the snowdrift transport rates continued to increase. This gap can be associated with advection of relatively moist air with RH values of up to 90% during this period. Figure 5d shows that the onset and termination of this humid interval were associated with relatively small but sudden changes in wind direction. During this interval, the winds were from the east, whereas the rest of the drier period featured more northeasterly winds. After the humid period, snowdrift sublimation rates increased strongly as a result of advection of drier air. The humid interval also initiated saturated conditions close to the surface; these conditions persisted after the humid period. This fact becomes clear by comparing the relative humidity values at 5 cm and 2 m. It is evident that the snowdrift sublimation rates were sufficiently high to saturate the lowest layers in strong winds. A sudden drop in wind speed at midnight on 9 January marked the end of the saturation period. This decrease in wind speed lowered the snowdrift transport and snowdrift sublimation rates, and the relative humidity close to the surface dropped as a consequence (the relative humidity at 2 m exhibited little systematic change). This result shows that diminution of snowdrift sublimation, caused by a decrease in wind speed and the associated reduction of moisture input, lowers the moisture content of the air. It also demonstrates the strong influence of snowdrift sublimation on the moisture budget close to the surface and reveals that both wind speed and humidity govern snowdrift sublimation rates. Mann et al. (2000) reported the occurrence of essentially the same mechanism (i.e., negative feedback) for observations made at Halley station in winter.

During most of 9 January, in the fiercest part of the storm, the vertical moisture gradient at the surface became zero as the near-surface air became saturated and an isothermal layer developed. The surface latent heat flux consequently became zero during this period, leaving only the snowdrift sublimation occurring above the saturated layer to contribute to the total surface–atmosphere moisture flux. In moderately strong winds, for instance, during large parts of 8 and 10 January, surface sublimation was only slightly less than snowdrift sublimation alone, because the near-surface air was still undersaturated and thus greatly contributed to the total moisture flux. However, in moderate winds, snowdrift and surface sublimation combined were usually less than snowdrift sublimation alone in the strongest recorded winds. Bintanja (2001a) simulated snowdrift sublimation and its effect on the near-surface moisture budget. He found that the negative feedback is able to reduce surface and snowdrift sublimation rates greatly in strong winds, so that the largest total moisture fluxes occur at moderate wind speeds. However, he also found that the negative feedback is weaker and that total sublimation (essentially snowdrift sublimation) may continue to increase with wind speed in the presence of mechanisms that effectively transport dry air toward the surface and thereby produce undersaturated environments. Examples of such mechanisms are horizontal advection and entrainment of dry air, processes that are known to be able to decrease the moisture content of katabatic flows, thereby enhancing sublimation (e.g., Gosink 1989; van den Broeke 1997). Such mechanisms may very well be important at site 3.

Figure 6 shows how snowdrift and surface sublimation components vary with u∗. Snowdrift sublimation increases with friction velocity over the entire u∗ range, whereas LH_s becomes zero for u∗ values larger than about 0.55 m s⁻¹ because of the formation of the saturated layer. As a result, the dominance of snowdrift sublimation increases with wind speed. It is interesting that snowdrift sublimation is always larger than surface sublimation except for wind conditions just above the threshold of snowdrift initiation. In such conditions, the number of snowdrifting particles in the air is still very low, so that Ŝ is low. Site 3 is a location at which katabatic winds prevail even in summer (Bintanja 2000a), which may explain why the total moisture fluxes at this location are largest in the strongest recorded winds despite the expected attenuating effect due to humidity increases caused by snowdrift sublimation. In contrast, similar analyses for data gathered at Halley, located on a flat ice shelf and therefore not subject to katabatic winds, indicate that the negative feedback may dominate the near-surface moisture budget there (Mann et al. 2000). During strong winds, a saturated layer of several meters developed owing to snowdrift sublimation, which considerably reduced the surface and snowdrift sublimation rates at Halley. Given that temperatures and wind speeds are about the same (in summer), this is the main reason for why the snowdrift sublimation rates at Halley are considerably lower (maximum values 23 W m⁻²) than those at site 3 in this study. The presence of the above-mentioned mechanism that reduces the influence of relative humidity of the surface air on the total surface–atmospheric moisture flux is clearly very important, which concurs with the modeling results of Bintanja (2001a). Any future attempt to evaluate continental-scale snowdrift sublimation rates should take this mechanism into account.

During snowdrift, the latent heat flux representing the particle–air moisture flux must be balanced by a sensible heat flux to the particle. This fact indicates that the ambient air will be cooled and a downward heat flux will develop (e.g., Déry et al. 1998; Bintanja 2000c). Figure 7 depicts the sensible and latent heat fluxes as a function of wind speed, as measured by the turbulence instruments. In strong winds, the upward latent heat flux (due largely to snowdrift sublimation) is about equal to the downward sensible heat flux. This result indicates that the balance between the two fluxes is reasonably well established, which concurs with modeling results of the effect of snowdrift on the moisture budget of the lowest layers of the atmosphere (e.g., Bintanja 2001a). There is no specific reason (for instance, related to the surface energy balance) for these two fluxes to be equal in nondrift flows, which indirectly indicates that snowdrift sublimation is probably the main contributor to the latent heat flux in strong winds. This result agrees with the results presented in Fig. 5.

Mean sublimation rates and sensitivity analysis

Table 5 shows the means of surface, snowdrift, and total sublimation rates at site 3 for the entire measuring period and for the stormy period. Surface (and hence total) sublimation rates taken from Bintanja (2000d; who used standard Monin–Obukhov similarity theory to evaluate turbulent fluxes in a surface energy balance model) were used for periods without snowdrift. The means of snowdrift and surface sublimation at site 3 [over the measuring period, with snowdrift occurring 21% of the time (for u∗_t = 0.3 m s⁻¹)] are about equally important at 6.7 and 6.4 W m⁻², respectively. The surface energy balance calculations of Bintanja (2000d) yield a surface sublimation rate of 13.4 W m⁻² during snowdrift episodes, and this rate can be regarded as an upper bound of LH_s. For a lower-bound estimate, one could reason that saturation of the air reduces LH_s to zero during snowdrift. All these calculations result in minimum, best, and maximum estimates of surface sublimation, integrated over the entire measuring period, of 5.1, 6.4, and 8.0 W m⁻², respectively, which indicates that the best estimate is approximately in the middle of the two extremes. The value of the best estimate obviously depends on the height at which LH_s is evaluated (currently 1 cm), as mentioned in section 3, but not much. If this height is increased to 5 cm, the mean surface sublimation only increases to 6.8 W m⁻².

The average snowdrift sublimation rate of 6.4 W m⁻² converts to about 0.2 mm water equivalent (WE) per day. To put this number in perspective, note that Mann et al. (2000) calculated a maximum snowdrift sublimation rate (for November) corresponding to 0.11 mm WE per day, from snowdrift data gathered at Halley. Using meteorological data from Halley, King et al. (2000) calculated 0.2–0.5-mm WE per day of snowdrift sublimation during the austral summer, with surface sublimation being about equally important. At Halley, snowdrift occurs about 30% of the time, which is more often than at site 3 in this study. This difference obviously contributes to the difference in mean snowdrift sublimation rates between the two sites. The fact that actual snowdrift sublimation rates at site 3 are larger than at Halley may be associated with its location in a katabatic wind zone, in which the thermodynamic feedback mechanism that reduces sublimation is weaker, in contrast to the location of Halley.

Table 5 also shows the sensitivity of the calculated sublimation rates to changes in a selection of parameters. Snowdrift sublimation is particularly sensitive to the mean diameter of the saltating particles, with a smaller diameter yielding a higher particle number density, which through (2) leads to higher snowdrift sublimation values. A reduction in threshold friction velocity has two effects, both leading to higher snowdrift sublimation rates: 1) snowdrift occurs more often and 2) more particles are being suspended. As a result, surface sublimation decreases but not by enough to balance the increase in snowdrift sublimation. A higher value of α in the saltation layer, indicating a more symmetric particle size distribution, reduces the snowdrift sublimation rate because the absolute amount of small particle decreases. Similar changes are found when only the stormy period is considered, although the magnitude of the fluxes and the changes therein are obviously much larger.

Validation of snowdrift sublimation

How can these calculations of snowdrift sublimation be validated? This is not an easy task, because there are currently no means by which snowdrift sublimation rates can be observed directly. Therefore one must rely on indirect ways to verify snowdrift sublimation rates. There are two potential candidates in the dataset: surface ablation measured using stakes and latent heat flux measured at 2 m by the turbulence instruments.

I will first discuss the usefulness of surface ablation readings. It is interesting to observe that during the stormy period surface sublimation is only slightly larger than the mean (Table 5), whereas snowdrift sublimation is significantly increased. Readings were taken at a stake close to site 3 just before and after the storm, covering the 4-day period indicated in Table 5. The stake readings indicated that the ablation during this period was 1.13 mm WE per day. The total calculated ablation by sublimation is 1.57 mm WE per day, which is about 40% more. A number of reasons may explain this discrepancy.

• It is likely that accumulation through precipitation reduced the net mass loss, given that some precipitation was observed during the event.

• Horizontal divergences and convergences in snowdrift transport probably affected the surface mass balance.

• Snowdrift sublimation and net erosion of snowdrift need not balance in nonsteady conditions, even though they should balance exactly when integrated over time when horizontal gradients can be ignored.

The other way to check the validity of the calculated sublimation rates is to compare them with the directly measured moisture flux. The eddy-correlation instruments were mounted at a height of 2 m. I recalculated integrated snowdrift sublimation rates from the surface to 2 m. In steady-state conditions and neglecting horizontal advection, the moisture flux at 2 m must be generated below this level by both surface and snowdrift sublimation. Figure 8 shows a comparison between the directly observed and the calculated moisture fluxes. Clearly, the measured moisture flux is larger than the calculated sublimation rates during the peak of the storm (Fig. 8a), even though the general forms of both agree well. The agreement is much better in moderate winds just before and after the storm. Especially for 7 and 11 January, but also for 8 January, the agreement is very good, indicating that in such situations the calculations yield accurate results. The agreement is also reasonable for a later period with lower sublimation rates, as depicted in Fig. 8b. A number of reasons may explain the large differences found during the peak of the storm on 9 January.

• The sonic measurements are in error, because drifting snow particles interfere with the sensor beams of the sonic anemometer and the Lyman-α hygrometer, which is most likely to occur often in strong snowdrift situations; even though I had argued that this effect cannot be very large, it may still contribute to the differences.

• The assumption that LH_s is equal to zero in snowdrift may be invalid, which would mean that surface sublimation may contribute significantly in strong snowdrift situations; this possibility, however, is difficult to reconcile with the extrapolated measured humidity and temperature data.

• Snowdrift sublimation may be underestimated in strong wind situations; at first thought, this is conceivable, given that its calculation depends on the applicability of standard expressions to evaluate drift density and transport rates for the specific situation at the site. These expressions were only tested for moderate wind situations (see Bintanja et al. 2001), during which they applied reasonably well. However, the fact that the comparison in Fig. 8a is very good for moderate snowdrift conditions (e.g., the first one-half of 8 January and the latter part of 10 January) suggests that the calculations are sufficiently accurate. There is no reason to assume that the formulations to calculate drift density are not applicable at high wind speeds.

I applied the parameterization of Bintanja (1998) to calculate snowdrift sublimation rates. This method involves a fit of total snowdrift sublimation in terms of air temperature, relative humidity, and wind speed, based on calculations with a fairly sophisticated snowdrift model. The parameterization allows one to evaluate snowdrift sublimation rates from standard weather station data. The results for site 3 are shown in Fig. 9. The parameterization of Bintanja (1998) evidently produces snowdrift sublimation rates that are significantly lower when compared with those evaluated here (see also Table 5). An explanation for this discrepancy may be that Bintanja (1998) excluded mechanisms that may carry dry air to the near-surface atmosphere through, for instance, horizontal advection or entrainment as discussed before. The negative feedback reducing the snowdrift sublimation rate may consequently be too dominantly present in the parameterization, at least for conditions at site 3 for which katabatic winds prevail, in turn yielding too-low values of Ŝ. Also, the negative feedback causes snowdrift sublimation to be most important in unstationary and/or inhomogeneous cases. Thus, parameterizations such as those of Bintanja (1998) cannot be straightforwardly applied to such regions and may thus be of limited general applicability. If true, it will be difficult to find a simple general expression to evaluate snowdrift sublimation rates accurately in katabatic wind regions from automatic weather station data.

Spatial variations in snowdrift sublimation

In this section we will evaluate snowdrift sublimation rates at the other snow sites (4, 6, and 7) to infer spatial variations. In principle, the analysis described in section 3 can also be applied to the data at these sites. However, at these sites, measurements of wind, temperature, and humidity were made at only three levels. Although this is, in principle, sufficient to define a log-linear vertical profile, the accuracy of the profile will be limited. In particular, close to the surface, where a saturated layer may develop, downward extrapolation of the profiles determined by only three levels is bound to result in large errors. I have chosen to use the parameterization of Bintanja (1998) instead, which I believe will yield more accurate results. In the previous section, it has been shown that this method results in sublimation rates that are too low when compared with sublimation rates calculated from the profiles. Therefore, I have corrected the calculated sublimation rates at sites 4, 6, and 7 by multiplying the outcome of the parameterization by the ratio of both values for site 3, which is equal to 6.7/2.9 = 2.3. I think that such a procedure is justified, because differences in quantities and mechanisms implicitly used in the parameterization (such as threshold velocity, effects of advection and entrainment on the near-surface moisture and heat budgets, and saltation drift density) are unlikely to be large among the four sites. In other words, the parameterization will probably be equally wrong for all sites, which allows for a simple correction as described above.

The snowdrift sublimation rates at the four sites are shown in Fig. 10, for a period of 10 days that contains the stormy period. The storm was a mesoscale event, during which snowdrift occurred over a large region. As a result, snowdrift sublimation occurred at all sites and at a magnitude comparable to that of site 3. Differences can be attributed mainly to variations in wind speed among the sites. For instance, near 5 January the high-altitude stations (6 and 7) experienced strong winds and considerable snowdrift sublimation while no snowdrift occurred at the low-altitude sites (3 and 4). During the storm of 8–10 January, the winds at site 7 were somewhat weaker than at the other sites, resulting in lower drift densities and sublimation rates. At site 4, the snowdrift sublimation rates were very high, which is probably caused by the higher wind speeds and temperatures and lower relative humidity during the storm as compared with site 3.

Table 6 shows average values of snowdrift sublimation rates at the various sites. Most horizontal snowdrift transport occurs at the high-altitude sites (6 and 7), at which the winds are relatively strong (see Table 1). Based on this fact alone, one would expect the highest snowdrift sublimation rates to occur at these windy sites. However, site 7 experiences the least snowdrift sublimation of the four sites. This is due to the low air temperature at this high-elevation site, which tends to reduce snowdrift sublimation rates considerably. Surface sublimation is also lowest at site 7, mainly because of the low temperatures. The surface sublimation rates were tentatively evaluated by multiplying the local mean latent heat flux calculated using an energy balance model (Bintanja 2000d) by the ratio of this calculation and the estimate based on the profiles (as described above) for site 3. The resulting “new” value will be lower because of the formation of the saturated layer in snowdrift (see section 3 and Table 5). Total sublimation is lowest at the coldest, windiest site. Snowdrift sublimation exceeds surface sublimation at all sites. It is interesting that, at sites 4, 6, and 7, about two-thirds of the total sublimation is due to snowdrift sublimation during this summer month. I therefore conclude that snowdrift sublimation causes a significant part of the surface–atmosphere moisture flux in this region of Antarctica in summer. Most of the total upward moisture flux occurred during relatively short stormy periods.

Concluding remarks

This paper presents an evaluation of snowdrift sublimation rates from meteorological and snowdrift measurements performed over Antarctic snow surfaces during 37 days in the austral summer. The main findings can be summarized as follows.

• Snowdrift sublimation dominated the surface–atmosphere moisture flux in strong winds (with values of up to 250 W m⁻² at site 3), because it is a very effective mechanism (e.g., airborne particles are exposed on all sides) and because the surface sublimation ceases during snowdrift.

• High snowdrift sublimation rates, which tend to saturate the near-surface layers (up to a height of 18 cm in the strongest recorded winds) and thus lower the vertical moisture gradient at the surface, caused a reduction in surface sublimation. That this effect is more important at, for instance, Halley station near the coast may be related to the specific location of the fieldwork area, namely in a katabatic wind region.

• The data clearly demonstrate the self-limiting nature of snowdrift sublimation, in that sublimation enhances the relative humidity of the ambient air, which in turn reduces sublimation.

• The calculated sums of snowdrift and surface sublimation rates compare reasonably well with the directly measured moisture fluxes, which indicates that the calculated snowdrift sublimation rates may be realistic.

• Averaged over the entire measuring period, snowdrift and surface sublimation were equally important at site 3, despite the fact that snowdrift occurred only 21% of the time. This result demonstrates the effectiveness of snowdrift sublimation as an ablative process.

• At three other sites, snowdrift sublimation accounted for two-thirds of the total sublimation rate, which stresses its importance in the surface–atmosphere moisture flux. The highest site with the largest wind speeds and transport rates experienced the lowest snowdrift sublimation rates because of the low ambient temperatures. In general, snowdrift sublimation is a complex function of wind speed, temperature, and humidity.