WIND-DRIVEN SUBLIMATION IMPACT ON SURFACE MASS BALANCE AND ICE CORE INTERPRETATION IN EAST ANTARCTICA

Transcription

1 1 WIND-DRIVEN SUBLIMATION IMPACT ON SURFACE MASS BALANCE AND ICE CORE INTERPRETATION IN EAST ANTARCTICA Massimo Frezzotti 1, Michel Pourchet 2, Onelio Flora 3, Stefano Gandolfi 4, Michel Gay 2, Stefano Urbini 5, Christian Vincent 2, Silvia Becagli 6, Roberto Gragnani 1, Marco Proposito 1, 7, Mirko Severi 6, Rita Traversi 6, Roberto Udisti 6, Michel Fily 2 1 Ente per le Nuove Tecnologie, l Energia e l Ambiente, Progetto Clima, Rome, Italy 2 Laboratoire de Glaciologie et Géophysique de l Environnement, CNRS, Saint Martin d Hères, France 3 Dipartimento di Scienze Geologiche, Ambientali e Marine, University of Trieste, Trieste, Italy 4 Dipartimento di Ingegneria delle Strutture, dei Trasporti, delle Acque, del Rilevamento, del Territorio, University of Bologna, Bologna, Italy 5 Istituto Nazionale di Geofisica e Vulcanologia, Dipartimento per lo Studio del Territorio e delle sue Risorse, University of Genoa, Genoa, Italy 6 Dipartimento di Chimica, University of Florence, Florence, Italy 7 Dipartimento di Scienze della Terra, University of Siena, Italy *corresponding author (frezzotti@casaccia.enea.it) INDEX TERMS: 1827 Hydrology: Glaciology (1863); 1863 Hydrology: Snow and Ice (1827); 3322 Meteorology and Atmospheric Dynamics: Land/atmosphere interactions; 3344 Meteorology and Atmospheric Dynamics: Palaeoclimatology; 9310 Information related to Geographic Region: Antarctica. ABSTRACT Surface mass balance distribution and its temporal and spatial variability is an input parameter in mass balance studies. It also has important implications for palaeoclimatic series from ice cores. Different methods were adopted, compared and integrated (stake farm, core analysis, snow radar, surface morphology, remote sensing) at eight sites along a transept from Terra Nova Bay to Dome C (East Antarctica). Cores were linked by snow radar and GPS surveys to provide detailed information on spatial variability in surface mass balance. Thirty-nine cores were dated by identifying tritium/β marker levels ( ) and nssso 2-4 spikes of Tambora and Unknown volcanic events ( ) in order to provide information on temporal variability. Spatial variability measurements show that maximum snow accumulation is strictly correlated to firn temperature and that it is homogenous at macro-scales (hundreds of km 2 ). Wind-driven sublimation processes, controlled by surface slope in the wind direction, have a huge impact (up to 85% of snow precipitation) on surface mass balance and are significant in terms of past, present and future surface mass balance evaluations. The redistribution process is local and has a strong impact on annual variability of accumulation (noise). The high variability of surface mass balance is due to wind-driven sublimation; only megadunes show depositional features. The spatial variability of surface mass balance at the km scale is one order of magnitude higher than its temporal variability (20-30%) at the century scale. The reconstruction of past climates based on ice core data from areas with high spatial variability is distorted; a periodicity of about 1500 years has been calculated in megadune areas. 1

2 2 INTRODUCTION Precipitation over Antarctica is recognised as an important climate variable. Snow accumulation or surface mass balance (SMB) on the Antarctic Plateau is the sum of precipitation, sublimation/deposition and wind-blown snow. Large gaps in observations mean that any estimate of the current mass input involves a large error factor [Genthon and Krinner, 2001; Rignot and Thomas, 2002]. SMB is known to vary greatly [e.g. Richardson et al., 1997; van den Broeke et al., 1999; Frezzotti et al., 2002a]. The snow deposition process is very complicated on the Antarctic Plateau, where blizzards are severe, accumulation is low and where snow remains as aeolian particles on the surface. Representative observations of the SMB are important in estimating the characteristics of spatial and temporal variability at local scales (<10 km 2 ) and at the scale of a drainage basin. Snow redistribution changes the topography, and the topography in turn alters the wind field in a feedback system between the cryosphere and atmosphere. On the local scale, there is continual interaction between processes such as wind, snow precipitation, sublimation and SMB variations; in particular, the surface-energy balance and katabatic wind patterns are closely inter-related. Antarctica is the highest and flattest of the Earth s continents, but small changes in slope have a strong impact on wind direction and speed [Frezzotti et al., 2002a]. It has long been known that slope and curvature can play an important role in the SMB; for example, concave depressions accumulate snow at the expense of convex rises [Black and Budd, 1964; Whillans, 1975; Pettré et al., 1986; van den Broeke et al., 1999; Liston et al., 2000; Frezzotti et al., 2002a]. A large area of the plateau, where the slope along wind direction is higher than 4 m km -1, has a nil or slightly negative SMB [Frezzotti et al., 2002b]. Ablation process of snow on short and long spatial scales has a significant impact on post-depositional losses of chemical species by re-emission [Waddington and Cunningham, 1996; Wagnon et al., 1999] and on the interpretation of ice core palaeoclimatic series [Fisher et al., 1985]. Interpretation of palaeoenvironmental records extracted from Antarctic ice cores depends upon knowledge of input precipitation values [e.g. Bromwich and Weaver, 1983; Jouzel et al., 1983]. SMB studies based on numerical weather prediction models [e.g. Bromwich, 1988; Genthon and Braun, 1995; Turner et al., 1999] yield results which are comparable to those based on the interpolation of observations, but the horizontal resolution of the meteorological fields is no better than one degree and is affected by systematic biases [Gallée et al., 2001; Genthon and Krinner, 2001]. It is known that sublimation in Antarctica is not negligible [e.g. Stearns and Weidner, 1993; Bintanja, 1998; Gallée et al., 2001], and methods for describing spatial and temporal variability must be further developed [Cullather et al., 1998]. The snowdrift process is not explicitly included in numerical weather prediction and general circulation models [Gallée et al., 2001; Genthon and Krinner, 2001]. One of the biggest areas of uncertainty regarding SMB is the role of surface and wind-driven sublimation [e.g. Genthon and Krinner, 2001; Turner et al., 2002]. The fraction of precipitation on the grounded ice which is returned to the atmosphere through sublimation ranges from 6 to 25 % in high-resolution models [van den Broeke, 1997; Genthon and Krinner, 2001]. Some authors even note that, due to small-scale spatial variability in net SMB, it is impossible to determine true precipitation or SMB in large sectors of the Antarctic [Turner et al., 1999]. Observation of SMB rate variability and the study of redistribution processes not only provide input for the mass balance term but are also essential for better interpreting surface elevation change signals from satellite altimeters [Rémy et al., 2002] and for improving climate and meteorological models [e.g. Cullather et al., 1998; Genthon and Krinner, 1998, 2001; Turner et al., 1999; Gallée et al., 2001]. Atmospheric/ocean warming in the coming century may lock up greater volumes of ocean water due to increased precipitation above the Antarctic ice sheet. The SMB trend on 2

3 3 a century scale is required to evaluate Antarctica s present mass balance and how it may change over the coming decades and centuries. As part of the ITASE (International TransAntarctic Scientific Expedition) project [Mayewski and Goodwin, 1999] and in the framework of the Franco-Italian Concordia Station collaboration (between 1998 and 2000), a traverse between Terra Nova Bay (TNB) and Dome C and research at Dome C (DC) were undertaken. The study aimed to better understand latitudinal and longitudinal environmental gradients, while documenting climatic, atmospheric and surface conditions over the last years in the eastern and north-eastern portions of the DC drainage area and in northern Victoria Land. The traverse [Frezzotti and Flora, 2002] went from TNB Station ( E S) to DC (123 23'E 75 06'S, 3232 m) and started from GPS1 ( E S) on November 19, 1998, reaching the Concordia Station at DC on January 5, 1999; a distance of 1300 km was covered (Fig. 1). The party performed several tasks (drilling, glaciological and geophysical exploration, etc.) during the traverse [Frezzotti and Flora, 2002; Proposito et al., 2002; Gay et al., 2003; Becagli et al., 2003; Traversi et al., in press]. GPR (Ground Penetration Radar; snow radar), GPS (Global Position System) and snow morphology surveys were regularly carried out along the traverse [Frezzotti et al., 2002a; 2002b]. Eleven shallow ice cores were drilled within a 25 km radius of the Concordia Station between 15 December and 25 December 1999 [Vincent and Pourchet, 2000]. In 2000, GPR and GPS surveys, covering a total of about 500 km, were used to link all core sites to DC in order to provide detailed information on the spatial variability of SMB. This paper combines geophysical surveys (GPR and GPS), field and remote sensing surface observations and firn core analyses to describe SMB along the traverse and at DC. It also provides new information on the SMB process and carries implications for SMB distribution and its temporal and spatial variability, and on palaeoclimatic series from ice cores. METHODOLOGY Twenty-three shallow snow-firn cores, up to 53 m deep, were drilled during the traverse at 8 sites between TNB and DC (at intervals of km); five more were drilled at DC using an electromechanical drilling system. In seven areas (Table 1) a m deep main core was drilled at the site camp and two 7-15 m deep secondary cores were drilled 5-7 km far. The location of the secondary sites was identified in the field after a detailed GPS-GPR profile along 15 km triangular shape (Fig. 1). Snow-radar processing in the field was used to identify variability in the internal layering of the snow pack and its maximum and minimum depth. Snow temperature profiles, at depths of 15 m, were measured at main core sites after a hour stabilisation period [Frezzotti and Flora, 2002]. Snow/firn density was determined immediately after retrieval by measuring and weighing core sections. Snow was poorly sintered in the uppermost meters; density was thus measured in a pit where stratigraphic studies and snow sampling were also performed [Gay et al., 2002]. Two well-defined reference levels (tritium peak resulting from thermonuclear atmospheric bomb tests [Jouzel et al., 1979]; volcanic sulphate signals of Tambora and an Unknown source were used to determine the mean accumulation rate at the main cores along the traverse and at DC (Table 2). After surface cleaning in a cold room (- 20 C), core sections were sub-sampled every cm for sulphate (SO 2-4 ) and every 3-5 cm for tritium. Tritium depth profiles were transformed into time series by comparing them with the tritium content of precipitation at the Kaitoke (New Zealand) International Atomic Energy Agency (IAEA) station. Peaks in SO 2-4 records related to the Tambora 3

4 4 eruption of 1815 AD and the Unknown volcanic event of 1809 AD constitute the well-known Tambora doublet. In Antarctic ice cores these signatures are dated 1816 and 1810, and represent the most reliable volcanic markers for dating the last two centuries [e.g. Legrand and Delmas, 1987; Dai et al., 1991; Cole-Dai et al, 1997; Udisti et al., 2000; Stenni et al., 2002]. The volcanic signal was obtained from the sulphate profile. No correction for sea-salt sulphate was made because its contribution is always lower than 15% [Becagli et al., 2003]. Analytical procedures for chemical and tritium measurements are described elsewhere [Udisti et al., 1994; Gragnani et al., 1998; Stenni et al. 2002]. The two well-defined β radioactive references of January 1955 and January 1965 were used (Table 2) to determine the mean accumulation rate in secondary cores along the traverse and at 6 m deep at DC (Table 2). The cores were sub-sampled every cm and analysed using methods described in Pourchet et al. [1983]. At seven sites (Table 2) 40 stake farms, centred on main core sites, were geometrically positioned about 100 m apart in a cross shape over an area of 4 km 2. At DC, 37 stakes were geometrically positioned at a radius of 2.5, 5, 12.5 and 25 km from the culmination of DC. The height of the stakes was measured at Middle Point (MdPt) in 1998, 1999, 2000, 2001 and 2002, at DC in 1996, 1998 and 2000, at others sites (31Dpt, D2, D4, D6) in 1998, 2000 and 2002, at M2 in 1998 and 2002, and at GPS2 in 1993, 1996, 1998 and 2000 (Table 3). The snow accumulation at the stakes was multiplied by the snow density measured in a pit up to 2.5 m deep to obtain water equivalents (we). The integration of GPS and GPR data yields the ellipsoidal height of both the topographic surface and firn stratigraphy (Table 1; Fig. 2). GPS and GPR surveys and analyses are described elsewhere [Frezzotti et al., 2002a; Urbini et al., 2002]. For electromagnetic wave speed calculations, the depth-density relation for the snow pack was established using the density profile of three firn cores and one pit in each triangular area, and using 17 firn cores and two pits at DC. Density data for each site were input into second order polynomial functions, yielding a determination coefficient of (R 2 ) >0.9. To facilitate comparison between firn cores and GPR, the eight calculated polynomial functions were used to convert firn depths in the seven triangular areas and at DC into water-equivalent depths. The depth of the snow radar layer was converted into SMB (GPR_SA) using the depth/age ratio from the main core tritium snow accumulation. The mean depth and standard deviation were computed for GPR_SA, and minimum and maximum accumulations were recorded. Snow accumulation data derived from core records are in good agreement with data derived from GPR_SA (Table 2). In line with others authors [e.g. Richardson et al., 1997; Vaughan et al, 1999a], we assume that the layers producing strong radar reflection are isochronous. Topographic description of the site (elevation, slope, etc.) was performed using the ERS Radar Altimeter Digital Terrain Model, with a pixel size of 1 km, provided by Rémy et al. [1999]. The integration of field observations and remotely sensed data yielded the prevalent wind direction and aeolian morphology along the traverse [Frezzotti et al., 2002b]. RESULTS Core site morphological and climatological characteristics The decrease in temperature with elevation shows a near-dry adiabatic lapse rate (1.0 C 100 m -1 ) with excellent correlation (R ) along the traverse [Frezzotti and Flora, 2002]. Meteorological data and continental-scale simulation of the wind field surface [Parish and Bromwich, 1991] show that the area between TNB and DC is characterised by a constant katabatic wind flow, with speeds ranging from 6 to more than 18 m s -1 (Table 1). The Automatic Weather Station at DC and at MdPt ( shows a wind speed average of 2.8 m s -1 (maximum 17 m s -1 ) at DC and of 10 m s -1 (maximum 30 m s -1 ) at MdPt. 4

5 5 The topographic profile of the traverse indicates three sectors: an area extending about 220 km from GPS1, characterised by a step with a slope of up to 25 m km -1, a second area about 1000 km long, with a slope of up to 4.5 m km -1, and the dome area in the last 250 km, with a slope of less than 2 m km -1. The slope profile shows very high variability in the first and second areas, and a homogenous slope in the third area [Frezzotti and Flora, 2002]. Analysis of the morphological conditions of the area shows that (Fig. 2 and Table 1): - GPS2, M2 and MdPt are characterised by relatively complex morphologies with abrupt changes in slope and wind direction; - 31Dpt area presents relatively steep slopes (> 4 m km -1 ), but wind direction is close to the direction of contour lines, and the slope in the wind direction is therefore very low; - D2 and D4 present low slopes, with a angle between the wind direction and the direction of the general surface slope; the D6 core site is located just a few km leeward of the megadune field [Frezzotti et al., 2002a]; - DC has the lowest slope, and the major axis of the dome is aligned in the direction of the prevailing wind. The morphological characteristics of core sites, where topography is relatively complex (GPS2, M2 and MdPt), are very different in core sites 5 km apart (Fig. 2, Table 1): GPS2A and M2A core sites are located leeward of the hill, where the change in slope is perpendicular to the wind direction; the MdPtC core site is located at the bottom of a relatively steep slope; the M2D and GPS2B core sites are located on the windward side of the change in Slope along the Prevalent Wind Direction (SPWD). Microrelief along the TNB-DC traverse consists of 31% erosional features (wind crust), 59% redistribution features (sastrugi) and only 10% depositional features [Frezzotti et al., 2002b]. Depositional microrelief occurs extensively only in the D4 area, a point close to the David Glacier ice divide and near DC [Frezzotti et al., 2002b]. The D2 and 31Dpt areas are characterised by the alternation of sastrugi fields with sporadic longitudinal dunes (10-20 m long, a few meters wide and up to one meter high), and by seasonal wind crust with sporadic sastrugi. The GPS2A, M2A, and MdPtC core sites are characterised by the extensive presence of wind crust, consisting of a single snow-grain layer cemented by thin (0.1 to 2 mm) films of sublimated ice, with cracks (up to 2 cm wide) and polygonal patterns. Snow grain size measurements were taken at the main core site and at MdPtC [Gay et al., 2002]. Larger grain sizes are found at the GPS2A, M2A and MdPtC sites along the traverse, and even on the surface, and smaller grains are found in sastrugi or depositional areas (31DptA, MdPtA, D2A, D4A, D6A and DC). Sites MdPtC and MdPtA are only 5 km apart but show quite different grain size profiles, both in terms of mean values and variability [Gay et al., 2002]. The MdPtA site shows sastrugi of up to 20 cm in height, but no permanent wind crust. The large difference in grain size at the two sites is due to different local-scale snow accumulation processes, and is reflected by the difference in SMB (see next paragraph and Table 2). Surface mass balance and surface morphology The average SMB based on stake farms, the relative standard deviations (computed as percentages with respect to mean values) and the percentage of accumulations 0 for 2 and 5 year intervals (4 year interval at DC) are reported in Table 3. The accumulation/ablation pattern resulting from the stake farm measurements shows large standard deviations, and largely reflects snow surface roughness (sastrugi) and a hiatus in accumulation (wind crust) at all sites. This variability or noise is important, as it limits the degree to which a single annual snow accumulation value may be temporally representative [Fisher et al., 1985]. Analysis of the MdPt stake farm surveyed each year from 1998 to 2002 reveals local-scale spatial variability, suggesting that the annual local noise (meter scale) in snow accumulation could be from less than 2 times to more than 4 times the mean SMB, with accumulation 0 for 32% of annual scale observations. 5

6 6 It appears that stake measurements and surface morphology are quite consistent for all sites. The lowest standard deviation values are present when the SPWD is low (31Dpt, D4 and DC). In contrast, site M2, with a high slope variability, shows the lowest accumulation value and the highest standard deviation of 144%, with accumulation 0 for a five-year period in 40% of cases. We generally obtained the same mean value for a five-year period but a much lower variability between individual points, thus concurring with the observation by Petit et al. [1982] at the old Dome C. This is due to the fact that the mean surface roughness tends to be conserved, implying a preferential deposition of snow at low-level points and erosion of sastrugi [Gow, 1965; Petit et al., 1982; Alley, 1988]. Temporal variability of snow accumulation The different techniques used for snow accumulation determination are representative of different time periods (Table 2): - Volcanic spikes: Unknown-Tambora ; Tambora: ; - Radioactive horizons (tritium/β): ; - Stake farms: Differences between the mean snow accumulation obtained from stake farms ( ) and that obtained from the tritium marker ( ) vary according to location. The GPS2, 31Dpt, MdPt, D2 and D mean values concur well with those of (Table 2). D6 and DC present snow accumulation values from stake farms which are up to 25% higher than the average. The M2 site presents the largest difference between the measurements and the tritium marker ( ), which records 50% lower snow accumulation values. Comparison at main cores between snow accumulation derived from the tritium marker and the Tambora-Unknown volcanic marker also reveals an irregular distribution of snow accumulation: - GPS2A: very high accumulation value for the period, decreasing in accumulation by up to 33% through and ; - 31DptA: increase of about 25% from to , with a difference of about 12% between and ; - M2A: decrease since the Unknown-Tambora value, with the most significant difference between and ; - MdPtA: up to 20% variation through the different periods; - D2A: about a 30% increase between and ; - D4A: very high values between 1810 and 1816, and accumulation increased by about 25% between and (tritium analysis does not work at this site); - D6A: 20% decrease in with respect to and stake farm values; - DC: about a 10% increase between and ; stake farm and core values differ by up to 25%. Local spatial variability of snow accumulation Tritium/β marker snow accumulation data also shows a highly variable, irregular distribution between main cores and secondary cores (Table 2): - GPS2A and GPS2C cores (drilled a few dozen meters apart) show differences of about 13%, core GPS2B, drilled 5 km upstream, accumulated approximately 2.5 times more snow than the main core (GPS2A); 6

7 7-31DptA and 31DptB cores show a difference of about 20%; - M2 site presents the cores with the largest difference, with snow accumulation for core M2D about 5 times greater than that of core M2A and 1.8 times greater than that of core M2C; - MdPt site also shows very high variability in snow accumulation, with snow accumulation at core MdPtC about 5 times greater than that of MdPtB and about 30% greater than that of the MdPtA core; - D2 cores show up to 23% variability; - D4 cores show up to 30% variability; - D6 cores show up to 40% variability, with the greatest difference between D6B and D6C; - DC site shows the lowest variability (cores are distributed over an area of 50 km in diameter), with the largest difference between DC4 and DC-Doris (30%). Surface mass balance spatial variability from snow radar measurements Richardson and Holmlund [1999] demonstrate the importance of determining the spatial representativeness of cores and making radar surveys prior to drilling. Snow radar is the most useful tool for detecting spatial SMB variability, and firn core time series have the best temporal resolution. Accumulation variability along the radar profile can be surveyed by integrating the two tools. Differences in snow accumulation between main cores and secondary cores (Fig. 2) are equally reflected in tritium/β marker and snow radar data (Tables 2 and 4). The two methods yield different results because they sample different areas: the core diameter is 10 cm, whilst snow radar works at the meter scale. Major differences between the results of the two methods (core and snow radar) are found where the spatial SMB presents the highest variability (GPS2, M2 and MdPt sites). The maximum difference between core and snow radar (20%), and the highest snow radar relative standard deviation (47%) are found at site GPS2. GPS2A and GPS2C cores were drilled a few tens of meters apart and show a 13% difference for the tritium/β marker (Table 2). A relatively higher standard deviation (24%) is also present at D6, a site in a megadune area. In contrast, the lowest standard deviation was detected at DC (3%). Two SW-NE, 50 km-long snow radar profiles at DC show that layer depth increases North-eastward by about 1 m (Fig. 2, from about 9 to 10 m). The depth increase is equivalent to a SW to NE increase in the snow accumulation rate of about 0.06 kg m 2 a -1 km -1. For comparison, the Tambora marker [Legrand and Delmas, 1987] revealed a SW to NE accumulation gradient of 0.01 kg m 2 a -1 km -1 between Vostok and DC (650 km), and of 0.08 kg m 2 a -1 km -1 between DC and old Dome C (55 km NE). A SW-NE ice penetrating radar transept (800 km long) centred at DC [Siegert, 2003] shows a relatively abrupt increase in the accumulation rate just NE of DC, and this area appears to be a key area of change in the SMB. DISCUSSION Surface mass balance Snow precipitation versus surface mass balance The maximum value of SMB revealed by snow radar along the traverse presents an excellent correlation with firn temperature (Fig. 3; R snow radar and R core data). The average core data value shows low correlation with firn temperature (R ). Core and snow radar minimum values show nosignificant correlation with elevation or 7

8 8 temperature (R and 0.32), and values are clearly correlated with local effects at the km scale (SPWD). Noone et al. [1999] found that precipitation reflects larger (synoptic) scale phenomena connected to circulation on a global scale. Several authors [e.g. Robin, 1977; Muszynski and Birchfield, 1985; Fortuin and Oerlemans, 1990; Giovinetto et al., 1990] suggest that the SMB is directly correlated with temperature, elevation, saturation vapour pressure and distance from the open ocean. Fortuin and Oerlemans [1990] found that the saturation vapour pressure is by far the most important parameter in determining the distribution of snow accumulation, particularly in the interior of the continent above 1500 m, where 72% of spatial variance in snow accumulation may be explained in terms of saturation vapour pressure and surface convexity (using a 20 km surface elevation grid). On the basis of our observations, we suggest that the maximum accumulation value at each site represents a value close to snow precipitation, which is in turn very strictly correlated with climatic conditions at each site (R ), according to the following function: SA= 12.35(Tc) (1) where SA is the maximum snow accumulation (kg m 2 a 1 ) and Tc is the firn temperature at 15 m ( C). This hypothesis is confirmed by the fact that snow radar and satellite image analysis along the traverse reveals that only megadune and some occasional transversal dunes present depositional features. Frezzotti et al. [2002a] pointed out that megadunes have a wave amplitude of 2 to 4 m and wavelength of 2 to 5 km; the waves formed by variable net accumulation, ranging from 25% (lee-ward slopes) to 120% (windward slopes), of the snow accumulated in adjacent non-megadune areas. Analysis of the maximum value surveyed by snow radar shows that only the megadune area site (D6) presents a value higher than the decrease trend from the coast inland (Table 4). All other sites show no clear accumulation morphology at the km scale, but a surface morphology prevalently determined by ablation processes [Frezzotti et al., 2002a and 2002b]. The analysis of δ 18 O, performed every 5 km along the traverse (1 m core), shows a regular trend with altitude and a good correlation (R ) between δ 18 O values and site temperatures of 0.99 C -1 [Proposito et al., 2002]. It follows that the snow sampled every 5 km was not blown from far away but felt close to the sample site. Snow accumulation data (No 41 cores) were collected along the transept between Dumont d Urville (DdU) and DC [Pourchet et al., 1997] using the β marker in cores (Fig 3C). The snow accumulation versus firn temperature plot shows a very scattered distribution of data points, but with a clear trend in maximum values. On the basis of observations along the TNB-DC transept, we suggest that also for DdU-DC transept the maximum accumulation value represents a value close to snow precipitation, which is in turn very strictly correlated (R ) with firn temperature at each site, according to the following function: SA= 15.17(Tc) (2) this equation is very similar to the one calculated along the TNB-DC transept (1). Snow ablation and sublimation versus surface mass balance measurements 8

9 9 Redistribution is a local process which has a strong impact on the variability of accumulation at the annual/meter scale (i.e. noise). The high variability of SMB is linked to the ablation process. Ablation is determined by the surface sublimation process (wind scouring and, subsequently, blowing sublimation) on the plateau and by the snow blown into the sea in coastal areas. Based on these observations, we suggest that the difference between the maximum and minimum SMB value at each site represents the ablation value (Table 4): GPS2, M2 and MdPt ablation exceeds about 80% of maximum snow accumulation, whereas D6 ablation represents 58%, 31Dpt, D2 and D4 range between 20 and 33% and DC represents 12% (the lowest value) of maximum snow accumulation. A clear decrease is observed from the coast to the plateau (GPS2-DC; Max-Min absolute value in Table 4). The sites where wind action has less impact present a standard deviation of less than 10% (31Dpt, D2, D4 and DC, Table 4) and a good correlation (R ) between ablation (maximum-minimum accumulation value) and firn temperature according to the following linear equation: Ab = 3.02(Tc) (3) where Ab is snow ablation (kg m 2 a 1 ) and Tc is the firn temperature at 15 m ( C). Equation (3) could be mainly correlated with light blowing sublimation in areas with no significant change in SPWD (DC, D4, D2 and 31Dpt). Blowing sublimation is negligible at DC, where the mean wind speed is 2.8 m s -1 and maximum wind speed is 17 m s -1, but it is still an important factor in SMB at D4 (26%). The sites where wind action has greater impact (GPS2, M2, MidPt, D6) show ablation values (from 48 to 156 kg m 2 a 1 ) ranging from 58% to 85% of SA (maximum snow accumulation value; Table 4). The standard deviation in SMB at these sites is higher than 10% (Table 4); the sites have a high variability of SPWD (GPS2, M2, MdPt and D6) and show a very good correlation (R ) between ablation (maximum-minimum accumulation value) and firn temperature, according to the following function (Fig. 4A): Ab = 10.87(Tc) (4) where Ab is snow ablation (kg m 2 a 1 ) and Tc is the firn temperature at 15 m ( C). Using the maximum and minimum values for snow accumulation along the DdU-DC traverse, it is possible to calculate the analogous function (4): Ab = (Tc) (5) The ablation functions (4 and 5) along the two transepts are very similar; furthermore, the values obtained by Bintanja [1998] along transept DdU-DC using a model to calculate snow sublimation rates for year-round automatic weather stations, are very similar to function (3). Using the maximum data from both transepts (TNB-DdU-DC), the correlation is very strictly correlated (R ) with firn temperature at each site, according to the following function: SA = 16.28(Tc) (6) where SA (maximum snow accumulation) represents snow precipitation minus ablation not induced by wind. 9

10 10 The two areas are very different: the DdU DC area is more greatly influenced by cyclonic activity and presents much higher snow precipitation than in the TNB - DC area, but the SA (functions 1 and 2) and ablation functions (3, 4, and 5) show correlation with firn temperature. Both areas are affected by katabatic winds originating inland of DC [Paris and Bromwich, 1991]. Function (3) represents from 30% to 45% of function (4), with higher values inland (reduction in wind speed and turbulence), making both surface and blowing sublimation negligible at DC; nevertheless, blowing sublimation is still an important factor in SMB at D6. The empirical functions (4 and 5) point out a potential for positive feedback episodes that may produce larger rates of snow ablation (surface and blowing) than previously estimated [Giovinetto et al., 1992]. Along the TNB-DC transept, the sites with higher standard deviation and higher ablation values are extensively covered by permanent wind crust (GPS2, M2, MdPt and D6). The increase in wind speed and turbulence leads to the formation of wind crust and of a hiatus in accumulation, in a positive feedback process which amplifies the ablation process. The increase in aerodynamic roughness length with friction velocity, which usually characterises snowdrift over snow (sastrugi), is absent over wind crust. This is because surface friction associated with the movement of saltating particles is negligible over smooth surfaces [Bintanja, 2001]. Wind crust is formed by wind scouring; surface smoothness reduces friction velocity, the low albedo and the lower relative humidity determine a latent heat flux which is much higher than over snow, thereby increasing the temperature gradient in the surface layer and promoting sublimation. Both these factors have a positive feedback effect on the development and maintenance of permanent wind crust. Depth hoar layer under well-developed wind crust with cracks clearly indicates prolonged sublimation due to a hiatus in accumulation and therefore long, multi-annual, steep temperature-gradient metamorphism [Gow, 1965; Watanabe, 1978; Goodwin et al., 1994]. Sublimation and upward transport of water vapour within the subsurface snow layer cause the condensation of vapour (recrystallization) beneath the ice crust. Wind crusts and snow surfaces have distinctly different albedos, with wind crusts absorbing more solar energy [Frezzotti et al., 2002b; Cagnati et al., 2003]. Solar radiation penetrates below the wind crust and warms the subsurface snow layer; this causes an upward transport of water vapour from the subsurface snow layer, condensation of vapour and the growth of loose depth hoar below the crust [Fujii and Kusunoki, 1982]. Thermal fractures, where circulation of vapour is facilitated, allow the water vapour to be transported upward from the hoar layer. Alley [1988] found that a depth hoar layer lost 25% of its mass, probably by migrating upward to the atmosphere. Very light winds (< 2 m s -1 ) could be sufficient to initiate snowdrifts on this smooth surface, particularly in the case of fresh snow falls. Wind scouring does not allow the snow accumulation (due to surface smoothness) during the winter period, while the lower albedo determines strong sublimation during summer. King et al. [2001] pointed out that, at Halley Station, surface and blowing sublimation make roughly equal contributions to ablation (about 82 kg m 2 a 1 ). Bintanja [1999] shows that there is a clear correlation between sublimation and air temperature in blue ice areas: sublimation is known to decrease with decreasing temperature and hence with decreasing air moisture content. Stearns and Weidner [1993] estimated a net annual surface sublimation of 35 and 95 kg m 2 a 1 respectively at the South Pole Station and Ross Ice Shelf, which amount to 45 and 80% of SMB. Kobayashi et al. [1985] and Takahashi et al., [1988] estimated that the precipitation rate at Mizuho Stations (East Antarctica) was between 140 and > 200 kg m 2 a 1. The measured surface sublimation was 54 kg m 2 a 1, with a SMB of 58 kg m 2 a 1 determined by strong sublimation in summer [Fujii and Kusunoki, 1982]. The deficit between precipitation and surface sublimation at Mizuho Stations can be explained by blowing sublimation [Takahashi et al., 1988]. 10

11 11 The roughness parameter of wind crust is smaller than that of snow surfaces with sastrugi, dunes and barchans. The lesser roughness leads to wind acceleration and divergence of drifting snow [Takahashi et al., 1988]. Measurements at the Charcot camp show that snow transportation by saltation starts at wind speeds no greater than 5 m s -1 [Pettré et al., 1986]. King and Turner [1997] used Ball s formula [1960] to calculate wind speed as a function of slope gradient and inter-layer potential temperature difference (20K); wind speeds increase rapidly from 2 m s -1 to 5 m s -1 for slopes between 1 and 2 m km -1. Changes in slope and roughness in the wind direction increase turbulence, thereby increasing the diffusion coefficient and breaking the stable profile of absolute humidity through air mass mixing [Takahashi et al., 1988], which also promotes sublimation of blowing snow. Sublimation of blowing snow also occurs due to the increase in temperature determined by compression-warming during the descent of katabatic winds. Eleven m s -1 appears to be the threshold wind speed at which the sublimation of blowing snow starts to contribute substantially to katabatic flows in a feedback mechanism [Kodama et al., 1985; Wendler et al., 1993]. An increase in drifting snow leads to an increase in air density due to cooling from drifting snow sublimation and to particle incorporation, thus increasing katabatic flow speeds by another 20-30% [Kodama et al., 1985]. The threshold wind velocity which differentiates drifting snow (with snow particles moving at low levels) from blowing snow (with snow particles moving at high levels) is m s -1 ; wind speeds less than 15 m s -1 produce transverse features such as ripples, waves and barchans (depositional features), whereas those greater than 15 m s -1 produce longitudinal features such as dunes and sastrugi [Kobayashi and Ishida, 1979]. The extensive presence of erosional (wind crust) and redistribution (sastrugi) features (up to 90%) along the TNB-DC traverse [Frezzotti et al., 2002b] indicates that the threshold wind velocity of 15 m s -1 is exceeded several times over for most of the traverse. Surface and blowing sublimation increase with temperature, and blowing sublimation also increases greatly with wind speed. Additional parameters significantly influencing snowdrift sublimation are: turbulence, humidity and friction velocity [Bintanja, 1998; Bintanja and Reijmer, 2001; Pomeroy, 1989]. Moreover, high variability of slope at local scales could drive higher wind turbulence due to changes in SPWD at the km scale. Snowdrift particles are more continuously ventilated on its entire surface in turbulent conditions, and this leads to steeper temperature and humidity profiles, thereby causing higher wind-driven sublimation. In agreement with Bintanja and Reijmer [2001], wind speed and turbulence are the most important factors affecting blowing sublimation. The authors calculate that from 50% to 80% of total sublimation is due to blowing sublimation. Observations show that the erosion and snowdrift threshold depends on the properties of the snow pack and that it changes as a function of snow metamorphism: the crystal shape of freshly fallen snow does not allow large grain cohesion in snow pack, and this is reflected in relatively high snow mobility values [Gallée et al., 2001]. Satellite analysis at TNB [Zibordi and Frezzotti, 1996] and field observations along the traverse reveal that the largest blowing snow phenomena occur just after snow precipitation, before sinterization process begins. Single strong wind events greatly decrease the SMB through snowdrift sublimation, especially during summer. Comparison with existing surface mass balance map compilations The average values of SMB at the 8 sites (Table 4) have been compared with previous SMB map compilations, revealing a general over-estimation of SMB, on average from 18% [Giovinetto and Zwally, 2000] to 65% [Vaughan et al., 1999b]. The greatest differences are present at the sites from MdPt to DC with lower accumulation (Table 4). Differences between the two map compilations, with a concentration of positive residuals centred at 78 S-140 E, were already observed [Giovinetto and Zwally, 2000; Genthon and Krinner, 2001]. The difference between the two 11

12 12 compilations and data collected between DdU and DC is less marked (data are used in the compilations), with values close to survey maximum values (Sa). However, values were over-estimated also in this case, with the greatest differences in the Vaughan et al. [1999b] compilation. The explanation for the large difference between our results and those reported in the SMB map compilations could be in relation to the following: - the only accumulation value available in the BTN-DC area before our study was surveyed by the U.S. Traverse using snow-pit stratigraphy [Stuart and Heine, 1961]; some authors [Frezzotti et al., 2000; Stenni et al., 2002] pointed out an over-estimation of SMB through the use of snow-pit stratigraphy; - the sublimation process is underestimated or not even taken into account [Giovinetto et al., 1992; Genthon and Krinner, 2001]. The relatively low precipitation value inferred along transept TNB-DC could be due to the presence of DC and the ice divide that runs to Talos Dome and is perpendicular to the cyclonic tracks coming from Southern Ocean, which in part deflect cyclones and in part induce a rain shadow effect on the eastern area of DC. The SMB from ECMWF (European Centre for Medium-Range Weather Forecasts) re-analysis also shows relatively low accumulation rates [e.g. Cullather et al., 1998; Turner et al., 1999; van Lipzig et al., 2002] for this region. Temporal variability in snow accumulation The analysis of spatial variability in SMB at different sites along the traverse shows an extremely high variability of snow accumulation also at the local scale, particularly for sites with a standard deviation in SMB greater than 10% (GPS2, M2, MdPt, D6). The very high spatial variability in SMB may influence the interpretation of firn core records. Comparison of snow radar data with core data shows that the snow radar layer is isochronous. GPS2A-GPS2B cores are located along the ice flow and the distance between the two cores is about 5 km (Figs. 1 and 2). The snow accumulated at GPS2B reaches GPS2A after about 250 years (ice velocity 19 m a -1 ; Vittuari et al., in preparation). Snow radar and core analyses reveal (Fig. 2, Table 2) that snow accumulation at site GPS2B is 3 times higher (137 kg m 2 a -1 ) than that at site GPS2A (54-60 kg m 2 a -1 ). A simple 1-D model can be used to calculate past accumulation rates at GPS2A using the snow accumulation rate derived from snow radar, the depth-age function of firn cores and ice velocity. The 1-D model allows evaluation of the submergence velocity (or burial rate) of the surface and simulation of the snow accumulation rate at core GPS2A (Fig. 5). We did not take into account layer thinning due to vertical strain, because the ratio of the investigated layer depth (20 m) to the entire ice thickness [more than 3000 m; Testut et al., 2000] is less than 1% and is therefore negligible. The analysis of snow accumulation (stakes, tritium, Tambora, Unknown) reveals an increase in accumulation with depth (presently 55 kg m 2 a -1 ; Tambora Unknown 161 kg m 2 a -1 ); simulation and present snow accumulation values at GPS2B are very similar. Core site D6 is downstream of a megadune with a wavelength of about 2-3 km and amplitude of 2 to 4 m [Frezzotti et al., 2002a]. A snow radar profile 45 km west of D6 follows the ice flow direction in the megadune area and reveals the presence of buried megadunes. As for GPS2, the 1-D model (snow radar and ice velocity: 1.5 m a -1 ; Vittuari et al., in preparation] can be used to estimate the snow accumulation rate in the megadune area (Fig. 6a). The simulated accumulation rate variability show a periodical variation of about 1500 years, with a standard deviation in the snow accumulation rate of about 28%. 12

13 13 The length of periodical variations due to megadunes is strictly correlated with ice velocity and snow accumulation, and can therefore vary in space and time. As a consequence, a lack of information on local conditions (snow radar) can lead to the incorrect definition of climatic conditions based on the interpretation of core stratigraphy alone. Snow accumulation variations also have impact on the chemical (re-emission NO - 3, nsscl - and MSA), isotopic (diffusion) and physical characteristics (grain size, snow porosity, close off, total air content, etc.) of snow [e.g. Alley, 1988; Wolf, 1996; Mulvaney et al. 1998; Delmotte et al., 1999; Wagnon et al., 1999; Traversi et al., 2000; Proposito et al., 2002]. Based on this observation the sites with high standard deviation (GPS2, M2, MdPt, and D6) are not useful in providing information on temporal variations in snow accumulation, because interpretation is very difficult when the snow originates from different SMB conditions. Temporal variations in snow precipitation The sites with less than 10% standard deviation (31Dpt, D2 and D4) generally show an increase in accumulation between Tambora-present and tritium/β-present (Table 2). DC shows a clear increase in accumulation from the Tambora marker (average 25.3±1 kg m 2 a -1 ) to tritium/β (average 28.3±2.4 kg m 2 a -1 ) and the stake farm (average 39 kg m 2 a -1 ). With respect to the Tambora marker, stake farm values show a 30% increase in accumulation during the period. The tritium/β marker also records about a 10% increment. Based on numerical data for age scale from the EPICA DC1 core, the snow accumulation average over the last 5000 years is 26.6 kg m 2 a -1, with a standard deviation of 1.0 kg m 2 a -1 [Schwander et al., 2001]. Stake measurements between 1996 and 1999 show a snow accumulation of 39 kg m 2 a -1, with a standard deviation of 12.3 kg m 2 a -1. An average value of snow accumulation of 34 cm, with a standard deviation of 12 cm, was measured during the period. The absolute elevations of the stakes (measured by GPS) change homogeneously, with an average value of 0.9 cm and standard deviation of 0.9 cm during the period [Vittuari et al., in preparation]. The homogeneous decrease in elevation (0.9±0.9 cm) with respect to the snow accumulation value (34 cm) indicates that the stakes are anchored to the bottom. Although snow compaction was not taken into account when calculating snow accumulation, it has often been found to be negligible [Lorius, 1983]. Recent increases in accumulation have been reported at other Antarctic locations. Pourchet et al. [1983] observed a general 30% increase in accumulation at 14 Antarctic sites (including Vostok, old DC, the South Pole and Ross Ice Shelf) in the period This general trend was also observed some years later in the South Pole area, with a 32% increase between 1960 and 1990 [Mosley-Thompson et al., 1995]. An 11% increase in accumulation was observed at Talos Dome during the 20 th century, while from the tritium marker to the present there has been a 7% increase with respect to the 800 year average [Stenni et al., 2002]. In Wilkes Land, Morgan et al. [1991] found a decrease in the accumulation rate from 1955 to 1960, with an increase during the subsequent period. Stenni et al. [1999] reported no significant accumulation change from a 200 year ice core record in northern Victoria Land. In Dronning Maud Land, Isaksson et al. [1996] observed a decrease in accumulation from 1932 to Oerter et al. [2000] produced a composite record of accumulation rates in Dronning Maud Land for the last 200 years by stacking 12 annually-resolved records. The authors reported a decrease during the 19 th century followed by an increase during the 20 th century, and these trends were linked to temperature variations derived from stable isotope records. Domes are the preferred sites for studying the temporal variability of climate using firn/ice cores, because interpretation is easier when all the snow originates from the same point on the surface. Stenni et al. [2002] found a good level of concordance between Talos Dome and DC isotopic profiles, which were correlated with similarly distributed storm tracks and sources of moisture. The clear increase in accumulation at DC and Talos Dome, as 13

14 14 indicated by stakes and tritium/β, is consistent with that observed at other sites in Antarctica, suggesting a regional scale phenomena. We observed that the increase in accumulation is mainly evident in the inner part of the plateau, where SMB, spatial variability and the ablation process (wind-driven sublimation) produce less impact. Mosley-Thompson et al. [1995] pointed out that the observed increase is not well understood, but that there are several dominant processes that affect snow precipitation, e.g. variability in the sea-ice extent, changes in moisture source regions, frequency, duration and seasonality of cyclonic storms. In addition to these effects, wind-driven sublimation processes must be considered as factors affecting the SMB. The analysis of a 20 year ( ) surface temperature record shows a general cooling of the Antarctic continent, warming of the sea ice zone, and moderate changes over the ocean [Kwok and Comiso, 2002; Doran et al., 2002; Torinesi et al., 2003]. Wind speeds over sloped terrains increase with decreasing temperature; cold temperatures are associated with strong inversions and hence strong gravitational flows [Wendler et al., 1993]. Cooler temperatures over East Antarctica and warming in sea ice areas increase the temperature gradient, and the persistence of katabatic winds and associated wind-driven sublimation. CONCLUSIONS This paper reports on a SMB study along the TNB Dome C traverse, one of the unknown areas of East Antarctica. Different survey methods were used, compared and integrated (stake farms, core analysis, snow radar, surface morphology, remote sensing, etc.). The results provide information at different scales of spatial (meter - kilometres) and temporal (annual-secular) variability, and new insight into surface and blowing sublimation processes which affect the SMB, its temporal and spatial variability, and have implications for palaeoclimatic series from ice cores. New data was compared with that of previous SMB studies. The main findings of the paper may be summarised as follows: The redistribution process is a local effect and has strong impact on the annual scale variability of accumulation (i.e. noise). The high variability of SMB is due to ablation processes driven by katabatic winds (wind scouring, surface and blowing sublimation); only megadunes and some occasional transversal dunes present depositional features. Winddriven sublimation can have a huge impact on the SMB, which is not negligible in terms of SMB evaluation. Snow precipitation is homogenous at the macro-scale (hundreds of km 2 ), but wind-drive sublimation phenomena controlled by SPWD have considerable impact on the spatial distribution of snow at short (tens of m) and medium (km) spatial scales. The increase in wind speed and turbulence leads to the formation of wind crusts and a hiatus in accumulation, in a positive feedback effect which amplifies the ablation process. The maximum value of snow accumulation (SA) is very strictly correlated (R ) with firn temperature and represents snow precipitation minus ablation not induced by wind. Along the transepts from DdU and TNB to DC, the sites where wind action has more impact show ablation values (from 48 to 156 kg m 2 a 1 ) ranging from 58% to 85% of SA. The extensive presence along the traverses of erosional (wind crust) and redistribution (sastrugi) features demonstrates that the threshold wind velocity of 15 m s -1 is frequently exceeded along most of the traverse, which is characterised by constant katabatic winds flowing at speeds of 6 to more than 18 m s -1. More than 90% of East Antarctica presents these characteristics, with wind speeds greater than 6 m s -1 [Paris and Bromwich, 1991]. Single strong wind events greatly decrease the mass through snowdrift sublimation, especially during summer. 14