Framework

The three-decade-old quest to recover ice samples from the interface zone between an ice shelf and the ocean has led investigators down several paths. Work was first done on shallower and thus more accessible ice tongues. The objective of earlier efforts was to demonstrate that sea water must be directly freezing on to the base of ice tongues due to upward heat conduction through the ice, forming congelation ice. Reference Gow and EpsteinGow and Epstein (1972) provided the first conclusive verification of this assumption. Their work was based on ice cores up to 13 m deep that had been drilled in the Koettlitz Ice Tongue, Antarctica. The same process, but on a much larger scale, was shown to take place beneath the Ross Ice Shelf (Reference Zotikov, Zagorodnov and RaikovskyZotikov and others, 1980) where the bottom-most 6 m of a 416 m ice core were composed of frozen sea water. The estimated average freezing rate at this location was inferred by Reference Zotikov, Zagorodnov and RaikovskyZotikov and others (1980) to be 2 cm a⁻¹ which was in good agreement with what Reference RobinRobin (1979) had already theoretically suggested for the Ross Ice Shelf.

Another access to interface ice was identified when Reference Kipfstuhl, Dieckmann, Oerter, Hellmer and GrafKipfstuhl and others (1992) and then Reference Warren, Roesler, Morgan, Brandt, Goodwin and AllisonWarren and others (1993) established the basal ice-shelf origin of green icebergs. The latter authors did not hypothesize on the formation mechanism of their green iceberg ice, but they did suggest the possibility that it had accreted at the base of the Amery Ice Shelf. This shelf was earlier the site of a 315 m deep coring project of which the preliminary results were presented by Reference MorganMorgan (1972). The lowermost 45 m of this core, known as G1, were composed of what Reference MorganMorgan (1972) at the time considered to be sea-water ice. However, the slow rate of this conduction-driven process made it inadequate to explain such thick accumulations of basal ice. Hence, a second mechanism was introduced when Reference RobinRobin (1979) linked for the first time the formation of basal ice at G1 with water circulation patterns in the sub-ice-shelf cavity. The process was theoretically elaborated by Reference Lewis and PerkinLewis and Perkin (1986) through their ice-pump model by which ice is melted at depth and deposited higher in the water column due to the freezing-point dependence on pressure. Later, Reference Engelhardt and DetermannEngelhardt and Determann (1987) explained how frazil-ice crystals would form in thermohaline circulation and subsequently accrete at the bottom of an ice shelf and consolidate. Modeling of melting and accretion processes at the ice-shelf/ocean interface has been further improved in recent years by several authors including Reference Hellmer and JacobsHellmer and Jacobs (1992), Reference Determann and GerdesDetermann and Gerdes (1994), Reference Bombosch and JenkinsBombosch and Jenkins (1995) and Reference Jenkins and BomboschJenkins and Bombosch (1995).

The number of available marine-ice samples and cores still remains limited. This could be explained in part by the relatively recent interest in the subject. Then there are the difficulties associated with the two possible sources of basal marine-ice samples. Green icebergs are rarely found in Nature, for the reasons discussed by Reference Warren, Roesler, Morgan, Brandt, Goodwin and AllisonWarren and others (1993) and Reference Grosfeld, Hellmer, Jonas, Sandhäger, Schulte, Vaughan, Jacobs and WeissGrosfeld and others (1998). On the other hand, drilling for marine ice at the bottom of ice shelves is confronted with the obvious necessity of having to penetrate hundreds of meters of meteoric ice. Other than at G1, bottom-ice accretion has also been found to constitute the lower 62 m of the 215 m core drilled in the Filchner–Ronne Ice Shelf at site B13 (Reference OerterOerter and others, 1992). Another 320 m core recovered further upstream from the same ice shelf at site B15 revealed the presence of a 167 m thick accretion layer at the bottom (Reference Oerter, Eicken, Kipfstuhl, Miller, Graf and OerterOerter and others, 1994). Layers of solid ice formed below meteoric ice at the bottom of ice shelves are now known by most authors as marine ice (Reference OerterOerter and others, 1992) and we follow this terminology.

As part of the 1995–96 Belgo-Italian collaboration program, a 45 m ice core was drilled out of the Nansen Ice Shelf (NIS), which is misleadingly identified as an ice sheet on official maps. The NIS core is comparable in length with the marine-ice sections of both the G1 and B13 cores. We present here the results of high-resolution multiparametric measurements showing that the properties of the entire core correspond to those of marine ice. We believe that this is the first time that a core of marine ice with a formation site so relatively near to the grounding line has been directly collected from the surface of an ice shelf. More importantly, the core’s proximity to the grounding line and its specific ice properties prompt us to consider and propose in this work the accretion of marine ice in basal crevasses opening where meteoric ice goes afloat as an active and previously undocumented process.

Setting

The NIS is located in Victoria Land, East Antarctica (Fig. 1). Its grounding line (Fig. 2) is thought to run in a roughly south-north direction across Reeves Glacier, along the eastern side of Teall Nunatak (Reference Frezzotti, Tabacco and ZirizzottiFrezzotti and others, 2000). The area of interest for this study is the branch of the Reeves Glacier flow which passes north of Teall Nunatak. In that section, surface velocities at the grounding line vary between < 50 m a⁻¹, near Teall Nunatak, and 100–150 m a⁻¹ further north, as illustrated in figure 6 of Reference Frezzotti, Tabacco and ZirizzottiFrezzotti and others (2000). Radio-echo sounding data provided by the same authors show that ice thickness ranges from 120–150 m, in the highly crevassed area located about halfway through, to 660 m further north. In addition to crevasses, rifts have opened near the grounding line. Some contain “islands” of continental ice chunks that have been frozen in place by the surrounding sea/marine ice (Fig. 3). From that point the shelf flows out into Terra Nova Bay for about 35 km to the front and is about 25 km across between Tarn Flat and Inexpressible Island. These two bedrock features together with the Northern Foothills laterally constrain the flow. The core was taken at 74°50.9′ S, 162°51.3′ E, as close as was logistically possible to the grounding line, about 7.5 km downstream from it. This position was chosen in ice outcrops shown to be of sea-water origin by preliminary tests. Typical ice-flow horizontal velocities in the vicinity of the core site were measured by Frezzotti (1992) to be about 160 m a⁻¹. In the same paper, the author estimates that the ice shelf covers an area of approximately 1800 km². Therefore, its area and average thickness make it a small to medium-sized ice shelf.

Fig. 1. Map of the NIS showing main surface features, the location of the drilling site and that of the meteorological station AWS 8931. Ice flows from the grounding line (located along the 200 m contour line around Teall Nunatak) to the front of the ice shelf indicated by the dashed line. The black area in the lower right corner of the picture is the open water of Terra Mova Bay. Background satellite image is taken from Reference Borfecchia and FrezzottiBorfecchia and Frezzotti (1991).

Fig. 2. Aerial photograph showing the morphological features of the NIS near its grounding line at the foot of Reeves Glacier. Approximate location of the drilling site is marked by a star. The larger rocky structure in the top left corner is Teall Munatak. Beyond it is Reeves Glacier flowing towards the viewer. Motice how the marine ice in the crescent-shaped outcrops occurs in a series that extends all the way from the grounding line. A more detailed photograph of the source area of these structures is shown in Figure 3. At the coring site, marine ice was at the same level as meteoric ice at the surface of the ice shelf

Fig. 3. Details of the fracture area at the origin of the flowline that passes through the location of the drilling site, 7.5 km downstream. The meteoric ice of Reeves Glacier (which is flowing towards the viewer) appears white in the upper half of the photograph, marine/sea ice is dark gray in the lower right area, and continental ice islands are visible as middle-gray flat surfaces delimited by white cliffs.

The phenomenon of lower strata of an ice shelf finding their way to the surface due to high ablation rates was first demonstrated by Reference Gow and EpsteinGow and Epstein (1972). Reference SouchezSouchez and others (1991) invoked such a process to explain the marineice nature of certain frontal sections of the Hells Gate Ice Shelf in the Terra Nova Bay area. Mass loss at the surface of an ice shelf could be induced by either melting and drainage or sublimation. The latter process is the one most likely to be prevalent in the NIS situation due to the intense and frequent katabatic wind activity. Wind velocity measurements are available from the weather station AWS 8931 (PAT) which is nearest to the core site, at 74°53′ S, 163°00′ E (Fig. 1). For the years 1989 and 1990, 41.2% of the wind blew from the southwest, the direction of the Antarctic plateau, with wind speeds exceeding 28 knots (52 km h⁻¹) for > 39% of the time (Reference BaroniBaroni, 1996). This has undoubtedly contributed to enhanced surface ablation rates in the area, estimated by Reference Frezzotti, Tabacco and ZirizzottiFrezzotti and others (2000) to be between 500 ± 100 and 400 ± 80 kg m⁻² a⁻¹ (56 ± 11 and 44 ± 9 cm a⁻¹).

Analytical Treatment

Traditionally, the three principal parameters used to establish the marine-ice identity of a body of ice are crystallography, salinity and stable isotopes.

All work on the ice core was done in a cold room kept at −25°C. Vertical thin sections 7–10 cm long were continuously prepared along the entire 45 m of core length. Sampling for salinity measurements was done at the same frequency by cutting, at positions corresponding to the top of each thin section, a volume of ice necessary to produce about 15 mL of meltwater. This high sampling frequency has never been attempted in previous Antarctic marine-ice-core studies and it insures a much enhanced insight into the variability with depth of the ice properties and a better chance of detecting any interceding layers of different properties/origin.

Thin sections were viewed and photographed between crossed polarizers and then their crystal sizes were calculated using the linear intercept method (Reference Tison, Thorsteinsson, Lorrain and KipfstuhlTison and others, 1994). Conductivity was measured with a Tacussel CD810 conductimeter used with probe XE110 (cell constant = 2.01 cm). During the conductivity measurements, the temperature of the melted samples was stabilized at 25.00°C by submerging their containers in a thermal bath. Since the temperature was not allowed to deviate in either direction by > 0.09°C, the biggest error source was the error in the conductivity of the standard KC1 solution used for calibrating the cell. Therefore, we estimate the error in the conductivity readings to be around ± 2.5%.

Mass-spectrometry analysis of the oxygen isotope composition relative to Vienna Standard Mean Ocean Water (V-SMOW) was conducted on 99 samples chosen more or less regularly along the core length and guided in part by the salinity results. Measurement accuracy is ± 0.05‰.

Evidence

The crystalline structure revealed by the thin sections is conspicuous by its complete lack of bubbles, which, for Antarctic ice, is a strong indication of its non-continental origin. Following the scheme outlined by Reference Tison, Lorrain, Bouzette, Dini, Bondesan, Stievenard and JeffriesTison and others (1998) for the classification of marine-ice types, two subcategories can be used to describe most of the facies exhibited by the NIS core crystals. Few thin sections are observed to exclusively contain one of the facies, and for the most part the two facies are observed together in different proportions.

A first facies, which is attributed a frazil-ice origin, is made of small equigranular crystals with rounded boundaries that could therefore be identified as “granular orbicular” (Fig. 4a). One of the mechanisms listed by Reference Weeks and AckleyWeeks and Ackley (1982) for the formation of frazil ice is the adiabatic drop in pressure of rising water as a result of deep thermohaline circulation processes described above. Cores B13 (Reference OerterOerter and others, 1992) and B15 (Reference Oerter, Eicken, Kipfstuhl, Miller, Graf and OerterOerter and others, 1994) also exhibit a granular facies, but some of the crystals show polygonal interlocking structure. According to Reference Eicken, Oerter, Miller, Graf and KipfstuhlEicken and others (1994), this is probably inherited from the specific time/temperature (and perhaps deformation) growth history of the ice crystals as they accrete at the bottom of the ice shelf.

Fig. 4. Crystallographic characteristics of the NIS core. Scale for photographs is shown on the bottom, (a) Granular/orbicular facies at 6.6 m depth; (b) string-lined facies at 16.6 m depth; (c) small-scale folding at 42.9 m depth; (d) profile of average crystal size with depth. Solid line in (d) is an 11-point running mean.

The second facies, which has not been reported for these other cores, is described by Reference Tison, Lorrain, Bouzette, Dini, Bondesan, Stievenard and JeffriesTison and others (1998) as “string-lined” and presents a striking feature of the NIS core. Grains belonging to this latter category are noticeable for their elongation which shows a clear preference to occur in a vertical or near-vertical direction. Most of these crystals have a distinct rectangular aspect with an elongation factor of 2.5–6 and appear in thin sections throughout the core (Fig. 4b). This is in complete opposition to what has been observed in the B13 core. Reference Eicken, Oerter, Miller, Graf and KipfstuhlEicken and others (1994) describe how most grains in the top part of B13 are elongated in a horizontal direction and how this elongation tends to disappear with depth. The occurrence of the string-lined facies in the NIS core is often accompanied by clear small-scale folding that has a wavelength and an amplitude both of the order of4 cm (Fig. 4c). Folding tends to be absent from the core segment at 17–27 m depth.

Crystal-size variation with depth is plotted in Figure 4d. The mean NIS core crystal breadth is 1.7 mm. If we were to use a rounded approximation, the corresponding average crystal cross-sectional area would be 2.7 mm². This value is distinctly lower than those reported for the marine-ice sections of the other cores. While crystal cross-sectional areas were not mentioned for G1, Reference Oerter, Eicken, Kipfstuhl, Miller, Graf and OerterOerter and others (1994) report values that vary mostly between 5 and 60 mm² for B13 and B15 crystals. Furthermore, with most NIS crystals having cross-sectional areas fluctuating between 1.1 and 3.8 mm², they also exhibit a much more confined range than that of B13 and B15. NIS core crystal size shows a very weak tendency to increase with depth in the lower third of the core, as manifested by the smoothed (11-point running mean) profile of Figure 4d. The B13 core (Reference Eicken, Oerter, Miller, Graf and KipfstuhlEicken and others, 1994), by contrast, shows a much clearer trend of increased crystal size with depth.

Both the quantitative values and the general qualitative behavior of the resulting conductivity profile with depth conform with those of marine ice. As can be seen from Figure 5a, most conductivity readings are clustered in the interval between 80 μS cm⁻¹ (0.035‰ salinity) and 300 μS cm⁻¹ (0.145‰). These values are considerably lower than typical sea-ice salinities which extend between 3‰ and 25‰ (Reference Weeks and AckleyWeeks and Ackley, 1982). Even congelation ice which is thought to have formed beneath the Ross Ice Shelf does not exhibit any salinities below 2‰ (Reference Zotikov, Zagorodnov and RaikovskyZotikov and others, 1980). On the other hand, the NIS conductivity results do overlap the ranges of 40–200 μS cm⁻¹ for B13 (Reference OerterOerter and others, 1992) and 100–210 μS cm⁻¹ for G1 (Reference MorganMorgan, 1972). Furthermore, the general trend of decreasing salinity with increasing depth echoes what has been observed in the above-cited studies. This can be seen from the smoothed (11-point running mean) profile of Figure 5a. On the other hand, it should be noted that the NIS core, with a maximum measured conductivity of 390 μS cm⁻¹, generally exhibits higher salinities than B13 and G1. The top 0.4 m of marine ice in the B15 core exhibit conductivity values as high as 390 μS cm⁻¹ (Reference Oerter, Eicken, Kipfstuhl, Miller, Graf and OerterOerter and others, 1994), but interaction with particle inclusions abundant in these layers could have occurred. Moreover, its conductivity profile rapidly drops back to a baseline value of around 40 μS cm⁻¹, similar to that of B13 (Reference Oerter, Eicken, Kipfstuhl, Miller, Graf and OerterOerter and others, 1994). The NIS core, by contrast, has a conductivity baseline value that falls from around 140 μS cm⁻¹ to 90 μS cm⁻¹ with increasing depth (Fig. 5a). However, significant deviations from the baseline of the salinity/conductivity signal occur as “bumps”, each extending over a few meters of depth along the profile. Furthermore, the high resolution of sampling has revealed the presence of smaller-scale (decimeter) fluctuations in the salinity signal which exceed the measurement error. The amplitude of this variation clearly increases with higher salinity values. Fluctuations have also been observed in sea ice, where Reference Weeks and AckleyWeeks and Ackley (1982) have emphasized the fact that even in the most homogeneous-seeming ice there is a small-scale, apparently random variation in the salinity.

Fig. 5. Profiles of conductivity (a) and δ ¹⁸O (b) for the NIS core. Solid line in (a) is an 11-point running mean through the data points.

The oxygen isotope composition results presented in Figure 5b do not show any clearly discernible trend with depth, nor do they show substantial variability (standard deviation = 0.09‰). The δ ¹⁸O values cover a range between + 1.80‰ and +2.37‰, with a mean value of +2.12‰. This corresponds well with the slightly positive values reported for G1 by Reference MorganMorgan (1972), and the value of +2‰ for B13 measured by Reference OerterOerter and others (1992). Such slightly positive δ ¹⁸O values are consistent with fractionation of sea water.

For comparison, δ ¹⁸O values for continental ice samples from Hells Gate Ice Shelf, which is part of the NIS, cover a range between −26‰ and −32‰ (Reference RonveauxRonveaux, 1992).

Plotting corresponding values in a δ ¹⁸O/conductivity diagram (Fig. 6) does not result in a significant correlation (r ² = 0.07, 99 points).

Fig. 6. δ ¹⁸O/conductivity relation for 99 samples of the NIS core. Solid line represents linear regression through all points.

Discussion

The evidence presented above strongly supports the idea that the NIS core is entirely composed of marine ice. This conclusion is reinforced by the comparison with B13, B15 and G1. It is excluded that this marine ice could have resulted from deep thermohaline circulation. Modeling work cited in the framework section predicts that melting prevails beneath an ice shelf near its grounding line. More specifically, Reference Frezzotti, Tabacco and ZirizzottiFrezzotti and others (2000) calculated discharge fluxes across a flow channel located north of the crevassed area where the marine ice originated. They deduced a mean net melting rate of the order of 0.26 ± 0.90 m a⁻¹ between the NIS grounding line and 16 km down-flow, which well includes the core site. It is difficult to envisage marine ice accreting under such conditions. Furthermore, a very crude calculation would show that even if marine ice were to form near the grounding line at 120–660 m depth, it would never have the time necessary to reach the surface, given the prevalent flow velocities and ablation rates. This last point implies to us that perhaps marine ice is being formed nearer to the surface in some feature of the bottom topography of the ice shelf. Considering the geographical location near the grounding line of the study area which is cut by several rifts reaching all the way to the bottom of the ice shelf, and that the formation of such deep rifts often requires the presence of bottom crevasses reaching the surface, we regard basal crevasses as the most plausible explanation of our observations. Bottom crevasses are common features of Antarctic ice shelves. This has been demonstrated using radar sounding methods (e.g. Reference Jezek, Bentley and CloughJezek and others, 1979). Reference Shabtaie and BentleyShabtaie and Bentley (1982) describe how bottom crevasses vary widely in distribution and dimensions, with some reaching heights up to 250 m and bottom widths of 100 m. Still more interestingly, Reference Jezek and BentleyJezek and Bentley (1983) observed that bottom crevasses in the Ross Ice Shelf generally disappear from the radar record within 100 km downstream from their point of formation. This could be taken as a strong indication of bottom crevasses being filled with ice. We therefore propose that a large basal crevasse near the grounding line would provide a sheltered setting permitting the formation of marine ice in the midst of the basal ablation zone at the grounding line. A volume of Ice Shelf Water, which should become buoyant as a result of ice-shelf melting, would enter and ascend the crevasse, thus becoming supercooled. The basal crevasse would eventually, in part at least, become filled with marine ice in a manner similar to that described above for the formation, accumulation and subsequent consolidation of frazil ice under ice shelves. A similar local ice pump, but at the front of an ice shelf, has been hypothesized by Reference Grosfeld, Hellmer, Jonas, Sandhäger, Schulte, Vaughan, Jacobs and WeissGrosfeld and others (1998) to account for the occurrence of marine ice at the bottom of green icebergs thought to have originated in the former Grand Chasm of the Filchner Ice Shelf.

A complementary mechanism that could contribute to ice formation in open rifts and basal crevasses is the possibility of ice directly forming on the walls, as a result of heat conduction into the ice shelf. Such accretion would be relatively small in the lower parts of the crevasse, where conditions are comparable to those at the interface between the bottom of an ice shelf and the water (Reference Jacobs, Gordon and ArdaiJacobs and others, 1979). In the higher parts, especially shortly after the crevasse had opened, water would be in contact with ice that could be as cold as −20°C, so more significant direct accretion is certain to occur. However, it is possible that, with time, the presence of the crevasse itself would transform the thermal regime of the ice shelf surrounding it to one more similar to that at the ice-shelf/ocean interface, thus gradually reducing the magnitude of congelation. Furthermore, in view of the generally large widths of rifts considered, congelation ice at the walls should only form a minor fraction of the total ice in the rift. Finally, in the particular case of the NIS core, the absence of ice with columnar texture, which is a distinctive sign of direct-freezing origin, shows that none of the marine ice examined here formed directly at the walls of the crevasse.

Many of the underlined observed discrepancies between the NIS core and the marine-ice segments of the other cores considered actually support the hypothesis of a local ice pump being active in the suggested morphological configuration. This includes the relatively higher general salinity of the NIS core compared with B13, B15 and G1. In the situation we are proposing, higher salinity would be explained by the higher freezing rate of the interstitial water existing among the frazil crystals after their accumulation at the top of the water column and before their consolidation. In the case of a basal crevasse, the ambient water-frazil mixture would be nearer to the surface, thus resulting in a more rapidly advancing freezing front and hence less efficient rejection of salt upon freezing. As the freezing front descends further away from the surface, the freezing rate would be reduced, resulting in better salt rejection and the observed trend of decreasing conductivity/salinity with depth. The direct relationship between the freezing rate and initial salt entrapment in sea ice was addressed by Reference Weeks, Lofgren and ŌuraWeeks and Lofgren (1967) and further developed by Reference Weeks and AckleyWeeks and Ackley (1982). On the other hand, work on stable-isotope fractionation in growing sea ice (Reference Souchez, Tison and JouzelSouchez and others, 1988; Reference Eicken and JeffriesEicken, 1998) showed an inverse relationship between the freezing rate and the δ ¹⁸O or δD) signal. The question might then be raised as to why the δ ¹⁸O isotopic profile with depth (Fig. 5b) does not show a clear analogous influence by the freezing rate, nor does it show a clear inverse correlation with salinity (Fig. 6). This could be explained by the fact that frazil crystals themselves are almost completely desalinated, such that the salinity signal results only from the intergranular brine inclusions, while the isotopic signal reflects both contributions. Therefore, even if isotopic fractionation does occur in the brine inclusions due to the advancement of a freezing front, the resulting signal will be dominated by the overwhelming and unmodified contribution from the crystals. As for the isotopic signal of the crystals themselves, it is plausible that the temperature and pressure conditions necessary for the formation of frazil ice in the successive ascending water masses are regularly satisfied at a certain specific depth in the water column, thus producing frazil crystals with more or less the same isotopic enrichment, which would account for the rather weak variability of the NIS core isotopic signal. Reference Eicken, Oerter, Miller, Graf and KipfstuhlEicken and others (1994), while noting the decreasing salinity with depth in the B13 core, exclude the advancement of a freezing front as an explanation of the salinity profile on the basis of thermodynamic constraints (conductive heat fluxes in the central Ronne Ice Shelf of the order of 0.1 W m⁻²) and of the “anomalously” low salt distribution coefficient (K _eff < 0.001) that would be needed to explain the observed salinity. The authors suggest that consolidation under the deviatoric buoyancy stress associated with the tens of meters of crystals accumulating beneath would lead to densification and expulsion of brine through fragmentation and settling of individual platelet crystals. However, in a companion paper (Tison and others, in press) we argue that both compaction and heat conduction are probably needed to account for the chemical properties of marine ice. In the near-surface formation setting for marine ice being proposed in this paper, thermodynamic growth through the progressively slowing descent of a freezing front can certainly not be neglected.

Although initial salt entrapment and stable-isotope fractionation of the NIS marine ice have been determined by thermodynamic growth, subsequent reworking has clearly taken place, as attested by the salinity profile (Fig. 5a). It is the location of the core and its geomorphological context which provide at least two possible explanations for the presence of large-scale conductivity/salinity deviations from the baseline value: the first is related to the dynamically active core-site environment, which was described above to be a zone where different ice streams converge with different velocities and thickness, thus creating a situation where the ice is subjected to lateral compression. Such a stress configuration would fold the ice body and disturb the initial stratification as sketched in Figure 7 and as witnessed by the small-scale folding features in the core (Fig. 4c). The question remains why large-scale compression would only selectively affect the part of the core above 28 m and spare the lower part, as inferred from Figure 5a. The answer could be that large-scale folding has also occurred in the lower parts of the core but is not as visible because ice strata there are characterized by low salinity contrast. A second possible explanation is related to the core’s location near the surface of the ice shelf, where its brine content could be affected by seasonal temperature changes. The resulting temperature gradients could produce brine-pocket migration (Reference Hoekstra, Osterkamp and WeeksHoekstra and others, 1965; Reference SeidenstickerSeidensticker, 1966), especially in the top few meters. The problem with this idea, however, is the slowness of the process (Reference Weeks and AckleyWeeks and Ackley, 1982; Reference Eicken, Oerter, Miller, Graf and KipfstuhlEicken and others, 1994), associated with the fact that the seasonal temperature signal does not significantly affect the temperature profile below a certain depth (in continental ice, its amplitude would only be 5% of its surface value at 10.2 m (Reference PatersonPaterson, 1994).

Fig. 7. Sketch of possible modification of the initial ice stratification due to lateral compression.

A basal crevasse, with its vertical spatial extension, would also account for the appearance of marine-ice layers on the ice-shelf surface despite proximity to the grounding line. This would also imply a relatively young age for the NIS core ice (on the order of 50 years) compared to the hundreds of years estimated for the Filchner–Ronne cores by Reference Eicken, Oerter, Miller, Graf and KipfstuhlEicken and others (1994). This would give the NIS grains less time to recrystallize, thus resulting in their relatively smaller sizes. On the other hand, the abundance of the string-lined facies and their vertical orientation could be due to grain recrystallization under the influence of a stress field applied by the walls of the crevasse. The presence of such stresses could in part be inferred from the appearance of folding in the core and its direction that suggest at least a horizontal component for the acting force.

It is worth noting that the proposed local ice-pump accretion process for marine ice in basal crevasses near the grounding line could easily be extended to the case of rifts opening in ice shelves, especially in the vicinity of ice rises, ice islands, on the side of large embayments and in the frontal regions of Antarctic shelves. Similarly, the process could occur in the regions of thinner ice that form between two converging ice streams (Reference Tison, Lorrain, Bouzette, Dini, Bondesan, Stievenard and JeffriesTison and others, 1998, especially their fig. 3). Strong evidence for such occurrence is provided by the recent work of Fricker and others (in press). The authors infer the presence of two main bands of marine ice beneath the Amery Ice Shelf; both are clearly associated with the convergence points of ice streams.

To conclude this discussion, we consider a possible alternative explanation for the properties observed in the NIS core, namely, that a body of marine ice, originally formed in a bottom crevasse upstream from the core location, is detached, tilted and lodged in another crevasse downstream with an angle relative to its initial position, as has been observed for continental ice chunks. However, such a process does not account for the general tendency of conductivity to decrease with depth, nor for the presence of vertically aligned crystals along the whole length of the core. Furthermore, ongoing work on another 45 m core, located 24.5 km downstream from the grounding line, reveals properties similar to those described in this work. That the results of such a peculiar alternative process are observed at two different sampling locations is highly improbable.

Conclusions

We have demonstrated in this work that marine ice could be more readily recovered and studied by exploiting a combination of climatic (high surface ablation rates) and accretion locations (nearer to the surface in a basal crevasse or rift) for certain ice shelves. Recent ice–ocean modeling and fieldwork efforts are increasingly underlining the importance of marine-ice formation and accretion beneath Antarctic ice shelves. Improved comprehension of these processes is indispensable for accurate mass-balance estimations of ice shelves. We have shown that an important segment of the thickness of small to medium Antarctic shelves, at certain locations, could be formed of marine ice. These same shelves could be among the first to show signs of regional warming because of their smaller thermal inertia and proportionally more rapid reduction of contact surfaces with pinning points and embayment sides. The thermal properties of Antarctic ice shelves, and hence their heat exchange with the atmosphere and/or the ocean, could be modified if a larger proportion of their mass was composed of saline, bubble-free ice instead of fresh, bubbly continental ice. Most importantly perhaps, we have proposed a new possible setting for the formation of marine ice in the vicinity of the grounding line, where basal melting would normally prevail. In this context, a modeling effort and the analysis of another core recovered from the same ice shelf are progressing. Basal crevasses and rifts are common features near the grounding lines of ice shelves and in their frontal zones as precursors for iceberg calving. Studies of the dynamic stability of Antarctic ice shelves and their fragmentation mechanisms would benefit from better insight into the interaction of shelf rifts and crevasses with the ocean. Reference HughesHughes (1983) explored the important role that these fracture features play in the disintegration of ice shelves, while Reference Stephenson and ZwallyStephenson and Zwally (1989) discussed the stabilizing effect that might result from the filling of rifts with ice. Recently, Reference Rignot and MacAyealRignot and MacAyeal (1998) and Reference MacAyeal, Rignot and HulbeMacAyeal and others (1998) have demonstrated how the dynamic properties of what they call the “ice mélange” in open rifts do play an important role in the calving process at the front and in the overall stability of an ice shelf. According to these authors, the melange is composed of multi-year sea ice, ice-shelf fragments and windblown snow. It is quite plausible that the dynamic properties of the material filling the rifts, and its response to temperature variation, would be different if the ice was mainly composed of a homogeneous body of marine ice resulting from a process such as the one described in this paper rather than the mixture described by the above authors.

Although Antarctic rifts and basal crevasses are common, their overall area probably does not exceed a small fraction of that of a big ice shelf. However, they could make a disproportionate contribution if they occur at the grounding line, which is a point of compulsory passage for continental ice on its way to the sea. There, fractures could be filled with marine ice and exported downstream before new fractures form at the same point again and the process is repeated. A hint of such a sequence of events can be seen in Figure 2. Furthermore, as indicated by Reference Corr, Popple, Robinson and OerterCorr and others (1995), considerable amounts of marine ice form at the confluence of individual ice streams that join together to form ice shelves. Such zones of steep lateral slopes are akin to a basal crevasse configuration and thus conducive to productive local ice pumps.

Finally, we believe that the high-resolution measurements presented in this work improve our perception of the variability patterns of ice-core properties with depth. Ongoing work on the three-dimensional variability of these signals should allow us to distinguish between the contribution to ice properties of the initial processes of consolidation and freezing and their subsequent modification through dynamic processes. Such an approach could help in revealing the mechanisms that lead to the comparatively very low salinities encountered in marine ice, still a subject of strong debate.