Debris-Rich Ice at the Base of the CAROLINE Ice Core

The structure of the BIL of the CAROLINE ice core consists of alternating layers of white bubbly ice devoid of particles and of pink bubbly amber ice with low to high debris/ice ratios. The thickness of individual layers varies between 10⁻² and 10⁻¹ m. The white bubbly ice looks very much like the glacier ice observed in the upper part of the core. Although it is generally particle-free, in a few cases it occurs with thin millimetre layers of dispersed particles regularly spaced every few centimetres. The pink colour of the amber ice is given by the fine silty and sandy rock particles which mainly occur as aggregates (clots) reaching 6 mm in diameter. The bubbles are smaller (<0.5mm) than in the glacier ice above and less dense. Sometimes, there is a gross internal layering, for low debris/ice ratios, with typical individual layer thicknesses from 20 to 40 mm. Rock fragments, up to 1cm in size, were also occasionally encountered in the profile. Their presence, together with the coarser fraction of the overlying Moraine Prudhomme, rules out surface deposition from the atmosphere.

Figure 2a shows the results of stable-isotope measurements in a δD–δ¹⁸O diagram. No co-isotopic measurements are presently available for the upper part of the CAROLINE ice core. However, seven samples of bubbly debris-free glacier ice were taken in the 60 cm just above the first debris-laden ice layers (open circles in Figure 2a). Although these boundary glacier-ice samples might be affected by the processes occurring at the interface, they provide our only estimate of the local precipitation slope. A linear regression for these sample points gives δD = 7.32δ¹⁸O – 30(r ² = 0.96). The slope of 7.32 is slightly lower than the usual value of 8 but this might be a bias from the small number of points. Deuterium excess calculated for each individual sample varies between −1.7 and +2.9. These values are compatible with the paleoclimatic interpretation of the base of the ice core given previously (Reference Yao, Petit, Jouzel, Lorius and DuvalYao and others, 1990), the very low isotopic values resulting from a pre-Holocene glacial maximum with expected higher relative humidity at the source area, explaining our low deuterium excess (Reference Jouzel and SouchezJouzel and others, 1982).

Fig. 2. SEM microphotographs of sand particles from the DIO ice core at 230 m depth: a. Extensively crushed quartz of the median fraction (250–500 μm) showing extensive gouges (g), steps (s) and numerous fractures, some of which are abraded; b. Quartz showing extensive dissolution etching (left) and fresh surface (right) which is heavily fractured and abraded.

All the BIL samples are distributed in a cloud centered on the glacier-ice samples. This dispersion is not peculiar to CAROLINE and was also observed amongst the most negative values measured at the base of Gl (Reference Merlivat, Lorius, Majzoub, Nief and RothMerlivat and others, 1967; their Figure 7. No obvious “freezing slope” indicating freezing-on or regelation processes at the ice-rock interface is observed in Figure 2a for the whole set of data. However, if individual debris layers, delineated by two successive white bubbly debris-free layers, are plotted separately (Figure 2b for thin laminations of dispersed debris in white bubbly ice; Figure 2c and d for two different sequences of pink amber ice); they define lower slopes (5.44, 4.79 and 4.54) but with considerable scattering (r ² = 0.61, 0.68 and 0.66, respectively), unusual for a typical refreezing event from a liquid reservoir (Reference Jouzel and SouchezJouzel and Souchez, 1982). Since, using the equation of Reference Jouzel and SouchezJouzel and Souchez (1982), theoretical freezing slopes should vary between 4.54 and 4.74 for the range of isotopic values observed above the BIL, these individual signatures could be interpreted as the result of modifications implying phase changes at the crystal boundaries, and yielding weaker co-isotopic relationships than in the case of freezing of a delimited water reservoir producing new crystals from a liquid layer. Indeed, the isotopic fractionation linked with the small-scale melting-refreezing process suggested in the present case will be more or less strongly dampened by the isotopic signal of undisturbed crystals in a given sample. Contact with freshly ground hydroxyl-bearing clay minerals or micas could also explain lower slopes on a δD–δ¹⁸O diagram (Reference Souchez, Lemmens, Lorrain, Tison, Jouzel and SugdenSouchez and others, 1990). They would, however, display much wider ranges of δs and exclusively towards more positive values, with stronger correlations and even lower linear regression coefficients.

The occurrence of phase changes along crystal boundaries of the debris-laden ice is further supported by gas measurements summarized in Table 1. There is a striking difference between debris-free and debris-laden ice both in total-gas content and CO₂ concentration. The mean total-gas value (0.097 cm³ g⁻¹ of ice for debris-free ice and 0.052 cm³ g⁻¹ of ice for debris-laden ice) are in close agreement with previous measurements on the Gl ice core performed by Reference Lorius, Raynaud and DolleLorius and others (1968). These authors gave values between 0.070 and 0.110cm³ g⁻¹ in white glacier ice and about 0.040 cm³ g⁻¹ in ice containing morainic debris. The same is valid for the CO₂ content, although recent developments seem to indicate that the “melting-refreezing” method for collecting CO₂, used by Reference Lorius, Raynaud and DolleLorius and others in 1968, might overestimate CO₂ concentrations by favouring possible reactions with carbonate particles during the experimental procedure (Reference Raynaud, Delmas, Ascencio and LegrandRaynaud and others, 1982). Nevertheless, these authors measured a mean value of about 240ppmV CO₂ in the debris-free ice and a single value of 80 ppm V in debris-laden ice, which are compatible with our concentrations listed in Table 1 (212–237 ppm V and 7–85 ppm V, respectively). The CO₂ concentrations found in debris-free ice are well within the range of the values measured in the Vostok core between 73 and 106 kyear BP, while the CO₂ values found in the debris-laden layers are much lower than the lowest value ( ≈180ppmV) measured in the entire Vostok core (0–160 kyear BP)(Reference Barnola, Raynaud, Korotkevitch and LoriusBarnola and others, 1987), thus indicating a strong CO₂ depletion in these layers. The comparison of the two main types of ice in Table 1 shows that the total-gas content is halved, whilst the CO₂ is divided by a factor of 4 when debris is present. This favours the hypothesis of the presence of a liquid phase, at ice-crystal boundaries, since only the higher dissolution rate of CO₂ with regard to the other atmospheric gases can explain its selective depletion.

Table 1. Gas composition of the different types of ice at the base of the CAROLINE ice core

The combined results of stable-isotope and gas analyses therefore require a peculiar mechanism for debris entrainment, allowing gas depletion in the presence of water, in a system different from that of progressive freezing in a finite reservoir at the ice-bedrock interface. Moreover, sedimentological studies on the particles forming the debris-rich ice layers suggest a local origin for the entrainment process which occurs well below the pressure-melting point (between −7 ° and −6 ° see Fig. 1). Indeed, Reference Nougier and LoriusNougier and Lorius (1969), in a geological and physico-chemical study of the debris in Gl and G2, showed that whilst the upper debris layers of both ice cores have chemical compositions close to that of a ferruginous mica-bearing sandstone (of the Beacon Formation observed inland in the Transantarctic Mountains), the bottom layers indicate a parent rock with a much lower proportion of silica and much richer in plagioclase feldspars. The latter can be compared either to the granite, diorite and migmatite cropping out near the coast, or to the dolerite boulders occurring in the ground moraine below G2 and at the surface of Moraine Prudhomme. Since the physico-chemical analyses were performed on disparate rock fragments, not identifiable macroscopically, and virtual minerals reconstructed from anhydrous oxides percentages, these results only demonstrate that there is an increasing proportion of local parent material in the debris-rich ice, as one approaches the ice-bedrock interface.

Comparative SEM analyses performed on sand particles from internal morainic layers at D1O (230m depth), on BIL from the CAROLINE ice core, and on debris-rich ice samples from Moraine Prudhomme strongly support the local origin for this debris entrainment process. Overall, the grains studied in the 20 sub-samples from D10–230 consist largely of quartz and feldspar material (90%) with ancillary heavy minerals (mainly amphiboles). The microtextures observed on quartz grains include very high relief and high degrees of edge sharpness consistent with glacially crushed grains reported elsewhere (Reference Krinsley and DoornkampKrinsley and Doornkamp, 1973; Reference Mahaney, Vortisch and JuligMahaney and others, 1988; Mahaney, 1990b, in press). About 80% of all grains studied showed extensive fracturing, over 70% of all glacially crushed grains were abraded, some very intensively. High- and low-frequency fractures, thought to be related to ice thickness and hence cryostatic pressure (Reference MahaneyMahaney, 1990a, Reference Mahaneyb), were observed on most of the grains studied. On the whole, high-frequency fractures (number of fractures per surface area) dominate, which may be related to the vibrational energy released upon low-velocity impact (Reference Mahaney and MenziesMahaney, in press). The commonest glacial-crushing features include sub-parallel linear and conchoidal fractures (Fig. 3a) of variable high to low frequencies (Reference MahaneyMahaney 1990a, Reference Mahaneyb), along with deep straight and curved grooves. The high degree of fracturing and the range of fractures observed suggests crushing by thick ice of continental proportions (Reference MahaneyMahaney, 1990b) (>800m thickness). V-shaped percussion cracks, long considered to have resulted from water transport in outwash channels or in streams (Reference Krinsley and DoornkampKrinsley and Doornkamp, 1973) were seen very infrequently on these sub-samples. This suggests that water transport within or at the base of the ice is minimal. About 30–35% of the quartz grains studied show dissolution effects followed by renewed crushing (Fig. 3b), thus indicating weathering prior to entrainment and crushing by ice. However, the age of this weathering is still highly conjectural. Such weathering-crushing sequences have been identified elsewhere in Antarctica (Reference Mahaney and MenziesMahaney, in press).

Fig. 3. SEM microphotographs of sand particles from the D10 ice core at 230 m depth: a. Extensively crushed quartz of the median fraction (250-500 µm) showing extensive gouges (g), steps (s) and numerous fractures, some of which are abraded; b. Quartz showing extensive dissolution etching (left) and fresh surface (right) which is heavily fractured and abraded.

Samples from the very bottom of CAROLINE and from the Moraine Prudhomme ice cores show different characteristics, especially a much higher proportion of plagioclases. Many particles from CAROLINE show old weathered surfaces side-by-side with freshly fractured and abraded surfaces and, in Moraine Prudhomme samples, most quartz is old, fractured and pre-weathered with only a few freshly fractured grains present. There is therefore an overall impression gained of a much more local provenance for the debris picked up in the ice as one moves from D1O to the coast with a higher proportion of autochthonous minerals, a higher degree of weathering and a lower proportion of fresh fracturing under thick ice masses, the last two characteristics being possibly coupled with déglaciation/glaciation cycles affecting the area closer to the ice-sheet margin.

Debris-Rich Ice from Surface Ice Core at Moraine Prudhomme

The five 2 m long surface ice cores (S1 to S5 in Figure 1 ) were sampled on a straight line along the central highest part of the arcuate Moraine Prudhomme in a down-glacier direction to the coast. They were taken 4–6 m apart from each other and correspond to most of the BIL sequence at CAROLINE, since no debris-rich ice layer could be found either up-glacier from S1 or down-glacier from S5. Ice texture and debris characteristics are quite similar to those observed in the CAROLINE BIL. However, thanks to the available thin sections, through debris-rich ice, a better understanding of the debris-entrainment processes can be gained.

Figure 4 summarizes the main characteristics of the different types of ice encountered in the cores. For each type of ice, the core sub-sample is shown in transmitted light together with a vertical thin section from the sub-sample in transmitted light and between crossed polars. Typical ice fabric for each type of ice is also shown in the vertical plane. Black circles on the thin sections are locations of the dirt aggregates. Three main ice populations are clearly visible.

Fig. 4. Main characteristics of the different types of ice encountered in the surface ice cores at Moraine Prudhomme. Photographs are shown in full scale. Black circles surround debris aggregates. Ice-fabric diagrams are shown in the vertical plane, the cross in the center of the diagram is the tail of the arrow pointing to the coast in the S1-S5 direction (see Fig. 1). For each diagram, the number of c-axes measured is mentioned in the upper left frame. Contour intervals are drawn at 2, 5, 10, 15 and 20% of c-axes per 1% unit area.

Crystal mean cross-sectional area (on at least 100 crystals), concentrations of bubbles and clots have been measured on enlarged thin-section photographs of each individual layer for a given ice type and the results are plotted on Figure 5. Mean crystal cross-sectional areas vary widely from 1 to 200 mm². The maximum value is higher than the one measured above the BIL in the CAROLINE ice core (85 mm² between 60 and 70 m depth; Reference Yao, Petit, Jouzel, Lorius and DuvalYao and others, 1990). This is probably a sign that the glacier ice near the margin of the ice sheet and close to the surface at Moraine Prudhomme has been submitted to relaxation and post-kinematic recrystallization in a nearly stagnant state. There is a clear limit in Figure 5, at about 50 mm² of cross-sectional area, beyond which debris-free ice with bubble concentration always higher than 25 cm⁻² only occurs. Debris-rich ice is therefore always small-grained and shows bubble concentrations ranging from values as low as 5 cm⁻² to values not much different from the debris-free ice. There is, however, in this small-grained population, no correlation between the mean crystal cross-sectional area and either the bubble concentration or the dirt content.

fig. 5 Bubble and clot concentrations (per cm²) as a function of the mean crystal cross-sectional area (in mm²) in surface-ice cores at Moraine Prudhomme.

Ice fabrics (Fig. 4) in the debris-rich ice differ from those observed in the interlayered debris-free ice, although a general pattern of a wide girdle in the vertical plane containing the S1-S5 direction (tail of the arrow in the middle of each diagram) is a common feature for all samples. Debris-free ice shows a multiple-maxima fabric which was also observed at 36.55 and 75 m depth by Reference Lorius and VallonLorius and Vallon (1967) in the Gl ice core. In our case, however, the maxima are grossly located in a girdle centered on the vertical and stretched in the vertical plane containing the S1-S5 direction. Following Reference AlleyAlley’s (1988) study of fabrics in polar ice sheets, this could be interpreted as a pure shear configuration with parallel flow (fig. 1c in Reference AlleyAlley (1988)) where the vertical compression is balanced by an extension roughly following the main flow component of glacier de l’Astrolabe (Fig. 1). This fabric was probably acquired further up-glacier when the ice was still in the active fast-flowing part of glacier de l’Astrolabe. Indeed, surface-velocity measurements performed by E.P.F. (personal communication from Expéditions Polaires Françaises) close to Moraine Prudhomme yield values as low as 0.045 m year⁻¹, with maximum values of about 2.3 m year⁻¹ for the surrounding debris-free ice (surface-ice velocity on the IAGP line between D24—14 km upstream — and D10 ranges from 36 to 7 m year⁻¹.

Fabrics in the debris-rich ice layers are quite similar to some of those observed by Reference AndertonAnderton (1974) in the basal cold ice of Meserve Glacier, Antarctica (“Dry Valleys” area, Victoria Land). Close to the ice-rock interface, in a BIL containing bubbly “amber” debris-laden ice, girdle fabrics containing a single maximum were observed (monoclinic symmetry). The single maximum was generally weaker than those measured higher in the basal sequence. Following Anderton, this pattern reflects flow perturbation due to irregularities in the glacier bed, the orientation of the single maximum being controlled by the local orientation of the ice-rock interface. A single-maximum fabric observed near the glacier bed is commonly interpreted in the literature as the signature of enhanced mechanical flow due to a dominating basal shear stress (Reference BuddBudd, 1972; Reference Kamb, Heard, Borg, Carter and RaleighKamb, 1972; Reference AlleyAlley, 1988). The deviation of the centre of the maximum at about 30 ° from the vertical, in this case, is thought to be the result of compressive stress superimposed on the simple shear, as the active glacier ice from upstream over-rides slow-moving ice near the margin of the ice sheet, and is furthermore buttressed by the bedrock outcrops along the coast. Maximum c-axis concentrations (S4F3) reach 25 points per 1% surface area, for example, a value in the range of those observed by Reference AndertonAnderton (1974) and by Reference Gow and WilliamsonGow and Williamson (1976) in deep Antarctic ice cores. The single maximum is clearly weaker in the diagrams S3A4 and S3 AÍ (maximum 14 points per 1% surface area). This could reflect locally lower-stress concentration than under thicker ice, enhanced disturbance due to the irregularities of the glacier bed, changing balance between simple shear and compressive stress as the active ice over-rides the dead ice or, in the case of S3A1, an increasing effect of post-kinematic recrystallization or relaxation near the surface.

The fine-grained texture of the debris-rich ice thus results from the activation of new local stress conditions near the glacier bed as the ice approaches the ice-sheet margin. Crystal-growth impedance by the debris load, another explanation for fine-grained textures, does not seem to play a major role in this case since samples with the same debris load show mean crystal sizes an order of magnitude apart (no correlation between dirt content and crystal sizes in Figure 5). Besides, dozens of individual ice crystals usually exist between two neighbouring debris aggregates in the fine-grained amber ice (S3A4 and S4F3 in Figure 4) and, at a microscopic scale, individual particles of a few tens of μm are often separated by several ice crystals.

Discussion

BIL at the base of the CAROLINE ice core and at Moraine Prudhomme are compared to those observed in some of the deep drillings to bedrock in the Northern and Southern Hemispheres in Table 2. Generally, debris structure and characteristics are very similar to the one described by Reference Gow, Epstein and SheehyGow and others (1979) for the Byrd Station ice core in Antarctica and by Reference Herron and LangwayHerron and Langway (1979) for the Camp Century ice core in northwest Greenland. However, the basal thermal regime clearly differentiates the Byrd Station ice core from the other two. In the first one, the interface is at the pressure-melting point and liquid water was even encountered at the ice-rock interface (Reference Gow, Ueda and GarfieldGow and others, 1968). In the two others, the interface was clearly at sub-freezing temperatures when the cores were retrieved. A similar contrast holds for the total-gas content of the three cores. Whilst the debris-rich ice in the Byrd BIL shows less than 0.002 cm³ of air g⁻¹ of ice (the ice is virtually devoid of air), the total-gas content is reduced in the same proportions (grossly halved) in the two other cores. The absence of air and the +3%o shift in δ¹⁸Ο between the basal meltwater and the BIL at Byrd Station were used as strong arguments by Reference Gow, Epstein and SheehyGow and others (1979) to propose “freezing-in” as the most likely process of debris inclusions at the base of the ice sheet. Unfortunately, the similarity between Camp Century and CAROLINE does not hold any more when qualitative gas measurements are considered. There is a five-fold increase of the CO₂ concentration in Camp Century against a four-fold decrease in the BIL of CAROLINE. The very high concentration levels (10 700–13 100 ppm V) observed by Reference Herron and LangwayHerron and Lang-way (1979) were interpreted by these authors as the major effect of the oxidation of methane that was found as traces in samples containing debris (samples Β and C). However, oddly enough, the CO₂ concentration in sample A, a few meters above BIL, is still an order of magnitude higher than those expected in “fossil” atmos-pheric air and entrapped in bubbles. This observation should be considered in the light of the more recent developments of CO₂ measurements in air from bubbles in polar ice (Reference Raynaud, Delmas, Ascencio and LegrandRaynaud and others, 1982) that demonstrate potential contamination in CO₂ by particle-dissolution processes, when using the melting-refreezing technique instead of the dry-crushing extraction method used nowadays. Nevertheless, argon measurements in the BIL of Camp Century did indicate an inverse relationship between the concentration of this gas and the total gas content. This observation has been used by Reference Herron and LangwayHerron and Langway (1979) to demonstrate the feasibility of a process in which gas diffusion occurred downward from bubbly glacier ice into an originally bubble-free zone of refrozen debris-laden ice formed upstream in a basal freezing zone. Since CO₂ is also a highly soluble gas component, supposing the natural (or experimental) contamination had not occurred at Camp Century, an increase of the CO₂ concentration would still have been observed in the BIL, a situation opposite to the one described at CAROLINE.

Table 2. Main characteristics of the basal ice layers observed in some of the deep drillings to bedrock in the Northern and Southern Hemispheres compared to those described in the present study. Numbers within brackets are calculated crystal surface areas, considering them as spherical objects. Asterisks indicate where no information is available

The overall characteristics of the debris-rich ice at the base of the CAROLINE ice core and at Moraine Prudhomme rule out pressure-melting regelation or freezing-on as possible mechanisms for debris entrainment at the glacier sole. Indeed, lithological and micromorphological signatures of the rock particles both indicate a local origin, and the low proportion of freshly fractured quartz grains in these debris samples do not support the hypothetical existence of the considerably greater ice thicknesses that would be required locally to set up pressure-melting conditions at the interface. Profiles in cores DIO, Gl and CAROLINE all show temperatures well below the pressure-melting point (between −7.50 ° and −8 °C). Therefore, neither is a fresh meltwater input locally available for freezing-on processes, nor is pressure-melting regelation possible around bed obstacles. This is further supported by the absence of a freezing slope on the δD–δ¹⁸Ο diagram for the whole debris-rich population and by the persistence of intracrystalline bubbles even in the fine-grained amber ice. There is thus a clear need for a debris-entrainment process acting at sub-freezing temperature but allowing phase changes at crystal boundaries to explain small-scale isotopic effects and considerable modifications in the total gas and CO₂ content of the ice.

Figure 6 illustrates such a possible mechanism that can be described as follows. Active glacier ice comes from inland, following the main flow direction of glacier de l’Astrolabe in “parallel flow” conditions. As it reaches the edge of the ice-sheet margin and as it is buttressed on the bedrock outcrops near the coast it is submitted to a strong compressive stress which combines with the basal shear stress near the interface. This local stress field causes the mylonitization of the active glacier ice as it over-rides the dead, slow-moving ice, near the margin. Stretched crystals with slightly elongated bubbles and sometimes showing undulose extinctions are typical of the clear-cut boundary between layers of debris-free and debris-laden ice in the BIL. While some of the new crystals keep the signature of the previous stress field, most of them rotate to form a single maximum perpendicular to the new principal shear plane at about 30 ° from the horizontal. Similar mylonitized zones were described by Reference EchelmeyerEchelmeyer and Zhongxiang (1987) in basal ice layers at sub-freezing temperatures at Urumqi Glacier No. 1 in China. In their case, thin sections taken just above the ice-sediment interface, where shear planes were proved to exist by detailed strain measurements, showed a non-gradational transition between coarse glacier ice and a 4–10 mm layer of extremely fine-grained ice (<0.5mm grain-size). In some places, like at the crest of a boulder embedded in the ice-laden basal drift and protruding 15 cm into the ice above, the fine-grained ice layer was underlain by a few thin bands of debris-ice laminations extending 1–2 m upward into the ice at an angle of 14–18 °. The soft sediment frozen at the sole of glacier de l’Astrolabe near its coastal margin is most likely to be affected by the new stress field mentioned above. Large particles, eventually protruding at the interface on a small scale, are entrained in the ice by differential movement across them along the new principal shear-plane direction. As they tend to roll on themselves, a coating of fine particles adhering to their surface is progressively formed. These clots are further entrained in the ice by the movement of the surrounding small ice crystals in the shear zone. Eventually, under the combined influence of high temperatures (≤10 °C) and high strain rates (Reference AlleyAlley, 1988), recrystallization will take place and form the coarser-grained debris-rich ice above, the fine-grained ice layer was underlain by a few thin bands of debris-ice laminations extending 1–2 m upward into the ice at an angle of 14–18 °. The soft sediment frozen at the sole of glacier de l’Astrolabe near its coastal margin is most likely to be affected by the new stress field mentioned above. Large particles, eventually protruding at the interface on a small scale, are entrained in the ice by differential movement across them along the new principal shear-plane direction. As they tend to roll on themselves, a coating of fine particles adhering to their surface is progressively formed. These clots are further entrained in the ice by the movement of the surrounding small ice crystals in the shear zone. Eventually, under the combined influence of high temperatures (≤10 °C) and high strain rates (Reference AlleyAlley, 1988), recrystallization will take place and form the coarser-grained debris-rich ice.

Fig. 6. Sketch for a possible mechanism of debris entrainment in basal ice at sub-freezing temperatures (see text for details).

Fracture of the large glacier-ice crystals into groups of smaller individuals will bring some of the intracrystalline bubbles in contact with water between the crystals. A liquid phase is sure to exist in pure ice at temperatures significantly below 0 °C, as demonstrated by studies of the liquid-like signal in PMR (proton magnetic resonance) spectra of ice (Reference Bell, Myatt and RichardsBell and others, 1971) and by NMR (nuclear magnetic resonance) spectroscopy (Reference Kvlividze, Kiselev, Kurzayev and UshakovaKvlividze and others, 1974). In both cases, a liquid-like signal was detectable down to −12 °C. It was also suggested by Reference Paren and WalkerParen and Walker (1971) that the formation of a liquid phase at the three-grain intersections (veins) contributes to the softening of polycrystalline ice at temperatures above −10 °C. Recent experiments on water veins in polycrystalline ice (Reference MaderMader, 1990) study the thermal behavior of the veins for temperature depressions down to 0.6 °C below the pressure-melting point. They show a strong relationship on a bilogarithmic plot between the temperature depression and the apparent vein width. A typical range of the apparent vein width for ice grown from singly-distilled water is between 100 and 10 μm for temperature depressions varying between 0.01 ° and 0.6 °C. Extrapolation of this linear relationship down to −7 °C (basal ice temperatures at CAROLINE) would give apparent vein width of the order of a few μm. The liquid-like layer adjacent to foreign solids in the ice or at the ice-bedrock interface, which is thought to play an important role in the sliding of glaciers at sub-freezing temperatures (Reference ShreveShreve, 1984; Reference EchelmeyerEchelmeyer and Zhongxiang, 1987), is also a possible water source albeit its size is probably three orders of magnitude less than the one of the veins described above.

If this liquid phase is undersaturated in atmospheric gases, part of these gases can dissolve, the CO₂ being selectively affected by the process, given its higher dissolution rate. For the gas depletion to be observed, there is however an obvious need of transfer of the liquid phase from one location to another. There are two possible paths for this transfer: either from the active shear zone, where the stresses are locally concentrated, to the surrounding undisturbed glacier ice, or from the basal ice to the underlying frozen sediment. Since the CO₂ content of the white bubbly debris-free ice in the BIL (samples 57, 92 and 98a in Table 1) is in the range of the value observed between 60 and 79 m depth in the CAROLINE ice core (210–250 ppm V CO₂) and do not therefore show any significant enrichment, the latter case is more probable. The pressure gradient set up around bedrock obstacles would help squeezing part of the liquid phase from the ice on the stoss side, and fracturing during mylonitization could ease the process. Frictional heat could also favour partial melting at grain boundaries, thereby enhancing the whole mechanism. Finally, although the efficiency of such a mechanism is still conjectural, one should bear in mind that it must only account for transfers of maximum 2 × 10⁻² cm³ of CO₂ per kg of ice. These phase changes at crystal boundaries and those which lead to the crystal growth by recrystallization, combined with water movements in the ice, are apparently strong enough to detect small-scale isotopic fractionation as shown in Figure 3 and discussed above, whilst no large-scale freezing effect, that would result from the progression of a freezing front in a liquid reservoir, is noticeable.