2.1. Snow measurements

Snow thickness was measured with a spacing of 1 m on representative 50–100 m subsections along the ice-thickness profiles with the aid of a ruler stick. Vertical profiles of snow temperature, wetness, density, salinity and grain-size were measured in up to three snow pits. Both thin snow on level ice and thick snow around pressure ridges were sampled. Snow temperature was measured with a Pt-100 thermometer at vertical spacings of 0.05–0.10 m, depending on snow thickness. In parallel, snow wetness was measured with a TOIKKA Snow Fork (Helsinki, Finland). Snow wetness in percentage liquid-water content is calculated from fork resonator measurements of the complex snow permittivity (Shivola and Tiuri, 1986).

After visual examination of the snow stratigraphy, samples were taken from the main layers with 0.05 m diameter, 0.5 L steel tubes. In the cold laboratory on board the ship, snow density was measured by weighing these samples. After visual grain-size estimation using a millimetre grid, the samples were melted and subsequently their salinity was measured with a Wissenschaflich-Technische-Werkstätten GmbH LF181 conductivity meter which was calibrated to display salinity. If the salinity was <0.1‰, the instrument gave a reading of 0.0‰.

2.2. Ice coring and analysis

From each floe, two 0.09 m diameter ice cores were drilled from level ice with a gas-motor-driven Kovacs Enterprise ice auger after the snow had carefully been removed. Such cores are referred to as “main cores”. In 1997, the top sections of the cores were drilled mostly by hand until the gap layer was encountered (see below). This “surface core” was retained from the core tube without turning it upside down, to avoid contamination of the ice with sea water and brine. After recovery, the surface cores were stored upright in pre-cooled cold boxes to minimize melting, brine drainage and contamination during the helicopter transport to the ship’s cold laboratory (<−25°C). Then cores of the remaining ice were taken and stored horizontally in plastic tubes for transfer to the ship.

The second core was immediately cut into 0.1 m sections and allowed to melt slowly for later chlorophyll-a measurements by standard fluorometric procedures (Reference Thomas, Lizotte and ArrigoThomas and others, 1998).

When surface gaps were encountered at the main coring site in 1997, up to four additional surface cores were drilled a few metres apart to determine the extent and abundance of these features. The thicknesses of the ice layers, the gap layers, as well as the freeboard height were measured. Water from the gaps was sampled for salinity. Ice crystals and slush were sieved out using a mesh size of 1 mm².

In the cold laboratory, detailed analyses of ice texture were performed on thick and thin sections of the cores (e.g. Reference LangeLange, 1988). The cores were cut according to stratigraphic units. The surface cores were cut with a high vertical resolution of 0.01–0.03 m. While two-thirds of the segments were subsequently melted for salinity measurements, the rest was transferred to the home laboratory. On a limited number of these samples, ¹⁸O concentrations were determined as described by Reference Eicken, Lange, Hubberten and WadhamsEicken and others (1994) and Reference Eicken and effriesEicken (1998).

3.1. General ice conditions

In February 1997, the ice extent in the Weddell Sea was rather high (Fig. 2), with highly deformed and thick ice in the south-east Weddell Sea (Reference Haas, Liu and MartinHaas and others, 1999; Table 2, shown later). The sampled floes were located across the entire Weddell Sea (Fig. 2), and were visited during a period of 42 days under varying climatic conditions (Fig. 3). Air temperatures measured on board Polarstern ranged between 0° and −5°C until day 55, when values dropped below −10°C when Polarstern transited a second time to the southeast, where minimum temperatures were as low as −22°C. However, at that location a slight cooling had already commenced by day 32, according to data logged by a drifting buoy in that region (Fig. 3b). The rather cold conditions started on day 49, about 7 days before the arrival of Polarstern. To distinguish between different climatic and ice regimes, the 1997 dataset is subdivided into three periods defined as summer (days 19–27) and early fall (days 56–61) in the southeast, and as summer in the northwest Weddell Sea (days 46–55; Figs 2 and 3). The 40° W meridian is taken as the boundary between the southeast and northwest Weddell Sea.

Fig. 3. Air temperatures recorded on board Polarstern along the cruise track in 1997 (a) and by a drifting buoy (Argos ID 8059) at about 40° W 73.2° S in the southeast Weddell Sea (b).

In February 1994 in the Bellingshausen and Amundsen Seas, air temperatures ranged mostly between 0° and −4°C. Only close to the coast were lower temperatures (−5° to −8°C) encountered. With respect to ice thickness (Reference HaasHaas, 1998; Table 2, shown later), we distinguish between the Bellingshausen Sea and Amundsen Sea.

In late January 1997 (days 19–27) the ship operated in closed pack ice in the southeast Weddell Sea, with ice concentrations mostly higher than 0.9 and floe diameters up to 500 m. During the second phase of the expedition in the inner marginal ice zone floe sizes exceeded 100 m only on days 53 and 54 as well as 57 and 58, while ice concentration varied between 0.5 and 1.0. The ice in the northwest Weddell Sea was characterized by a mixture of thick deformed and thin level floes, indicating the presence of old and young ice, respectively (Reference Lange and EickenLange and Eicken, 1991). The latter carried only a thin snow cover and their surfaces appeared greyish. In 1994, most floes were located well within the perennial ice zone (Reference Haas and ViehoffHaas and Viehoff, 1994; Reference Morris, Jeffries, Li and JeffriesMorris and others, 1998), although their size was mostly <100 m.

Numerous surface ponds were observed throughout the southeast and northwest Weddell Sea, mostly originating from ridge loading (Reference Ackley and SullivanAckley and Sullivan, 1994). However, aerial photographs (C. Haas, unpublished information) showed that they covered only 0.3% of the ice surface. The mean salinity of 19 pond-water samples was 19.4 ± 7.1 (range 8.3–30.9), indicating a mixture of sea water and melted snow and ice.

3.2. Snow properties

Table 1 summarizes the average snow thicknesses in each region calculated from the thickness-profile data of each floe. The snow layer was generally much thicker in the Bellingshausen and Amundsen Seas than in the Weddell Sea, where the thinnest snow was observed in the northwest. There, in contrast, winter observations of Reference Eicken, Lange, Hubberten and WadhamsEicken and others (1994) and Reference Massom, Drinkwater and HaasMassom and others (1997) report the highest snow thicknesses of the Weddell Sea. Glearly, our findings indicate strong snowmelt in the northwest during summer, which will be discussed later in more detail.

Table 1. Mean snow properties

Figure 4 shows typical vertical profiles of snow temperature and wetness. The differences in air temperatures (Fig. 3) were clearly reflected in the snow (Fig. 4a). While temperatures in the southeast Weddell Sea were higher at the surface than at the snow/ice interface (positive gradients), positive as well as negative gradients were found in the northwest Weddell Sea. In both summer situations, temperatures at the snow/ice interface were close to or above the freezing point of sea water (−1.8°C). In contrast, strong negative gradients ranging from −8° to −66°C m⁻¹ were observed in the southeast Weddell Sea in early fall, 75% of these being less than −25°C m⁻¹. Depth hoar was found in every snow pit. The ice surface had cooled below −5°C at several sites, and dendritic ice growth underneath the surface layer into the gap layer was observed (see below).

Fig. 4. Typical vertical snow profiles of (a) temperature obtained on level ice and (b) wetness (percentage liquid water by volume) obtained on level ice and ridge flanks in three different regions in 1997. Note different scale in upper right graph.

Figure 4b shows that snow wetness seldom exceeded 4% This means that during most of the sampling period the snow was in the pendular regime where liquid water cannot freely drain but is held at the triple junctions of ice grains and air. This regime promotes snow coarsening by the growth of grains into clusters. Snow wetness increased with depth on most floes and was highest at the bottom (Fig. 4b). This was most pronounced in the southeast Weddell Sea in fall.

Of all 169 salinity samples, only 31 had a salinity equal to or higher than 0.1‰ (Table 1). Only three samples which were obtained from the snow/ice interface had salinities higher than 1‰ in 1997. These summer values are significantly lower than measurements in other seasons, where mean snow salinities range between 1‰ and 16‰, with more than 10% of all measurements ranging from 10‰ up to 40‰ (Reference Eicken, Lange, Hubberten and WadhamsEicken and others, 1994; Reference Massom, Drinkwater and HaasMassom and others, 1997; Reference Sturm, Morris, Massom and JeffriesSturm and others, 1998; Reference Worby, Massom, Allison, Lytle, Heil and JeffriesWorby and others, 1998).

Snow density ranged between 130 and 518 kg m⁻³, with means of 300–400 kg m⁻³. Grain-sizes ranged between 1 and 10 mm, with mean values of 2.5–3.1 mm (Table 1). Apart from the depth hoar observed in the southeast Weddell Sea in fall, and fine new snow that mostly melted within 1 or 2 days, the snowpack consisted of coarse and very coarse grain clusters (seasonal-snow-on-the-ground classification 6cl/6mf; Reference ColbeckColbeck and others, 1990). The snowpack was very icy at the bottom. Just above the snow/ice interface, grain-size increased rapidly and the grains were partially frozen onto the underlying ice. Therefore, the snow/ice interface was very rough on the centimetre scale. As the ice underneath was mostly superimposed ice (see below), the rough surface was a transition zone of snow being metamorphosed into ice.

Observed grain-sizes and densities are similar to values reported by others for snow which was subject to melting phases. However, they are much higher than values of cold Antarctic winter snow (Reference Eicken, Lange, Hubberten and WadhamsEicken and others, 1994; Reference Massom, Drinkwater and HaasMassom and others, 1997; Reference Sturm, Morris, Massom and JeffriesSturm and others, 1998; Reference Worby, Massom, Allison, Lytle, Heil and JeffriesWorby and others, 1998).

3.3. Ice properties

Table 2 summarizes all EM and drillhole thickness measurements. Note that these include both level and deformed ice, the fraction of the latter being rather high (Reference Haas, Liu and MartinHaas and others, 1999). The thickness of undeformed, level ice is 0.4–1.3 m (cf. Figs 5 and 6). However, mean thicknesses in the southeast Weddell Sea were much greater than those recorded in previously published winter data (Reference Eicken, Lange, Hubberten and WadhamsEicken and others, 1994; Reference Massom, Drinkwater and HaasMassom and others, 1997), and are much closer to values reported by Reference Strass, Fahrbach and JeffriesStrass and Fahrbach (1998). In the northwest Weddell Sea, thicknesses were comparable to those measured for undeformed second-year ice in spring in the same region (Reference Lange and EickenLange and Eicken, 1991). The ice in the Bellingshausen and Amundsen Seas is some of the thickest Antarctic sea ice reported (Reference HaasHaas, 1998).

Table 2. Mean ice properties

Fig. 5. Top: Vertical cross-sections typical for floes encountered in the southeast and northwest Weddell Sea in summer (a, b) and in the southeast Weddell Sea in fall (c) obtained from thickness drilling (see (a) for legend). Bottom: Salinity, δ¹⁸O and texture for typical surface and main cores drilled along the profiles above at locations indicated by thick vertical lines. Data are plotted at centre position of individual core segments with respect to the water level (z = 0.0 m).

Fig. 6. Photographs of a thin section (a: 1994) and two thick sections (b, c: 1997) of three typical surface cores, taken with the samples between crossed polarizers. In addition, (b) shows the rotten ice between the gap and the bottom, and arrows mark saltwater droplets penetrating into the superimposed ice. The salinity profile of each core, plotted at the centre position of individual core segments with respect to the water level (z = 0.0 m, y axis), as well as the texture interpretation are also shown.

Figure 5 shows high-resolution vertical cross-sections typical for floes encountered in the three ice regimes in the Weddell Sea, obtained from thickness drilling. While the floes in Figure 5a and c were rather level and much larger than 100 m, the floe in Figure 5b was only 40 m in diameter and comprised an old pressure ridge. All three profiles show the typical sequence of a snow layer above a surface ice layer overlying a slush- or sea-water-filled gap. Underneath, the main ice was encountered, which was often extremely rotten and porous (Figs 1 and 6b).

The frequency of occurrence, the thicknesses and drafts of the surface ice layers, as well as the gap thicknesses, are summarized in Table 2. In total, gap layers were observed on 64% of all 33 sampled floes in 1997. The surface ice layer in the southeast Weddell Sea was much thinner than in the other regions, and its underside was at or just below sea level (Fig. 5a; Table 2). This is similar to earlier summer observations of a crusty snow layer above the slush (Reference Lytle and AckleyLytle and Ackley, 1996). Two samples of gap water had salinities of 14‰ and 17‰. In the other two regions, by contrast, the thick surface ice layer extended on average 0.14 m into the water, with most of its thickness not rising above the water level. The mean salinity of 37 gap-water samples was 29.3 ± 2.6‰ (range 20.6–31.3‰). As with the pond water, the low gap-water salinities indicate that melting was taking place within or adjacent to the gaps.

The lower portion of Figure 5a–c shows vertical profiles of texture and salinity of the main and some surface cores, as well as four δ ¹⁸O profiles. All cores consisted of superimposed PG ice at the top. The typical PG ice texture with isometric grains with planar boundaries can be seen in the thin- and thick-section photographs in Figure 6. Grain-size often decreased with depth. In the thick-section photographs (Fig. 6b and c), PG ice is less clearly recognizable once the grain-size becomes smaller than the sample thickness. Therefore, fractions of PG ice given below are probably underestimates of the real amount. The downward-decreasing grain-sizes also make it difficult to distinguish it from underlying fine-grained ice with higher salinities occurring in some instances (Figs 5 and 6). That ice is probably of snow-ice origin.

PG ice contributed significantly to the overall surface core lengths (see, e.g., Fig. 5b). It was found at the top of 76% of the 33 main cores taken in 1997, and of 73% of the 15 cores taken in 1994 (Fig. 2). Table 2 summarizes the mean PG ice-layer thicknesses. The smallest values occurred in the southeast Weddell Sea in summer; they doubled there by early fall. The greatest thicknesses were observed in the Bellingshausen and Amundsen Seas. The amount of PG ice growth during summer is roughly comparable with fractions of snow ice formed during flood–freeze cycles in the cold season (Reference Eicken, Lange, Hubberten and WadhamsEicken and others, 1994; Reference Jeffries, Worby, Morris and WeeksJeffries and others, 1997).

Figures 5 and 6 also show typical vertical high-resolution salinity and δ ¹⁸O profiles of the respective cores. Salinities of all surface and main cores are presented in Figure 7. They ranged between 3 and 4 for the main ice underneath the gap. In contrast, salinities were very low in the surface cores, and mostly zero for PG ice (Table 2). Only immediately above the gap were high salinities sometimes measured, coinciding with a fine-grained granular texture. Additionally, in some locations in the southeast Weddell Sea in fall, PG ice salinities rose at the top (see also Table 2), when the freeboard was less than about 0.01 m, or where the gap had started to refreeze. Then, the surface cores often revealed a C-shaped salinity profile (e.g. at 4 m in Fig. 5b), comparable to the salinity distribution of young nilas ice in winter. Apparently, ice formation within the gap leads to some upward brine expulsion.

Fig. 7. Salinity and δ¹⁸O profiles of all surface and main cores obtained in 1994 and 1997 plotted at centre position of individual core segments with respect to the water level (z = 0.0 m).

The δ ¹⁸O values of the surface ice layers were extremely small (Figs 5 and 7). At the very top, they were more like values for snow, ranging between −3‰ and −28‰ in the Weddell Sea in winter (Reference Eicken, Lange, Hubberten and WadhamsEicken and others, 1994). With increasing depth, they slowly became more typical of sea ice, at 0–2.5‰, indicating the presence of snow ice (Reference Eicken, Lange, Hubberten and WadhamsEicken and others, 1994; Reference Jeffries, Worby, Morris and WeeksJeffries and others, 1997).

Two other common observations are worth mentioning (Fig. 6): First, PG ice was often underlain by a thin layer of columnar ice, either immediately below, as in Figure 6a, or further below, as in the case of dendritic new ice growing into the underlying gap (Fig. 6c). This columnar ice layer indicates the former presence of liquid water, in contrast to the finegrained granular snow ice forming from slush during flood–freeze cycles. Secondly, in the southeast Weddell Sea in fall, surface ice with small freeboard often had a C-shaped salinity profile (Fig. 6b). Below the surface of this ice, salt-water droplets were observed within the fresh superimposed ice, indicating downward melting of sea water which had flooded the ice surface. The salt water caused the salinity increase at the top, and destroyed the PG texture of the ice, making the PG origin of the ice difficult to recognize. This observation was common on very rotten, thin ice, where the main sea ice underneath had melted almost completely and the floes consisted only of the surface ice layer (Fig. 6b). These features indicate the final stage of an ice floe in late summer, when it would either soon deteriorate completely or become restrengthened once subject to fall freezing. The rotten, highly porous structure of the ice underneath the gap (Fig. 6b) permits large convective transport of nutrients, salt and heat at the onset of fall freeze-up (Reference Fritsen, Lytle, Ackley and SullivanFritsen and others, 1994; Reference Lytle and AckleyLytle and Ackley, 1996).

Finally, in Figures 5–7, note that PG and very low-salinity ice occurred below the water level. As this ice originated from the refreezing and metamorphosis of snow above sea level, it must have moved downward relative to the water level during summer.

Figure 8 presents chlorophyll-a concentrations of the upper metre of all cores obtained in the Weddell Sea. Unfortunately, the averaging over 0.1 m long segments does not allow detailed study of surface concentrations. However, it can be seen that chlorophyll-a concentrations were generally highest at 0.1–0.3 m depth below the surface, i.e. in the ice surrounding the gap. This demonstrates the importance of the surface-layer/gap structure for the primary productivity in perennially ice-covered regions. The highest concentrations of up to 439 μg L⁻¹ were found in the dendritic new ice forming from refreezing gaps in the southeast Weddell Sea in fall. Similar observations have been reported by Reference Fritsen, Lytle, Ackley and SullivanFritsen and others (1994).

Fig. 8. Chlorophyll-a concentrations of ice cores obtained in 1997, plotted at centre depth of individual core segments with respect to the ice surface. For clarity, data from below 1 m were omitted.

4.1. Snowmelt and refreezing

We have presented numerous snow and ice data obtained from perennial Antarctic sea ice. Although the dataset is heterogeneous, several common features emerge which can be considered typical for the summer situation. Most importantly, there is considerable surface melting in the Antarctic pack ice. The amount of melt is much less than in the Arctic, where the snow melts completely and the ice surface is extensively covered with melt ponds. Nonetheless, internal snow melting on Antarctic sea ice is strong enough to induce extensive snow metamorphism and the formation of superimposed ice. In the northwest Weddell Sea, where the thickest snow occurs in winter, melting was strongest and resulted in both the thinnest residual snow cover and the thickest surface ice layer. Average superimposed ice thicknesses of 0.08–0.25 m correspond to late-winter snow thicknesses of 0.24–0.75 m if a snow density of 330 kg m⁻³ is assumed. The low snow salinities show that there was considerable meltwater percolation through the snow, transporting salt downwards. When measured, however, the snow was in the pendular regime. Nevertheless, there must have been episodic events of enhanced surface melt and percolation, for example during periods of warm-air advection (Reference Massom, Drinkwater and HaasMassom and others, 1997; Reference Morris, Jeffries, Li and JeffriesMorris and others, 1998). Snowmelt and wetness can also be much higher earlier in the season, when the amount of incoming solar radiation is highest. Very likely, wetness varies diurnally with changing solar elevation, resulting in melt–freeze cycling (Fig. 3b).

Although surface melting was regarded as atypical in earlier studies (Reference Andreas and AckleyAndreas and Ackley, 1982; Reference Jeffries, Shaw, Morris, Veazey and KrouseJeffries and others, 1994), the regional and temporal distribution of our observations and those of Reference Fritsen, Lytle, Ackley and SullivanFritsen and others (1994) and Reference Lytle and AckleyLytle and Ackley (1996) suggests that it is actually quite common, not only in the marginal ice zone, but also within the inner pack.

Melting occurs even at air temperatures below 0°C through absorption of solar radiation within the snow (Golbeck, 1989; Reference Launiainen and GhengLauniainen and Gheng, 1998). Temperature gradients in a melting snowpack on sea ice are positive, i.e. temperatures decrease towards the ice surface, where they are close to the freezing point of sea water. This results in the refreezing of fresh meltwater further down at the snow/ice interface, and finally in the formation of superimposed ice with its PG texture. However, the measured low snow wetnesses could also indicate that superimposed ice may form just by growth and merging of grain clusters. The processes described have also been observed on Arctic fast ice (Reference Holt and DigbyHolt and Digby, 1985; Reference Onstott and CarseyOnstott, 1992; Reference Barber, Thomas, Papakyriakou, Tsatsoulis and KwokBarber and others, 1998), although melting is much stronger there and more continuous. Therefore, the snow and successively the superimposed and underlying ice layers melt completely.

Generally, superimposed-ice formation should affect every spot of an ice floe, including level ice and pressure ridges. However, depending on the amplitude of melting and on the amount of meltwater, the water will drain to the lowest topographic parts of a floe where the thickest superimposed ice would then be formed. When melting is not strong enough to produce freely draining water, as indicated by our wetness measurements, lateral differences in the amount of superimposed-ice formation might be caused by heterogeneous snow properties. Glearly, the local variability in snow thickness, stratigraphy, grain types and iciness (Reference Massom, Drinkwater and HaasMassom and others, 1997; Reference Sturm, Morris, Massom and JeffriesSturm and others, 1998) will result in different light penetration and scattering, causing variable intensities of internal melt.

4.2. Gaps

The other common summer feature is the widespread occurrence of slush- or water-filled porous layers or gaps underneath a surface ice layer. As outlined in the introduction, their origin has been explained as the result of summer warming and the subsequent increase in brine volume, which fosters a biophysical feedback mechanism (Reference Ackley, Buck and TaguchiAckley and others, 1979; Reference Ackley and SullivanAckley and Sullivan, 1994). In fact, the high salinities near the top of first-year ice (Reference Maykut and UntersteinerMaykut and Untersteiner, 1971; Reference EickenEicken, 1992) will result in large brine volumes once the ice warms up. However, the results of Reference Zeebe, Eicken, Robinson, Wolf-Gladrow and DieckmannZeebe and others (1996) show that a positive feedback mechanism by absorption of solar radiation would only be effective with rather high chlorophyll-a concentrations in the gap of > 150 mg chi a m⁻² and a snow cover thinner than 0.05 m. Although our snow and chlorophyll data suggest that absorption by algae might become significant in the course of the summer, these conditions are unlikely to prevail at the onset of summer warming. For example, Reference Günther and DieckmannGünther and Dieckmann (1999) observed no accelerated rise in nutrients or chlorophyll in the upper ice layers until the end of December in a time-series study on snow-covered fast ice.

However, the formation of gap layers can also be explained by purely abiotic processes. First, the widespread formation of snow-ice layers during flood–freeze cycles in the cold season provides a mechanism to form layered structures with strong salinity gradients (Reference Fritsen, Ackley, Kremer, Sullivan, Lizotte and ArrigoFritsen and others, 1998; Reference Maksym and JeffriesMaksym and Jeffries, 2000). Although some of the saline brine might be expelled downwards during flood–freeze events because the underlying ice is warm and permeable (Reference Maksym and JeffriesMaksym and Jeffries, 2000), it is likely that some high-salinity layer remains at the former slush/ice interface on completion of the freeze cycle. These high saline layers would reopen as soon as ambient temperatures rise. As outlined by Reference Eicken, Fischer and LemkeEicken and others (1995) and Reference Maksym and JeffriesMaksym and jeffries (2000), due to energetic constraints, the slush layer will only incompletely congeal if flooding and freezing commence after early September. Thus, there is a high probability that slush-filled gaps exist already at the end of winter. Where these are present, superimposed ice will form on top of snow ice overlying a gap, as shown in Figures 5 and 6.

Secondly, the surface-ice/gap structure might be formed due to melting of the snow above a slush layer of flooded snow. At negative freeboard, surface flooding is most probable and widespread in summer, as the ice is most permeable and the warm snow allows for long-range lateral flow of the sea water at the snow/ice interface. Initially, however, the temperature of the saline slush layer will be close to the freezing point of sea water. The less dense and fresh meltwater percolating through the overlying snow onto the cold slush surface will therefore refreeze at the interface, and will thicken by the addition and refreezing of more meltwater from above. The resulting structure was often observed in the southeast Weddell Sea in summer and is shown in Figure 5a. In this case, a snow-ice layer between the superimposed ice and the gap was missing (Figs 5a and b and 6a). The situation is comparable to the formation of ice below under-ice melt ponds in the Arctic (Reference EickenEicken, 1994), where an ice layer forms at the boundary between a fresh- and a salt-water layer both at their freezing point.

Once the porous or gap layers have formed, the gaps are widened both by absorption of solar radiation and by advection of sea water from the floe edges. Absorption might be enhanced by high chlorophyll-a concentrations which could develop in the gap habitat once it had formed. Slight increases of the gap water temperature above its freezing point lead to melting of the slush within it or of the adjacent ice, as indicated by our gap-water salinity measurements. While in this case the gap water is diluted and might still be in phase equilibrium with its environment, advection of, and exchange with, more saline sea water from the floe edges leads to further melting, finally resulting in the rotten, honeycomb-like ice structure. This will be particularly important closer to the ice edge, where water motion within the floes can be induced by swell penetrating into the ice. Clearly, as also shown by nutrient data (e.g. Reference Fritsen, Lytle, Ackley and SullivanFritsen and others, 1994; Reference Thomas, Lizotte and ArrigoThomas and others, 1998), at the end of summer the permeability of the ice is high enough to allow for considerable exchange with the surrounding sea water. Another source of energy available for melting results from the downward flux of latent heat released by meltwater refreezing during superimposed-ice formation. As with the under-ice melt ponds, a small fraction of this heat is consumed by melting at the very bottom of the surface ice layer.

4.3. Summer surface processes

Our measurements were made while moving through both space and time. While the data from the southeast Weddell Sea represent conditions at the beginning of summer, the northwest Weddell Sea data are typical of the peak of the ablation season. As indicated by low air temperatures and the refreezing of gaps, measurements in the southeast Weddell Sea in fall show the transition back to winter conditions. By summarizing the observations in these regions, we suggest a heuristic model of processes on perennial Antarctic ice from late spring to early fall, shown in Figure 9.

Fig. 9. Schematic drawing of summer surface processes described in the text.

The late-spring situation describes the preconditioning of the ice surface before the onset of surface melt. Freeboard is either positive (Fig. 9a) or negative. In the case of negative freeboard, flood–freeze cycling with the formation of snow ice may be present (Fig. 9c), or the snow may remain dry because the underlying ice is impermeable and flooding from the sides does not reach the interior of floes through the cold snow (Fig. 9b). Either way, the salinity at the top of the ice is high.

In summer, the ice and snow temperature gradients reverse. As a consequence, the porosity at the top of the ice increases drastically (Fig. 9a), incompletely frozen slush layers widen (Fig. 9c), or the ice surface becomes widely flooded (Fig. 9b). Simultaneously, snow metamorphoses and meltwater percolates downward, and superimposed ice forms above the water level, on top of either ice or slush. Generally, the summer season is also characterized by desalination of the original sea or snow ice below the gap. Among these processes, the “freeboard layer” mechanism (Reference Ackley, Buck and TaguchiAckley and others, 1979; Reference Ackley and SullivanAckley and Sullivan, 1994) is just one possibility related to the porosity increase (Fig. 9a).

Finally, as air temperatures drop in fall, and snow and ice temperature gradients become negative again, superimposed-ice formation and widening of gaps cease. Depending on the amount of slush left in the gaps, they will congeal from top to bottom, developing either a columnar or granular texture. This final freezing, with all the consequences of high primary productivity and large vertical heat fluxes (Reference Fritsen, Lytle, Ackley and SullivanFritsen and others, 1994; Reference Lytle and AckleyLytle and Ackley, 1996), concludes the summer season. Following continued accumulation of snow, the ice will begin to undergo the well-known processes of flood–freeze cycling.

4.4. Submergence

One more process complicates the transformations described above (Fig. 9): over summer, the ice slowly submerges further into the water due to changes in the isostatic equilibrium. First, during summer the ice melts from below. Depending on the amount of ocean heat flux and on the initial ice thickness, a thinning of 0.2–0.6 m can occur during summer (Reference Martinson, Iannuzzi and JeffriesMartinson and Iannuzzi, 1998). Secondly, mass is added as either snow or rain to the ice surface. According to data presented by Reference Eicken, Lange, Hubberten and WadhamsEicken and others (1994), snow accumulation from December to February amounts to 0.12–0.16 m at the continental coast of the southeast Weddell Sea, and is much higher in the northwest. Finally, submergence is also due to internal melt, leading to increased brine volumes and therefore higher bulk densities. While density increases of 10–20 kg m⁻³ are possible (based on calculations using the equations of Reference Cox and WeeksCox and Weeks (1983) ) for warming of the ice to close-to-melting temperatures, even higher bulk density rises are likely for the ice-water mixture of the very rotten, honeycomb-like ice.

The amount of submergence in relation to changes in these variables can easily be deduced by differentiating the equation for ice freeboard Z _fb

(1)

for snow and ice thickness z _s and z _i, as well as for ice density ρ _i. Characteristic values are given in Table 3. As can be seen, the three variables in total could yield submergence distances of >0.1 m during the summer months. Thus, drafts of the surface ice layer of 0.14 m (Table 2) can be explained by submergence alone, if the superimposed ice has formed immediately above a slush layer. On the other hand, as explained earlier, the gaps may form already at some depth below the water level.

Table 3. Sensitivity of ice freeboard dz _fb to changes in ice and snow thickness dz _i and dz _s and in ice density dρ _i as calculated by differentiating Equation (1) for each respective variable

The data presented in Figures 5–7 show that the low surface ice salinities and δ ¹⁸O ratios are not changed much when the ice submerges. However, once freeboard becomes very small or even negative, as in Figures 5c and 6b, flooding might again take place, either causing a new cycle or ending superimposed-ice formation completely at the end of summer. Then, salinities at the very top are raised and eventually salt-water droplets may melt themselves into the superimposed ice, destroying the original PG texture (Fig. 6b). Thus, during summer there is a competition between upward superimposed-ice formation and downward ice submergence. Ultimately, the latter takes over towards the end of the ablation season unless it is balanced by ice growth at the ice underside.

Internal snowmelt and superimposed-ice formation will take place both in the perennial and in the seasonal ice zone. In the latter, however, submergence and surface flooding are so strong that no significant amounts of superimposed ice can form above the water level even before new flooding takes place and the ice finally deteriorates.