Ice-core δ ¹⁸O profiles, snow-ice occurrence and snow fractions

The snow ice is differentiated from frazil ice on the basis of differences in their isotopic composition, which arise because frazil ice originates from sea water, while snow ice originates from a mixture of sea water and snow. This concept is illustrated by the ^1 ⁸0 values for snow and sea water (Table 2) and the composite ice-core 6*0 profiles (Fig. 2) for each cruise. The δ ¹⁸O profiles share two basic features. First, the values at the bottom of the ice are all similar and about 2‰ more positive than the sea-water values (Table 2) due to isotopic fractionation during freezing, which enriches the ice in the heavy isotope. Second, the δ ¹⁸O values in the uppermost ice layers are all more negative than those in the lower ice layers and the sea water. This reflects the entrainment of snow into the floes as the base of the snow is flooded with sea water and snow/sea-water mixture subsequently freezes to form snow ice. The δ ¹⁸O values of snow ice are intermediate between those of snow and sea water and depend on the fraction of each component, and the extent of isotopic fractionation that occurred as the snow/ sea-water mixture froze and snow ice formed.

Table 2.

Mean (±7 standard deviation) δ ¹⁸O values for snow and sea water. the number of measurements is given in parentheses

Fig. 2.

Composite δ¹⁸O profiles representing the average of all the δ¹⁸O profiles in all the ice cores obtained during each cruise. because the ice cores were sampled for stable isotopes at irregular intervals according to their stratigraphy, the following procedure was adopted for compiling the profiles: the depth for each δ¹⁸O value was normalized by dividing by the ice-core length; the δ¹⁸O values were binned between the surface and the base of the ice at 11 normalized depth intervals of 0.05, 0.1, 0.2 and so on at intervals of 0.1; and a mean δ¹⁸O was calculated for each bin. each data point in a profile represents the mean δ¹⁸O value for each bin.

The δ ¹⁸O profiles indicate that there was a significant amount of snow ice in the cores. A majority of cores examined during each cruise did indeed contain snow ice (Table 3) and it occurred in all thickness categories (Fig. 3). The bulk of the snow ice occurred on ice of intermediate (0.3−1.2 m) thickness. Occasionally, a relatively large amount of snow ice also occurred on thin (<0.3 m) and thick (>1.2 m) ice. This suggests that sufficient snow can accumulate quickly on young ice to overcome the buoyancy of the ice, leading to flooding and snow-ice formation. Sufficient snow can also accumulate on the thickest ice to overcome its buoyancy. Regardless of the differences between ice-thickness categories, and the spatial and temporal variability evident from cruise to cruise, the essential aspect of the snow-ice data is that, regardless of time, location and ice thickness, snow ice was ubiquitous and had made a substantial contribution to the thickening of much of the ice cover prior to sampling. A total length of 297 m of ice core had been examined after five cruises: 24.5% of that ice was snow ice (Fig. 3d).

Table 3.

Number of ice cores containing snow ice, frazil ice and congelation ice (figures in parentheses represent the proportion ofthe total number of cores)

Fig. 3.

(a-h) Amounts of snow ice as a proportion of the total length of core examined infour ice-thickness categories and in all those categories combined during each cruise. the amount of snow ice in each thickness category as a proportion of the total length of core examined during all the cruises is shown in (d). the ice-thickness categories are based on the wmo sea ice nomenclature (wmo, 1970) used in standardized ship-based observation and characterization of the antarctic pack ice (Reference Worby and Allisonworby and allison, 1999).

To solve for f^s, the fraction of snow in the snow ice, required δ ¹⁸O values for sea water (δ_sw), snow (4) and the snow ice (6). The 6 values are those of the snow-ice sample for which f_s is being determined. For δ_sw we had to use a δ ¹⁸O value of 0‰ rather than the values in Table 2, since we assume that all granular ice with a value <0‰ was snow ice Reference Jeffries, Worby, Morris and WeeksJeffries and others, 1997) For δ_s for each cruise we used the mean and standard deviation values shown in Table 2 (e.g. for cruise NBP 93−5 we used values of −9.4‰, −13.2%, and −17.0‰, i.e. −13.2 ±3.8‰, to calculate f_s values). Using a range of 4 values allows for the variability of snow δ ¹⁸O values that is likely to arise due to atmospheric processes that affect the isotopic ratio of the moisture prior to and during precipitation events, and from factors such as metamorphism, wind erosion and redeposition that affect the isotopic composition of the snow cover once it has been deposited on the floes.

The mean f_s and f_m values calculated using three different δ_s values for individual cruises are illustrated in Figure 4. Depending on δ_s, f_s varied between 4% and 18%, while f_m varied between 0.6% and 8%. The lowest f_s and f_m values occurred in the NBP 98−3 cores. Why those f_s values were so low is unclear, but the low f_m values can be attributed with certainty to the fact that the snow-ice layers were particularly thin for this cruise, as will be shown later. Combining all the cruise data gives mean f_s values of 6.25−12.5% and mean f_m values of 3.65−7.3% (Fig. 4: All). Whether one considers the snow fractions for individual cruises or for all the cruises combined, the f_s and f_m values are similar to those reported in the Weddell Sea (Reference Lange, Schlosser, Ackley, Wadhams and DieckmannLange and others, 1990; Reference Eicken, Lange, Hubberten and WadhamsEicken and others, 1994, Reference Eicken, Fischer and Lemke1995). All published f_s and f_m values are probably conservative estimates because some of the snow entrained in the sea water during flooding dissolves in the melt and is subsequently transferred out of the incipient snow-ice layer as it freezes and some of the rejected brine flows into the underlying ice (Reference Maksym and JeffriesMaksym and jeffries, 2001).

Fig. 4.

Bar graphs showing the mean snow fraction of the snow-ice layers (f_s) and the mean snow fraction of the entire ice thickness (f_m) during each cruise. the short, cross-hatched bars represent f_m. f_s is represented by the height of the open bars plus the cross-hatched bars below.

Though the /^s and f_m values may be conservative, it is apparent from the snow fraction data that the snow entrained in floes as a consequence of snow-ice formation was a modest proportion of the total ice mass in the Ross, Amundsen and Bellingshausen Seas. However, the snow ice itself comprised a more significant proportion of the ice mass (Fig. 3). Just how significant snow-ice formation was to the thickening and mass budget of the ice cover relative to the other two main ice types, frazil and congelation ice, is discussed in the next subsection.

Snow ice, frazil ice and congelation ice: temporal and spatial variability and their role in the thermodynamic thickening of the ice cover

A majority of ice cores on each cruise also contained frazil ice and congelation ice (Table 3). Snow ice, frazil ice and congelation ice commonly occurred together in most cores, but there was considerable spatial and temporal variability in terms of their contribution to the total ice mass (Fig. 5). This variability can be condensed into three main points:

1. 1. ice composition in late winter in the eastern region (80−110° W; NBP 93−5, 95−5B) was almost identical in two different years, and the eastern and central (110−170° W; NBP 94−5) regions had less congelation ice than the western region (164°E −150° W; NBP 95−3, 95−5A, 98−3);

2. 2. there was less snow ice in the western region as a whole in 1998 than in 1995, but in both years there was strong zonation of the ice cover, with an outer pack ice dominated by snow ice and frazil ice (NBP 95−3Ai, 98−3i) and an inner pack ice dominated by congelation ice (NBP 95−3Aii, 98−3ii). The inner pack ice had much less snow ice in 1998 (NBP 98−3h) than in 1995 (NBP 95−3Aii);

3. 3. snow-ice formation was not temporally dependent, i.e. as much snow ice occurred in autumn as in winter.

Fig. 5.

Proportional representation of the amounts of snow ice, frazil ice, congelation ice and other ice observed in ice cores. each column represents the total length of ice core normalized to 1 or 100%. the proportion of each ice type is represented by the length of the space it occupies within the column and by the value in each space. column 1 represents the total length of core analyzed on all the cruises combined. columns 2−8 represent the total length of core analyzed on each cruise and they are arranged in sequence from right to left, east to west, to show the variability in the study area. columns 9−12 represent the total length of core analyzed in the outer and inner pack ice of the ross sea in may/june 1995 and 1998. other ice includes fragmented ice and cavities; the latter are slush- or water filled gaps between ice blocks sometimes encountered in floes where the ice has not consolidated completely since being deformed. cavities are identified as an ice type because they are apart of the total ice thickness and represent the consequences of a particular mode of thickening.

The zonation of the Ross Sea according to the composition of the ice floes in autumn 1995 was reported by Reference Jeffries and AdolphsJeffries and Adolphs (1997). That equally strong zonation should be observed 3 years later indicates that 1995 was not an anomalous year, and confirms that the Ross Sea is unlike other Antarctic sea-ice zones, including the central and eastern regions of this study, because of the large quantity of congelation ice that occurs there. In 1995, the boundary between the outer and inner pack ice, as defined by the contrasts in the amount of snow ice and congelation ice, was at 71° S, and in 1998 it was at 70° S (Fig. 5).

Reference Jeffries and AdolphsJeffries and Adolphs (1997) attributed the differences in ice composition in the Ross Sea inner and outer pack ice to differences in environmental conditions. The outer pack ice is more stormy and dynamic, which favours frazil-ice and pancake-ice formation, and there is greater snowfall, which favours flooding and snow-ice formation. The inner pack ice has colder, calmer conditions, which favour congelation-ice growth, and a lower oceanic heat flux over the continental shelf, which reduces bottom melting. The greater snowfall on the outer pack ice is due to its proximity to a moisture source, i.e. the Pacific Ocean, from which there is a moist airflow. U.S. National Centers for Environmental Prediction re-analysis data for autumn 1998 show that periods of airflow from the Pacific Ocean were shorter and less frequent, and did not penetrate as far into the Ross embayment as they did in 1995 (K. Morris and others, unpublished data). Consequently, the snow cover on the Ross Sea pack ice was thinner and flooding of the snow/ice interface was less widespread in 1998 than in 1995 (Reference Morris and JeffriesMorris and Jeffries, 2001). Hence, in 1998, the amount of snow ice in the Ross Sea pack ice was smaller overall, and the snow-ice-dominated outer pack ice did not extend as far south as in 1995.

Apart from the Ross Sea inner pack ice over the continental shelf, the ice investigated in this study was on the deep ocean, where the oceanic heat flux would be expected to be higher than in shallower waters. Reference McPheeMcPhee and others (1996) have described the important role that the oceanic heat flux at the underside of the ice plays in determining the character of the ice cover through bottom melting. In this study, bottom melting, represented by the number of cores with a smooth, scalloped base and no skeletal layer, was common in autumn and winter, but was spatially and temporally variable (Fig. 6). It was least common in the inner Ross Sea in autumn (NBP 95−3Aii, 98−3ii), and most common in late winter in the outer Ross Sea pack ice (NBP 95−5A) and the central region (NBP 94−5). The amount of snow ice was moderately well correlated with the occurrence of bottom melting (Fig. 6). Bottom melting was much less widespread and there was a smaller amount of snow ice in the Ross Sea inner pack ice (NBP 95−3Aii, 98−3ii) than in the outer pack ice (NBP 95−3A1, 95−3B, 95−5A) and the eastern region (NBP 93−5, 95−5B). The amount of snow ice was greatest in the central region (NBP 94−5), where the widespread bottom melting was attributed to the lateness of the season and a high oceanic heat flux (Reference Jeffries, Jaña, Li and McCullarsJeffries and others, 1995, citing personal communications from H. H. Hellmer and S. S. Jacobs).

Fig. 6.

Scatter plot of the amount of snow ice on each cruise as a function of the occurrence of bottom melting, as represented by the proportion of the total number of ice cores with a smooth, scalloped ice/water interface. the correlation coefficent (r) is statistically significant at the 95% confidence level.

The effect of bottom melting has been likened to a conveyor belt (Reference Ackley, Lytle, Kuehn, Golden and DarlingAckley and others, 1995), where ice that melts off the bottom is replaced by snow ice at the snow/ice interface. Bottom melting, in effect, increases the snow load and the likelihood of flooding and snow-ice formation. It appears that high effective snow loads and the bottom-melting/ snow-ice conveyor belt were particularly common in the Ross Sea outer pack ice, and the central and eastern regions of this study in autumn and winter.

If we consider only the bulk composition of the ice, it might be concluded that, in general, frazil-ice and/or congelation-ice formation dominated the development of the ice cover because one or the other, and sometimes both, exceeded the amount of snow ice (Fig. 5). However, the bulk composition gives a misleading impression of the importance of each ice type, as it does not account for the relative contributions of thermodynamic and dynamic thickening. Throughout the Antarctic sea-ice cover it has been found that ice cores are often composed of many layers of frazil and congelation ice (e.g. Reference Eicken, Lange and DieckmannEicken and others, 1991; Reference Lange and EickenLange and Eicken, 1991; Reference Worby and MassomWorby and Massom, 1995; Reference Jeffries, Worby, Morris and WeeksJeffries and others, 1997, Reference Jeffries, Li, Jaña, Krouse, Hurst-Cushing and Jeffries1998). The individual layers of frazil and congelation ice represent the original, thermodynamically formed building-blocks of the ice cover, which subsequently thickened further by dynamic processes, i.e. rafting and ridging.

The relative importance of snow ice to the thermodynamic thickening of the ice cover can be judged by comparing the thickness of snow-ice layers to those of frazil ice and congelation ice (Fig. 7). We consider only the snow ice found as the uppermost layer of the ice cores, i.e. where it originally formed thermodynamically by the freezing of the mixture of snow and sea water/brine that resulted from flooding at the base of the snow cover. The few "buried" snow-ice layers that were observed have been excluded from the analysis as they are found below the surface as a result of dynamic processes (Reference Lange, Hubberten, Maeno and HondohLange and Hubberten, 1992; Reference Worby, Jeffries, Weeks, Morris and JafiaWorby and others, 1996). As Figure 7 shows, on five occasions the congelation-ice layers were thicker than the frazil-ice layers; on four occasions the snow-ice layers were thicker than the congelation-ice layers; and on six occasions the snow-ice layers were thicker than the frazil-ice layers. Snow ice and congelation ice, then, were often more important to the thermodynamic thickening of the ice cover than frazil-ice formation, and on some occasions snow ice was more important than congelation ice.

Fig. 7.

Mean thickness of individual layers of snow ice, frazil ice and congelation ice in ice cores. fra_zil-ice layer thickness data are not available for cruise nbp 94−5.

This is not to suggest that frazil-ice formation was unimportant in the development of the ice cover. On the contrary, here, as elsewhere in Antarctic waters, we believe that frazil ice played a vital role in the initial expansion of the pack ice through the "pancake cycle" in which rafting of the pancakes plays a large role in the dynamic thickening of the ice cover (Reference Wadhams, Lange and AckleyWadhams and others, 1987; Reference Lange, Ackley, Wadhams, Dieckmann and EickenLange and others, 1989). However, once the initial frazil/pancake-dominated ice cover had formed and stabilized, snow ice and congelation ice were often more important to the subsequent thermodynamic thickening of the ice.

The importance of frazil ice to the thermodynamic thickening of the ice cover should not be underestimated, however. As Figure 7 shows, when the data from all five cruises are combined, the frazil-ice layers were thicker on average than the congelation-ice layers. But the data from all five cruises also show that, on average, snow ice made a greater contribution to the thermodynamic thickening of the ice cover than either frazil ice or congelation ice (Fig. 7).

Snow ice: a circum-Antarctic perspective

In the Weddell Sea, snow ice has been reported to make up as much as 16% of the ice volume (Reference Lange, Schlosser, Ackley, Wadhams and DieckmannLange and others, 1990; Reference Lange and EickenLange and Eicken, 1991), and numerical simulations indicate that snow-ice formation is most likely to occur during the latter part of the ice-growth season (Reference Eicken, Fischer and LemkeEicken and others, 1995). In the East Antarctic pack ice, the amount of snow ice "is seasonally variable, with a lower percentage early in the growth season compared to spring" and has been reported as comprising 12−46% of the total ice volume during different cruises, and 13% for all cruises combined (Reference Worby, Massom, Allison, Lytle, Heil and JeffriesWorby and others, 1998).

The pack ice in the Ross, Amundsen and Bellingshausen Seas is quite unlike that elsewhere, in that there is a greater amount of snow ice at any time of the winter compared to the Weddell Sea and East Antarctic pack ice. We attribute the greater amount of snow ice in the Ross, Amundsen and Bellingshausen Seas to greater snow loads and more extensive flooding. This is not because the ice is thinner than elsewhere. We have shown previously that in autumn and winter the ice in the Ross, Amundsen and Bellingshausen Seas is at least as thick as, and sometimes thicker than, in the Weddell Sea and the East Antarctic pack ice (Reference Jeffries and AdolphsJeffries and Adolphs, 1997; Reference Jeffries, Li, Jaña, Krouse, Hurst-Cushing and JeffriesJeffries and others, 1998). Ship-based observations and remote sensing indicate that the snow depth on the ice in the Ross, Amundsen and Bellingshausen Seas is greater than elsewhere (Reference Markus, Cavalieri and JeffriesMarkus and Cavalieri, 1998). This, coupled with the fact that more snow ice forms in this region than elsewhere, indicates that there is greater snow accumulation and thus a greater snow load on the ice.

Snow-cover depletion and total snow accumulation

As the base of the snow cover floods and refreezes to form snow ice that contributes to the thermodynamic thickening of the ice cover, the depth of snow is reduced and the remaining snow does not represent the total amount of snow accumulation. This was illustrated with the NBP 94−5 and 95−3 data (Reference Sturm, Morris, Massom and JeffriesSturm and others, 1998) and predicted with a Weddell Sea ice model (Reference Fritsen, Ackley, Kremer, Sullivan, Lizotte and ArrigoFritsen and others, 1998). In this subsection we estimate for each of the five cruises the proportion of the total snow accumulation that remained on the floes after snow-ice formation, and use that information to adjust the Set B snow-depth observations to derive estimates of the total snow accumulation on different ice-thickness categories.

To estimate the proportion of the total snow accumulation that remained on the floes, we assume that the thickness of the snow-ice layers is the same as the depth of snow lost to snow-ice formation. Then we regress the Set A mean snow depth against the sum of the Set A mean snow depth and the mean snow-ice layer thickness on each floe. The results (Fig. 8) show that there is some scatter in the data because: (1) we do not have as many snow-ice layers to average as we have Set A snow-depth measurements on each floe; (2) the Set A data include data from the tops of ridges, where the snow depth is reduced by wind erosion, and the flanks of ridges where snow depth increases due to deposition/drifting; (3) there might have been further snow accumulation since the last flooding and snow-ice formation event; (4) some fraction of snow might have been transferred off the floes into leads by the wind; and (5) there might have been some snow settling.

Fig. 8.

Scatter plots of the mean snow depth as a function of the sum of the mean snow depth and the mean snow-ice layer thickness on each ice floe during each cruise. the correlation coefficents (r) for the regression equations/lines are statistically significant at the 95% confidence level. the standard error (s.e.) of the estimate of snow depth is given with each regression equation. the regression lines have been forced through the origin at zero on the assumption that, if there is no snow, neither flooding nor snow-iceformation will occur.

Despite the scatter, each regression line is statistically significant and the results indicate that as little as 42% of the total snow accumulation remained on the floes during cruise NBP 94−5 (Fig. 8f), while as much as 70% remained on the floes during cruise NBP 93−5 (Fig. 8a). When the data from all five cruises are combined, the regression equation, y = 0.49× (r = 0.95), indicates that 49% of the total snow accumulation remained on the floes. The results of this analysis corroborate model predictions that flooding and snow-ice formation will reduce the snow depth by >50% (Reference Fritsen, Ackley, Kremer, Sullivan, Lizotte and ArrigoFritsen and others, 1998).

The slopes of the regression lines offer the means to predict the total snow accumulation on the ice from the observed Set B snow-depth data. The results, presented in Figure 9, show the observed and predicted snow depth on three different Set B ice-thickness categories (<0.3 m, 0.3−0.7 m and 0.7−1.2 m) during each cruise. Ice >1.2 m thick has been excluded, as it includes an unknown amount of ridged ice, which is not part of this analysis of snow-ice formation and snow-cover depletion on the "level", unridged ice.

Fig. 9.

Observed and predicted estimates of mean snow depth on three ice-thickness categories and on all categories combined during each cruise. the observed data, the set b hourly snow-depth estimates, are represented by the short, closed cross-hatched bars. the predicted data, estimated using the regression equations in figure 8, are represented by height of the open cross-hatched bars plus the closed cross-hatched bars below. the error bars are twice the standard error of the estimate of snow depth (see fig 8) and represent error estimates for the predicted snow-depth values at a 95% confidence level.

During each cruise, the observed and predicted snow depth increased as ice thickness increased (Fig. 9). This is to be expected, since thicker ice is generally older ice and has had more time to accumulate snow. That older ice accumulates more snow than younger ice is demonstrated by the ice in the central region, which was investigated in very late winter through very early spring and had the highest total snow accumulation of the entire study: a mean value of 0.38 m for all ice-thickness categories combined (Fig. 9a).

However, snow accumulation is not determined solely by the age of the ice. In autumn 1995 in the western region (Fig. 9d and e), the total snow accumulation was almost as great as it was in the eastern region in late winter 1993 and 1995 (Fig. 9b and c). The resultant early, high snow load would explain why there was as much snow ice in autumn in the western region, particularly in the outer pack ice, as there was in winter in the eastern region (Figs 3 and 5). The greater amount of precipitation on the younger ice cover in the western region may be due to regional and interannual differences in atmospheric circulation and moisture transport. As noted earlier, year-to-year differences in atmospheric circulation and moisture transport probably account for the lower observed and total snow accumulation in the western region in autumn 1998 (Fig. 9g) than in autumn 1995 (Fig. 9d and e), and the resultant differences in amount of snow ice, particularly in the inner pack ice (Figs 3 and 5).