2.1. Texture

The Siple Dome A ice core is stored at −35°C at the US National Ice Core Laboratory (NICL). Seven samples were transferred from NICL to Dartmouth at transport temperatures not exceeding −10°C. In Dartmouth’s Ice Research Laboratory, the samples were stored and prepared for analysis at −20°C.

The Siple Dome core was dated using annual layering through 514 m (i.e. through the Holocene period) (Reference TaylorTaylor and others, 2004a). Below this depth, approximate dating can be derived from control points calculated by comparing methane and δ¹⁸O gas age with that from GISP2 (Reference BrookBrook and others, 2005). There are 22 control points beginning at 514.78 m (8.33 ka) and ending at 919.88 m (56.70 ka) (Reference BrookBrook and others, 2005). Sample depths are shown in Table 1 with their approximate ages calculated by our group using simple linear interpolation from the control points given in Reference BrookBrook and others (2005). This method produces dates different from those in the discrete depth ion-chemistry data (P. Mayewski, http://nsidc.org/data/waiscores/pi/mayewski.html) but adequate for our purposes.

Table 1.

Depths of Siple Dome ice-core samples examined and approximate age, based on linear interpolation using Reference BrookBrook and others (2005)

Vertical thin sections were prepared and then photographed on a light table through crossed polarizers. Grain size was measured using multiple images of each sample (taken at different angles of rotation between the polarizers) and using the pixel-counting area measurement utility of an image-processing application, Image SXM (Reference RasbandRasband, 2006; S.D. Barrett, http://www.liv.ac.uk/∼sdb/ImageSXM). No correction was made for grains intersected by thin-section edges or for the limitations associated with measuring grain size in only two dimensions.

2.2. Fabric

For orientation analysis, electron backscatter diffraction (EBSD) on an FEI XL-30 environmental scanning electron microscope (SEM) with a liquid-nitrogen-chilled cold stage was used. The technique for sample preparation and scanning electron microscopy of uncoated ice is described in detail in Reference Iliescu, Baker and ChangIliescu and others (2004) and Reference ObbardObbard (2006). Samples for analysis were rough-cut slightly larger than the desired final dimension using a bandsaw, then transferred to a HEPA-filtered laminar-flow clean-room hood for surface smoothing with a pre-cleaned standard safety razor. The SEM specimens were stored in a sealed container at −25°C for ∼24 hours before examination in the SEM. In the SEM, a specimen was initially chilled to a temperature of approximately −75 ± 5°C while vacuum conditions were achieved. Then the SEM chamber was placed in environmental mode, and the cold-stage temperature increased to −45 ± 5°C for the duration of analysis.

The EBSD data are reported in two ways. Pole figures display c- and a-axis information for each depth (Fig. 3, further below). Misorientation is defined as the smallest angle by which one crystal can be rotated around a common axis to achieve lattice coincidence with a second crystal. The misorientation angles are presented in a histogram (Fig. 4, further below) which compares the distribution of rotations between adjacent grains (correlated data, shown with bars), with the distribution of misorientation angles between random pairs of grains in the dataset (uncorrelated data, dashed line) and the theoretical random distribution curve for a system of randomly oriented hexagonal grains (solid line). The theoretical distribution curve is produced by the HKL software according to the procedures of Reference MackenzieMackenzie (1958) and Reference MorawiecMorawiec (1995). An uncorrelated distribution that differs from the theoretical random indicates a preferred orientation (or fabric) overall in the material, which would also be observed on the pole figures. A distribution of correlated misorientations that differs from that of the uncorrelated misorientations, however, suggests a special relationship between adjacent grains such as that resulting from polygonization. Although every effort was made to collect only one EBSD pattern from each grain observed in the secondary electron image while the sample was in the SEM, the misorientation between adjacent grains was used to eliminate patterns that might represent the same grain. Reference ObbardObbard (2006) found that within a single crystal, there was a change in specimen orientation of ∼1° over a span of 2 mm corresponding to a ∼1° orientation noise, consistent with that indicated by Reference Bate, Knutsen, Brough and HumphreysBate and others (2005). Therefore, all EBSD patterns differing in orientation by <1° were assumed to represent the same grain.

Due to the large crystal size, a manual (rather than automatic) mapping procedure was used to collect nine EBSPs in a 3 × 3 array from each of nine contiguous 600 μm × 600 μm areas across each sample. Thus, a pattern was collected every 200 μm when possible. Patterns could not be collected from pores or grain boundaries. This mapping method resulted in superfluous data when adjacent points fell within the same grain. In order not to over-represent an orientation, the extra points needed to be identified and nullified. Misorientation data were used to identify adjacent points having the same orientation. If two points had a misorientation of 1° or less, they were considered to be identical, and all points representing the same orientation were nullified except for one. The resulting subset of unique orientations was used to diagram grain boundaries for the sample, with lines between separating unique orientations. This process was repeated in each area examined, producing a 0.6 mm × 5.4 mm diagram of grain boundaries.

The pole figure analysis performed for each sample is described below, using one sample as an example. Figure 1 includes upper-hemisphere pole figures showing the c- and a-axis orientations, {0001} and {11-20}, respectively, for one of six samples from 670 m. Due to hexagonal symmetry, each unique crystal orientation is represented by one point on the c-axis pole figure and three points on the a-axis pole figure. Note that these pole figures contain distinct clusters of poles. These clusters, identified by colored circles in Figure 1, were compared to the grain boundary diagrams, and both in this sample and many others (700–790 m), the poles within each cluster happen to originate from adjacent grains on the sample.

Fig. 1.

Pole figures (top) and grain-mapping schematic (bottom) for one sample from 670 m. 43 unique orientations are plotted in the pole figures. Poles from adjacent grains are circled.

We then compared the angular spread of the points within each cluster and the misorientation between separate clusters (their centers). Misorientation within a given c-axis cluster for the sample shown in Figure 1 ranged from <1° to 16° (mean 7°), while misorientation within a given a-axis cluster ranged from <1° to 8° (mean 3°). The c-axis misorientation between different clusters ranged from 36° (light green to yellow in Fig. 1) to 79° (orange to blue), with a mean of 51°.

2.3. Radar data

Radar data were acquired along the northern flank of Siple Dome as part of a different study focused on grounding-line behavior along the northern and southern edges of Siple Dome (Reference Catania, Hulbe and ConwayCatania and others, 2010). The radar system is custom-built, low-frequency, ground-based and is useful for imaging deep internal layers with a center frequency of 3 MHz. At this frequency the radar antennae provide a pulse wavelength in ice of ∼56 m. To improve the signal-to-noise ratio, our data consist of 1600 waveforms recorded over a horizontal spacing of ∼17 m. As a result, steeply dipping stratigraphy within this horizontal distance can be aliased. Further processing includes removal of the mean waveform (to limit interference due to instrumentation artifacts) and application of a bandpass filter. We convert the two-way travel time to depth by assuming a radar wave speed in ice of 168 m μs⁻¹ . We account for faster wave speeds in the firn column using depth–density measurements from Siple Dome (Mayewski, http://nsidc.org/data/waiscores/pi/mayewski.html) and the Looyenga mixing equation (Reference Glen and ParenGlen and Paren, 1975).

3.1. Texture

The vertical thin sections we produced were very similar in appearance to those shown in Reference DiPrinzio, Wilen, Alley, Fitzpatrick, Spencer and GowDiPrinzio and others (2005) and Reference Gow and MeeseGow and Meese (2007a). Grain boundaries were irregular (sutured, tortuous or wavy), and extinction (color within a grain) was undulose, indicating strain and hence bending or buckling of the basal plane in response to stress (Reference Gow and MeeseGow and Meese, 2007a). Grain size (area) for each depth is shown in Table 2, along with grain sizes derived from Reference DiPrinzio, Wilen, Alley, Fitzpatrick, Spencer and GowDiPrinzio and others (2005), using their graph of grain radius versus depth. Reference DiPrinzio, Wilen, Alley, Fitzpatrick, Spencer and GowDiPrinzio and others (2005) reported grain size as the radius of a circle having the same area as the average grain size measured. They also used a correction factor (in the larger-grained samples) to account for the grains that were cut off by the boundaries of the thin section, and noted that this correction may have affected mean grain size (but not the data trend), which may explain the large apparent grain size at 640 and 670 m.

Table 2.

Mean and median grain area, along with standard deviation for each depth sampled, and mean grain area based on grain radius found in Reference DiPrinzio, Wilen, Alley, Fitzpatrick, Spencer and GowDiPrinzio and others (2005). Because the grain size was large relative to sample size, 100 grains could not be measured in each sample

The large standard deviations in the 640–670 m samples reflect a mixture of smaller and much larger grains. Reference Gow and MeeseGow and Meese (2007a) also noted this bimodal distribution at 640 m. Table 3 shows the mean, standard deviation and median values for grain area excluding large grains (>60 mm²). The samples from 727 and 750 m did not contain the larger grains and had a more consistent, smaller grain size. For the other samples, excluding large grains results in the mean grain size approaching the median.

Table 3.

Grain area mean and median for each depth sampled excluding grain areas greater than 61 mm²

Figure 2 breaks down the grain sizes into 10 mm² bins. The 640 m sample has a bimodal distribution, where a significant number of grains have an area of 61 mm² or greater, while the majority have an area between 0 and 30 mm². The 670 and 700 m samples also have a few much larger grains. Nonetheless, the grain size in all three samples is skewed towards the smaller end, and the sample distributions become narrower and centered in the small grain area bins as the depth increases to 750 m. Below 750 m, the distributions become slightly broader and begin to include larger area grains, again shifting the median size to greater values.

Fig. 2.

For each of the seven depths tested, (a) fraction of all grains measured having a given area (range), and (b) fraction of total sample cross-sectional area that was found in each grain area range.

3.2. Fabric

The data gleaned from EBSD are useful for distinguishing between recrystallization mechanisms, as discussed here and seen in our results which follow.

Polygonization occurs when interactions of dislocation stress fields cause the dislocations to align, producing subgrain boundaries and additional new grains that differ in orientation from the c-axis of the original grain by only a few degrees, with rotation around an a-axis. In fact, because of its nature, we would expect polygonization to produce c- and a-axis clusters with somewhat less misorientation within the latter. Polygonization can also explain the lack of increase in mean grain size, despite increasing depth and age, because larger grains split into smaller ones (Reference Alley, Gow, Meese, Fitzpatrick, Waddington and BolzanAlley and others, 1997; Reference Castelnau, Canova, Lebensohn and DuvalCastelnau and others, 1997; Reference GowGow and others, 1997). Polygonization has been specifically identified between 1551 and 1745 m at GISP2 (Reference ObbardObbard, 2006), at 650 m in the European Greenland Icecore Project (GRIP) (Reference Castelnau, Canova, Lebensohn and DuvalCastelnau and others, 1997), and between 400 and 1200 m at Byrd, Antarctica (Reference Alley, Gow and MeeseAlley and others, 1995). Polygonization may cause a widening of an existing c-axis fabric (e.g. a single maximum) (Reference De La Chapelle, Castelnau, Lipenkov and DuvalDe La Chapelle and others, 1998), and, because individual larger grains split into two or more smaller grains of similar orientation, small clusters of poles on both c- and a-axis pole figures may also be produced.

Migration recrystallization, the nucleation and growth of new grains amongst larger, older ones, can also produce multi-maxima or clustering in pole figures (Reference AlleyAlley, 1988). An examination of nearest-neighbor misorientations is useful for distinguishing between migration recrystallization and polygonization. The former tends to produce large grains with tortuous grain boundaries, and the repeated intersection of a single large, complex grain by a plane might resemble a set of similarly oriented smaller grains, with clusters on both {0001} and {11-20} pole figures. However, there would be no pattern in the location of these grains, i.e. similarly oriented grains would not necessarily be adjacent to one another. Thus, it is unlikely that migration recrystallization is the cause of the pole figure clusters associated with adjacent grains.

Due to the large grain size, insufficient grain orientation data were available from the 640 m sample to construct a pole figure. Reference Gow and MeeseGow and Meese (2007a), however, report a multi-maxima fabric at this depth. Pole figures for the other samples we tested are shown in Figure 3.

Fig. 3.

Pole figures for (a) 670 m (6 samples, 131 grains), (b) 700 m (7 samples, 335 grains), (c) 727 m (5 samples, 391 grains), (d) 750 m (6 samples, 367 grains) and (e) 770 m (6 samples, 378 grains).

Pole figures for 670 m (Fig. 3) reveal clustering of both c- and a-axes. The misorientation histogram (Fig. 4) shows a significant number of low angle misorientations (<10°) between adjacent grains, as well as some high angle misorientations (35–80°). The low misorientation angles within pole figure clusters represent adjacent grains mis-oriented by ∼2° such that there was a net orientation change from one end of a region of adjacent grains to the other. The angular spread was dependent on the size of the region, with the larger regions tending to have larger angular spreads. These low angle misorientations suggest polygonization, according to Reference Alley, Gow and MeeseAlley and others (1995), who established that the misorientation angle distribution reveals the dominant grain-size reducing mechanism.

Fig. 4.

Misorientation angle distribution for (a) 670 m, (b) 700 m, (c) 727 m, (d) 750 m and (e) 770 m, based on orientations shown in Figure 3. Correlated values (blue) are between user-defined adjacent grains; uncorrelated values (dashed red line) are between 1000 randomly chosen pairs of measured grains. The random (theoretical) orientation (black line) is for a randomly oriented hexagonal crystal.

Interestingly, even the large misorientations (35–80°) between different clusters are primarily due to differences in c-axis orientation. Often, several different clusters on the c-axis pole figure shared some a-axis locations on the {11-20} pole figure. This may mean that even larger misorientations originated in polygonization around a common a-axis.

The sample from 700 m is different in that it has a single-maximum c-axis fabric (Fig. 3) and a significantly smaller mean grain size. This depth had a high frequency of low angle misorientations (<10°) and a significant number of misorientations between 10° and 30° (Fig. 4). As in the 670 m sample, the low angle misorientations were primarily from c-axis variations between adjacent grains with a common a-axis. Larger misorientations were generally associated with differences in a-axis orientation.

These results agree with those obtained by Reference DiPrinzio, Wilen, Alley, Fitzpatrick, Spencer and GowDiPrinzio and others (2005) and Reference Gow and MeeseGow and Meese (2007a), which also show a reduction in grain size accompanied by a single-maximum fabric at 700 m. The polygonization process would produce the reduction in mean grain size seen here. Typically, the transition from a vertical girdle to a single-maximum fabric has been attributed to a change in the stress state, from pure shear to simple shear (Reference AlleyAlley, 1992).

In the sample from 727 m, the c-axis fabric is again a single maximum (Fig. 4) and the increased strength of the fabric is reflected in the peak in uncorrelated misorientations. A higher proportion of the misorientations are low-angle for this depth than at 700 m (Fig. 4). The low angle misorientations can be attributed to polygonization. The larger misorientations are due to differences in a-axis orientation.

The single-maximum c-axis fabric is also found at 750 m (Fig. 4). Overall, the fabric of this sample was similar to those from 700 and 727 m, but with an even greater proportion of low angle misorientations (Fig. 4). Based on their distribution, the low angle misorientations can be attributed to polygonization within larger original grains. In two of the six specimens tested, regions having adjacent grain misorientations, attributed to differences in c-axis orientation, between 45° and 50° were found. Importantly, this was also observed in the deeper sample (770 m) where in situ temperatures would permit migration recrystallization.

The c-axis fabric at 770 m (Fig. 4) may be classified as either a broken vertical girdle or a multiple maxima. The former would suggest a return to a state of uniaxial longitudinal tension; hence the latter seems more likely. A multiple maxima may represent grains produced by migration recrystallization, with large misorientations between those that are themselves experiencing polygonization, thus producing widely separated clusters. There is again a high proportion of low angle misorientations (<10°), but this time other misorientation angles are spread across the range 15–90° and are due to c-axis tilting in some cases and a-axis rotation in others (Fig. 5).

Fig. 5.

3 MHz radar data across the northern flank of Siple Dome. Data in the upper ∼50 m (shown as a black band) are obscured by the direct wave from the transmitter. Inset shows location relative to Siple Dome. The grounding line was crossed at ∼60 km and is characterized by basal crevasses and downwarped internal stratigraphy.

EBSD data were not collected from the 790 m sample due to large grain size (few grains). Reference DiPrinzio, Wilen, Alley, Fitzpatrick, Spencer and GowDiPrinzio and others (2005) document a single-maximum fabric at this depth, but with a significant number of off-vertical poles, similar to our 770 m c-axis pole plot (Fig. 4). This is consistent with the onset of migration recrystallization while polygonization is still taking place.

3.3. Impurities

Glacial ice contains soluble and insoluble impurities, which are both of climatological interest and an important factor in the microstructural evolution of ice (Reference Fisher and KoernerFisher and Koerner, 1986). Continental dust, sea-salt aerosols, and species formed in the atmosphere are most common, but volcanic eruptions contribute both insoluble particles and H₂SO₄ (Reference Kreutz and MayewskiKreutz and Mayewski, 1999; Reference Siggaard-AndersenSiggaard-Andersen, 2005). In SEM studies of ice from Antarctica (e.g. Byrd and Vostok), 5–10 μm white spots are commonly observed on crystallographic facets in the grain interiors and along grain boundaries, and sometimes coalesce to form filaments in grain boundaries during sublimation (Reference CullenCullen, 2002). EDS reveals that these contain predominantly Na, S, Cl and Mg. Larger particulates (∼60 μm) are sometimes found within the lattice and these consist primarily of Si and Al with K, Na and C and occasionally Cl, Mg, Fe, S and Ti. These are generally attributed to continental dust or volcanic ash.

The presence of certain impurities above critical levels affects glacier flow and ice deformation, so it is important to analyze the type of impurities present and their location within the microstructure. In general, the effect of the impurities on mean grain size is dependent on the relationship between their volume fraction and size (Reference Humphreys and HatherlyHumphreys and Hatherly, 2004). There have been numerous studies of the effect of impurities on mechanical processes in ice (Reference Jones and GlenJones and Glen, 1969; Reference Nakamura and JonesNakamura and Jones, 1970, 1973; Reference Hooke, Dahlin and KauperHooke and others, 1972; Reference Baker and GerberichBaker and Gerberich, 1979; Reference Iliescu, Baker and LiIliescu and others, 2003; Reference Jacka, Donoghue, Li, Budd and AndersenJacka and others, 2003; Reference Song, Cole and BakerSong and others, 2004, Reference Song, Cole and Baker2005a,Reference Song, Cole and Bakerb, Reference Song, Cole and Baker2006, Reference Song, Cole and Baker2007a,Reference Song, Cole and Bakerb, Reference Song, Baker and Cole2008; Reference Li, Iliescu and BakerLi and others, 2009). However, ion chromatography of discrete samples of the pre-Holocene Siple Dome A core suggests that no unusually high concentrations of any major ion are present at the depths tested (Mayewski, http://nsidc.org/data/waiscores/pi/mayewski.html). There is a somewhat higher concentration of Ca (10–20 ppb) at 700, 727 and 750 m than elsewhere between 640 and 790 m (Mayewski, http://nsidc.org/data/waiscores/pi/mayewski.html), but not nearly as much as found in fine-grained layers of the Vostok core (117–409 ppb). Nor did the SEM reveal impurities on the grain boundaries of these Siple Dome samples, except at 727 and 750 m, where there was a single location in each that contained Na, Cl, K and S. Although the Ca may play a role in limiting grain growth, as it seems to in the Vostok core, evidence that polygonization is the primary cause of reduced mean grain size in the Siple Dome A layers between 700 and 800 m is overwhelming.

A number of 20–100 μm (greatest dimension) Si-, Na- K- and Al-containing particles were also found in the crystal lattice. These are probably insoluble dust or ash. In their study of 300 specific volcanic ash and dust layers in the 700–800 m depth range in Siple Dome, Reference Gow and MeeseGow and Meese (2007b) measured an average particle size of 50 μm in ash layers and 5 μm in dust bands and identified peralkaline trachyte (KAlSi₃O) and alkaline feldspar (NaAlSi₃O₈), plus accessory minerals containing K, Na, Mg, Fe, Mn, Ti, Zn, Li, Cr, Al, Ca and Si. The samples studied here did not contain visible debris layers, but it is not surprising that they did contain some particles in the size range determined by Reference Gow and MeeseGow and Meese (2007b).

Inert, second-phase particles have been shown, both theoretically and empirically, to inhibit grain growth and grain-boundary migration in polycrystalline materials, including ice (Reference Dunn, Walter and MargolinDunn and Walter, 1966; Reference Gow and WilliamsonGow and Williamson, 1976). Indeed, grain sizes are often found to be smaller in visually dirty ice than in clean ice of the same age (Reference Gow and WilliamsonGow and Williamson, 1976; Reference Duval and LoriusDuval and Lorius, 1980). Initially, a causal relationship was drawn between the increase in particle concentration and the decrease in grain size across the Holocene–Wisconsin boundary in several ice cores (Reference Koerner and FisherKoerner and Fisher, 1979). Later, however, two separate groups (Reference Duval and LoriusDuval and Lorius, 1980; Reference Alley, Perepezko and BentleyAlley and others, 1986) compared the intrinsic driving force for grain growth with the drag force from particles in ice. Reference Alley, Perepezko and BentleyAlley and others (1986) concluded that at the particle concentrations present in typical clean glacier ice, the particles had little effect on grain growth. Even in visibly dirty glacier ice, which the samples examined here were not, the higher particle concentration does not always fully explain decreases in grain growth rate (Reference Alley, Perepezko and BentleyAlley and others, 1986). Hence, while it is not surprising to find debris in the Siple Dome samples we studied, it is not the cause of the differences in microstructure observed.

3.4. Radar data

Previous radar data collected on Siple Dome found continuous isochrones to 700 m and an absence of any visible stratigraphy below that depth (Reference Nereson, Raymond, Waddington and JacobelNereson and others, 1998). Our improved radar has a higher sampling frequency allowing layers to become visible to nearly the base of the ice sheet and allows comparison to ice-core characteristics at greater ice thicknesses. We find that the deepest continuous layer across Siple Dome is at ∼757 m depth. Below this, we observe an abrupt transition to discontinuous stratigraphy that appears to extend to the base of the ice sheet across the entire northern flank of the dome (Fig. 5). Other radar data previously reported for Siple Dome with the same radar system show similar stratigraphy on the southern flank of Siple Dome, close to its grounding line (Reference Catania, Hulbe and ConwayCatania and others, 2010, fig. 3). The lack of continuous layers below 757 m is consistent with electrical conductivity measurements of the ice core by Reference Taylor and AlleyTaylor and Alley (2004) who report finding disturbed layers at 803 m depth. Layered stratigraphy is found below this disturbed zone (Reference Taylor and AlleyTaylor and Alley, 2004), but our radar system is unable to detect these layers, likely because the signal has been strongly attenuated through the disturbed stratigraphy.

The deepest continuous layer varies with depth in such a way as to create undulations that are not related to the bed or surface topography and instead may be due to deposition over the deeper disturbed stratigraphy. Because of the coarse horizontal spacing of our radar waveforms during acquisition, we believe that the disturbed stratigraphy has been aliased. As such, it is difficult to determine whether the stratigraphy is continuous and steeply dipping with respect to the present-day ice-sheet surface, or whether the radar is returning diffraction hyperbolae. Hyperbolae in radar data may result from scatter due to crevasses that are filled or bridged, and then buried by subsequent accumulation (e.g. Reference Retzlaff and BentleyRetzlaff and Bentley, 1993; Reference Smith, Lord and BentleySmith and others, 2002; Reference Catania, Conway, Raymond and ScambosCatania and others, 2005) or from isolated pockets of contrasting permittivity, such as from sediment or water inclusions in the ice (Reference Clarke, Liu, Lord and BentleyClarke and others, 2000). We rule out this latter hypothesis because freezing temperatures are found at the base of Siple Dome (Reference EngelhardtEngelhardt, 2004) that have persisted for long enough to refreeze any available water. Further, sediment is found in only the lower 2 m of core (Reference Gow and MeeseGow and Meese, 2007a). Tephra deposits found at the depths of the disturbed stratigraphy (Reference Gow and MeeseGow and Meese, 2007a) are widely deposited and will not cause the layer shape to be altered. Instead, these same tephra deposits have been used to explain the occurrence of a bright continuous reflector present throughout most of the West Antarctic ice sheet in radar data (Reference Jacobel and WelchJacobel and Welch, 2005).