Ice shelves provide an important buttressing mechanism that restrains the ice flow from ice sheet interiors towards the ocean. The disintegration of the Larsen A and B ice shelves on the east coast of the Antarctic Peninsula resulted in substantial acceleration of the glaciers that formerly flowed into the area lost (De Rydt et al., 2015; Hulbe et al., 2008; Rignot et al., 2004; Rott et al., 2011; Scambos et al., 2004; Shuman et al., 2011). These events have led to considerable current interest in the mechanisms of ice shelf fracture and break-up and the associated consequences for the far-reaching upstream impacts of such changes (Reese et al., 2018). Work on several ice shelves has demonstrated an association between internal structure and the rate and direction of rift propagation. On the Ronne Ice Shelf, Hulbe et al. (2010) noted that crack tips coincided with suture zones between ice originating from adjacent outlet glaciers through most of the advective path, with rifts propagating through the suture zones only near the ice front. On Amery Ice Shelf, Walker et al. (2015) showed rifts changing their propagation direction several times on entering the suture zone between two ice streams. A number of studies have focussed on the Larsen C Ice Shelf since the break-up of its northern neighbour (Glasser et al., 2009; Jansen et al., 2013; Kulessa et al., 2014; McGrath et al., 2012). Here, the details of the formation of suture zones with a high proportion of marine ice in areas immediately downstream of coastal promontories have been documented (Holland et al., 2009; Jansen et al., 2013). The suture zones are composed of accreted marine ice, sea ice, fallen meteoric blocks and accumulated snow from both drift snow captured in the surface depression and direct snowfall (Jansen et al., 2013; Leonard et al., 2008; McGrath et al., 2014). Numerous rifts in the meteoric ice bands terminate against the suture zones, indicating that the heterogeneous ice within the zones has higher fracture toughness, providing resistance to rift propagation through the ice shelf (Borstad et al., 2017). When the large rift in Larsen C Ice Shelf broke through the suture zone, the speed of propagation increased markedly as the rift traversed meteoric ice. Suture zones are thinner, warmer and more heterogeneous than the meteoric ice and may have differences in water content and crystal fabric, all of which likely vary spatially throughout the zone. Which combination of these factors is most important in determining fracture toughness is still unknown. These examples from various ice shelves show that it is important to understand the internal composition and structure of ice shelves as this may impact on the rate and path of fracture. However, direct observations of the control exerted by internal structure on crack propagation are limited. Here we present for the first time direct evidence that the path of propagation is directly influenced by deep-lying structures within an Antarctic ice shelf.

The data were collected between December 2015 and February 2016 using a commercial ground-penetrating radar system (Sensors and Software pulseEKKO PRO) operating at a centre frequency of 50 MHz. The system was mounted on a sledge, which was towed behind a snow mobile travelling at 15 km h-1, with traces recorded continuously. Positioning information was recorded using a dual-frequency GPS receiver, and the satellite range information was processed through the Canadian Geodetic Service Precise Point Positioning service in kinematic mode. This methodology provided radar profiles with positions accurate to around 0.5 m. The radar data were processed using ReflexW software by applying a bandpass filter (with corner frequencies 10/20/60/120 MHz), a spherical spreading correction, a horizontal filter to remove the direct arrival, and time migration. A fixed wave speed of 0.168 m ns-1 was used for both migration and depth conversion because the construction of a detailed wave speed model for the ice shelf using multiple common mid-point determinations was beyond the scope of the survey. We therefore elected to use the fresh water ice wave speed of 0.168 m ns-1 throughout. We estimate that our figures for the thickness of the ice shelf where most of that thickness is made up of meteoric ice have an uncertainty of about 20 %. Where there is significant thickness of pure firn (i.e. not infiltrated with seawater), the use of a uniform wave speed underestimates the overall ice shelf thickness by approximately 5 %.

We have established that the “railway track” bands of higher elevation originated at the grounding line in locations where there are troughs in the bed topography. Conversely the bands of lower topography originated at ridges in the bed topography beneath Coats Land. Figure 8 summarizes the situation in cross section. Ice flowed from the troughs in bedrock at a higher rate than off the ridges, providing a steady stream of large, thick blocks of meteoric ice that formed a closely packed flow band within the ice shelf. Ice that flowed off the ridges in the bed topography was thinner and supplied to the ice shelf at a slower rate. As a result, the meteoric ice blocks were both thinner and more spaced out, resulting in isolated icebergs surrounded by large areas of sea ice (Fig. 2, topographic profile B–B′; Figs. 3c, 6b).

The majority of Antarctic ice shelves are formed by the coalescence of glaciers or ice streams that flow across the grounding line in structurally coherent bodies, perhaps with some surface or bottom crevassing, but otherwise intact. The Brunt Ice Shelf is one of a class of ice shelves (other examples are Thwaites Glacier Tongue, the western sector of Cook Ice Shelf and the ice shelf lying off the Leopold and Astrid Coast) comprising ice which retains no structural integrity when flowing across the grounding line; as a result, the blocks of meteoric ice are cemented together by sea ice and drift snow to form the ice shelf.

The process which cements together the ice shelf involves the gaps between meteoric ice blocks being filled first by sea ice and then by drift snow which consolidates into firn (Fig. 9a and b). The isostatic loading by snow accumulation pushes the sea ice downwards until it lies below sea level. At this point, the process of firn consolidation has not advanced to pore close-off and the firn is still porous and permeable, allowing seawater to soak the firn. Freezing cycles then concentrate the salt to leave a brine horizon (Fig. 9c). In Fig. 5 the topography on the high-amplitude reflector provides evidence that the brine infiltration may have occurred by sinking of the firn rather than by horizontal spreading of salty water. Warm periods in the summer can produce melt horizons within the firn, creating an impermeable layer. Our hypothesis is that the soaking of the firn from below as it sank to ocean level may have been blocked by the impermeable layer in the region between 50 and 150 m along line 05 (Fig. 5). This would explain why the brine infiltration reflector is a syncline conformable with the firn layering in this section of the profile. The same process may have occurred between 670 and 800 m along the profile, although it is also possible that enhanced accumulation in this region created or deepened the syncline there after the brine horizon had frozen in place. Another possible explanation of the syncline in the brine reflector between 670 and 800 m is horizontal shortening induced by variations in the flow speed of the ice shelf.

It is not clear what controls the termination of the process of brine infiltration and locks the high-amplitude reflector in place within the ice shelf so that it descends below sea level as further accumulation takes place. It is likely that a balance between pore space reduction by compaction of the firn and freezing of seawater in the gaps between ice crystals eventually reduces the permeability of the firn to nothing at a depth shallower that the dry firn pore close-off would be. Another factor may be the formation of marine ice below these sections of developing ice shelf in a similar fashion to the formation of marine ice in rifts described by Khazendar and Jenkins (2003). While marine ice is porous and permeable when first formed, it thickens and consolidates over time and could eventually close off access for seawater to the lower sections of firn.

The lower-elevation bands between the “railway tracks” in the Brunt Ice Shelf have many of the characteristics of ice mélange; that is, they combine sea ice, marine ice, firn and scattered meteoric ice blocks. Ice mélange has been identified as a prominent feature of the Brunt Ice Shelf–Stancomb-Wills Ice Tongue system (Khazendar et al., 2009), where large areas were identified on either side of the Stancomb-Wills Ice Tongue cementing together large tabular icebergs. Thought of in this way, the Brunt Ice Shelf is a series of ice tongues cemented together by ice mélange to create a single mass with highly heterogeneous properties.

This heterogeneity of structure has a number of potential implications. For example, the icebergs of meteoric ice may have a different bulk density compared to the adjacent mixture of firn, sea ice and marine ice in the mélange. If the bulk density were significantly different, it would affect the thickness-from-freeboard calculations that are carried out over ice shelves where there are no ground-truth thickness measurements (Griggs and Bamber, 2011). Khazendar et al. (2009) estimated the temperature of ice mélange at between -11 and -7 ∘C and that of meteoric ice at between -21 and -15 ∘C. This has implications for melt rate calculations, and the different types will produce meltwater of different salinity.

The propagation of a large rift through the Larsen C Ice Shelf was shown by Borstad et al. (2017) to be faster through meteoric ice than through suture zones, implying greater fracture toughness in the suture zones. It is not known whether the mechanical properties of the ice in suture zones constrained between large homogeneous bodies of meteoric ice such as those in the Larsen C Ice Shelf are similar to the mechanical properties of the ice mélange in the Brunt Ice Shelf. There are strong similarities in the way the material forms but post-formation history may be important in the development of the mechanical properties. For example suture zones can experience high shear strain (Jansen et al., 2013), whereas the ice mélange areas of the Brunt Ice Shelf probably do not.

The heterogeneous structure of the ice shelf influences the rate and direction of the propagation of fractures. Over about half the area of the ice shelf, the location of the meteoric icebergs can be identified from the surface topography (Fig. 2). Where firn accumulation has buried the icebergs completely in the outer part of the ice shelf, the location of the meteoric ice can be mapped using SAR satellite imagery. The top panel in Fig. 10 shows a radargram of the upper 90 m of ice shelf in the central “railway track”, following the red line in Fig. 11 from its most western point towards the grounding line. The black line in Fig. 10 marks the interface between radar facies 1 (firn) and facies 2 and 3 (brine and meteoric ice). The bottom panel shows the backscatter amplitude along the same section from a Sentinel-1A radar image acquired on 29 October 2017. A Gaussian filter with a radius of 40 m was applied to suppress small-scale noise. There is a very high correlation (r=0.607, p=0) between peaks in the backscatter amplitude and the zones of thin firn overlying the crests of the icebergs. This correlation is not a coincidence, as the C-band sensor on Sentinel-1A is known to penetrate the surface of the ice (Bingham and Drinkwater, 2000), and it is therefore capable of picking up the spatial variability in the structure of the ice near the surface. However, the details of this mechanism remain subject to future study, and here we merely note its existence. Using radar backscatter from Sentinel-1A as a proxy, the spatial distribution of meteoric ice can be mapped across the entire ice shelf, as illustrated in Fig. 11.

The complete map of ice shelf heterogeneity provides a unique opportunity to interpret observed changes in the trajectory and propagation speed of two major rifts (Chasm 1 and Halloween Crack) in the ice shelf, and to relate these changes to the internal structure of the ice shelf. To highlight the different ways in which the rifts interact with the internal structure, we focus on two small regions outlined by the black boxes (A and B) in Fig. 11. Region A (Fig. 11) covers the tip of Chasm 1 in November 2017; a detailed view is shown in Fig. 12. The overall direction of propagation of Chasm 1 is dictated by the large-scale stress field in the ice shelf, with a trajectory that is perpendicular to the direction of maximum tensile stress (De Rydt et al., 2017; Gudmundsson et al., 2017). However, Fig. 12 shows that at smaller length scales the exact trajectory is dictated by the location and shape of the meteoric icebergs, and Chasm 1 follows pre-existing lines of weakness along the edges of the icebergs, in particular within the “railway track”. Only at one instance did Chasm 1 propagate through an area of meteoric ice (red circle in Fig. 12), which coincided with a period of significantly slower propagation rates (De Rydt et al., 2017).

Region B (Fig. 11) covers part of the Halloween Crack where the rift propagated through the northern “railroad track”; a detailed view is shown in Fig. 13. The overall direction of propagation is again dictated by the stress distribution in the ice shelf, which forces the Halloween Crack to cross an area with a high concentration of meteoric ice, where the edges of the iceberg are misaligned with the overall direction of propagation (red ellipse in Fig. 13). This has resulted in a more complex propagation behaviour, where the rift followed edges of the icebergs for short periods and broke through the icebergs in places of weakness, such as discontinuities in the “railway sleepers”. For the remainder of the trajectory shown in Fig. 13, the rift, while generally following a path dictated by the stress field, in detail weaves its way around scattered icebergs within the ice shelf, avoiding fracturing through areas of meteoric ice. Thus, fracture through the Brunt Ice Shelf progresses at contrasting rates to rifts through other documented ice shelves. Elsewhere rift extension rates increase in meteoric ice and decrease in suture zones comprising ice mélange, whereas recent Brunt rifts slow in meteoric ice and speed through ice mélange. This suggests that there is a difference in the physical properties of ice that was formed in similar ways but which had different subsequent histories, e.g. amount of shear strain, during advection though the ice shelf.

Unusually the Brunt Ice Shelf is composed of alternating bands of ice of two types which have different origins and compositions. The first type is identifiable on visual satellite imagery by a “railway track” appearance and comprises large blocks of meteoric ice originating in the ice sheet in Coats Land which are cemented together by thin strips of sea ice and firn. This type is mostly thick freshwater ice. The second type has a random appearance on satellite images and comprises an ice mélange of relatively thin scattered blocks of meteoric ice from the continent separated by large areas of sea ice (probably underlain by marine ice) and firn that has been soaked by seawater and refrozen.

While a fracture propagating through the Larsen C Ice Shelf sped up when traversing through meteoric ice and slowed in suture zones, fractures observed in the Brunt Ice Shelf slowed when going through meteoric ice blocks and were often routed through the ice mélange in the gaps between the meteoric ice. This contrast in styles of fracture propagation in different ice shelves needs to be better understood so as to improve predictive modelling of ice shelf stability.