Borehole‐Based Characterization of Deep Mixed‐Mode Crevasses at a Greenlandic Outlet Glacier

Optical televiewer borehole logging within a crevassed region of fast‐moving Store Glacier, Greenland, revealed the presence of 35 high‐angle planes that cut across the background primary stratification. These planes were composed of a bubble‐free layer of refrozen ice, most of which hosted thin laminae of bubble‐rich “last frozen” ice, consistent with the planes being the traces of former open crevasses. Several such last‐frozen laminae were observed in four traces, suggesting multiple episodes of crevasse reactivation. The frequency of crevasse traces generally decreased with depth, with the deepest detectable trace being 265 m below the surface. This is consistent with the extent of the warmer‐than‐modeled englacial ice layer in the area, which extends from the surface to a depth of ∼400 m. Crevasse trace orientation was strongly clustered around a dip of 63° and a strike that was offset by 71° from orthogonal to the local direction of principal extending strain. The traces’ antecedent crevasses were therefore interpreted to have originated upglacier, probably ∼8 km distant involving mixed‐mode (I and III) formation. We conclude that deep crevassing is pervasive across Store Glacier, and therefore also at all dynamically similar outlet glaciers. Once healed, their traces represent planes of weakness subject to reactivation during subsequent advection through the glacier. Given their depth, it is highly likely that such traces—particularly those formed downglacier—survive surface ablation to reach the glacier terminus, where they may represent foci for fracture and iceberg calving.

. (a) Basemap image (Sentinel-2 taken on July 7, 2018) of Store Glacier, Greenland, and (b) expansion of (a) with layers showing: surface velocity presented as colored flowlines; principal surface strain rates (extending in red, compressing in blue); crevasse trace azimuth from borehole OPTV (black dashed line); and crevasse rotation (A to A′ and B to B′) from the surface velocity field. The red diamond marks the borehole location. Surface velocity flowlines and strain rates are derived from the 2017 MEaSUREs velocity data set (Joughin et al., 2010).
In the absence of lake-drainage driving hydraulic fracture, at least four general lines of evidence suggest the presence of deep, englacially terminating crevasses. First, such crevasses have been considered responsible for englacial warming as a result of energy released by meltwater refreezing within them (e.g., Colgan et al., 2011;Phillips et al., 2010;Poinar et al., 2017). For example, Lűthi et al. (2015) favored refreezing of meltwater in 300-400-m deep crevasses to explain englacial ice temperatures measured in boreholes near the western margin of the GrIS. These ice temperatures were recorded to be 10°C-15°C warmer than modeled by heat diffusion alone. More recently, Seguinot et al. (2020) invoked the same process to explain measured englacial warming at a rate of 0.39°C/a in englacial ice advancing toward the terminus of tidewater Bowdoin Glacier, Greenland. Here, it was considered that the crevasses, which possibly initiated in association with band ogives several kilometers upglacier, may have warmed the glacier's full thickness of ∼250-300 m. Such inferences are not limited to the margins of the GrIS. For example, Gilbert et al. (2020) suggested that meltwater refreezing in crevasses may be responsible for deep warming, inferred from surface ground-penetrating radar, at Rikha Samba Glacier, Nepal, a process that may also contribute to anomalously warm englacial ice measured in boreholes across the debris-covered tongue of Khumbu Glacier, Nepal (Miles et al., 2018). Second, deep englacially terminating crevasses have also been invoked to explain englacial radar scattering. For example, Catania et al. (2008) interpreted single englacial diffractors from 1 MHz surface radar transects in Greenland as reflections from the apex of deep crevasses, in this case from depths of ∼100-200 m (their Figure 2). Third, inferred crevasse traces have been observed in ice near the glacier bed during exploration of ice-marginal subglacial cavities and tunnels (e.g., Hubbard et al., 2009;Lovell et al., 2015), implying initial propagation to depths of greater than some tens of meters. Hambrey (1976) calculated this depth (from cumulative overburden mass loss) to be at least 80 m from observations of surface crevasse traces in the ablation area of Charles Rabots Bre, Norway. However, in these cases there is some uncertainty involved in determining confidently whether such features were originally deep surface crevasses or local basal crevasses. Finally, Fountain et al. (2005) used a borehole camera to investigate the interior of Storglaciären, Sweden, and reported the presence of water-filled englacial fractures to a depth of 131 m.
Strong inferential evidence therefore exists for the occurrence of crevasses extending more than tens of meters below the ice surface. However, gaining direct access to such crevasses is difficult. Here, we use borehole optical televiewing to investigate the presence and properties of subsurface crevasses at Store Glacier, a fast-moving outlet glacier of the GrIS.

Field Site and Methods
Store Glacier (Greenlandic name Sermeq Kujalleq) is a fast-moving outlet glacier draining the central-west sector of the GrIS into Davis Strait via Ikerasak Fjord in the Uummannaq Fiord system (Figure 1).
The borehole investigated herein (BH18c) was drilled in July 2018 and was located within a crevassed area ∼30 km from the terminus of Store Glacier and within 1 km of a supraglacial lake that drains rapidly via hydrofracture during most summers (Chudley et al., 2019). Although the area is generally crevassed, BH18c was drilled ∼5 m away from the nearest open crevasse, both for reasons of operator safety and to minimize the chances of the vertical borehole intersecting the (assumed vertical) crevasse. Locally, ice surface velocity is ∼700 m/a, achieved largely through basal motion focused at a temperate interface  between ice and an underlying deformable subglacial sediment layer . Although drilling HUBBARD ET AL.  high angle sinusoids (planes in 3D space), each marked by a red arrow, cutting across low-angle background layers. With a borehole diameter of ∼12 cm, the width of these OPTV images is the borehole circumference, ∼0.38 m. The reflective (brighter) approximately vertical streaks at ∼160° and ∼340° are drilling artifacts, probably caused by contact between the hose and borehole wall. Images are orientated to magnetic north. 3D, three-dimensional; OPTV, optical televiewer.
(by hot pressurized water) was halted temporarily to allow logging, the borehole was eventually drilled to the bed at a depth of ∼949 m. The uppermost 325 m of this borehole was logged by optical televiewer (OPTV) on July 18, 2018, providing a geometrically accurate RGB image of the complete borehole wall at a resolution of ∼1 mm both vertically and laterally. The OPTV instrument also houses a 3-axis accelerometer and a 3-axis magnetometer, enabling orientation of the image log to magnetic north. Such OPTV images have the capacity to inform on annual accumulation (Philippe et al., 2016), the presence and scale of refrozen layers within firn , marine ice accretion (Hubbard et al., 2012), and visible structural features intersecting the borehole wall (Roberson & Hubbard, 2010). For example, the last of these studies identified eight separate ice structures, including high-angle crevasse traces to a depth of some tens of meters, in the ablation area of Midre Lovénbreen, Svalbard. Since OPTV images extend around the entire (cylindrical) borehole wall, they are typically unrolled for two-dimensional (2D) presentation, extending left to right around the compass (N-E-S-W-N or 0°-90°-180°-270°-0°). Planar layers appear as sinusoids on such 2D images, with the amplitude of the sinusoid representing the layer's dip and the phase of the sinusoid its azimuth (Hubbard et al., 2008). Structural analysis and plotting were carried out using software packages BiFAT , Well-CAD, and Stereonet v.11 (Cardozo & Allmendinger, 2013). In the analysis presented herein, local declination of −30° (west) was corrected for when mapping the orientation of OPTV-imaged features. All feature azimuths are therefore presented (as three digits) clockwise from grid (NSIDC Polar Stereographic North) north.
Borehole temperatures were recorded by sensor strings installed into two boreholes, BH18b and BH18d, both located <10 m from BH18c. Englacial temperature was measured using Fenwal UNI-curve 192 thermistors (see Doyle et al., 2018) and DS18B20 digital temperature sensors (Table S1 and Figure S1). Undisturbed ice temperatures were estimated following established methods Humphrey & Echelmeyer, 1990). A theoretical borehole temperature profile was also created by inverting the 2016 annually observed MEaSUREs surface ice velocity using the higher-order Community Ice Sheet Model 2.0, which conserves momentum, mass and thermal energy in three dimensions. The approach follows that used by Price et al. (2011) for Greenland and Bougamont et al. (2019) for Antarctica. The observed motion was simulated by first prescribing a noslip basal boundary condition, and then subtracting the derived internal ice deformation from the observed surface velocities. We then iterated basal traction and sliding rates to an equilibrium in which ice temperature, effective ice viscosity, and ice velocity fields converged with the target velocities. To constrain the inversions, we used BedMachine v.3 (Morlighem et al., 2017) ice thickness and bed topography resampled to the 1 km fixed grid of our model. Surface air temperature was specified at the same 1 km resolution using output from the RACMO regional climate model (Noël et al., 2018).

OPTV Log
The complete 325 m long OPTV log of BH18c ( Figure 2a) shows two general characteristics.
First, the log shows an overall decrease in luminosity with depth. This is common to many OPTV logs recovered from glacier boreholes , with light scattering decreasing with depth as the material forming the borehole wall progresses from highly reflective snow and firn through bubble-rich ice HUBBARD ET AL.   Figure 2), illustrating high-angle planar layers, apparent as a sinusoid in these 2D representations: (a) 33.8-34.5 m depth, (b) 79.7-80.7 m depth, and (c) 183.8-184.9 m depth. Note the general decrease in luminosity with depth and the central bubble-rich (bright) planar layers bisecting the high-angle layers imaged in (a and b) and the presence of multiple such layers in (c). Images are orientated to magnetic north. 2D, two-dimensional; OPTV, optical televiewer. of intermediate reflectivity to almost transparent bubble-poor ice at depth. Since BH18c is in the ablation area of Store Glacier, it intersects no near-surface snow or firn, and the darkening in this case represents a general decrease in bubble content with depth. Second, numerous alternating decimeter-scale light and dark bands signify alternating bubble-rich "white" ice and bubble-poor transparent ice ( Figure 2b). This banding extends throughout the full depth of the OPTV log and, although appearing uniform and horizontal at the (vertically compressed) scale illustrated in Figure 2a, an expanded section ( Figure 2b) reveals that the banding dips shallowly and is of variable thickness around the borehole.
Closer inspection of the OPTV log of BH18c also reveals the presence of 35 distinct high-angle layers (examples of which are indicated by red arrows in Figure 2b) that cut across the less distinct, lower-angle banding noted above. Enlargements of the log (Figure 3) reveal that these layers form regular sinusoids, indicating uniform and planar layers in 3D space. Each layer is formed of one or more millimeter-scale laminae that are bright (reflective) and hence inferred to be bubble-rich. In 28 cases, these laminae are enveloped by centimeter-to decimeter-thick layers of ice that appear dark and are hence transparent and devoid of reflective bubbles. Of these 28 layers, 24 host a single central lamina (e.g., Figures 3a and 3b), one hosts two such laminae, two host three such laminae, and one (a ∼0.25-m thick layer at a depth of ∼184 m) hosts nine such laminae (Figure 3c  Analysis of the distribution of the 35 high-angle layers intersected by BH18c ( Figure 4a) shows a general decline in frequency with depth, with the deepest observed plane being 265 m below the glacier surface. It is quite possible that the borehole intersected similar planes below this depth but that they were indistinguishable from the background ice. This is because (visibly bright) included bubbles are progressively expelled from all ice as pressure increases with depth, homogenizing the darkening OPTV image.
Geometrical analysis of the orientation (relative to true north) of these planes reveals a strong single maximum dipping approximately WNW to ESE (mean azimuth of pole to plane = 287°) at a mean dip of 63° (Figure 5 and Table 1).
Accordingly, eigen analysis of the poles to these planes (green dots on Figure 5) indicates a strongly clustered fabric with a dominant principal eigenvalue of 0.83 at a vector of 287° (Table 1). Considering the strength of this mode, the second and third eigenvalues (of 0.1 and 0.07, respectively) have little physical meaning.

Englacial Ice Temperature
The borehole sensor strings recorded 32 discrete ice temperatures between the near-surface and the ice-bed interface ( Figure S1). These temperatures decrease from ∼−5°C at a depth of ∼25 m (the uppermost thermistor location in the borehole) to ∼−22°C at 600-700 m depth, to rise again sharply to temperate at the bed at a depth of ∼949 m. Over most of the borehole length, the modeled temperatures correspond closely with the measured temperatures ( Figure S1). However, modeled temperatures diverge from the observed record in the depth ranges 0-∼400 m and 770-850 m. The former of these is relevant to the present study, and temperature data over the depth range 0-500 m are plotted in Figure 4b. Here, measured temperatures are notably higher than modeled temperatures, a difference ("residual temperature") that decreases with depth from ∼10°C near the surface to zero at ∼400 m (Figure 4c).

Interpretation and Discussion
The background (host) ice is formed of alternating subhorizontal, but irregular, bands of bright bubble-rich ice and darker bubble-poor ice at the scale of centimeters to decimeters. This is typical of the remnants of primary stratification in englacial ice, with the bands having originally been laid down as snowfall in the accumulation area of the GrIS. Such primary stratification has been identified elsewhere, both at the glacier surface (see review by Hambrey & Lawson, 2000) and within borehole OPTV logs (Roberson & Hubbard, 2010). Superimposed onto this primary structure, the high-angle planes have the physical properties of healed crevasses or crevasse traces (Hambrey & Müller, 1978). These properties include their secondary status (cutting across the primary stratification), high angle, and planar geometry. The clear nature of the transparent ice forming these layers is consistent with formation by the refreezing of a water-filled crack some cm to tens of cm across, similar in dimension to the fractures intersected by boreholes at Storglaciären, Sweden (Fountain et al., 2005). Such ice is typical of refrozen ice from which gas has been expelled during the freezing process (e.g., Pohjola, 1994), as are their bubble-rich central laminae.
Here, gas is rejected at the advancing ice front, increasing its concentration in the remaining reservoir of unfrozen water, eventually reaching saturation. At that point, bubbles form and are incorporated as a lamina into the last-frozen ice layer (Hubbard, 1991). Planes are plotted as green lines and poles to those planes as green dots, contoured by standard deviation following Kamb (1959). Eigen analysis data are given in Table 1. Note these planes have been corrected for local declination and are orientated to grid north for consistency with Figure 1. Thick arrows give the direction of local surface motion (black) and the azimuth of local principal surface extending strain (red). OPTV, optical televiewer. Abbreviations: 3D, three-dimensional; OPTV, optical televiewer. bubble-rich last-frozen lamina we observe in most of the OPTV image of BH18c (Figures 3a and 3b) is consistent with water freezing inwards at a similar rate from both edges of a crevasse. Similar features, also interpreted as crevasse traces, have been identified cropping out at the glacier surface (e.g., Figure 6).
Where exposed at the glacier surface, such clear ice crevasse traces commonly erode preferentially, providing channels for supraglacial drainage (Hambrey & Müller, 1978) (also see Figure S2). We interpret multiple last-frozen laminae within a crevasse trace, such as that imaged at a depth of 184 m (Figure 3c), in terms of the effects of multiple separate refreezing events. Although multiple central laminae have not, to our knowledge, been reported elsewhere, reactivation of crevasse traces as thrust faults has been proposed (e.g., Goodsell et al., 2005). At Store Glacier, we envisage trace reactivation to involve opening along its bubble-rich last-frozen lamina and splitting that layer into two thinner bubble-rich laminae. The open crevasse then refills with meltwater that eventually refreezes, forming a new (third) bubble-rich last-frozen lamina. After that, any subsequent reactivation can open any of the three pre-existing last-frozen laminae and the process is repeated, each time adding a pair of new laminae and therefore generally resulting in an odd total number of such laminae. This is consistent with our observations, and in particular with the feature at 184-m depth which has nine such laminae (Figure 3c), indicating that the initial crevasse trace was reactivated four times. In contrast, we imaged one high-angle layer with an even number of central laminae (two). In this case, the crevasse could have reactivated along a new plane rather than along the pre-existing central laminae. If this interpretation is correct then, of the 28 filled crevasse traces imaged in the uppermost 325 m of BH18c, 24 showed no clear evidence of reactivation, three showed evidence of a single reactivation phase, and one showed evidence of four reactivation phases. Of these seven inferred reactivation phases, six occurred along a pre-existing bubble-rich last-frozen lamina. The seven bright laminae that were not visually bounded by transparent (refrozen) ice likely formed HUBBARD ET AL.   either as unopened fractures or as dry crevasses, the traces of which consequently did not host a refrozen ice layer.
Geometrical analysis of the crevasse traces intersected in BH18c (Figures 1b and 5) indicates a strongly clustered distribution with a dip of 63° and a strike of 017°, neither of which varies systematically with depth. Reference to Figure 1b shows that this strike differs by 87° from that of local surface crevasses, which trend ∼110° (none of which was intersected by the borehole). Furthermore, the local azimuth of principal extending strain is ∼008°, so the orientation of the OPTV-imaged crevasse traces diverges by ∼81° from orthogonal to this. The crevasse traces intercepted by BH18c are therefore aligned ∼81° away from the azimuth at which local crevasses should form if solely by fracture Mode I "opening" (Figure 7a).
Indeed, even local open surface crevasses deviate from orthogonal to the azimuth of local principal extending strain by 12°. While the orientation of principal stress can rotate a few degrees with depth (Pfeffer et al., 2000), a devaition of 12° is compatible with local surface crevasses opening under local strain conditions, with mixed mode fracture explaining the slight offset. However, the englacial traces imaged by our OPTV log are highly unlikely to have formed locally by any combination of fracture mode. This interpretation is supported by our inference of multiple trace reactivation since each such event requires three phases: (i) crevasse opening, (ii) water filling, and (iii) water freezing. It is unlikely that all three phases could occur multiple times within a single year and-although no independent evidence is currently available to evaluate this-variability in the requisite fracturing and temperatures are greatest, and therefore most likely, at the annual timescale. The traces we image-one with four reactivation phases-are therefore likely to be at least 4 years old, and probably older. With a local ice velocity of ∼700 m/a, that places their origin at least ∼3 km upglacier. Figure 1b shows the presence of two areas of opening crevasses upflow of the borehole location, one ∼4 km distant and the other ∼8 km distant. Since extending strain rates are small between the first of these sites and the borehole (and crevasse reactivation is therefore unlikely along this path) this location is unlikely to serve as the origin of the traces intercepted by the borehole. In contrast, crevasses formed at the more distant site both have farther to travel (allowing more time for reactivation) and pass through the lower site of locally high extending strain rates. Indeed, plotting the trajectory of crevasse azimuth at the upper site to the borehole (illustrated by the paths of A to A′ and B to B′ on Figure 1b) rotates this original azimuth to within 7° of the strike of the englacial traces imaged at the borehole. We therefore consider that the most likely location for the formation of the crevasses whose traces we image by OPTV in BH18c was ∼8 km upglacier. It is possible that these crevasses formed yet farther upglacier-at least as far as the tensile strain-rate remains sufficient to induce brittle fracture and water is available to fill those fractures (i.e., probably the top of the wet snow zone). However, without further evidence, and given the orientation match noted above, we have no reason to argue that formation was >∼8 km upflow. Figure 1b reveals that neither the crevasse field at the borehole location nor that located ∼8 km upglacier is aligned orthogonal to the local azimuth of principal extending strain, suggesting mixed-mode crevasse formation at both sites (Colgan et al., 2016). The occurrence of Mode II "sliding" (Figure 7b) and/or mode III "tearing" (Figure 7c) is also consistent with 7 of the 35 imaged traces having no discernible refrozen ice component; in such cases the fracture either opened and closed without water ingress and freezing (considered possible but unlikely) or formed largely by Mode II and/or Mode III fracture. Certainly, Mode I "opening" (Figure 7a) predominates, as evidenced by (open) surface crevasses on the glacier and by the presence of refrozen ice in 28 of the 35 traces we image. However, none of these crevasse traces appears to displace the pre-existing background stratification vertically (Figure 3). While it is possible that vertical displacement of lower-than-detectable magnitude did occur, we estimate-based on the nature of the OPTV images and the irregularity of the stratification-that any such displacement would have been less than a few millimeters in all cases. We therefore consider initial crevasse formation involving a significant Mode II component as unlikely and instead favor Mode I combined with Mode III fracture (Figure 7d) at, and upglacier of, our study site.
We also note that the traces we image dip at a mean angle of 63° and show no systematic change in dip with depth. While the dip of open crevasses is difficult to measure directly for more than a few meters below the surface, shallowly dipping crevasse traces have been reported from valley glaciers. For example, Roberson and Hubbard (2010) identified transverse fractures (their "S2") in OPTV borehole logs from midre Lovénbreen, Svalbard, which dipped as shallowly as 60°. Similarly, Hambrey and Müller (1978) mapped several sets of crevasse traces exposed at the surface of White Glacier, Canada. This mapping showed that crevasse traces were typically near vertical ∼3 km upglacier from the terminus, but that they dipped progressively during advection downglacier until they became almost horizontal at the terminus. It is not, therefore, uncommon for crevasse traces to dip as they advect through an ice mass. That the primary stratification has remained close to horizontal in our OPTV images at Store Glacier (Figures 2 and 3) suggests that the crevasse traces have dipped without rotating the pre-existing stratification; thus, the crevasses either formed initially at a mean dip of 63°, or formed vertically and subsequently shallowed under simple shear. If the latter, then the mean trace dip of 63° at the borehole site also supports our interpretation that the antecedent crevasses formed several kilometers upglacier. Initially vertical crevasse formation, followed by strain-induced shallowing during passage along the glacier, is also consistent with BH18c not intersecting any local crevasses, despite them being only some meters distant at the glacier surface.
It takes ∼40 years for ice to move from the borehole location to the front of Store Glacier, during which time ∼100 m of surface ice would be lost to ablation (assuming a surface ablation rate of 2.5 m/a). Thus, crevasse traces deeper than this at the borehole site, as well as those formed deep enough farther downglacier, would survive to the glacier's terminus, some to depths of hundreds of meters. Given several of the crevasse traces we report herein show evidence of reactivation-implying they are weaker than adjacent host ice-and the continued formation of similar crevasses downglacier, nearer to the terminus, it is possible that crevasse traces reaching the terminus present subvertical planes of weakness there, offering locations of preferential fracture. If this is the case, then models of fracture mechanics may need to account for the presence of such deep and geometrically recurring planes of weakness.
The frequency of intersected crevasse traces shows a general decrease with depth (Figure 4a), consistent with their antecedent crevasses forming at the surface and terminating at different depths. Assuming this depth distribution represents that of the original crevasses, then only 7 (or 20%) of the 35 intersected by the borehole were shallower than 40-m depth and 12 (34%) were deeper than 100 m. Although specific to our study site, the relationship between the crevasses present (%) per 30-m depth range (C(30) % ) and depth (D, m), as illustrated in Figure 4a, can be approximated (R = 0.69) as a logarithmic function: This fit both suggests the presence of crevasses below the lowermost crevasse trace imaged by our OPTV log (at a depth of 265 m) and is consistent with the difference between measured and modeled englacial temperatures at our field site (Figure 4c), suggesting that crevasses warmed englacial ice to a depth of at least ∼400 m. Fitting a logarithmic curve to this residual temperature (T e , °C) against depth, as illustrated in Figure 4c, yields a close match (R = 0.95) described by: The similarity of the relationships between crevasse frequency and depth (Figure 4a; Equation 1) and residual temperature and depth (Figure 4c; Equation 2) lends support to our interpretation that crevasses propagate to at least 400 m below the surface of Store Glacier, warming ice to that depth by the presence and refreezing of crevasse-filling meltwater.

Summary and Conclusions
Our OPTV log of a borehole drilled in a crevassed area of fast-moving Store Glacier, Greenland, has successfully imaged deep crevasses in the GrIS for the first time. Combining analysis of this log with thermomechanical modeling has revealed the following: • Thirty-five traces of surface crevasses were imaged directly to a maximum depth of 265 m. It is possible that the borehole intersected crevasses below this, but that they could not be distinguished from the host ice. Although trace separation increased with depth, approximately one third of all traces were intersected deeper than 100 m. • The borehole intersected traces of crevasses that were highly unlikely to have formed locally. Comparison of trace orientation with the surface strain rate field of Store Glacier indicates their antecedent crevasses most likely formed ∼8 km upglacier.
• Despite drilling in an active crevasse field, and between two open crevasses spaced ∼10 m apart, the borehole intersected no local crevasses. This indicates that the active local crevasses were shallow and/ or did not deviate sufficiently from vertical to intersect the uppermost 325 m of the borehole. • Crevasse fields analyzed are not aligned perpendicular to the orientation of principal extending strain rate, indicating mixed-mode formation, likely Modes I and III since the traces did not appear to displace primary stratification vertically. • Of the 35 traces imaged, 28 showed evidence of having been filled with meltwater that subsequently refroze. This meltwater and its refreezing released heat, assumed to be responsible for warming the ice to a depth of ∼400 m, suggesting crevasses extended to this depth-although not evident below 265 m in our OPTV log. • Refreezing of crevasse-fill water creates a distinctive ice layer formed of a planar bubble-free ice layer some centimeters to decimeters thick that envelopes a planar mm-thick central layer of bubble-rich last-frozen ice. • We hypothesize that, once formed, crevasse traces represent a plane of weakness that may be reactivated during advection through the glacier. Visual analysis of the traces imaged by borehole OPTV suggests that reopening occurs preferentially along the bubble-rich last-frozen layer(s), which then refill and refreeze, creating additional new last-frozen layers. The time required for crevasse opening, filling and refreezing is not known but is likely to be some years, consistent with the ∼8 km distance separating the borehole from the proposed location of initial crevasse formation. • The crevasse traces imaged by our OPTV log, supplemented by new ones formed downglacier, extend deep enough to survive ablation and reach the glacier terminus. Here, it is possible that these relatively weak traces precondition the precise location of ice fracture, a process that may need to be addressed by fracture models.