The topography of Dome C (Fig. ) was first defined from the joint SPRI/NSF/TUD airborne surveys of the 1970s . These pioneering airborne radar altimetry and radar sounding observations predated GPS, and aircraft positions were constrained by pressure altimetry and inertial navigation systems with large uncertainties; however, subsequent ground-based traverses and satellite radar altimetry confirmed the presence of the dome. As a site of thick ice, low accumulation, and slow ice flow, it was a promising site for ice coring, with the first cores in the region acquired in 1977–1978 . These early surveys also revealed the presence of an extensive population of subglacial lakes in this region .

Site selection work for the EPICA Dome C ice core took place in the mid 1990s with an Italian survey grid covering the Dome C region . This work, a combination of ground and airborne (Twin Otter) based surveys using a 60 MHz incoherent radar system with a 1 µs pulse width, covered most of the Dome C region with 10 km line spacing. Ice thickness measurements from these surveys form the bulk of the data coverage for this region in the Bedmap2 compilation (; see Fig. a and b).

The coarse subglacial geography revealed by the Italian survey is comprised of a deep subglacial trough (the Concordia Subglacial Trench) to the northeast of Dome C (see lower left of Fig. a), with indications of a flat subglacial plateau near mean sea level under the center of the primary dome, and a massif to the southwest along the line of northeastward ice flow. EPICA Dome C targeted the center of the dome on the basis of apparently flat topography and its isolation from surrounding ice flow . Additional analysis , however, revealed broad, shallow channels trending north–south within the subglacial plateau region.

The final EPICA Dome C ice core succeeded in obtaining ice dated as old as 800 kyr; however, the lower 75 m of the ice column was either undatable or not drilled to prevent contamination of a wet bed , and extrapolation of the borehole temperatures indicated that melting was likely occurring at the bed . Analysis of the composition and structure of the lower portion of the ice core showed that focusing of ice flow by the broad channels on this plateau may have resulted in stretching and recrystallization of the lower part of the ice column, implying that an ideal old ice target may require a very flat ice–bed interface around a flowline tracing back toward the ice divide, characterized by a horizontal size of several ice thicknesses .

In 2008, 2009, and 2011, the ICECAP project conducted survey flights using the HiCARS family of radar sounders , mounted on a DC-3T Basler. These radar systems provided coherent, focusable 60 MHz data with a 0.08 µs pulse width. The goal of these flights was improving the radar stratigraphy between the EPICA Dome C and Vostok ice core sites . Included in these ICECAP flight lines was a transect along the ice divide from Dome C to subglacial Lake Vostok, which was also flown by a range of other radar sounders, as well as a number of sparse lines of the Vostok–Concordia–Dumont d'Urville corridor (VCD), typically 20 to 40 km apart, parallel to the ice divide.

developed an ensemble model for predicting regions of frozen bed using a combination of remote sensing and teleseismic estimates for geothermal heat flow combined with a thermomechanical ice sheet model calibrated by observations of subglacial lakes. When thresholds for ice thickness (>2000 m) and of the horizontal component of the ice velocity (<2 m yr-1) were applied, a map of possible old ice candidates was produced (Fig. ).

In the Dome C region, five candidate sites exist, which we term A, B, C, D, and E (Fig. ). Notably, none of these sites overlaps with the EPICA Dome C ice core site near Concordia Station – consistent with the likely basal melting there implied by extrapolation of borehole temperatures. Sites B, C, and D are located on the steep and poorly sampled subglacial peaks on the northeastern side of the Concordia Subglacial Trench; basal ice in this region has likely traversed the deep, wet Concordia Subglacial Trench. Site E lies on a small subglacial high downstream on the Totten Glacier side of the dome; this site also lies down flow of a deep subglacial trough, thus raising substantial doubt to its suitability as an old ice coring site.

Candidate A is by far the largest site in the Dome C area and lies under the ice divide on a subglacial massif, minimizing both ice thickness and ice velocity. The ice surface above Candidate A forms a topographic extension to the south of Dome C informally termed “Little Dome C”. The central part of Candidate A lies 40 km southwest of Concordia Station. The size of Candidate A compared to the 5 km model cell size also makes it more likely that the model captured basal temperatures correctly. Because of its characteristics, Candidate A represents a near-term primary goal of European and Australian old ice site selection.

The 2011 airborne survey line (VCD/JKB2g/DVD01a; location shown in green in Fig. ) crossed the middle of the Candidate A site. Focusing of the radar data showed that the southern flank of the Candidate A massif ended in a steep cliff over which englacial layers dive (Fig. ). Coherent, continuous englacial reflectors are present in the upper 80 % of the ice column , while in the bottom 500 m, a region of more diffuse englacial scattering is present. This distinct zone of basal ice is also apparent in Operation IceBridge (OIB) radar data that operates at a higher frequency , and in appearance is similar to the “valley wall” accretion ice seen in Dome A .

Uncertainties in the older data sets, a lack of resolution appropriate for the small-scale processes near the base of the ice sheet, and inadequate knowledge of subglacial hydrology and geothermal heat flow drive a need for greatly increasing resolution over these old ice targets.

Key objectives of the survey were to define the ice thickness at high resolution, infer basal roughness across the target region, and map the distribution of subglacial water. Improving the englacial stratigraphy especially deep layers; and correlating it to the existing EPICA Dome C core site were also high priorities. In addition to the radar data, we acquired laser altimetry, gravity, and magnetics data, along with complementary GPS and inertial measurement unit (IMU) data. Instruments are detailed in Table .

The results of this survey have been used for follow-up high-resolution ground work, using the BAS DOLORES ground radar for further bedrock mapping, ApRES phase-tracking radar to track vertical strain rates, and the BAS Rapid Access Isotope Drill UK-RAID; to acquire thermal gradients in the upper ice sheet, for geothermal heat flow inversions. A location just northeast of the ice divide in central Candidate A (122∘12′ E, 75∘18′ S) was selected for further investigation. If the ground investigation proves fruitful, the SUBGLACIOR drilling probe will be deployed to measure the in situ oxygen isotope record, as a pathfinder for a full eventual ice core retrieval.

The OIA survey was designed to sample Candidate A at high resolution, with 110 km long longitudinal-to-slope “Y” survey lines at separations of down to 1 km cutting across the ice divide, and ∼ 65 km long transverse-to-slope “X” tie lines with separations of 5 km parallel to the ice divide (Fig. d). Some of the X lines extend true northeast to cross the Concordia Subglacial Trench and candidates B, C, and D, while the Y lines extend far enough to the true northwest to cover Candidate E.

Two lines were added to cut obliquely across the grid: one that tracked over the EPICA Dome C site in order to connect the ice chronology to the grid and a second line to better constrain an oblique topographic ridge crossing the divide. Flight lines were designed to avoid Concordia's clean air sector to the south of the station, as well as to allow the aircraft to make very high frequency (VHF) communications with the station before landing. Typical aircraft speeds were 85 m s-1, and flights at altitude were typically 4 h in duration.

Four flights of 4 h each were carried out from Concordia Station in late January 2016 – the first two (ICP7/F11 and ICP7/F12) focused on 2 km line spacing Y lines over Candidate A, followed by one flight (ICP7/F13) targeting X lines extending past Concordia to Candidate B, and lastly one flight (ICP7/F14) focused on increasing the line density over the primary target to 1 km line spacing. Initial interpretation of the radar data was performed during the field program, and helped to refine the later flight plans. GPS base station data were collected during the survey flights.

After the field season, GPS data were processed using Waypoint Inertial Explorer, using precise point positioning (PPP) loosely coupled to the acceleration and roll rate data from the SPAN IMU system. Internal estimates of uncertainty for these data have a 2 cm vertical standard deviation and a 4 cm horizontal standard deviation. Apparent surface elevation differences between survey lines at crossovers were minimized to obtain laser altimeter pointing biases using the methods described in.

Radar data were stacked in acquisition 32 times and were coherently recorded at 196 Hz; these data were range-compressed resulting in a range resolution in ice of 8.4 m . The radar data were first processed using a very short synthetic aperture to extract the surface return and for initial quality control. This processing (called “pik1”) retains the unmigrated along-track hyperbolae that characterize many earlier radar sounding data sets. The data were then processed using the 1-D focused synthetic aperture radar (SAR) approach of , where focusing of the along-track Doppler phase variations within each range resolution cell was employed to improve the along-track resolution to approximately 10–20 m for scattering targets. The data were resampled to 4 Hz along-track sampling (∼ 22 m along-track sampling), and the logarithm of signal power was displayed for manual interpretation.

To obtain ice thicknesses, we systematically select a window around the earliest bed return, and then automatically select the best fitting pulse waveform within that window (assumed to be a paraboloid power profile in decibels), for both the surface and the bed. The surface time delay is subtracted from the bed time delay to obtain the two way travel time in the ice column and, using an appropriate refractive index for ice (3.15), we convert to ice thickness. We choose not to apply a firn correction to ice thicknesses; as shown in , a firn correction is not required for our focusing, and will not affect the conclusions in this paper firn correction is however critical for isochrone interpretation. Bed elevations are derived by subtracting the ice thickness from concurrently collected laser or radar altimetry; all elevations are referenced to the WGS-84 ellipsoid.

We do not attempt to reconcile ice thickness interpretations at crossover points, and maintain a strict first return policy. The first return represents a stable interface to interpret in radar, but has a high likelihood of selecting off nadir echoes in steep topography. As detailed in Appendix , preserving crossover differences provides important information on understanding the interactions between radar geometry, processing, and bedrock roughness, and allows us to extrapolate these statistics to intervals without crossover constraints.

Specularity content of the basal return was extracted by comparing the echo strengths of the bed from 1-D focused SAR to the results of range-migrated 2-D focused SAR, following the approach outlined in . Regions with a specularity content of greater than 0.2 were classified as subglacial lakes.

Hydrostatic pressure is important for the context for subglacial lakes, and is often represented by hydraulic head (equivalent to the height of a water column with the same basal pressure as the ice load). Gradients in hydraulic head control basal water flow direction; the magnitude of the slope of hydraulic head controls the expression of water flow. In this case, hydrostatic equilibrium, indicated by zero hydraulic head gradient, was not used for subglacial lake identification (as was the case for the “lake detector” in ); however all subglacial lakes that were identified had low hydrostatic gradients (Fig. ).

We combined the OIA results processed as described above with older data sets from Italy , Germany's Alfred Wegener Institute (AWI; ), Operation IceBridge (OIB) data , and British Antarctic Survey (BAS) data from Jordan et al. (2010; of these data sets, much of the Italian data and AWI data had been included in Bedmap2). We excluded the poorly positioned SPRI/NSF/TUD data.

We compared these surveys to a 1 km resolution grid derived from OIA focused bed elevation data, and found good visual matches between the coherent, focused BAS and OIB data (see Table ). The negative bias in OIB data is likely due to the resolution of small-scale valleys in the OIB profile data that are not resolved in the 1 km grid we used for this comparison.

Both the Italian and AWI data sets were acquired incoherently without focusing or migration, which will induce range hyperbolae in the radargram that will tend to reduce the measured ice thickness. This does not appear to have been a major effect on the biases, however. For the Italian data, the lower mean value is due to the coarse resolution of the pulse, combined with noise in the picker used for the survey; for incorporation into the compilation, we select the shallowest return in each 1 km block of data. For AWI, ice thickness measurements at peaks are systematically ∼ 50 m smaller than for OIA. This corresponds to the length of the high energy pulse used for deep ice sounding. We add 50 m to the AWI ice thicknesses for incorporation into this grid. In all cases, the standard deviation of the trackline data compared to the OIA-only grid was better than the comparison to Bedmap2 (see Table ), likely related to the loss of spatial resolution in Bedmap2.

A compiled grid using all of these data were generated by first extracting the median value for the data at 500 m cells, and used a natural neighbor interpolator (nnbathy; ) on the data. We apply a 2 km Gaussian filter, and mask out data more than 5 km from a data point.

While the outlines of the terrain at the 10 km length scale were visible in Bedmap2 (Fig. a), which was largely derived from the survey (Fig. b), the addition of the OIA data set delineates the key features of this landscape (Fig. b, c). The Concordia Subglacial Trench is bound by a sharp, west-facing dissected escarpment approximately 2000 m relief that hosts Candidates B, C, and D. In this new compilation, with the pre-GPS data removed, this escarpment is no longer discordant with the surface data (left yellow box in Fig. ). Instead we see an array of hanging valleys, consistent with a glaciated terrain. The minimum ice thickness of 2383 m in this region occurs on a sharp peak between two of these hanging valleys.

The massif underlying Candidate A is bound by a southwest facing 200–300 m high system of scarps to the southwest, which capture a system of perched lakes. The massif dips gently to the northeast, and is marked by a series of 200 m deep, 2–3 km wide valleys running toward the north, divided by occasionally large ridges. Under the divide, there is a local 700 m elevation peak where ice thickness is under 3000 m. The minimum ice thickness within Candidate A (2600 m) occurs on a local peak adjacent to the southern escarpment.

To the southeast, a complex series of troughs with extensive water bodies emerges from the Candidate A massif and opens out into the Concordia Subglacial Trench.

Using specularity content, we map out 54 subglacial lakes in the OIA survey, 50 of which were not included in the compilation. Details on these lakes are provided in the Supplement. The largest of these lakes is 11.5 km long and lies in a hanging valley on the northeast side of the Concordia Subglacial Trench (located at 1310 km, -915 km). A second large lake, at least 4.5 km long, lies within the Candidate A region, in the escarpment to the southwest of the massif that underlies Little Dome C (located at 1367 km, -852 km); 50 % of segments of specular bed that were 1 km or greater in length had hydraulic head gradients less than 0.1 %, meeting the criteria for a lake in , and 71 % were less than 0.2 %. This result is consistent with flexural support of small gradients around the edges of these small lakes . In all, 19 lakes are now observed in the region predicted to be at least as cold as -5 ∘C, only one of which was known to .

Small-scale roughness, at length scales of the line spacing and below, is relevant for four reasons: (1) roughness gives insight into the pathways that basal ice must traverse; (2) roughness may provide information on past ice sheet behavior and basal conditions, (3) roughness is a key control on the uncertainties inherent in profiling radar systems, and (4) basal roughness forces the overriding ice sheet to develop a complex deformation pattern in the lower part of the ice column, and this deformation field could disturb stratigraphic continuity of the ice core record.

We calculate the along-track roughness as the deviation in detrended elevation between points of a given length scale . For a given cell size, we calculate the root mean square deviation (RMSD). We choose 800 m length scale for this investigation, as it both provides insight into sub-line spacing roughness, but is also relevant for the cross-track beam pattern for understanding uncertainties in the ice thickness data (see Appendix ).

Typical RMSDs at 800 m length scale are 40 to 50 m toward the center of Candidate A, and are lower toward the margins (Fig. c), although locally smoother regions 3–5 km across exist in places. One of these locations is the EPICA Dome C ice core site. Much of the base of the Concordia Subglacial Trench, and the regions surrounding the massif, are very smooth, consistent with deformable sediments.

On the massif, smoother regions also tend to correlate with regions of subglacial lakes, although subglacial lakes are also found in extremely rough regions with deep incisions (for example location 1320 km, -860 km in Fig. ). Much of the regions of higher roughness in central Candidate A are sinuous, and appear to follow local valleys, with smoother regions between.

We used the OIA laser altimetry to validate available satellite-based DEMs, using the WGS84 datum. We compared this with both the Bedmap2 surface DEM , which in the interior is largely based on the combined ICESat/ERS radar altimetry product of , and the Cryosat-2 DEM , wholly derived from radar altimetry. We transformed all DEMs into WGS-84, and used the Generic Mapping Tools (GMT) grdtrack to extract points of comparison for each laser point (we removed one anomalous point over Concordia Station itself). Results are shown in Fig. .

We find that both DEMs have a significant bias, outside the previously demonstrated accuracy of the ICECAP laser system, with the laser altimetry at Dome C (consistent with some penetration by the radar altimeters); however both have very low levels of noise in this very flat region. There is a slight preference for the Cryosat-2 DEM, which is what we used as our reference ice surface for this paper and for calculating the driving stresses in the following section (e.g., Fig. c).

Disturbed ice at the basal interface will be more likely if the ice is under horizontal gravitational stress due to surface slopes; however, decreased horizontal stresses may also be due to basal melting, which would also destroy the sought climate record. We calculate the driving stress to investigate this potential impact on old ice.

We derived surface slopes from a smoothed version (15 km Gaussian filter) of the Cryosat-2 DEM, and combine this with the new ice thickness compilation to derive driving stress (τ, Fig. a) using the following formula: τ=ρiceghsin⁡(θ), where ρice is 910 kg m-3, g is the acceleration due to gravity of 9.8 m s-2, h is the ice thickness, and θ is the slope of the ice sheet surface. Surface slopes (compare with Fig. c) dominate the driving stress map.

The englacial reflectors imaged as part of this survey represent isochrones that can be dated at the existing EPICA Dome C site using the methods outlined in , and inverted for basal age. This work is now in progress, with the stratigraphy developed by being propagated through the entire OIA survey, and is the subject of follow-up papers .

used the presence of subglacial lakes to calibrate their geothermal heat flow model, and the significant amount of free water in the Candidate A area may cause some doubt as to the prediction of basal freezing. However, it is clear that most of these lakes lie within local valleys in the bed rock. Most lakes do not lie under ice less than 3000 m thick (Fig. b), and Fig.  demonstrates that the majority of lakes lie below the enveloping surface of the massif.

When projected in terms of hydraulic head (the height of water consistent with the overburden pressure at the bed), few lakes appear above 2950 m head, implying the presence of a limiting hydraulic or thermal “water table” (Fig. ). The implication for these lakes on regional geothermal heat flow may be limited by local topographic focusing, which may locally double basal heat flow . This factor is not taken into account in the model due to its 5 km spatial resolution, but may be significant in the interpretation of subglacial lakes in this deeply incised region. Notably, there should be conservation of energy; geothermal heat flow will be reduced in the regions between valleys, and latent heat absorbed by melting ice in the valleys will not be available for melting on the highs between valleys.

The primary target for ground work is 10 km upstream from the nearest subglacial lake at a similar hydraulic level (Fig. ), further indicating the need for high-resolution work in this area.

Only Candidate A lies over the divide. The other frozen bed candidates lie well off the divide, and lie downstream of significant subglacial topography. In addition, observed driving stresses (derived using a validated surface elevation DEM), provide additional context, although the interpretation is not straightforward.

Without reliable velocity data, a formal inversion for basal shear stress is not useful; however, driving stress and basal shear stress are often well correlated . We see in Fig.  elevated driving stresses correlated with thinner ice (Fig. b) and elevated bed topography (Fig. c). The relationship with bed structure implies that this configuration of driving stresses have not evolved much over time in response to divide migration.

We see elevated driving stresses over Candidates B, C, D, and E, as these candidates lie on local mountains, and over the southern edge of the Candidate A massif. Basal ice in these locations may be disturbed by the application of these elevated driving stresses. Within Candidate A, regions with low driving stress on either side of the ice divide have a population of subglacial lakes, implying an element of basal lubrication, and thus melting may have modified the base of the ice. In the upland region on the divide, driving stresses increase over the southern escarpment. The targeted region (Fig. a) has intermediate to high driving stresses that may be an unavoidable consequence of targeting a shallow area where the ice surface is at its highest.

While the results of indicate that basal ice may be sensitive to elevated basal roughness, conversely low basal roughness may also indicate conditions not favorable to the preservation of coherent basal ice. We observe a spatial relationship between subglacial lakes and larger smoother areas, including near the EPICA Dome C ice core site (Fig. ).

Areas of elevated basal roughness appear to be associated with valleys in the topography, and again subglacial water. However, there does appear to be regions between the incised, rough valleys with reduced roughness, in the interior of Candidate A on the order of 5 km across.

In the Dome C region, we found no ice thinner than 2500 m that lay over subglacial topography that was appropriately flat. However, we found a significant region with overlying ice thinner than 3000 m that was lacking in free water bodies and contained kilometer-scale flat regions. This was consistent with the coldest region of the model.

The Dome C region, as mapped by the OIA survey, challenges intuition regarding old ice targets. Subglacial lakes are common, the bed is rugged in key places, and the ice is not as thin as suggested , although this general guideline was to help avoid meltwater, which is here identified by specific means. However, a detailed inspection of the data is encouraging.

The best compromise target is the center of the massif, near the ice divide of Little Dome C, in the Candidate A region (Fig. ). Flatter regions lie between incised, rough valleys that serve to capture geothermal heat flow and melt water; and thus the rough topography of the Candidate A region may serve to preserve old ice on elevated areas between the valleys.

However, a trade off is that maintaining a simple flow path for basal ice in such an rough environment will be difficult, and the mountainous region also induces relatively large driving stresses in the overlying ice. The paths taken by basal ice elements in such an environment may be torturous, and result in stratigraphic complexity. These compromises may be a requirement of finding the necessary “flat mountain” for ice from 1.5 Ma. Detailed site selection work (currently in progress), and careful, 3-D modeling of geothermal heat flow in the context of rough topography will be required.