Main

Few archives constrain the history of the Greenland Ice Sheet in its modern interior. Cosmogenic radionuclides, organic material and stable isotopes from subglacial sampling provide evidence for a smaller ice sheet sometime during the Pleistocene1,2, and sea level estimates indicate substantial mass loss from Greenland during the last interglacial period3. However, these proxy data provide very limited spatial information, leading to major questions about the size and configuration of the Greenland Ice Sheet over the last few glacial cycles. Although challenging to interpret, the most spatially comprehensive and easily measured record of Greenland’s history lies in the stratigraphy of the ice sheet. In this Article we build on ongoing efforts to both map4 and understand5 Greenland’s internal structure, with the goal of improving our knowledge of Greenland’s behaviour during retreat and regrowth.

Within most of the Greenland Ice Sheet the layering is simple, draping the subglacial topography at the ice-sheet base, transitioning to smooth, sub-horizontal layering near the ice-sheet surface. However, across Northern Greenland, ice-penetrating radar data have revealed many large structures that disturb the deep stratigraphy. Using only geometric information from radar data, two categories of models for their formation have developed. The first are those that describe these features as ‘folds’, arguing they are composed of deformed meteoric ice, and radar images capture overturned (but otherwise typical) layering6,7,8,9. These models rely on non-steady flow behaviour (changes in the spatial pattern of glacier sliding, shear or flow direction through time) to explain the structure geometry, with folding enhanced by heterogeneous or anisotropic ice viscosity10. The second set of models describe these structures as ‘plumes’, and argue they represent refrozen meltwater accretion at the base of the ice sheet11,12. Because both models are consistent with measured structure geometries, the debate over formation mechanisms cannot be resolved without new types of data.

Geophysical methods for measuring ice sheets have advanced greatly since those first interpretations. Beamforming techniques applied to multi-channel ice-penetrating radar enable swath mapping of subglacial topography13, dramatically increasing the resolution of our best digital elevation models of the glacier substrate14,15. Here, we use that same technology to more completely describe the nature of scattering from these structures, to infer their material properties rather than just their geometry. We show that scattering within these features is likely generated by entrained subglacial debris, concentrated in discrete scattering horizons that we refer to here as ‘debris trains’. This observation challenges some interpretations of structure formation in Greenland, and brings new evidence to an unresolved debate about how debris is emplaced within glacier systems that invigorated the glaciology community in the 1960s16.

In the following, we collect observations of debris trains from Greenland and Antarctica in an effort to address major outstanding questions about ice flow generally and the history of Greenland specifically, including ‘when did these structures form?’, ‘what explains their pervasiveness in Northern Greenland but relative absence in similar environments elsewhere?’ and, finally, ‘what implications might they have for projections of future ice flow?’

Insights from radar swath imaging

Ice-penetrating radar captures the structure of glaciers by transmitting an electromagnetic wave and measuring backscattering that arises from contrasts in the electromagnetic properties of subsurface materials17. Beamforming techniques make it possible to determine both the back-azimuth and distance to subsurface targets, turning two-dimensional (2D) scattering profiles into 3D scattering volumes. This technology has been used to produce high-resolution maps of subglacial topography, but the nature of backscattering from within the ice in 3D radar volumes, mostly ignored in earlier studies, can help us better understand the physical make-up of imaged layers, and thereby help us better understand the anomalous structures measured in northern Greenland.

For meteoric ice, subtle variations in density and chemical impurities produce specular (mirror-like) reflections within the ice sheet18. Backscattering from specular surfaces arrives from only one back-azimuth, the vector normal to the scattering interface. Because most of the energy is directed downward, steep specular layers are notoriously difficult to image19, with vertical boundaries essentially impossible to capture (a challenge in all fields of exploration geophysics). Yet, in complex ice-sheet structures like those investigated here, imaged horizons appear to turn from horizontal towards and eventually past vertical. Such scattering behaviour violates our typical model for electromagnetic waves within ice sheets, and must arise from diffuse rather than specular scattering.

Direction-of-arrival analysis, both along-track using delay-Doppler analysis20 (Fig. 1a) and across-track using the multi-channel array13 (Fig. 1b), confirmed that meteoric layering in the upper part of the ice column is specular (arriving from a single back azimuth). However, from the deep ice, we see two other types of scattering behaviour. The first we call type 1—diffuse scattering with an intensity weaker than the overlying meteoric layering, not apparently localized to discrete horizons, with directions-of-arrival restricted to within ±5∘ of nadir (Extended Data Fig. 1). The second we call type 2—diffuse but intense scattering, in many places brighter than the ice-bottom reflector (30+ dB higher than isochrons at comparable depths, Extended Data Fig. 2), concentrated in distinct horizons that scatter at angles of ±20∘ off nadir. By producing radar images that include only signal from near nadir (Fig. 1c) and images that average signals from off-nadir (Fig. 1d), the qualitative difference in scattering behaviour between the specular scattering in the upper half of the ice column and the two types of diffuse scattering in the lower half stands out.

Fig. 1: Beamforming products generated from radar data collected in Northern Greenland.
Fig. 1: Beamforming products generated from radar data collected in Northern Greenland.
Full size image

a,b, Backscattered energy as a function of two-way travel time and direction-of-arrival (DOA), both along-track (a) and across-track (b). c,d, Focused along-track profiles, showing energy scattered from targets near nadir (c) and energy scattered from targets ±20∘ off nadir (d).The trace associated with a and b is marked by the orange vertical line in c and d. An explanation of the different radar images and profiles is shown top left.

Type 1 scattering appears consistent with transitions in ice fabric, arising from the dielectric anisotropy of individual ice crystals (although low concentrations of small, entrained point scatterers cannot be ruled out). Type 2 scattering requires horizons of more dielectrically distinct, densely concentrated scattering targets (consistent with debris-rich ice, Extended Data Fig. 3), and in many places is associated with significant power losses, as is evident from reduced backscattering from features that lie beyond the type 2 scattering horizons (Extended Data Fig. 4). Type 2 scattering defines the features we describe as debris trains.

In addition to characterizing the diffuse scattering in the deep ice column, the two direction-of-arrival techniques together allow us to calculate the layer-normal vector for specular horizons in the top half of the ice column. These layer normals contain information about layer deformation. Englacial layer slopes are a product of spatial and temporal patterns of accumulation, the flow over complex basal topography, and the so-called ‘Weertman effect’—differential strain with depth in the ice column and along flow21. In the shallow part of the ice column, we see a gradual and linear increase in layer slope with depth, followed by a rapid change in the apparent layer slope in the layers immediately adjacent to debris trains (Fig. 2). This break in slope is also typically associated with a zone of type 1 scattering. Together, these indicate localized strain and ice-fabric reorganization, probably due to strength contrasts induced by the debris train itself.

Fig. 2: Three examples of along- and across-track layer slope measurements above debris trains in Northern Greenland.
Fig. 2: Three examples of along- and across-track layer slope measurements above debris trains in Northern Greenland.
Full size image

The top panels show conventional radar profiles capturing the structure geometry (with flow oriented primarily left to right, with components into, ⨂, and out of, ⨀, the page). The lower panels show along- and across-track direction-of-arrival images for the trace indicated by the vertical dashed line in their associated top panels. Best-fit directions-of-arrival (indicating layer slopes for specular layering) are shown in red, offset by red arrows from the true image maxima to facilitate comparison.

These observations indicate past and/or present ice-flow processes capable of emplacing basal ice and its entrained substrate material hundreds of metres off the ice-sheet bed. They also imply that debris trains locally modify ice strength, localize englacial shear, and affect ice-fabric development.

The spatial distribution of debris trains

In Fig. 3 we consider the spatial distribution of observed debris trains. There are few published studies describing debris trains in major ice sheets (past studies have been focused in coastal Greenland22 and around nunataks and subglacial mountain ranges in Antarctica23,24,25,26,27), but they are widely described in surging alpine glacier systems28,29,30,31.

Fig. 3: Maps showing subglacial topographic roughness where evidence of debris trains has been identified.
Fig. 3: Maps showing subglacial topographic roughness where evidence of debris trains has been identified.
Full size image

Several lines of evidence for debris trains are presented, including: shear localization (black symbols), intense, diffuse, englacial scattering (red symbols), englacial scattering losses apparent from reduced amplitude of the bed reflector (blue symbols) and overturned or otherwise anomalous ice layer geometries (white symbols). Debris trains identified in previous studies are plotted as triangles22,23,24,27,49. Subglacial roughness maps are derived from Bedmap3 for Antarctica50 and BedMachine for Greenland51.

Proposed mechanisms for debris-train formation (referred to as ‘inner-moraines’) were hotly debated more than 60 years ago. Two hypotheses were developed that could explain their presence: (1) compression, shear localization, and thrusting associated with debris-rich planes of weakness or (2) subglacial meltwater refreezing that underplates debris-rich basal ice, elevating it into the ice column16. As originally defined, both models share one requirement, that debris trains form at thermal transitions, with sliding, temperate ice upstream and sub-temperate, stagnant ice downstream. This requirement, together with their spatial distribution, gives us a basis for evaluating whether or not observed debris trains could have initiated under modern ice-sheet conditions.

Fig. 4: Schematic describing ice-sheet thermal evolution and debris-train development in Greenland.
Fig. 4: Schematic describing ice-sheet thermal evolution and debris-train development in Greenland.
Full size image

a,b, Although the exact size of the MIS 5e ice sheet is debated (with model realizations from refs. 2,34,35,36,37,38,39,40 showing divergent histories represented by the model extents shown in b), reductions in thickness and extent from the penultimate glacial maximum (a) generate the preconditions for debris-train formation. c, At newly developed thermal transitions (i), which could form in newly deposited glacial ice or interior to the remnant ice from the last interglacial, either thrusting or subglacial meltwater accretion emplaces debris-rich ice high into the ice column, forming a debris train (ii).

Transitions in the basal thermal state are a product of spatial variability in one or more of the controlling terms in the energy balance: geothermal energy flux, ice thickness, ice surface temperature, rate of vertical ice motion or basal strain rates. Thick ice, high geothermal flux, and high flow speeds promote temperate basal ice. Because flow speed and surface temperature tend to increase toward the ice margins, steady-state transitions from temperate to frozen basal ice typically require either significant thinning of the ice along flow or significant reductions in the geothermal heat flux.

As is apparent from Fig. 3, all published debris trains identified in Antarctica exist in areas of high subglacial relief, where transitions in the basal thermal state of the modern ice sheet can be easily explained by the ice thickness. Where the variability in ice thickness is small in the continental interior, there is no evidence of debris-train formation. This is different from the distribution across Greenland. Debris trains and deformed basal ice are observed in marginal ice environments, which have comparable subglacial relief to the areas where debris trains are observed in Antarctica, but the majority of features are found in the ice-sheet interior where subglacial relief is low. In addition, they are concentrated in fast-flowing outlet glaciers (probably advected there from the ice-sheet interior) with known thawed beds32. Their distribution is seemingly inconsistent with formation in the modern ice sheet, but their prevalence suggests a process that generated along-flow thermal transitions across large parts of the ice sheet.

Implications for the history of the Greenland Ice Sheet

Greenland is a terrestrial ice sheet, and is less climatically isolated than the younger, marine West Antarctic Ice Sheet and the older, stable, terrestrial East Antarctic Ice Sheet. Because of these characteristics, the Greenland Ice Sheet has probably experienced more significant variation in its basal thermal state over glacial–interglacial cycles, driven partly by thickness change and partly by surface air temperature change. We argue here that the ubiquity of structures across Northern Greenland (and their absence in the smooth interiors of West and East Antarctica) is a product of the unique thermal structure of the Greenland Ice Sheet during terrestrial readvance following the last interglacial period (‘The Eemian’, or Marine Isotope Stage 5e (MIS 5e)) (Fig. 4).

Several lines of evidence indicate a smaller Greenland Ice Sheet in the recent past. Sea-level proxies indicate that the global mean sea level was 1.2 to 5.3 m higher during MIS 5e than today3. However, studies of Greenland’s contribution to that sea-level rise span a wide range, from contributions of ~1 m sea-level equivalent to contributions as high as 5 m sea-level equivalent33, and ice-sheet models forced by the known climate yield a wide range of ice-sheet configurations2,34,35,36,37,38,39,40. Ice-core records and cosmogenic nuclide dating of subglacial samples provide some constraint on model performance, but because of non-unique exposure histories and limited spatial sampling, there is still large uncertainty about the size of the Greenland during MIS 5e. Data from the GISP2 bedrock core indicate ice-free conditions at the modern Greenland Summit at some point within the last 1.1 Myr (ref. 2), with inferred ages from the deep ice in the GISP2 ice core41 and basal debris from the NEEM ice core1 indicating persistent ice for much of that time. Meanwhile, data from subglacial sediments collected at Camp Century in Northwest Greenland indicate it was last ice-free during MIS 11 (424–374 ka)36.

During MIS 5e, Greenland experienced its highest summer temperatures of the last 600,000 years42, which tended to increase both precipitation and melt43. The net effect of warming was negative surface mass balance, thinning and marginal ice loss. The ice sheet that survived the last interglacial would therefore be the relatively warm core of the much larger ice sheet of the penultimate glacial period. This configuration provides the first necessary precondition for widespread debris-train formation—a thawed and readily deformable ice-sheet interior. Readvance of the MIS 5e ice sheet occurred during cooling, with net positive snow accumulation across the continent. Thin, cold, marginal ice provides the second necessary condition, a contrast in the basal thermal state along flow. We argue this enabled the formation of debris trains in regions of low relief, possibly via surge-like dynamics. The debris and fabric developed modified local ice fluidity, which would enhance subsequent deformation and enable the growth of more complex folds with time10.

So what explains the absence of debris trains in Southern Greenland and across Antarctica? It is possible that similar structures formed in Southern Greenland; however, high accumulation rates during the last glacial period and the Holocene yield low residence times for ice in the south, and thus those structures (and associated sediments) would be largely fluxed out of the system. In Antarctica, the transitory conditions required for debris formation are not expected. Weaker variations in Antarctic climate on glacial–interglacial timescales fail to produce large-scale marginal retreat of terrestrial basins in East Antarctica. If deglaciation of the West Antarctic Ice Sheet occurred in our recent past, readvance into its marine basin cannot be accomplished by cold, marginal deposition and grounded readvance with high basal stresses.

Because both proposed mechanisms for debris-train formation require similar thermal conditions, knowledge of the precise mechanism is not required to infer this thermal history. However, the precise mechanism has implications for the nature of the ice found within the structures. Given our proposed timing for structure formation, convergence-driven thrusting (as opposed to accretion of refrozen meltwater) would produce repeated sequences of Eemian or early glacial ice within these structures (as opposed to refrozen subglacial meltwater), making them exciting targets for subglacial drilling for palaeoclimate research. It is thus worth considering which mechanism is more likely given the evidence available.

In his early work, Weertman favoured the explanations of debris trains that called on subglacial meltwater accretion, arguing that the spacing and observed lack of displacement across debris bands were inconsistent with thrusting16. Subsequent work by Moore and colleagues44, focused on brittle fracture, also questioned published observations of thrusting, citing their conclusion that brittle faulting should not occur at slip/no-slip glacier transitions. However, subglacial hydrologic models fail to reproduce the size and position of structures observed in Greenland45, and field observations of debris bands do show displacement across debris trains24,31. Rather than resolve Weertman’s original debate, new observations have reopened it.

The idea of localized englacial shear is no longer controversial. In East Antarctica, incoherent scattering in radar data together with ice-flow modelling have been used to infer shearing flow over a stagnant basal layer46, potentially a product of fabric transitions in the deep ice47. Across blue-ice moraines in the Antarctic Dry Valleys, exposure ages and boron isotope concentrations indicate the progressive stagnation of marginal ice, with subglacial debris transported from bed to surface along shear planes along an internal thermal boundary24. Field data collected after the surging of Kuannersuit Glacier show repeated sequences of meteoric ice (identified by repeated sequences of oxygen isotopes) separated by debris trains22, again, the product of stagnant marginal ice. Given the complex thermal structure of a terrestrial ice sheet during regrowth, we believe the more parsimonious explanation for debris trains calls on endogenous shearing processes rather than exotic subglacial hydrology, with debris trains representing thrusting within the ice sheet.

Finally, observations of diffuse scattering and layer slopes above debris trains show they have an impact on ice properties, localizing deformation and modifying the ice fabric. This apparent property heterogeneity has implications for the accuracy of current prognostic ice-sheet models, which most commonly assume either a constant ice fluidity or an ice fluidity that only varies with temperature. Debris and fabric development also introduce fluidity variability, and, as long as debris concentrations are low enough to prevent debris-grain interactions, debris can lead to profound weakening of the ice48. When initialized to match modern surface velocity, errors in model-assumed ice fluidity will result in compensating errors in basal friction, shunting motion actually accommodated by ice deformation into the basal sliding term (or vice versa). Because sliding and ice deformation have different functional relationships with stress, failing to include elevated fluidity in ice can counter-intuitively produce model results that are too responsive to changes in stress, depending on the prescribed glacier sliding law (Supplementary Section 2 provides a full description of this effect). This is important to consider when describing uncertainty in prognostic models of North Greenland.

Conclusions

Beamforming techniques applied to ice-penetrating radar data collected in Greenland reveal subglacially sourced debris more than 1,000 m from the ice-sheet bed. This observation reopens the debate over debris transport through glaciers and ice sheets dating back to the 1960s16, and challenges previous interpretations of structures in Northern Greenland. We argue here that these debris trains are a record of the transient thermal state of the Greenland Ice Sheet during regrowth following the last interglacial period.

We believe our observations may support models of debris-train formation by thrusting rather than subglacial meltwater refreezing. If this is the case, ice below observed debris trains may contain repeated sequences of Eemian or early Glacial ice. Both to ground-truth the radar-inferred properties and as a potential source of information about Greenland’s climate history, direct sampling of these structures could prove invaluable.

Finally, englacial layer slopes and apparent fabric transitions in the meteoric layering above debris trains in Greenland indicate they modify the ice fluidity and promote localized englacial shearing. Assuming homogeneous fluidity for ice in Northern Greenland will produce compensating errors in the model prescribed basal friction during typical model initialization. This can result in model ice sheets that are more responsive to stress changes than the real ice sheet, depending on the functional form of the implemented sliding law.

Methods

In this Article we infer three different types of scattering differentiated by backscattering amplitude and direction-of-arrival information. Direction-of-arrival synthesis was performed following the methods of ref. 20 for along-track imaging and using the Multiple Signal Classification algorithm (MUSIC) for across-track imaging (described in ref. 13 and implemented in the Open Polar Radar toolbox (https://gitlab.com/openpolarradar/opr)). The scattering types we identified are as follows: specular scattering from meteoric layering, weak incoherent scattering from a narrow angular range (type 1), and strong incoherent scattering measured across a wide angular range (type 2).

Measured backscattering amplitude is a product of both scatterer properties and path effects. For debris trains, we attempt to isolate scatterer properties by comparing their signal character to other targets at similar depth within the ice sheet, assuming that path effects (ohmic and scattering losses) do not vary substantially in the shallow ice (an assumption we re-evaluate for deeper targets). From amplitude alone, debris trains are distinct, with reflection powers up to 30 dB higher than meteoric layering at comparable depths (Extended Data Fig. 2).

This difference must be explained by dielectric or geometric differences in the nature of the scattering targets. Here, we calculate the non-dimensional backscattering cross-sections for specular meteoric layering and for volumetric scattering from debris-rich layers to evaluate whether or not 30 dB can be explained by entrained debris. We start with the following simplified form of the radar equation:

$${P}^{r}=\frac{S}{4{\rm{\pi }}{({z}_{\mathrm{air}}+{z}_{\mathrm{ice}})}^{2}}\,\,A{\sigma }_{0}\,\,\frac{1}{4{\rm{\pi }}{({z}_{\mathrm{air}}+{z}_{\mathrm{ice}})}^{2}}\,\,{{\rm{e}}}^{-2\alpha {z}_{\mathrm{ice}}}$$
(1)

Because the source properties described by S (transmit power, antenna gain, receiver aperture size) do not change along a radar profile and we are comparing targets at equal range (zair + zice) and depth (zice), the observed power variability must arise from changes in the radar cross-section (Aσ0) or the attenuation rate (α). If we assume that attenuation rates in the top half of the ice column are small and relatively homogeneous (typical for cold ice), the extra backscattered energy from debris trains (30+ dB) must be explained by differences in the radar cross-section of the targets. For specular, dielectric planes with power reflection coefficient R, the radar cross-section is defined in terms of signal wavelength (λ) and illumination area (A):

$$A{\sigma }_{0}=R\frac{4{\rm{\pi }}}{{\lambda }^{2}}{A}^{2}$$
(2)

For most radar systems, the illumination area is not beam-limited, and instead is treated as the first Fresnel zone of the system:

$$A=\frac{\pi \lambda }{4}\left({z}_{\mathrm{air}}+\frac{{z}_{\mathrm{ice}}}{\sqrt{{\epsilon }_{i}}}\right)$$
(3)

This produces a total radar cross-section of the following form:

$$A{\sigma }_{0}=R{\pi }^{3}{\left({z}_{\mathrm{air}}+\frac{{z}_{\mathrm{ice}}}{\sqrt{{\epsilon }_{i}}}\right)}^{2}$$
(4)

Mie scattering theory gives us a framework for calculating the non-dimensional radar cross-section for volumetric scattering layers52. Here, we use ‘miepython’53 to calculate the backscattering efficiencies (Qback) for debris of different sizes (defined by particle radius r). This can then be used to find the integrated radar cross-section using an assumed particle number density (N, generated from an assumed volume fraction, p). Defining the illuminated area as the first Fresnel zone and assuming some debris-rich layer thickness (zl), we can produce the total radar cross-section for Mie scattering from debris:

$$N=\frac{3p}{4{\rm{\pi }}{r}^{3}}$$
(5)
$$A{\sigma }_{0}=\mathop{\underbrace{\pi {r}^{2}{Q}_{\mathrm{back}}}}\limits_{\text{particle backscattering}}\mathop{\underbrace{{NA}{z}_{{\rm{l}}}}}\limits_{\text{number of particles}}$$
(6)
$$A{\sigma }_{0}={\rm{\pi }}{r}^{2}{Q}_{\mathrm{back}}N\frac{{\rm{\pi }}\lambda {z}_{{\rm{l}}}}{4}\left({z}_{\mathrm{air}}+\frac{{z}_{\mathrm{ice}}}{\sqrt{{\epsilon }_{i}}}\right)$$
(7)

Further discussion of radar amplitude analysis is provided in Supplementary Section 1, and the theoretical framework for considering model error is presented in Supplementary Section 2.