The Greenland Ice Sheet (GrIS) displays large spatial variations in surface velocity, with a few fast-flowing outlets draining most of the interior . It is therefore critical to capture the complex flow pattern of GrIS in models used for future sea level projections. Recent developments in ice sheet models such as efficient parallel computation , better representation of flow equations , detailed basal topography and the inclusion of subglacial hydrology have contributed to greatly improving the representation of this spatially varying flow . In addition to these advances, inversion for basal friction using surface velocities has proved to be a powerful tool , and models are now able to capture most of the complex flow pattern of the ice sheet. Inversions are useful to capture present-day velocity, but they mask information that is needed to evolve these conditions in time. Therefore, we cannot fully rely on inversions for future projections, as basal conditions may evolve as a result of a changing climate and in turn influence ice dynamics.

The Northeast Greenland Ice Stream (NEGIS) drains more than 10 % of the GrIS and is exceptional by displaying high velocities all the way to the ice divide . Despite its large impact on the GrIS mass balance, NEGIS is not accurately represented in ice sheet models without inverting for basal friction . simulated NEGIS in the Parallel Ice Sheet Model, capturing high velocities using a simple hydrology model, however lacking the far inland onset of the ice stream. used the basal melt rates from the model by in a more sophisticated hydrology model to reproduce NEGIS in the Ice Sheet System Model (ISSM). They capture the high velocity flow of the outlets well, but the representation of the transition areas outside of the main trunk are more diffuse compared to the observed values. These studies illustrate how we are getting closer to reproducing present-day NEGIS in ice sheet models. However, the characteristic clearly defined shear margins, and high velocities upstream at the onset of the ice stream are still lacking.

To understand why high upstream velocities are not reproduced in models, one must look into how the ice stream is initiated. The origin of NEGIS has been explained by a geothermal heat flux (GHF) anomaly left behind by the passage of the Icelandic plume . Interpretation of radar data points to unusually high basal melt rates at the head of the ice stream, corresponding to an exceptionally high GHF of 970 mW m-2 . A local increase in GHF intensifies basal water production and potentially enhances basal sliding. Unfortunately, GHF maps for Greenland display a large spread of values . These large uncertainties in the estimates of the GHF have been shown to dominate the uncertainty on the ice flux in this region . In addition, the GHF maps are coarse and may not capture local anomalies like the one suggested to exist at the head of NEGIS . Accurately capturing such a feature and explicitly representing the effect of high melt rates on basal sliding are key to reproduce the distinct velocity pattern of NEGIS in ice sheet models.

Here, we study the impact of the presence and intensity of a mantle plume, at the head of NEGIS on the ice flow structure. We do not suggest the presence of a mantle plume, but rather use an existing mantle plume model to generate feasible GHF scenarios in the model sensitivity study. We use a sophisticated hydrology model coupled to ice dynamics in the Ice Sheet System Model ISSM; to capture the influence of enhanced basal melt on ice dynamics. We first describe the models and different plume experiments. Finally, we present and discuss resulting basal conditions and surface velocities corresponding to the various plume configurations.

In the control simulation we use the GHF from (Fig. a), and the corresponding basal melt rates are shown in Fig. f. Melt rates at the head of the ice stream (at EGRIP) are 1–2 mm yr-1, and the highest basal melt rates (600 mm yr-1) occur at the grounding line of Zachariæ, with surface velocities reaching 1500 m yr-1. Friction is the dominating heat source in the fast-flowing regions, and melt rates thus increase with increasing velocities towards the grounding line. Low melt rates in regions with high velocity are due to low-lying bed topography causing low basal drag and hence less frictional heat. The effective pressure for the control experiment is shown in Fig. k, and the values increase upstream toward the ice divide as ice thickness increases and basal melt decreases. The lowest values of effective pressure coincide with low bed elevation in the main trunk, 100 km upstream of the grounding line.

The resulting velocity field for the control simulation captures the main features of NEGIS: the three outlets with high velocities across the grounding lines and sharp shear margins (Fig. p). The northern branch feeding into 79N is slower and less defined than in the observed velocities, and the velocities of Storstrømmen are also slower than observed. Velocities of the floating tongues of 79N and Zachariæ are not well represented, and floating shelves are not shown here. The western branch, feeding into the main trunk of NEGIS, shows a more diffuse pattern with higher velocities than observed.

To evaluate how well the model simulations reproduce the observed velocity pattern, we plot the 50 m yr-1 velocity contour (black contour in Fig. ), and we compare how far upstream this contour reaches (in kilometres from the ice divide) relative to the observed velocity (white contour in Fig. ). The modelled velocity contour in the control simulation reaches 305 km from the ice divide (Fig. p) and thus further downstream than the observed velocity (120 km, Fig. a, f, k). The control simulation does not capture the characteristics of NEGIS, with high upstream velocities close to the ice divide.

To capture the upstream velocities, we enhance the GHF locally at the onset of the ice stream in the plume970 simulation to reach the maximum magnitude proposed by . The addition of the mantle plume results in high GHF, with values up to 970 mW m-2, rapidly decreasing to the values used in the control simulation (Fig. b) within a radius of less than 100 km. High geothermal heat leads to high basal melt rates, with ∼ 100 mm yr-1 above the plume (Fig. g), compared to 1–2 mm yr-1 in the control experiment. The increase in basal melt rates causes a reduction in effective pressure to 1.2 MPa directly above the plume, resulting in a local floatation fraction (ratio of water pressure over overburden pressure) of 0.95. The resulting velocity field in the plume970 experiment is similar to the control experiment, except for the higher velocities simulated at the head of the ice stream. In the plume970 simulation the 50 m yr-1 velocity contour reaches 131 km from the ice divide (black contour Fig. q), which is close to the observed 120 km. However, the spatial pattern upstream is more diffuse and the ice stream is wider than observed. The Storstrømmen outlet shows higher velocities relative to the control simulation, but still lower than observed. The 79N and Zachariæ outlets, on the other hand, display higher velocities than observed. Overall, with this approach, we capture most of the characteristics of NEGIS, although the ice stream is more diffuse and displays velocities slightly higher than the observations.

To determine whether a lower GHF may induce a similar high-velocity pattern, we run three simulations with a less intense mantle plume. Figure c–e show the GHF values computed by increasing the plume depth to 3000, 4000 and 4500 km, respectively, obtaining maximum basal melt rates of ∼ 70 (Fig. j), ∼ 85 (Fig. i) and ∼ 95 mm yr-1 (Fig. h). The modelled effective pressure for the three plumes (Fig. m–o) results in slower velocities than plume970, with 50 m yr-1 velocity contours reaching 253, 245 and 210 km from the ice divide, respectively (Fig. r–t). This shows that GHF values of 677, 836 and 909 mW m-2 produce weaker ice stream signatures than observed and, given our model set-up, are not sufficient to induce the upstream fast flow of NEGIS.

To investigate the sensitivity of the position of the plume in plume970, we moved the plume 75 km to the southwest, southeast, northeast and northwest (Fig. ). The computed GHF distribution is shown in Fig. a–d and the basal melt rates are of the same magnitude as in plume970. The computed effective pressures for the southwest and southeast (plume970SW and plume970SE, Fig. i, j) have minimum values of 3.2 and 2.9 MPa above the plume, which are not sufficient to initiate fast flow (Fig. m, n). When the plume is located further downstream, the effective pressure reaches lower values (Fig. k, l) and the ice stream flows faster than in plume970 (Fig. o, p), however, with the 50 m yr-1 contour only reaching 204 km from the ice divide. The plume970NE induces the fastest flow, and the plume970NW creates an interesting double-branched ice stream starting from the ice divide. The experiments in Fig.  indicate that the elevated heat required to initiate the NEGIS in our model must be located close to EGRIP.

To determine whether a lower GHF value over a larger area could induce high upstream velocities, we investigate the influence of four weaker plumes with larger plume radii (Fig. ). The weakest but most extensive plume (plume494, Fig. a) produces basal melt rates of a maximum of 51 mm yr-1 (Fig. e), resulting in a large area of low effective pressure (minimum 0.2 MPa; Fig. i). The corresponding surface velocity for the plume494 displays a faster and wider ice stream (Fig. m) relative to the observations. Plume594 gives basal melt rates of 60 mm yr-1 (Fig. f) and the ice stream becomes wide, reaching all the way to the ice divide (Fig. n). Plume775 is twice the size of plume970 (Fig. c), and with melt rates of ∼ 75 mm yr-1 over a larger area (Fig. g) the velocity of the ice stream (Fig. o) is similar to plume970. However, the 50 m yr-1 velocity contour reaches too close to the ice divide and the ice stream is wider than the observed one. Plume792 produces melt rates of ∼ 75 mm yr-1 (Fig. d), resulting in velocities similar to those of plume594 (Fig. p). This shows that plumes with a restricted extent, ∼50 km × 50 km, produce model results more consistent with the observed flow behaviour in the upstream reaches of NEGIS.

Finally, we investigate the influence of varying the parameters in the friction law (Eq. ), presented in Fig. . The noHydro simulation with an effective pressure approximated to the hydrostatic pressure shows very little resemblance to the observed NEGIS (Fig. a), with too slow velocities. The simulation with a uniform friction coefficient and no mantle plume captures the main feature of NEGIS (Ctrl-uni, Fig. b): with a main trunk, the northern branch and three outlets, with the fastest flow in Zachariæ. However, the velocity pattern is more diffuse than the observed pattern (Fig. e). The high upstream velocities are better captured in the simulation with plume970 and a uniform friction (plume970-uni, Fig. c). For plume970-uni, high velocities reach slightly closer to the ice divide than plume970, but the velocities of the main trunk are less confined than in experiment plume970 (Fig. d) and the observations (Fig. f).

Most of the spatial velocity pattern of NEGIS is represented in our control run, apart from the upstream one-third of the main trunk. This indicates that the downstream area of the NEGIS catchment is largely controlled by topography, while the upstream area is controlled by its basal conditions, which is in agreement with . The control simulation captures the main outlets and the observed snake-shaped velocity pattern of the trunk. High velocities coincide with low-lying bed elevation. However, we do not capture the high velocity of Storstrømmen, or the floating tongues of the Zachariæ and 79North outlets. This could be caused by the simple friction coefficient approach not being representative of these areas, where basal properties display a more complex pattern.

We performed experiments with various mantle plume configurations introduced at the head of NEGIS to assess if the presence of an anomalously high GHF can explain the pattern of ice flow of this region. The different plume configurations vary in intensity, position and extent. In the control simulation we use present-day surface velocity and GHF from . Without the presence of a plume, the GHF does not reach more than 54 mW m-2 and leads to underestimating velocities in the upstream part of the catchment. These low values of GHF are not sufficient to initiate the onset of NEGIS close to the ice divide. By testing with four mantle plume configurations of increasing intensity (Fig. ), we find that the GHF (GHF) needed to induce the observed upstream velocity of NEGIS in our model is ∼ 970 mW m-2.

A GHF of 970 mW m-2 is consistent with the maximum value presented in and for regions in proximity to EGRIP, where plume970 is located. It also compares well to the anomaly modelled by in the trunk of NEGIS but does not include the high GHF that they find upstream. These GHF values are imposed based on basal melt estimates from radar internal stratigraphy. Our modelled basal melt rates (∼ 100 mm yr-1) are thus consistent with their proposed values. By directly comparing the basal melt rates of our plume970 experiment to the basal melt rate estimates from in Fig. , it can be seen that our plume produces a basal melt pattern that matches the position, extent and values of the northeastern branch of their anomaly. The sensitivity simulations in Fig. m, n show that more than 970 mW m-2 is needed to initiate high velocity, when the plume is located further upstream in a region with thicker ice relative to downstream. This suggests that the area of high basal melt estimated by in the trunk of NEGIS is probably more consequential than the larger melt anomaly that they modelled closer to the divide.

The GHF at the head of NEGIS is suggested to be high due to lithospheric thinning as a result of the Iceland plume passage . However, 970 mW m-2 is an extremely high GHF value, 10 to 20 times higher than the values suggested by GHF models for Greenland . derived GHF values for five deep ice core boreholes in Greenland, using the SICOPOLIS model (SImulation COde for POLythermal Ice Sheets; http://www.sicopolis.net/, last access: 4 March 2020), such that the simulated and observed basal temperatures match. This resulted in a local elevated GHF anomaly around NGRIP of 135 mW m-2, located at the ice divide ∼ 150 km away from the head of NEGIS. Our GHF anomaly has a magnitude 7 times higher than that of and 3 times as high as the highest current GHF observations in Greenland . In summary, plume970 produces a basal melt pattern with magnitude and extent in line with previous estimates from the radar data for the region within the 50 m a-1 isoline; however there is a large discrepancy between the necessary GHF to produce this melt and the GHF estimates for Greenland.

To explain the high GHF value of 970 mW m-2, we need to investigate processes that may locally elevate the GHF. and explained high GHF in this region by the passing of the Iceland plume, leaving behind partly molten rock that may have migrated up in response to glacial–interglacial cycles, as the crust is loaded and unloaded. A study showed that glacial rebound may have caused young intraplate volcanism in Greenland, despite the old age of the tectonic plate and no mantle plume present . The plume passage could have lead to shallow magma emplacements, which may feed hydrothermal systems, causing hot fluid percolation that enhances high heat transport to the base of the ice sheet . It is important to note that the term GHF is defined as the heat flux from the Earth's interior as a purely conductive heat transfer. Hence, the 970 mW m-2 heat flux can not be explained by GHF alone but rather also with surface heat flow from locally elevated GHF due to advective heat transfer from the processes mentioned above .

Comparing the velocity field in the plume970 experiment to previous studies without inversion shows that combining a basal hydrology model with an elevated GHF at the head of NEGIS captures the observed high, confined, upstream velocities of the NEGIS. The simulations in show that the ice flow models capturing the upstream onset of NEGIS all rely on inversions to initialise the basal drag in the simulations (Elmer/Ice, ISSM, BISICLES, GRISLI and f.Etish). The models without inversion underestimate the velocities in the upper part of the NEGIS catchment and lack the sharp velocity gradients. simulated the high upstream velocity of NEGIS without inverting for basal conditions in Parallel Ice Sheet Model (PISM), but their simulation lacks the clearly defined main trunk and underestimates the high upstream velocity. further improved the simulation by using a subglacial hydrology model to compute effective pressure, which allowed higher velocities in the outlets. However, high upstream velocities are still lacking, similar to our control simulation. The last two studies used GHF from , which proposed slightly lower values at the head of NEGIS compared to the values of used in our study.

used the same friction law as we use in ISSM, but with a uniform friction coefficient. We tested a uniform friction coefficient, which led to a more diffuse ice stream (Fig. b, c), but with more confined outlets compared to the study. The difference can be explained by different basal melt rates used as input and different hydrology models. In order to capture sharp gradients in the velocity field, we find it important that the areas without any basal melt have effective pressure equal to the ice overburden pressure.

We invert for basal friction to get the basal melt rates that are used to initialise the subglacial hydrology model, and the model is then free to evolve. We do not use the inverted friction in the forward ice flow simulation; instead we use the simple friction coefficient from Eq. (). To investigate whether the modelled velocity pattern is caused by the effective pressure distribution or the friction coefficient, we run the simulation noHydro, where the effective pressure is approximated to the hydrostatic pressure, commonly used in ISSM. The modelled velocity pattern (Fig. a) does not resemble the observed pattern, and we conclude that including the subglacial hydrology model is responsible for the improved velocity pattern in the control simulation and plume970. By using our friction coefficient distribution, combined with initialising with present-day basal melt from velocity observations, both the control and plume970 experiments display velocity patterns similar to the observations (Fig. d, e).

The middle western branch of the ice stream displays too high velocity in both the control and plume970 experiments, correlating with low-lying bed elevation (Fig. ). Too high velocities in this region were also modelled by using PISM and a similar bed-elevation-dependent friction law. When performing additional simulations with the GHF values from , this branch becomes more pronounced in velocity (not shown here). This may indicate that the GHF values in this region of Greenland are even lower than those in and , and the glacier base is frozen to the ground. This region is recognised as “uncertain” in the synthesis of Greenland's basal thermal regime by . Other explanations for too high velocities in this branch may be a higher bed roughness, errors in the bed topography or “sticky spots”.

Given the model configuration, an exceptionally high heat flux of 970 mW m-2 is needed to reproduce NEGIS. We acknowledge that this value may be overestimated due to uncertainties and assumptions in our model set-up, and we discuss these in the following sections. We use a simple friction law linearly dependent on effective pressure, and we are aware that the results are likely to change with a different choice of friction law. For example, in the friction law used in the MISMIP+ experiments , effective pressure is included only where the coulomb criterion is met, normally a few kilometres upstream of the grounding line. This may result in a smaller dynamic response from the mantle plume in the slow upstream regions of NEGIS. However, the use of a non-linear friction law may enhance the sensitivity of the ice dynamics to effective pressure, also upstream, as we compute low effective pressure above the plume. This implies that the use of a non-linear friction law may result in a lower GHF needed to sustain NEGIS in a model.

By using a coarse model mesh we may underestimate the softening occurring due to strain heating in the shear margins and hence overestimate the lateral drag. Refining the mesh and inducing damage softening of the ice in the shear margins would decrease the lateral drag. In this case, the observed high upstream velocity of NEGIS may have been reproduced with higher basal drag and hence lower GHF. The underestimation of modelled ice softness may also explain why our modelled upstream velocity field is wider and more diffuse than the observed field.

In the simulations where we investigate the influence of an increased plume radius (Fig. ), we show that lower values of GHF can induce even faster flow, when the plume is more extensive (Fig. ). However, with a larger mantle plume the ice stream becomes wider and does not match the observed velocity of NEGIS (Fig. e). The basal melt pattern of in Fig.  consists of two melt anomalies near EGRIP. It would be interesting to investigate the velocity response of two weaker elevated GHF anomalies closely located. There is also room for improvement of the model in the treatment of the shear margin or the use of a non-linear friction law . Both those improvements would lead to sharper transition from slow to fast velocities and might allow a plume with a larger radius.

We parameterise the friction coefficient with a simplified estimate linearly dependent on the bed elevation. In other studies this coefficient is inverted for by matching observed surface velocity, producing low values in the main trunk of NEGIS . By lowering the friction in the main trunk, we may reproduce fast flow with a lower GHF value. However, this would make the friction coefficient relate to the velocity, which we are trying to avoid. The bed topography used is from BedMachine , so datasets used to create this map impact the choice of friction. A uniform lowering of the friction coefficient, also outside the trunk, would increase velocities all over the domain; hence we would lose the sharp velocity gradients and overestimate the outlet velocity even further. Additionally, the modelled ice surface in the control experiment is lower than the observed ice surface , and a uniform reduction of friction will enhance this mismatch. We do not observe a local depression in the surface topography above the 970 mW m-2 plume, which agrees with the observed ice surface for the region .

Hydrology parameters are unfortunately highly uncertain, and different choices would lead to a more or less responsive hydrological system and hence possibly a lower GHF value to sustain the fast flow. However, we have a rather low transmissivity of the inefficient drainage system, resulting in low efficiency in water evacuation, causing our system to be sensitive to an increase in water input. If the transmissivity was lowered further, the efficient drainage system is likely to activate in the GHF anomaly region, lowering the water pressure and becoming less sensitive to increased water input. For this reason, we do not expect that a different hydrology configuration would reproduce NEGIS with a lower heat flux. In addition, the subglacial hydrology is only one-way coupled to ice dynamics, so we do not capture the positive feedback expected with higher velocities leading to more melt, and lower effective pressure, giving even higher velocities. With a more responsive and fully coupled system, one might be able to reproduce NEGIS with lower heat flux.

With a simple bed-elevation-dependent friction and hydrology model forced by melt rates from GHF, we capture the overall pattern of NEGIS velocity. This has implications for studies trying to predict the response of NEGIS to a future climatic warming. Basal friction may not remain constant in time, and thus we cannot fully rely on inversion as it masks unknown time-varying basal properties. By using our approach (with or without the GHF anomaly) one can capture complex velocity patterns and then invert for the remaining basal properties. These may in turn be assumed to be constant in time, while the subglacial hydrology will evolve with a changing climate, accounting for varying basal conditions. Unfortunately, observations and estimates of GHF and subglacial hydrology are challenged by large uncertainties. Therefore, it is critical for future observational and modelling studies to better constrain the basal conditions of the Greenland Ice Sheet.

Present-day basal melt rates from GHF maps and frictional heat are not sufficient to sustain the observed upstream velocities of the Northeast Greenland Ice Stream (NEGIS). The downstream velocities appear to be driven by topography, and the spatial pattern is well captured by the subglacial hydrology model. Our findings suggest that a local heat flux anomaly may explain the characteristic high upstream velocity of NEGIS and hence is consistent with previous studies . To reproduce high upstream velocities at the onset of NEGIS, a sustained basal melt rate of 100 mm yr-1 is needed in a local region close to EGRIP, where observed present-day velocities reach 50 m yr-1. Hence, the minimal heat flux value needed to initiate the ice stream in our model is 970 mW m-2, as proposed by . This magnitude is too high to be explained by GHF alone, and we suggest that processes such as hydrothermal circulation may locally elevate the heat flux of the area.