Yearly ice-surface elevation change was determined from 2009 to 2017 by differencing the annually averaged ArcticDEM DSMs. This revealed three distinctive quasi-circular features, hereafter referred to as “anomalies”, all within 2 km of the lateral margin of Isunguata Sermia, that were characterized by periods of subsidence followed by uplift (Fig. 1a). Time series of relative elevation change for each anomaly, which are not advected towards the margin, were calculated from sub-annual ArcticDEM DSMs by subtracting the mean ice-surface elevation of the anomaly from the mean elevation of a 500 m buffer around it (Fig. 1b). This approach was used to isolate the dynamic effect and to remove the influence of systematic vertical and horizontal offsets (of up to 3–5 m) between DSMs. Anomaly 1, located < 5 km from the terminus of Isunguata Sermia where the ice is ∼ 325 m thick (Lindbäck et al., 2014), formed a 0.93 km² depression between 19 August 2010 and 3 August 2011 with a mean depth of 5 m and maximum depth of 17 m. The ice surface then rose 1 m by November 2011 before recovering back to its 2010 elevation by February 2013. Anomaly 2 (∼ 370 m ice thickness), about 1 km further up-ice, formed a 0.88 km² depression between 2 August 2015 and 21 September 2015, with a mean depth of 13 m and maximum depth of 30 m. It then rose 9 m between 2015 and 2017. Anomaly 3 (∼ 450 m ice thickness), immediately up-ice of anomaly 2 and ∼ 9 km from the terminus, formed a 0.67 km² depression between 17 August and 19 September 2014, with a mean depth of 4 m and maximum depth of 14 m, before the surface rose 3 m between 2014 and 2017. Surface structural features indicative of localized ice fracture during subsidence, such as crescentic crevasses, are not apparent in any of the depressions.

Ice velocities derived from feature tracking of Sentinel-1 radar data (see Appendix) do not show a consistent area of fast flow over any of the anomalies. However, during the subsidence of anomaly 2 in late July to early August 2015 (Fig. 2), ice flow immediately downstream of the anomaly experienced a net slowdown to roughly winter values (Fig. 2b, c), coincident with an overall regional slowdown in ice flow. This was followed by a return to pre-subsidence flow speeds by 19–31 August 2015 (Fig. 2d). During the period of subsidence ice flow also converged into the depression, leading to localized rapid “up-glacier” flow against the regional westerly flow direction, which manifests as a strong positive ice flow change immediately above the anomaly (Fig. 2c).

Figure 2Ice dynamic response to subglacial lake drainage. Ice velocity anomalies (% change) are relative to winter 2015 (3 January–5 April).

Figure 3Proglacial signature of the 2015 subglacial lake drainage event. Panels (a) and (b) show True Colour Landsat 8 images of Isunguata Sermia and the foreland before (15 July 2015) and after (25 August 2015) the drainage event. The blue/purple coloured line in (a) represents the predicted subglacial drainage routeway. IS denotes Isunguata Sermia. Panel (c) is change in normalized difference water index (NDWI) between 7 July and 25 August 2015. Panel (d) reveals the change in elevation from two ArcticDEM DSM tiles (co-registered to remove systematic vertical offsets using the mean vertical difference between common bedrock surfaces). The dotted line demarcates the mapped wetted area in (a).

Download

Landsat 8 OLI satellite images acquired before and after the 2015 ice-surface subsidence (anomaly 2) reveal a major change in the 1.8 km wide proglacial braided river system (Fig. 3). On 15 July 2015 the river plain is characterized by a single channel emanating from the front of Isunguata Sermia, which then bifurcates downriver into multiple braids and intervening bars (Fig. 3a). Dry areas above the water level are demarcated by a sharp change in colour, with wetted areas being darker and dry areas lighter. Using this demarcation, a major flood plain directly in front of the main portal, which causes the river emanating from the glacier to divert northwards and then westwards, is identified. On the basis of a qualitative change from light to dark, on 25 August 2015, and a quantified positive change in normalized difference water index (NDWI) (Appendix) of up to +0.23 (mean: +0.09) between July and August, the dry areas (bars and floodplain) became inundated by water and the braided river system re-organized (Fig. 3b, c). Differencing ArcticDEM DSMs of the proglacial area before (4 May 2015) and after (21 September 2015) the ice-surface elevation change of August–September 2015 associated with anomaly 2 reveals 3 m of mean net sediment aggradation across a 5 km stretch of the main proglacial channel (Fig. 3d). Aggradation was up to 8 m close to the outlet and declined to < 1 m 5 km from the glacier terminus.

We identify three ice-surface elevation anomalies on Isunguata Sermia, which we interpret as subglacial lake drainage and filling (Fig. 1). From this point onwards we therefore refer to these anomalies as “Subglacial Lakes 1–3”. Confirmation of a subglacial lake origin is provided by flooding of the proglacial outwash plain in August 2015, which coincided with the timing of ice-surface elevation anomaly 2, evidencing the release of meltwater (Fig. 3). This inundation (wetting) of the flood plain is not replicated at the nearby Leverett–Russell Glacier (Fig. 3b), ruling out a common external forcing (e.g. heavy rainfall). All three subglacial lakes are located under 325–450 m thick (Lindbäck et al., 2014), warm-based ice on a reverse gradient bed slope (15 m per km); the reverse slope may be trapping the water causing the lakes to form. Although subglacial hydrological analysis (e.g. Chu et al., 2016) does not produce hydraulic minima in the locations where we identify lakes, this is likely to be the result of the limited and relatively poor-quality airborne radar ice-thickness measurements across the thin, near-marginal area of Isunguata Sermia and the resolution of the bed DEMs.

The three subglacial lakes underwent one drainage event each over the 8-year data period (Fig. 2). Differencing of the DSMs either side of the drainage events, over the area of the ice-surface anomalies, gives total lake volume changes of $\mathrm{6.5}\pm \mathrm{0.52}\times {\mathrm{10}}^{\mathrm{6}}$ m³, $\mathrm{1.3}\pm \mathrm{0.05}\times {\mathrm{10}}^{\mathrm{7}}$ m³, and $\mathrm{3.5}\pm \mathrm{0.38}\times {\mathrm{10}}^{\mathrm{6}}$ m³ for Subglacial Lakes 1–3 respectively. Drainage of Subglacial Lake 2 in 2015 and Lake 3 in 2014 was triggered in August and drained in < 1 month for both, which is consistent with other larger subglacial lake drainage events identified in Greenland (Palmer et al., 2015; Willis et al., 2015; Bowling et al., 2019), but contrasts with the longer (months to years) drainage period of those in Antarctica (e.g. Siegfried and Fricker, 2018). If the vertical displacement of the ice surface is equivalent to the depth of the subglacial lake, this gives a mean minimum discharge of 6.5 m³ s⁻¹ for Subglacial Lake 2, which is the largest and best-constrained lake, by available DSM and satellite imagery, in this study. This is, however, well below the likely maximum discharge given the enormous mobilization of sediment that resulted from the drainage and the unknown period of drainage within the observational window.

Lake recharge is on the scale of a few years, and it is noticeable that the largest subglacial lake drainage event (Lake 2) subsequently refilled at the fastest rate (∼ 5 m yr⁻¹ uplift), while the smallest drainage event (Lake 3) filled at the slowest rate (∼ 1 m yr⁻¹ uplift). The lakes are at the lower end of the ablation zone and therefore would be expected to be dominated by upstream surface meltwater inputs and the seasonal melt signal. Despite this, lake drainage is not associated with high-melt years (e.g. the 2011 drainage event coincided with a low-melt year). All three lakes exhibited quiescent periods of extended high stand, which might occur when water flow into the lake is balanced by outflow and suggests an external threshold controlling lake drainage initiation.

Although the three subglacial lakes are in close proximity, drainage of an upstream lake does not trigger a cascade of drainage in downstream lakes. In addition, the filling of Lake 3 did not limit recharge of Lake 2 just downstream (Fig. 1b). This suggests that the lakes are not hydraulically well-connected, consistent with subglacial hydraulic routing analysis, which indicates that the main subglacial drainage axis is just to the north of the two upstream subglacial lakes (Fig. 3a). Both the 2014 and 2015 drainage events were initiated in August at a time when drainage system connectivity is envisaged to be high and water preferentially drains towards efficient channels along a hydraulic gradient. Thus, rapid drainage could be a response to lakes infrequently connecting with the main subglacial channel.

Drainage of Subglacial Lake 2 in August 2015 resulted in a net slowdown in downstream ice flow of Isunguata Sermia over a short period of about 1 month, before ice flow subsequently returned to pre-drainage ice-flow speeds in late August (Fig. 2). This slowdown occurs abruptly at the location of Lake 2, suggesting its drainage has impacted ice flow (Fig. 2b, c), but is also coincident with a broader regional slowdown that must at least partly overprint the dynamic signal of the subglacial lake. Although we cannot rule out a short-term speed-up in response to initial drainage, any acceleration must have been extremely rapid (< the 12 d repeat period imagery used to calculate the velocities) and of smaller overall magnitude and/or duration than the subsequent slowdown in ice flow. This response to subglacial lake drainage is the opposite of temporary ice accelerations observed in Antarctica (e.g. Scambos et al., 2011), and is likely because any initial speed-up would have been dampened by efficient drainage of the lake through existing large subglacial channel(s) near the ice margin. Melting of the channel sides and erosion of the channel bed caused by turbulent water flow would have enlarged the channel's cross-sectional area leading to reduced water pressure, likely causing deceleration of overlying ice. This is based on the timing of the event at the end of the melt season, the position of the lake close to the ice margin (Fig. 1), and the large subglacial drainage catchment of Isunguata Sermia (Chu et al., 2016), all of which would act to promote the development of large and efficient channels (e.g. Chandler et al., 2013). This inference is supported by the pattern of thickest sediment deposition at the southern end of the glacier terminus (Fig. 3), which suggests that the subglacial drainage event was at least partially focused into a channel rather than an unconstrained sheet flood.

The August 2015 subglacial lake drainage event flooded the foreland and resulted in substantial net ice-proximal sediment aggradation (7.5×10⁶ m³) of the outwash plain (Fig. 3). Deposition was greatest in the main channel, with up to 8 m of net aggradation close to the outlet decreasing to < 1 m 5 km from the terminus. This near-margin pattern of aggradation is consistent with the geomorphic impact of jökulhlaups observed in Iceland (e.g. Dunning et al., 2013) and demonstrates the potential of episodic subglacial lake drainage events to erode the subglacial bed and modify the proglacial environment. Given the subglacial lake is located just 8 km from the glacier terminus, the subglacial erosion necessary to produce the sediment volume deposited on the foreland is equivalent to a 10 m deep and 100 m wide channel cut into the bed.

The three subglacial lakes beneath Isunguata Sermia do not exhibit the seasonal pattern of winter storage and summer drainage that was previously thought to dominate in the ablation zone (Chu et al., 2016; Bowling et al., 2019). Instead, they share many of the drainage characteristics of the hydrologically active subglacial lakes identified near the EL of the Greenland Ice Sheet (e.g. Palmer et al., 2015; Willis et al., 2015; Bowling et al., 2019). This indicates the potential for multi-year storage of subglacial water towards the margin of the ice sheet in regions dominated by surface meltwater inputs to the bed. This storage could delay the delivery of meltwater to the margin and influence the ice dynamic response to surface meltwater downstream.

Although the presence or absence of sediments in these lakes or the thickness of any water column has yet to be tested, these three subglacial lakes present an extremely accessible target for future geophysical characterization and active lake exploration. Ice-surface elevation changes suggest the lakes are at least 14–30 m deep and have minimum volumes of 3.5–13×10⁶ m³. Ice cover is relatively thin (325–450 m), and the lakes are clustered and in close proximity to the ice margin (< 2 km), road (< 5 km), and key logistical support including a major airport (Kangerlussuaq). Key questions that could be addressed through detailed investigation of these lakes include the following: what triggers subglacial lake drainage and how does drainage evolve downstream? How do the lakes interact with other components of the subglacial drainage system? What geomorphological and sedimentological signatures of similar drainage events might be recorded in the proglacial area?

Further datasets used in this paper can be found in the Appendix.

AS and RS initially identified the subglacial lakes. SL, AS, and DH processed and analysed the data. SL wrote the paper with input and ideas from all co-authors.

The authors declare that they have no conflict of interest.

This work was supported by PhD studentships awarded to Devin Harrison through the IAPETUS Natural Environmental Research Council Doctoral Training Partnership (NE/L002590/1) and Jade Bowling through the ENVISION Natural Environmental Research Council Doctoral Training Partnership (EAA6583/3152).

This paper was edited by Kenichi Matsuoka and reviewed by Katrin Lindbäck and Martin Lüthi.