Surface outburst of a subglacial ood from the Greenland Ice Sheet

20 As Earth’s climate warms, surface melting of the Greenland Ice Sheet is projected to intensify, contributing to rising sea 21 levels 1–4 . Observations 5–7 and theory 8–10 indicate that meltwater generated at the surface of an ice sheet can drain to its bed 22 via crevasses and moulins, where it ﬂows relatively unhindered to the coast. This understanding of the movement of water 23 within, and beneath, ice sheets, underpins theoretical models which are used to make projections of ice sheet change 11 . In 24 this study, we show the ﬁrst evidence of a disruptive drainage pathway in Greenland, whereby a subglacial ﬂood – triggered 25 by a draining subglacial lake – breaks through the ice sheet surface. This unprecedented outburst of water causes fracturing 26 of the ice sheet, and the formation of 25-metre-high ice blocks. These observations reveal a complex, bidirectional coupling 27 between the surface and basal hydrological systems of an ice sheet, which was previously unknown in Greenland. Analysis 28 of over 30 years of satellite imagery conﬁrms that the subglacial lake has drained at least once previously. However, on that 29 occasion the ﬂoodwater failed to breach the ice surface. The two contrasting drainage regimes, coupled with the increased rates 30 of ice melting and thinning that have occurred over the past three decades years, suggest that Arctic climate warming may have 31 facilitated a new, disruptive mode of hydrological drainage on the ice sheet. As such, our observations reveal an emerging and 32 poorly understood phenomenon, which is not currently captured in physical ice sheet models.

its downstream behaviour 40 . For example, following the 2015 drainage of a subglacial lake in southwest Greenland, a ∼25% 61 reduction in ice flow speed was observed 36 . Given projected increases in Arctic atmospheric temperatures, ice melting and 62 run-off during the 21 st century 41, 42 , subglacial lake drainage events may be expected to increase in extent and frequency 32,43 . 63 However, the impact of such extreme forcing upon the Greenland Ice Sheet remains highly uncertain, due to a paucity of  (Figure 2). We interpret this dynamic feature to be the surface signature of a subglacial lake filling, 83 then rapidly draining, similar to events that have been previously observed in Greenland 32-35 , Antarctica 48 and Iceland 49, 50 .

84
Assuming the volume of the feature was equivalent to the volume of water lost during the subglacial lake drainage event, 85 the outburst flood had a total volume of 9 ×10 7 m 3 . This equates to a mean rate of water discharge of 101 m 3 s −1 during the 86 10-day period between satellite acquisitions (22 nd July -1 st August 2014); albeit the drainage duration may have been much 87 3/11 shorter, and the peak discharge higher. Nevertheless, this mean discharge rate is still approximately 2 orders of magnitude greater than that of an Antarctic subglacial lake of the same volume 43 . This newly identified active subglacial lake represents 89 the largest such event recorded beneath the mainland Greenland Ice Sheet, albeit smaller than the 4×10 8 m 3 subglacial lake 90 drainage under the neighbouring Flade Isblink Ice Cap 35 . What is remarkable, and unprecedented for Greenland, is the observed 91 behaviour of the ice sheet downstream of the subglacial lake.

92
As the subglacial lake drained suddenly, a ∼1 km wide rupture with crevasses to a depth of at least 40 m and ice blocks up to 93 25 m in height appeared in the ice surface approximately 1 km downstream of the collapse basin (Figure 1f & 2). Immediately 94 downslope of these ice blocks, a ∼6 km 2 region of the ice surface became scoured clean. Together, these observations indicate 95 that a substantial volume of water had flooded across the ice surface ( Figure 1). Similar to subglacial lake jökulhlaups in 96 Iceland characterised by extremely rapid linear rises in lake discharge, we suggest that a turbulent sheet flood, produced by the 97 subglacial lake drainage, propagated to the surface via englacial routeways and hydrofracturing due to basal water pressures 98 greatly in excess of ice overburden pressure 51-54 . Upstream of the ice blocks there is additional disturbance of the ice surface, 99 with a newly formed ∼1.6 km 2 fan-shaped feature bounded by two raised linear ridges extending ∼800 m in length and up to 5 100 m in height (Figures 1 & 2). We propose that some water was also forced up through a concentric ring fracture at the rim of  This is the first time that such a phenomenon has been observed on the Greenland Ice Sheet, and demonstrates a previously 105 unknown level of complexity and interconnectedness between its surface and basal hydrological systems. In particular, contrary 106 to current understanding of the ice sheet's hydrological system, it provides evidence that water flow is not always unidirectional 107 from the ice sheet surface to its base, but instead can travel from the surface to the bed, and back again, over short spatial and 108 temporal scales. Although this type of behaviour has previously been observed on much smaller, geothermally-active, Icelandic 109 ice caps 53, 54 , it has not, until now, been resolved as a mechanism affecting the larger ice masses of Greenland or Antarctica. past 32 years) occurred at the main glacier terminus, leading to a 500-600 m retreat of the glacier's calving front (Figure 3e-g),

115
(2) an ice-marginal lake broke through its lateral moraine dam and emptied in its entirety (Supplementary Material Figure 3c Figure 3a). Given 117 the destructive nature of the outburst flood, it is likely that at least some of these events may have been connected, although the 118 10-day temporal sampling of the optical satellite imagery makes it impossible to determine the chronology of these events 119 precisely.

120
As the subglacial lake is likely to have been filled by summer supraglacial meltwater input, it is probable that the lake had 121 been gradually filling for some time. As the lake filled, subglacial water pressure would have increased, and we hypothesise have provided an additional rapid injection of water into the system causing a rapid linearly rising discharge event, irrespective 126 of whether it triggered, or was triggered by, the subglacial lake drainage itself. It is also notable that the drainage event occurred                       z(x, y,t, h) =z + a 0 x + a 1 y + a 2 x 2 + a 3 y 2 + a 4 xy + a 5 h + a 6 t We solve for model coefficients using an iterative least-squares fit to minimise the impact of outliers, and discard any 292 unrealistic estimates from poorly constrained solutions using a set of statistical thresholds which include: a minimum of 70 293 data points, a minimum time series length of 2 years, a maximum root mean square difference of 12 m, a maximum elevation 294 rate magnitude of 10 m a −1 , and a maximum surface slope of 5°. 295 We account for temporal variations in range due to changes in radar echo shape using an empirical correction based upon correlated changes in elevation and backscattered power 12 . Using a linear fit, we compute the gradient in elevation as a function of power in order to determine a height correction: which we apply to time series of height change in each grid cell. We compute time series of height evolution by averaging year.

304
Here, we use release 34 GLAS/ICESat Level-2 Global Antarctic and Greenland Ice Sheet Altimetry Data (GLAH12) 14 305 processed with the plane-fitting method 15 to estimate and subtract the local topography from the elevation data, to obtain an 306 estimate of the temporal evolution of the ice surface (see Figure 3d). The GLAH12 data have been pre-processed to remove the 307 intercampaign bias following 16 and the saturation biases as provided in GLAH12.

309
In order to calculate ice thickness change as a result of surface mass balance processes, we calculate cumulative SMB anomalies 310 using daily surface mass balance (SMB) from the 1 km Regional Atmospheric Climate Model RACMO2.3p2 17 to give a mass 311 change, which is then converted to height, assuming that the change occurs at the density of ice (917 kg m 3 ).

312
Nunatak supraglacial lake area 313 The supraglacial lake beneath the nunatak was manually delineated using cloud-free Landsat and Sentinel-2 imagery for the 314 period 1988-2020. Maximum lake area was calculated for each available image.  Before the 1990 subglacial lake drainage event on 16 th June 1990, the oval feature has wind-scoured bare ice with snow surrounding it suggesting it is higher than the surrounding topography and therefore domed. Supraglacial lake (SL1) is filled c After the 1990 drainage event on 1 st August 1990, showing tension fractures caused by downwards motion of ice. The nunatak supraglacial lake has drained d Sedimented meltwater is present in the deepest part of the collapse basin on 3 rd August 1990 e Following a snowfall event, shadowing on the southwest side of the collapse basin indicates that the feature has subsided f Before the 2014 drainage event on 28 th June 2014, some surface meltwater is evident around the rim of the basin and beyond the feature g Landsat scene acquired on 7 th July 2014 showing the supraglacial lake beneath the nunatak has an ice lid h Shadowing on the southwest side of the collapse basin indicates that the surface feature is still collapsed on 15 th August 2014. A fan-like feature with raised ridges occurs between the basin and uplifted ice blocks caused by subglacial outburst of a subglacial lake which leaves behind an outwash plain downstream where the ice is cleaner. Supraglacial lake (SL1) has now drained i Shadowing again is present on the southwest side of the collapse basin, and the fan-like feature is prominent.