Evidence of Cascading Subglacial Water Flow at Jutulstraumen Glacier (Antarctica) Derived From Sentinel‐1 and ICESat‐2 Measurements

Migration of subglacial water underneath thick Antarctic ice is difficult to observe directly but is known to influence ice flow dynamics. Here, we analyze a 6‐year time series of displacement maps from differential Sentinel‐1 SAR interferometry (DInSAR) in the upstream region of Jutulstraumen Glacier. Our results reveal short‐term (between 12 days and 1 year) interconnected subsidence‐ and uplift events of the ice surface, which we interpret as a pressure response to the drainage and filling of subglacial lakes. This indicates an episodic cascade‐like water transport with longer quiescent phases in a dynamically stable glacial setting. Abrupt events appear in the DInSAR time series and are confirmed by ICESat‐2 altimetry. The events can be traced for a 1‐year period along a ∼ 175 km flow path. We are able to observe the migration of subglacial water with unprecedented spatial and temporal resolution, providing a new observational baseline to further develop subglacial hydrological models.

• Sentinel-1 DInSAR estimates reveal a ∼175 km flow path of a cascadelike chain of subglacial water propagation over a 1-year period • High-resolution ultra-wideband radar data show no characteristic lake bed reflections and suggests water transport via a local trough system • We find evidence of highly dynamic water transport in a static glacial stetting providing a benchmark dataset to improve modeling results

Supporting Information:
Supporting Information may be found in the online version of this article. may not show temporal variations at the ice surface (Ashmore & Bingham, 2014). However, the detection of active lakes requires observation periods of months to years to coherently map ice surface elevation changes (Siegfried & Fricker, 2018). First studies indicate that the majority of active lakes are located beneath fast flowing ice streams in Antarctica (Fricker et al., 2014;Gray et al., 2005;Malczyk et al., 2020;Siegfried & Fricker, 2018;Smith et al., 2017). Goeller et al. (2016) found evidence for possible subglacial lake locations in Dronning Maud Land (DML) in RES data. However, no active lakes have been reported for central DML so far. Consequently, little is known about the subglacial hydrology, water transport and the impact on local ice dynamics (Thoma et al., 2012). In this study, we derive short-term changes in ice surface displacement lasting between 12 days and 1 year by means of differential SAR interferometry (DInSAR) on Sentinel-1 data. These displacement anomalies reveal a cascade-like pattern starting in the onset region of Jutulstraumen Glacier (JG, Figure 1). Additional ICESat-2 repeat-track measurements capture more gradual elevation changes at the same locations lasting for up to 1 year. However, when integrating the DInSAR displacements over the repeat period of ICESat-2, we find similar results for both datasets. This indicates vertical movement of the ice surface, which we interpret as filling and drainage of subglacial lakes. We further use a dense grid of ultra-wideband (UWB) RES data to obtain detailed information about the bed topography. The combination of these datasets indicates that the subglacial water transport generally follows the hydraulic gradient and migrates downstream in a localized trough system.

Study Site
Jutulstraumen Glacier is the largest ice draining glacier in DML. Ice flows from the polar plateau to the lower coastal section of the East Antarctic Ice Sheet (EAIS) and follows the bearing of the Jutulstraumen Graben through the DML escarpment (Andersen et al., 2020). The JG trough is 40-50 km wide, 1.6 km below present sea level (Fretwell et al., 2013) at its deepest location, and ice flow velocity accelerates to 760 m 1 a E at the grounding line (Mouginot et al., 2019). Our survey area is located at the onset of JG where ice flow is convergent and accelerating from 5 to 100 m 1 a E (Figure 1c). Possible locations for subglacial lakes have been reported in central DML by Goeller et al. (2016). However, their analysis is solely based on radio-echo sounding (RES) surveys and the potential lake locations are restricted to the margins of the JG drainage basin. Ice thickness and bed topography have been extensively mapped in this region (Ferraccioli et al., 2005;Riedel et al., 2012;Steinhage et al., 1999Steinhage et al., , 2001) and indicate a spatially variable and preserved alpine landscape, which has been most likely generated by relief-controlled glacial erosion, sub-aerial weathering and fluvial erosion from mountain glaciers Näslund, 2001).

Data and Methods
In order to detect localized vertical ice-surface displacements we employed satellite borne estimates from Sentinel-1 InSAR and ICESat-2 laser altimetry. To complement our findings, we used airborne radar data and the REMA DEM for high-resolution mapping of the hydropotential. The applied datasets and methods are described in the following.

Ice-Surface Displacements
We applied InSAR processing to Sentinel-1 Interferometric Wide (IW) swath mode data to detect anomalous surface displacements in the satellite's Line Of Sight (LOS). We highlight local anomalies over short time scales by canceling the background ice flow using double-differences as commonly done for grounding line estimates (e.g., Friedl et al., 2020;Joughin et al., 2016;Rignot et al., 2011;Figures 1d -1f). We computed 247 double-differential interferograms between 2015-05 and 2020-09, covering large parts of our study region ( Figure 1b). Steady horizontal displacements were removed from the time series by (a) subtracting two interferograms from a 12-day baseline and (b) calculating the anomaly with respect to an average multi-year interferogram (Text S1 and Figure S1). As the majority of the interferograms showed no bull's-eye fringe patterns, method (b) was more suitable for isolating the vertical displacement and better constrains the timing of the events. Hence, all vertical DInSAR displacement values throughout this manuscript are based on the long-term displacement anomaly with unwrapped results projected from LOS to vertical (e.g., Figures 2d-2f). The fact that most double-differential interferograms showed no anomalous fringe patterns either in the form of distinct bull's-eyes nor in the downstream direction of the drainage events is picked up later to discriminate potential large-scale changes in horizontal ice flow velocities. In addition, we used data from the Advanced Topographic Laser Altimeter System carried on board the Ice, Cloud and land Elevation Satellite-2 (ICESat-2) to find elevation anomalies over the DInSAR-detected events. We employed level 3A Land Ice Height (ATL06) Version 3 data , which are available for the time period starting on October 14, 2018 . We modified the repeat-track analysis approach introduced by Fricker et al. (2014) to address the data acquisition characteristics specific to ICESat-2 (six beams instead of one nadir beam; Text S2). Based on this analysis, we were able to detect elevation changes over the locations of the DInSAR-detected bull's-eye fringe patterns (Figures 2g and 2h). In this study, we used ICESat-2 data with the Reference Ground Track (RGT) 0732 and Ground Track (GT) GT3l as well as RGT 1235 and GT1r (see Text S2 for details).

Hydropotential Mapping
The DInSAR-detected bull's-eye fringe patterns in Figures 1d -1f are also covered by multiple airborne radar data profiles. The data were acquired at the onset region of JG during the austral summer of 2018/19, using a multichannel ultra-wide band (UWB) radar system operated by the Alfred Wegener Institute We provide a description of the specific acquisition geometry and corresponding radar processing steps (Text S3) and refer to Hale et al. (2016) and Rodriguez-Morales et al. (2014) for further system specifics. The radar data cover areas that are not covered in Antarctic-wide ice thickness and bed topography maps (Fretwell et al., 2013;Morlighem et al., 2020), and we include this additional information to resolve the bed in our area of interest in finer detail (see also Franke et al., 2021, for further details on the data and methods used for the refined bed topography). The grid size of our refined bed topography is 1 km. to RGT 1235 GT1r. Elevation anomalies (difference to the mean) for multiple ICESat-2 repeat passes are shown in (g) and (h). The cumulated DInSAR anomalies between the time periods indicated by the arrows are shown in the lower part of (g) and (h) together with the difference of the time-consistent ICESat-2 anomalies (black curve).
For mapping preferential water flow paths, we used the locally improved bed topography and the Reference Elevation Model of Antarctica (REMA, Howat et al., 2019) to estimate the glaciological hydraulic potential in our study area (e.g., Shreve, 1972;Smith et al., 2017). A flow accumulation grid was generated on the basis of the depression filled refined bed topography using the algorithm proposed by Tarboton (1997) (Text S4). In this approach, flow direction is defined from the steepest downward slope in the hydraulic potential of each pixel's eight triangular facets. The output is displayed as the number of up-slope grid cells (Figures 2d -2f).
in the hydraulic potential ( Figure S8a). Most events are located over V-shaped valleys (see Figures S5, S6, and S8), the depth of which constantly increases from (hydrologically) upstream to downstream.

Discussion
Building on previous large-scale studies based on satellite altimetry (e.g., Fricker et al., 2007;Siegfried & Fricker, 2018, the combination of Sentinel-1 and ICESat-2 measurements offers new insights into the inter-connectivity of subglacial lake drainage events. Here we present a highly resolved chain-like pattern stretching from the onset of JG toward the grounding line (Figures 3a and 3b, Movie S1). The setting of JG is such that the subglacial water generated in the large upstream catchment must be funneled through a comparatively narrow constriction starting at the ice-stream onset. This favors the development of efficient channelized drainage systems similar to what has been observed at other ice streams (e.g., at Recovery Ice Stream, Dow et al., 2018). Our observations indicate that this drainage is not steady but contains episodic events of locally increased water storage resulting in negative effective pressure manifested in localized surface uplift. After sufficient build-up, the pressure then abruptly changes in a way that facilitates more . The complete time series is available in Figure S2 and Movie S1. Panel (b) shows the initial uplift events for each location. The propagation of subglacial water along the bed is indicated with a blue dashed line as well as the filling of local sinks. The pressure-related uplift is indicated with upright arrows with the respective color for the timing. The length of the arrow is proportional to the maximum uplift. The DInSAR-detected uplift at the ice surface is indicated with a blue outline. Panel (c) indicates the temporal sequence of filling and drainage of two adjacent sinks and the respective response at the ice surface.
10.1029/2021GL094472 7 of 10 downstream transport of water to a different low in the hydraulic potential where this process is then reiterated (Figures 3b and 3c). These observations are in line with previous modeling based assertions (Dow et al., 2018) and observational studies at Recovery (Fricker et al., 2014) and Thwaites Glacier (Hoffman et al., 2020;Smith et al., 2017).
The proposed mechanism indicates that water is efficiently transported via transient channels. Whether these channels are Röthlisberger channels that melt into the ice (Dow et al., 2018) or subglacial canals incised into the sediments (Carter et al., 2017) remains open to discussion. The idea of a channelized drainage system with efficient water transport is further supported by the observed lack of large-scale changes in ice flow dynamics after the drainage events ( Figure S4). However, here we have to admit that the geometry of the available SAR acquisitions is rather unfavorable for mapping ice velocities as the main ice flow direction is close to the azimuth direction of the satellite. Therefore, these estimates are restricted to the sensitivity of the SAR sensor.
Instead of following the prevailing ice flow direction, the chain of interconnected DInSAR anomalies largely follows the hydraulic gradient (Figure 3a). Small deviations from the hydraulic gradient can be attributed to several reasons: (1) uncertainties in ice surface elevation or bed topography ( Figure S7b), (2) the assumption that basal water pressure equals the ice overburden pressure might not hold true in all places, (3) interpolation artifacts in the input datasets influence the accuracy of the derived hydraulic potential (e.g., data gaps in the ice thickness data, Figures S7a and S8b), (4) algorithm artifacts caused by the choice of the flow routing algorithm (Desmet & Govers, 1996), and (5) the sink filling algorithm might flatten some regions and hence can also impact the subglacial routing pathways.
In comparison to the lakes found in this study, most active lakes so far detected in Antarctica are clustered in regions of elevated ice flow velocity and are subject to larger lake volume changes (Smith et al., 2009;Siegfried & Fricker, 2018. The latter is also reflected in the timing and duration of the drainage events. During the 2017/2018 chain of events the duration of the drainages varied between 12 days (Sentinel-1 repeat pass time, event 1 E E ) and  E 5 months (events 2 E E and 1 E D ) along its 175 km flowpath. The cascading chain of events is therefore not linear but dependent on basin size with the largest basin taking up to 2 months (event 2 E E ) of filling before draining. The 2019/2020 drainage events can not be followed as far downstream as during the 2017/2018 chain of events. In 2019/2020, the longest period of lake drainage was found for lake 1 E F , which showed a drainage duration of  E 1 year, which is in accordance with both DInSAR and ICE-Sat-2 estimates (Figure 2g). The duration of the drainage events is significantly shorter than reported from ICESat measurements elsewhere in Antarctica (Fricker & Scambos, 2009;Fricker et al., 2007). However, also lake size and drainage magnitudes are smaller than in the above studies.
To our knowledge, this is the first study, which finds active lakes in central DML. However, this might be attributed to the fact that compared to the altimetry measurements of previous studies, InSAR is not restricted to sparse repeat tracks and is capable of detecting surface elevation changes at the wavelength scale of the sensor. This makes DInSAR sensitive to small drainage events, which are possibly not resolved in altimetry based estimates. Such highly resolved DInSAR time series both in space and time will also be of great value for other regions in Antarctica where indirect measurements of small scale elevation changes are missing so far, but will improve our knowledge of local drainage systems and are of great value to inform modeling studies (e.g., Napoleoni et al., 2020).
Many locations where active subglacial lakes have been identified using satellite altimetry have also been surveyed by airborne RES campaigns (e.g., Christianson et al., 2012;Humbert et al., 2018;Siegert et al., 2014;Wright et al., 2012), which did not identify any characteristic [strong, flat and specular] bed reflection at the lake sites (Carter et al., 2007). Such reflections are also absent in our study region and sometimes the bed reflection is absent completely, which may be due to several factors such as high englacial attenuation rates, which reduce the receiving signal power. Since englacial attenuation is a function of temperature, it is possible that warm ice in the troughs underneath our active lake catchments or a temperate layer of basal ice might be the cause for the reduction in, and in some cases total absence of, basal reflectivity (Humbert et al., 2018;Siegert et al., 2014). Christianson et al. (2012) and Siegert et al. (2014) argue that, in order to create a dielectric contrast of 10-20 dB higher than the surrounding material, a minimum water table of several meters' thickness is required. However, our centimeter-scale elevation changes combined with the odds of a radar survey coinciding with a filled lake (Siegert et al., 2014), strongly indicate that RES-based detection of subglacial water is very unlikely in our survey area (see Figures S5 and S6). This agrees with the lack of evidence for a deep water lake at JG. Nonetheless, it may still be possible that larger-than-anticipated amounts of subglacial water are being transported. Carter et al. (2017) showed that subglacial lakes in Antarctica could drain through sediment-floored canals, which might serve as a further explanation for the missing evidence in the RES data.

Summary and Outlook
We identified a subglacial hydrologic network at Jutulstraumen Glacier in which subglacial water is periodically trapped and released. We are able to trace the propagation of subglacial water over a distance of  E 175 km within 1 year using a combination of different remote sensing methods. DInSAR estimates from Sentinel-1 data reveal abrupt but localized events occurring between 12 days and 1 year. Individual events are interconnected in a cascade-like pattern of short-term surface uplift and subsidence following lows in the hydraulic potential indicating an episodic transport of excess water across subglacial lakes. ICESat-2 repeat-tracks capture the long-term lake drainage patterns and match with the cumulated DInSAR displacements, but undersample the short term dynamics. Additional airborne radar data constrains the hydraulic potential and show that the inter-connected chain of subsidence and uplift events occurs along projected pathways of subglacial water flux. The subglacial water is not apparent in the bed reflection amplitudes either due to its transient nature or because of a lack in system sensitivity. Despite the widely discussed influence of subglacial water on local ice flow variability, we find no evidence of associated changes in downstream ice velocities from the DInSAR dataset. The lake areas and surface displacement magnitudes found in this study are relatively small and could only be detected due to the high spatial and temporal resolution of the Sentinel-1 DInSAR estimates. This enabled us to find evidence of highly dynamic water transport in a rather static glacial setting. Such findings might now also be possible for other slow moving areas in Antarctica where no water movement was detected before. Our results contribute to the understanding of the subglacial hydrology of Antarctica in regions with small scale water movements, where observations were not possible so far. This knowledge gain will be valuable for improving hydrological models and is needed to capture the entire spectrum of hydrological processes.