Glacial lake outburst floods (GLOFs) or
Ice-dammed lakes can form either in supraglacial, subglacial, or ice-marginal
positions (Tweed and Russell, 1999). Globally, proglacial lakes (including
ice-marginal lakes) contain up to 0.43 mm of sea level equivalent (Shugar et
al., 2020), and recent studies show that ice-marginal lakes in Greenland have
increased in both number and size (Carrivick and Quincey, 2014; Shugar et
al., 2020). Currently, there are more than 3300 ice-marginal lakes in
Greenland. These are predominately found around peripheral mountain glaciers
and ice caps (PGICs) as well as along the southwest Greenland ice sheet
(GrIS) margin (Carrivick et al., 2022; How et al., 2021). The outflow of
ice-dammed lakes can vary substantially from a gradual near-steady discharge
to sudden outburst floods called
Carrivick and Tweed (2019) review the status of knowledge on GLOFs and ice-dammed lake drainages in Greenland and show that continuous multidecadal observations of transient lake water levels (i.e. pre- and post-drainage), lake drainage dates, and released flood volumes are extremely rare. Nevertheless, such time series are important for revealing spatio-temporal patterns in lake drainage and the timing and magnitude of flood events. Furthermore, long-term data improve our understanding of drainage triggers and mechanisms, provide important context for the scale and frequency of current and future GLOFs, and aid in the mitigation of downstream effects. The primary aim of this paper is to (re)calculate and analyse the lake water level and drainage volume of 14 historical GLOFs observed from 2007 to 2021. Secondly, we investigate geomorphological changes supporting a shift in the proglacial GLOF drainage route observed following the recent GLOF on 22 August 2021.
One of the most intensively monitored and widely studied ice-dammed lakes in
Greenland is located on the northern flank of Russell Glacier in West
Greenland (Fig. 1) (Carrivick et al., 2017; Lamsters et al., 2020;
Mikkelsen et al., 2013; Russell, 1989, 2007; Russell et al., 2011), and so it
is a key site for understanding GLOF behaviour. The lake is
Fieldwork at Russell Glacier was carried out between 3 and 6 September 2021, 2 weeks after a GLOF on 22 August 2021. Two uncrewed-aerial-vehicle (UAV) missions were undertaken to produce DEMs and orthophotos of the drained lake basin topography as well as the flood drainage route (Fig. 1). As the lake did not fully drain, we were unable to survey the entire lake topography; however, a standing water level of 408.8 m was surveyed in the lake, which is almost identical to the minimum lake levels observed after other previous GLOF events (Russell et al., 2011). Russell et al. (2011) produced a DEM of the lake basin bathymetry from interpolation of kinematic dGPS tracks surveyed in February 2008, finding a minimum elevation of 410 m. In this study, our UAV surveys enabled a highly accurate and high-resolution DEM without surface interpolation. From this DEM, we are able to precisely estimate the pre- and post-GLOF water level, the lake area, and the likely drainage volume of both historical and future events. All elevations are reported as height above the WGS84 ellipsoid, unless otherwise stated.
The UAV flights were conducted on two different dates using two different UAVs, due to the battery capacity and weather conditions (Table 1, Fig. 1):
Overview of the two UAV missions.
Both UAVs have direct georeferencing capabilities provided by an onboard
GNSS receiver (Table 1), which records the positional data of each image as
it is captured. To achieve centimetre-level accuracy in both the vertical
and horizontal direction of the camera positions, we kinematically
post-processed the positional data from the UAV GNSS receivers. Compared to
real-time-kinematic (RTK) correction, post-processed-kinematic (PPK)
positioning is considered more accurate and does not depend on a reliable
real-time connection to a GNSS base station (Chudley et al., 2019). The UAV GNSS data were post-processed using WingtraHub (v. 2.2.0) and KlauPPK (v. 7.17) software
relative to the fixed Greenland GPS Network (GNET) base station, located in
Kangerlussuaq (KLSQ) approximately 30 km from the field site (Bevis et al., 2012).
The processed camera position for both UAV surveys had a vertical and
horizontal accuracy of
For the purpose of validating the accuracy of the produced DEMs, we placed a
combined total of 33, 0.3
The UAV images were processed using a structure-from-motion (SfM) workflow
in Agisoft Metashape Pro (v. 1.7.4). We follow the general processing workflow
described in the official Agisoft guidelines (Agisoft LLC, 2020). The camera
calibration was set as “precalibrated”, and the calibration parameters were set
according to the calibration report of the camera used. Instead of GCPs, we
used the post-processed, geolocated camera positions to georeference the
point cloud. During the bundle adjustment, we performed a refined camera
calibration, which is recommended when other variables are well constrained
(Chudley et al., 2019). DEMs and orthomosaics for Mission I were then
exported at resolutions of 0.1 and 0.04 m, respectively, while for Mission
II both were exported at a resolution of 0.1 m (Table 1). The RMSE of the
Previous studies, using a similar setup and approach (Chudley et al., 2019;
Jouvet et al., 2019) reported horizontal and vertical uncertainties in the
range of 0.1–0.4 m, without the use of GCPs. By measuring the horizontal
and vertical displacement between the two fix solution GCPs and their
observed location in the Mission II orthomosaic and DEM, we estimated the
accuracy to be 0.14 and 0.35 m, respectively. Due to a lack of reliable
GCPs, we applied an additional method for determining the uncertainty.
Inspired by similar studies (Chudley et al., 2019; Jouvet et al., 2019), we
estimated the uncertainty by calculating the relative offset between the
Mission I and Mission II DEM over stable bedrock, assuming no change in the
topography. We applied the Python module PyBob (McNabb, 2019) based on the
co-registration method developed by Nuth and Kääb (2011), which
determines the
Due to image gaps at the western part of the lake, we were not able to
produce a complete UAV-derived DEM of the drained lake topography. Thus, the
missing regions were filled with elevation data from two ArcticDEM strips
acquired on 19 September 2014 and 2 August 2015,
respectively (Fig. 1). The ArcticDEM strips have a resolution of 2
Using the co-registered DEM mosaic, as well as the Mission II orthomosaic,
we digitized lake area and extracted elevation points every 5 m along the
digitized eastern lake margin to estimate a water level of 408.8 m
To estimate the lake water level at different temporal intervals we used
satellite images from PlanetScope, Landsat 7 and 8, and Sentinel-2. The
satellite images were manually georeferenced to the high-resolution UAV
orthophotos to adjust for small offsets. Inspired by the approach of
previous studies (e.g. Carrivick and Tweed, 2019) the pre- and post-drainage
water level was determined by manually placing 30 points along the
All drainage estimates from 2017–2021 are based on PlanetScope satellite images, whereas estimates of previous events are based on mainly panchromatic images from Landsat 7 and 8 with a resolution of 15 m as well as RGB images from Sentinel-2 with a resolution of 10 m (Table 2). In contrast to the relatively coarse (10 and 16 d) temporal coverage of the Landsat and Sentinel images, Planet images have a much finer spatial resolution of 3 m and a temporal resolution of approximately 1 d (Planet Team, 2017). This enables the detection of short-term changes in water level, albeit during clear-sky conditions.
Based on the Mission I DEM we determined the main surface drainage routes
for the 2021 GLOF event from the glacial drainage outlet (i) to the outlet
lakes and (ii) across the ice margin. The drainages routes were calculated
as the paths of least resistance from the source (drainage outlet) to the
locations (i) and (ii), assuming that water is flowing to the neighbouring
pixel with the lowest elevation. The calculations were based on a 2
We estimated the drainage volume from a gauging station deployed in Watson River at Kangerlussuaq, 27 km downstream of the lake (van As et al., 2017). Here, pressure transducers record changes in water pressure, which subsequently is corrected for atmospheric pressure before being converted into hourly averages in water level. Water discharge was then obtained using a rating curve, based on discharge measurements at various water levels (van As et al., 2017), and is associated with a conservative uncertainty value of 15 %. Due to diurnal fluctuation in discharge, we estimated the daily minima and maxima on the day of the drainage event by fitting a linear trend through the equivalent low and high stage values on the day before and after. This allowed estimates of the baseflow and thus estimates of the volume associated with lake drainage to be made.
Air temperature data were obtained from the KAN_L automatic weather station, part of the PROMICE automated weather station (AWS) network, which is located on the ice sheet at 670 m a.s.l., 18.5 km from the study site. We use hourly average data that are based on measurements recorded every 10 min (Fausto et al., 2021; GEUS Dataverse, 2022). For each of the analysed periods, the air temperature data contained no missing values.
Figure 2 illustrates how lake volume and area change with variations in
water level, as calculated based on the 2021 post-drainage DEM. The lake has
a theoretical maximum water level of 433 m, after which water overspills the
ice dam, hereby indicating the elevation of the damming glacier. The 2021
theoretical water level maximum produces a lake surface area of 0.79 km
Lake area (km
Since the lake entered its new drainage cycle in 2007, we have observed annually
reoccurring events, with the exception of 2009. The 2007 GLOF had the
largest observed drainage volume, with a value of 37.73
Drainage dates, pre- and post-drainage water levels, lake areas, and volumes for 15 GLOFs spanning 1987 to 2021. The 1987 estimates are adopted from Russell et al. (2011). The 2007-to-2021 estimates are reconstructed using the 2021 post-drainage DEM in combination with selected optical satellite images as well as from downstream hydrograph observations. The table includes references to previous studies of the lake, including estimated drainage volumes.
Pre-drainage water level, drainage volume, and drainage day of year
(DOY) for 14 GLOFs spanning 2007 to 2021. Pre-drainage water levels (blue
circles) are estimated using the 2021 post-drainage DEM. Drainage volumes
are estimated using both the 2021 post-drainage DEM (red triangles) and
downstream hydrograph observations (green squares). The grey bars indicate
the day of year (DOY) of the drainage and refer to the leftmost
Figure 4a and b illustrate the two main routes of drainage for the GLOF event to exit the drainage outlet. Drainage route I channels the water into an ephemeral river channel and into two outlet lakes connected to the downstream river network. In contrast, in drainage route II the water flows across the ice margin and into an ice-marginal meltwater drainage system before reaching the river network, thus bypassing the two outlet lakes. There is a 0.4 m elevation difference between the drainage threshold of drainage route I (390.2 m) and drainage route II (390.6 m) (Fig. 4b).
The high resolution of the orthomosaic and DEM produced through UAV Mission
I has enabled us to observe a number of important geomorphologic features
across the drainage region which are not visible in the 3 m resolution Planet
imagery. For example, large blocks of ice up to 5 m in length are observed
scattered across both drainage route I and II (Fig. 4c and d). On the
western part of the ice margin we observe a
For all GLOFs except 2014, the DEM- and hydrograph-based methods produce
volume estimates that are within each other's margins of error (Table 2,
Fig. 3) with a total mean difference of 10 %. This indicates that the
two methods used to obtain drainage volumes can serve as independent
validation for one another. For the 2014 GLOF the hydrograph estimate is 2 times larger than the DEM-derived volume. This could partly be
because the cloud-free Landsat 8 image captured closest to the drainage date
on 3 August was acquired 13 d prior to the GLOF, on 21 July. Using the max 2010-inflow rate of 1.3 m
When the lake drains below the 2021 post-drainage DEM reference elevation of
408.8 m (2007, 2008, 2011–2015, 2018, 2020, 2021), we underestimate the
volume release, as we cannot measure the precise post-drainage water level.
As annual differences in the post-drainage area are minimal (Table 2), the
changes in volume are also expected to be limited. Additionally, the total
lake area during these instances is at its minimum. Russell et al. (2011)
reported the post-drainage water level of the 2007 event to be 404.5 m,
which is 3.34 m lower than our 2021 reference minimum. Assuming that the
entire 2007 post-drain area (Table 2) is lowered by an additional 3.34 m, it
would give an extra volume release of 1.59 Mm
With the documentation of seven new GLOFs and the recalculation of seven known GLOFs, we are now able to re-evaluate the proposed drainage-triggering mechanisms. Previous studies have suggested several different mechanisms that control GLOFs at Russell Glacier, such as flotation of the ice dam (Carrivick et al., 2017), fluctuation in subglacial meltwater (Russell and de Jong, 1988; Russell, 1989), incomplete resealing of the subglacial conduit (Russell et al., 2011), and subglacial drainage through an incised bedrock-walled Nye channel (Russell et al., 2011). Recent data from ground-penetrating radar surveys, however, revealed no evidence of a Nye channel incised into the bedrock but instead found evidence of at least one englacial tunnel running parallel to the ice margin (Lamsters et al., 2020).
Plot of hourly temperature measurements from 10 d prior to the
drainage event from KAN_L. The green line shows the mean air
temperature (MAT) 41–10 d prior to the drainage, the red line shows the MAT 10–0 d prior, and the orange line shows the MAT 5–0 d prior to drainage. The red circle
denotes the start of the GLOF, and for the 2018 event the circle is larger
due to uncertainty about the timing. All plots share the same
Had the lake been draining due to floatation of the ice dam, we would expect to see a gradual decrease in the release volume and pre-drainage water level as less water is required to float the thinning ice dam. We do observe a lower drainage volume compared to the 2007 and 2010 maximum, but the lake is still able to drain at both similar and higher water levels than observed in 2008 (Fig. 3, Table 2). The two largest GLOFs (i.e. 2007 and 2010) both occurred following a year of no drainage, and they indicate that in order for the lake to reach such a high-water level an additional (or multiple) melt season is required. However, due to thinning of the damming glacier the lake is unable to reach its previous peak drainage water level and volume that was observed in 2007 and 2010. As a result, and based on its current configuration, the lake can only reach a maximum water level of 433 m, at which point it overspills the ice dam.
Russell (1989) suggested the internal drainage network of Russell Glacier
and a possible reduction in (sub)glacial meltwater as the main trigger for
the 1984 and 1987 GLOFs. This closely aligns with the majority of the
observed GLOFs occurring late in the melt season when subglacial and englacial
water pressure is lower. However, the partial drainage events of 2014, 2015,
2017, and 2020 occur earlier in the melt season, indicating a different
drainage mechanism or an additional means by which to lower the water
pressure. The water pressure can also be lowered as a consequence of a
sudden reduction in meltwater production (Tweed and Russell, 1999; Russell
et al., 2011). Russell et al. (2011) suggested a link to an observed drop in
air temperature prior to the 2007 and 2008 GLOFs. For 7 of the 12 GLOFs that
occurred between 2010–2021 (2010–2011 and 2015–2019), we observe a similar
drop in mean air temperature (MAT) when comparing the MAT of the 10 d
prior to the GLOF with the MAT of the month prior (Fig. 5). The difference
ranges from
The fluctuation between short periods of relatively high and low drainage
volumes (Fig. 3, Table 2) suggests other factors may influence the
triggering threshold. The partial 0.9 Mm
Previous observations of the lake drainage system (e.g. Carrivick et al., 2018; Mernild and Hasholt, 2009; Russell, 1989) coincide with the estimated location of drainage route I. In this study, the scattered ice blocks and fractured ice surface observed in Fig. 4c, d, and f indicate a considerable flow of surface water along drainage route I, as well as the new route II during the 2021 GLOF. The roughly circular and nearly vertical holes, exemplified in Fig. 4e, are likely created by the collapse of the ice surface above an empty englacial or subglacial cavity. They may also be a result of pressurized englacial or subglacial water flow being forced upwards and breaching the ice surface, causing a localized collapse. There are multiple potential explanations for the parallel fractures observed on the ice margin (Fig. 4g and h), such as a propagation of basal crevasses towards the surface, stretching of the ice surface from increased basal sliding, and a temporary uplift and/or (subsequent) falling of the ice surface. In combination with the additional observed surface features, we consider the latter hypothesis the most plausible; however, all explanations can be linked to a subglacial or englacial flow of drainage water.
From 3 m resolution Planet satellite images (Planet Team, 2017) captured immediately before and after the 2021 GLOF, we also observe geomorphological changes along the ice-marginal meltwater drainage system which channels the GLOF drainage water from drainage route II into the downstream river network. As a results of this observation, we reanalysed previous drainage events. The reanalysis showed no evidence of geomorphic change along the ice-marginal meltwater drainage system after the 2020 and 2019 events. However, we did observe standing water on the ice margin and changes in the ice colour (black to white) after the 2019 drainage, indicating water flow on the ice surface.
On the basis of these observations, we hypothesize that the new drainage pattern is predominantly caused by the thinning and retreat of the ice margin in the vicinity of the outlet, allowing floodwater to more easily run over and into the ice margin. The 0.4 m elevation difference between drainage route I and II (Fig. 4b) suggests that route I is still the primary path. However, as the ice margin gradually thins, drainage route II will likely become the dominant path taken. This shift is very profound, because it bypasses the two outlet lakes (Fig. 1) that currently act as a buffer and slow the downstream flow of water. Thus, this shift will affect downstream geomorphology and potentially cause hazards to local infrastructure. Therefore, we strongly suggest that a comprehensive investigation of the potential downstream consequences of GLOFs along the new route is undertaken.
This study presents one of the longest and continuous known records of GLOF
drainage estimates in Greenland. We (re)analyse 14 GLOFs spanning 2007 to
2021 to provide a new evaluation and a greater understanding of the drainage
patterns and trigger mechanisms. Our time series reveal annual GLOFs, with
the exception of 2009, and considerable variations in both the date of
drainage, ranging from 31 May to 15 September, and the
overall volume, ranging from 0.9 to 37.7 Mm
We hypothesize that when the ice-dammed lake episodically drains, it does so through an englacial tunnel created by the 2007 GLOF. In contrast, the ensuing annual drainages are likely caused by a syphoning drainage mechanism within the pre-existing englacial conduit. This syphoning is likely triggered by a reduction in meltwater, driven by late-season drainage and sudden reductions in mean air temperature, as well as annual variations in the configuration of the drainage system of the damming glacier. The observed fluctuations between short periods of relatively high and low drainage volumes suggest that the large GLOFs potentially weaken the ice dam, causing it not to seal during winter and thus allowing the following event(s) to drain at a lower water level.
This study also reports geomorphological evidence from UAV and satellite data that reveals an altering of the proglacial drainage route with a new subglacial or englacial flow pathway, as well as the supraglacial flow of drainage water across the ice margin. We suggest that the new drainage route has developed as a result of thinning and retreat of the ice margin and that further thinning will cause the new drainage route to eventually become dominant. As the new route bypasses the two buffering outlet lakes, the delivery of drainage water to the downstream system will be faster and less attenuated, with significant consequences for the surrounding geomorphology and the potential risk of flooding hazards.
Contour map with 5 m intervals based on the 2021 post-drainage DEM. Background is a four-band Planet Team (2017) acquisition from 23 August 2021.
Position of ice margin digitized from satellite images. Background is a four-band Planet Team (2017) acquisition from 23 August 2021.
DEMs and orthophotos of the drained lake topography and outlet drainage
route are publicly available via figshare
(
MD led the data analysis and wrote the main manuscript. MD, FH, and AAB planned the study and carried out the fieldwork. KKK collected the hydrograph data and performed the hydrograph volume estimates. SAK carried out GPS data processing. JLC provided guidance on the interpretations and assessment of the drainage triggers and water rerouting. All authors contributed to the data analysis and interpretation of results and provided inputs for the manuscript.
The contact author has declared that none of the authors has any competing interests.
Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was funded by the Villum Foundation (Villum Young Investigator grant no. 29456).
This paper was edited by Kang Yang and reviewed by Nathaniel Baurley, Brianna Rick, and one anonymous referee.