Nioghalvfjerdsfjorden, also known as 79∘ North Glacier (henceforth 79∘ N Glacier), is a marine-terminating glacier on the north-eastern coast of Greenland. Approximately 8 % of the Greenland Ice Stream (GIS) drains into 79∘ N Glacier through the North East Greenland Ice Stream (NEGIS), making it the largest discharger of ice in northern Greenland (Mouignot et al., 2015; Mayer et al., 2018). Prior to the 21st century, NEGIS, which extends 600 km into the interior of the GIS (Fig. 1), was believed to be stable, with little change in ice dynamics (Khan et al., 2014; Mayer et al., 2018). However, since 2006 NEGIS has undergone pronounced thinning of 1 myr-1, and the floating tongue of 79∘ N has retreated by 2–3 km since 2009 (Khan et al., 2014). Recently, over 100 km2 of ice was lost through calving of a tributary glacier to 79∘ N Glacier, Spalte Glacier (Fig. S1 in the Supplement), following record-breaking summer air temperatures in 2019 and 2020, highlighting the vulnerability of this region to climate change and surface melt.

The surface of 79∘ N and the NEGIS feature persistent meltwater ponds, or supraglacial lakes (SGLs), and meltwater drainage channels (Fig. S1). SGLs are a frequent summertime feature on many glaciers in Greenland (Pope et al., 2016), on ice shelves (e.g. Larsen C; Luckman et al., 2014), and on sea ice (Perovich et al., 2002). The albedo of SGLs is between 0.1 and 0.6, depending on their depth (Malinka et al., 2018), and therefore they absorb much more shortwave radiation than the surrounding solid ice (Buzzard et al., 2018a). SGLs influence both the surface mass balance (SMB) and the dynamical stability of glaciers by lowering the albedo at the surface and draining water to the base, which reduces friction and influences ice flow velocity (Zwally et al., 2002; Vijay et al., 2019). Both ice velocity increases and decreases have been linked to the drainage of SGLs across Greenland. Short-lived velocity increases have been observed during summer in several marine-terminating glaciers, including 79∘ N Glacier (Rathmann et al., 2017). Both Rathmann et al. (2017) and Vijay et al. (2019) hypothesise that the summer speed-up of 79∘ N Glacier occurs when SGLs drain to the base and alter the subglacial hydrology. Conversely, on land-terminating glaciers, SGL drainage has been shown to reduce ice velocity on the seasonal to decadal timescales (Sundal et al., 2011; Tedstone et al., 2015). SGLs are a key component of the SMB and yet rarely feature in mass balance models or estimates (Smith et al., 2017; Yang et al., 2019). Despite the high number of studies focusing on surface mass loss from the Greenland Ice Sheet (e.g. Lüthje et al., 2006; Das et al., 2008; Tedesco et al., 2012; Stevens et al., 2015), the relationship between SMB, runoff, and SGL development remains unclear.

Despite the widespread occurrence of SGLs, very few studies have investigated the relationship between the seasonal evolution of SGLs and the atmospheric processes required for their formation in this region. Previous studies have largely focused on Antarctic ice shelves (Langley et al., 2016; Arthur et al., 2020; Leeson et al., 2020) and southern and western Greenland (Lüthje et al., 2006; Das et al., 2008; Tedesco et al., 2012; Stevens et al., 2015). Recently, more northerly locations have been investigated, including Petermann Glacier (Macdonald et al., 2018). Multispectral satellite products now provide observations of SGLs over north-eastern Greenland at both high temporal and spatial resolutions, and in many cases free of charge. The north-east of Greenland, and specifically the NEGIS region, has, until recently, lacked such detailed analysis of SGLs; however, this region is likely to show an inland expansion of SGL and ablation zones in the near future (Leeson et al., 2015; Igneczi et al., 2016; Noël et al., 2019). Sundal et al. (2009) used MODIS data to assess the lake area between 2003 and 2007 for 79∘ N Glacier amongst other locations. However, as the ASTER images were acquired at a later stage in the melt season, the percentage of unidentified lake area at the start of the summer is likely to be higher than 12 % (Sundal et al., 2009). Winter estimates of liquid water area on 79∘ N Glacier are also now available from Schröder et al. (2020). Recently, Hochreuther et al. (2021) developed an automated melt detection algorithm for Sentinel-2 satellite data. This provides a near-daily, 10 m spatial resolution time series of SGLs on the NEGIS during summertime.

Widespread summer melting was observed over Greenland in 2007, 2010, and 2012, which prompted research into large-scale teleconnection patterns (Tedesco et al., 2013; Lim et al., 2016; Hanna et al., 2014a). These patterns of atmospheric variability have been found to influence the air temperature and precipitation in Greenland. The North Atlantic Oscillation (often termed NAO) is the dominant mode of variability for Greenland and the Arctic, defined as the “seesaw” of atmospheric surface pressure changes between Iceland and the Azores (Hildebrandsson, 1897; Hanna et al., 2014b). Three other modes of atmospheric variability were found to be important for specifically the north-east and east of Greenland by Lim et al. (2016): the Arctic Oscillation, the East Atlantic pattern, and the Greenland Blocking Index. Generally (for the whole of Greenland), a negative phase of the North Atlantic Oscillation and Arctic Oscillation is associated with a warm and dry atmosphere over the GIS and often leads to mass loss at the surface (Lim et al., 2016). Furthermore, a positive Greenland Blocking Index (especially when combined with a positive East Atlantic pattern and negative North Atlantic Oscillation index) also leads to positive temperature anomalies over the GIS. Extreme Greenland-wide melt seasons, such as in 2012, have been linked to specific teleconnection patterns (Tedesco et al., 2013); however, no studies have assessed the potential role of teleconnections in the development of SGLs or localised melt conditions.

Along with large-scale teleconnection influences, smaller-scale mesoscale processes also influence the climate and melting over Greenland. Recently, atmospheric rivers, or narrow filament-like regions of intense water vapour transport in the atmosphere, have been investigated in response to extreme surface mass balance variations in the north-west of Greenland (Bonne et al., 2015; Mattingly et al., 2018, 2020). In most cases, the north-east of Greenland, especially the coastal regions and marine-terminating glaciers, have received little or no attention during extreme melting years, possibly due to weaker teleconnection signals (Lim et al., 2016) or due to low spatial resolution data (Oltmanns et al., 2019). Similarly, prior to the mid-2010s, the majority of melting was located in the southern and western parts of Greenland, leading to vast research for these regions (e.g. van de Wal et al., 2005, 2012; Tedstone et al., 2017; Kuipers Munneke et al., 2018). However, after the mid-2010s, the highest melt anomalies were located in northern Greenland, especially in 2014 and 2016 (Tedesco et al., 2016). Recently, a low-permeability ice slab was identified in north-eastern Greenland and within 79∘ N Glacier (MacFerrin et al., 2019). The metres-thick, englacial layers of refrozen meltwater enhance melting and runoff processes and are sustained with relatively small amounts of meltwater from drainage of SGLs (MacFerrin et al., 2019). With a warming climate, it is likely that the ice slabs will become more widespread and persistent, although more research is required to investigate the glaciohydrology in these regions. In a recent review paper, Flowers (2018) highlighted that further investigation into surface meltwater volume, drainage, and runoff from marine-terminating glaciers is required.

The specific aims of this study are to investigate (1) the spatial distribution of SGLs over 79∘ N Glacier, (2) the life cycle of lake development, (3) the atmospheric and topographic controls on melt pond evolution in the north-east of Greenland between 2016 and 2019, and (4) whether and how conditions have changed since the Sundal et al. (2009) study in the early 2000s. To accomplish this, we use a combination of Sentinel-2 data, high-resolution (1 km) atmospheric modelling output from the Polar Weather Research and Forecasting (PWRF) model, surface mass balance estimates from the COSIPY model, as well as in situ observations.

In Sect. 2, we introduce the automatic detection algorithm and data used in the study, followed by the results (Sect. 3). These are separated into topographic (Sect. 3.2) and climatic (Sect. 3.3) controls of the SGL formation and spatial distribution. The discussion continues in Sect. 4 and the research concludes in Sect. 5.

Summer 2016 saw the largest loss of glacier area over the GIS since 2012, which was the standout, record-breaking melt year since records began (Hanna et al., 2014a). Summer 2016 also experienced the largest average individual SGL size (0.07 km2), second largest total SGL area and second fastest rate of SGL area growth in our 4-year record. A combination of above-average air temperatures, particularly in mid-June and July, and a large amount of liquid precipitation during summer was likely responsible for the rapid SGL development and peak in total SGL area in late July. Despite the early observation of Ta above freezing in 2017, the earliest in our 4-year period, the June 2017 average Ta was slightly below average. This, combined with the slightly above-average July 2017 temperatures, likely led to the slower rate of increase in SGL area in 2017 compared to 2016 (Fig. 2) and peak in maximum area in early August 2017. The thinner snowpack and limited amount of liquid precipitation falling during summer contributed to the lower maximum SGL area of 153.26 km2 in 2017 compared to 265.39 km2 in 2016. In 2018, the spatial distribution of SGLs was different to the other 3 years, with the largest SGL area at elevations between 300 m and 400 ma.s.l. (Fig. 4). Very few SGLs were observed at elevations greater than 900 m, leading to smaller average individual SGL area, as no larger lakes at higher elevations were identified (Fig. 3). Average individual lake size in 2018 was 0.02 km2 compared to 0.07 km2 in 2016 and 0.06 km2 in 2017 and 2019. A combination of the cooler air temperatures at the start of summer (see Sect. 3.3.1) and thick snowpack led to the delayed onset of SGL development, lower maximum altitude of SGLs and lower total SGL area in 2018 (Fig. 3). Total SGL area was largest in 2019, even though the average size of individual SGLs was the same as in 2017 (0.06 km2). A combination of higher air temperatures, more days above freezing and a smaller snowpack at the start of the melt season all contributed to a significantly higher total SGL area in 2019 (Fig. 4). The peak melt pond area at the start of August 2019 coincides with an air temperature peak of 6.9 ∘C on 2 August at KPC_L, the warmest daily Ta ever recorded here (Fig. 5).

Several years within the last decade have been characterised by high air temperature and extreme melting, including two years within our study period (2016 and 2019). With a projected increase in air temperatures and inland expansion of SGLs into the year 2100 (Leeson et al., 2015; Igneczi et al., 2016), it is important to understand the linkages between different climatic variables and the spatial distribution and temporal evolution of SGLs in the north-east of Greenland. Furthermore, the role of supraglacial melting within the glacial–hydrologic system is in need of further assessment. In a number of studies, enhanced surface melting has contributed to accelerated glacier velocity (Bartholomew et al., 2011; Rathmann et al., 2017); however, in other Greenlandic glaciers, especially land-terminating glaciers, ice velocity has decreased with warmer summers (Sundal et al., 2011; Tedstone et al., 2015).

The spatial spread of the SGLs on 79∘ N Glacier from lower to higher elevations as the melt season progresses is also seen on Leverett Glacier in south-western Greenland (Bartholomew et al., 2011). In south-western Greenland, as the melt season develops, runoff from up-glacier (higher-elevation) regions contributes to subglacial discharge at the base of the land-terminating glacier due to a larger melt area and higher air temperatures (Bartholomew et al., 2011). A similar transport of meltwater from surface to base is hypothesised for 79∘ N Glacier also. Rathmann et al. (2017) observed a seasonal increase in ice velocity following the particularly warm summer of 2016. An extension of the Rathmann et al. (2017) study and estimation of the volume of water potentially interacting with the base of the glacier are important areas of future research.

The rapid increase in SGL area over 79∘ N Glacier during June in most years was also observed at Petermann Glacier at 81∘ N in the north-west of Greenland. Other similarities in SGL characteristics are found between the 79∘ N and Petermann glaciers, including the spatial distribution of the SGLs and the onset of above-freezing air temperatures at the start of June (Macdonald et al., 2018). The only summer with overlap between the current study and the Macdonald et al. (2018) study is 2016. In both locations, this year was characterised by larger total SGL area and warmer-than-average air temperatures, highlighting the relationship between SGL development and climatic factors across the north of Greenland. However, as noted by Macdonald et al. (2018) and observed in the current study in 2018, the low elevation of these regions dictates that, even in cool years, SGLs still form on the lower sections of the glaciers.

Langley et al. (2016) hypothesised that SGL expansion in the early part of the season is particularly rapid, as even small changes in air temperature can increase the total lake area. A rapid increase in lake area was seen at the start of the 2016 and 2019 melt seasons over 79∘ N Glacier; however, in 2017, late-summer temperatures led to later expansion of SGLs. The large rate of increase at the start of summer 2016 (Fig. 2) is likely skewed by the slightly lower temporal resolution in 2016 (approximately 3–7 d) compared to the other years (1–2 d). In 2016 and 2017, there was a lower temporal coverage than the following years as only one Sentinel satellite was in orbit and data quality was poorer (Hochreuther et al., 2021). However, upon visual inspection of the satellite images, 2016 also saw a rapid expansion in SGLs similar to 2019.

Warmer summer air temperatures alone do not always lead to enhanced melting. For Shackleton Ice Shelf in Antarctica, the years with largest SGL area and volume were not always in the same years as the highest summer near-surface air temperatures (Arthur et al., 2020). With only 4 years of data in the present study, no major conclusions can be drawn on this; however, it is clear that precipitation also had an impact on SGL area and development over north-eastern Greenland. In Buzzard (2018b), the relationship between snowfall and melt pond depth was not simple or linear. A small amount of snowfall will promote melt pond development, as there is more water available at the surface; however, a high amount of accumulation can bury the melt pond (especially if the surface has started to freeze towards the end of summer) and reduce melting (Buzzard et al., 2018b). We also see evidence of this non-linear response. A combination of a large amount of snowfall prior to the 2018 melt season and below-average summer air temperatures led to a lower total area of SGLs and positive mass balance over the majority of the glacierised area (Figs. 9 and 10). With a thicker snowpack, it took longer for the SGLs to form, as there was more pore space for water to percolate through before pooling. A thick snowpack was also responsible for a 20 d delay between above-zero air temperatures and runoff in the south-west of Greenland, as the meltwater initially refroze within the cold snowpack (Bartholomew et al., 2011). In the present study, the duration between above-zero air temperatures and melt pond development varies from 7 to 14 d at KPC_L and from 1 to 16 d at KPC_U. The shortest duration was observed in 2019, which had the thinnest snowpack of the 4-year period. Similar conclusions were found for Tibetan glaciers (Mölg et al., 2012) and Shackleton Ice Shelf in the Antarctic (Arthur et al., 2020).

Conversely, the year with the smallest snowfall amount (2018–2019 accumulation season) was not followed by the summer with the fewest melt ponds. However, the much higher air temperatures and late summer freeze-up of SGLs in 2018 played a bigger role. Summer 2016 saw the second-largest total SGL area and spatial distribution of SGLs. This year also saw a large amount of precipitation fall as rainfall in summer. Rainfall is additional liquid for the surface of the glacier, provides heat to the snowpack, and refreezes into solid ice lenses, which preconditions the glacier surface for further SGL development (Machguth et al., 2016). Rainfall associated with summer storms has been linked to extreme melting events in southern Greenland by Oltmanns et al. (2019) and enhanced ice velocity in western Greenland by Doyle et al. (2015). Similarly, Tedesco and Fettweis (2020) concluded that low snow accumulation was also partly responsible for the extensive melting along much of the coast of Greenland in 2019.

Relationships between large-scale temporal and spatial anomalies within the atmosphere, termed teleconnections, have been found to influence the climate and mass balance of Greenland (Tedesco et al., 2013; Lim et al., 2016). With only 4 years of data in the current study, we are unable to draw conclusions about the role of teleconnections in the development of SGLs; however, there is evidence that combinations of teleconnection indices play a role in the near-surface climate and therefore SGL development in the north-east of Greenland. In 2016 and 2019, the average summer (JJA) North Atlantic Oscillation (NAO) index was strongly negative (-1.36 for 2016, -1.23 for 2019) (see Supplement for teleconnection data). Simultaneously, both the summer East Atlantic (EA) index and the Greenland Blocking Index (GBI) were strongly positive in both of these years. In summer 2016, the EA (GBI) summer average was 1.44 (1.73). Similarly, in 2019 the JJA average EA index (GBI) was 1.1 (2.26). This combination of strong negative NAO and strong positive EA also occurred in both summer 2010 and 2012, when extensive melting was observed over the GIS (Lim et al., 2016). In terms of teleconnections, the biggest differences between the 2016/19 and the 2017/18 summers were the NAO and GBI summer indices. In 2017 the NAO index was positive in June and July. In 2018 the summer NAO index was strongly positive (1.74), with all summer months observing a positive NAO signal. The GBI for summers 2017 and 2018 was weakly negative (-0.03) and negative (-0.57) respectively. In terms of the teleconnection indices evaluated here, summer 2017 appears to the be the intermediate or transition year between a particularly strong negative NAO in 2016 and a strong positive NAO in 2018. A decreasing trend in summer NAO since 1981 has been previously identified and is believed to be partly responsible for record-breaking warm temperatures over Greenland in the most recent decade (Hanna et al., 2014b).

The relationship between teleconnections and precipitation is more complicated and is often only significant in the southern part of Greenland, where the majority of the precipitation falls. Bjork et al. (2018) identified a positive relationship between NAO and precipitation in eastern Greenland: there is more precipitation during positive NAO years. The year with the largest cumulative precipitation amounts was the 2017–2018 accumulation season, which was also characterised by a strong positive NAO index. However, the relationship between NAO and precipitation for north-eastern Greenland cannot be assessed with certainty in this study.

Although we present only 4 years of results here and previous studies in this region are sparse, we are confident that SGLs are a persistent feature in the NEGIS and 79∘ N region. Sundal et al. (2009) observed SGLs between 2003 and 2007 using MODIS data. With the availability of Sentinel data (10 m resolution), the SGL areas are less erroneous than previously stated using lower-resolution MODIS data (250 m) (Hochreuther et al., 2021). There is an increase in the maximum altitude of SGL detection between the early 2000s study of Sundal et al. (2009) (1200 ma.s.l.) and the results presented here (1600 ma.s.l.). The lakes at these higher elevations are larger and therefore would have been detected by the MODIS data in the Sundal et al. (2009) study had they been present. Therefore, it is likely that maximum lake altitude has increased over time.

This is not surprising given an increasing air temperature trend of 0.8 ∘Cdecade-1 over 79∘ N Glacier (Turton et al., 2019a) and model suggestions of inland expansion in this area into the 21st century (Ignéczi et al., 2016). Leeson et al. (2015) concluded that maximum lake altitude could reach up to 2221 ma.s.l. with RCP8.5 future projections. Although there are a number of assumptions made in our comparison to Sundal et al. (2009), it is possible that inland expansion of lakes is occurring under increased air temperatures in this region.

Under certain high-melt years, surface rivers have been observed for a number of northern Greenland glaciers, including 79∘ N (Bell et al., 2017). While we do not consider meltwater channels in our analysis and focus only on SGLs, a number of linear features similar to rivers are clearly visible in the Sentinel data (Fig. S1). This highlights that more liquid water is likely present on and within the glacier than discussed here. There is even some evidence of the persistence of liquid water in melt lakes during the winter season on 79∘ N Glacier (Schröder et al., 2020). It is hypothesised that lakes beneath the surface were formed in particularly warm years (such as 2019) and then subsequently covered by a thin ice lens or snow (Schröder et al., 2020).

Estimates of the SGL volume are not provided in this study, which is unusual for these types of studies (e.g. Pope et al., 2016.; Arthur et al., 2020). We hypothesise that SGLs in this region are much deeper than those observed in the west of Greenland (of the order of 0–10 m). Neckel et al. (2020) recorded the depth of an SGL on 79∘ N Glacier, which, at the edge of the lake, had a depth of 10.8 m. The same lake drained suddenly in September 2017, and analysis of the height difference from a full to empty lake using DEMs revealed a subsidence of 50 m in the centre of the lake (Neckel et al., 2020). Therefore, applying the same albedo-depth calculation to the lakes in north-eastern Greenland as in western Greenland would largely underestimate the volumes. In situ observations of these lakes are required to calculate depth and volume with a different albedo-depth coefficient. Fieldwork is planned for this region to observe the depths of SGLs.

In this study we provide a multi-year analysis of the area of SGLs over 79∘ N Glacier (north-eastern Greenland) and investigate the atmospheric and topographic controls of the evolution of the lakes. SGLs have been automatically detected using Sentinel-2 data from 2016 to 2019. The melt detection algorithm implemented here and developed by Hochreuther et al. (2021) is automated, meaning that this work can be continued in the future to analyse a long-term time series of SGL evolution. Our findings would ideally now be expanded to include volume estimates and to model the surface and subglacial hydrology to provide an estimate of the volume of freshwater entering the ocean.

Whilst the SGL location is primarily determined by topographic depressions and the slope of the ice sheet, the occurrence of lakes within these depressions relies on the local meteorology and SMB. Similarly to the spatial distribution, the maximum size of individual lakes is controlled by topography. At higher elevations, larger lakes form due to a lower slope angle (Fig. 4). The larger total SGL areas in 2016 and 2019 were due to lakes developing at higher elevations than in 2017 and 2018, as opposed to individual lakes becoming larger. SGLs refreeze and melt in the same locations above the grounding line each year, but maximum inland expansion of the lakes depends on climatic conditions.

The two key climatic variables controlling the development of the SGLs are air temperature and snowfall. Below-average air temperatures and high snowfall accumulation prior to the melt season of 2018 contributed to reduced lake extent, a reduced amplitude in the seasonal cycle of lake evolution, and late-season freeze-up of the SGLs. These climatic conditions led to a largely positive mass balance at all altitudes except the very lowest-lying regions. Conversely, in the prior two years, surface mass balance was negative for a large portion of 79∘ N Glacier and the surrounding area. Largely this was driven by the above-average air temperature, evident in both the in situ AWS data (Fig. 5) and regional atmospheric modelling output (Figs. 6 and 7). The duration between onset of above-zero air temperatures and SGL development varies between 1 and 16 d, depending on the year and elevation. The snowpack thickness prior to the warm air temperatures likely also has an influence on this duration.

The role of clouds in melt production over the Greenland Ice Sheet is often studied (e.g. Bennartz et al., 2013). Within the 4 years, the warm summer of 2016 coincided with a positive bias in SWin and a reduction in cloud cover; however, the warm June of 2019 was characterised by negative biases in SWin. Similarly, the relatively cool summer of 2018 was characterised by positive anomalies in SWin and higher-than-average cloud cover in June. With just 4 years of data in the current study, no clear conclusions can be drawn about the role of clouds in the development of SGLs in this region.

Whilst 2019 was record breaking in terms of melt over much of the Greenland Ice Sheet, in fact second only to 2012 (Tedesco and Fettweis, 2020), the summer of 2016 was only warm and extreme in the north-eastern region. The extreme summer temperatures led to extensive SGL formation over 79∘ N Glacier as well as subsequent ice velocity acceleration (Rathmann et al., 2017). Similarly, 2019 was not a record-breaking melt year in the north-east of Greenland, and at lower elevations, the number of melt days (TSKmelt) and the duration of melting were less than in other years. This highlights the importance of regional studies of extreme melting as well as Greenland Ice Sheet-wide studies.

There is some evidence of inland expansion of lakes between the Sundal et al. (2009) study, which looked at SGLs between 2003 and 2007, and the present findings from 2016 to 2019. The highest elevation of SGLs in the early 2000s was 1200 ma.s.l., whereas in the late 2010s, SGLs above 1600 ma.s.l. were observed. This is in line with global climate model projections for inland expansion of SGLs and the ablation zone under climate change (Ignéczi et al., 2016). The north-east of Greenland is expected to undergo the largest changes in SMB and SGLs by 2100 and therefore should feature in future atmosphere–glaciohydrology studies.