Ocean forcing of the Greenland Ice Sheet: Calving fronts and patterns of retreat identified by automatic satellite monitoring of eastern outlet glaciers



[1] We have developed an automatic method to identify changes in the position of calving glacier margins using daily MODIS imagery. Application of the method to 32 ocean-terminating glaciers in East Greenland produced 26,802 margin positions for a 10 year long period (2000–2009). We report these high-resolution data and show that the glaciers exhibit seasonal cycles with magnitudes of advance and retreat proportional to glacier width. Despite similar seasonality there is a distinct difference between the interannual trends of calving front positions north and south of 69°N. All glaciers above this latitude showed very limited or no change when seasonality was excluded, while glaciers south of 69°N retreated significantly between 2001 and 2005 (∼2.3 km on average). Approximately 26% of the retreat of southern glaciers was regained by readvance from 2005 to 2009. To explain the latitudinal boundary of glacier dynamics, we review basic climatic factors, including summer and winter atmospheric forcing, sea ice conditions, and ocean temperature. We conclude that the southern retreats were strongly influenced by warm oceanic conditions associated with increased transport of subtropical waters to the Irminger Sea and to fjords and coastal regions south of 69°N. Northern glaciers remained stable despite significant increase in runoff in this region because fjords at latitudes higher than 69°N are less exposed to subtropical waters. The southern retreats illustrate sensitive behavior of calving glaciers, and we hypothesize that the calving fronts retreated because they were exposed to rapid ice-front melting.

1. Introduction

[2] The high rates of thinning observed along the coastal margin of the Greenland Ice Sheet clearly show that its mass balance is strongly influenced by the flow of marine-terminating outlet glaciers [Krabill et al., 1999, 2000; Abdalati et al., 2001; Krabill et al., 2004; Thomas et al., 2003; Rignot et al., 2004; Pritchard et al., 2009]. Rignot and Kanagaratnam [2006] quantified ice losses in individual drainage basins and estimated that a third of the net annual mass loss of approximately 220 km3 in 2005 occurred in the catchments of the three largest glaciers (Jakobshavn Isbræ and Kangerdlugssuaq and Helheim glaciers). Howat et al. [2007, 2008] estimated net mass losses from the discharge of ice by Kangerdlugssuaq and Helheim glaciers to be 16 and 14 km3 yr−1, respectively, in 2002–05, while numerous smaller glaciers farther south collectively lost 50 km3 yr−1.

[3] The synchronous speed-up of glaciers between 2001 and 2005 was at first interpreted to be a response to atmospheric warming [Rignot and Kanagaratnam, 2006], which has been significant in southeast Greenland since 1995 [Chylek et al., 2006; Box et al., 2009]. Kangerdlugssuaq and Helheim glaciers did, however, slow down after 2005 [Howat et al., 2007; Joughin et al., 2008a], reducing the mass losses from ice discharge while surface meltwater production and runoff remained high [Rignot et al., 2008; van den Broeke et al., 2009]. The abrupt retreat and subsequent stabilization of large outlet glaciers in East Greenland resembles a classic pattern of tidewater glacier retreat [Joughin et al., 2008a]. However, whereas the classic theory of tidewater glaciers attributes cycles of fast retreat (lasting 10–100 years) and slow advance (∼100–1,000 years) to long-term variations of mass balance of interior drainage basins, i.e., an indirect and dampened response to climate fluctuations [Meier and Post, 1987], the recent rapid fluctuations of marine outlet glaciers show an immediate and sensitive response to environmental perturbation [Rignot and Kanagaratnam, 2006; Luckman et al., 2006; Howat et al., 2007, 2008; Stearns and Hamilton, 2007; Joughin et al., 2008a, 2008b].

[4] Zwally et al. [2002] observed changes of surface velocities on >1 km thick ice in West Greenland and proposed that fluctuations were a result of basal lubrication caused by drainage of surface meltwater to the bed. This mechanism has received much attention due to a potential positive feedback whereby atmospheric warming leads to increased surface meltwater production, acceleration of ice flow through basal lubrication, increase of the ablation area due to dynamic thinning and therefore further increase of surface meltwater production [Parizek and Alley, 2004]. Recent studies have confirmed this mechanism [Das et al., 2008; Shepherd et al., 2009], but it appears to mainly influence the slow-moving, land-terminating parts of the ice sheet [Joughin et al., 2008b]. The response of the terrestrial ice margin to surface meltwater inputs is, however, not straightforward [van de Wal et al., 2008]. Although ice flows faster during the melt season compared to winter months due to hydrological connections between surface and bed [Bartholomew et al., 2010], summers with high surface-melt inputs can result in reduced ice flow compared to less warm summers because the transition from ineffective (distributed) to effective (channelized) subglacial drainage occurs sooner [Sundal et al., 2011].

[5] Moon and Joughin [2008] examined glacier terminus positions and found negligible change for glaciers terminating on land compared with those terminating in the ocean. Sole et al. [2008] also studied different types of outlet glaciers, concluding that land-terminating outlets primarily thin according to rates of surface melt while ocean-terminating outlets thin because internal ice dynamics respond to perturbations near calving termini. The main conclusions drawn by Moon and Joughin [2008] and Sole et al. [2008] are supported by Nick et al. [2009] who used a numerical model of Helheim Glacier to illustrate that recent behavior is more likely to be driven by calving dynamics than basal lubrication from penetration of surface meltwater to the bed. These recent studies of outlet glaciers illustrate a likely sensitivity of the Greenland Ice Sheet to variations of the positions of calving glacier termini. It is, however, not yet known whether accelerated calving rates during retreat occur as a response to atmospheric or oceanic forcing, or from a combination of both. Atmospheric warming can accelerate the calving rate if surface meltwater fills and deepens crevasses because calving of large icebergs may occur when crevasses reach sea level [Benn et al., 2007; Nick et al., 2010]. Ocean warming will increasingly undercut the ice front and this may also accelerate the calving rate [Benn et al., 2007].

[6] The unresolved influence of atmospheric and oceanic processes on the frontal position of calving glaciers has been discussed in several studies. Moon and Joughin [2008] show that wide-spread retreat of ocean-terminating outlet glaciers between 2000 and 2006 coincided with atmospheric warming of 1.1°C during June, July and August, concluding that the glaciers respond to atmospheric forcing. Sole et al. [2008] used a statistical relationship between thinning of land-terminating glaciers and surface melt and the absence of this relationship for marine-terminating outlets to suggest that oceanic processes may be the primary control on calving. This argument is supported by the observed warming of West Greenland coastal waters since 1996 and the coincidental retreat of Jakobshavn Isbræ [Holland et al., 2008]. The studies by Moon and Joughin [2008] and Sole et al. [2008] provide excellent spatial coverage, but limited temporal resolution. Other studies have been made with high temporal resolution, but with limited spatial coverage [Joughin et al., 2004; Thomas, 2004; Howat et al., 2005, 2007; Luckman et al., 2006; Stearns and Hamilton, 2007; Joughin et al., 2008a]. Howat et al. [2008] examined the retreat of numerous tidewater glaciers in southeast Greenland and found that glacier acceleration is proportional to the magnitude of interannual retreat. This study also examined variations of surface air temperature (SAT) and sea surface temperature (SST), both of which experienced major peaks in 2003. Murray et al. [2010] expanded the use of SSTs to show that stabilization of calving front positions after 2005 coincided with cooling of coastal surface waters. Straneo et al. [2010] report the widespread presence of subtropical waters in Sermilik Fjord, East Greenland, and rapid circulation of fjord waters after intermittent storms. Rignot et al. [2010] acquired hydrographic data near calving termini in Torssukatak Fjord in West Greenland and found large variations of submarine melt rates, from 0.7 m d−1 to 3.9 m d−1 for adjacent glaciers, concluding that ice/ocean interactions need to be documented more completely and in greater detail.

[7] In this paper, we present results from automatic satellite-based tracking of the calving fronts of 32 glaciers in East Greenland (Figure 1). We base our study on Moderate Resolution Imaging Spectroradiometer (MODIS) data, which are available daily, and we attain unprecedented detail in time series of margin position change. The development and validation of an automated method to identify calving fronts in MODIS imagery allows us to produce a data set that combines high temporal resolution and large spatial coverage. We show that outlet glaciers across East Greenland advance and retreat annually and that the amplitude of seasonality is similar in the north and south. We also document a distinct behavioral boundary at 69°N. All surveyed glaciers south of this latitude experienced retreat significantly beyond their seasonal average and the magnitudes of retreat were proportional to glacier width. None of the northern glaciers retreated significantly beyond their seasonal average during the period of observation. We use this distinct boundary to examine atmospheric and oceanic forcing factors. We conclude that ocean warming provided a critical influence on the widespread retreat of southern outlet glaciers in 2001–2005, but we hypothesize that increased runoff of surface meltwater in the same period may have contributed to glacier retreats due to its effect on meltwater-plume-induced ice-front melting as well as calving.

Figure 1.

NASA Blue Marble image of Greenland. Names refer to glaciers surveyed in this study.

2. Methods

[8] The method used in this study was designed to gain high temporal resolution margin positions for a large number of Greenland outlet glaciers. Previous studies using manual or semiautomatic edge tracing techniques and manual cloud recognition have been operator intensive. The studies have therefore been restricted to a biannual sampling frequency [Howat et al., 2008], a relatively small number of years [Moon and Joughin, 2008] or to a relatively small number of glaciers [Joughin et al., 2004; Howat et al., 2005, 2007; Luckman et al., 2006; Stearns and Hamilton, 2007; Joughin et al., 2008a, 2008c]. The method developed for this study maximizes automation to allow for the processing of daily images of 32 glaciers, totaling 105,536 images acquired in 2000–09.

2.1. Imagery

[9] Glacier margin positions were tracked using MODIS imagery collected from the NASA EOS Terra satellite launched in December 1999. We chose MODIS imagery from this satellite simply due to the fact that it was launched earlier than the other satellite carrying the MODIS sensor (i.e., Aqua). MODIS data were used in this study due to its high temporal resolution (daily) and satisfactory (250 m) spatial resolution in the visible wavelength bands. This allowed for high degrees of continuity in margin position tracking. Such data are however subject to cloud interference, although the high repeat pass frequency of the satellite increased the probability of returning cloud-free data over a given period of time. The visible imagery was also limited by insufficient solar illumination during winter, creating seasonal periods where margin positions could not be determined in high temporal resolution. These periods were around 2–3 months long for glaciers south of 69°N and 3–4 months long for glaciers farther north. We applied a number of preprocessing quality control routines to individual images to improve the accuracy of margin detection. Image subsets capturing glacier termini were taken from the full MODIS scenes, reducing computation time and memory usage. This also allowed cloud classification to be conducted on a localized basis, avoiding the dumping of viable images with high levels of cloud situated away from the glacier. The Normalized Difference Snow Index (NSDI) (following Klein et al. [1998]) was used to determine the presence of low-level clouds, identifiable by their shortwave infrared (1.6 μm) response, which is higher than that of snow. Higher altitude cirrus clouds have a different temperature profile and are therefore not detected by this index. To identify images overlain by cirrus cloud, we used the MODIS cloud identification product, which detects cirrus clouds based on their spectral response across many bands [Vermote et al., 2008]. The proportion of cloud free pixels, as denoted by both indexes, was calculated for each image. Images were rejected if they contained more than 58% cloud pixels identified by the NSDI measure or 42% cloud pixels under the cirrus measure. A further quality control routine was implemented, based on the quality assurance layer provided with the MODIS imagery. This removed images with sensor noise or missing data. All images were converted to a Lambert-Azimuthal projection to reduce image distortion.

2.2. Automatic Edge Detection

[10] To facilitate automatic detection of calving fronts in imagery, we applied edge detection algorithms to image subsections containing calving fronts. These image segments are referred to as ‘active regions’. The active regions span the width of the glacier (excluding the surrounding rock), and a distance in the direction of flow, containing the glacier front, which is sufficient to capture glacial advances and retreats. The pixels contained within the active regions were subsequently used for margin position detection. Active regions were determined by the viewing of a high frame rate animation of the images (as can be seen in Animation S1 in the auxiliary material) and all images were rotated so that ice flow direction was from left to right.

[11] We tested many commonly used edge detection algorithms on a sample of images and found the Sobel method to be the most successful at identifying the glacier front. The less successful algorithms included the Canny Prewitt, Roberts, Log and Zero-Cross methods described by Canny [1986], Lim [1989] and Parker [2010]. The Sobel method operates by estimating the derivative of brightness changes over horizontal space by convolving a 3 × 3 pixel kernel with the satellite image. A threshold is then taken to identify pixels classified as ‘edge.’ Conventionally an application of the Sobel edge detection uses a second kernel convolution to detect vertical edges, although this was not done as the images had been rotated to make only one edge orientation relevant. The number of pixels containing edges in each cross-glacier column was then summed to give the locations of the glacier edge.

[12] The Sobel method worked well because calving margins in imagery of Greenland fjords are typically defined by a change from bright snow or glacier ice to darker surfaces of open water or sea ice. A different change in brightness can, however, occur when relatively bright glacier ice meets even brighter ice mélange or sikussak, which are semirigid mixtures consisting of icebergs, iceberg fragments and sea ice [Reeh et al., 2001; Joughin et al., 2008a, 2008c; Amundson et al., 2010]. The brightness change was in this case in the opposite direction. Both types of automatically detected edge were included in our assessment.

[13] For increased robustness, we also examined calving margins using a different type of image brightness profiling method. In this method image brightness was averaged across the glacier, i.e., perpendicular to the direction of flow. We then determined the gradient of the brightness distribution by applying a difference calculation across a 3-pixel window, moving along the line of flow. The peak modulus of the output identified the steepest brightness gradient, which pertained to the most likely glacier margin position. Taking the modulus of the output ensured changes of brightness in both directions were identified.

[14] The two methods are both based on brightness, but the principal difference is that the columnar averaging in the second method took place before the brightness was analyzed. The Sobel method was best at identifying changes in brightness present over a small area, for example when open water appears in the sikussak or ice mélange in front of glacier termini. The brightness profiling method was better at detecting subtle changes present over the whole length of the terminus, e.g., when sea ice in front of a glacier was only of a slightly different brightness compared to the glacier ice itself. The two edge-detection methods were collectively used to identify the pixels most likely to contain a glacier edge. As a result of curvature and the general shape of calving glacier fronts, detected edges tended to be contained within a normal distribution rather than a single pixel. The glacier margin positions were interpolated through the automatic fitting of a Gaussian distribution and subsequent identification of the peak frequency. This procedure also allowed for subpixel accuracy refinement of the edge detection methods. The key steps of image processing, data handling and edge detection are illustrated in Figure 2, and an animation of the automatic method applied to Kangerdlugssuaq Glacier can be found in the auxiliary material.

Figure 2.

Flow diagram describing the key procedures of the method used to automatically identify calving glacier margins in daily MODIS imagery (see text for details).

2.3. Noise Reduction and Removal of Erroneous Data Points

[15] Automatic edge detection was subject to error from multiple sources. These included sensor and image registration variation, sensor noise, image distortion, incorrect edge detection location, subpixel estimation errors and noise classification error. Sensor and image registration variation was the result of geolocation errors resulting in a displaced image. Investigation found this to be present on less than one in 1000 images and to be limited to displacement of just one pixel (250 m). Sensor noise to varying degrees was present on approximately half of the images. However, it largely occurred as isolated pixels, which was insufficient to trigger false edge detections. Aerially extensive sensor noise was rare, but in a few cases it did cause misidentification of terminus positions. Similar false detections were also triggered by small clouds and icebergs. As these image features were not persistent they generally produced isolated outliers that were easily identified. Fjord wall shadows also presented coherent contrasting lines across the images and in a few cases were liable to be detected as margin position. These features occurred on a repeated annual cycle, which also allowed straightforward identification.

[16] Removal of erroneous data points was also conducted automatically (Figure 2). Glacier front positions were more likely to have been correctly identified when they had been identified in that position by both edge detection methods and did not differ unrealistically from previous position detections in space and time. Positional variation within 1.75 km from a moving average of 7 days and within 2.5 km from a moving average of 3 months were assumed to be realistic and given high scores. Points located further from the moving averages were not excluded, but simply given a lower score. An automatic filtering system was created to take advantage of these principles and remove erroneous data points. A matrix was created containing the probability of the glacier front position being located at every point in time and space for each glacier. The likelihood of glacier front position occurring at each point in the matrix was calculated as a function of the degree of coherence between the two edge detection methods and the spatiotemporal coherence of a detected edge with other detected edges. A cut off probability was then defined for each glacier's matrix, below which identified edges were assumed to be inaccurate and were removed from the data. Further anomalies were removed from the data if they lay too far from a robust spline curve fitted to the margin position time series and occasionally manually, if there was reason to believe they were erroneous.

2.4. Validation and Error Assessment

[17] A validation exercise was performed against margin positions for Kangerdlugssuaq Glacier derived by Joughin et al. [2008a]. The latter data set was generated by running an edge-detection algorithm on cloud-free MODIS and ASTER images and using visual inspection to accept or reject the delineated positions. This comparison showed that our automatic method captured very similar margin positions (Figure 3). Deviations are seen in 2004 and 2005 where eight margin points differ considerably from the positions reported by Joughin et al. [2008a]. It is important to note that the sampling intervals in this study and the one by Joughin et al. [2008a] are not identical and that our method may have captured previously undetected changes in the position of glaciers fronts. However, manual inspection confirmed that the eight deviating data points are spurious margin positions captured erroneously by the automatic technique. Nevertheless, a comparison of margin positions for identical days shows that the root-mean-squared error was 0.19 km. This error is small compared to the amplitude of margin fluctuations and it is a satisfactory result because it is less than the spatial resolution of MODIS imagery (0.25 km). We performed the same test against margin positions reported by Joughin et al. [2008a] for Helheim Glacier, which also yielded a root mean squared error smaller than the image resolution. The Joughin et al. data set is based partly on MODIS imagery, so a general agreement between our margin positions and theirs was expected. However, given the exclusion of high-resolution ASTER imagery in this study, the use of different edge detection methods and the elimination of extensive visual inspection, we consider the comparison to reflect a validation of automatic glacier edge detection based exclusively on MODIS imagery and automatic techniques. The method has been shown to work well for the vast majority of data, but eight of the 670 data points shown in Figure 3 do nevertheless deviate from the Joughin et al. [2008a] data set. We consider this to be satisfactory given that the spurious points represent only 1.2% of the entire data series from Kangerdlugssuaq Glacier. To reduce the possible influence of occasional false data points, we use the high number of data points generated by the automatic method to calculate mean monthly margin positions and we use maximum and minimum values from mean monthly time series to estimate the intra-annual variability of terminus position and determine the average range of seasonal advances and retreats during the period of investigation. We also use the mean monthly estimates of margin position to characterize the interannual trend of each glacier. The latter was achieved by filtering out seasonal variation with a 12 month running average. Below we use the term seasonal range to characterize the intra-annual variability of terminus position while interannual retreat is used to characterize year-on-year variations, i.e., the trend derived by exclusion of seasonality.

Figure 3.

Comparison of glacier front positions derived by the automatic method of this study (solid blue crosses) and by Joughin et al. [2008a] (open red circles). The positions are for Kangerdlugssuaq Glacier (68.5°N). Arrows point to eight deviating data points discussed in text.

2.5. Application

[18] In this study, we applied the automatic edge detection method to 32 ocean-terminating glaciers in East Greenland fjords (Figure 1). The glaciers were selected evenly along the coast and their widths range from 1.3 km to 15 km. In fjords with numerous glaciers we selected long and wide glaciers over small and narrow ones as they have a clearer presence on the MODIS imagery. The aim of this widespread application was to examine the magnitude of seasonal cycles of advance and retreat and determine the geographic extent of recent glacial retreat. The names, abbreviations and locations of examined glaciers are shown in Figure 1.

3. Results

[19] The automatic margin detection method produced 26,802 calving front positions from 105,536 MODIS images encompassing the 32 glaciers shown in Figure 1. The average sampling frequency of the data set was 4.75 days, but it is important to note that the sampling period was not evenly distributed. Persistent cloud cover produced intermittent data breaks on a daily to weekly scale while poor illumination meant data were not continuously available over winter. However, whereas no margins were detected for glaciers north of 69°N during a 3–4 month long winter season, scattered margin positions were often identified in winter periods for glaciers farther south due to better solar illumination at lower latitudes. The temporal and regional trends of margin position change in East Greenland are shown in Figure 4. The time series of calving margin positions often display distinct seasonality. Dynamic seasonal cycles are known for Kangerdlugssuaq, Helheim and Jakobshavn Isbræ glaciers [Luckman and Murray, 2005; Joughin et al., 2008a, 2008c], but this study shows that seasonal response to environmental conditions is common from the north to the south.

Figure 4.

Changes of the relative position of calving glacier fronts (solid dots) derived from automatic edge detection and brightness in MODIS imagery. (a) Jakobshavn Isbræ (J). (b) Mogens 1–3 (M1–M3) and Tingmjarmiut (T1). (c) AP Bernstoffs Gletscher (APB), Storebjørn (S), and Fimbulsgletscher (Fim). (d) Helheim Glacier (HH), Kangerdlugssuaq Glacier (KG), Courtaulds Gletscher (CG), and Frederiksborg Gletscher (FG). (e) Borggraven (BG) and Daugaard-Jensen Gletscher (DJ) and seven glaciers in between (see Figure 1 and Table 2 for details). (f) Hisingers Gletscher (HiG) and Nordenskjolds Gletscher (NG). (g) Jætte Gletscher (JG), Gerard de Geer Gletscher (GdG), Walterhausen Gletscher (WG), and Adolf Hoel Gletscher (AH). (h) Heinkel Gletscher (HG), Lars Bistrup Bræ (LBB), Storstrømmen (SS), and Kofoed-Hansen Bræ (KH). The thin solid line superimposed on margin positions is the trend derived from calculation of mean monthly positions and interpolation.

[20] The seasonal cycle is superimposed on interannual trends. We derived the latter for each glacier by computation of 12 month moving averages from monthly mean margin positions. The seasonal variability was determined by subtracting the interannual trends from monthly mean margin positions. Taking a monthly mean canceled out short-term stochastic variation within the data, producing a more robust result and avoiding the possibility of identifying possible outliers as maximum or minimum seasonal position. The maximum and minimum positions of calving fronts were typically observed in the early spring to late autumn season where our data points are numerous. The only exception to this trend is Jakobshavn Isbræ (Figure 4a), which is not part of our analysis, but results for this west coast outlet glacier are included nonetheless because they indicate a difference in behavior. For Jakobshavn Isbræ, the transition from advance to retreat occurs in February or March because the iceberg mélange in this fjord weakens in midwinter, possibly because open water forms when trapped icebergs are able to escape through a topographical constriction [Joughin et al., 2008c].

[21] The seasonal variations of east coast glaciers are listed in Tables 1 and 2 together with values of interannual change before and after 2005. Regular seasonal cycles with amplitude of 1 km or more were observed on 22 glaciers and the amplitudes did not vary much between glaciers in the south (Table 1) and glaciers in the north (Table 2). For instance, the seasonality of glaciers north of Kangerdlugssuaq Fjord (68.5°N) averaged 1.2 km and the average seasonality for the southern glaciers was found to be almost the same (1.1 km). The largest mean seasonal cycle of 3.4 km was observed on Kangerdlugssuaq Glacier, followed by Helheim Glacier (2.3 km) and Nordenskjölds Gletscher (1.6 km). Calving fronts with seasonal variations less than 1 km were typically less than 2 km wide. The relationship between glacier width and the amplitude of seasonal advance/retreat is highly significant at r = 0.73 when two anomalously wide glaciers are excluded (Figure 5a). The excluded glaciers were Walterhausen Gletscher and Lars Bistrup Bræ, which have widths of 12 and 16 km, respectively. All other glaciers were less than 7.5 km wide. We do not know why the relationship between front position and glacier width breaks down for glaciers wider than 7 km, but it may be related to width-dependent role of sidewall friction in the force budget of glaciers [Raymond, 1996].

Figure 5.

(a) Scatterplot showing relationship between glacier width and seasonal range of advance and retreat (all glaciers). (b) Scatterplot showing relationships between width of glaciers south of 69°N and the magnitude of glacier retreat in 2000–2005 (red) and readvance in 2005–2009 (blue).

Table 1. Names and Calving Front Characteristics of Glaciers South of 69°Na
GlacierWidth (km)Seasonality Advance/Retreat 2000–2009 (km)Front Change July 2001 to July 2005 (km)Front Change July 2005 to July 2008 (km)
  • a

    Seasonality refers to the differences between maximum (summer) and minimum (winter) annual positions, and front change refers to the differences in interannual trend when seasonal variations are excluded.

Mogens 12.900.76−2.380.53
Mogens 21.270.70−1.420.07
Mogens 31.650.59−3.420.25
AP Bernstorffs Gletscher4.651.53−4.030.61
Helheim Glacier7.282.27−4.522.17
Kangerdlugssuaq Glacier6.603.44−5.001.93
Courtaulds Gletscher1.860.69−0.12−0.16
Frederiksborg Gletscher1.510.70−0.380.07
Table 2. Names and Calving Front Characteristics of Glaciers North of 69°Na
GlacierWidth (km)Seasonality Advance/Retreat 2000–2009 (km)Front Change July 2001 to July 2005 (km)Front Change July 2005 to July 2008 (km)
  • a

    Seasonality refers to the differences between maximum (summer) and minimum (winter) annual positions and front change refers to the differences in interannual trend when seasonal variations are excluded.

Vestre Borggletscher3.
Vestfjord Gletscher2.571.18−0.100.03
Rolige Bræ4.631.10−0.130.07
Eielson Gletscher3.681.00−0.08−0.01
Daugaard-Jensen Gletscher3.951.82−0.59−0.13
Hisingers Gletscher3.201.37−0.28−0.29
Nordenskjölds Gletscher3.401.590.110.07
Gerard de Geer Gletscher2.770.99−0.310.16
Adolf Hoel Gletscher5.011.04−0.44−0.31
Waltershausen Gletscher11.870.75−0.130.15
Heinkel Gletscher2.950.99−0.29−0.28
Lars Bistrup Bræ15.141.08−0.760.29
Kofoed-Hansen Bræ5.911.08−0.760.29

[22] Although seasonal dynamics appear widely similar, the interannual trends of the data show a distinct behavioral boundary at 69°N. All observed glaciers south of this latitude experienced significant interannual retreats in 2001–05 (Figures 4a4d), while those observed above showed very little or no interannual change over the period of observation (Figures 4e4h). All southern glaciers wider than 2 km retreated and their average interannual retreat was 3.3 km, more than twice their average amplitude of seasonality (1.5 km). Smaller glaciers retreated by 1.3 km on average, which is also twice their seasonal average (0.65 km). Two glaciers in Kangerdlugssuaq Fjord whose widths were 1.5 and 1.9 km did not retreat significantly beyond their seasonal average. These glaciers were Courtaulds and Frederiksborg glaciers, located at 68.5°N and 68.3°N respectively. All 20 glaciers north of 69°N remained stable in 2000–09, although the mean width (4.8 km) was higher than the mean width of glaciers in the south (3.4 km).

[23] To estimate the average magnitude of interannual retreats, we computed the average retreat for glaciers south and north of 69°N, respectively. For the southern glaciers this average showed mean interannual retreat of 2.3 km between 2001 and 2005 (Table 1), compared to just 0.26 km for northern glaciers (Table 2). After this period the southern glaciers stopped retreating and some readvanced. Helheim Glacier, for instance, advanced 2.2 km between 2005 and 2009. This readvance is, however, considerably higher than the regional interannual average of 0.43 km. The retreat and readvance of southern glaciers averaged 3.3 km and 0.91 km, respectively, if we exclude glaciers narrower than 2 km. Overall, we found that southeast glaciers stopped retreating after 2005 and readvance in 2006–09 amounted to 26% of the retreat encountered in 2001–05. The extent of retreat in 2001–05 (r = 0.79) and the extent of subsequent readvance (r = 0.93) are significantly correlated with glacier width (Figure 5b).

4. Assessing Possible Causes of the Dynamic Boundary at 69°N

4.1. Surface Ablation and Summer Atmospheric Forcing

[24] The similar seasonal patterns of outlet glacier advance and retreat (Figure 5a) and the contrasting interannual trends of glaciers north and south of 69°N (Figure 4) offer a means to examine environmental forcing factors. To examine whether calving in the north and south are influenced by different trends in surface air temperature, we calculated surface air temperature anomalies from permanent weather stations located in Tasiilaq (65.60°N), Ittoqqortoormiit (70.48°N) and Danmarkshavn (76.77°N) [Cappelen et al., 2008]. Summer air temperature variations (July, July and August) at these locations are significantly correlated at the 99% level (Figure 6). This shows that air temperature changes have been uniform across East Greenland during summer months. On a basin scale, Box et al. [2009] also report similar summer warming trends for northeast (+0.6°C), east (+0.7°C) and southeast (+0.8°C) Greenland when averages for 1994–2007 are compared to averages for 1840–2007. The synchronous retreats of southeast Greenland glaciers occurred when summer air temperatures were unusually high [Moon and Joughin, 2008], but equally high air temperature anomalies occurred in the northeast where glaciers have remained stable. A comprehensive review of all atmospheric forcing factors is beyond the scope of this study, but we contend that the uniform variability of summer air temperature across East Greenland is still a good indicator of a uniform change in summer atmospheric forcing. This is supported by van den Broeke et al. [2009] who report very similar increases in the runoff of meltwater from northeast (+20 Gt yr−1) and southeast (+25 Gt yr−1) sectors of the Greenland Ice Sheet in 2003–2008 (relative to the means for 1961–1990), when the surface mass balance is fully assessed. Past studies therefore indicate that the 69°N boundary of glacier change is unlikely to be a result of regional differences in summer surface air temperature or runoff.

Figure 6.

(a) Summer air temperature anomalies (July, August, and September) derived from measurements in Tasiilaq (65.60°N, red line), Ittoqqortoormiit (70.48°N, blue line), and Danmarkshavn (76.77°N, black line). The anomalies represent differences from averages of period shown. (b) Same as Figure 6a but for winter months (December to March). Data are from the Danish Meteorological Institute [Cappelen et al., 2008].

4.2. Sea Ice and Winter Atmospheric Forcing

[25] Figure 6b shows anomalies of mean winter (December to March) air temperature in Tasiilaq, Ittoqqortoormiit and Danmarkshavn, and these vary considerably more than the summer equivalent (Figure 6a). Box et al. [2009] show that winter air temperature anomalies in northeast, east and southeast Greenland were +2.8°C, +3.0°C and +3.9°C respectively in 1994–2007 compared to 1840–2007. These anomalies are far greater in magnitude than the summer equivalent, i.e., +0.6°C, +0.7°C and +0.8°C respectively, and they indicate a greater contrast between northern and southern basins. Although the retreat of southeast Greenland glaciers occurred in a period when summers were warm [Luckman et al., 2006; Howat et al., 2008; Joughin et al., 2008a], the geographic pattern reported here with retreating glaciers confined to latitudes below 69°N is also consistent with a response to winter climate. It is feasible that the annual rate of calving is sensitive to mild winters as well as warm summers because mild winters may promote glacial retreat if sea ice in fjords forms over shorter periods than usual [Reeh et al., 2001]. Cold winters may similarly promote glacier advance if sea ice cover fjords over a comparatively long period. Joughin et al. [2008c] shows that calving of icebergs ceases when sea ice forms in front of Jakobshavn Isbræ and that seasonal advance and retreat are a result of seasonal strengthening and weakening of sikussak or ice mélange near the calving glacier front. The role of sea ice and sikussak was discussed in earlier work by Sohn et al. [1998] and Reeh et al. [2001] and more recently by Amundson et al. [2010] who show that ice mélange in front of Jakobshavn Isbræ moves down the fjord as a rigid mass in winter when it is tied together by sea ice, but not in summer when absence of sea ice makes the mélange less rigid. We cannot confirm whether this also is the case in East Greenland, but warm air temperatures in October–December, which is when new sea ice first forms, or in May–July, which is when sea ice breaks up, can conceivably extend the calving season and result in retreat beyond the seasonal average [Joughin et al., 2008a].

[26] To evaluate the role of seasonal strengthening of ice mélange and sikussak we examined the breakup and formation of sea ice inside fjords in the MODIS imagery used to detect glacier margins. Although the resolution of the images is too coarse to allow direct observation of sea ice formation, it was possible to infer from sequences of images whether collections of icebergs near calving termini were moving as a coherent mass, indicating the presence of a mechanically strong sikussak, or if the icebergs were moving independently, indicating weaker and more broken-up mélange. While this method has a certain level of temporal uncertainty, we found that in most cases the transition from rigid to loose mélange is abrupt and that breakup dates can be identified with confidence. Figure 7a shows margin positions for Kangerdlugssuaq Glacier together with the latest date of each year when imagery shows rigid motion of the sikussak in Kangerdlugssuaq Fjord. Figure 7 also shows the earliest dates when the sikussak has lost its integrity. Figure 7 illustrates that glacier retreat and breakup of sikussak coincide. Jakobshavn Isbræ displays similar behavior, but with sikussak breakup occurring in midwinter rather than in spring [Joughin et al., 2008c].

Figure 7.

Time series of mean monthly margin positions for (a) Kangerdlugssuaq Glacier, (b) Daugaard-Jensen Gletscher, and (c) Storstrømmen. Dash-dotted lines illustrate the latest date when MODIS imagery indicates rigid motion of sikussak near calving fronts (black) and the earliest date when sikussak has lost rigidity (red). A red line superimposed on a black line represents well-constrained timing of the loss of sikussak integrity. See text for details.

[27] We were not able to clearly identify the dates when a rigid sikussak first formed because suitable images acquired prior to the onset of dark winters with poor solar illumination (October) were typically ice-free, while images acquired when solar illumination returned (February) typically showed a rigid sikussak. The only winter when this was not the case was in 2004–05 when Kangerdlugsuaq Fjord was ice-free when sufficient solar illumination returned on 14 February 2005. Further inspection of the MODIS data revealed that the fjord was relatively ice-free on 14–28 February and that freeze-up occurred at some point between 28 February and 10 April. This unusual condition coincided with the sudden retreat of Kangerdlugssuaq Glacier. The margin data we acquired for Kangerdlugssuaq Glacier show that retreat of 2.2 km occurred between July and September 2004 and that additional 4.6 km of retreat occurred between September 2004 and March 2005 (Figure 3). This sudden retreat caused the glacier to double its speed and flow as fast as 40 m/day [Luckman et al., 2006; Stearns and Hamilton, 2007; Howat et al., 2007].

[28] A study of sea ice breakup dates for other glaciers was used to elucidate winter atmospheric forcing further. In Sermilik Fjord, which contains Helheim Glacier, sea ice forms and breaks up intermittently and open water is common in various parts of the fjord throughout winter months. This makes it difficult to infer the influence of sea ice in strengthening the sikussak. Sea ice conditions in southeast Greenland may thus be similar to sea ice conditions in Disko Bugt near Jakobshavn Isbræ where winter concentrations have dropped to 50–80% since 1997 compared to ∼100% in the early 1990s [Joughin et al., 2008c]. In the northeast, we examined sea ice in front of Daugaard-Jensen Gletscher and Storstrømmen. Daugaard-Jensen Gletscher retreats when sea ice is absent, but the retreat typically commences prior to the annual breakup of sea ice (Figure 7b). For Storstrømmen, there was no temporal coincidence between sea ice breakup and seasonal retreat of the glacier (Figure 7c).

[29] The fast retreat of Kangerdlugssuaq Glacier between July 2004 and March 2005 occurred during a warm summer and the retreat continued during a mild winter when the onset of sea ice formation was delayed. It is possible that the delayed formation of sea ice prolonged the weak state of sikussak well into the winter of 2004–05, causing a substantially increased length of seasonal retreat, as suggested by Joughin et al. [2008a]. However, the retreat of many southeast glaciers, including Helheim located about 80 km from Tasiilaq, was already underway in 2001–02, i.e., prior to peak atmospheric temperature forcing, which occurred in the winter of 2002–03 and the subsequent summer (Figure 6). The role of atmospheric temperature forcing and sea ice, which seem apparent for Kangerdlugssuaq Glacier, is less apparent for other glaciers.

4.3. Ocean Forcing

[30] The properties of coastal water masses influence the rate of submarine melting [Motyka et al., 2003] and submarine melt can undercut the front of calving glaciers and thereby influence the rate of calving [Benn et al., 2007]. Ice-ocean interactions in Greenland are, however, documented in few locations only. Holland et al. [2008] show that coastal water along the west coast of Greenland warmed significantly in 1997 and they suggest that the rapid subsequent thinning of Jakobshavn Isbræ and the disintegration of its 15 km long floating ice tongue was caused by submarine melting from increased inflow of subtropical waters to the fjord. The latter has not been observed directly, but warming and salinification of waters in the Irminger Sea and the northwest North Atlantic have been documented by Falina et al. [2007], Yashayaev et al. [2007] and Sarafanov et al. [2007, 2009], after sudden weakening of the North Atlantic Oscillation (NAO) in the mid-1990s. The NAO is a major recurring pattern of atmospheric variability [Hurell et al., 2003] and the transition from a strong positive phase in the late 1980s and early 1990s to an unusually weak state in 1996 caused slowdown and contraction of the subpolar North Atlantic gyre [Flatau et al., 2003; Hakkinen and Rhines, 2004], resulting in advance of subtropical waters to the Irminger Sea and to Greenland coastal waters [Holland et al., 2008].

[31] Although the proposed inflow of subtropical waters to the fjord containing Jakobshavn Isbræ has not been confirmed by observations, high subaqueous melt rates (up to 3.9 m per day) for smaller adjacent outlet glaciers have been estimated by Rignot et al. [2010], while fast wind-driven exchange of cold polar waters and warm subtropical waters has been observed in Sermilik fjord, containing Helheim Glacier [Straneo et al., 2010].

[32] To examine ocean forcing of the Greenland Ice Sheet we used subsurface ocean temperatures in a 22 year long ocean reanalysis (1987–2008) with the NEMO ocean model version 2.3 [Madec, 2008], at an eddy-permitting 1/4° resolution on the tripolar ORCA025 grid [Barnier et al., 2006]. The reanalysis was produced by the OPA9 ocean model and the LIM2.0 sea ice model [Fichefet and Maqueda, 1997; Goosse and Fichefet, 1999] and with assimilation of salinity (S) and temperature (T) data. Surface atmospheric forcing for the reanalysis was obtained from the DRAKKAR Forcing Set 3 [Brodeau et al., 2010], which is a hybrid data set making use of the ERA-40 atmospheric reanalysis, ECWMF operational analyses and the Common Ocean Reference Experiment data set [Large and Yeager, 2004]. The parameter settings of the eddy-permitting 1/4° resolution NEMO ocean model are discussed by Barnier et al. [2006] and Penduff et al. [2010]. The model experiment used for this analysis is referred to as UR025.1.

[33] In situ observations were assimilated from the UK Met Office quality controlled ENACT/ENSEMBLES data set EN3-v1c, which includes S(T) data from the World Ocean Database '05 (http://www.nodc.noaa.gov/OC5/WOD05/pr_wod05.html, the Global Temperature-Salinity Profile Program (http://www.nodc.noaa.gov/GTSPP/), and Argo (http://www.argo.net). There is a good coverage of observations from the Irminger Sea as well as the East Greenland shelf seas in EN3-v1c and independent hydrographic observations show that the S(T) assimilation method results in accurate representation of water mass properties [Haines et al., 2006; Gemmell et al., 2008, 2009; Smith and Haines, 2009; Smith et al., 2010]. The performance of UR025.1 compared to a control run without assimilation in terms of water mass properties and circulation in the Arctic Ocean is discussed in detail by Zuo et al. [2011]. It is appropriate to utilize UR025.1 to investigate changes in ocean temperature along the east coast of Greenland where hydrographic observations are more frequent than those from the Arctic Ocean. To ensure that the correction from data assimilation was achieved, we exclude the first two years of the reanalysis and focus on the period 1989–2008.

[34] Figure 8 shows locations of surveyed glaciers as well as sites where ocean conditions were examined in the reanalysis. Margin positions for glaciers south and north of 69°N are shown in Figures 9a and 9b, respectively. Figure 9c shows 12 month moving averages of monthly mean temperatures averaged from 100 m depth to the bottom for southeast sites in the reanalysis. Figure 9 clearly shows that temperatures of deep coastal waters change at 69°N, which is coincident with the geographical boundary of glacier dynamics described above. In the southeast, sudden warming of water columns (∼2°C) occurred in 1996 and further gradual warming (∼1°C) lasted until 2004. Between 2004 and 2008, subsurface coastal temperatures declined by ∼2°C, which is coincident with cessation of retreat and partial readvance of southeast glaciers. Figure 9d shows the equivalent coastal water temperatures for glaciers north of 69°N. The temperature of coastal waters north of 69°N rarely exceeds 0°C when seasonality is excluded and the region where the subsurface ocean is persistently cool is the same region where calving fronts of marine-terminating outlet glaciers have remained unchanged when seasonality is excluded.

Figure 8.

Map of Greenland and adjacent seas with color scale illustrating mean annual sea surface temperatures in 2004 in the ocean model reanalysis UR025.1. Colored squares show locations of surveyed glaciers and colored dots show locations where subsurface temperatures were examined in the ocean reanalysis. The same colors are used to display margin positions in Figure 4 and ocean temperature in Figure 9. White dashed arrows illustrate main currents, which are North Atlantic Current (NAC), Irminger Current (IC), East Greenland Current (EGC), and East Icelandic Current (EIC).

Figure 9.

Margin positions for glaciers (a) south and (b) north of 69°N. Colors correspond to margin positions shown in Figure 4 and locations shown in Figure 8. Dashed lines mark glaciers narrower than 3 km. Thin solid lines represent glaciers with widths of 3–6 km. Bold lines denote glaciers wider than 6 km. (c) Temperature of coastal waters near surveyed glaciers in southeast Greenland. The temperatures are from ocean reanalysis UR025.1 and are averaged from 100 m to bottom. (d) Same as Figure 9c, but temperature is for coastal waters near glaciers in northeast Greenland. Seasonal variation of ocean temperatures has been removed by filtering monthly means with a 12 month moving average.

5. Discussion

[35] The boundary at 69°C marking the geographical extent of recent calving retreat is similar to the boundary of recent ice velocity change reported by Rignot and Kanagaratnam [2006]. The consequences of region-wide glacial retreats is underscored by van den Broeke et al. [2009] who attribute a net ice loss of 70 Gt yr−1 in 2003–08 to discharge of ice from southeast Greenland outlet glaciers, which retreated significantly in 2001–05. This is 74% of the ice sheet–wide mass imbalance by ice discharge and 30% of the total net annual ice loss. In contrast, there was no net loss from discharge of ice by glaciers in the northeast where we detected no major interannual change of ice-front positions and where subsurface ocean temperatures are generally cold. Howat et al. [2008] show that the increased velocity of southern glaciers in 2000–06 were proportional to the magnitude of glacier retreat, so the difference in ice discharge reported by van den Broeke et al. [2009] for the southeast sector of the ice sheet may be explained by the observed boundary of interannual margin change at 69°N.

[36] It is crucial to correctly identify the underlying cause of the geographical boundary separating the region of sustained calving retreat with regions where glaciers have remained relatively unchanged. For instance, if the cause is atmospheric, the boundary will likely migrate north as climate warms, as suggested by Rignot and Kanagaratnam [2006]. If the cause is oceanic or associated with air-sea interactions, future changes will depend on changes in wind forcing, coastal currents and the exchange of water masses between deep sea, shelves and fjords [Holland et al., 2008; Straneo et al., 2010].

[37] This study shows that recent high rates of mass loss from southeast Greenland glaciers occurred from calving glaciers whose fronts retreated in 2001–05 and that a regional pattern of retreat for glaciers south of 69°N corresponds to a region where mean annual temperatures of deep coastal waters have warmed by >2°C due to inflow of subtropical waters from the Irminger Sea. The warming seen in the ocean reanalysis in 1995–96 is the equivalent of that reported by Holland et al. [2008] on the west coast in 1997. Thomas et al. [2000] report thinning of Kangerdlugssuaq Glacier's terminus by 50 m between 1993 and 1998 and this major episode of thinning may have been an immediate response to warming of coastal waters after 1995–96. However, widespread and synchronous retreats did not occur until 2001 when mean annual subsurface temperatures along the southeast coast of Greenland were up to 3°C higher than those encountered in 1995. We do not know why there was a lag of more than 5 years between the warming in 1995–96 and widespread glacier impact. Although the reanalysis is well-constrained by assimilation of extensive observational data sets from the North Atlantic including the Irminger Sea, we cannot firmly exclude the possibility that simulated coastal subsurface temperatures are overestimated. The simulated temperatures are, however, supported by observations from West Greenland, which confirm that coastal waters originating from the east coast were substantially warmer in 1997 compared to 1991–1996 [Holland et al., 2008]. This observation and subsequent observations from East Greenland, which show that stormy periods with strong northeasterly (along-shore) winds result in fast exchange of water masses inside eastern fjords [Straneo et al., 2010], indicate that the 5 year lag between the onset of ocean warming and glacier response is unlikely to be explained exclusively by slow rate of water mass transport. Our preferred interpretation is that glaciers remained stable in the late 1990s despite warm oceanic conditions because air temperatures were relatively cold.

[38] Atmospheric warming became sustained in Greenland only after 1995 [e.g., Chylek et al., 2006; Box et al., 2009]. Hanna et al. [2008] show that the four warmest summers in southern Greenland in 1958–2006 were 2003, 2005, 2006 and 2001. We therefore propose that contemporaneous increases in both air and ocean temperatures caused the dramatic and synchronous retreat of southeast Greenland outlet glaciers in 2001–05. Supporting this are calculations indicating high rates of submarine melting of Greenland outlet glaciers [Rignot et al., 2010] and results from theoretical modeling, which show that the rate of submarine melting increases significantly in the presence of subglacial meltwater plumes in front of calving termini [Motyka et al., 2003]. The plumes promote melt because they introduce turbulent transfers at the ice/ocean boundary [Jenkins et al., 2010]. The velocity of a plume, and hence the associated submarine melt rate, is driven by the rate at which glacial meltwater is discharged from the subglacial water system. High rates of submarine melt should therefore occur in warm summers when high runoff rates increase discharge from the subglacial drainage system. The relationship between the plume velocity and the melt rate is furthermore modulated by the temperature of the ambient fjord water, which is mixed into the plume as it rises. We expect that a large fraction of the runoff generated in catchments is routed to the terminus zone of outlet glaciers [Andersen et al., 2010]. The cumulative effect of this may be plume-induced frontal melting, intensification of stresses near the ice-front and calving at high rates, particularly if subglacial discharge is high while fjord waters are warm [Holland, 2010]. Atmospheric forcing may alternatively influence the calving processes if surface-melt-induced hydrofracturing deepens crevasses near the terminus [Benn et al., 2007; Nick et al., 2010], but the driving mechanism of this effect should be limited to melt occurring near the terminus rather than meltwater produced in the entire catchment, e.g., as shown by Andersen et al. [2010].

[39] Runoff in the northeast sector of the Greenland Ice Sheet increased by +20 Gt yr−1 in 2003–2008 compared to 1961–1990 and this change is similar to the increase in runoff in southeastern drainage basins (+25 Gt yr−1) [van den Broeke et al., 2009]. We hypothesize that northern terminus positions remained stable despite increased runoff because fjords north of 69°N are much less exposed to inflow of subtropical waters compared to glaciers in fjords farther south. One of the main consequences of subtropical waters in southern fjords may be plume-induced ice-front melting in addition to hydrofracturing from surface meltwater entering crevasses near glacier termini. While surface ablation and runoff from the Greenland Ice Sheet are likely to remain high due to climate change, warm oceanic conditions are likely to be influenced by short-term climate variability including air-sea interactions in the North Atlantic.

6. Conclusions

[40] The development of an automated method for the extraction of glacier margin positions from daily MODIS imagery enabled data to be collected over large spatial area and in high temporal resolution. On average, we determined margin positions every 5 days for 32 calving glaciers in Greenland, located from 62.4°N to 77.5°N, but data points from winter months were irregular in the south and absent in the north because of insufficient solar illumination. The automatic data processing method performed well in validation exercises for Kangerdlugssuaq and Helheim glaciers against data collected in a previously published study [Joughin et al., 2008a]. Our method eliminates labor-intensive processes used in other studies.

[41] We used the automatic edge detection method to map changes in the position of calving glacier fronts across East Greenland. Our data showing retreat of southeast glaciers corroborates earlier findings [Howat et al., 2008; Moon and Joughin, 2008; Murray et al., 2010], and we report new spatial characteristics of these retreats with data acquired in high-temporal resolution. The data show similar seasonal glacier-front variability from the north to the south when two unusually wide glaciers were excluded. For the remaining 30 glaciers, the amplitude of advance and retreat was proportional to glacier width. Significant interannual retreats were confined to glaciers south of 69°N, as glaciers north of this latitude remained unchanged when seasonality was excluded. We reviewed basic climatic factors, including summer and winter atmospheric temperature forcing, but these forcing factors did not alone explain our observations. Instead, we found that the boundary of dynamic glacier change at 69°N correspond to a distinct boundary in the temperature of subsurface coastal waters.


[42] This study was supported by the Natural Environment Research Council through grants NE/G00692X/1 and NE/H020667/1 to P.C. and the IPY ASBO program. R.M. thanks Keith Haines and Greg Smith for their help with the NEMO ocean model.