Glacier outlines and tidewater margins

In order to assess central East Greenland glacier area, tidewater margin fluctuations and sensitivity to climate change, a detailed glacier inventory of the Geikie Plateau region was compiled. Using semi-automated digitization, glacier outlines were extracted from 68 Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) L1B scenes with a spatial resolution of 15 m, and 6 Landsat 7 Enhanced Thematic Mapper Plus (ETM+) pan-sharpened scenes with a resolution of 14.5 m. The global Landsat 7 panchromatic band is orthorectified to an accuracy of 14.25m (Reference Tucker, Grant and DykstraTucker and others, 2004). Most ASTER scenes were from late summer 2004/05. May and August 2000 scenes were used to fill in gaps. Landsat 7 scenes were in a coastal strip from 12 July 2000 and an inland strip from 28 August 2001. ASTER scenes were georeferenced to Landsat 7, with the exception of two regions where ASTER-to-ASTER georeferencing was applied due to an absence of cloud-free Landsat 7 scenes. We matched 20 ground-control points distributed in a star-shaped pattern, with a horizontal root-mean-square (RMS) error less than 3.0 m. By georeferencing ASTER to Landsat 7 scenes, the ASTER scenes were resampled to 14.5 m.

Although the band ratio method (Reference Paul, Kääb, Maisch, Kellenberger and HaeberliPaul and others, 2002) is commonly used for semi-automated glacier delineation, we used supervised classification as the separate extraction of snow, ice, fresh water and sea water enabled us to obtain snowlines and lakes for further study. Glaciers were identified from mosaics of two to four ASTER scenes of the same date and gain setting. Training areas of ≥1500 pixels were selected for ten surface classes within each mosaic. These were ultimately grouped into a binary glacier/non-glacier class. The glacier class contained snow and exposed glacier ice. The non-glacier class contained sediment, bedrock, fresh water, vegetation, shadow and sea water. Shaded ice was included in the glacier class, and shaded bedrock/sediment in the non-glacier class, but this shadow filtering was only moderately successful. Systematic comparison of supervised classification tools in ENVI 4.3 revealed that the Maximum Likelihood and Mahalanobis Distance classifications performed best in identifying the glacier class. The Mahalanobis Distance method was ultimately applied because it is faster and less sensitive to the quality of training data (Reference Mutlu, Popescu, Stripling and SpencerMutlu and others, 2008). Small polygons and irregularities were removed using the Enhanced Lee Filter with a threshold of 3×3 pixels. Filtered raster images were converted to polygons and appended into one shapefile in ArcGIS 9.3. A minimum glacier size threshold of 2 km² was applied because of difficulties distinguishing seasonal snow from glaciers, the small proportion of overall glacierized area and the absence of calving margins in this size group, and for time-management reasons. Because the shapefile exceeded 4GB, polygons were cleaned using the generalize function with a minimum distance of 100 m. The original file was archived for future use. Significant generalization errors, debris-covered ice, shadow zones and nunataks were manually improved using the Landsat 7 mosaic and the ASTER global digital elevation model (GDEM), a product of the Ministry of Economy, Trade and Industry (METI), Japan, and NASA (http://www.ersda-c.or.jp/GDEM/E/). Ice divides were manually delineated from illumination differences, DEM aspect and glaciological interpretation (Reference Racoviteanu, Paul, Raup, Khalsa and ArmstrongRacoviteanu and others, 2009). Ultimately, roughly half the glacier outlines were manually digitized.

Each glacier unit was assigned a GLIMS-ID (Global Land Ice Measurements from Space), based on the east and north decimal degree coordinates of the glacier flowline intersect with the late-summer snowline (http://www.glims.org/MapsAndDocs/guides.html). GLIMS-ID G333549E68976N (Storbræ) corresponds to point location 68.976˚ N, 26.451˚W. Glacier inventory attributes, glacier names and a hydrological basin code (Reference Weidick, Williams and FerrignoWeidick, 1995) were also assigned. We spatially joined our glacier inventory to its predecessor (Reference JiskootJiskoot, 2002; Reference Jiskoot, Murray and LuckmanJiskoot and others, 2003) and transferred its surge classification, where surge class 0 = normal glacier (no morphological evidence for surge behaviour), 1 = possibly surge-type (one or two equivocal morphological features of past surge behaviour (e.g. unusual crevasses, some potholes, contorted moraines)), 2 = probably surge-type (two or more unequivocal morphological features of surge behaviour (e.g. tear-shaped moraines, widespread potholes, surge bulge)) and 3 = surge-type (glaciers with an observed surge). For new glaciers, these surge classes were also added, based on the full range of morphological characteristics from Reference Jiskoot, Murray and LuckmanJiskoot and others (2003).

Tidewater glacier margins were manually extracted from July 2000 Landsat 7 and August 2001 ASTER scenes. Where possible, margins were snapped to the glacier outlines derived with the semi-automated method. Tidewater margins from 2004 and 2005 were digitized from ASTER scenes. GEUS map database ice polygons, based on Kort og Matrikelstyrelsen (KMS; Danish Cadastral Survey) aerial photographs from August 1981 and 1987, were used to trace tidewater margins from the 1980s. Because these ice polygons were mapped at a range of scales (1 : 100 000, 1 : 250 000 and 1 : 500 000) and by unknown cartographers using aerial photographs of different years, we verified tidewater margins on original aerial photographs (Greenland Aerial Photo Database: (http://kmswww3.kms.dk/gronland/gronland_english.htm). Where confluent tidewater margins formed a continuous calving front, they were assigned a ‘fjord system’ code and the GLIMS-ID of the main flow unit. Thus, areal and effective length changes (terminus area change divided by the 1980s or 2000/01 margin widths) could be summed for individual glaciers and fjord systems.

Polygons representing areas of retreat or advance were constructed from the overlay of tidewater margin lines from the 1980s, 2000/01 and 2004/05. Because changes were often irregular along tidewater fronts, multiple polygons of advance and retreat occurred at many termini. By assigning all advance/retreat polygons the glacier GLIMS-ID, each advance (positive) or retreat (negative) polygon area was summated using pivot tables in Excel. In this manner, calving-margin area change between the 1980s and 2000/ 01 was obtained for 113 glaciers. Because it was not always clear whether the 1980s margin was mapped from the 1981 or 1987 KMS aerial photographs, we took 1984 as the beginning of this time period, so the time interval 1980s– 2000/01 was 16 or 17 years. Using the same advance/retreat area summation method, the net areal tidewater margin change was obtained for 78 glaciers for the period 2000/01 to 2004/05 (3–5 year interval).

We minimized the effects of seasonal and interannual fluctuations by using mid- to late-summer scenes only, and by measuring change over relatively long periods. Errors in georeferencing were 0–4 pixels with an average of 2 pixels (29 m). Manual digitization errors were within 1 pixel (14.5 m). The average cumulative error for the ASTER-to-ASTER or Landsat-7-to-ASTER calving margin is

Extrapolated over the entire calving margin width (196 km), this results in an absolute maximum regional error in margin retreat of 8 km², and a one standard deviation error of ~4km².

Fjord morphology

Fjord width, depth, length and shape can potentially influence tidewater margin stability (Reference Benn, Warren and MottramBenn and others, 2007) as well as the stability of ice melange, fjord water circulation and the intrusion of deep warm waters (Reference ChristoffersenChristoffersen and others, 2011). Due to lateral drag, a widening fjord may lead to a fluctuating ice margin (dynamic equilibrium), while a narrowing fjord may lead to a more stable equilibrium (Reference Benn, Warren and MottramBenn and others, 2007). Fjord width was approximated by the 1980s calving-margin width. For confluent tidewater margins, the sum of the widths of adjacent calving margins was taken as a proxy. Fjord shape was classified for the first 50–500m directly in front of tidewater margins, with class 1 = tidewater margin at fjord mouth, 2 = fjord width is constant, 3 = fjord widens, and 4 = fjord narrows away from the margin. Fjord length (distance between tidewater margin and fjord mouth) was measured by digitizing a centre line from the 2000s tidewater terminus to the fjord mouth. Coastal glaciers are more exposed to sea-ice, wind and wave action than inland fjords, secondary fjords and fjords on the lee side of islands. Fjord location may therefore determine the forces acting on calving fronts, including ice melange stability, fjord circulation and sea and air temperature (Reference MurrayMurray and others, 2010). Coastal glaciers terminating on Blosseville Kyst are directly exposed to the Atlantic Ocean and were classified as ‘outer fjord’, unless they were located on the lee side of coastal islands or in secondary-order fjords. These, and glaciers terminating in Kangerdlugssuaq fjord and Scoresby Sund were classified as ‘inner fjord’.

Glacier inventory

The Geikie Plateau region glacier inventory contains 332 glaciers, with a total area of 41 591 km². Glaciers range in size from 2 km² (thresholded) to 11 079 km² for Kong Christian IV Gletscher, which partly drains the Greenland ice sheet (Fig. 1). Glacier area is log-normally distributed, and the 2 km² glacier area threshold underestimates the total glacierized area by only 0.5–1%. Glacier types include ice caps, snowfields, and mountain, valley and outlet glacier systems, of which many are tidewater-terminating. Many glaciers form complex systems, with multiple tributaries and confluent units. Surge evidence is diagnostic for 56 glaciers (classes 2 or 3), and another 75 glaciers were assigned class 1 and are possibly of surge type (see previous section). Hence, we estimate the percentage of surge-type glaciers in the Geikie Plateau region to be 19–44%.

Most large outlet glaciers drain from ice masses at elevations above 2000ma.s.l. (Geikie Plateau ice cap, icefields around Watkins Bjerge, and the Greenland ice sheet (Kong Christian IV Gletscher)). Smaller glaciers drain from steep coastal cliffs to sea level. Tidewater margin widths range from 0.1 to 10 km, with a total calving-front width of 196 km. Roughly 90% of the total glacierized area drains through 120 tidewater glaciers (Fig. 1). It is the first time that these parameters of calving have been quantified in this region.

Calving-margin dynamics

Between the 1980s and 2000/01, 84 glaciers retreated (–41.3km²) and 29 glaciers advanced (+10.6km²), resulting in a net calving area loss of 30.7 ±4km². The overall yearly retreat rate for these 113 glaciers was 1.9 km² a^–1, but the vast majority (93) did not change significantly (<0.5km² over 14–20 years). However, 3 glaciers advanced significantly and 17 glaciers retreated significantly, and 3 small glaciers changed to land-terminating. The maximum retreat occurred in Storbræ S (7.6km²; Table 1), and the maximum advance in Sortebræ (6.7km²; Table 1; Fig. 2) due to its 1992–95 surge (Reference Jiskoot, Pedersen and MurrayJiskoot and others, 2001). Most other glaciers with retreats more than 1 km² have some indication of pre-1981 surging behaviour, although no observations of significant surge advances exist (Reference WeidickWeidick, 1988; Reference Jiskoot, Murray and LuckmanJiskoot and others, 2003). The largest glacier with retreat more than 1 km², the non-surge-type Kong Christian IV Gletscher, has a wide tidewater margin (~10 km) and limited effective length change (Table 1; Fig. 4 further below). Of all other glaciers only four advanced >0.5km². One of these is a newly detected surge of an unnamed former tributary of Bredegletscher (Table 1; ‘Tributary Brede’). On KMS aerial photographs from 1981 and 1987, the glacier appears depleted and is land-terminating, hence the surge took place sometime between 1987 and 2000, changing its margin to tidewater-terminating. Due to the lack of ice thickness data in this region, volume losses due to retreat are unattainable.

Fig. 2.

Tidewater margin changes of (a) unnamed complex in Kivioq Fjord (‘Rosenborg Basin’), (b) Kong Christian IV Gletscher, (c) Sortebræ and (d) Borggraven. See Figure 4 for locations. The yellow line is the glacier polygon from the Landsat 7 mosaic of 2000/01. Margin lines are from GEUS map polygons of the 1980s (orange) and ASTER scenes from 2000 (purple) and 2005 (green). Coloured circles correspond to areal retreat or advance between the 1980s and 2000 (Fig. 3a). Backgrounds are 2005 ASTER scenes.

Fig. 3.

Tidewater margin area changes for (a) 1980s–2000/01 and (b) 2000/01–04/05. Tidewater margin changes indicated with circles. Colours represent direction (red = retreat, green = advance, yellow = no significant change) and circle size represents magnitude of change.

Table 1.

Terminus fluctuations of 16 fjord systems and 11 of the largest non-confluent tidewater glaciers. Glacier names in italic are unofficial. See Figure 4 for locations. Retreat is the areal retreat (negative) or advance (positive) over the time period indicated. L _eff rate is the yearly rate of change in effective length (area divided by tidewater margin width). SUM tallies glaciers with measurements over both time periods (asterisked SUM values exclude Sortebræ). Glaciers are sorted from maximum retreat to maximum advance in the 1980s– 2000s period

Between 2000/01 and 2004/05 the cumulative net retreat loss (n = 78) was 26.3 km² with a cumulative annual areal retreat rate of 5.7 km² a^–1. This is approximately a threefold increase in retreat rate compared to the 1980s–2000/01 period (1.7km² a^–1 for the same 77 glaciers). However, when Sortebræ’s surge and retreat are disregarded, these values are 2.1 km² a^–1 (1980s–2000/01) and 3.9 km² a^–1 (2000/01–04/05), giving a twofold increase. A total of 46 glaciers retreated >0.1km² (maximum 7.4 km² for Sortebræ), 44 had no significant change (<0.1 km²) and only one glacier advanced >0.1km² (unnamed G329460E68679N in Kong Christian IV basin). ‘Rosenborg Basin’, a fjord system with four confluent tidewater margins (Fig. 2; Table 1), doubled its effective length retreat rate from 53ma^–1 to 100ma^–1. The westernmost flow unit, a tributary of Rosenborg Gletscher, increased its up-glacier flow speed between 2000/01 and 2005/06, which Reference Joughin, Smith, Howat, Scambos and MoonJoughin and others (2010) suggested may be indicative of surge activity. The rapid multi-year retreat and near-separation of its calving margin may also influence these upstream dynamics.

From the spatial distribution of retreat and advance over both time periods (Fig. 3), it is evident that glaciers along Blosseville Kyst generally have faster retreat rates than those in Scoresby Sund and Kangerdlugssuaq fjord. Increasing retreat rates after 2000/01 occur both in near-stationary margins of the ‘inner’ fjord-terminating glaciers (e.g. Magga Dan, Kista Dan, Nordfjord and Courtauld) and the near-stationary (e.g. Rosenborg) to significantly retreating (e.g. Bartolinbræ) glaciers along Blosseville Kyst (Fig. 4; Table 1). Exceptions to the increased retreat are Storbræ N and S which decreased their retreat rates by ~50% (Table 1). However, these are still some of the fastest-retreating glaciers, and may either have retreated into shallower waters (or a shoal), increasing the basal drag by reduced flotation, or may have slowed their post-surge thinning.

Fig. 4.

Effective length rate of change between the 1980s and 2000/01 for 16 fjord systems and 11 of the largest single glaciers. Names correspond to those in Table 1 and throughout the text.

Glacier type and fjord morphology

Of the 120 tidewater glaciers, 48 are outlet glaciers, 48 valley glaciers and 24 mountain glaciers. Average glacier areas are 514km² (without Kong Christian IV Gletscher), 31km² and 4.5 km² respectively. Confluence is common, and 41 tidewater margins make up 16 fjord systems (each with two to four confluent glaciers; Table 1). The majority of fjords are widening (70) or parallel (19). Only 10 glaciers are in fjords that narrow away from the calving front, and 21 are at the fjord mouth. Fjord lengths range from 0 to 24 km (excluding the distance to Scoresby Sund and Kangerdlugssuaq fjord mouths). Tidewater glaciers that retreated the most since the 1980s are all in widening or parallel fjords, and most are of the outlet type (Fig. 5). While the largest non-surge advance (Borggraven) occurred in a narrowing fjord, its margin is located at a widening point in the fjord. This 1 km² advance between the 1980s and 2000 (11ma^–1) was followed by a rapid retreat of 1.67km² between 2000 and 2005 (–51ma^–1 effective length change) (Fig. 2; Table 1). Surge-type glaciers evidently have much higher frontal variability than normal glaciers (Fig. 5). The magnitude of variability appears larger than the general glacioclimatic response over the two decades presented in this paper, as surge-type glaciers with some evidence of pre-1981 surges have amplified retreat rates during their quiescence as compared to normal glaciers. In conformity with the spatial pattern in Figure 3, coastal glaciers (‘outer’ fjord-terminating) have greater retreat and margin variability than glaciers in inner fjords (Fig. 5). Retreat rates were also correlated with fjord width, length to fjord mouth, glacier area and glacier length (Kong Christian IV, Sortebræ and ‘Tributary of Brede’ were removed from this analysis). Pearson’s correlation statistics suggest that only fjord width has a weak but significant (R ² = 0.23, α = 0.05, p = 0.000004) correlation with retreat; some glaciers in wider, deeper fjords retreat faster.

Fig. 5.

Box plots of 1980s–2000s change in tidewater margin for different classes of (a) glacier type, (b) surge type (ST), (c) fjord location and (d) fjord shape away from glacier margin. Kong Christian IV Gletscher is omitted.

Possible causes of spatial variability in retreat rates

While annual variability is large, long-term Greenland Sea summer sea-ice extent has decreased 11% per decade between 1978 and 2006 (Reference Parkinson and CavalieriParkinson and Cavalieri, 2008). As sea-ice concentration and surface melt are strongly correlated in central East Greenland (Reference Rennermalm, Smith, Stroeve and ChuRennermalm and others, 2009), it may be difficult to separate margin fluctuations resulting from changes in SST and sea ice, from those related to surface mass balance. The spatial pattern of inner and outer fjord tidewater margin behaviour may help identify additional controls. The Geikie Plateau region has large north–south and coast–inland climatic and sea-ice gradients (Reference Cappelen, Jørgensen, Laursen, Stannius and ThomsenCappelen and others, 2001). Yearly melt days along Blosseville Kyst (40–60) exceed those along Scoresby Sund (20–30) (Reference MoteMote, 2007). Between 1988 and 2004, melt along Blosseville Kyst increased by 18–24 days, melt in surrounding coastal areas increased by 6–18 days, while melt days were unchanged or decreased in Scoresby Sund (Reference BoxBox and others, 2006). Meltwater production changed accordingly, increasing up to 2000 mm along south Blosseville Kyst, and decreasing up to 500 mm on Geikie Plateau and along inner Scoresby Sund (Reference BoxBox and others, 2006). Additionally, Scoresby Sund and Kangerdlugssuaq experience strong katabatic ‘föhn’ (piteraq) winds from the ice sheet, while alongshore winds dominate the coast (Reference Cappelen, Jørgensen, Laursen, Stannius and ThomsenCappelen and others, 2001). Wind redistributes sea ice, influences ocean swell and waves, and can be an important factor in fjord water circulation (Reference StraneoStraneo and others, 2010). Therefore, coastal glaciers and those in inner fjords may have different atmospheric and sea-ice forcings. Moreover, although sea ice is more prevalent along the coast, SST in inland fjords is generally lower due to the relatively high contribution of glacial discharge.

Summer advection fog is frequent along the central East Greenland coast and is related to sea-ice break-up (Reference BrooksBrooks, 1979). At coastal stations Ittoqqortoomiit and Aputiteeq (Fig. 1), the May–September average number of fog days per month is 10–21, with the highest frequency in July (Reference Cappelen, Jørgensen, Laursen, Stannius and ThomsenCappelen and others, 2001). Sea breezes bring fog further into the fjords, but there it is quickly dissolved by radiation from the ice-free land. Further, strong off-ice winds remove sea fog and sea ice from fjords, forming a yearly polynya at the mouth of Scoresby Sund. As a result, coastal glaciers are more likely to be influenced by fog, which is known to suppress melt (Reference BrooksBrooks, 1979; Reference Mernild, Hansen, Jakobsen and HasholtMernild and others, 2008). It may be that the strong July correlation between sea-ice concentration and surface melt (Reference Rennermalm, Smith, Stroeve and ChuRennermalm and others, 2009) is in part due to the occurrence of sea fog. We observed coastal fog on our satellite imagery and on July– August MODIS quicklooks in other years. In years with low summer sea-ice concentration, coastal glaciers may experience enhanced melt while fjord-terminating glaciers will not, as sea fog is always more sporadic in fjords. Thus, variability in sea ice not only influences SST and wave action (dynamics at the tidewater terminus), but also the amount and frequency of coastal fog (surface melt). Trend analysis of fog days in East Greenland, and correlations to glacier behaviour and other factors may help unravel causes for differences in coastal and fjord-terminating glacier behaviour.

Another potential explanation of the differences may be derived from hypsometric analysis. From the ASTER GDEM we derived that glaciers north of the Geikie Plateau ice divide are more top-heavy and have steeper slopes at low elevations than along Blosseville Kyst. Outlet glaciers along Scoresby Sund and Kangerdluggsuaq have their lower 10– 20 km below 1000ma.s.l., while along Blosseville Kyst it can be 30–50 km. This, in combination with a lower albedo and a southerly aspect, results in a wider zone of negative mass balance along Blosseville Kyst (e.g. Reference BoxBox and others, 2006). Thinning and meltwater production may therefore be more important for the coastal region. It may therefore also be useful to study average glacier slope (Reference Joughin, Smith, Howat, Scambos and MoonJoughin and others, 2010) and hypsometry in relation to the response of tidewater glaciers.