The Cryosphere Multi-decadal marine-and land-terminating glacier recession in the Ammassalik region , southeast Greenland

Landsat imagery was applied to elucidate glacier fluctuations of landand marine-terminating outlet glaciers from the Greenland Ice Sheet (GrIS) and local landterminating glaciers and ice caps (GIC) peripheral to the GrIS in the Ammassalik region, Southeast Greenland, during the period 1972–2011. Data from 21 marine-terminating glaciers (including the glaciers Helheim, Midgaard, and Fenris), the GrIS land-terminating margin, and 35 GIC were examined and compared to observed atmospheric air temperatures, precipitation, and reconstructed ocean water temperatures (at 400 m depth in the Irminger Sea). Here, we document that net glacier recession has occurred since 1972 in the Ammassalik region for all glacier types and sizes, except for three GIC. The land-terminating GrIS and GIC reflect lower marginal and areal changes than the marine-terminating outlet glaciers. The mean annual land-terminating GrIS and GIC margin recessions were about three to five times lower than the GrIS marine-terminating recession. The marineterminating outlet glaciers had an average net frontal retreat for 1999–2011 of 0.098 km yr −1, which was significantly higher than in previous sub-periods 1972–1986 and 1986– 1999. For the marine-terminating GrIS, the annual areal recession rate has been decreasing since 1972, while increasing for the land-terminating GrIS since 1986. On average for all the observed GIC, a mean net frontal retreat for 1986– 2011 of 0.010± 0.006 km yr−1 and a mean areal recession of around 1 % per year occurred; overall for all observed GIC, a mean recession rate of 27 ± 24 % occurred based on the 1986 GIC area. Since 1986, five GIC melted away in the Ammassalik area.

The underlying mechanisms of GrIS marine-terminating glacier dynamics remain somehow unclear (Straneo et al., 2010;Johannessen et al., 2011).The frontal recessions at the calving front are highly due to changes in the force balance due to thinning, reduced resistive force, and speed-up based on warming of oceanic subsurface waters and warming of the atmosphere, where warm subsurface waters are suggested by Luthcke et al. (2006), Velicogna and Wahr (2006), Holland et al. (2008), Howat et al. (2008), Thomas et al. (2009), van den Broeke et al. (2009), Velicogna (2009), Murray et al. (2010), Rignot et al. (2010), Straneo et al. (2010), and Andresen et al. (2011) to play a significant role.However, Johannessen et al. (2011) stated, based on statistical correlations, that penetration of snow and ice melt water to the glacier bed might play an important role, influencing the GrIS sliding and dynamic processes.The mechanisms suggested for land-terminating GIC recession are less complex.In the Ammassalik region, Southeast Greenland, studies of GIC have shown that mass loss and margin retreat have Published by Copernicus Publications on behalf of the European Geosciences Union.been driven mainly by higher surface temperatures (increasing surface ablation) and decreasing precipitation (decreasing snow accumulation) since the mid-1900s (Mernild et al., 2011a).
The glacier contribution to sea-level rise from Greenland marine-terminating glaciers has been analyzed by satellite.Howat and Eddy (2011) identified changes in ice-frontal positions from 210 GrIS marine-terminating glaciers with fronts wider than 1 km (spanning nearly four decades, 1972-2010).These results show a trend of accelerated recession, where 90 % of the observed glaciers receded between 2000 and 2010.Box and Decker (2011) identified areal changes at 39 of the widest Greenland marine-terminating glaciers (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010).Collectively, the 39 glaciers lost a cumulative area of 1368 km 2 .For the Ammassalik region -a region including the Ammassalik Island, the Sermilik Fjord (the largest fjord system in Southeast Greenland) and its surrounding landscape -only frontal changes of major GrIS marine-terminating glaciers, such as Helheim, Fenris, and Midgaard glaciers, have been observed in earlier studies; however, at least 21 marine-terminating glaciers have been identified; all 21 were included in this study (Fig. 1).
For the land-terminating GIC in Greenland, margin retreat has been sparsely observed (Yde and Knudsen, 2007), and the only currently published time series of whole glacier insitu mass balance observations (since 1995/1996) is from the Mittivakkat Gletscher, located in the Ammassalik region (WGMS, 2009;Mernild et al., 2011a), even though thousands of individual GIC are located on the land-strip between the GrIS and ocean, of which several hundred are situated in the Ammassalik region (Mernild et al., 2012).Thus, there is a need for more information about contemporary glacier fluctuations of Greenlandic GIC and their coupling to climate change.
The ability to assess GrIS and GIC margin changes in the Ammassalik region has been improved through the use of Landsat imagery dating back to 1972.The imagery gives us the possibility to map "snapshots" and the averaged behaviour of glacier changes for the past four decades for the identified marine-terminating glaciers, the GrIS landterminating margin, and the GIC during a period of climate warming.The average multi-decadal glacier recession in the Ammassalik region (65 • N, 37 • W) was examined, rather than the annual range of variability, even though recent observations suggest that major changes in the dynamics of Greenland marine-terminating glaciers take place over timescales of 3-10 yr (Howat et al., 2007;Nick et al., 2009;Andresen et al., 2011;Johannessen et al., 2011), rather than over several decades or centuries as previously believed (Truffer and Fahnestock, 2007).
Here, on approximately decadal scale we examine net frontal position and area fluctuations using multispectral Landsat satellite data, observing 21 marine-terminating glaciers from 1972-2011, land-terminating glacier frontal positions for parts of the GrIS, and of 35 GIC from 1986-2011 for the Ammassalik region -a region including the thermodynamic transition zone from the North Atlantic Ocean into the Arctic Ocean through the Denmark Strait.Changes were considered in the context of meteorological observations and reconstructed ocean water temperature time series.As part of the study, the cumulative net area changes for the marine-terminating glaciers were examined.Finally, we investigate differences in marginal change rates between the marine and the terrestrial glacier environments.

Data and methods
The Landsat and Advanced Spaceborne Thermal and Reflection Radiometer (ASTER) Global Digital Elevation Model Version 2 (GDEM v2) scenes were selected and acquired through WIST (http://reverb.echo.nasa.gov/reverb/redirect/wist: EOSDIS, 2009): Landsat 1 carried the four-band Multispectral Scanner (MSS); Landsat 5 carried the seven-band Thematic Mapper (TM); and Landsat 7 carried the eightband Enhanced Thematic Mapper Plus (ETM+), having a ground resolution of ∼60, 30, and 30 m (15 m in the panchromatic band 8), respectively.The ASTER GDEM v2 provides a ground resolution of 30 m (Table 1) and is used to estimate minimum, mean, and maximum GIC elevations.
Data were obtained from seven relatively cloud-free scenes (>25 % cloud free) covering the Ammassalik region at the end of the ablation period (mid-August through mid-September) for the time series analysis (i.e., 1972, 1986, 1999, 2007, and 2011, Table 1).For 1972 the GrIS and GIC land-terminating margin analyses were omitted due to difficulties in separating ice cover from snow cover; the year 1972 was one of the coldest years during the study period, and the year with the lowest satellite-derived melt extent cover for GrIS (Mernild et al., 2011b).For 2011, scenes from 2007 were used for filling gaps, due to the ETM+ sensor Scan Line Instrument (SLI) malfunction.Because of the four year gap between 2007 and 2011, the gapfilled procedure was used only when absolutely necessary to Fig. 2.An example of the multi-criteria analysis for the Landsat 5 TM and Landsat 7 ETM+ scenes.For the latter the corresponding values were 1999 NDSI > 0, NDWI > 0.16 and <0.45, and NDVI < 0, and for 2011 NDSI > 0, NDWI > 0.15 and <0.5, and NDVI < 0. Band ratios were the same for all scenes.The analysis was carried out using the tool BandMath in ENVI ™ .minimize misclassification.All imagery and features were projected in WGS84, UTM Zone 24 N, and the selected scenes were radiometricly calibrated using the Landsat calibration tool in ENVI ™ software package (http://www.ittvis.com/ProductServices/ENVI.aspx), converting the band values to "At Surface Reflectance".The individual bands (TM and ETM+ bands 1-5, and 7, and MSS bands 2-7) were standardized using the ENVI ™ Dark Subtract (DS) tool before ratio and indices were calculated.Standard pixel errors associated with the different scenes and sensors were ±30 m (MSS), ±15 m (TM), ±15 m (7.5 m panchromatic) (ETM+), and ±30 m (ASTER GDEM v2).The error associated with the ASTER GDEM v2 is expected to be ∼ ±12.9 m vertically (Table 1); however, larger uncertainty might occur vertically in steep terrain and in areas of poor contrasts (Tachikawa et al., 2011).
The image co-registration errors associated with the individual sensor types were 39.86 m (MSS-TM) and 11.33 m (TM-ETM+) (due to the root mean square, RMS), based on 27 near-sea ground control points (GCP) for each sensor type (Table 1).The supervised classification process used for the 1986, 1999, and 2011 scenes was based on a multi-criteria analysis involving the calculation of a set of indices (Fig. 2): normalized difference snow index (NDSI; Dozier and Warren, 1982); normalized difference water index (NDWI; Gao, 1996); NDVI (normalized difference vegetation index; Rouse et al., 1973); and RATIO bands 3/5 TM/ETM+.The RATIO was used in reference due to better performance than the NDSI index in mountainous areas (like the Ammassalik region) capturing ice and snow covered areas influenced by shadows and debris (Paul, 2004).The NDVI was used to filter out vegetation and the NDWI to identify and filter out lakes in the margin area of the GrIS and GIC.The resulting classifications were converted to (1) polygon files and cleaned up manually in ESRI ™ ArcMap by visual inspection, (2) poly-line files for visual presentation; and (3) point shapes with a 30-m distribution along the lines and used to examine the mean and median center of the margin distribution for each year.The 1972 scene was not included in the classification process since the spectral bands of the Landsat 1 MSS sensor used different wavelengths and band distribution than both the TM and ETM+ sensors, and thus is not eligible for this type of snow/ice classification.The raw model precision errors for each year compared with the cleaned up classification were found to be 5.2 % (overestimate by 3.4 % due to snow patches and underestimate by 1.8 % due to heavily debris covered terrain and shadow regions) overall, with 1999 being the most troublesome year (Table 1).
The GrIS and GIC margin positions were digitized for each of the years, and the distance between the margins was calculated at the dominant glacier's flow direction using a centerline method approach.The expected errors related to the classification and determination of the glacier margin positions are shown in Table 1.
The selection of the 35 GIC was randomly chosen: (1) following the regional distribution due to size, aspect, and elevation, and (2) trying to avoid area where the 2011 scene SLI failure could influence the classification.The 2007 scene gap-fill was used in reference only when absolutely necessary.
The Landsat-derived 2011 (14 August) GIC margin was validated for the Mittivakkat Gletscher against the 2011 (9 August) GPS-observed margin; only the lower elevated margin of the Mittivakkat Gletscher was observed.The location of the observed margin was obtained from portable single-frequency GPS measurements having a relative uncertainty of about ±5 m (Mernild et al., 2011a).Overall, the RMS difference between the 2011 satellite and GPS margin observations was 22 m (Fig. 3 and Table 1).The observed meteorological data, air temperature and precipitation were obtained from the Danish Meteorological Institute (DMI) station in Tasiilaq for 1972-2011, located less than 10 km from the Sermilik Fjord outlet, and reconstructed ocean subsurface water temperatures at 400 m depth in the Irminger Sea were used as a proxy for the variability of the subsurface warm Atlantic water in the Sermilik Fjord (Johannessen et al., 2011).
Several authors (e.g., Holland et al., 2008;Murray et al., 2010;Straneo et al., 2010;Andresen et al., 2011) suggest that ocean warming and warm subsurface waters caused large changes to submarine melting, marine-terminating glacier frontal positions and thinning, reduced resistive stress and ice discharge acceleration.In recent decades the 21 marineterminating glaciers in the Ammassalik region have experienced increasing frontal recession rates and decreasing area exposure rates that are synchronous with both increasing mean annual air temperature (MAAT) (∼0.06 • C yr −1 , significant at p < 0.01, where p is level of significance) from the DMI meteorological station in Tasiilaq and reconstructed annual ocean water temperature (∼0.12 • C yr −1 , significant p < 0.025; at 400 m depth in the Irminger Sea penetrating into the Sermilik Fjord and exposing the lower part of glaciers such as Helheim and Fenris to warm waters with temperatures up to 4 • C; Johannessen et al., 2011) (Figs.5b  and 6).The mean glacier retreat was more widespread for 1999-2011 (approximately the first decade of the 21st century) than for the earlier sub-periods (1972-1986 and 1986-1999).The observed accelerated recession of the marineterminating glaciers in the Ammassalik region coincided with the onset of a warming trend in the sub-polar North Atlantic Ocean (Myers et al., 2007;Straneo et al., 2010), likely initiated by the influx of warmer deep water originating in the Irminger Sea (Holland et al., 2008;Hanna et al., 2009).This supports the hypothesis that ocean warming associated with shifts in the Irminger and East Greenland currents caused increasing submarine melt at the iceocean interface, and retreat, thinning, and acceleration as the loss of resistive stress at the terminus recedes (Joughin et al., 2010).However, oceanographic studies have demonstrated that although subtropical ocean waters reach glacial fjords in Southeast Greenland, there is no proof that they come into direct contact with glaciers (Walsh et al., 2012).Mechanisms driving the circulation of warmer North Atlantic waters are, however, still not well understood (e.g., Straneo et al., 2010).On the other hand, Johannessen et al. (2011) have argued (due to statistical correlation) that, based on annual frontal positions of Helheim Gletscher, 24 % of the ice-front fluctuations could be accounted for by ocean temperatures and 56 % by air temperatures, even though changes in frontal positions were influenced by a number of local factors, e.g., up-glacier ice dynamics and bed geometry.Overall, several studies using a range of different methods show that GrIS marine-terminating glaciers recede and mass loss might be influenced by atmospheric and probably more importantly by oceanographic impacts, especially in the southeastern part of the GrIS (Luthcke et al., 2006;Velicogna and Wahr, 2006;van den Broeke, 2009;Velicogna, 2009).

The land-terminating ice sheet
In Fig. 7 the GrIS land-terminating margin and changes within the Ammassalik region are illustrated for 1986for , 1999for , and 2011for . Since 1986 the GrIS area has decreased in size within the Ammassalik region from 1166 km 2 (1986), 1153 km 2 (1999) to 1124 km 2 (2011), indicating a net area loss of 4 % (equal to an area exposure rate of 0.15 % yr −1 ) (Table 3).As such for the land-terminating GrIS, the area exposure rates were 1.0 km 2 yr −1 (13 km 2 ) and 2.4 km 2 yr −1 (29 km 2 ) for the sub-periods 1986-1999 and 1999-2011, Table 3. GrIS land-terminating margin-position change rate statistics for each survey year and period for the Ammassalik region.

Years
GrIS margin length for the highlighted section in Ammassalik region (km) GrIS area for the section for the highlighted section in Ammassalik region (km respectively, indicating an increasing trend in area exposure since 1986.The land-terminating area exposure was unevenly distributed for the GrIS (Fig. 7).A division of the ice sheet into 100-m elevation bands indicated that the largest GrIS area recession occurred at the elevation between 701-800 m a.s.l. for both survey periods, with rates of 0.22 km 2 yr −1 (1986-1999) and 0.50 km 2 yr −1 (1999-2011) (Figs. 7 and 8).Along with this area reduction, the GrIS landterminating margin decreased ∼20 % in total length from 686 km (1986) to 544 km (2011), because the 2011 margin was less curved -had fewer land-terminating outlets -than in 1986 and 1999.When the area recession is compared to changes in margin length, the largest GrIS area length recession ratio occurred at the elevation >800 m a.s.l., most pronounced for the period 1999-2011 (Fig. 8).The spatial area recession seems to be highly influenced by local topography, hypsometry, shadow effects, climate variability, glacier dynamic processes within the GrIS, increasing ELA elevation (the ELA is the spatially averaged elevation of the equilibrium line, defined as the set of points on the glacier surface www.the-cryosphere.net/6/625/2012/The Cryosphere, 6, 625-639, 2012 where the net mass balance is zero), and the margin elevation distribution, where approximately 20 % of the margin was located between 701-800 m a.s.l.(1986-2011) (Fig. 8).
The mean net recession rate of the GrIS land-terminating margin was 0.018 ± 0.009 km yr −1 (equal to a net recession of 0.443 km) for 1986-2011, comprised of a mean recession rate of 0.010 km yr −1 (0.127 km) for 1986-1999, and 0.026 km yr −1 (0.316 km) for 1999-2011 (Table 3).This land-terminating recession rate for the GrIS over this period of 1986-2011 is about three times lower the mean rate of recession of the marine-terminating GrIS.Sohn et al. (1998) measured recession rates of the GrIS landterminating margin near Jakobshavn Isbrae, West Greenland, of 0.016-0.040km yr −1 , averaging 0.026 km yr −1 for 1962-1992.This may suggest that the recession rate can be expected to be within this order of magnitude along many parts of the GrIS land-terminating margin.
Net GrIS land-terminating marginal recession for the Ammassalik region, including increasing area exposure, occurred for the period 1972-2011, during a period of increasing MAAT (∼0.06 • C yr −1 ) and decreasing annual precipitation (−7.0 mm water equivalent (w.e.) yr −1 , significant at p < 0.025; Fig. 5b) -probably heading towards future warmer and drier conditions in the region (Mernild et al., 2011a).The average increase in MAAT generally favors surface ablation (evaporation, sublimation, and melt), and an earlier start of the ablation season by decreasing the "cold content" of the snowpack (Bøggild et al., 2005;Mernild et al., 2011a), whereas a decrease in annual precipitation may lead to earlier exposure of glacier ice melt and summer firn surface of previous years (having a lower albedo than fresh snow, promoting increased solar absorption).Therefore, the combination of increasing air temperature and decreasing precipitation is likely to increase ablation and GrIS margin thinning and recession, and if MAAT and precipitation continue to follow these trends, then it is expected that the GrIS land-terminating margin will continue its recession, leading to increased area exposure.However, changes in the hypsometric distribution along the GrIS margin may influence recession rates on a decadal timescale.

Land-terminating glaciers and ice caps
Peripheral to the GrIS, 35 land-terminating GIC were chosen (Fig. 1) to assess area exposure for the Ammassalik region for 1986, 1999, and 2011 based on Landsat imagery.In Fig. 9 the size, mean elevation, and aspect distribution are illustrated for the 35 GIC, indicating that the majority of the GIC is below 5 km 2 , located between 400-800 m a.s.l., and facing south, west, and northwest.The GIC are non-surging glaciers located south of the East Greenland surge cluster (Jiskoot et al., 2003).For the Ammassalik region, the observed GIC indicated a relative mean area exposure of 4 ± 18 % for 1986-1999, and 27 ± 24 % for 1986-2011, which is equal to a mean net area exposure rate of 0.04 km 2 yr −1 per glacier (or 1.07 % yr −1 per glacier) (Fig. 10a and b).For small GIC (n = 32; <10 km 2 ), the net area exposure rate was on average 1.08 % yr −1 , and for large GIC (n = 3; >10 km 2 ) was a comparable rate of 0.81 % yr −1 .For 1986-1999, eleven individual GIC (around 30 %, mostly below 2 km 2 ) had a net increase in area, while for 1986-2011 there were only three GIC all facing towards the west (L11, L14, and L9; and around 10 % -all less than 1 km 2 ) (Fig. 10b).As illustrated in Fig. 10c, GIC having a mean elevation height higher than 705 m a.s.l. had in general a net area increase from 1986 to 1999, while glaciers with a mean elevation lower than 705 m a.s.l. had a net area decrease (based on the significant linear regression; r 2 = 0.38; p < 0.01).The height of 705 m a.s.l. was around the observed average ELA of 690 m a.s.l. at the Mittivakkat Gletscher in the late 1990s (Knudsen and Hasholt, 2004;Mernild et al., 2011a; 1986-1999, 1999-2011, and overall for 1986-2011.For 1999-2011 the linear regression shown in Fig. 10c indicates the opposite trend for GIC in the Ammassalik region: an increase in area recession for GIC at high elevation ranges, and vice versa.This shift in trend occurred simultaneously with an increase in the average observed ELA for the www.the-cryosphere.net/6/625/2012/The Cryosphere, 6, 625-639, 2012 Mittivakkat Gletscher to 750 m a.s.l.Overall for 1986-2011, the observed GIC faced a general net area loss that was highest at low elevations, and vice versa (based on the linear regression; Fig. 10c).However, as previous mentioned, three minor GIC (<1 km 2 ) had a net area gain during this 25-yr period, indicating that glacier fluctuations may vary on local scales.
Since the GIC on average had a mean net area exposure rate of about 1 % yr −1 , it may be expected that GIC in the Ammassalik region could melt substantially in the 21st century under ongoing climate change.For the period 1986-2011, there are examples of five glaciers that completely melted away: all located at different mean elevations within the region from 460 to 1110 m a.s.l., and with different aspects facing from east to west (Table 4).The recession seems therefore not to be limited to low elevated areas only, but more likely to occur for north-facing GIC.Also, in Fig. 11, examples of eight GIC are shown to illustrate the spatial changes in margin location from 1986 to 2011.
For the largest GIC -the Mittivakkat Gletscher (26.2 km 2 in 2011) -the area extent had diminished about 18 % since 1986 (lower than the mean GIC area exposure for the Ammassalik region of 27 ± 24 %).The terminus has retreated by 1.6 km (0.015 km yr −1 ) since the maximum extent of the Little Ice Age around 1900, by 1.3 km (0.017 km yr −1 ) since 1931 (Humlum and Christiansen, 2008;Mernild et al., 2011a), and by 0.3 km (0.013 km yr −1 ) since 1986.This is almost of the same magnitude as the GIC of the Ammassalik region's mean net recession rate of 0.010 ± 0.006 km yr −1 (1986-2011), and of the regional GrIS landterminating margin of 0.018 ± 0.009 km yr −1 (see Sect. 3.2): the mean Ammassalik GIC land-terminating recession rate  is about five times lower than the mean GrIS marine-terminating recession rate.Also, for Mittivakkat Gletscher the annual mass balance measured continuously since 1995/1996 illustrates a 16-yr average mass loss of 0.970 ± 0.190 m w.e.yr −1 , and an accumulation-area ratio (AAR: the ratio of the accumulation area to the area of the entire glacier) of ∼0.10 (updated from Mernild et al., 2011a), indicating that the glacier is significantly out of balance with the current climate.The glacier will likely lose at least 70 % of its current area extent and 80 % of its volume even in the absence of further climate changes (Mernild et al., 2011a).Since the initiation of the mass balance observation program in 1995/1996, Mittivakkat Gletscher had in 14 out of 16 yr a negative surface mass balance, while the general climatic trend in the region has been towards higher temperatures, less winter precipitation, and more negative glacier mass balances and continuous marginal recession (Fig. 11).Consecutive record glacier mass loss occurred for the years 2009/2010 and 2010/2011 of −2.16 and −2.45 m w.e.yr −1 , respectively.The 2011 mass loss was not only the largest annual loss of volume in the history of the mass balance observational program, but also the largest annual loss in simulations of glacier mass balance changes back to 1898 (Mernild et al., 2008).The marginal recession, mass balance and AAR observations suggest that recent Mittivakkat mass losses, which have been driven largely by higher surface temperatures and less solid precipitation, are representative of the broader region, which includes the 35 observed GIC in Figs. 10 and 11 -glaciers of different sizes and elevation ranges.This is confirmed since the Mittivakkat Gletscher net area exposure rate closely follows the average rates for the Ammassalik region.
Glacier fluctuations and area exposure have been studied in other parts of Greenland.North of the Ammassalik region, in central East Greenland (68-72 • N) land-terminating GIC peripheral to the GrIS have receded at a mean rate of 0.010 km yr −1 for a wide range of glacier sizes (2002-2009) (Kargel et al., 2012).Also, on Disko Island in West Greenland, Yde and Knudsen (2007) estimated mean GIC terminus retreat rates of 0.008 km yr −1 for non-surging GIC, and 0.020 km yr −1 for quiescent phase surge-type GIC, reflecting a higher non-climatic-driven recession rate after glacier surges.These studies are in accordance with the findings for the Ammassalik region and indicate that the current mean recession rate for GIC in Greenland (probably excluding North Greenland where no data are currently available) is likely to be on the order of 0.008-0.010km yr −1 .

Summary and conclusion
The satellite observations show net glacier recession since 1972 for the Ammassalik region for all glacier types and sizes, both at the marine-terminating and land-terminating GrIS, and land-terminating GIC.However, the landterminating GrIS and GIC reflect slower area exposure rates than the faster marine-terminating outlet glaciers.This could likely be due to a combination of effects, as the marineterminating GrIS was influenced by the onset of a sea-water warming trend in the North Atlantic Ocean (Myers et al., 2007;Straneo et al., 2010), and by atmospheric impacts from regional trends in MAAT and precipitation, while GIC were only influenced by the latter.For the marine-terminating GrIS outlet glaciers the mean annual area exposure rate has decreased since 1972, whereas it has increased for the landterminating GrIS margin since 1986, even though both parts of the GrIS have undergone substantial area changes in the past decades.The observed land-terminating GrIS and GIC indicate a net area recession of 4 % (equal to an area exposure rate of 0.15 % yr −1 ) and 27 ± 24 % (around 1 % yr −1 ), respectively, and margin recession rates of 0.018 ± 0.009 and 0.010 ± 0.006 km yr −1 .These mean net margin recession rates are about three to five times lower than the GrIS marineterminating margin rates.If these GIC recession trends were extrapolated, it would indicate that a substantial amount of the GIC in the Ammassalik region might melt away within the 21st century under ongoing climate warming.So far, five GIC in the Ammassalik region have melted away since 1986.

Fig. 1 .
Fig. 1.The Ammassalik region, including the Sermilik Fjord and parts of the southeastern sector of the GrIS.The marine-terminating glacier margins are marked with positions (lines) for each of the survey years: 1972 (red color), 1986 (green), 1999 (blue), and 2011 (yellow) (the location names are written in turquoise).The 35 local land-terminating glaciers and ice caps (GIC) peripheral to the GrIS are written in yellow, where eight are marked with shaded areas; marginal changes for those eight GIC and the GrIS are illustrated in detail.The GIC marked with L33-L37 are examples of GIC that melted away during 1986-2011.The inset figure indicates the general location of the Ammassalik region in Southeast Greenland (source: Landsat 7 ETM+ Mosaik, 7 September 1999/9 September 2000).

Fig. 3 .
Fig. 3.A comparison between the Landsat-derived Mittivakkat Gletscher margin and the observed margin based on portable GPS observations (green line with dots) for August 2011.The inset figure indicates the location of the observed margin at Mittivakkat Gletscher.The oblique black lines are due to the SLI malfunction.

Fig. 5 .
Fig. 5. (a) Box plots of frontal-position change for the sample 21 marine-terminating glaciers in the Ammassalik region with measurements in each of the1972-1986, 1986-1999, and 1999-2011  survey periods.The edges of the boxes denote 25 % and 75 % percentiles and the vertical line mean.Data points outside this range are considered outliers and are plotted as crosses and labeled; (b) mean annual air temperature anomaly (observed at the DMI meteorological station in Tasiilaq), mean annual ocean water temperature anomaly at 400 m depth in the Irminger Sea(Johannessen et al., 2011), mean annual precipitation anomaly (uncorrected) (observed at the DMI meteorological station in Tasiilaq) and standard deviations are shown.

Fig. 6 .
Fig. 6.Mean annual area change rates and cumulative net area change for all 21 marine-terminating glaciers, the five GrIS outlet glaciers, and other sixteen outlet glaciers for the Ammassalik region 1972-2011.

Fig. 7 .
Fig. 7.The location of the GrIS land-terminating margin for the survey years 1986 (green), 1999 (blue), and 2011 (yellow) in the Ammassalik region, and the marginal changes between the survey years.The topography has black and gray shaded colors (source: ASTER GDEM v2 and Landsat 7).

Fig. 8 .
Fig. 8. GrIS land-terminating area recession rates for different elevation intervals, Ammassalik region, for the sub-periods 1986-1999 (light gray) and 1999-2011 (dark gray), area recession rate related to changes in margin length for 1986-1999 (black dashed line, long dashed) and 1999-2011 (black dashed line, short dashed), and the percentage of margin elevation for the different elevation intervals for the years 1986 (green), 1999 (blue), and 2011 (yellow) (see Fig. 7 for location of the margin).

Table 1 .
Satellite platform, sensors, band information, scenes used in the analysis, and uncertainties related to the satellite classification process.

Table 4 .
Characteristics of land-terminating GIC L33-L37 in the Ammassalik region, which have melted away during the period 1986-2011 (see Fig.1for location of the doomed glaciers).