1. Introduction
Glacial retreat is a dynamic process largely influenced by the internal glacier dynamics [
1,
2], bedrock topography [
3,
4] and the response to climatic changes [
5,
6]. Even though the front position changes of fast flowing glaciers can be asynchronous to climatic forcing [
3,
7,
8,
9], over a long-term period, tidewater glacier termini retreat can be assumed to be an indicator of the atmospheric temperature rise [
10,
11]. Therefore, a globally observed retreat of glacial termini [
12,
13] is considered one of the most important indicators of climate change [
14].
Since the 1950s, the Antarctic Peninsula region was subjected to an unprecedented air temperature rise of
C/year [
15]. This led to a large scale retreat of glacial fronts an rapid glacial change, including the disintegration of the Larsen A and B ice shelves [
16,
17]. However, it was shown by Turner et al. [
18], Oliva et al. [
19] that since 1999 a regional cooling has been observed, which has also influenced the behaviour of the cryosphere [
19,
20]. Air temperature records from the Bellingshausen Station (
Figure 1; [
21]) confirmed that for the 2012–2015 period a significant cooling was observed during summer months on the Fildes Peninsula, King George Island. Overall, the ice caps and glaciers of this region are one of the major sources of current sea level rise contributing 0.22 mm
[
22], compared to 0.47 ± 0.23 mm
for the entire Greenland Ice Sheet between 1991–2015 [
23]. Nonetheless, there is a large uncertainty concerning the volume change of glaciers in the Antarctic periphery, as the ground based measurements are relatively scarce [
20,
24].
King George Island is the largest island in the South Shetland Islands archipelago (
Figure 2a,b). A major part of the island is covered by an ice cap divided into five main ice basins: three connected domes along the main ridge (A-Bellingshausen, B-Arctowski and C-unnamed), and Kraków and Warszawa Icefields on the southern peninsulas [
25]. Despite there being a number of published glacial extent change studies (e.g., [
25,
26,
27,
28,
29]), only a few studies of glacier volume change in the South Shetland Islands area have been conducted (e.g., [
20,
26,
30]). The majority of ice volume change studies covering this region are still based on volume-area scaling (e.g., [
22,
31]), with little in situ data that could provide robust validation. The study by [
26] recorded surface lowering over the entire Bellingshausen Dome on King George Island, including at the summit, at 270 m a.s.l. However, given that the maximum elevation of the King George Island ice cap is much higher, approximately 700 m a.s.l. [
25], there is a large uncertainty regarding overall volume change for the island, and mass balance modelling indicates that it is out of balance [
32]. There have been several attempts to quantify the surface mass balance of King George Island glaciers (e.g., [
28,
32,
33]) but little is known about their dynamic response and hence overall ice elevation change.
Located only 1 km south of the Arctowski Polish Antarctic Station, Ecology Glacier is the northernmost outlet glacier of Warszawa Icefield (
Figure 2c), south of the ice-free area of Point Thomas Oasis. It calves into the Suszczewski Cove on the western coast of the Admiralty Bay. According to the Randolph Inventory, Ecology Glacier had an area of 5.21
and was 4.2 km long [
34]. The bedrock topography of Ecology Glacier remains unknown, nonetheless it can be assumed that subglacial relief is highly complicated with multiple subglacial landforms, as has been observed beneath the neighbouring Bellingshausen and Arctowski Icefields [
35,
36]. This assumption can be supported by the presence of significant undulations of the ice surface of the glacier. The glacier has been a subject of interdisciplinary research for almost 40 years, including studies of glacier mass balance [
28,
33], glacio-meteorology [
37] and glacial microbiology [
38,
39]. Although the mass balance [
28,
33] and ice extent change over several periods has been reported elsewhere [
26,
28,
40,
41,
42], a detailed assessment of ice elevation change is still lacking.
The main goals of this paper are to quantify ice elevation change of Ecology Glacier, an outlet glacier of King George Island ice cap between 1979 and 2016, and to partition ice elevation changes related to frontal retreat and surface lowering.
2. Materials and Methods
This study computes ice elevation changes by differentiating Digital Elevation Models (DEMs) constructed using archival maps, aerial panchromatic images, high-resolution satellite imagery, terrestrial laser scanning and bathymetric data.
2.1. 1979 DEM—Aerial Photographs
During the austral summer of 1978/1979, a series of airborne imagery was taken to map the Admiralty Bay area [
43]. These images were subsequently used by Pudełko [
44] to create an orthophotomap of Antarctic Specially Protected Area (ASPA) 128. 1:3600 scale images were taken with a spy camera AFA-21 mounted on a Mi-2 helicopter. The original images were scanned by Pudełko [
45] and stored in digital format. Agisoft Photoscan was used to orthorectify 14 photographs and to create a 3D model of the terminal section of Ecology Glacier and the ice-free area north of it using the Structure-from-Motion method [
46,
47]. Twenty two ground GCPs located on bed outcrops and Arctowski Station buildings were used for georeferencing, their positions were measured on the 2016 DEM and yielded a mean elevation error of 2.43 ± 7.24 m.
2.2. 2001 DEM and Archival Maps
Three maps were used as the basis for this study: (1) 1:50,000 scale topographical map of the Admiralty Bay [
43] that was further improved with topographic works conducted in 1988/89 [
48]; (2) 1:12,500 scale topographical map of SSSI-8, now called ASPA 128 [
45]; and (3) 1:10,000 scale orthophoto map of Western Shore of Admiralty Bay, King George Island, South Shetland Islands [
44]. For construction of the DEMs based on maps [
44,
45], the original isolines were provided and used, while isolines on the 1990 map were digitized manually in ArcGIS. The DEM for 2001 was computed in ArcGIS software by the TopoToRaster interpolation of contours taken from a 1:12,500 topographical map [
45], based on data collected during the XXV Antarctic Expedition (2000/2001) with use of two double frequency GPS receivers Ashtech Z-12. According to [
45], the precision of surveyed points was better than 100 mm. However, in some cases obtaining the points was impossible because of the loss of communication with satellites or due to difficult terrain. In these situations, gaps were interpolated based on aerial photos from 1979 and theodolite measurements (for more details see [
45]).
Maps [
43,
48] have large distortions and could not be used for the purpose of this study. For further analysis only the maps of [
44,
45] were used, for which the interpolation error was calculated as the elevation difference with 2016 DEM over unglaciated areas of Point Thomas Oasis north of Ecology Glacier, yielding mean elevation error of 1.26 ± 6.61 m. However, it cannot be excluded that the error over ice-covered areas might be higher.
2.3. 2012 Pléiades DEM
The DEM for the lower Ecology Glacier, and surroundings, derived for 2012 was based on an along-track, tri-stereo set of Pléiades 1A panchromatic images collected on 25 December 2012 (Centre National d’Etudes Spatiales, Paris, France) and a set of six ground control points (GCPs). The tri-stereo set was selected because it presented the lowest percentage of cloud cover for available images during the 2012/13 summer, and it only contains a small seasonal snow coverage which facilitates the delineation of the glacier. For DEM construction, the original 12 bit encoding of the panchromatic band was used instead of the optional 8 bit encoding, as the former provided improved image contrast over flat and featureless areas of the glacier [
49,
50].
The GCPs used to refine the DEM from relative to absolute elevations were collected in December 2015 utilizing a Trimble Zephyr antenna and Pathfinder ProXRT receiver differential GPS (dGPS) system (Trimble Inc., Sunnyvale, CA, USA). The mean horizontal and vertical precision of the GCPs were 0.19 m and 0.21 m, respectively, which were largely influenced by poor satellite configuration due to the latitude, even though occupation times were relatively long (10–20 min).
A bundle-adjustment was undertaken using Imagine Photogrammetry 2014 software (Hexagon Geosystems, Heerbrugg, Switzerland), which incorporated the panchromatic images, the six GCPs, forty tie points and the initial orientation data of each image from the original Rational Polynomial Coefficients (RPCs). The overall accuracy of the tri-stereo bundle-adjustment was a root mean squared error (RMSE) of 0.45 pixels (∼0.25 m). An automatic DEM extraction procedure was employed with all geometric parameters associated with the three overlapping images, by matching conjugated points from image overlaps and retrieving their 3D coordinates. It is important to note that by using triplet matching, redundancy is introduced into the derivation of elevation which improves the accuracy of the final DEM product [
50]. Finally, the irregularly distributed 3D points were interpolated to a raster grid DEM using a non-linear interpolator with a sampling resolution of 0.5 m.
2.4. 2016 Terrestrial Laser Scanning DEM
In the austral summer 2015/2016, a detailed survey of the main trunk of Ecology Glacier and its surroundings was made with a Riegl VZ-6000 long-range terrestrial laser scanner (RIEGL Laser Measurement Systems GmbH, Horn, Austria). The measurement principle of this instrument is time-of-flight with online waveform processing. This instrument is characterized by a very long effective range over the ice and snow surfaces due to the used wavelength of the laser beam (1065 nm) and has been used in various glaciological applications (e.g., [
51,
52]). The glacier surface was surveyed from seven positions (
Figure 2,
Table A1): three near the front (Ecology E, N and S), one near the equilibrium line (Wróbel Hill) and three overlooking the accumulation zone (Dutkiewicz Cliff E, N and S) on three separate dates: 27 January, 17 and 21 March, 2016. The position of the Terrestrial Laser Scanner (TLS) was measured with a differential GNSS (GPS+GLONASS) receiver Leica GS14. The base station (Leica GS10 receiver) was located at Jasnorzewski reference point in the vicinity of Arctowski Polish Antarctic Station (
Figure 2,
Table A1). Measurements were made in static mode and post-processed with Leica Geo Office software, yielding mm-scale 3D accuracies. As the glacier surface could only be surveyed from the northern side of the glacier, a series of additional TLS surveys of the surrounding terrain was conducted in order to increase the accuracy of georeferencing (
Figure 2). This provided a larger baseline (4269 m in N-S and 8546 m in E-W direction) for the survey, decreasing the georeferencing error.
TLS data was processed with RIEGL RiSCAN PRO, a proprietary software provided by the manufacturer of the scanner. First, point clouds were processed with the Multiple-Time-Around (MTA) package to account for long-range multiple echoes enhancing the effective range of the measurements to 6000 m. Next, artifacts due to reflection from the presence of falling snow during the data acquisition, reflection from sea surface and sea ice present in Suszczewski Cove were removed. Snow and ice surfaces were eliminated manually from point clouds. Subsequently, all point clouds were aligned into a common global coordinate system using the Multi-Station Adjustment (MSA) plugin. The overall georeferencing error as reported by the MSA plugin was 85 mm. Finally, point clouds were merged into one polyobject, exported to the LAS data format (American Society for Photogrammetry and Remote Sensing LASer file format), and interpolated to a 1 × 1 m rectangular grid with CloudCompare software, resulting in a DEM of Ecology Glacier and its surroundings.
2.5. Proglacial Lagoon Echosounding and Sediment Sampling
Lagoon bathymetry data were collected during March 2016 using a float-mounted acoustic Doppler current profiler (ADCP, RiverSurveyor S5, manufactured by SonTek, San Diego, CA, USA). This device was equipped with a vertical single-beam echo sounder (working at a frequency of 1 MHz and accuracy 1% of measuring values), and four-beams (working at a frequency of 3 MHz, with slant angle and accuracy 0.25% of measuring values), enable the 3D visualization of “surface” water velocity. The dGPS provides the ADCP horizontal position with an accuracy of 0.5 m.
Measurements were made along the survey lines on 2, 17, 21 and 29 March 2016 with a total surveyed length of approximately 20 km (see
Figure 3). It should be noted, however, that the Admiralty Bay experiences significant tides with maximum amplitudes of 1.81 m. Therefore, it was important to take account the short-term tidal level changes during the day of measurement.
Table 1 shows the water level during the start and the end of the field work, which were obtained from the numerical data available at
www.tide-forecast.com. Due to the small differences of up to 0.2 m, we assumed linear model changes, and the obtained values were treated as an amendment to the measured values of bathymetry. They were all related to a depth of 0.7 m as a reference point.
The measured depths refer to the present geometry of the lagoon (2016), but it could have been deeper in the past if the marine or glacial sedimentation has decreased its depth over time. While glacial sedimentation rates in this area are relatively well constrained (e.g., [
53]), the marine sedimentation rate is unknown and can provide additional source of uncertainty in long-term lagoon bathymetry estimation. Therefore, in order to determine the presence and origin of sediments in the lagoon, sediment samples were collected from three locations: (1) close to the front of the glacier; (2) in the central part which is close to the islands; and (3) finally in the small bay (
Figure 3). For each of these cases, the samples were taken with use of Ekman bottom grab and at least 1 kg of bed material was collected for further analysis and determination of granulometric distribution curves.
2.6. DEM Differencing and Ice Elevation Changes
Ice elevation changes were computed by differencing subsequent DEMs for each time interval (e.g., [
54,
55,
56]). Each DEM was interpolated into a common reference system and grid, co-registered with use of several GCPs derived from the 2016 TLS DEM. Then, ice elevation change was computed as the difference in height over the glaciated area. Given that the lagoon bathymetry was only known for 2016, it was assumed that it reflects the same level as the glacial bed when it was still covered by ice. Such assumption is supported by relatively low (6.6 mm
) sediment accumulation rates reported for another outlet glacier of the Warszawa Icefield, the Fourcade Glacier [
53].
4. Discussion
An important shift in the ice elevation change of Ecology Glacier was detected in years 2012–2016, when both frontal retreat and surface lowering due to ablation have decreased. This can be partially explained by observed positive surface mass balance, at least over the season 2012/2013 [
28]. Whereas the surface elevation change is close to zero, the ice front retreat reported by [
26] is still ongoing, although it has decelerated. The boundary between ice elevation loss and gain over the years 2012–2016 is located at the elevation of 200 ± 25 m a.s.l., slightly above the recent ELA of 150 m a.s.l. reported by Sobota et al. [
28]. While these two values cannot be directly compared, as the former includes the ice dynamics component (emergence/subsidence flux) and the latter does not, this indicates that the glacier is currently out of balance in its ablation zone.
On neighbouring Livingston Island ice cap, a significant deceleration of the mass loss caused by positive surface mass balance, has also been reported and associated with air temperature change [
20]. Comparatively, in the case of the Antarctic Peninsula outlet glaciers, the major fraction of ice elevation changes have been attributed to frontal retreat, with little contribution of ice surface lowering [
12]. On the contrary, only ∼40% of the ice elevation loss in the terminal part of Ecology Glacier can be attributed to the long-term frontal retreat (
Table 3). This can be explained by the shallow depths of the proglacial lagoon that do not contribute significantly to the overall ice cliff height. On long term, this limits the contribution of frontal retreat to ice elevation change to approximately 40 m. Hence, long-term ice elevation change has to be driven mainly by surface melt and ice dynamics. Surface melt is directly dependent on climate and should decrease with recently observed cooling (
Figure 1, while ice dynamics should indirectly respond to decreasing ice thickness, increased slope and change in sliding speeds caused by varying supply of meltwater to the glacier base.
Locations where the Ecology Glacier ice front position was relatively stable and the retreat rate was low coincide with a presence of the bedrock outcrops (
Figure 10), indicating that the glacier retreat may be governed by the presence of pinning points [
58]. Despite shallow depths of the proglacial lagoon, it seems that the water depth controls termini retreat rate through enhanced calving. This may be caused by undercutting of the ice cliff [
59,
60,
61] by relatively warm water, reaching
C (
Figure 7). However, given that the elevation change caused by front retreat is limited by low ice cliff height due to small water depths, this should not affect the interpretation of long-term climate effects.
Proglacial lagoon bed sediment granulometry (
Figure 8) suggests that the bottom sediment in the outer part of proglacial lagoon is maritime whereas the one located near the glacier front is a product of local glacial sedimentation. This may be the result of backfill of the Suszczewski Cove and higher influence of the marine processes on the outer part of the proglacial lagoon, principally the Bransfield Strait swell entering the lagoon and depositing coarse sediment. Consequently, this can lead to an erroneous estimation of the bedrock depth when the lagoon was still covered by ice and thus constitute a possible source of underestimation of the calculated ice elevation change in this area.
Although the 1979 DEM reconstruction from aerial photos and the 2001 map [
45] have relatively large errors (
Table 2), the observed glacier elevation changes are typically an order of magnitude higher (
Figure A1,
Table 3). Thus, as recently shown by [
47], performing Structure-from-Motion analysis of archival aerial imagery can be a valuable source of information concerning long-term glacial change.
The Warszawa Icefield shows generally low dynamics comparing to other parts of the King George Island and is not contributing significantly to the overall mass balance of the entire ice cap of King George Island [
27]. Therefore it is more sensitive to the surface mass balance changes than to ice flow variations, showing a more direct response to climate signals than the more dynamic neighbouring icefields. Our data shows slight thickening in the accumulation area in recent years (2012–2016). Unfortunately, it was not possible to determine previous elevation change in the upper reaches of Ecology Glacier due to an insufficient spatial coverage of produced DEMs.
It must be stressed that the mass balance of glaciers on King George Island is highly dependent on the large-scale circulation, mainly by advection of warm, humid air from the north [
62]. Therefore, it is important to consider long term changes that are independent of the short-term fluctuations. Whereas recent years showed significant air temperature cooling and thus positive net surface mass balance, future changes of climate in the region are not clear given its high natural variability [
18].
5. Conclusions
Ecology Glacier experienced highly negative mass balance over 1979–2016 driven both by high surface melt and frontal retreat. The ice elevation change rate for the common ablation area over all analyzed periods (1979–2001–2012–2016) has changed from −1.7 ± 0.4 m/year in 1979–2001 and −1.5 ± 0.5 m/year in 2001–2012 to −0.5 ± 0.6 m/year in 2012–2016.
Ice surface lowering of Ecology Glacier has significantly decelerated in the beginning of the 21st century, especially in 2012-2016 when the glacier was close to equilibrium. Ice flow velocities of the terminal part of Ecology Glacier have been very low at least since 1995 [
27,
63]. Therefore, we do not expect large variation in ice flux due to termini retreat and the latter should be driven mainly by an enhanced calving when the front retreats to deeper water in the proglacial lagoon. Thus, we associate the recent deceleration of ice thinning in the terminal part with the observed atmospheric cooling, rather than with a dynamic response to a rapid retreat to a new pinning point. However, such a possibility cannot be ruled out based on the collected data and needs further investigation.
For the long term, the frontal retreat rate is mainly controlled by the bedrock topography: water depth of the proglacial lagoon and the presence of pinning points. Additionally, the shallow depths of the proglacial lagoon (mean depth 3.6 ± 1.9 m, maximum 11.8 m) limit the ice thickness of termini and, hence, the long-term contribution of frontal retreat due to calving to observed ice loss.
The only elevation data that was able to provide glacier wide coverage of Ecology Glacier was the high resolution satellite image and DEMs from Pléiades satellite. Structure-from-Motion analysis of archive aerial imagery did not provide reliable results in the snow-covered areas due to the low contrast and oversaturation of the photographs, while Terrestrial Laser Scanning DEM coverage suffered from shadows cast by highly undulated glacier surface.
As the bedrock topography of Ecology Glacier remains unknown, our results show only the absolute values of elevation change and cannot be expressed as percentage of the actual volume of the glacier as has been reported for other sites in South Shetland Islands (e.g., [
30]). Measurements of bedrock topography would help to shed light on the future stability of the calving front and whether the observed deceleration of termini retreat will prevail or if it is just temporary and linked to the presence of an isolated pinning point [
26,
58] as experienced in the years 1988–2003 (
Figure 10). Therefore, there is a need to extend ground penetrating radar surveys of the central part of the King George Island [
36] to its peripheral icefields and outlet glaciers. This would also provide glacial geometry required for ice dynamics modelling needed for determination whether ice elevation change is driven mainly by the surface mass balance or the ice dynamics response to enhanced calving.