Introduction

Drastic changes have been detected in glacial systems of the Antarctic Peninsula in the last few decades. The observed phenomena comprise the disintegration of ice shelves (Reference Rott, Rack, Nagler and SkvarcaRott and others, 1998; Reference Skvarca, Rack, Rott and DonángeloSkvarca and others, 1999; Reference Rack and RottRack and Rott, 2004), acceleration and thinning of glaciers (Reference De Angelis and SkvarcaDe Angelis and Skvarca, 2003; Reference Rignot, Casassa, Gogineni, Krabill, Rivera and ThomasRignot and others, 2004; Reference Scambos, Bohlander, Shuman and SkvarcaScambos and others, 2004) and retreat of glacier fronts (Reference RauRau and others, 2004; Reference Cook, Fox, Vaughan and FerrignoCook and others, 2005). Although the time response of such processes to climate change remains uncertain, the dependency of the detected glacial changes on variations in climate parameters seems evident. However, due to the lack of consistent systematic observations of the higher parts of the glacial systems in particular, it is difficult to predict further responses of glaciers to climate change. To expand the current monitoring of a few glaciers, unevenly distributed along the peninsula, to a representative set of glaciers in the near future, Reference Arigony-Neto, Saurer, Jaña, Rau, J.C.Simões and GoßmannArigony-Neto and others (2006) proposed a method to simplify the detection of glacial zone boundaries (i.e. snowline, wet-snow line and dry-snow line) using remote-sensing data.

The evolution of glacier zones is greatly influenced by local and regional climatic and meteorological settings. Their variations in response to changes in energy or mass balance are considered as good indicators of climatic changes (Reference RauRau and others, 2000). In addition, several authors have demonstrated the potential of radar remote sensing to identify distinct zonal boundaries within the glacier (e.g. Reference Fahnestock, Bindschadler, Kwok and JezekFahnestock and others, 1993; Reference Forster, Isacks and DasForster and others, 1996; Reference Braun, Rau, Saurer and GoßmannBraun and others, 2000; Reference RauRau and others, 2000; Reference König, Winther and IsakssonKönig and others, 2001). A time series of satellite synthetic aperture radar (SAR) images (European Remote-sensing Satellite-1 and -2 (ERS-1/-2) SAR, Envisat advanced SAR (ASAR) and RADARSAT-1 SAR) enables the compilation of a database of glacier parameters for the last 14 years.

In this paper, we describe the results of knowledge-based image analysis algorithms used for classifying the climate-related boundaries between radar glacier zones (i.e. snowline and dry-snow line) on test areas located on Trinity Peninsula (TPE) and Marguerite Bay (MGB) in the Antarctic Peninsula (Fig. 1).

Fig. 1.

Location map of the study areas on the Antarctic Peninsula.

Radar Glacier Zones

The concept of superficial zones or facies on glaciers was first presented by Reference BensonBenson (1962) and Reference MüllerMüller (1962), and reviewed by Reference PatersonPaterson (1994). In this work, we use the reviewed concept from Reference PatersonPaterson (1994), illustrated in Figure 2. The analogous radar glacier zones (Fig. 2) correspond to the classification scheme proposed by Reference RauRau and others (2000). The boundaries between the radar glacier zones are addressed as different snowlines (i.e. snowline, dry-snow line). This paper follows the terminology given in Figures 2 and 3.

Fig. 2.

Glacier zones as described by Reference PatersonPaterson (1994) and corresponding radar glacier zones. Modified after Reference Rau, Braun, Friedrich, Weber and GoßmannRau and others (2001).

Fig. 3.

Subset of the ERS-2 SAR image acquired on 23 February 2000 (orbit 25328, frame 4941) showing the position of boundaries between radar glacier zones along the glacier centre line.

The backscatter signal recorded on glaciers and ice sheets depends on snowpack parameters (e.g. liquid-water content, snow density, stratigraphy, grain size and surface roughness) (Reference Rau and BraunRau and Braun, 2002). These parameters drive backscatter mechanisms such as volume scattering for the dry-snow radar zone and frozen percolation radar zone, low surface scattering with the radar signal being mainly absorbed by liquid water (wet snow radar zone) and surface scattering of a rough and dense surface (bare-ice radar zone) (Reference Braun, Rau, Saurer and GoßmannBraun and others, 2000). The different backscatter values for each successive radar glacier zone are given in Figure 2.

Data and Methods

Satellite data used for analysis consist of 32 images acquired in a descending orbit from 1992 to 2005 (Table 1) by the ERS-1/-2 SAR (30 images) and Envisat ASAR (2 images). Both are C-band VV polarization instruments. Figure 1 shows the area covered by the images on TPE (17 images) and MGB (15 images). Normalized backscattering coefficients (σ ⁰) were calculated using the Basic Envisat and ERS SAR Toolbox (BEST) from the European Space Agency (ESA). BEST uses the SAR calibration algorithm developed by H. Laur and others (http://earth.esa.int/pub/ESA_DOC/ers_sar_calibration_issue2_5f.pdf), who estimated that resulting values of σ ⁰ have an accuracy of ±0.4 dB. The excellent stability of the SAR sensors enables direct comparison of the calibrated data (Reference Meadows, Laur and SchättlerMeadows and others, 1998). To reduce the speckle effect, a 5×5 median filter was applied. Finally, speckle-filtered images were orthorectified using the digital elevation model (DEM) from the RADARSAT Antarctic Mapping Project (RAMP), which was created from various datasets by H. Liu and others (http://bipolar.colorado.edu/data/nsidc-0082.html).

Table 1.

Satellite imagery used in this work. TPE: Trinity Peninsula; MGB: Marguerite Bay; DSL: dry-snow line; SL: snowline

The SAR dataset was analyzed using the methodology proposed by Reference Arigony-Neto, Saurer, Jaña, Rau, J.C.Simões and GoßmannArigony-Neto and others (2006). This involves simplification of the image analysis by considering only areas located along glacier centre lines (Fig. 3). The so-called ‘centre-line approach’ minimizes the need for high-resolution DEMs for terrain correction near the lateral ice divides on steep ridges. Furthermore, glaciers were selected for analysis when they were wide enough so as to avoid layover effects or shadows occurring in SAR data (Reference Arigony-Neto, Saurer, Jaña, Rau, J.C.Simões and GoßmannArigony-Neto and others, 2006). As the backscatter values of dry-snow and frozen percolation radar zones result from volume scattering, small changes in terrain slope and aspect along the centre lines will not affect the classification of these zones. The separation between wet-snow radar zone and bare-ice radar zone, whose thresholds in backscatter values are relatively close (Fig. 2), will profit from the restriction of the analysis to areas with smooth slopes near the centre line.

Image analyses along the glacier centre lines were carried out using a knowledge-based image analysis algorithm (Fig. 4). Rules for pixel classification were based on backscattering thresholds determined by Reference Rau, Braun, Friedrich, Weber and GoßmannRau and others (2001) (see Fig. 2) and the following thresholds in altitude:

Fig. 4.

Schematic representation of the knowledge-based algorithm developed for classifying radar glacier zones on the Antarctic Peninsula using SAR imagery.

Elevation information was derived from the RAMP DEM. Misclassified pixels usually correspond to steep slopes where the plateaus break down to the glaciers, or crevasse fields inside major classes such as the wet-snow radar zone. After the classification, small patches of <2000 pixels were eliminated, removing the misclassified areas. Boundaries between classified glacier zones were extracted automatically by the algorithm. Whenever the boundary was located within 50 m of the threshold in altitude, a record was added to the classification report and the position of the boundary was revised by the analyst.

Results and Discussion

In total, 124 glaciers were investigated: 71 glaciers on TPE and 53 in MGB. TPE covers the northern tip of the Antarctic Peninsula (Fig. 1). There, 36 of the 71 glaciers studied flow to the eastern side, and the other 35 flow westwards. Glaciers monitored in the MGB area include 12 glaciers flowing to the Larsen C ice shelf, and 41 glaciers flowing into MGB itself. The results of the image analysis are as follows.

Snowline

The development of a bare-ice radar zone was detected on 45 glaciers in total (40 glaciers on TPE and 5 in MGB). However, for glaciers on the west Antarctic Peninsula, the occurrence of the bare-ice radar zone is patchy and irregularly distributed in time. Therefore, only results from 26 glaciers in the east TPE are shown in Figure 5 and discussed here. This region coincides with the area of highest annual positive degree-days in the Antarctic Peninsula, modelled by Reference VaughanVaughan (2006). The maximum value of mean snowline altitude (i.e. mean value of snowline altitude calculated for each balance year using only data from glaciers where a bare-ice radar zone was detected) was 392m (standard error (SE) = 41 m) for the balance year 1999/2000. This agrees with the lowest value of annual net mass balance (–0.56mw.e.) recorded for this balance year by Reference Skvarca, De Angelis and ErmolinSkvarca and others (2004) for the time series 1999/2000–2002/03 of ‘Glaciar Bahía del Diablo’, Vega Island (63˚50′ S, 57˚25′ W). In addition, the mean snowline altitude on the east coast of the TPE shows a good correlation with variations in the mean summer (December–February) air-temperature records at Marambio station (Fig. 5). Mean summer air temperatures are responsible for glacier surface melting and run-off in this region (Reference Skvarca and De AngelisSkvarca and De Angelis, 2003). However, the mean snowline altitude (233 m; SE = 58 m) detected for the balance year 1992/93 does not agree with the mean summer air temperature recorded for this period. The lack of precipitation data or field information in addition to images acquired prior to the end of the annual ablation season makes it difficult to interpret this phenomenon, because a new snowfall event followed by snowmelting and metamorphism could mask the bare-ice zone detectable by SAR sensors.

Fig. 5.

Mean values of snowline altitude (m) and standard error bars as detected on glaciers on the eastern side of TPE. The numbers indicate the sample size of glaciers for each balance year. The rhombs correspond to mean summer air temperature (˚C) recorded at Marambio station. Temperature data are from the British Antarctic Survey database (http://www.antarctica.ac.uk/met/metlog).

The variability in the averaged position of boundaries between glacier zones is related to local variations in mass and energy balance (Reference SchneiderSchneider, 1999). The good correlation of the mean snowline altitude with the mean summer air temperature shows that the SAR-derived mean snowline altitude can be used to test the hypothesis that glaciers on the northeast TPE are reacting to the positive trends in annual positive degree-days described by Reference VaughanVaughan (2006) for this region.

Dry-snow line

Variations in location of the dry-snow radar zone were observed on 47 glaciers located on Detroit Plateau (30 glaciers) and the central plateau of MGB (17 glaciers). On Detroit Plateau, the dry-snow line is located between 1292 and 2050 m altitude, whereas it reaches from 785 to 1714 m on the central plateau of MGB. The differences of dry-snow line altitude between the two study areas agree with the difference in mean annual surface temperature as recorded for this region (i.e. ~4˚C; Reference Morris and VaughanMorris and Vaughan, 2003).

As the basis for the statistical analysis, the altitudinal differences of the dry-snow line in metres per year were calculated for each glacier. Then annual changes in dry-snow line altitude were averaged in the western and eastern sectors of both study areas for three time periods between 1992 and 2005 (Fig. 6). Different patterns of change in the dry-snow line altitude were detected for all sectors over the period 1992–2005 (Fig. 6a–d). While the northern sectors had a positive trend in average change of dry-snow line altitude, the recorded change shows a slight downward trend in the southernmost parts of the Antarctic Peninsula (Fig. 6c–d).

Fig. 6.

Average change in dry-snow line altitude over time (1992–2005) and by sectors for each study area: (a) TPE west; (b) TPE east; (c) MGB west; and (d) MGB east. Negative and positive values correspond to mean downward and upward shifts of the dry-snow line respectively, and are relative to the mean dry-snow line altitude in the first year of the period. The numbers indicate the sample size of glaciers for each period.

Considering glaciers flowing to the Larsen C ice shelf (Fig. 6d), the observed increase in dry-snow line altitude agrees with findings from Reference RauRau (2003) for the period 1992–2000. This author shows the relationship of detected upward shift in dry-snow line altitude with an increase in high-temperature events at San Martin station (Fig. 1) during this period. Here, and during the last period recorded for TPE, it seems that extreme snowmelt events reach high-altitude areas in response to the increase in air temperature (Reference Morris and VaughanMorris and Vaughan, 2003) and duration of warm periods (Reference Torinesi, Fily and GenthonTorinesi and others, 2003).

A mean decrease in dry-snow line altitude was detected in the northern sectors for periods from 1992 to 2000. By modelling the backscattering (C-band; VV) characteristics for dry-snow zones, Reference Rau, Braun, Friedrich, Weber and GoßmannRau and others (2001) found that an increase in snow accumulation is necessary to move the dry-snow line downwards. Therefore, as high accumulation (Reference Turner, Lachlan-Cope, Marshal, Morris, Mulvaney and WinterTurner and others, 2002) and precipitation (Reference Turner, Colwell and HarangozoTurner and others, 1997) are recorded for the western sector of the TPE, the dry-snow line in this area appears to have reacted to increased accumulation from 1992 to 2000. Despite the high standard error values of the mean dry-snow line altitude recorded for glaciers on southern Avery Plateau, MGB, between 2001 and 2005, it is clear that the dry-snow line did not shift upward during the last period analyzed.

Conclusions

The implementation of a simple knowledge-based image analysis algorithm enabled the semi-automated analysis of a multitemporal set of satellite SAR images acquired on two test areas on the Antarctic Peninsula. The centre-line approach minimizes the need for high-resolution DEMs for terrain correction, facilitating the pre-processing of SAR data prior to image classification. However, due to the geometric effects inherent to SAR images, glaciers flowing through narrow valleys cannot be analyzed using this approach. We expect to overcome this in the near future, by using data from higher-resolution SAR sensors (e.g. TerraSAR-X).

A preliminary analysis of the spatial and temporal distribution of satellite-derived glacier parameters on the Antarctic Peninsula shows that climate-related boundaries between glacier zones (i.e. the dry-snow line and the snowline) have different spatial patterns of change in this region. These patterns vary within the study period (1992–2005), confirming the assumption by Reference Rignot and ThomasRignot and Thomas (2002) that changes in climatological and glaciological conditions on a relatively short timescale are usual for this region. Furthermore, the good correlation of the mean snowline altitude with the mean summer air temperature shows that the mean snowline altitude as detected by SAR sensors can be used to test the hypothesis that glaciers on the northeast TPE are responding to the positive trends in annual positive degree-days described by Reference VaughanVaughan (2006) for this region.