skip to primary navigation skip to content

Department of Geography

 

 

INTEGRAL

FP6 Logo

Austfonna, Nordaustlandet

Nordaustlandet in the northeast Svalbard archipelago has two major ice domes: Austfonna to the east and Vestfonna to the west (see below). The Austfonna Ice Dome (8,120 km) is the most prominent ice cap in the Svalbard archipelago with a mean ice thickness of around 300 m (http://dib.joanneum.at/integral). In the 1980s, the total length of its seaward margins was 251 km. The most relevant test glacier of the INTEGRAL project is Duvebreen in the north of the ice cap.

Image as described adjacent

Map of the Austfonna ice cap (Norway) with the position of stakes (?), surface GPS and GPR profiles (-) and ice drainage basins (-) of the Norwegian Polar Insitute (NPI).

Image as described adjacent

Example of NOAA browse image used for estimation of weather conditions. Left: near infrared (channel 2); right: thermal channel (channel 4)

NOAA AVHRR data quasi-synchronous with SAR Data acquisitions were used to estimate atmospheric conditions (cloudiness) in the SAR mapping area. This was done in order to avoid noise in phase measurements caused by tropospheric heterogeneities. NOAA images, available for all 8 JERS SAR acquisitions used in the INTEGRAL project, confirm a quiet and homogenous tropospheric state in the area of the test sites.

A number of GPS-measured stake velocities are available from NPI field work over Austfonna. The table below shows a number of these velocities, and the image below right shows the stake network. See above for map of NPI surface GPS/GPR profiles on Austfonna.

Image as described adjacent

Image as described adjacent

Image as described adjacentImage as described adjacent

Radio-Echo Sounding (RES) flights were conducted over Nordaustlandet in 1983 and 1986 by SPRI, to measure the ice thickness along transects over the ice caps. RES data is available as point measurements, and as a digital grid of ice thicknesses interpolated over the ice cap at 1km spacings from the original point data. Cyan points mark where both surface elevation and ice thickness was recorded green points record only a surface elevation.

Surface and bed elevations are shown along a (broadly) West-East transect of the ice cap, along with SAR-derived 3D velocity along the same profile.

Image as described adjacent

Map shows approximate location of satellite image sections (red boxed area)

Image as described adjacentImage as described adjacent

The images above show the ice margin position for the SW part of Austfonna in 1962 (Declassified Satellite Imagery, DSI) and 2001 (Landsat ETM+). In combination with ice velocity data derived from SAR processing (DINSAR / Offset Tracking) and ice thickness data, use of optical / Near-Infrared satellite imagery to track margin changes allows for assessment of ice volume flux at the margins. When combined with data regarding the surface mass balance of the ice cap, it may then be possible to estimate the proportion of mass balance changes due to ice melting and to iceberg calving.

The image right is a composite of mosaiced aerial photography from 1970 (background) and from 1990. The retreating front of Etonbreen is clearly demonstrated.

Click the image below to see an image map of Etonbreen frontal changes (opens in new window)

Image as described adjacent

Image as described adjacent

SAR data from the ERS-1 and 2 satellites has been used, along with JERS-1 SAR data within the INTEGRAL project for study of the Nordaustlandet ice caps.

Image as described adjacent

Frame coverage of ERS-1 and ERS-2 scenes over Austfonna and Vestfonna.

Image as described adjacent

The JERS SAR mission (Nemoto et al. 1991) was developed jointly by the Japanese Ministry of International Trade and Office and the Japan Aerospace Exploration Agency (JAXA). JERS was operated between 1992 and 1998 collecting a huge amount of SAR data at L-Band (l = 24.3 cm) during the six and a half years mission period. As already demonstrated for land subsidence monitoring (Strozzi et al. 2003), landslides survey (Strozzi et al., 2005) and active rock glaciers (Strozzi et al. 2004), L-band interferometry has the capability of complementing the existing applications based on C-band (l = 5.7 cm), because its larger wavelength is more appropriate for mapping rapid displacements. Furthermore, the greater penetration of the radar signals into the snow and firn at L-band than at C-band should result in reduced decorrelation. JERS SAR data are archived both at JAXA and at the European Space Research Institute (ESRIN), where there are more than 100,000 JERS SAR frames over Europe acquired at the Troms and Fucino ground stations (De Groote and Greco 2003).

Image as described adjacentImage as described adjacent

Image as described adjacentImage as described adjacent

L-Band JERS-1 SAR data (1.3 GHz, 23.5 cm wavelength, HH polarization) was obtained from the ESRIN archive as long stripes Level-0 raw data on DLT. Some adaptation of GAMMA SAR processing software was required to handle this data format. The two JERS SAR interferometric pairs of Nordaustlandet mainly cover Vestfonna in winter 1994, and Austfonna in winter 1998. It can be seen that the backscattering energy at L-Band is higher over the central parts of the ice caps.

A number of Altimetery (ALT) datasets were obtained and investigated within INTEGRAL. These are described below, concentrating on their applicability to INTEGRAL aims of combination with InSAR processing.

Click here for LEGOS document Studies of Austfonna Ice Cap (Venice 06) [pdf, opens in new window], which gives further details regarding investigation of ALT data in combination with other data sources. This document was prepared for the conference 15 Years of progress in Radar Altimetry Symposium, Ocean surface topography science team (OSTST), International Doris Service (IDS) Workshop, Argo Workshop, 13-18 March 2006, Venice, Italy.

To be assured of the number and accuracy of elevation points used in InSAR baseline constraint or DEM construction for topographic phase simulation, one ideally requires a good number of crossover points (where data is available from ascending and descending orbits) spread across the ice cap.

The ERS-2 RA mode mask shows that Svalbard is set to Ocean mode (not Ice mode). Analysis of the available ERS-2 RA ALT data shows that there is limited data suitable for combination with InSAR processing.

Single cycle crossovers (AscCn - DescCn)

The figure to the right shows the Density of Crossovers (single cycles 48 - 50), where data points have been assigned to the nearest 10km2 bins

Image as described adjacent

Image as described adjacent

Image as described adjacent

Density maps show almost no crossover points over Austfonna. The figure to the left is an orbit plot (cycle 48, representative of others) showing the tracking mode and crossover locations. This partly demonstrates the reason for the lack of crossover points. In fact, there are less crossover points than might be supposed, since there are some locations which show that the instrument may be tracking in both orbits but no crossovers are displayed. Further investigation shows that in most of these locations, the OCOG retracker failed to retrack and the crossover was rejected.

Analysis of Envisat RA2 ALT data over Austfonna indicates a greater availability of crossover data than for ERS-2 RA.

Single cycle crossovers (AscCn - DescCn)

The figure to the right shows the Density of Crossovers (single cycle 40), where data points have been assigned to the nearest 10km2 bins. Crossovers are thresholded at a level of +/- 3m difference between Ascending and Descending elevation measurement. This shows more interior ice cap crossover points than the ERS-2 RA ALT data.

Image as described adjacent

Image as described adjacent

Image as described adjacentImage as described adjacentImage as described adjacent

Image as described adjacent

The figure to the left shows the Residuals of Crossovers (for a single cycle, 40), where data points have been assigned to the nearest 10km2 bins. Left all points, Middle land-only, Right ocean-only.

Multi cycle crossovers [(AscCRef - DescCn) - (AscCn - DescCRef)]/2, single reference cycle

The figure to the right shows the Density of Multi-Cycle Crossovers for a single (40), where reference cycle (Ref) has been chosen as cycle 25 (~March 2004; chosen from density plots), and data points have been assigned to the nearest 10km2 bins.

Image as described adjacent

Image as described adjacent

Corrections

A number of corrections need to be applied for multi-cycle crossover analysis versus a reference cycle. The applied corrections are summarized below.

Image as described adjacent

Correction of delta elevation due to Dry Tropospheric differences.

Image as described adjacent

Image as described adjacent

Correction of delta elevation due to Wet Tropospheric differences.

Image as described adjacent

Image as described adjacent

Correction of delta elevation due to Modelled Ionospheric differences.

Image as described adjacent

Image as described adjacent

Correction of delta elevation due to Solid Earth Tide differences.

Image as described adjacent

Other contributions to range differences from reference cycle elevations include Retracker (OCOG) [+/- 5m ], Altitude [+/- 200m display] and Range [+/- 100m display].

Analysis of elevation change over time

Temporal analysis of crossover elevation differences versus reference may reveal rates of growth or lowering of ice cap surface. This may be relevant to InSAR processing if the SAR data used for InSAR pairs is not co-incident with the date of acquisition of data used in a DEM (for topographic phase simulation) or as topographic baseline constraint. Thinning or thickening rates may differ over the ice cap, and the rate of elevation change determined from ALT crossover analysis may be applied to initial elevation data to correct them to values expected at the time of SAR data acquisition.

Image as described adjacent

Elevation change trend (m / yr)

Image as described adjacent

Elevation change trend (m / yr) smoothed at 20 km radius to produce a continuous correction field over the ice cap. Note the relative (circa zero) homogeneity of the data over the ocean surface area around the ice cap.

The above analysis is repeated for a number of other Reference cycles, chosen from density plots as having reasonable coverage. In order to combine timeseries from multiple Reference cycles, the 10 km grid of dH timeseries for each reference cycle are merged into a single 10km grid using a timeseries averaging process that fills in missing values using a modelled sinusoidal trend where necessary

The CryoSat mission was intended to make use of a new kind of Radar Altimeter called SIRAL (Synthetic Interferometric Radar ALtimeter). SIRAL uses synthetic apature techniques to increase along-track resolution, and has 2 receiving antennae providing an across-track interferometric capability. To cater for these new operational modes UCL designed new algorithm chains and implemented them within the CryoSat ground segment. The new processing algorithms are described in :-

Wingham, D.J., Phalippou, L., Mavrocordatos, C. and Wallis, D., 2004. "The mean echo and echo cross product from a beamforming interferometric altimeter and their application to elevation measurement". IEEE Transactions on Geoscience and Remote Sensing, 42(10), 2305-2323.

This paper can be obtained on-line by visiting URL http://www.cpom.org/ (requires Internet opens in new window) and following the links Research -->Publications. The paper also describes the new retracking algorithms for SARin (see 325s)

Image as described adjacent

Due to the failure in launch of the Cryosat satellite, it has not been possible to obtain SIRAL Altimetric data during the INTEGRAL project. Instead, we have obtained ASIRAS airborne data over the Austfonna ice cap. ASIRAS functions in the same way as the spaceborne SIRAL radar, with suitably-scaled parameters for airborne application. The Cryosat replacement mission, Cryosat2, has the potential to supply swath-mode altimetric elevations (SARIN mode), giving a number of points at ~ 250 metres spacing along- and across-track, in addition to the point of closest approach within an ~10 km wide area at each ~250 m along-track step (SAR mode). Formation of swath-mode Altimetric data is dependant on phase coherence of the returned energy. This figure demonstrates our experience with ASIRAS data, which indicates that swath processing may be feasible, although swath formation is not uniformly achieved or predictable.

It is evident, that from the experience of ASIRAS data processing a Cryosat swath-mode elevation product is potentially viable. The ability to process swath returns across-track is spatially variable and does not appear to clearly depend on a single parameter such as platform attitude angle or surface elevation. A maximum swath width of 241 m is observed, with an average of 19 m and standard deviation of 26 m. The theoretical maximum across-track swath distance in this example data is ~ 159 m, for the average flying height above surface of 1139 m, assuming an onboard tracker which places the first surface return exactly in the middle of the range window. The greater maximum width actually recorded is due to a combination of greater terrain clearance (upto ~1540 m) in places, and actual tracker performance. To see what this might imply for Cryosat swath formation, we assume that a similar pattern of swath returns is recorded, and that footprints / bin spacings may be directly scaled up from the airborne ASIRAS parameters, and that perfect tracker mid-range performance is achieved. This gives estimates for Cryosat swath formation potential of maximum ~13 km, average ~1.5 km, and standard deviation ~2 km.

Platform hgt (m)

Max bins

Swath bins

dR/bin cm

Max Range (m)

Max across-track (m)

Avg. across-track (m)

Std dev. across-track (m)

ASIRAS

1139

256

128

8.6

1150

159

19

26

Cryosat

717242

512

256

46

717360

12998

1556

2129

Estimated CryoSat Swath formation performance with relation to ASIRAS

An amount of GLAS laser altimetry data is also available over Austfonna, as shown by red points over figure to right. This may have use for interferometric baseline constraint, along with field-based GPS data along NPI transects.

Image as described adjacent

The listings below show the perpendicular baselines of combinations of ERS SAR scenes available to the INTEGRAL project. The values highlighted indicate interferograms actually processed (GAMMA).

Image as described adjacent

Image as described adjacent

Image as described adjacent

Image as described adjacent

Image as described adjacentAbove: Flattened, filtered, geocoded differential interferogram of January 6 and 9, 1994. Look direction (ascending mode) is indicated by diagonal arrow (incidence angle ~23). The perpendicular baseline is -70 m.

Line-of-sight surface displacement map of the Austfonna and Vestfonna ice caps in map geometry for January 6 to 9, 1994.

Image as described adjacent

Image as described adjacent

Image as described adjacentAbove: Flattened, filtered, geocoded differential interferogram of January 7 and 10, 1994. Look direction (ascending mode) is indicated by diagonal arrow (incidence angle ~23). The perpendicular baseline is 34 m.

Line-of-sight surface displacement map of the Austfonna and Vestfonna ice caps in map geometry for January 7 to 10, 1994.

Image as described adjacent

Flattened, filtered, geocoded JERS interferogram of December 11, 1997 and January 24, 1998. Look direction (descending mode) is indicated by the arrow (incidence angle ~35), the perpendicular baseline is 230 m.

Our SAR interferometric results indicate a very good coherence, with well preserved fringes over slow moving glaciers. With a longer temporal baseline (44 days) than for ERS data, decorrelation effects might be expected due to surface changes (such as melt, snow accumulation or wind drift), or due to volume effects from differing microwave penetration of dry snow cover and ice. In practice, these effects are limited in this dataset. Decorrelation is mainly observed over areas with high displacement gradients, in particular along outlet glacier margins with excessive strain rates.

Image as described adjacent

3-dimensional displacement map for the part of the Austfonna and Vestfonna ice caps derived from dual-azimuth interferometry, using ERS-1 3-day repeat SAR data of January 6 and 9, 1994 (ascending orbit) and January 7 and 10, 1994 (descending orbit).

Three GPS stakes around the main flow line of Duvebreen in Austfonna were surveyed in 2004 and 2005 (see right), and their displacements are compared with those derived from InSAR processing. Results for these three points are compared with DINSAR records of displacement in the Table below. When the displacement rate is larger than 10 m/year (Nord 1) the correspondence between the two surveying techniques is very good, both in magnitude and direction, for both 1 day and 3 days ERS SAR data. For slow displacement rates of less than 10 m/year there is a certain difference between DINSAR and GPS, with a bias of about 5 m/year in DINSAR velocities. Unfortunately, GPS stakes were not precisely located over the main flow line of the glacier in areas with larger displacement velocities.

Image as described adjacent

GPS 2004-2005

DINSAR January 1994

DINSAR January 1996

Stake

Rate (m/y)

Orientation

Rate (m/y)

Orientation

Rate (m/y)

Orientation

Base

1.1

22

5.2

70

5.3

-35

Nord 1

9.6

-6

11.2

49

7.8

37

Nord 2

11.5

18

13.4

27

11.4

8

The successful use of DINSAR is limited by the phase noise, usually characterized by the coherence. Over glacier surfaces the coherence is affected by meteorological and flow conditions and generally diminishes with increasing time interval between the acquisitions of the two SAR images used in the interferogram. Meteorological sources of decorrelation include ice and snow surface melt (Strozzi et al., 1999) and possibly snowfall and wind through the redistribution of snow and ice. Decorrelation causes related to the motion of the glacier are incoherent displacements of adjacent scatterers and rapid flow if local deviations from the overall image registration function are not taken into account.

When differential SAR interferometry is limited by loss of coherence, i.e. in the case of rapid and incoherent flow and of large acquisition time intervals between the two SAR images, offset-tracking procedures of SAR images are a welcome alternative to differential SAR interferometry for the estimation of the glacier motion (Rott et. al, 1998, Gray et al., 1998, Michel and Rignot, 1999, Derauw, 1999, Gray et al., 2000, Gray et al., 2001). With offset-tracking procedures the registration offsets of two SAR images in both slant-range (i.e. in the line-of-sight of the satellite) and azimuth (i.e. along the orbit of the satellite) directions are generated and used to estimate the displacement of glaciers. The estimated offsets are unambiguous values which means that there is no need for phase unwrapping, one of the most critical steps in SAR interferometry.

Within the INTEGRAL project we generally distinguished between two image-to-image patch offset techniques for estimating motion between satellite SAR images: intensity-tracking and coherence-tracking. As intensity-tracking we understand the cross correlation of image patches of detected real-valued SAR intensity images (Rott et. al, 1998, Gray et al., 1998, Michel and Rignot, 1999, Gray et al., 2000, Gray et al., 2001). As coherence-tracking we understand the coherence optimization procedure, also known as fringe visibility algorithm (Derauw, 1999), applied to complex SAR images.

For the Nordaustlandet ERS SAR data set intensity tracking was applied. With intensity-tracking the offset fields are generated with a normalized cross-correlation of image patches of detected real-valued SAR intensity images. The successful estimation of the local image offsets depends on the presence of nearly identical features in the two SAR images at the scale of the employed patches. When coherence is retained, the speckle pattern of the two images is correlated and intensity tracking with small image patches can be performed to remarkable accuracy. Incoherent intensity tracking is also feasible, but requires large image patches. Because intensity-tracking is the only method that can be applied for incoherent features, often it is also referred to as feature-tracking. In order to increase the estimation accuracy, oversampling rates are applied to the image patches and a two-dimensional regression fit to model the correlation function around the peak is determined with interpolation. The location of the peak of the two-dimensional cross-correlation function yields the image offset. Confidence in the offset estimate is measured by comparing the height of the correlation peak relative to the average level of the correlation function to supply a correlation Signal-to-Noise Ratio (SNR). Coarse information on the slant-range and azimuth offsets is used to guide the search of the cross-correlation maximum.

The image offsets in the slant-range and azimuth directions estimated with intensity tracking are related to the different satellite orbit configurations of the two SAR images and to the displacement occurring between the acquisition time interval of the image pair. The estimation of glacier motion, or surface deformation in general, requires therefore the separation of these two effects. The orbital offsets in the slant-range direction are related to the baseline, i.e. the separation in space of the antennas between the two SAR acquisitions. Orbital offsets in the azimuth direction are affected by the change of the baseline along the orbit. The estimation of the baseline from the orbit parameters supplied by the space agencies is normally not precise enough either for the determination of the orbital offsets or in general for SAR image co-registration. It is therefore preferable to determine the orbital offsets from stable reference points. This is achieved, for instance, by fitting a bilinear polynomial function to offset fields determined globally from the SAR images if the assumption of no displacement for most parts of the image is satisfied. After subtraction of the orbital offsets, pixel shifts in slant-range and azimuth directions are related to glacier motion only and are transformed to displacements in meters with knowledge of the SAR image pixel spacing.

Image as described adjacent

Image as described adjacent

3-dimensional velocity map for the Austfonna-Vestfonna ice caps derived from offset-tracking in March 1994

Slant-range and azimuth displacements provide only two components of the 3-dimensional displacement vector. In order to obtain a 3-dimensional displacement map, the slant-range and azimuth displacements can be combined with a DEM by assuming flow parallel to the surface of the glacier. If a DEM is not available, but it can reasonably be assumed that the surface is horizontal, slant-range and azimuth displacements may provide a 2-dimensional ground displacement field without using multiple image pairs from dual-azimuth angles (e.g. ascending and descending passes).

As an example (left), the velocity map for March 1994 combines offset fields in the azimuth direction from ERS SAR data of ascending (6 to 12 January) and descending (4 to 10 January) orbits. The velocities of the fast flowing units are well represented. On the other hand, for slow velocities artefacts related to the precision of the technique, and in particular to ionospheric streaking (Grey et al., 2000), are evident.

Image as described adjacentHere, we see a velocity map for Austfonna derived from JERS offset-tracking between SAR images of December 11, 1997 and January 24, 1998. The results highlight the fastest moving glaciers around the ice caps, but again indicate azimuth shift modulations, possibly related to auroral zone ionospheric disturbances (Grey et al., 2000). L-band SAR data is much more sensitive to ionospheric conditions along the SAR swath path than C-band data.

DINSAR vs JERS Offset tracking

JERS offset-tracking records of surface ice velocity are compared below with ERS DINSAR results along the approximate main flow line of Duvebreen. The Duvebreen glacier was considered for this evaluation because it is partially visible with both JERS image pairs of March/May 1994 and December 1997 / January 1998.

Image as described adjacent

Ice surface velocities Duvebreen from ERS DINSAR in February 1992, January 1994 and January 1996 and from JERS offset-tracking in April 1994 and January 1998.

The ERS DINSAR records indicate an increase of ice velocity from 1992 to 1996, but seasonal variations may also account for this. The JERS offset-tracking results in 1998 are similar to the DINSAR outcomes of 1996. In particular, between 10 and 25 km there is a very good correspondence, with differences within 10 m/year. Strong azimuth shift modulations are observed in the JERS results between 0 and 10 km, where differences with ERS DINSAR results are larger than 10 m/year. The JERS SAR images of 1994 cover only the front of Duvebreen, where measured surface ice velocities are smaller than in 1998 and comparable with the DINSAR rates of 1994. However, at the front of the glaciers large variations of ice velocities are observed.

DINSAR vs ERS Offset tracking

ERS offset-tracking records of surface ice velocity using dual-azimuth offset-tracking with 6 days acquisition time interval are compared below with ERS DINSAR results along the approximate main flow line of Duvebreen.

Image as described adjacent

Ice surface velocities along Duvebreen from ERS DINSAR and 6-day ERS offset-tracking in January.

The ERS offset-tracking results are within 10 m/year of the DINSAR outcomes between 10 and 27 km, where velocities are larger than 20 m/year.

Between 0 and 10 km there is a bias of the offset-tracking results of about 2010 m/year.

ERS Offset tracking vs JERS Offset tracking

ERS offset-tracking records of surface ice velocity using dual-azimuth offset-tracking with 1 and 6 days acquisition time interval are compared below with JERS range-azimuth offset-tracking results along the approximate main flow line of Eltonbreen, a very fast flow unit in Vestfonna.

Image as described adjacent

Ice surface velocities along Eltonbreen from JERS offset-tracking in April 1994 and from ERS dual-azimuth offset-tracking in January 1994 (6-days interval) and January 1996 (1-day interval).

The ERS offset-tracking results with 6-days acquisition time interval and the JERS offset-tracking results are within 10 m/year.

The ERS offset-tracking results with 1-day acquisition time interval show larger variations, at least 50 m/year when velocities are larger than 200-300 m/year and at least 100 m/year for minor rates.

Image as described adjacentImage as described adjacent

3-dimensional velocity map for the Austfonna-Vestfonna ice caps derived from the combination of DINSAR and offset-tracking in March 1994.

The DINSAR technique allows surface velocities for relatively slow moving glaciers to be measured with high precision, but may fail on areas moving at higher velocities where coherence is greatly diminished or lost. Offset tracking can be very successful for measuring the velocities of these higher-velocity glaciers, but are not sufficiently precise for slower-moving ice (see above). It is therefore clear that a combination of these two methods may yield a more complete overview of glacier displacement over the full range of ice displacement velocities. The velocity map for the Austfonna-Vestfonna ice caps shown left was obtained using DINSAR results wherever successfully obtained, with offset tracking results considered in order to fill gaps. Tracking results were only used where they exceeded a velocity of 50 m/year.