RADARSAT 1 synthetic aperture radar observations of Antarctica: Modified Antarctic Mapping Mission, 2000



[1] The RADARSAT 1 Antarctic Mapping Project (RAMP) is a collaboration between NASA and the Canadian Space Agency to map Antarctica using synthetic aperture radar (SAR). RAMP comprises two distinct mapping missions. The first Antarctic Mapping Mission was successfully completed in October 1997. Data from the acquisition phase of the 1997 campaign have been used to achieve the primary goal of producing the first high-resolution SAR image map of the entire Antarctic continent. The Modified Antarctic Mapping Mission (MAMM) occurred during the fall of 2000. The acquisition strategy concentrated on collecting highest-resolution RADARSAT 1 data of Antarctica's fast glaciers for change detection, feature tracking estimates of surface velocity, and interferometric analysis of velocity and coherence over the entire viewable region, which extends north of 80.1°S latitude. This paper reviews the MAMM project and describes the techniques to be used in processing the data. An example of data acquired over the Drygalski ice tongue, Antarctica illustrates how MAMM data will further benefit investigations of the icy continent.

1. Introduction

[2] The RADARSAT 1 Antarctic Mapping Project is a collaboration between NASA and the Canadian Space Agency (CSA) to map Antarctica using synthetic aperture radar (SAR). SAR is used because of its ability to image the surface in all weather conditions and during the day or night. SAR also yields high-resolution imagery with good contrast between sea ice, glacier ice, and rocky outcrops as well as between discriminating features on the ice sheet surface such as snow facies, crevasses, flow stripes, snow dunes, and even evidence of human activity such as aircraft landing strips and traverse tracks.

[3] The first Antarctic Mapping Mission (AMM 1) was successfully completed in October 1997 and yielded the first high-resolution radar mosaic of Antarctica (Figure 1) [Jezek, 1999]. The Modified Antarctic Mapping Mission (MAMM) occurred during the fall of 2000 [Jezek, 2002]. MAMM had two goals, which complemented scientific objectives for understanding the mass balance of the polar ice sheets and the response of the polar ice sheets to changing climate. The MAMM goals were (1) to produce high-resolution image mosaics of Antarctica north of 80°S latitude for change detection measurements and studies to understand the response of the ice sheet to climate change and (2) to measure the surface velocity field over coherent and/or trackable areas of the ice sheet north of 80°S latitude for ice dynamics studies and for exploring time variations in the surface velocity through comparisons with earlier data sets.

Figure 1.

1997 RADARSAT 1 Antarctic Mapping Project mosaic of Antarctica. The largest-scale tonal patterns are related to snow accumulation variations. Bright zones around the margin correspond to refrozen summer melt zones [Jezek, 1999].

2. MAMM Acquisition Phase

[4] The RADARSAT 1 synthetic aperture radar is a C band (5.3 GHz) instrument flown in a 24-day repeat orbit. To maintain the 24-day repeat cycle, the satellite elevation is routinely adjusted by firing its single thruster to compensate for deceleration due to drag. As drag acts on the satellite, it slows and eventually descends. As it does so, the satellite position drifts laterally relative to the nominal nadir track. During low-drag periods, as experienced during AMM 1, lateral displacements over time are small and gradual. Consequently, there is a relatively infrequent requirement for accelerating the satellite back to a higher elevation, and thruster firings can be separated by over 40 days.

[5] Unlike AMM 1, the MAMM campaign occurred during the peak of the solar activity cycle. Variable and, at times, high-drag levels increase the rate and magnitude of the lateral displacement relative to the nominal satellite track. To compensate for these conditions, more frequent, custom-designed adjustments to the satellite orbit are necessary in order to meet navigation requirements for interferometric applications. The basic navigational requirement for RADARSAT 1 interferometry is that the relative distance between repeat satellite observation positions be less than ∼250 m in order to maintain acceptable coherence, a quantity which improves with increasing radar system bandwidth, signal wavelength, and incidence angle. To meet this requirement, the Canadian Space Agency flight dynamics team implemented a series of orbit maneuvers based on the measured satellite orbit and the predicted solar drag. Maneuvers occurred with periods as short as 6 days. Frequent orbit maneuvers can only be executed in north looking mode because of the spacecraft design, which includes only a single thruster. Thus north looking operations restricted coverage to regions north of 80.1°S latitude but enabled the satellite to be routinely positioned to within several hundred meters of a modified nominal track, which was defined by the position of the satellite during the first cycle of acquisitions.

[6] RADARSAT 1 fine-1 beam data were acquired during MAMM to optimize the data set for change detection and surface velocity measurements. These data provide an unprecedented opportunity to make detailed images of many of Antarctica's fast glaciers, the extent of which were revealed through AMM 1 data [Jezek, 1999]. The data set was also optimized for interferometric analysis of the surface velocity field by acquiring fine-1 beam data, which, for a single look, have an azimuth resolution of 8.4 m and a range resolution of 5.2 m (J. Lipscomb, personal communication, 2001). This is because interferometric coherence becomes less sensitive to the satellite baseline as the range resolution improves. Said alternatively, the coherence improves as the radar system bandwidth is increased for a fixed baseline separation. RADARSAT 1 fine-1 beam data are collected with a bandwidth almost twice that of standard-1 and -2 beams and almost 3 times that of standard-6 beam.

[7] MAMM began on 3 September 2000 and lasted until 14 November 2000, an interval corresponding to three repeat cycles. Acquisitions were scheduled by the CSA, and data were downlinked to the Alaska SAR Facility (ASF), the Prince Albert satellite station, the Gatineau satellite station, and the McMurdo ground station. The acquisition plan was developed by the Jet Propulsion Laboratory and was designed to maximize fine beam coverage over fast glaciers, to obtain ascending and descending coverage of the viewable area using a combination of fine, standard, and extended low beams, and to obtain three complete cycles of data for interferometric analysis.

3. MAMM Processing Phase

[8] A custom-designed processing system is being developed to convert the MAMM data into map-quality image mosaics, coherence mosaics, and velocity fields. The RADARSAT Antarctic Mapping System 2 (RAMS 2) is an augmentation to the existing RAMS 1 developed for AMM 1 (Figure 2) [Norikane et al., 1996]. The key RAMS 1 functions are to (1) divide the map into manageable data blocks, (2) utilize block processing, which performs a block adjustment by reconciling orbit ephemeris data using ground control points, (3) perform orthorectification and radiometric balancing of the scenes and create a preliminary map tile, (4) perform grand adjustment to remove residual radiometric seams, (5) perform a final block-to-block radiometric adjustment, and (5) process tiles to create individual map tiles ready for DVD mastering. The RAMS 2 system will support the existing map formation capabilities of RAMS and, in addition, will have the capability to process interferometric SAR (IFSAR) acquisition pairs for a variety of IFSAR products, including coherence maps and ice velocity maps (Figure 3).

Figure 2.

Schematic of RAMS 1 functions.

Figure 3.

RAMS 2 processing diagram. The shaded elements are RAMS 1 functionalities.

[9] The development and operation of RAMS 2 software is divided into two phases. Phase 1 upgrades the existing RAMS system to ingest the new data formats and to produce image mosaic products identical to the RAMS 1 products (exceptions include the creation of higher-resolution “minimosaics” which retain details captured in fine-1 beam images). In addition, interferometric processing is carried out by RAMS 2 phase 1 software for the creation of coherence images and raw interferometric products (interferograms and registration fields). Phase 1 software will produce a coherence mosaic map similar to the image mosaic map. The raw interferometric results will be archived for use in phase 2. Phase 2 RAMS software will ingest the interferometric results and produce a global ice velocity map of the continent.

3.1. Interferometric Frame Planning and Processing

[10] The initial processing plan treats the AMM 1 and MAMM data quite differently. In AMM 1, with only one cycle of data considered, the framing was standard 100-km frames assigned by ASF. RAMS could form equivalent mosaic products regardless of how the data were framed as long as there was some overlap between scenes along the track. Because MAMM interferometric products are to be derived from the data across three cycles, there are several new restrictions placed on the data processing and framing plan.

[11] A process was established in which the ASF level 0 (raw data products), corresponding to three entire data takes of a “triplet,” along with low-resolution quick-look imagery are loaded into RAMS 2. RAMS 2 ingests this data and forms “chop” files for each of the three data takes so that the along-track frames are geo-coincident and have appropriate Doppler characteristics for interferometry and frame boundaries that do not bisect glaciologically interesting regions. Figure 4 shows this process at a high level.

Figure 4.

Schematic of the process of loading level 0 ASF data into RAMS 2. The appropriate framing of triplets is critical for interferometric processing. A new function of the RAMS 2 system is the definition to ASF of the required data framing for each cycle as well as any special processing considerations to be applied to these data. This new RAMS 2 function is carried out using level 0 data and metadata provided from ASF.

[12] Many of the processing parameters required to guarantee the highest-quality interferometric processing across the three cycles produce less than optimal image products. For instance, the azimuth processing bandwidth used for each of the single look complex (SLC) products was considerably larger than the nominal 900 Hz so as to maximize the Doppler overlap and hence maximize the potential coherence between the three cycles of data corresponding to a triplet. Therefore further additional processing is required within RAMS 2 upon the ingestion of the SLC data in order to produce high-quality multilooked imagery data for mosaicing. This processing includes Kaiser weighting and band-pass filtering to reduce sidelobes to acceptable levels, antenna gain corrections along track to compensate for a drifting antenna beam, and radiometric adjustments to compensate for variable processing parameters (such as processing bandwidth) as determined by the planner to ensure a radiometrically calibrated image mosaic product.

3.2. Image Registration for Coherence, Phase, and Speckle-Matching Measurements

[13] A challenge in automatically registering IFSAR image pairs is that sometimes the best results are obtained using coherent matching techniques (locking up on the coherent SAR signals), while other times only incoherent matching techniques are successful (matching image features). We developed a combined approach for registering the two complex images, which attempts to be efficient while ensuring that as many registration offsets as possible are measured as accurately as possible throughout the scene.

[14] Our approach takes advantage of the fact that incoherent offset measurements are much faster to calculate than coherent measurements and that they can be used to limit the search window for the more accurate coherent technique. Our technique attempts to first find incoherent offsets at different scales and then to measure the coherent offsets over a limited region. The computational efficiency of this approach is data dependent; however, we have found that it usually takes less time than the coherent technique alone and results in the most accurate offset measurements possible. The implementation within the RAMS 2 system permits the operator to choose the combined technique or either the amplitude or the coherent technique alone. The operator also controls the search window size. In this manner, the technique that produces the best results can always be utilized.

[15] During the development of the offset measurement techniques described above we made an interesting observation. Figure 5 shows an example of an image formed using the coherent technique and displaying the magnitude of the correlation peak, computed using 64 × 64-pixel chips and by using the same fringe rate to flatten the interferogram formed by the reference and slave single-look complex data chips. Aviator glacier, a fast-moving glacier which flows from upper left to lower right in the image, can be seen to have a fine structure on its surface. The source and explanation of this structure is not yet understood, and the filamentary patterns are not observable on the SAR amplitude image data or in the final coherence map (obtained after interpolating the registration offsets and resampling the slave image). We suspect that it is not an artifact of our measuring technique. The structure is reminiscent of melt patterns seen, for example, on the George VI Ice Shelf [Swithinbank, 1988, p. B116], but we hesitate to speculate too much about the pattern's physical significance.

Figure 5.

Magnitude of the coherence peaks used to calculate registration offsets, showing fine structures in Aviator Glacier (74°S, 165°W), which is ∼10 km wide. The reference orbit was acquired on 6 October 2000. The Doppler difference between frames is 130 Hz, and the baseline is 105 m. Black corresponds to low correlation peak values and white corresponds to high peak values. AG is Aviator Glacier.

4. MAMM Observations of Northern Victoria Land, Antarctica

[16] We illustrate an application of the MAMM data acquired over northern Victoria Land, Antarctica. The Drygalski ice tongue is shown in Figure 6. It is a long, relatively narrow glacier-ice tongue floating on the Ross Sea. It is the seaward extension of David Glacier, which is the largest outlet glacier of northern Victoria Land. Figure 6a shows 1997 standard-7 beam imagery orthorectified and terrain corrected using the RAMS 1 system. Figure 6b shows 2000 fine-1 beam data orthorectified and terrain corrected using the RAMS 2 system. Because only one swath is being analyzed, only ephemeris data were used to block adjust the MAMM data. No image-to-image tie points were used to readjust the geometry between the AMM 1 and MAMM data.

Figure 6.

RADARSAT 1 orthorectified images of Drygalski ice tongue, extending from the eastern edge of northern Victoria Land, Antarctica (inset) into the Ross Sea. (a) Acquired on 20 September 1997 with standard-7 beam. (b) Acquired on 11 October 2000 with fine-1 beam. East is to the left of each image. Darker tones in the left upper portion of each image are sea ice; lighter tones are glacier ice. Triangles and squares indicate the locations of velocity measurement point measurements.

[17] The serrated margin is typical of floating ice tongues and provides easy references for comparing common points between images acquired at different times. Displacements of 10 points were measured starting ∼20 km down-glacier from the grounding line and ending at the tip of the ice tongue (Figure 6). For display in Figure 7 the velocity at the tip was extrapolated to a range point on the southern flank in line with the previous nine measurements. We estimate the total uncertainty in identifying common points to be ∼150 m (ephemeris errors and picking errors). This corresponds to a velocity error of 70 m yr−1. Were we to use tie points to register the AMM 1 and MAMM data, we estimate that our velocities would systematically increase by ∼35 m yr−1. However, as noted below, that step also introduces uncertainties because tie points are limited to rocky outcrops away from the tongue.

Figure 7.

Three-year average velocities (bold line with circles) measured along the southern margin of Drygalski ice tongue. Ranges are measured from the most westerly (landward) point shown on Figure 7. Error bars indicate combined velocity uncertainties due to errors in the satellite ephemeris and uncertainties in picking common points for displacement measurements on the sequential images. The thin line is a linear fit to the data.

[18] The Drygalski ice tongue has been studied since the early 1900s. Holdsworth [1985] compared historical, airborne photographic, and Landsat data to estimate a velocity of 730 ± 36 m y−1 for a point 50 km from the coast. This is similar to our result of 682 ± 70 m y−1 to within the estimated errors. Several other investigators analyzed sequential Landsat scenes by coregistering images with tie points. Our result at the landward end of the velocity profile (588 m yr−1) compares favorably with Swithinbank's [1988] estimate of 580 m yr−1 but is less than Lucchitta et al.'s [1993] estimate of ∼640 m yr−1. Our maximum value (703 m yr−1) is similar to the 15-year average maximum velocity (719 m yr−1 over 1973–1988) quoted by Frezzotti [1993; Frezzotti et al., 2000] but is again less than Lucchitta et al.'s [1993] value, whose estimated 15-year average velocity (1973–1988) near the tip is ∼800 m yr−1. Our maximum values are also less than Frezzotti's 13-month average maximum velocity of 912 m y−1 (1988–1990). Frezzotti et al. [1998] compare their Landsat-derived velocities with Global Positioning System data collected on the ice tongue. They find that the GPS velocities agree with the Landsat feature-retracking data on the landward half of the ice tongue. Like our data, the GPS-derived velocities (714 m y−1) are slower than the Landsat feature-retracking velocities (750–800 m y−1), possibly indicating problems in registering Landsat imagery over the extreme parts of the ice tongue using data from tie points assembled on the land. Finally, our estimated average longitudinal strain rate is 1.2 × 10−3 yr−1 based on the line fit to the data in Figure 7. This is about a quarter of the value estimated by Frezzotti [1993].

[19] Even in view of measurement uncertainties, the differences between the measurements are intriguing and, hypothetically, may represent temporal changes in glacier speed. The MAMM data may help reconcile the observations by illuminating whether there are short- and longer-term variations in glacier speed. The AMM 1 and MAMM data yield a 3-year average velocity. The three cycles of MAMM fine-1 beam data will be able to yield short-term (48 day) estimates of average velocity using feature retracking (range and azimuth resolutions are 5.2 and 8.4 m, respectively (J. Lipscomb, personal communication, 2001); this corresponds to ∼9 pixels of displacement over 48 days at a speed of 600 m y−1). The MAMM IFSAR data will provide a 24-day average velocity estimate.

5. Summary

[20] Through the efforts of the Canadian Space Agency and NASA, extraordinary amounts of Antarctic data have been acquired during the RADARSAT 1 Antarctic Mapping Project. A total of 2179 and 4742 min of data were acquired during AMM 1 and MAMM, respectively. This corresponds to 898 and 2380 separate acquisitions for each observation epoch, respectively. The shear volume of data requires the development of automated analysis procedures as described in section 3. As we have attempted to illustrate, the effort is resulting in a new view of Antarctica that depicts in great detail the largest-scale features associated with continental scale patterns of accumulation rate, smaller-scale features such as the intricacies of cracks developed as ice from the continental interior spreads outward onto the ocean, and the motion of the ice itself. Mosaicing of the AMM 1 data is now complete, and the results have been widely distributed to the scientific community. The results from MAMM will require several more years to process completely; however, once done, the data will be similarly distributed through the NASA data centers.


[21] This research was supported by a grant from NASA's Pathfinder Program and from NASA's Polar Oceans and Ice Sheets Program. We thank James Miller of Vexcel for help in producing the images of the Drygalski ice tongue.