Using traditional methods, it is currently not possible to detect and control the melting of ground ice before landslides are initiated. 

The properties of satellite interferometric synthetic aperture radar (SAR) data were assessed to determine whether it may be suitable to detect the small changes in the Earth’s surface associated with the initial stages of ground ice melting. Amplitude, coherence, and interferometric phase images were generated from the SAR data from ERS-1/2, RADARSAT, and JERS-1 satellites, and A high-resolution digital elevation model, created from airborne cross-track SAR interferometry.

In this instance, RADARSAT data appear to be best for detecting the small surface changes associated with melting ground ice. RADARSAT has a suitable repeat pass period and the images retained enough coherence that areas of ground melt could be detected.
 
 

Recent climatic warming has initiated the melting of ground ice throughout permafrost regions. This is resulting in increased landslide activity and significant terrain disturbance. A direct consequence of these alterations to the land surface is damage to ecosystems and infrastructure (e.g. roads, buildings and pipelines).

Before landslides or significant surface disturbance is initiated, melting ground ice is thought to cause small changes in the morphology and elevation of the ground surface. With the detection of these surface changes, mitigative measures can be taken to retard the ground ice melt before landslides occur. Unfortunately, conventional techniques cannot provide the three essential elements required to detect the initial stages of ground ice melt: extensive spatial coverage (tens of square kilometres), frequent measurements (monthly or less), and fine-resolution change detection measurements (sub-metre).

Cross-track airborne interferometric SAR is now commonly used to create DEMs of the land surface (Gray et al., 1995; Orwig et al., 1995; Moorman et al., 1998). In certain situations, repeat-pass satellite interferometric SAR can also be used to generate DEMs of the Earth’s surface and produce associated temporal derivatives (e.g. glacier flow rates (Vachon et al., 1996)). One of the limiting factors in using satellite SAR data for interferometry are the changes in the character of the ground surface over the time period between satellite passes. These changes can result in a reduction in the coherence between images and thus limit the potential for generating a DEM (Vachon et al., 1995).

This decorrelation between SAR images is of interest in detecting melting ground ice. Areas where there is relatively greater change in the character of the ground surface over time (e.g. were ground ice is melting) will be highlighted in a coherence image as areas of localized decorrelation. However, due to the slow rate of surface subsidence associated with the initiation of ground ice melt, longer observation periods are required for a detectable amount of surface change to develop. Unfortunately, as the repeat period of the satellite increases, so does the potential for widespread scene decorrelation due to other processes (e.g. vegetation growth, snow accumulation/ablation, or changes in the soil moisture content).

The coherence between SAR scenes (i.e. the ability to interferometrically correlate the two images) is dependent on a number of factors other than just ground surface change (Vachon et al., 1995). In generating interferometric images from satellite SAR data, both the orbital and sensor characteristics of the satellite play a major role in the suitability of the data for interferometric processing. Orbital considerations include the repeat period of the satellite and its across-track repeatability. Sensor considerations include spatial resolution, incidence angle, radar wavelength, and signal-to-noise ratio. For example, the longer wavelength of the JERS sensor is theoretically less sensitive to the small changes that would effect the ERS and RADARSAT sensors, thus decreasing resolution, but at the same time decreasing the level of background noise.

The optimal system for locating melting ground ice would show a loss of coherence in the areas of melt, while coherence is retained throughout the rest of the scene. The different system configurations of the sensors considered (see Table I), enabled examination of the loss of coherence within a scene due to geomorphological processes which occur at different rates.

For this project, amplitude, coherence, and phase images were generated from ERS-1/2, RADARSAT, and JERS image pairs for roughly the same area. The characteristics of the three systems are compared in relation to observed scene changes and the rate of geomorphic activity. The suitability of each sensor for detecting ground ice melt is discussed.
 

A study area on southern Bylot Island in the eastern Canadian Arctic was chosen for this project as it is known to contain a variety of proglacial massive ground ice bodies that are currently melting at varying rates (Klassen, 1993). The area’s cold and dry climate results in little vegetation growth or snow accumulation, thus the potential for detecting ice-melt related changes is enhanced (Fig. 1).

Currently, 10 of the 18 larger glaciers within the southern portion of the island are retreating. The other eight show no signs of change within the last century (Moorman, 1998). Generally, the moraines surrounding the retreating glaciers are ice cored. Ground ice bodies observed on the island range in size from 1 m³ to over 200 000 m³. As the mean annual air temperature in the region is approximately -9.5ûC, meltout of the ice-cored moraines does not occur spontaneously following deglaciation as in more temperate environments. Some ground ice bodies were found to still exist in areas not glaciated for tens of thousands of years. The ultimate result of ground ice melt in this environment is the development of thermokarst lakes, many of which can be observed throughout the lowlands.
 
 

The character of the interferometric fringes generated for three data sets from the sensors of Tzable I are displayed in Fig. 2. The ERS-1/2 image has a high degree of coherence in this region and the grey cycles depicting the phase fringes can be easily seen. Only in the high relief terrain at the base of the image is there a decrease in the coherence. The loss of coherence within the JERS and RADARSAT images at this location makes delineation of the phase fringes more difficult.

In all three data sets the coherence is variable across the scene, and is well correlated with terrain type (Fig. 3). The relationship between coherence and terrain type can be explained by the rate of surface activity within each terrain unit (Table II). As can be seen in Fig. 3, there is a lower coherence for the areas covered by the (relatively rapidly moving) glaciers. The small dark spot in front of the terminus of the glacier on the left is an area of open water or slush formed from the runoff from a subglacial spring (the location is indicated in Fig. 1.)

The dark area indicated by the arrow in Fig. 3 is a retrogressive thaw flow. Ground ice melts very rapidly in the summer and the resultant saturated mudflows can continue to move well into the winter. However, the amount of surface change expected from ground ice melt before a retrogressive thaw flow is initiated would be much less. Thus, a 1 day repeat period is of minimal use in detecting pre-landslide surface changes.

In the RADARSAT coherence image shown in Fig. 4, there is a lower overall coherence relative to the ERS image. This is due to the longer repeat period between the image pair. However, this image displays the same patterns as the ERS image. The glacial valley shows the greatest coherence while the glaciers show the least. Consequently, local areas where coherence is lost can still be detected in the valley (e.g. the wet area in front of the glacier of the left). Note the ground ice is most frequently found in glacial valleys.

This JERS data had low coherence values with only the valley floors and some of the upland terrain showing appreciable coherence. Differentiating between local areas having rapid surface change would be difficult with this data set.
 
 

This preliminary investigation into the suitability of repeat-pass satellite interferometric SAR for detecting ground ice melt reveals a number of considerations in application of this technique to geomorphological analysis: specifically,

• By comparing coherence images from three sensors, the rate of surface change in different terrain types was estimated, and its impact on interferometric analysis determined
• ERS tandem mode data has the greatest coherence due to its short repeat period, but this is too short a period for detectable change to occur.
• RADARSAT data was less coherent than the ERS data, but the 24 day repeat pass period makes it more suitable for detecting melting ground ice before landsliding is initiated.
• Even with the longer wavelength of the JERS data, there was minimal coherence between the images. The major factors likely being the lengthy repeat pass period of 44 days and the seasonal timing may of the imagery.

We thank A.L. Gray (CCRS), K. Mattar (Intermap), and D. Geudtner (DLR) for their contributions to the airborne and satellite InSAR processing at CCRS. We thank B. Armour (Atlantis Scientific) for helpful discussion. The RADARSAT data were acquired through CSA ADRO project #500 and are copyright CSA. The JERS-1 data were acquired through the NASDA JERS-1 Research Invitation and are copyright ESA. The ERS data are copyright ESA. C. Livingston (CCRS) expedited the acquisition of the CV580 data. We also thank F. Michel, L. Moorman, D. Kliza, and M. Elver for assistance in the field. Field logistical support was supplied by the Polar Continental Shelf Project. Thanks also go to the Hamlet of Pond Inlet for supporting this project.

Gray, A. L., Mattar, K. E. & van. der Kooij, M. W. A.., 1995. Cross-track and long track airborne interferometric SAR at CCRS. In: Proceedings, 17th Canadian Symposium on Remote Sensing, Vol. 1, pp. 232-237. Canadian Remote Sensing Society, Saskatoon, Canada., 232-237.

Klassen, R. A., 1993. Quaternary geology and glacial history of Bylot Island, Northwest Territories, Geological Survey of Canada Memoir 429, pp. 93.

Moorman, B. J., 1998. The Development and Preservation of Tabular Massive Ground Ice in Permafrost Regions, Ph.D. thesis, Carleton University, Ottawa, 308 pp.

Moorman, L. A., Geile, W., & Mercer, B., 1998. New Frontiers: Environmental applications of high accuracy DEMs. To appear in: Proceedings, 27th International Symposium on Remote Sensing of Environment, Tromso, Norway.

Orwig, L. P., Aronoff, A. D., Ibsen, P. M., Maney, H. D., O'Brien, J. D. & H.D. Holt Jr., 1995. Wide-area terrain surveying with interferometric SAR. Remote Sensing of the Environment, 53, 97-108.

Vachon, P. W., Geudtner, D., Gray, A. L. & Touzi, R., 1995. ERS-1 synthetic aperture radar repeat-pass interferometry studies: implications for RADARSAT. Canadian Journal of Remote Sensing, 21, 441-454.

Vachon, P. W., Geudtner, D., Mattar, K., Gray, A. L., Brugman, M. & Cumming, I. G., 1996. Differential SAR interferometry measurements of Athabasca and Saskatchewan Glacier flow rate. Canadian Journal of Remote Sensing, 22, 287-296.
 

Table I
SENSOR PARAMETERS

Table II
THE LOSS OF SCENE COHERENCE OVER TIME OF DIFFERENT TERRAIN TYPES
 

Rate of Decorrelation	Terrain Type	Cause
Rapid	streams and slush flows	Movement of water or slush
Moderate	glaciers and steeply-sloping terrain	Ice movement and snow accumulation/erosion
Slow	barren, gently-sloping upland terrain	Change in moisture content or snow accumulation/erosion
Very slow	valley bottoms, well-drained gravel	Lack of vegetation and moisture, stays windswept and dry

 

Figure 1. A SAR image of the study area on southern Bylot Island acquired with the CCRS CV580 in March 1996. The arrow indicates the location of a spring emerging from near the terminus of the glacier on the left. The region shown in Fig. 2 is outlined.
 

Figure 2. Phase images of a location in the southern portion of the study area. E =ERS-1/2, J = JERS, and R = RADARSAT. See Fig. 1 for the location of these images. The lack of coherence within the JERS and RADARSAT images makes it difficult to determine the location of the fringe boundaries.
 

Figure 3. ERS-1/2 coherence image. The gently sloping valley bottom and plateau areas have a higher coherence than the glaciers and steep valley slopes. The dark area just in front of the glacier on the left is open water produced from a subglacial spring.
 

Figure 4. RADARSAT coherence image. The glaciers and steep slopes have lost all coherence in the 24 days between images, while the valley floor still shows some coherence.