Uncertainties in Firn-Line Monitoring in Winter SAR Data

By Ian A. Brown, Department of Physical Geography and Quaternary Geology, Stockholm University, Sweden

Since the launch of ERS-1 in 1991, SAR imagery has become widely used in glaciology. The ability to image a surface, regardless of cloud cover and solar illumination, is particularly useful in a discipline which is often concerned with polar regions or areas with a prevalence of cloud cover. Sea ice monitoring in North America and northern Europe has been one of the great success stories of spaceborne SAR. Now operational glacier monitoring using optical-thermal and SAR systems are underway. For example, the Global Monitoring for Environment and Security (GMES) Polarview service provides products describing the position of facies boundaries at periodic intervals to support in situ mass-balance and runoff observations. The improvement in resolution and increased flexibility in imaging, inherent in the next generation systems, may lead to greater adoption of SAR in glacier monitoring.

One key parameter used in operational monitoring and glaciological research is the firn limit; the lower boundary of firn accumulation. Firn is snow left over from previous seasons that has been compacted and recrystallized. The firn limit equates to the equilibrium line: the boundary at which accumulation and ablation are equal and the glacier is in balance. On a valley, this boundary is generally assumed to conform laterally to the topography and can, therefore, be described by an altitude; the equilibrium line altitude (ELA). On an icecap, ELAs may be determined for different outlet glaciers or aspects to allow for differences in accumulation and ablation patterns. The ELA is frequently used to describe the annual ‘state’ of a glacier and is a common forcing mechanism in glacier-climate modelling. In SAR images with dry snow conditions, the firn limit, a steady-state
proxy of the ELA, can be identified as the lower limit of the bright scattering zone associated with firnification and percolation. This boundary may shift up-glacier as a result of increased melting on an annual basis. However, as firnification can take several years, increased accumulation may not be immediately identifiable in annual SAR images (as dry snow generates low volume scattering). The firn limit in SAR images is, therefore, marked by a degree of uncertainty unless there is a known net mass loss. Futhermore, the sensitivity of a SAR to firn thickness is not entirely understood. In optical images, snow a few centimeters thick may have the same signature as snow tens of metres thick covering 2 kilometers of glacier ice. In SAR data, dry snow, a few centimeters thick, may be undetectable and at what depth of firn the firn-limit threshold is reached may depend on the imaging geometry, snow/firn density and grain-size distribution. While firn-limit mapping is useful, further investigation is needed to refine the method.

Since 1998 we have collected RADARSAT-1 data on Blåmannsisen, an ice cap in north Norway. In 1998 and 2005, ground penetrating radar profiles (GPR) were taken in situ and an ASTER image from August 30th, 2001 was also acquired. The GPR profile from 1998 traversed the glacier from south to north. In the centre of the glacier, at an altitude of 1160 m, well above the firn limit, the GPR data indicated zones in which the snow and firn layering appear to have been planed off (removed by surface melting). The ASTER data from late summer 2001 also showed patches in which the annual snow cover had melted away exposing firn or ice. These zones, along with other zones of higher surface melt extracted from the ASTER image provide the opportunity to test whether the SAR data are sensitive to small changes in firn thickness. Additionally, by measuring backscatter near the firn limit relative to backscatter above and below the limit, we can refine our zonation of backscatter and ultimately may be able to determine the gradient in backscatter related to the gradient of firn depth from the firn limit up-glacier.

Six images were used in the analysis of backscatter signatures (Table 1). The images were calibrated and georeferenced using the ASF convert tool. The georeferencing was refined using a Rational Functions approach with a 30-m Digital Elevation Model (DEM) and 1:50000 scanned maps as the reference layers. The ASTER image was georeferenced using the Geomatica implementation of the Toutin method (satellite orbital modelling) with the same reference data. The SAR images were gereferenced to approximately 2 pixels (25-m) although the ASTER image appeared to have an offset of perhaps 5 pixels. It was not possible to determine whether there was a georeferencing bias in the middle of the icecap, so it was assumed that errors were more or less evenly distributed where surface slopes were low to moderate. From the ASTER data, a set of polygons was delimited describing different surface zones. Reference zones of firn and bare glacier ice were selected along with small patches of melt determined from ASTER and GPR data and other melt zones from the ASTER data alone. Given the small size of the melt patches observed in the ASTER and GPR data speckle filtering was not performed so as to limit the smoothing and resampling of the SAR data. A further class of “thin-firn” near the firn limit was delimited from the ASTER reference data. The polygons in the central icecap were digitized from the ASTER image and from the region identified in the GPR data. The polygons were small; in all, 11 polygons contained 466 pixels. The mean backscatter from these polygons was -3.56 dB, statistically inseparable from the firn classes. Larger polygons exhibiting melt derived from ASTER data alone had lower backscatter (mean -6.47 dB). The polygons adjacent to the firn limit, assumed to be containing thin firn, had mean backscatter values of -5.11 dB, -5.07 dB and -3.87 dB for northern, eastern, and western samples respectively (Figure 2). The original small patches could not be distinguished from the background signal of the firn area. The larger melt patches and thin firn regions could be distinguished by their lower backscatter, although not in all images.

Our dataset comprises a range of beam types acquired during different seasons, testing the wider applicability of the method. The best performance was from the ST2 beam images acquired in late winter 2000 and the spring 2000 image. The image from November 2000 (2000-304) may have been influenced by recent melt-refreeze events producing anomalously bright signals from the region below the firn limit or a very rough surface causing greatly enhanced surface scattering than is normal. Radiometric correction of topographic effects would probably improve these results or at least dampen the effect of some of the anomalies. The georeferencing of the ASTER data should also be readdressed. Speckle filtering might also improve the signal analysis, but would result in smoothing, reducing the viability of small image samples.

These preliminary results suggest that scattering above the firn limit is strongly affected by the depth of firn and that melt effects might be detected in the firn area using SAR data. In all the images, the firn limit could be clearly delineated, but it is probable that this boundary equates to at least one meter of firn rather than the lowest limit of firn. A Rayleigh volume scattering model coupled with a surface scatter Small Peturbation Model estimated one meter of firn to have a backscatter of -6.78 dB (using in-situ measurements near the SAR firn limit). Below the “SAR firn limit” a polygon of 109 pixels averaged -7.09 dB in the images (range -6.57 dB to -8.74 dB) suggesting that the firn limit is better regarded as the firn gradient.

Click here to download a copy of the newsletter featuring this article