|
|
 |
Satellite data synergies for monitoring Arctic ice masses
Abstract
-
Mass balance is a key parameter for assessing the health of a glacier or ice cap. By routinely locating
the position of the snow line, and hence equilibrium line, at the end of the ablation season it is possible
to monitor changes in mass balance. Currently this is not possible using single satellite data sources,
but we demonstrate that by combining different satellite data sets it is possible to generate a DEM and
locate the position of the snow line.
Keywords:
glaciers, ice caps, Landsat, radar altimeter, SAR
The mass balance of a glacier or ice cap at any time is defined as the sum of the ablation and
accumulation. Measured at the end of the ablation season, it is a key parameter for monitoring
the health of a glacier and hence possible changes in the climate. Due to the inaccessability of
the polar regions, satellite remote sensing provides an ideal tool for monitoring these changes.
In this paper we explore possible synergies between ERS-1 radar altimetric (RA),
synthetic aperature radar (SAR) and Landsat visible data, and how these might be used to monitor
changes in mass balance. To date, our work has been focused on the smaller terrestrial ice
masses of the High Arctic and here we present data from Austfonna and Finsterwalderbreen, a
glacier and ice cap on Svalbard (9-27 E, 77-81 N).
There are three possible approaches to measuring mass balance from space: direct measurement
of the accumulation and ablation terms, direct monitoring of the glacier volume or monitoring the equilibrium
line that separates the ablation zone from the accumulation zone. Direct measurement of the accumulation and
ablation terms has proven difficult. Zwally, 1977 and
Rotman et. al., 1982
have shown that it is possible to determine the accumulation rate from passive microwave observations,
but only to an accuracy of 20 %. In any case, the spatial resolution available from spaceborne passive
microwave radiometry is extremely coarse and is therefore only of use for monitoring the ice sheets of
Greenland and Antarctica. Similarly, the only ablation term that may be measured from space is calving
volumes estimated from iceberg areas. Direct measuremnt of surface melting and run-off is currently not
possible.
In order to measure changes in the glacier volume it is necessary to make observations
of the surface topography with a repeatable accuracy of better than 0.1 m
(Rees and Squire, 1989).
Possible techniques for achieving this include the use of RA data, shape-from-shading
algorithms applied to visible/near infrared imagery and interferometric SAR (InSAR).
However, these methods have all proven problematic and are not currently capable of
reaching the required accuracy. The RA is a low data-rate sensor with poor spatial resolution.
Uncertainties in the penetration depth can introduce errors of the order of 8 m
(Ridley and Partington, 1988)
and slope-induced errors of the order of 50 m for slopes of 0.01 radians are possible.
Over steep slopes the onboard tracker can lose lock which consequent loss of data.
Over dry snow areas a shape-from-shading algorithm can generate slope information from visible imagery.
However, it is both necessary to calibrate the slope map and define a constant of integration to produce
a topographic map. Both of these require some form of ground control data which is generally not available.
InSAR techniques have potential but they are difficult to process. The technique requires at least two
successive images that are highly correlated.
This can be problematic because weathering and other environmental effects, such as glacier
movements, can act to cause temporal decorrelation between the imagery. Furthermore ground
control points are required for phase unwrapping otherwise the resulting DEM ‘floats’ in a similar
way to the shape-from-shading algorithm.
The final method for measuring mass balance involves monitoring the position of the equilibrium line.
At the end of the ablation season the position of the equilibrium line is close to the snow line
(Paterson, 1994). If the snow line can be routinely detected
and superimposed on a sufficiently accurate digital elevation model (DEM), changes in the equilibrium line
altitude (ELA), and hence mass balance can be monitored.
It is well established that the snow line can be identified in visible imagery
(Williams et. al., 1991),
but these data can not be routinely used in the High
Arctic because of severe cloud problems.
Marshall et. al., 1994 have shown that the likelihood of
acquiring cloud free visible imagery of Svalbard at the end of the ablation season is less than
six percent. However, this problem of timing can be circumvented by the use of SAR data that are
potentially available in all weather
conditions and at any time of year. Interpretation of SAR imagery is notoriously difficult because
variations in the backscatter are caused by changes in the surface dielectric properties, roughness
and topography. Despite this
Jezek et. al., 1993 and
Marshall, 1995 have used in situ data to
demonstrate that it is
possible to delineate the snow line in SAR imagery acquired at the end of the ablation season. Without
the benefit of in situ data we explore the possiblity of using data synergies to monitor the snow line.
Data synergy involves the combination of independent datasets to extract new or to obtain more
accurate information about a parameter being measured. Here we aim to demonstrate the power of
this approach to (a) develop a DEM by using RA data to calibrate a shape-from-shading
algorithm applied to visible imagery, and (b) to combine visible and
SAR data to locate the position of the snow line, and consequently the equilibrium line at the end of the
ablation season .
In this work we recognise that the absolute accuracy of the DEM will not be accurate enough for
monitoring temporal changes in mass balance. Nevertheless, we suggest that the DEM will be sufficiently
accurate to monitor relative changes in the ELA, which may be deduced from visible and SAR imagery.
To date, airborne radio echo sounding surveys have been used to map selected ice-covered
areas of the world, but these surveys are one-off and expensive. More recently Seasat, Geosat
and ERS-1 RA data have been used to map the ice sheets of Greenland and Antarctica
(Zwally, et. al., 1983;
Rapley, et. al., 1993), demonstrating the role of satellite
remote sensing for mapping the polar regions. However, satellite altimetric data have not been used
to map smaller ice caps because of limitations in available data.
Figure 1 shows the available data on
Austfonna after averaging three cycles of ERS-1 35-day repeat data. Only summer data were used in order
to reduce penetration effects. The resulting DEM (Figure 2) contains
predominately low spatial frequencies and is consequently of little value. In order to improve the
spatial frequency we have applied a shape-from-shading algorithm
(Rees and Dowdeswell, 1988)
using a Landsat MSS image to model the topography of the ice cap and the ERS-1 RA data to constrain
the model. In order to acheive this, profiles of Landsat MSS band 6 brightness values were
generated from transects running parallel to the solar azimuth and separated by one pixel in the
east--west direction. Whenever a transect crossed an ERS-1 RA data track the surface height measured
by the RA and position along the transect were recorded and used to constrain the modelled DEM.
At least three tie-points are required to calculate the coefficients necessary to invert the algorithm.
When this was not possible, coefficients from the nearest available constrained transect were used.
Figure 1: Available ERS-1 altimetric data over Austfonna, for 3 cycles
worth of 35-day repeat data acquired in the summer of 1992.
 |
Figure 2: DEM of Austfonna produced entirely
from ERS-1 radar altimetric data.
 |
The new DEM (Figure 3) combines the high spatial frequency information
contained in the MSS imagery with the low spatial frequency altimetric data. The result is a DEM that is a
significant improvement on the DEM derived from altimetric data alone. The absolute accuracy of the DEM
is still poor due to contamination by slope-induced errors, although in principle these my be removed
when appropriate techniques are available. Moreover, with the development of the EOS GLAS many of
these problems will be eliminated. Comapared with radio echo sounding data, the new DEM, even using
slope-uncorrected RA data, has an RMS error of less than 30 m.
Figure 3: DEM of Austfonna produced from a
shape-from-shading algorithm applied to a Landsat/MSS image and constrained
with ERS-1 altimetric data.
 |
Having demonstrated the possibility of generating an accurate DEM by combing ERS-1 RA and visible
wavelength data, we now consider the possibility of defining the position of the snow line at the end of the
ablation season, and hence the equilibrium line, on a routine basis.
Figure 4 shows a SAR image of Finsterwalderbreen acquired at the
end of the 1995 ablation season. On the basis of tonal variations, the glacier can be readily divided into
a number of altitudinal zones with well defined boundaries. These boundaries (SL1, SL2 and SL3 in
Figure 4) are possible locations for the snow line. However, without
other information it is impossible to confirm which of these, if any, is the real snow line.
Figure 5 shows a Landsat/TM image of the same area acquired during the
middle of the 1993 ablation season. Although this image was obtained in a different year and at a
different time in the ablation season, there are clear spatial similarities with
Figure 4. This permits an unambiguous attribution of the 'potential' snow lines,
SL! and SL2, to morphological features and the real snow line as SL2. Verification of this has further been
obtained from aerial photography coincident with the SAR image.
Figure 4: SAR image of Finsterwalderbreen acquired at the end of the
1995 ablation season. SL1, SL2 and SL3 are possible locations of the snow line.
 |
Figure 5: Landsat/TM image of Finsterwalderbreen acquired during the mid 1993 ablation of the
ablation season.
 |
No single satellite remote sensing technique can currently give mass balance measurements with
sufficient accuracy. However, the work presented here suggests that there is scope for combining
RA, visible and SAR imagery to (a) generate a DEM and (b) combine the end of
ablation season snow line on with the DEM in order to infer changes in the ELA and
consequently changes in the mass balance with sufficient accuracyn any case, the spatial resolution available from spaceborne passive
microwave radiometry is extremely coarse and is therefore only of use for monitoring the ice sheets of
Greenland and Antarctica. Similarly, the only ablation term that may be measured from space is calving
volumes estimated from iceberg areas. Direct measuremnt of surface melting and run-off is currently not
possible.
In order to measure changes in the glacier volume it is necessary to make observations
of the surface topography with a repeatable accuracy of better than 0.1 m
(Rees and Squire, 1989).
Possible techniques for achieving this include the use of RA data, shape-from-shading
algorithms applied to visible/near infrared imagery and interferometric SAR (InSAR).
However, these methods have all proven problematic and are not currently capable of
reaching the required accuracy. The RA is a low data-rate sensor with poor spatial resolution.
Uncertainties in the penetration depth can introduce errors of the order of 8 m
(Ridley and Partington, 1988)
and slope-induced errors of the order of 50 m for slopes of 0.01 radians are possible.
Over steep slopes the onboard tracker can lose lock which consequent loss of data.
Over dry snow areas a shape-from-shading algorithm can generate slope information from visible imagery.
However, it is both necessary to calibrate the slope map and define a constant of integration to produce
a topographic map. Both of these require some form of ground control data which is generally not available.
InSAR techniques have potential but they are difficult to process. The technique requires at least two
successive images that are highly correlated.
This can be problematic because weathering and other environmental effects, such as glacier
movements, can act to cause temporal decorrelation between the imagery. Furthermore ground
control points are required for phase unwrapping otherwise the resulting DEM ‘floats’ in a similar
way to the shape-from-shading algorithm.
The final method for measuring mass balance involves monitoring the position of the equilibrium line.
At the end of the ablation season the position of the equilibrium line is close to the snow line
(Paterson, 1994). If the snow line can be routinely detected
and superimposed on a sufficiently accurate digital elevation model (DEM), changes in the equilibrium line
altitude (ELA), and hence mass balance can be monitored.
It is well established that the snow line can be identified in visible imagery
(Williams et. al., 1991),
but these data can not be routinely used in the High
Arctic because of severe cloud problems.
Marshall et. al., 1994 have shown that the likelihood of
acquiring cloud free visible imagery of Svalbard at the end of the ablation season
|
EO Data Access
