variety of methods exists for finding corresponding points in two images
(Goshtasby 2012). Successful examples of mapping the kinematics of glaciers,
rock glaciers and other mass movements by remote sensing techniques are given,
for example, by Ka ¨a ¨b (2005).
In the present study we focused on the kinematics of the tongue of Pasterze
Glacier. Feature tracking, i.e., finding corresponding points in the multi-temporal
dataset, was accomplished by means of automatic image matching maximizing the
normalized cross-correlation coefficient (NCC) at pre-defined grid points.
Sub-pixel accuracy was achieved by interpolation of a parabola at the position of
the peak-value of the correlation function. Back-matching, i.e. applying the image
matching algorithm in the reverse direction for consistency check, helped to sort out
most of the gross errors. Remaining outliers of the 2D displacement vectors were
identified visually and eliminated manually. The accuracy obtained was quantified
at stable regions, e.g. bedrock, where no surface movements can be expected.
Two morphologically interesting areas at Pasterze Glacier were investigated
(Figs. 9.3 and 9.4). The image matching technique applied will fail if (1) the
geometry of the two patches to be compared has changed excessively, (2) the
surface textures have decorrelated in time, or (3) sufficient surface texture is
completely lacking. The results of both test sites show that areas with bare ice are
prone to rapid decorrelation of surface texture, and thus image matching fails. Best
results are obtained on completely debris-covered areas and on bedrock. In order to
fully benefit from the high resolution of the original photographs additional
orthophotos with a spatial resolution of 0.25 m were computed for both test sites.
Lower test site (LTS) The window size for image matching was 41 pixel Â
41 pixel. Flow velocities obtained are accurate to Æ0.17 m a
À1 (2003–2006) and
Æ0.12 m a
À1 (2006–2009).
Upper test site (UTS) Here the surface texture is mostly determined by
supraglacial debris and ogive-type structures. The flow velocities are much higher
than in the lower test site, resulting in faster surface texture decorrelation. This
problem was overcome by re-computing the orthophotos at a relatively large
GSD of 2 m, and increasing the window size for successful image matching to
101 pixel  101 pixel. Subsequently, accuracies of flow velocities obtained are
lower than for the lower test site, i.e., Æ0.24 m a
À1 in the best case for stable areas.
9.3.2.6 Glacier Surface Elevation Change
Glacier mass balance can be computed using the geodetic method or the glaciological method. Both methods have advantages and disadvantages (Benn and Evans
2010; Fischer 2011; Zemp et al. 2013). The geodetic method is based on simple
glacier surface elevation change. The numerical transformation of volume change
to mass loss or gain requires spatial information on the density of the material
involved, i.e. ice, firn and snow. Most glacier studies assume a mean density of
900 kg m
À3 for glacier ice (Huss 2013). Specific mass balances are often calculated
in mm water equivalent (w.e.).
9 Glaciological Studies at Pasterze Glacier (Austria) Based on Aerial Photographs
181
(Goshtasby 2012). Successful examples of mapping the kinematics of glaciers,
rock glaciers and other mass movements by remote sensing techniques are given,
for example, by Ka ¨a ¨b (2005).
In the present study we focused on the kinematics of the tongue of Pasterze
Glacier. Feature tracking, i.e., finding corresponding points in the multi-temporal
dataset, was accomplished by means of automatic image matching maximizing the
normalized cross-correlation coefficient (NCC) at pre-defined grid points.
Sub-pixel accuracy was achieved by interpolation of a parabola at the position of
the peak-value of the correlation function. Back-matching, i.e. applying the image
matching algorithm in the reverse direction for consistency check, helped to sort out
most of the gross errors. Remaining outliers of the 2D displacement vectors were
identified visually and eliminated manually. The accuracy obtained was quantified
at stable regions, e.g. bedrock, where no surface movements can be expected.
Two morphologically interesting areas at Pasterze Glacier were investigated
(Figs. 9.3 and 9.4). The image matching technique applied will fail if (1) the
geometry of the two patches to be compared has changed excessively, (2) the
surface textures have decorrelated in time, or (3) sufficient surface texture is
completely lacking. The results of both test sites show that areas with bare ice are
prone to rapid decorrelation of surface texture, and thus image matching fails. Best
results are obtained on completely debris-covered areas and on bedrock. In order to
fully benefit from the high resolution of the original photographs additional
orthophotos with a spatial resolution of 0.25 m were computed for both test sites.
Lower test site (LTS) The window size for image matching was 41 pixel Â
41 pixel. Flow velocities obtained are accurate to Æ0.17 m a
À1 (2003–2006) and
Æ0.12 m a
À1 (2006–2009).
Upper test site (UTS) Here the surface texture is mostly determined by
supraglacial debris and ogive-type structures. The flow velocities are much higher
than in the lower test site, resulting in faster surface texture decorrelation. This
problem was overcome by re-computing the orthophotos at a relatively large
GSD of 2 m, and increasing the window size for successful image matching to
101 pixel  101 pixel. Subsequently, accuracies of flow velocities obtained are
lower than for the lower test site, i.e., Æ0.24 m a
À1 in the best case for stable areas.
9.3.2.6 Glacier Surface Elevation Change
Glacier mass balance can be computed using the geodetic method or the glaciological method. Both methods have advantages and disadvantages (Benn and Evans
2010; Fischer 2011; Zemp et al. 2013). The geodetic method is based on simple
glacier surface elevation change. The numerical transformation of volume change
to mass loss or gain requires spatial information on the density of the material
involved, i.e. ice, firn and snow. Most glacier studies assume a mean density of
900 kg m
À3 for glacier ice (Huss 2013). Specific mass balances are often calculated
in mm water equivalent (w.e.).
9 Glaciological Studies at Pasterze Glacier (Austria) Based on Aerial Photographs
181