ith all three
Nun modern
^ preserve
me images
rmat by the
ition scans,
rmat size is
is available
relating to
though the
eters were
nation was
n was that
as (Spriggs,
ible. Hence,
a is not
t of Nemo
roximately
the front of
roximately
assified as
at the front.
, which is
te, modern
id. ASTER
cal studies
ifrared and
nportantly,
nich makes
h a spatial
ly detailed
track nadir
27.7? off-
spectrum,
extracted.
A's Land
AAC). For
ial stereo-
s available
calibration
.1 and the
€ 3B) with
n 47) A
e pair was
ints were
emoved if
ven that no
ised on the
1 projected
ellipsoid.
Additionally, an ortho-image was produced using the extracted
DEM. Processing of the ASTER data was relatively
straightforward compared to the archive imagery.
For the photogrammetric processing (interior, exterior, and
absolute orientation) of the USGS aerial imagery, BAE Systems
SOCET SET 5.5.0 was used. For automated DEM extraction the
SOCET SET NGATE module was used. Individual frames and
image blocks for each site were processed using the multi-
sensor triangulation (MST) module within SOCET SET. The
interior orientation of the USGS imagery is not trivial as
fiducial mark coordinates were not available. However, the
focal length and lens distortion parameters were known. The
radial lens distortion parameters for the two Metrogon USGS
camera lenses pertaining to the selected imagery are detailed in
Spriggs (1966) and plotted graphically in Figure 2. The USGS
frames contain four observable fiducial marks in the centre of
each image side. For their reconstruction the image coordinates
for each fiducial mark were measured using Adobe PhotoShop
CS2 and the centre of the image was determined as the
intersection between lines of opposing fiducial marks. The
principle point was assumed to be located at (0,0) and the
calculated intersection point gave the respective offsets to
reduce the fiducial mark coordinates. The final fiducial marks
coordinates were then calculated as the average of the reduced
fiducial marks. The described lack of calibration information is
not uncommon in studies that include archive imagery. The
work-around as described here has been successfully adopted in
other studies (Miller et al, 2007; Fox and Czifersky, 2008;
Dornbusch et al, 2010). A new camera calibration file was
created and the final interior orientation RMS was within 0.5
pixel.
0.15 % T T T
| — T11 DS2294
: T11 RF4224
0.05 F
0.00 [em
-0.05
Displacement [mm]
-0.10 Fr
015k i L i 1 À i i L 3 Jj
0 5 10 15 20 205 30 535 40 145 50
Field angle [degrees]
Figure 2: Lens distortion for the two Metrogon T11 lenses
For the exterior orientation, at least 25, regularly distributed tie
points per image per were measured automatically using the
automatic tie point matching facility within SOCET SET. The
resulting tie points were manually examined and adjusted or
deleted if necessary. Additional points at varying elevations
were added to strengthen the orientation solution. Distinctive
features such as mountain ridges, rock outcrops or features on
stable sea-ice were utilised. This produced a final exterior
orientation RMS value of less than one pixel.
Because no ‘true’ GCPs were available, the absolute orientation
of the historical image strips proceeded by using information
from the ASTER data and 2005 aerial photography. To achieve
an approximate absolute orientation, at least six ‘artificial’
GCPs were extracted from the present-day ortho-imagery and
DEMs, at the location of distinct surface features. After
achieving a successful absolute orientation solution, TIN-based
DEMs were extracted from the historical imagery. Point
extraction on a grid basis was avoided because oversampling
can occur in areas with low image contrast such as snow
covered areas (Fox, 1997). Depending on the reference data
accuracy, the RMS,,, values of the absolute orientation of the
historical DEMs were within + 30 m when initiated by ASTER
and + 10 m for the modern aerial imagery.
ASTER Image {Band 3H & 3B} USGS Archive Stereo Imagery
Y 1
Reconstruction of interior
DEM and Oftho-image Generation Si SEO aou oon
i Y
Extraction of "artificia GCPs initial absolute orientation
from ASTER With extracted ASTER GCPs
Relative ASTER DEM Historical DEM
Surface Matching
DEM Differencing
Figure 3: Workflow of DEM production and assessment using
ASTER imagery and historical archive stereo-photography
without ground control.
However, despite the successful extraction of DEMS, the initial
fit between the historic and the modern DEMs is too poor for
the precise measurement of volumetric glacier change. This is
because it is only possible to extract a minimal number of
artificial GCPs from the present-day ASTER/aerial imagery,
and the accuracy of the extracted points is relatively low in
comparison to that which would be achieved through field
survey. Consequently, the absolute orientation of the archival
imagery using these points results in only an approximate
alignment of the datasets. To overcome this limitation and to
correct for remaining offsets between the DEMs a robust
surface matching technique was applied. The complete
workflow is summarised in Figure 3.
4. SURFACE MATCHING
Surface matching has become an increasingly relevant
technique in geoscience applications in recent years. A growing
number of multi-temporal, multi-scale and multi-sensor datasets
are now available for environmental studies concerned with
assessing surface changes over long time periods (Mills et al.,