je of the laser
neasured by a
n thresholding
he increase in
(scattering en-
lationship be-
ye, and the re-
libration. The
the rotational
shelf mounted
nit. Data from
rplane renders
nonitoring the
nematic GPS
ual frequency
r located over
ceiver on the
are integrated
one suggested
veen the laser
collected over
The following
g: geographic
aser footprint
ch and roll of
1g, and to the
Jifferent tech-
)ent accuracy,
aging airports
| on a truck.
usly surveyed
towed behind
veling;
ng points” of
an be reliably
acy of 20 cm,
n hundred km
y manageable.
approximately
atial distribu-
ges, and small
sed by the re-
g or blowing
cient thinning
. The recom-
eo Transformation of the ellipsoidal coordinates into a suit-
able projection system
e Blunder removal
e Data thinning.
Ice sheet surfaces are typically very smooth and they can be
approximated by small planar patches. A patch size of 25 m was
selected based on the a priori analysis of the surfaces and the
scanning geometry. Because the largest distance between con-
secutive ellipses of the scan pattern is approximately 20 m (in
1991, much less later), every grid cell contains at least one scan
line. After transforming the original ATM data into a suitable
projection system, all points within a 25 meter grid cell are used
to determine a best fitting plane by least squares. Blunders are
removed and the elevation of the tilted plane at the cell's cen-
ter is used as a representative point.
Using the centroid of the grid cell as a representative point (first
order approximation) also works well. The standard deviation,
O, obtained in this procedure comprises two errors: observa-
tion error of the ATM data points and approximation error. It
follows that larger G values are to be expected on heavily cre-
vassed areas, on rocks, on the calving front of glaciers, or over
fjords with floating icebergs. Data collected during bad weather
(ice fog, blowing snow, etc.) also are characterized by higher o
values. Examples shown in (Csatho et al., 1996) confirm the
remarkably high accuracy of the ATM system. The standard
deviation computed during the data thinning process is consis-
tently below 0.15 m in smooth areas.
Greenland is quite extensively covered by ATM. The raw point
elevation data are thinned and stored in the Greenland Airborne
Precise Elevation Survey (GRAPES) data base. The data are
available on anonymous ftp on "gdglas.gsfc.nasa.gov".
4. APPLICATION OF ATM DATA FOR MAPPING
The ATM data used in this study were acquired in Greenland
along the so called ERS-1 line in September 1991 (Thomas et
al., 1994).
4. 1 Large Scale Mapping
Contour maps were generated over small areas to demonstrate
the inherent potential of the ATM system for feature extraction.
In order to preserve the details, the TIN model of the irregularly
distributed original data points was contoured without interpo-
lation. In Figure 3 the perspective view of such a contour map
shows a gently sloping surface. The elongated, low ridges lo-
cated perpendicularly to the prevailing wind direction are most
likely sastrugi.
4.2 Digital Elevation Model (DEM) Generation from ATM
Data
DEMs are particularly suitable for further analysis of the sur-
face. By the application of well-known techniques, such as
Fourier transform, filtering, and scale-space theory, interesting
features can be identified and delineated. Some experiments
towards the automatic detection of lakes and other features
from ATM data are presented in (Csatho et al., 1995a). DEMs
also facilitate the comparison between the laser altimetry data
and other elevation data sets, such as those derived from aerial
45
photogrammetry, Synthetic Apperture Radar (SAR) interfer-
ometry, and satellite radar and laser altimeters.
Contour interval: 0.1 m
"862
8%; tT
78e
== e22 e
gs.
Prevailing wind direction
Figure 3: Perspective view of a large scale contour map gener-
ated from ATM data (swath width: 200 m).
DEMs were generated from the raw ATM using the following
procedure:
e Coordinate transformation.
e Data thinning and blunder detection.
e Creation of TIN (Triangulated Irregular Network) model.
e Interactive editing and smoothing of the TIN model
(optional).
e Interpolation of the TIN model.
In the examples shown in this paper the geographical coordi-
nates were transformed into Universal Transverse Mercator
(UTM) projection system. Blunders were removed by interac-
tive editing or by applying the procedure described in Section
2.4.. The TIN models were created by Delauney triangulation
and optionally smoothed by least squares. A neighborhood-
based planar interpolation was used to create the DEMs and
contour maps. The contour lines of the final contour maps were
smoothed by a weighted average method, and drawn by B-
splines. The data processing was performed on an Intergraph
workstation using the MGE (Modular GIS Environment) Mod-
eler software package from Intergraph Inc.
4.3 An Example - Jakobshavns Drainage Basin, West
Greenland
To obtain a DEM in the Jakobshavns drainage basin in West
Greenland, five parallel ATM swaths were bridged together.
The 200 m wide flight swaths are 1 km apart from each other.
The elevations in the 800 m wide gaps were found by interpo-
lation.
The velocity of the Jakobshavns glacier, which reaches 8
km/year at the floating terminus, is the highest recorded veloc-
ity of any non-surging glacier (e.g. Echelmeyer et al., 1991).
Within the Jakobshavns drainage basin, there is considerable
surface melting in summer below altitudes of about 1400 m.
Fast-flowing melt streams cut into the ice, and large lakes of
meltwater are formed in depressions. The lakes drain periodi-
cally through moulins at the bottom or through rivers and
streams on the ice sheet surface (Thomsen et al., 1989). An
abundance of the surface features and extensive knowledge
acquired during years of intensive research makes the Jakob-
shavns drainage basin an excellent test field. The DEMs and
contour maps derived from laser altimetry are described in de-
tail in (Csatho et al., 1996).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996
