In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
combination of stereo pairs, data acquired outside the full 
performance range (15 to 60 degrees) were also used in the 
project. 
The following acquisition scenarios were used for testing: 
• Acquisition with StripMap mode, single polarized 
data (HH) 
• Acquisitions in both orbit directions (in order to avoid 
layover and shadow effects where no stereo matching 
is possible) 
• Acquisition at incidence angles of -25°, -35°, ~45°, 
~58° in ascending orbit direction 
• Acquisition at incidence angles of-29°, -45°, -56° in 
descending orbit direction 
• Two acquisition campaigns: one in July / August 
2009 (in parallel to the field campaign), a second one 
in October 2009 
With help of the different acquisition scenarios, stereo pairs 
with different disparity ranges were composed and used for 
digital surface model (DSM) calculation by the automated 
radargrammetry processor integrated into Infoterra’s production 
infrastructure. 
2.3 DEM Evaluation 
During the development phase of the TerraS AR-X 
ELEVATION product, Infoterra followed a strict validation 
approach, which was also applied to the results of this 
development project. 
The evaluation was performed on the results of the different test 
scenarios: 
• Verification based on the raw DSM product for each 
orbit direction, i.e. DSM product without any 
filtering, interpolation of smaller gaps or filling of 
larger gaps with an external DSM source 
• Verification of the raw DSM merged from both orbit 
directions, i.e. no filtering, but gaps are reduced due 
to availability of height information from the alternate 
orbit directions. 
• Verification of the edited DSM, i.e. TerraSAR-X 
ELEVATION DSM product, which is produced with 
the best suited acquisition scenario. It includes outlier 
removal, filtering, interpolation of smaller gaps, 
filling of larger gaps with an external DSM source 
and edited water bodies [1]. 
• Verification of the edited and calibrated DSM. 
2.3.1 Verification methods: The following verification 
methods are applied to the data: 
Visual inspection 
Visual inspection is performed on a shaded relief representation 
of the DSM. This step helps to identify structural irregularities 
in the data processing, deviations in comparison to other DSM 
datasets, systematic artifacts, and outliers inside the elevation 
model. 
Additionally a linear profile plot with the available DEM 
sources is drawn and visually analyzed [3]. A regular shift and 
irregular undulations in the DEM can easily be identified with 
this method. 
Statistical analysis 
In addition to the visual inspection of a DEM, the statistical 
analysis is the most important step of the validation process. 
The statistical calculations are based on a 90 % linear error 
(LE90) for the vertical accuracy [4]. In this project, input to the 
statistical calculations was the DGPS measurements acquired 
during a campaign in July and August 2009. A total of 739 
points was available. 
For point based data like DGPS measurements a difference 
between the DSM and the height values from the reference data 
is calculated. Generally, all reference points are taken into 
account for statistical analysis independent of slope and sensor 
dependency. No selection of reference points according to 
selection criteria was carried out and only data of inconsistency 
is excluded from the process. 
In the standard DEM evaluation procedure a classification of 
different slope and land cover classes is accomplished if a large 
number of points with a regular distribution over the entire area 
are available. In case of the Juneau Icefield, all available 
reference points were acquired over the glacier, thus falling into 
the same slope and land cover class. Consequently, no 
differentiation of classes was possible. 
Figure 1. TerraSAR-X StripMap images over the Juneau 
Icefield: A: Acquired in July 2009, B: Acquired in October 
2009 
2.3.2 DEM evaluation results: During the visual inspection 
of the input scenes it was noted that the backscatter of the areas 
covered by snow and ice was very low for the acquisition 
performed in the summer season (July / August) due to the 
warm weather conditions and wet snow and ice (see Figure 1, 
A). Therefore, it was assumed that the DSM produced with 
these scenes might have some quality deficiencies in 
comparison to the DSM calculated on basis on the scenes 
acquired in autumn (October) (see Figure 1, B). The visual 
inspection of the DSM confirmed these assumptions. The DSM 
calculated with the scenes acquired during the summer season 
show more noise whereas the results received from the autumn 
scenes looks homogenous (see Figure 2).