tanbul 2004 
ig has been 
in square y- 
11m with a 
| (pixel size 
ts have not 
for the y- 
points have 
hole area of 
hern model 
[he spacing 
in the orbit 
| models. A 
oroximately 
also only a 
n. The root 
9m for the 
| of the x- 
cies at the 
(slope) 
5 xtana 
7xtana 
  
2 x tan à 
against the 
ce 
ure 2, only 
e test areas 
km x 5km. 
in accuracy 
ion of full 
iy which is 
he range of 
h is located 
F(slope) 
).6 x tan a 
9xtan a 
  
4 x tan o 
4 x tan o 
  
6xtan à 
  
  
.9 X tan « 
"test areas 
canner 
to a DSM 
> trees and 
ure Radar 
id, but also 
th InSAR is 
s, the large 
> explained. 
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004 
The DSM includes several points not located on the bare 
ground like the reference DEM. This is indicated also by the 
negative bias, which has the sign of the correction (reference 
height — matched height), so a negative sign is indicating a 
location of the matched points on top of the vegetation and 
buildings. A filtering of the DSM to a DEM by RASCOR 
removes between 35% and 55% of the points depending upon 
the type of filtering. The DEM-results in table 2 are based on 
45% removed points. Also the frequency distribution of the 
discrepancies demonstrates the reason of the chosen data 
handling (figures 11 and 12). 
not very probable - the orientation of the HRS-scene has been 
based on trigonometric points from the survey administration, 
which are also base for the topographic maps. Another reason is 
the reduced accuracy of matching inclined objects, which are 
  
  
Ty 
Rteighariholr "2% 
Spy, J ars- Bahnhof 
ier Sper 
EM 
A 
  
    
  
   
iki d CS Fin 
figure 14.1 
matched DSM 
test area Taching 
  
  
  
Figure 10: test area Gars, 
left topographic map right: forest layer 
figure 14.2 
DSM after filtering 
with RASCOR to a 
DEM 
  
  
  
figure 11: 
frequency 
distribution 
of AZ before 
filtering: left 
open areas, 
right forest 
figure 12: 
frequency 
distribution 
of AZ after 
filtering: left 
open areas, 
right forest 
  
  
The frequency distribution of the discrepancies (figures 11 and 
12) also do show the influence of remaining points on top of the 
vegetation in the forest area. Also RASCOR is not able to 
identify points on top of the forest if all neighboured points are 
elevated. Only if at least some points are located on the bare 
ground, the points not belonging to the bare ground can be 
identified and removed. Nevertheless, also for the forest there is 
a clear advantage of the filtering. 
  
  
  
Figure 13: accuracy 
depending upon 
tangent of slope 
test area Gars 
slope 2 
  
  
vertical: rmse AZ 
  
  
  
  
There is a clear dependency of the accuracy from the terrain 
inclination as visible in figure 13. It shows the situation of all 
points after filtering by RASCOR. The dependency upon the 
terrain inclination can be caused by errors in the horizontal 
location between the reference and the HRS-data. This reason is 
  
  
figure 14.3 
reference DEM 
  
  
The filtering of the DSM to a DEM is very important — not only 
in the forest, but also in the open areas. The open areas do 
include settlements and several single trees and bushes, so also 
here a higher percentage of matched points are not located on 
the ground. This cannot be seen only with the standard 
deviation, but also at the bias. The remaining systematic height 
difference of -1.3m for the open areas after filtering with 
program RASCOR, is small but realistic. The filtering of the 
DSM to a DEM requires height differences larger than the 
noise. Small objects like bushes cannot be identified as not 
belonging to the bare ground if they are below the root mean 
discrepancies. On the other hand there are several points located 
in the forest which are on top of closed areas where the ground 
cannot be seen. 
Also a visual comparison of the original DSM with the filtered 
DEM and the reference DEM shows the justification of the 
filtering. The morphologic structure of the filtered DEM is quite 
more similar to the reference DEM like the matched DSM (see 
figure 14). 
5.2 Test area Vilsbiburg 
In the larger test area Vilsbiburg the reference DEM has a 
spacing of 50m and only a limited accuracy of 2m, but this 
should be sufficient. Only 13.7% is covered by forest and the 
forest areas are small. 
The a little larger rmse for the test area Vilsbiburg can be 
explained by the lower accuracy of the reference together with 
the interpolation over 50m. The smaller bias of the forest areas 
after filtering has been caused by the smaller size of the 
individual forest areas.