tanbul 2004 
> Principal 
ny) and is 
001 pixels, 
001 x 1001 
m 
ls, spacing: 
401 pixels, 
t site have 
odesy and 
'echnology, 
grammetry 
, Germany) 
. Aplicada, 
eodesy and 
1 Agency — 
ssia) 
res and K. 
their work 
their paper 
analyzed as 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
  
  
  
  
North South 
P. Reinartz Global -spacing = 50m | Global -spacing = 50m 
D. Poli Global -spacing = 25m 
K. Jacobsen Global- spacing = 15m 
  
R. Kaczynski | Global -spacing = 20m | 4 areas — spacing = Sm 
2 areas— spacing = 25m 
  
  
  
A. Suchkov Global -spacing = 20m 
Table 25: Analyzed DEM over Bavaria. 
  
  
In south area, D. Poli produced two DEM with two different 
kind of modelisation (polil: orientation with rigorous sensor 
model, poli2: orientation with rational polynomial functions) 
(Poli Daniela, 2004, SPOT-5/HRS stereo images orientation 
and automated DSM generation; ISPRS Congress Istanbul 
2004). 
As for the Barcelona site, the providers chose different 
sampling so all the results are going to be given with the 
finest sampling between received DEM and reference. In the 
same way, we were not able to compare the five of them in 
the same time because they have not produced all the same 
areas. Then we compare the results by area (North and South) 
also because references are very different between those two 
areas. 
3.3 Statistical results 
All the following results are given as for the Barcelona site 
for 98% of dots (the invalid values in received DEM and/or 
reference, like zero, have been taking off statistics), all the 
differences have been computed in the same way : 
received DEM - Reference. 
3.4 Results on North Area 
3.4.4 The Reference N 50 (Fig 26) is given with a 50 m 
spacing. Its size is 30 km x 50 km and the altitude is between 
360 and 570 m. 
  
    
e m t 
Fig 26: Reference N 50 
  
We can mention that the reference spacing is coarser than the 
received DEM spacing (R. Kaczynski - 20m and A. Suckov - 
20m) but the area is quite flat and the accuracy is about 2m, 
so it is still interesting to calculate statistics on the difference. 
min max mean | Standard dev. 
P. Reinartz -3 28 7,8 5.6 
R. Ka | -13 26 204,0 7,0 
A. Suchkov -9 9 0,0 3,8 
Table 27: Overall results on North Area 
  
Min and max values are relative values for R. Kaczynski 
results. 
3.4.3 P. Reinartz DEM analysis is based on statistics and 
visualization of the difference between the produced and 
reference DEMs (Fig 28) and on some profile (Fig 29) 
EM —ref N 50 
    
Fig 28 Difference: P. Reinartz D 
The visualization of the difference image confirm the 
altimetric bias between P. Reinartz DEM and the reference 
(nearly 8 m) and some local variations corresponding to the 
relief shape but this is slight. In fact, this DTM is very 
smooth with very few artefacts, we can suppose that some 
filters have been used after the matching processing. 
  
Fig 29: Above: Profile in P. Reinartz DEM (purple line) 
Below: blue 7 P. Reinartz DEM; black ^ ref N. 50 
370« reference height<470 height exaggeration = 4 
3.4.4 R. Kaczynski DEM analysis has shown an important 
bias of 203 m or more which has not been explained yet. To 
be able to see local defaults this bias has been taken off (Fig. 
  
1 
@ 10 
&o 
e -10 
fs @ -100 
3 = 
des CIS add 7 
ki DEM - ref N 50 
To be able to see local defaults this bias has been taken off. 
This bias is incomprehensible and not noticed in R. 
Kaczynski report, and we could not find any convincing