International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
  
  
    
    
     
   
    
   
    
   
    
   
    
    
    
   
    
      
   
  
   
   
   
   
   
   
    
     
    
   
   
   
   
   
   
   
   
   
    
  
  
   
   
  
  
  
  
  
  
  
    
  
  
  
  
  
  
  
  
  
  
  
  
Pt ID DEM Residual Z for 
heights(m) DEM (m) 
98 46.0 5.7 
2011 224.6 -3.2 
3043 458.1 0.9 
3048 291.3 1.2 
3053 246.0 1.4 
31441 23.0 4.2 
31432 73.0 1.] 
31433 324.0 4.2 
32433 492.0 1.9 
32443 501.0 5.3 
  
  
  
  
  
Table 7. ICP accuracy in the DEM covering IGN DEM area 
Having in mind that the covered area is mainly mountains, 
from these initial results the following points could be 
extracted: 
e The correlation quality of the dem is 91% better than 
the 85% that it is expected in mountains areas (Valorge, 
2003). 
e The accuracy of the ICPs also meets the expectations. 
e However the RMSE of the height difference between 
the reference DEM and the LPS DEM of 16m, is worse 
than it is expected. The expected value for the RMSE in 
mountains areas is better than 10m (Valorge, 2003). 
The above conclusions indicate that there is correspondence 
between the height difference and the slope of the terrain. The 
correlation between the acquisition characteristics of HRS 
images and the slope should be examined in more detail. An 
examination of the errors was carried out and it was found 
that: 
e There is a correlation between the height difference and 
the slope. 
e In some areas the slope is bigger than 35°. However as it 
is shown in the height error can be better than 10m. In 
other areas where the slope is smaller than 35°, the height 
error in that area is bigger than 20m. 
e The most interesting area is the Pitkin area (Figure 2 — 
blue borders) where the slope is almost perpendicular to 
the flight direction and parallel to the acquisition 
direction. In this area the height difference is very high. 
This area will be examined deeply in the section 5.3.4. 
e Finally, it seems that the height difference not only 
depends on the slope itself but also to the direction of the 
slope. It seems that is main reason for the loss of the 
accuracy is the steepness of the along track slope. This is 
expected and it will be discussed in more detail in the 
following sections. 
Another test was done covering the area with the green 
borders in image 1 (DEM2). In this area although the 
maximum slope is 40°, the along track slope is smaller than 
the 20°. The accuracy of the DEM2 covering an area of 
695.97 sq. km is described in table 8. For the image matching 
process exactly the same strategy parameters were used as the 
DEMI. From table 8 it is obvious that the RMSE which is 
about 8 meters, is better than the expected value of the 10m. 
At this point it is also very helpful to introduce for this DEM2 
a detailed height difference distribution in order to realize the 
improvement of the total accuracy. In table 9 this detail height 
difference distribution is introduced. 
  
  
  
  
  
  
Min diff -115.81m 
Max diff 111.64m 
Mean diff -0.83m 
RMSE 8.19m 
  
  
Table 8. Accuracy of DEM2 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Height difference Percentage 
Min/-50m 0.039% 
-50m/-20m 0.768% 
-20m/-10m 3.681% 
-10m/-5m 20.415% 
-5m/-2m 26.912965 
-2m/-1m 6.579% 
-Im/1m 11.900945 
1m/2m 5.375% 
2m/5m 10.936% 
Sm/10m 7.181% 
10m/20m 3.989% 
20-50m 2.021% 
50m and above 0.204% 
-10m/10m 89.2989 
  
  
  
  
Table 9. Details height difference distribution of DEM2 
From the above table it is shown that the almost the 90% of 
the points are within 10m, while in the DEMI, 81% of 
height difference are within «10m. Having in mind that the 
same matching parameters were used, it is obvious that the 
relief itself is responsible for this improvement. It is assumed 
that the DEM2 is the more characteristic of our evaluation 
process than the DEMI. 
5.3.4. HRS images comparison in the PITKIN area. In this 
section some examples are given of the distortions created by 
steep slopes in the along track direction on HRS images, 
where the along track slope is from about 28? to 43?. The 
different representation of the cliffs in HRS images is obvious. 
A lot of information is missing in HRS front image which is 
represented in HRS back image. As a result, it seems that it is 
impossible to extract heights correctly in those areas and 
generally in areas with large along track slope. The critical 
value for the along track slope it seems to be about 30°. 
Finally, it seems that with manual editing it is possible to 
extract the borders of the cliffs, but in any case, it is very 
difficult to extract the heights in the cliffs. 
  
Figure 3. Example 1- HRS front image. Slope in cliff 1 
is around 43° along track 
   
Inter: 
  
In t] 
the 
the 
In tl 
This 
and 
Pho 
that 
con