International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
  
  
Figure 3. Sample points imposed on a sampled image. 
3.2 Bidirectional dataset 
As the end product, a large amount of directionally defined data 
was generated. A single entry in a dataset consists of the 
following data: sensor id (1-9), camera and sample point 
locations, slope, aspect, sampled intensity, sensor azimuth, 
nominal measuring angle, sun azimuth, sun zenith angle, 
measuring azimuth, measuring zenith angle, relative azimuth 
(sun-camera), brdf azimuth, brdf zenith angle, sun azimuth and 
sun zenith angle relative to surface normal, measuring azimuth 
and measuring zenith angle relative to surface normal, relative 
azimuth using the inclined surface and finally the primary and 
secondary classes according to the corresponding 
classifications. 
The total amount of data was 4.3Gbytes of ASCII files named 
according to date, the sensor bandwidth, and the flight strip 
number. The wavelength range of the data extends from 395 to 
1015 nanometres and the radiometric resolution of the data is 8 
bits. 
  
a b 
Figure 4. Bidirectional effect for (a) pine and (b) spruce copse 
using the HRSC-A based data set. Surface normalised sun 
zenith in (a) 90° and in (b) 60° (blue circle). 
  
Figure 5. Grey level variation of spruce, birch and pine as 
function of viewing angle. Sun zenith angle 50 degrees. 
In figures 4 and 5, some of the sampled target reflectance 
anisotropy are presented. In these presentations the surface 
compensated BRDF is used, an eight bit image DN as a 
anisotropy unit. Presentations show potential of providing 
additional information for tree species determination. 
3.3 Analysis of accuracy 
Because of the short image acquisition period of a single strip, 
the sun movements during the image acquisition can considered 
to be constant. Due to this assumption and the distance from the 
feasibility point used in sun angle calculations the maximum 
azimuth error was estimated to be 0.30° and the zenith angle 
error to be 0.06° at the both ends of an image strip. For the 
selected intensive areas errors are obviously smaller. 
The accuracy of the viewing direction is a function of the 
accuracy of the attitude data of the sensor. Sensor attitude was 
systematically corrected, based on empirical measures, on an 
average of 0.80° for pitch and 0.20° for yaw, most likely due to 
the map projection. The need could not be specified exactly, 
however, and must thus be included as a source of uncertainty 
in the data. The linear densification of the positioning and 
attitude data adds some minor uncertainty to the viewing 
direction. This is considered to lie within acceptable limits 
relative to the imaginable range of applications using the data. 
The surface normal direction is more problematic. The angular 
behaviour of the calculated slope and aspect layers were studied 
using surface normal vectors for the two sequential digital 
surface models. Surface normal vectors were constructed at 
each point and the angular difference was calculated. In Figure 
6, this difference is represented by means of a raster image. The 
calculated mean difference was 9.2? and the standard deviation 
15.1?. At maximum, the obtained angular difference was 164.8? 
for some forest pixels. 
  
Figure 6. Surface normal uncertainty. 
The surface normal uncertainty is mainly caused by the 
smoothness of the terrain models, e.g. the model fills in small 
gaps between trees and other structures. The edges between 
forests and low vegetation seems also to be problematic. 
The angular variability was most prominent in forested areas, 
while for farmlands and water areas it was much smaller. This is 
due both to mutual occlusion and shadowing of trees, which 
affect the DSM processing, and to the automatic-image-based 
DSM creation itself. The spatial and height resolutions of the 
DSM also play a role in this behaviour. 
4. CONCLUSIONS 
In this study, multi angular HRSC-A images and digital surface 
models were used in examination of bidirectional reflectance 
from multi angular high-resolution HRSC-A images. Method 
uses high-resolution DSM to approximate the orientation of the 
    
    
  
  
  
    
    
  
  
   
    
   
   
   
     
   
   
    
   
    
    
    
     
     
     
   
   
    
   
  
    
    
    
   
   
   
  
   
   
    
   
   
   
   
    
Inter 
— 
surfa 
utiliz 
relat 
inve: 
natu 
imag 
geon 
Dire 
in € 
estal 
First 
in th 
angt 
this 
obje 
stud; 
from 
com 
Date 
data 
pres 
each 
view 
seen 
relia 
The 
digi 
may 
surf 
(Hy 
This 
BRI 
heig 
ima; 
Arn 
Sou 
atm 
Inte 
Bei: