EDC conducted computing of NDVI before 
rectification, i.e. before changes of pixels original 
positions on images according to specific 
geographic projection and scale. For that reason 
we used their values which are stored as "6th 
channel’ in source data. NDVI values were scaled 
to —1.0 to +1.0 range, with discretization step of 
0.01. Computed NDVI was of great value for 
elimination of composites with unperceptible cloud 
cover over target areas (very low NDVI values 
occurred). 
Geometric registration with Interrupted Goode 
Homolosine projection [3] was conducted at EDC 
  
Figure 2 An example of spring composite, 11-20 
April 1992, first leafing. 
  
752 
  
  
  
Figure 4 An example of autumn composite 21-30 
October 1992. Yellowing of leaves. 
too. We did additional georeferencing, especially in 
the vicinity of observed woods. Chosen woods 
were represented with samples of 14 pixels 
(Spaëva) and 9 pixels (Cesma, llovski dol, Opeke, 
Srnolov). Samples were picked from the center of 
the woods to avoid point spread influences of 
pixels which lie outside of the wood boundaries. 
For this approach we chose distinct woods in the 
lowlands of Croatia in order to avoid additional 
ortocorrecting. Pictures 2 to 5 show changes on 
the 10-day AVHRR composites (first two channels 
according to key RGB=221 are used only) for 
typical scenes of four seasons during observed 
period. In colour, differences are more distinctive. 
Figure 3 An example of summer composite, 21-30 
July 1992. The greatest values of NDVI. 
  
  
Figure 5 An example of winter composite 11-20 
February 1993. Mostly under cloud cover. 
  
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
  
  
  
  
gri 
Ta 
llo 
Sr 
fron 
dev 
obs 
loce 
of ^ 
yell