Roeland de Kok 
  
In this case study 4 layers are defined: 
A GIS analysis layer. 
A Remote sensing layer based upon image objects derived from the NDVI and Panchromatic image 
A Layer with Inventory points derived from the forest GIS. 
A layer with sub-objects of similar aspect, a DTM derivative. 
B uw mM: 
  
RS RS+GIS 
  
As subalpine Spruce forests 
>> mountainous Spruce forests 
safle Spruce/Fir-forests 
     
  
  
Coniferous - old 
Coniferous- Coniferous - open 
stand Coniferous - young i Fir/Spruce-forests 
Pine-forests 
Pinus-mugo Pinus-mugo 
high-mountainous Broadleaved forests (Acer ps., Sorbus auc.) 
Beech forests 
Hardwood forests (Acer ps., Ulmus g., Fraxinus ex.) 
Broadleaved - old 
Broadleaved- Broadleaved - open 
stand Broadleaved - young 
  
   
Mixed-stand I (Picea a./Abies a./Acer ps/Sorbus auc.) 
im 
n». Mixed-stand II (Fagus s/Abies a/Picea a.) 
Mixed - stand 
Alpine pasture 
Non-forest vegetation | Meadow high-NDVkwsæiÿ}> Non-forest vegetation 
Meadow low NDVI 
Non-vegetation areas “mA Non-vegetation areas 
  
  
  
  
  
  
Figure 1, Reclassified remote sensing layer and GIS. The latter using GIS info like height, exposition, forest 
management map, stand info etc. 
Using this particular version, the strategy of modeling is focused upon image objects with homogenous forest cover in 
layer 2. The sequence of segmentation is layers 2,1,3,4. In layer 2, tolerance parameters are set in such a way, that small 
gaps around 0.5 Hectare in the forest stands are registered as the smallest single objects. For multi level segmentation, 
the Panchromatic band receives a weight value of 1, the NDVI of 0.5 and Spot 2, 3 and 4 a weight factor of 0.2. It is 
important in this study, that weight and tolerance factors are set according to objects of interest and adapted through 
interactive experiments. The intention is to use layer 2 as a classification layer for pure spectral and textural analysis. 
The contents of layer 1 is segmented with the same parameters as layer 2. Therefore it is based on identical objects with 
equal size as layer 2. This layer 1 however will we used to analyze GIS attributes and relies on the classified results of 
layer 2 (see figure 1) Layer 3 contains inventory points from the forest GIS. They are used in a further GIS analysis. 
They should be placed in a separate layer to prevent data corruption. Layer 4 contains aspect. This layer should respect 
the outer boundaries of the image objects from layer 2 and should not corrupt point data from layer 3. As features 
within multi-spectral data and panchromatic data are direct related to the forest stands, the aspect should not disturb the 
shape and size of these image objects, because in the further GIS analysis, there is a need to work with absolute aspect 
values, therefore this separate layer is required. For other GIS information such as height, the absolute value is of minor 
importance as for this attribute the range is required. 
4.2 Object classification: A selected object within it's sub-class. 
The remote sensing part is fully concentrated on the 2nd layer and uses spectral and textural attributes. There is a need 
for training area's to define the decision curves. The class hierarchy is listed in figure 1. 
The panchromatic band is highly correlated with the red and green band from SPOT. The considerations of including 
the mean value of the panchromatic band should be made clear beforehand as the use of the panchromatic brightness is 
ambivalent in many image scenes. In imagery with simple classes (forest, non-forest) it could be useful from a practical 
point of view. In alpine environment, terrain induced illumination has a huge impact on panchromatic brightness values 
and therefore this study only uses derivatives from the panchromatic band such as standard deviation per object as a 
textural feature and a combination with relative brightness of an object towards the entire scene. For certain classes, the 
textural features are very important. The standard deviation per object of the panchromatic band is such a textural 
measurement. For a typical example, figure 2 shows the separation of coniferous-old versus coniferous-open in the 
feature’ standard deviation of the panchromatic band’. The curve for coniferous-open (on the right) is edited in favor of 
coniferous-old (on the left). 
  
226 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.