04 
  
a 
> 
An 
hite 
  
——À 
A 
).10848 
Tam 
two 
387; 
  
International Archives of the Photogrammetry, Remote S 
Serpentinite (fig. 5). Min value: -0.12; max. value: 
*0.11; mean: 0; st. dev.: 0.014. 
Observing fig. 4a and 5a and considering the given statistical 
information, we would like to say that: 
  
Biotite. Checking the pixel numbers of fig. 4 we 
found out that the pixels « 0 are 1286080 over a 
whole of 3428880. We think to explain this error by 
the topographic features of the image (which caused 
some over lighting problems in the satellite scene). 
The normal distribution of the absolute frequencies 
(fig. 4a), suggests to classify the thematic layer in 
three classes. Medium class is built summing and 
subtracting from the mean value the st. dev. value 
(0.005 + 0.041): high class is made by values 
>=0.046; low class by values <= -0.036. You can see 
the said classification in fig. 6. 
  
alla BE bos -0.046 
  
  
  
  
  
  
  
  
  
media -0.036<bio<0.046 
bassa bio<=-0.036 
Figure 6. Three classification levels of the Biotite 
concentration (see text for more details) 
We think that the high Biotite concentration areas 
could be enough correct while the low Biotite 
concentration areas may be not. We think that some 
pre processing procedures applied to the ASTER 
sensor image could correct the classification of the 
low concentration areas. These are: atmospheric and 
topographic corrections (which should low the errors 
caused by the morphology of the land — shadow areas, 
CEM pixels «0). 
Serpentinite. Similarly to the Biotite case, the high 
number of pixel «0 (1478102 over a whole of 
3428880) is also attributed to the light condition of 
the image (land morphology). 
We classified the image in three classes, using the 
same rules of the previous example  (Biotite). 
Considering the mean and the st. dev. values 
(respectively 0 and -0.014) these are the three 
obtained ranges: low: serp. <= -0014; medium: - 
ensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
0.014<serp.<0.014; high: serp. >= 0.014. You can see 
the said classification in fig. 7. As in the previous 
case, we think that the high Serpentinite concentration 
areas are enough correct, while the low Serpentinite 
concentration areas should be corrected with the said 
pre processing procedures. 
  
  
  
  
  
  
alla serpy =0.014 
media [ 014<serpel. 014 
Dassa serpc 21.014 
  
  
  
  
Figure 7. Three classification levels of the Serpentinite 
concentration (see text for more details). 
4. Results and conclusions 
At last we ran the so called “Matrix” procedure. This procedure 
overlays the thematic layers of fig. 6 and 7 (Biotite and 
Serpentinite concentrations by the CEM algorithm) in order to 
point out all the possible combinations among the classes. Fig. 8 
shows in pink the high Biotite and low Serpenite areas; in violet 
are drawn instead the low Biotite and high Serpentinite areas. 
In order to make a first classification check, we overlaid on the 
pink and the violet areas two vector layers, which were drawn 
on the basis of the PNRA map. The yellow vector layer shows 
the Biotite areas (GHGr), while the green one, the Serpentinite 
areas (GHGa). 
We would like to note that: 
e Especially in the case of the. low mineral 
concentrations, we think that the pink and violet areas 
may be overestimated. This fact may have happened 
because of shadow zones existence in the raw data 
satellite scene. In fact we saw that the CEM algorithm 
tends to consider the shadow zones as mineral low 
concentration ones. 
e In the case of the isolated pixel existence inside the 
icy areas, we think that these pixels may show 
morenic settlements instead of emerged rocks. We 
intend at this aim to refine our classification on the 
basis of this hypothesis. 
1237