OW Ones 
ation and 
in fuzzy 
ly to find 
uring the 
is linked 
ig their 
| the end 
s become 
a. Raster 
sides are 
as these 
he years 
aracterise 
different 
"rule" for 
reas with 
covered 
) degrees 
available 
this new 
rated and 
e human 
lowledge, 
to decide 
| matches 
| the pixel 
to green 
the pixel 
- low) for 
ud IS low 
v AND IF 
d-field IS 
N AND IF 
WRISK is 
ibed by: 
ield, 
features 
order to 
formation 
  
  
This new classification is called multisource classification 
because it considers simultaneously wide-range elements 
in order to determine the final class assignment. 
Multisource classification compares, pixel by pixel, the 12 
bands values to the rules substance and then assigns the 
pixel to a high / medium / low snow possibility class. 
The final result of classification is a snow-risk map 
containing for all the pixels: 
- the snow risk class (high-medium-low), 
- the matching value to rules. 
If no rule describes a situation or the matching degree is 
lower than a threshold value the pixel is not classified. 
  
Since the goal is only the snow layer localisation the rules 
regard only the snow 
- to adjust snow-risk value where there can be snow, 
for direct knowledge and snow membership degree is 
from O to 1, 
to remove snow where there cannot be snow, for direct 
knowledge, and snow membership degree not equal to O . 
Figure 15 shows the final classification result. 
  
Figure 15: final localisation of blanket of snow. White for high snow-risk, light grey for medium risk, dark grey for low risk 
black for non classified areas. 
13 CONCLUSIONS 
The fuzzy classification approach gives very good results 
especially for the mixed pixels treatment in this case, 
while classical approach does not produce good results. 
The multisurce classification improves the already good 
fuzzy result applying the human decision mechanism. 
The final result of the analysis is a map of snow-risk areas 
on Alps region showing in white colour the high snow-risk 
areas. These white areas are the reliable snow covered 
areas that have to be regarded for the water equivalent 
estimation. 
Acknoledgements 
Special thanks to Dott. Anna Rampini and all the ITIM 
staff where the second part of the described analysis were 
performed using UCLA and FIREMEN softwares. 
UCLA and FIREMEN software are home made programs 
made by ITIM laboratory (Istituto Tecnologie Informatiche 
Multimediali) at CNR (Consiglio Nazionale Ricerche) of 
Milan (ENVIRONMENT CEC Project N° EV5V CT94 
0521). 
References 
A. Bellaciucco, “Modelli per il trattamento dei dati 
satellitari", Sistema Terra, Ill, 2, 1994 
E. Binaghi, P. Madella, M.G. Montesano, A. Rampini, 
"Fuzzy contextual classification of multisource remote 
sensing images", IEEE transaction on geoscienze and 
remote sensing, vol. 35 n? 2, march 1997 
R. E. Bellman, L.. A. Zadeh, "Decision-making in fuzzy 
environmental" Fuzzy sets and application, R. R. Yager, 
S. Ovchinnikov, R. M. Tong, H. T. Nguyen, Eds. New 
York: Wiley, 1987 
E. Binaghi, A. Rampini, "Fuzzy decision-making in the 
classification of multisource remote sensing data", Optic. 
Eng, vol. 6, n°32, 1993 
J. A. Richards, D. A. Landgrebe, P. H. Swaoin, "A means 
for utilizing ancillary information in multispectral 
classification”, Remote Sensing Environ. N° 12, 1992 
P. Slater, “Remote sensing”, Addison-Wesley Publishing 
Company, Massachussetts, 1980 
F. Wang: Fuzzy supervised classification of remote 
sensing image”, IEEE Trans. Geo. Remote Sensing, vol. 
28, March 1990 
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 335