multivariate data set consisting of intercorrelated 
spectral bands into an uncorrelated data set, with a 
geometrical dimension less than the geometrical 
dimension of the original multispectral image,we 
applied this technique on LANDSAT MSS images, 
and we verified the signal - to - noise ratio theory 
(Santiesteban & Munoz - 1978). 
t 
   
Fig.7. Landsat TM - 1989 image(left), compared to SAR - ERS - 1 
(right) showing the morphological changes of the Sacalin Island. 
4.3. The new image band, the first principal 
component, is used in image enhancement, for 
change and edge detection. For edge detection we 
have used a photogrammetric technique described 
in the following: 
- considering the remote sensing image as a 
"positive image " we compute its " negative " by 
subtracting from an established maximum grey 
level the grey level values of every pixel; 
- adding the "positive image" to this "negative", 
shifted with one line and one column, we obtain the 
edge of the phenomena appearing in the remote 
sensing image. 
4.4 Classification of different turbidity level waters: 
(a) a bitmap has been drawn, by thresholding the 
MSS 7 band in - between 0-4 grey level values and 
TM 5 band in - between 1-16 grey level values. 
  
Fig.8. Classification of the different turbidity level waters on the 
Landsat MSS image of July 1975. 
The best spectral bands to detect the suspended 
sediment concentration are the MSS 4, 5 and TM 1, 
  
2, 3; the MSS 6 is adequate for chlorophil pigment 
detection; (b) an unsupervised classification Was 
performed (ISOCLUS program, together With 
extracting the spectral signature for each class; 
(c)the spectral signatures obtained were used for à 
maximum likelihood classification (MLC program); 
(d) an average filter (FAV) and a modal filter 
(FMO) were used for eliminating the "striping" type 
noise. 
For each of the 2 images classifications (see fig. 4, 
fig.5) the confusion matrices were drawn; the 
maximum confusion values obtained were 87-88% 
- An interpretation of the two classified images can 
be performed having the meteorological conditions 
in the moment of the images acquisition: for the 
Landsat TM image (fig. 4) the N-NE wind direction 
and 10 m/s speed, can explain the narrowing and the 
direction of the suspended sediments plume (lower 
part of fig.4), as well as merging of the waters with 
different turbidity - upper part of fig.4. In the 
Landsat TM classified image, both of these 
suspended sediment plumes were much enlarged 
and the different turbidity level classes very well 
separated, in good correlation with the wind 
direction and speed (W,2 m/s). 
- Atmospheric corrections were performed on the 
TM 6 thermal band, in order to extract temperature 
information (ATCORT O program); the grey levels 
were transformed into grey levels, corresponding to 
temperatures, after the atmospheric correction - see 
fig.6. 
- Isodensity lines were drawn (fig.9) over the 
Landsat MSS, following the algorithm: 
= 2 T 
I TR x 
^ 2 3 > 
he Zn > 
E 
a 5 
  
T Jd 
Fig.9. Isodensity lines over the Landsat MSS image, separating the 
calsses of water with different turbidity levels. 
  
560 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996