GER 
ons. 
the 
nts. 
1rce 
Ver 
aSiC 
ing 
'ost 
ant 
itly 
the 
10d 
ind 
his 
ata 
ras 
)ot 
im 
he 
he 
ng 
Bacchi, Gustavo Sanches 
  
2. Geographic and geometric compatibilization 
This compatibilization was done through the digital registration procedure, which in the first phase was applied only for 
the SPOT image. In this phase, topographic maps were used to extract the geographical coordinates, also recognizable 
in the digital image. These coordinates were used as referential points for geometric corrections. The registered SPOT 
image was used in the digital aerial photo registration through the *image to image" register mode. Submitted to this 
procedure, the digital aerial photo resultant had the same spatial resolution of the SPOT image, which reduced the 
details of the original aerial photo. These procedures made both remote sensing data compatible because they had the 
same size and the same georeference system. It is important to note that after this compatibilization, both remote 
sensing data were sampled in order to distinguish different land cover classes. These classes were preconceived from a 
classification system with four land cover classes: “tropical rainforest (Ca)”, “degraded tropical rainforest (Cb)", 
“herbaceous (Ps)", and “grass (Pa)” (ANDERSON, 1979). The first and second classes are two types of forest. The 
third and the fourth are two types of grass. In this way, it was possible to establish a pre-classification of the Ibaté- 
Mirim watershed land cover map based on those samples, which is shown in Table 1. 
Table 1: Gray level intervals found in samples of both remote sensing source data. 
  
Aerial photograph SPOT image 
Classes | Min. Max. Mean S.D. CV | Min. Max. Mean SD. CV. 
Ca 29 67 43,26 7,64 17,66 15 65 35,32 7,97 22,56 
Cb 61 121 83,37 15,92 18,65 37 77 61,80 4,59 7,43 
Ps 73 150 102,09 16,58 16,24 65 90 74,37 5,08 6,83 
Pa 148 211 176,97. 15,66 . 885 88 140 2108,68 11,96 11,00 
Mean - - - 13,87. 15,33 - - ^ 7,4 11,95 
S.D. = standard deviation, C.V. = coefficient of variation. 
  
  
  
  
  
  
  
  
  
  
  
Note through the dispersion measures, standard deviation and variation coefficient, that the variability of the gray level 
values of the aerial photo are higher than those of the SPOT image. This was clearly noticed in all the land cover classes 
with the exception of the “Ca” class. This variability is related to the mapping effort or with the class limits delineation 
when producing the land cover map from a digital image. 
23 The “true” land cover map 
The same vertical aerial photographs were used to produce the true land cover map of the same watershed. For that, the 
Kartoflex-Zeiss stereoscopic device was employed in order to rectify the pair of stereo photos. The land cover map 
obtained by this process was used to analyze the accuracy of both pre-classification data described above. It is important 
to notice that the map scale adopted for the “true” land map cover was the 1:25000. 
3 THE ACCURACY ANALYSIS 
Both remote sensing data sources pre-classifications were compared with the true land cover map through the 
misclassification matrix method (GOODCHILD & KEMP, 1991; JENSEN, 1986). Using this method it was possible to 
evaluate the accuracy level of each remote sensing source through the omission and comission error indexes. The 
accuracy coefficient and the kappa index were calculated from the misclassification matrix too. It is important to 
consider that each land cover class described previously was individually compared with the true land cover map. This 
accuracy analysis was facilitated by the Envi software. 
4 RESULTS 
The results are presented in the Table 2. These results are the arithmetic mean calculated from each land cover class 
misclassification matrix. In this way, these results summarize four matrixes for each land cover class from each remote 
sensing data source. Observing the results, it is possible to notice that the digital aerial photograph presents a higher 
variability of the gray level values in comparison with the SPOT image. This consideration is based on the accuracy 
coefficient: 91.11 % for the SPOT image and 88.62 % for the aerial photograph, and on the kappa index: 0.62 for the 
SPOT image and 0.57 for the aerial photo. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 99