for 
n at 
ajor 
otal 
and 
gh 
ng/l 
t be 
asin 
atial 
to a 
WFI 
p in 
0m 
n to 
for 
sion 
M 3 
ask. 
ked 
e to 
r of 
the 
pe), 
the 
Ms 
n of 
e in 
  
Novo, Evlyn 
both red and near infrared region. Black Water, seen as black, is responsible for strong absorption of red and infrared 
radiation due to the small concentration of suspended particles and high concentration of dissolved organic matter 
(Kirk, 1993; Mobley, 1994). In the floodplain lake, one can see a patch of white water indicating the advection of river 
water into the lake as reported by Fisher and Parsley (1979). 
  
30mx30 m 100mx 100m 258mx258m 
   
      
  
  
  
  
  
  
  
  
Figure 1 — Effect of the spatial resolution on the identification of water types. 
It is interesting to observe that the reduced spatial resolution prevents the detection of the smaller channels (paranas) 
flowing between the Amazon River and the floodplain lakes, but does not prevent the detection of river water in the 
lake. The results also suggest that the 100 m x 100 m resolution data preserve most of the information found in the 30 m 
x 30 m resolution image. Therefore, improving the spatial resolution up to 100 m by applying image restoration 
algorithms can improve the application of WFI/CBERS camera for floodplain water type monitoring. 
Figure 2 shows the classification of water types derived from the simulated WFI image. Reaches selected for 
comparison with the 100 m by 100 m resolution image are specified on the figure. 
  
  
     
  
  
s MEC e EUN 
>. Antonio do Ica 
  
  
  
  
VE. 
Ds 
  
Figure 2 — Water Type Classification (Yellow — White Water; Cyan — Clear Water; Dark Grey — Black Water) 
Figure 3 presents the classification results for 100 m and 258 m resolution images at Santo António do Icá, in the upper 
Amazon River basin. In this reach, both the main stem and tributaries are narrower. The average width of the main 
channel near this reach was reported as 2000 m (Mertes, 1985). The secondary channel width is quite variable ranging 
from tenths of meters to hundreds of meters. Some of them can not even be detected at 30 meters resolution because of 
they run beneath a dense forest cover. The classification results at this reach are not affected by scaling up the resolution 
from 100 m to 258 m. 
Figure 4 presents the classification results at a reach at Madeira River. It is clear that at the reach some information is 
lost when scaling to 258 m resolution. This reach has a very complex geomorphology with a mixture of narrow 
channels, scroll bars, levees organized at fine spatial scale. Some white water channels resolved at 100 m resolution are 
lost at 258 m resolution. Figure 5 shows the classification results for the reach at Trombetas River. 
The effect of the resolution on the discrimination and mapping of water types can be better assessed by looking at Table 
2. Three trends are observed in Table 2: 1) Reaches where the spatial resolution does not affect the amount of pixels 
classified as water (Santo Antonio do Iga, Manaus, and Madeira); 2) Reaches where the increase in spatial resolution 
from 100 m to 258 m results in an overestimation of the amount of pixels classified as water; 3) Reaches where the 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1029