Urban extent 
of a COC 
Querying scenes overlapping on urban extent —— 
  
  
Test necessity by decrement order of 
percentage of cloud contamination 
  
  
~ Steps to test necessity of a scene 
| 
  
+ 
_ Testing Scene À X 
  
  
  
Scenes overlapping on uran extent 
  
   
| Scene À is assigned to be unecessary 
X because complete coverage is x 
assured without Scene A. — — ^ 
+ 
_ Testing Scene B 
   
       
/ 
/ 
"Scene Bis asslgned to be necessary 
. because the urban extent is partly / 
"..dropedfromthe coverage. |... 
  
  
Test all scenes overlap- Order scenes with 
ing on the urban i 7» less cloud to be 
ping upper layer 
  
Figure 1. Steps to test necessity of scenes overlapping on urban extent of a crowd of cities. 
In some intensively populated region, the coordinates of the 
cities are mutually so close that those would be within an extent 
of a satellite images. Such cities are often continuously 
connected as a cluster of urban area. However, clusters larger 
than swath of satellite images could be divided if the catalogues 
are constructed by cities which are separated into different 
scenes. Therefore, we defined clusters of urban area from GSP. 
We assembled the coordinates of the cities close to each other 
into a COC. The distance was set to 30 km considering the 60- 
km swath widths of the ASTER/VNIR. Finally, clusters of 
urban area were constructed with the extent of 30-km buffer 
from the assembled point coordinates of the COC. 
2.2 Determining combination of scenes 
To determine a combination of satellite images, manual 
selection with human decision is often required because the 
satellite images are not assured to be aligned regularly as tiles 
and have cloud contaminations which are only identified by 
percentiles. 
To simplify selection procedure of satellite images, we assumed 
that merging satellite images with zero cloud contamination and 
little cloud will achieve a combination free of cloud or less 
cloud. That assumption indicates that choosing the least cloud- 
contaminated satellite images would provide the least cloud- 
contaminated combination. Once picking up the least cloud- 
contaminated scenes, we may remove redundant scenes which 
are overlapping with each other so that necessary scenes are 
preserved for completing coverage of urban extent of a city of 
interest. 
With that assumption and strategy, we proposed an algorithm to 
determine the least cloud-contaminated combination using 
metadata database of satellite images with percentile of cloud 
contamination (Figure 1). 
l. Order scenes overlapping with urban extent of a city 
by percentage of cloud contamination in decrement. 
2. Test the necessity of the first scene for keeping 
complete coverage on the urban extent. If the urban extent 
is still covered without the scene, the scene is assigned 
unnecessary, and vice versa. 
3. Test the necessity of all the scenes. The scenes 
assigned to be necessary are to be included in the 
combination with the least contaminated with cloud. 
4. Order the scenes included in the combination by 
percentage of cloud contamination incrementally. 
3. RESULT AND DISCUSSION 
For the experiment, we used the 3734 data points of the subset 
with more than 0.1 million from the GSP. By the 30-km rule, 
the 3734 data points were assembled into 2214 COCs. The 
combinations of the scenes for the COCs were determined by 
the procedure of checking necessity. To reject apparently 
unusable scenes, we set constrain on cloud contamination less 
than 20%. We also considered incidental failure of SWIR 
sensor, which is used for assessing cloud contamination, from 1 
April in 2008, and set search period from the beginning to 31 
March in 2008. As a result, for 1951 of the 2214 COCs, the 
combinations of the 11802 scenes were successfully assigned. 
For the other COCs, there was not any scene with condition of 
cloud contamination less than 20% over the search period. In 
this experiment, we rejected the 372 cities of the 263 COCs 
with incomplete coverage of ASTER/VNIR. 
Figure 2 shows the examples of the results of merging the 
scenes with the combination assigned to be least-cloud 
contaminated. For 1340 of the 1951 COCs, the mosaic of 
ASTER/VNIR images were successfully organized; however, 
for the other 611 COCs, the combinations were incomplete to 
cover the extent of the COC or considerably contaminated with 
cloud. For the 611 COCs, we manually selected ASTER/VNIR 
images and ordered them (Figure 3). By the manual procedure, 
the visual appearances of the merged images were improved; 
    
   
     
     
   
   
   
   
    
    
     
    
     
     
   
    
    
   
   
    
   
    
  
    
   
    
   
    
   
   
    
   
Fig 
howe 
haze. 
In the 
main 
image 
cloud 
reject 
assess 
2010) 
cloud: 
Pixel- 
reject 
urban 
woulc 
cities 
In th 
catalo 
propo 
best c 
exper: