— < 
wu gc 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
associated with collapsed buildings exhibits a higher spectral 
return than the standing structure. Although considerable 
damage was also sustained in western Golcuk, reduced 
differences may be due to suppressed reflectance values where 
smoke from the burring Tupras oil refinery was present in the 
upper atmosphere. Positive differences are limited to the ‘Sunk’ 
coastal stretch, where reflectance values have fallen following 
widespread inundation. 
  
    
LEGEND 
-200 
-100 
  
-50 
-25 
100 
200 
rro 
a) Difference damage 
0.25 
0.50 
0.75 
>0.75 
: TOT HR "ui TOA apn CAA ur 4 
(b) Correlation damage map 
  
     
x 
3 UN 
os 
ge map 
a 
cS. 
(c) Block correlation dama 
  
  
  
  
Figure 4 SPOT damage maps. Areas of high positive difference 
and low correlation correspond with the Golcuk city center, 
which from Figure 3b sustained severe and widespread building 
damage. See text for explanation of annotations C1-C3. 
Results for the block and window-based correlation (Figure 4b 
and Figure 4c) are overlaid with a base map of Golcuk. For 
visualization purposes, all values are displayed as positive, 
since the magnitude rather than the direction of change is of 
interest. For both block and sliding window-based results, areas 
of low correlation (displayed in red) are concentrated in central 
Golcuk (see annotation C1). As for the difference values, the 
damage map in Figure 3b confirms that building collapse was 
widespread. A low level of correlation around the subsidence 
613 
zone (C2) is due to the change in reflectance following 
inundation. Low correlation offshore (C3) is probably 
attributable to the random or chaotic patterns of surface 
reflectance associated with wind-driven wave action. 
Figure 5 depicts the SAR intensity responses for Golcuk. For 
visualization purposes, the difference image in Figure 5a was 
thresholded at -7.0 « dif « 7.0, with intermediate values 
displayed across an 8-bit range using a linear contrast stretch. 
Similarly, correlation images in Figure 5b and Figure 5c were 
thresholded at 0.2 « cor « 0.6. Block correlation statistics were 
further classified into categories of: low (0 « bk cor « 0.2); 
moderate (0.2 « bk cor « 0.4); high (0.4 « bk cor « 0.6); and 
very high (bk cor » 0.6). The coherence image in Figure 5d 
was thresholded at 0.3 « coh « 0.6. 
  
t 
A 
  
0 0.2 04 0.6 20.6 0.3 0 0.6 
(c) Block correlation map (d) Coherence map 
  
  
  
Figure 5 SAR damage maps. Areas of low correlation 
correspond with Golcuk city center, which from Figure 3b 
sustained severe and widespread building damage. See text for 
explanation of annotations C1-C2 
From Figure 5a, the magnitude of SAR intensity difference is 
low compared with its optical counterpart. It is difficult to 
discern any obvious regularity in response, with the urban 
centre exhibiting an amalgam of positive and negative values. 
Low correlation (see annotation Cl in Figure 5c) is evident 
throughout central areas of Golcuk. Once again, this 
corresponds with high levels of building collapse (Figure 3b). 
Low correlation outside the urban area is concentrated around 
Izmit Bay (C2), where changing water surface conditions cause 
differences in backscatter. In Figure 5d, low coherence is 
present throughout both urban and rural areas, suggesting that 
this measure has limited ability to distinguish earthquake 
building damage from other modes of change. 
3.1.3 Damage severity: From the zone-based damage profiles 
in Figure 6, panchromatic imagery yields an encouraging trend 
between difference and damage state (Figure 6a). As the 
percentage of collapsed buildings increases from class A to E,