The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
1310 
(i) A high proportion of built-up areas and of buildings 
on gentle slopes corresponds to low SV. 
(ii) A high proportion of buildings on medium slopes, 
buildings exposed to landslide hazard, and the 
abundance of two specific roof types, corresponds to 
high SV values. 
(iii) A high amount of infrastructure corresponds to high 
SV, although at low significance. This is somewhat 
surprising and most likely this variable provides a 
correction on the other variables, compensating some 
of the variation in the variables with a higher 
significance, such as buildings at landslide hazard. It 
may also be the case that the amount of service 
infrastructure and lifelines is higher around buildings 
exposed to landslide hazard because of planning 
considerations by the city developers in the past. 
(iv) No significant influence was found for the selected 
texture measures, the amount of buildings at flood 
risk and the distance to infrastructure (e.g. lifelines). 
Legend 
Pansharpened QB intage 
0.6 m ree., from 12/2000 
Legend 
Social Vulnerabitrty 
Score 
ШШ frie - 1.x 
ШШ 1-31 - 1.41 
Figure 3. Scores from the social vulnerability index per 
neighborhood (bottom) with the pansharpened Quickbird image 
as reference (top). 
6. VALIDATION, CONCLUSION, DISCUSSION AND 
OUTLOOK 
on a larger scale (e.g. using data from house-to-house surveys), 
the proxy variables could have been delineated on a smaller and 
more detailed scale as well. To validate the results obtained in 
this study and to quantify their contribution to risk assessment 
better, the validation could be improved by the comparison with 
results from detailed traditional methods, e.g. from house-to- 
house surveys in the same area. Unfortunately, they were not 
available for this research. 
Nevertheless, the potential of the method becomes obvious: 
relevant information, consisting to a large proportion of 
valuable contextual information, can be delineated from remote 
sensing data in combination with other digital data sets (Table 
1). The information that can be extracted ranges from building- 
specific information (location, roof material, etc.) to 
neighbourhood information (neighbourhood composition, living 
standards, etc.). 
In terms of transferability, the delineated proxies can be adapted 
and evaluated to fit different places and different hazards. The 
main precondition of this approach is the initial understanding 
of SV in a particular place, mainly based on the specific local 
conditions and the understanding of those factors. For example, 
if it is not the case that poor people live in areas with 
unfavourable terrain, the consequent evaluation of this proxy 
variable can be changed accordingly or combined with other 
information such as neighbourhood composition. 
For the assessment of physical vulnerability, it is important to 
refer to a certain natural hazard (e.g. flood, earthquake, 
landslide, storm, etc.) as physical characteristics of buildings 
imply different degrees of vulnerability to different hazards. As 
discussed in the literature (e.g. Cobum et al., 1994) indicators 
for SV do not necessarily relate to a certain hazard, but do 
express an intrinsic incapability and lack of resilience to cope 
with natural hazards in a more general way. This enables the 
application of this method for SV assessment for hazard- 
specific studies as well as for multi-hazard studies. 
Some indicators, e.g. construction material, are used for both 
the assessment of physical and social vulnerability. On the one 
hand, it can be used to express the resilience or the lack thereof 
to withstand the power of a hazardous event (physical 
vulnerability). On the other hand, it is being used to describe 
indirectly the social standard and wealth of a household and 
thereby their vulnerability and coping capacities in case of a 
disastrous event (social vulnerability). 
Another advantage is that medium scale approaches such as 
ours consider spatial relations of objects and their environment, 
and thereby provide a spatial framework in which information 
can be interpreted, whereas individual, very small scale 
information lacks this context and topology information. Thus, 
a combination of approaches at different levels is needed. 
In terms of data interchangeability, previous studies showed that 
the use of lidar data for the extraction of building heights can be 
substituted by the use of stereo- or orthophotos (Fraser et al., 
2002). 
Eight out of 47 variables explained almost 60 % of the variance 
of the SVI. While this is from the statistic perspective a 
satisfactory result, it has to be noted that the applied SVI was 
calculated based on a small amount of variables from census 
data not ideally suited for SV assessment. The data only include 
a limited amount of indicators relevant for SV assessment, and 
also allow validation only at the level of a census unit 
(neighbourhood). If the validation could have been performed 
In this paper we present a new method for SV assessment based 
on contextual analysis of remote sensing and GIS data. An 
approach based on physical proxy variables and spatial metrics 
that were derived from high resolution optical and laser 
scanning data, in combination with elevation information and 
existing hazard data. Object-oriented image analysis was 
applied for the definition and estimation of proxy variables that 
describe the living environment and living standards, such as