Abstract
A crucial basis for risk assessment forms the characterisation of an exposed building stock with respect to its vulnerability. Commonly used approaches are based on in-situ screening methods by structural engineers. These methods proved to be sufficient in characterising individual buildings and determining their vulnerability, but usually they are highly time- and cost-consuming. Given high urbanization rates and increasingly high spatio-temporal variability in many present-day cities, local governments are often unable to keep track of the building inventory exposed to seismic hazard. As a result, exposure and vulnerability information is often out-of-date, highly aggregated or spatially fragmented, leading to large unknown uncertainties introduced into seismic risk assessments. Moreover, if information is available, the characterisation of exposed assets is often not standardized and can hardly be compared between regions or even between surveys. Based on these observations, this dissertation aims at developing tools and methodologies for a cost- and time-efficient exposure estimation over large areas that can be applied in a standardized and comparable way to different urban environments and be efficiently scaled depending on the desired level of detail in order to provide a valuable input for rapid vulnerability assessments. A sampling approach is introduced that employs a novel multi-source method for evaluating exposure characteristics. In a top-down analysis, the processing scheme moves from an aggregated neighbourhood scale to a detailed per-building scale involving three analysis tiers that are based on analysis of different image types. In a tier 1 analysis, at an aggregated neighbourhood scale medium resolution satellite images are analysed to stratify an urban environment into areas of relatively homogeneous urban structure. The tier 1 exposure information layer provides aggregated information about the spatial distribution of predominant building types with associated structural characteristics and their approximate construction date. A hierarchical classification scheme for exposure elements is developed that builds up on taxonomy standards of the Global Earthquake Model (GEM). A tier 2 analysis at a per-building scale based on high resolution satellite images provides with location, area and shape of individual buildings. It can further enrich the tier 1 exposure layer with information about the number and density of buildings aggregated for each area of homogeneous urban structure. In combination with census information also detailed population distributions are mapped. The tier 1 or tier 2 exposure information layers, moreover, concur to define the strata composing a stratified sampling scheme to identify representative areas for a more detailed tier 3 analysis. To overcome limitations of a purely satellite-based approach, a ground-based mobile mapping system with omnidirectional camera is introduced that provides with information about the street view of the objects of interest. An exemplification of automated analysis of the omnidirectional images to extract building characteristics is given and a novel Remote Rapid Visual Screening (RRVS) techniques for a manual image analysis is introduced. To provide the necessary tools for exposure estimation from multi-source images, a modular image processing chain is developed based on free and open-source solutions. It follows a state-of-the-art object-based image analysis (OBIA) approach and consists of several modules including image preprocessing, segmentation, multi-scale segmentation optimization and evaluation, feature selection and machine learning based classification. Moreover, an exposure information integration with Bayesian networks is introduced as exemplification of a novel probabilistic vulnerability assessment based on multi-source (imaging) information. For each building successfully characterised by satellite- and ground-based imaging a posterior probability distribution of its vulnerability classes according to the European Macroseismi Scale 1998 (EMS-98) is derived. Given the assumption that each stratum is composed of relatively homogeneous urban structure, the sampled information can be back-propagated in a subsequent bottom-up approach from per-building scale to neighbourhood scale to derive detailed exposure and vulnerability distributions for a whole city. The proposed methodologies for exposure estimation have been successfully adjusted and tested within the activities of the Earthquake Model Central Asia (EMCA) for the city of Bishkek, Kyrgyzstan. Accuracy assessments indicate overall good quality of the resulting exposure information layers at the different levels of detail. Comparison with data from commonly used Rapid Visual Screening (RVS) surveys of the study area could confirm high accuracies of the results. Moreover, the methodologies proved to be transferable to other study areas and accuracy assessments of the derived products for other Central Asian (Dushanbe, Tajikistan, Osh and Jalalabad, Kyrgyzstan) and European (Cologne, Germany) cities could provide highly accurate results.
