Articles | Volume 16, issue 2
Research article
10 Feb 2016
Research article |  | 10 Feb 2016

Using open building data in the development of exposure data sets for catastrophe risk modelling

R. Figueiredo and M. Martina

Abstract. One of the necessary components to perform catastrophe risk modelling is information on the buildings at risk, such as their spatial location, geometry, height, occupancy type and other characteristics. This is commonly referred to as the exposure model or data set. When modelling large areas, developing exposure data sets with the relevant information about every individual building is not practicable. Thus, census data at coarse spatial resolutions are often used as the starting point for the creation of such data sets, after which disaggregation to finer resolutions is carried out using different methods, based on proxies such as the population distribution. While these methods can produce acceptable results, they cannot be considered ideal. Nowadays, the availability of open data is increasing and it is possible to obtain information about buildings for some regions. Although this type of information is usually limited and, therefore, insufficient to generate an exposure data set, it can still be very useful in its elaboration. In this paper, we focus on how open building data can be used to develop a gridded exposure model by disaggregating existing census data at coarser resolutions. Furthermore, we analyse how the selection of the level of spatial resolution can impact the accuracy and precision of the model, and compare the results in terms of affected residential building areas, due to a flood event, between different models.

Short summary
The building exposure component of risk models is frequently based on census data at coarse resolutions. Spatial disaggregation into finer resolutions is usually performed based on proxy variables, which is a reasonable but not ideal procedure. The availability of open data is increasing and these data can be taken into account in order to generate more accurate exposure models, which in turn can improve the results of risk models. A method to do so is proposed and its limitations are analysed.
Final-revised paper