Defining area at risk and its effect in catastrophe loss estimation: a dasymetric mapping approach

Abstract

Catastrophe loss estimation for natural hazards combines both hazard and exposure data. While hazard attributes such as intensity distributions are usually represented at a spatially explicit raster (or pixel) level, exposure data such as population, dwellings and insurance portfolios are usually only available at spatially lumped census tracts. In current loss estimation studies, this spatial incompatibility is often inadequately addressed and a uniform distribution of exposure data within an areal unit assumed. As a result, loss estimation models overlook a great deal of spatial disparity. This paper defines occupied residential area as the area at risk and uses a dasymetric mapping approach to obtain this from areal census tracts. Using Sydney (Australia) as an example, residential areas at risk were produced through street buffers. The effect of incorporating area at risk in loss estimation models in this manner was then tested for two hazards (earthquakes and hailstorms) that impose very different-sized damage footprints. Total numbers of separate houses (as exposure data) were represented at two hierarchically nested areal unit forms (postcode and census collection district—CCD) and their corresponding residential area forms. The spatial distribution of calculated losses for these different forms was then pairwise compared. For earthquakes, estimated losses were insensitive to the manner of delineating area at risk and the use of finer-resolution exposure data. This follows because the affected zone is much larger than the areal unit to which exposure data are attached. Hailstorms, on the other hand, have relatively small affected zones, and loss estimates at a coarse postcode level were considerably different to those at postcode-based residential, CCD and CCD-based residential levels. Differences between CCD and CCD-based residential area levels were relatively small in both cases because the distribution of fine CCD units closely reflects the underlying residential areas. Our empirical findings suggest that improved delineation of the area at risk and employing exposure data based on finer areal units are important for improving loss estimation from catastrophic events, particularly those that affect only a small proportion of the area under consideration. The results also have significance for other multidisciplinary studies concerned with the integration of spatially explicit environmental data and spatially lumped socioeconomic data.