Probabilistic Disaggregation Model with Application to Natural Hazard Loss Estimation of Portfolios

Abstract eng:
In natural hazard loss estimation, a resolution mismatch between hazard data and aggregated exposure data is often observed. A possible solution to this issue is the disaggregation of exposure data to match the spatial resolution of hazard data. Disaggregation models available in literature are usually deterministic and make use of auxiliary indicator, such as land cover, to spatially distribute exposures. As the dependence between auxiliary indicator and disaggregated number of exposures is generally imperfect, uncertainty arises in disaggregation. This paper proposes a probabilistic disaggregation model and illustrates its relevance in the context of natural hazard loss assessment. Whereas deterministic disaggregation models only consider one possible geographical distribution of exposures, a probabilistic disaggregation model accommodates uncertainty on spatial distribution of the exposures. This allows for assessing tail-risks of aggregated portfolios facing natural hazard more precisely. The proposed model utilizes auxiliary data and models the characteristics of uncertainty of the disaggregated variables through a scaled Dirichlet distribution. The performance of the model and its relevance to natural hazard risk assessment is illustrated at the example of a flood hazard scenario in the Canton Bern, Switzerland. The number of buildings is given aggregated at a communal level and is to be disaggregated to a 100m raster using land cover data as auxiliary information. The disaggregated portfolio is intersected with a recent flood scenario to determine the number of buildings affected by the flood. The probabilistic disaggregation model performs as expected and disaggregation uncertainty is reasonably reproduced. The relevance of the probabilistic disaggregation for natural hazard loss estimation is demonstrated; i.e. by considering disaggregation uncertainty, the risk tails of portfolios subject to natural hazard increase.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket