Abstract
Abstract. Probabilistic seismic risk analysis is widely used in the insurance industry to model the likelihood and severity of losses to insured portfolios by earthquake events. The available ground motion data – especially for strong and infrequent earthquakes – are often limited to a few decades, resulting in incomplete earthquake catalogues and related uncertainties and assumptions. The situation is further aggravated by the sometimes poor data quality with regard to insured portfolios. For example, due to geocoding issues of address information, risk items are often only known to be located within an administrative geographical zone, but precise coordinates remain unknown to the modeler. We analyze spatial seismic hazard and loss rate variation inside administrative geographical zones in western Indonesia. We find that the variation in hazard can vary strongly between different zones. The spatial variation in loss rate displays a similar pattern as the variation in hazard, without depending on the return period. In a recent work, we introduced a framework for stochastic treatment of portfolio location uncertainty. This results in the necessity to simulate ground motion on a high number of sampled geographical coordinates, which typically dominates the computational effort in probabilistic seismic risk analysis. We therefore propose a novel sampling scheme to improve the efficiency of stochastic portfolio location uncertainty treatment. Depending on risk item properties and measures of spatial loss rate variation, the scheme dynamically adapts the location sample size individually for insured risk items. We analyze the convergence and variance reduction of the scheme empirically. The results show that the scheme can improve the efficiency of the estimation of loss frequency curves and may thereby help to spread the treatment and communication of uncertainty in probabilistic seismic risk analysis.
Highlights
Seismic risk analysis is widely used in academia and industry to model the possible consequences of future earthquake events, but it is often limited by the availability of reliable earthquake event ground motion data over a longer period of time, resulting in the necessity for many assumptions and a wide range of deep uncertainties (Goda and Ren, 2010)
Building on a framework proposed in a recent study, in the present paper we describe a novel variance reduction scheme designed to increase the computational efficiency of stochastic treatment of portfolio location uncertainty in probabilistic seismic risk analysis (PSRA)
The variance reduction and speedup obtained with the proposed adaptive location uncertainty sampling scheme are analyzed using the western Indonesia hazard model described in Sect. 3.1 in conjunction with a vulnerability model for regional building stock composition
Summary
Seismic risk analysis is widely used in academia and industry to model the possible consequences of future earthquake events, but it is often limited by the availability of reliable earthquake event ground motion data over a longer period of time, resulting in the necessity for many assumptions and a wide range of deep uncertainties (Goda and Ren, 2010). Probabilistic seismic risk analysis (PSRA) is the means of choice to model the likelihood and severity of losses to insured portfolios due to earthquakes. In this context precise exposure locations are often unknown, which can have a significant impact on scenario loss, as well as on loss frequency curves (Bal et al, 2010; Scheingraber and Käser, 2019). For PSRA in the insurance industry, uncertainty is usually taken into account by means of Monte Carlo (MC) simulation (e.g., Pagani et al, 2014; Tyagunov et al, 2014; FoulserPiggott et al, 2020) This is a computationally intensive process, because the error convergence of MC is relatively slow and a high-dimensional loss integral needs to be evaluated with a sufficient sample size.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call