Post-earthquake fire ignition model uncertainty in regional probabilistic shaking–fire cascading multi-hazard risk assessment: A study of earthquakes in Japan

Abstract

Fires following earthquakes have significant impact on urban communities in earthquake-prone countries with wooden buildings, in addition to the main earthquake hazard of ground shaking. Regional-scale post-earthquake fire risk assessments are important for developing effective risk reduction strategies. Such risk assessments require the consideration of epistemic uncertainty, that is, uncertainty caused by lack of knowledge concerning the best models, in addition to aleatoric uncertainty. This study focused on epistemic uncertainty in post-earthquake fire ignition estimations, and incorporated it into risk assessments by using alternative empirical ignition models. The key objectives of this study were as follows: (1) correlating the probability of ignition per person to the ground motion intensity based on data for different past large earthquakes in Japan; (2) investigating the impact of using ignition models calibrated on different earthquake events in risk assessments through a realistic case study considering possible large earthquakes. A hierarchical Bayesian Poisson regression analysis revealed that the empirical relationship between the ignition probability and ground motion intensity greatly varies across six major large earthquakes in Japan from 1995 to 2022. Particularly, the effects of ignition prevention measures that have been widely implemented in households since the 1995 Kobe earthquake were inferred from much smaller ignition probabilities in approximately the last 10 years. The effects of this high epistemic uncertainty manifested as the large variability of the loss exceedance curve for a portfolio of buildings. These results indicate that the new ignition models can help to foster risk-informed decision-making under epistemic uncertainty.