A stochastic model for time series prediction of the number of post-earthquake fire ignitions in buildings based on the ignition record for the 2011 Tohoku Earthquake

Abstract

This article presents the development of a stochastic model for time series prediction of the number of post-earthquake fire ignitions in buildings for use in post-earthquake fire risk assessment. Two kinds of Poisson regression models with an explanatory variable of JMA instrumental seismic intensity were applied to 126 ignitions affected by ground motion, which were extracted from the ignition record for the 2011 Tohoku Earthquake: (1) a time-dependent occurrence model for the ignitions from electricity-related sources, which is coupled with a statistical model for electrical supply rate after an earthquake, and (2) a time-independent occurrence model for the ignitions from gas-related sources, oil-related sources, and others. In order to verify the models, time series prediction of the number of ignitions in the 2011 Tohoku Earthquake was conducted using Monte Carlo simulation. From the calculated results, we concluded that the models could reasonably explain the occurrence tendency of ignitions in the 2011 Tohoku Earthquake.