[S-SS25_2PM1] Earthquake prediction

[SSS25-04] Comments on a Bayesian approach to earthquake probabilities of the Poisson model

*Masajiro IMOTO¹, Hiroyuki FUJIWARA¹ (1.NIED)

Keywords:Earthquake probability, Poisson model, Bayesian statistics, Seismic hazard maps for Japan, Kanto

In making national seismic hazard maps for Japan, earthquake probabilities are estimated based on past seismicity with the Brownian passage time model and the Poisson process model. With a small number of past earthquakes, unreliable model parameters produce large uncertainties of estimated values. In the present paper, we discuss a Bayesian approach to the problem for the Poisson model. When n earthquakes were observed in period T₀, a Bayesian approach gives the probability that m earthquakes are observed in period T₁ in the form of a binomial distribution. We compared Bayesian probabilities with those obtained by the maximum likelihood estimate (MLE) for n less than 5 and found the following significant differences between them. 1) When T₁ is the average interval of the past earthquakes, Bayesian probabilities of at least one earthquake increase 3 to 12% over those of MLE. 2) For a somewhat smaller T₁ than that in 1), the differences become larger. The Bayesian approach presented here could be tested by a simulation study.