階層ベイズモデルを用いた地震火災の出火件数予測手法とその応用

抄録

In this paper, we discuss our estimation methods for the number of ignition fires after a great earthquake. In the Great East-Japan Earthquake, occurred on March 11 2011, 404 fires happened in Eastern Japan area. We proposed three types statistical models (Generalized Linear Model, Generalized Linear Mixed Model and Hierarchical Bayesian Models) and estimated their parameters using Maximum Likelihood Estimation or Markov Chain Monte Carlo Methods. And, we calculated several risk maps for post-earthquake fires. For example, according to Population Exposure to Earthquake Fires, people in Tokyo, Kawasaki and Osaka have extremely high ignition risk after a great earthquake. Finally, we proposed realtime estimation method for the number of post-earthquake fire using these statistical models.