Modelling non-volcanic tremor observations using autoregressive hidden Markov models
Chia-Chuan (Willie) Huang and Ting Wang
Department of Mathematics and Statistics, University of Otago, Dunedin
Non-volcanic tremor is a series of low frequency seismic vibrations detected in the tectonic areas. Such tremor events are associated with slow slip events which are commonly observed within the same source region of megathrust earthquakes. Understanding the connections between these three types of seismic activities may therefore aid forecasts of large destructive earthquakes. Insights into such mechanism can be extracted by studying the spatiotemporal migration pattern of tremors.
A hidden Markov model was proposed recently to study the spatiotemporal variation of tremor activities in Southwest Japan. However, the dependence structure of that hidden Markov model was unable to capture the correlation between tremor clusters. We developed a 1D autoregressive hidden Markov model to analyse the spatiotemporal behaviour of the tremor activity in the Tokai region, Southwest Japan. To facilitate the analysis, we reduced the dimension of the data by projecting the 2D locations onto a line northeast of the region. This new model captured major correlation arose in the tremor activity. It classified the tremor systems into three types: active, transit and background. By comparison, the forecasting ability of the new model outperforms the hidden Markov model developed in earlier research. Future studies will relax the model assumptions further to provide a more robust model.
This presentation is eligible for the NZSA Student Prize.