Abstract by Hayden Ringer
Reconstructing Historical Seismic Events Using Bayesian Monte Carlo Methods
Modeling historical seismic events is a key pillar of seismic hazard analysis. Profiles of past earthquakes allow identification of fault zones capable of supporting dangerous earthquakes. Typically, forming such a profile requires detailed data from instruments (such as seismometers). However, due to the centuries-long timescale of the "earthquake cycle", a large fraction of the Earth's dangerous fault zones have not experienced a large earthquake during the modern era of seismic data collection. Anecdotal accounts of large seismic events in these regions exist (such as from newspaper articles or personal journals), however they do not directly provide enough information to determine the intensity and geometry of the event accurately.
Here we present a method to infer the source parameters of a tsunami-producing earthquake from anecdotal records. This method uses a robust forward model for simulating a tsunami as well as the Metropolis-Hastings algorithm to produce an approximate posterior distribution for the Bayesian inverse problem corresponding to estimating the earthquake parameters. We demonstrate this method for the 1852 Banda Arc Earthquake.