Embracing Uncertainty in "Small Data" Problems: Estimating Earthquakes from Historical Anecdotes

Abstract

We apply the Bayesian inversion process to make principled estimates of the magnitude and location of a pre-instrumental earthquake in Eastern Indonesia in the mid 19th century, by combining anecdotal historical accounts of the resultant tsunami with our modern understanding of the geology of the region. Quantifying the seismic record prior to modern instrumentation is critical to a more thorough understanding of the current risks in Eastern Indonesia. In particular, the occurrence of such a major earthquake in the 1850s provides evidence that this region is susceptible to future seismic hazards on the same order of magnitude. More importantly, the approach taken here gives evidence that even "small data" that is limited in scope and extremely uncertain can still be used to yield information on past seismic events, which is key to an increased understanding of the current seismic state. Moreover, sensitivity bounds indicate that the results obtained here are robust despite the inherent uncertainty in the observations.