Methodological Reconstruction of Historical Landslide Tsunamis Using Bayesian Inference

TL;DR

This paper reconstructs the potential sources of historical landslide tsunamis in Indonesia using Bayesian inference, providing insights into their likelihood and geological plausibility to improve risk assessment.

Contribution

It introduces a Bayesian inference-based method to estimate landslide tsunami sources from historical accounts, advancing understanding of tsunami risk in densely populated regions.

Findings

A reasonable match between simulated landslide tsunamis and historical reports.

Most plausible landslides challenge geological plausibility.

Future joint models could improve source attribution.

Abstract

Indonesia is one of the world's most densely populated regions and lies among the epicenters of Earth's greatest natural hazards. Effectively reducing the disaster potential of these hazards through resource allocation and preparedness first requires an analysis of the risk factors of the region. Since destructive tsunamis present one of the most eminent dangers to coastal communities, understanding their sources and geological history is necessary to determine the potential future risk. Inspired by results from Cummins et al. 2020, and previous efforts that identified source parameters for earthquake-generated tsunamis, we consider landslide-generated tsunamis. This is done by constructing a probability distribution of potential landslide sources based on anecdotal observations of the 1852 Banda Sea tsunami, using Bayesian inference and scientific computing. After collecting over 100,000 samples (simulating 100,000 landslide induced tsunamis), we conclude that a landslide event provides a reasonable match to the tsunami reported in the anecdotal accounts. However, the most viable landslides may push the boundaries of geological plausibility. Future work creating a joint landslide-earthquake model may compensate for the weaknesses associated with an individual landslide or earthquake source event.