Bayesian near-field tsunami forecasting with uncertainty estimates

Abstract

Tsunami waveforms can be observed at offshore locations such as sea-bottom pressure gauges or GPS-mounted buoys. Recent work has focused on using these observations to make near-field tsunami forecasts in real time. However, existing forecasting methods are limited in that they do not provide uncertainty estimates. This study develops a near-field tsunami forecasting method with uncertainty estimates. The method embeds a conditional autoregressive model in a hierarchical Bayesian inverse model. Since we sample from the posterior distribution of interest using a Markov Chain Monte Carlo algorithm, not only the mean but also the variance for forecasts can be readily obtained. The proposed method is validated through simulation-based experiments for four historical earthquakes in the Nankai Trough, Japan. Key Points Near-field tsunami forecasting with uncertainty estimates Conditional autoregressive model in hierarchical Bayesian inverse model Validation through simulation-based experiments © 2014. American Geophysical Union. All Rights Reserved.