Calibrating Probabilistic Tsunami Hazard Analysis workflows for subaerial landslide sources

Landslide tsunami hazard analysis is associated with high uncertainty. In other words, predicting the temporal exceedance probability for a given tsunami height will involve a very large uncertainty. As such, a common hazard methodology does not exist for landslide tsunamis. Most approaches are based on scenario analysis, while the Probabilistic Tsunami Hazard Analysis (PTHA) methods are rarely employed. A reason for the lack of a streamlined approach is arguably the uncertainty, related to lack of past landslide tsunami data that can provide a statistical background for most areas across the world. Modelling procedures also need a higher degree of sophistication than for earthquake tsunamis, particularly for the subaerial landslide sources producing impact tsunamis.

The authors of this abstract have previously developed a Landslide PTHA (LPTHA) that was used for tsunami hazard mapping in Norway. It combines landslide rates derived from slope stability assessment, with an event tree analysis of landslide kinematic parameters. To model the LPTHA uncertainty, it was necessary to run thousands of simulations using a range of values for the landslide parameters with highest influence on generated tsunami (e.g. runout distance, impact velocity, and frontal area). To accomplish this, the models were simplified, and hence, a significant model uncertainty is propagated. A persistent shortcoming of this method was that the chosen probabilities, landslide parameters, and outputs from the hazard models were not tested towards observations of previous events. To constrain the uncertainties, we propose here a new method comparing run-up observations of past events with simulations based on sets of uncertain parameter values. In this presentation, we combine a block source model with a linear dispersive tsunami propagation model coupled with a non-linear shallow water inundation model. By simulating a few historical rockslide tsunami events with this procedure, we analyse how LPTHA event sets match past models. We finally analyse which parameter datasets that are presently considered most suitable for future LPTHA forecasts.