Data-driven probabilistic tsunami risk assessment from incomplete and uncertain historical impact records in Mediterranean coastal sites

Probabilistic tsunami risk assessment is a puzzling issue due to the many uncertainties involved. Several approaches have been tried and a variety of risk metrics were used so far. A data-driven method is an alternative approach, which was tested successfully for the entire Mediterranean region and its main oceanographic basins (Triantafyllou et al., PAGEOPH, v. 180, 2023). We continue this effort by testing the approach to a set of discrete coastal spots that have historically been affected by past tsunamis. The impact metric of a tsunami is expressed in terms of tsunami intensity values, K, assigned on a 12-degree scale similar to macroseismic scales. In a coastal spot tsunami risk was calculated on the basis of the past impact data in terms of tsunami intensity. The probabilistic model adopts that the tsunami intensities observed along a stretch of coastline are continuous and independent random values, with activity rate, r, distributed according to an exponential law similar to Gutenberg-Richter law with slope b. The so-called Hard Bounds Model was followed to account for the uncertainty involved in tsunami intensity determination, implying that the real, unknown tsunami intensity is assumed to occur within fixed boundary limits. The coastline-characteristic tsunami risk parameters r, b, Kmax are estimated using a maximum likelihood estimation technique. The procedure allows utilization of the entire data set consisting not only from the complete (recent) part of tsunami catalogue but also from the highly incomplete and uncertain historical part of the catalogue including palaeotsunami data, if any. Aleatory and epistemic uncertainties in the occurrence model are approached using a mixing Poisson-gamma distribution based purely on empirical data as an alternative formalism to the classic Bayesian method. The method was applied to a series of test-sites including the cities of Rhodes, Heraklion, Aegion, Zakynthos in Greece; the Augusta bay (east Sicily) and the volcanic island of Stromboli in Italy, and Algiers in Algeria. Tsunami risk is assessed in terms of probabilities of exceedance and return periods of certain intensity values in specific time frames.