An Empirical Earthquake Ground-Motion Model Based on Truncated Regression: A Case Study in the Middle East

Session: Alpine-Himalayan Alpide Shallow Earthquakes and the Current and the Future Hazard Assessments
Type: Oral
Date: 4/30/2020
Time: 02:15 PM
Room: 215 + 220
Description: 

We present an empirical earthquake ground-motion model for the Middle East. The model is developed on a database of about 9500 strong-motion records from the Middle East. The model is developed as a Bayesian multi-level model. The multi-level structure accounts for event-specific, site-specific and regional effects. Prior distributions for the parameters are based on physical considerations and predictions from global ground-motion models.

The strong-motion observations come from stations that only record data if the peak ground acceleration (PGA) exceeds a certain trigger threshold. This presents a problem, since no observations with PGA smaller than the trigger threshold are available. This results in a truncated distribution for PGA and spectral acceleration. If this is not taken into account during the model development, the model will be biased. Typically, this is done by using only data up to a certain (magnitude-dependent) distance, making the assumption that the effect of data truncation only has an influence at larger distances on the model. Such an approach would lead to a great reduction in the usable number of data [records?] for the database, which would lead to a poorly constrained model. Hence, we directly model the truncation of PGA and spectral periods in the regression. Conditional on the parameters of the model, the likelihood of the target variables (e.g. logarithmic PGA) is given by a truncated normal distribution. Using the truncated likelihood in the regression accounts for the biased observations.

The estimated model, based on truncated regression, is in reasonable agreement with previously published global models. The data is not spatially sufficient enough to estimate nonergodic effects; however, we take systematic path effects into account. Due to the relatively sparse nature of the data, deviations of path effects from the nonergodic mean attenuation are on average small, but associated with large epistemic uncertainties.

Presenting Author: Nicolas M. Kuehn

Authors