← Back to Gallery

Session: Current and Future Challenges in Engineering Seismology [Poster]

Type: Poster

Date: 4/25/2019

Time: 06:00 PM

Room: Grand Ballroom

[Withdrawn] The Applicability of Lognormal Uncertainty in Ground Motion Prediction Equations: Analysis, Implications and a Proposed Alternative

Lognormal uncertainty around ground motion prediction equations is ubiquitous in the literature. However, little research into the durability of lognormal uncertainty or its applicability for forward analysis has been performed. By “durability,” I mean the likelihood that the original characterization of uncertainty remains applicable as more validation data are considered. This presentation reports the results of a study based on the NGA-West2 database (with a total of approximately 15,800 records after censoring) which evaluated the stability of the moments of lognormal uncertainty as validation data accumulates, as well as whether or not the original descriptions of model uncertainty continued to accurately describe the distribution of errors. Three initial database sizes were considered: approximately 1,500, 3,000, and 4,500 records. Lognormal uncertainty had stable moments only for relatively large initial databases (4,500 records), and was not durable for any scenario considered. Lognormal uncertainty may therefore not be suitable for use in forward analysis in some situations. Logistic distributions formulated to approximate the original normal distributions around the median were shown to be more likely to survive additional validation data due to their increased tail thickness. Use of alternative probability distributions may make ground motion prediction equations more resilient to possible outliers in the future and may therefore offer a more realistic description of model uncertainty. Implications for probabilistic seismic hazard analysis (PSHA) and performance-based earthquake engineering (PBEE) and the challenges presented will be discussed.

Presenting Author: Zach Bullock

Additional Authors