Towards an Informed Decision Making in Validation Metrics

Naeem Khoshnevis, & Ricardo Taborda

Submitted September 4, 2017, SCEC Contribution #7442

We approach the problem of response history analysis validation from a ground motion perspective, and focus on the validation of physics-based simulation synthetics. Within this framework, we adopt an application-independent approach. That is, we assume that a good prediction (i.e., a satisfactory validation) of ground motion synthetics should inherently lead to a good prediction of the system’s response, independently of the system type (unless there exist strong ground-system interaction effects). With this in mind, we are interested in identifying the dominant characteristics in ground motion time series—as revealed by quantitative metrics used in goodness-of-fit (GOF) evaluations—that lead to the correct assessment of the validity of simulations; and how these can inform the choice of input models and parameters. An appropriate choice of validation metrics is particularly relevant in high-frequency deterministic (0–5 Hz) and broadband non-deterministic (0–20 Hz) simulations, for which modelling uncertainties render synthetics incapable of matching data on a wiggle-by-wiggle basis, but where quantitative comparisons (in both time and frequency) are still necessary. It is not clear, however, how such metrics should be chosen or why. We seek to inform the choice of such metrics on data analysis and not on personal preferences or intended applications. Here, we present some of our current efforts towards taking better informed decisions on the choice of validation metrics, and how such a choice can help rank simulations in a statistically stable manner.

Khoshnevis, N., & Taborda, R. (2017, 09). Towards an Informed Decision Making in Validation Metrics. Oral Presentation at QuakeCoRE Annual Meeting.