Methods for assessing the epistemic uncertainty captured in ground-motion models

Abstract

A key task when developing a ground-motion model (GMM) is to demonstrate that it captures an appropriate level of epistemic uncertainty. This is true whether multiple ground motion prediction equations (GMPEs) are used or a backbone approach is followed. The GMM developed for a seismic hazard assessment for the site of a UK new-build nuclear power plant is used as an example to discuss complementary approaches to assess epistemic uncertainty. Firstly, trellis plots showing the various percentiles of the GMM are examined for relevant magnitudes, distances and structural periods to search for evidence of “pinching”, where the percentiles narrow excessively. Secondly, Sammon’s maps, including GMPEs that were excluded from the logic tree, are examined to check the spread of the GMPEs for relevant magnitudes and distances in a single plot. Thirdly, contour plots of the standard deviation of the logarithms of predicted ground motions from each branch of the logic tree (σµ) are compared with plots drawn for other relevant hazard studies. Fourthly, uncertainties implied by a backbone GMM derived using Campbell (2003)’s hybrid stochastic empirical method are compared to those of the proposed multi-GMPE GMM. Finally, the spread of the percentile of hazard curves resulting from implementing the GMM are examined for different return periods to check whether any bands of lower uncertainty in ground-motion space result in bands of lower uncertainty in hazard space. These five approaches enabled a systematic assessment of the level of uncertainty captured by the proposed GMM.

Ground-motion models for earthquakes occurring in the United Kingdom

“…nuclear power plants), have generally used a suite of groundmotion models (GMMs) from both ACRs and SCRs as well as UK-speci c models (e.g. Tromans et al, 2019;Villani et al, 2020;Mosca et al, 2022;Aldama-Bustos et al, 2023). Adjusting models from other regions (e.g.…”