Efficient Numerical Methods for Computational Challenges Arising from Non-Ergodic Probabilistic Seismic Hazard Analysis
Video not found
Please check back later. Videos are usually posted within 24 hours.
Maxime Lacour
UC Berkeley
- Date & Time
- Location
- Hybrid In Person and Online-only seminar via Microsoft Teams
- Summary
Recent developments in non-ergodic ground-motion models account for spatially-varying systematic source, site, and path effects, offering significant improvements over traditional ergodic approaches for probabilistic seismic hazard analysis (PSHA). However, the implementation of the prediction of these spatially varying terms and the corresponding PSHA calculations involve major computational challenges. Predicting spatially-varying adjustment terms at new locations requires conditioning on available datasets of observations and simulations. When these datasets become large (> 100,000 records), the predictions involve large computational and memory requirements, making the method impractical on standard computers. The same issue arises when predictions are needed at many locations, such as finely discretized distributed systems, areal source zones, virtual fault representations, or national-scale seismic hazard maps requiring coverage of large geographic areas.
Non-ergodic models also introduce epistemic uncertainty in the estimation of spatially-varying terms due to limited data coverage. Propagating this uncertainty through PSHA calculations using traditional logic-tree approaches becomes computationally expensive, often requiring to reduce the analysis to just a few branches and compromising the accuracy of the epistemic uncertainty in the hazard. This presentation describes efficient numerical methods that address these challenges, using analytical approximations and sparse matrix techniques that reduce computation and memory requirements by several orders of magnitude, making non-ergodic ground-motion modeling and PSHA practical on standard computers.