CyberShake physics-based simulations for machine learning ground motion forecasting in South Iceland
- Name: CyberShake physics-based simulations for machine learning ground motion forecasting in South Iceland
- EuroHPC machine used: MareNostrum 5
- Topic: Earth and related environmental sciences, Computer and Information Sciences
Overview of the project
This project generates a large-scale synthetic earthquake database (35,300 scenarios, up to 2 Hz) for the South Iceland Seismic Zone (SISZ-RPOR) using the CyberShake platform with its AWP-ODC solver. The resulting simulations are used to train machine-learning surrogate models (Random Forest and Neural Networks) as fast, accurate alternatives to classical ground motion models (GMMs), enabling rapid post-event hazard assessment. This is the first European implementation of the MLESmap methodology, developed within the ChEESE-2P Center of Excellence for Exascale in Solid Earth.
How did EPICURE support the project and what were the benefits of the support?
“We requested support to install and validate the GPU version of CyberShake’s AWP-ODC solver on MareNostrum 5. The EPICURE team provided invaluable assistance with this installation, which was essential to enable GPU execution of the code. Without EPICURE’s support, the GPU installation and validation process would have taken significantly longer, and we would not have been able to reach simulation frequencies of up to 2 Hz within the project timeline, a key scientific requirement for capturing the impact of earthquakes on structures of different sizes. The GPU migration also reduced our dependency on CPU resources, shifting the heavy computational load of the AWP-ODC solver to GPU and improving overall time-to-solution.” – Marisol Monterrubio