Machine Learning based EStimator for ground shaking maps workflow applied to New Zealand

Earthquakes are among the most frequent yet unpredictable natural hazards, posing substantial risk to human safety and infrastructure globally, particularly, when large-magnitude earthquakes occur. This highlights the urgent need to develop innovative and alternative methodologies for rapidly assessing the intensity of ground shaking following an earthquake.

This study explores the application of the Machine Learning Estimator for Ground Shaking Maps (MLESmap) methodology in New Zealand, a region characterized by  high seismic activity.

MLESmap utilizes extensive datasets of high-fidelity, physics-based seismic scenarios to rapidly estimate ground-shaking intensity in near real-time following an earthquake. This methodology has demonstrated evaluation times similar to those of empirical ground motion models, while offering superior predictive accuracy in the two previously tested regions: the Los Angeles basin and the South Iceland Seismic Zone (SISZ).

To adapt MLESmap for New Zealand’s seismicity, seismic simulations tailored to the unique geological and tectonic context of the region are implemented. Specifically, we use the dataset generated by CyberShake NZ, a probabilistic seismic hazard analysis (PSHA) software developed by the University of Canterbury. Using this software, a total of 11,362 finite-fault rupture simulations were performed across the region and seismic hazard results were calculated on a grid of 27,481 synthetic seismic stations. A ‘forward’ simulation approach was adopted due to the large number of output locations relative to rupture locations, the optimisation of the grid for each rupture and the intention to include plasticity.

The expected results aim to demonstrate the applicability of MLESmap to New Zealand, providing ML-based tools for rapid response actions. This study also takes the first steps in applying cascading effects to MLESmap, in order to improve the overall risk assessment and to advance prevention efforts through innovative and multidisciplinary methodologies.

 

 

©2023 ChEESE-2P Funded by the European Union. This work has received funding from the European High Performance Computing Joint Undertaking (JU) and Spain, Italy, Iceland, Germany, Norway, France, Finland and Croatia under grant agreement No 101093038.