Estimation of the location of the earthquake from single station recordings using Physics-Informed Neural Networks (PINNs)
Abstract
We propose a novel approach to estimating the earthquake's location from the single seismic station recordings using Physics-Informed Neural Networks (PINNs). Large numbers of the dataset are required to train the neural networks, but the number of earthquake recordings for any particular region is limited. So, we have generated the synthetic seismogram using SPECFEM3D software to train the model. Different realistic earthquake scenarios are simulated by varying the Strike angle, Dip angle, and Rake angle in the source parameter of the earthquake. This study presents an algorithm for PINNs applied to the acoustic wave equation to determine the earthquake's location using a single data station. PINNs are a meshless method, so they are less computationally expensive. We find that the current PINNs provide good results for estimating earthquake location with the limited amount of seismic recordings. We discuss the limitations of the PINNs for the different complex earthquake scenarios.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMNV44A..03C