← Back to Sessions

Exploring Paired Neural Networks to Identify Similar Waveforms

Description: 

Similar event waveforms, such as from repeating earthquakes, aftershocks, or explosions, are often identified using cross-correlation methods. However, these methods can require a careful curation of template libraries or more computational resources than pre-trained, deep-learning models, and this motivates us to explore the effectiveness of using a paired neural network (PNN) to identify similar event waveforms. The PNNs were trained to recognize similar earthquake pairs from a global dataset of real earthquakes augmented with different levels of background noise and showed success when tested against a similar dataset composed of noise-augmented, similar earthquake pairs. When transportability of the PNN model is explored, by testing against a real dataset of repeating earthquake pairs (Schaff and Richards, 2011), we find the PNN model still performs well, but to a lesser degree than when tested against data that are more like the original training dataset. We explore how subsequent PNN model tuning can improve the performance against the repeating earthquake dataset. We also investigate a hyperparameter optimization process and whether it may be used to recognize when a specific training dataset is a good representation of the target test dataset. This Ground-based Nuclear Detonation Detection (GNDD) research was funded by the National Nuclear Security Administration, Defense Nuclear Nonproliferation Research and Development (NNSA DNN R&D). SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525

Session: Advancements in Forensic Seismology and Explosion Monitoring [Poster]

Type: Poster

Date: 4/17/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Erica

Student Presenter: No

Invited Presentation: 

Poster Number: 86

Authors