Global Nuclear Blast Discrimination using a Convolutional Neural Network

Abstract

Using P-wave seismograms by Barama et al. (2022), we trained a seismic source classifier using a Convolutional Neural Network. We trained for three classes: earthquake P-wave, nuclear P-wave, and noise. Seismograms with low signal to noise ratios (SNR) adversely affect the model performance, thus a threshold was applied, limiting the training set size. Our method can accurately characterize most events, finding over 95\% signals in the validation set, even with the SNR-limited training data. We applied the model on holdout datasets of the North Korean test blasts to evaluate the performance on unique region and station-source pairs. Additionally, we tested on the Source Physics Experiment events to investigate the potential for chemical blasts to act as a surrogate for nuclear blasts. We anticipate that machine-learning models like our classifier system can have broad application for other seismic signals including volcanic and non-volcanic tremor, anomalous earthquakes, ice-quakes or landslide-quakes.