Explosions produce low‐frequency acoustic (infrasound) waves capable of propagating globally, but the spatio‐temporal variability of the atmosphere makes detecting events difficult. Machine learning (ML) is well‐suited to identify the subtle and nonlinear patterns in explosion infrasound signals, but a previous lack of ground‐truth data inhibited training of generalized models. We introduce a physics‐based method that propagates infrasound sources through realistic atmospheres to create 28,000 synthetic events, which are used to train ML classifiers. A simple artificial neural network and modern temporal convolutional network discriminate synthetic events from background noise with >90% accuracy and, more importantly, successfully identify the majority of real‐world explosion signals recorded during the Humming Road Runner experiment. ML models trained entirely on physics‐based synthetics advance explosion detection capabilities and make ML more viable to related fields lacking training data.

Similarity search is a popular technique for seismic signal processing, with template matching, matched filters, and subspace detectors being utilized for a wide variety of tasks, including both signal detection and source discrimination. Traditionally, these techniques rely on the cross‐correlation function as the basis for measuring similarity. Unfortunately, seismogram correlation is dominated by path effects, essentially requiring a distinct waveform template along each path of interest. To address this limitation, we propose a novel measure of seismogram similarity that is explicitly invariant to path. Using Earthscope’s USArray experiment, a path‐rich dataset of 207,291 regional seismograms across 8452 unique events is constructed, and then employed via the batch‐hard triplet loss function, to train a deep convolutional neural network that maps raw seismograms to a low‐dimensional embedding space, where nearness on the space corresponds to nearness of source function, regardless of path or recording instrumentation. This path‐agnostic embedding space forms a new representation for seismograms, characterized by robust, source‐specific features, which we show to be useful for performing both pairwise event association as well as template‐based source discrimination with a single template.

The detection of seismic events at regional and teleseismic distances is critical to Nuclear Treaty Monitoring. Traditionally, detecting regional and teleseismic events has required the use of an expensive multi-instrument seismic array; however in this work, we present DeepPick, a novel seismic detection algorithm capable of array-like detection performance from a single-trace. We achieve this performance through three novel steps: First, a high-fidelity dataset is constructed by pairing array-beam catalog arrival-times with single-trace waveforms from the reference instrument of the array. Second, an idealized characteristic function is created, with exponential peaks aligned to the cataloged arrival times. Third, a deep temporal convolutional neural network is employed to learn the complex non-linear filters required to transform the single-trace waveforms into corresponding idealized characteristic functions. The training data consists of all arrivals in the International Seismological Centre Database for seven seismic arrays over a five year window from 1 January 2010 to 1 January 2015, yielding a total training set of 608,362 detections. The test set consists of the same seven arrays over a one year window from 1 January 2015 to 1 January 2016. We report our results by training the algorithm on six of the arrays and testing it on the seventh, so as to demonstrate the generalization and transportability of the technique to new stations. Detection performance against this test set is outstanding, yielding significant improvements in recall over existing techniques. Fixing a type-I error rate of 0.001, the algorithm achieves an overall recall (true positive rate) of 56% against the 141,095 array-beam arrivals in the test set, yielding 78,802 correct detections. This is more than twice the 37,572 detections made by an STA/LTA detector over the same period, and represents a 35% improvement over the 58,515 detections made by a state-of-the-art kurtosis-based detector. Furthermore, DeepPick provides at least a 4 dB improvement in detector sensitivity across the board, and is more computationally efficient, with run-times an order of magnitude faster than either of the other techniques tested. These results demonstrate the potential of our algorithm to significantly enhance the effectiveness of the global treaty monitoring network.

This paper presents the adaptive noise cancellation model as a method for isolating the seismic response of an infrasound sensor. While this response has been thoroughly characterized in the laboratory, to date there has not been an effective method for isolating it in the field. The technique developed in this paper utilizes a collocated seismometer as a reference signal and then applies an adaptive filter to isolate the infrasound sensor's seismic response. In this paper, two adaptive algorithms are investigated for both cost and performance, including Least Mean Squares (LMS) and Block LMS (BLMS). The technique is applied to a real-world data set obtained from the October 2011 Turkish earthquake, and the results demonstrate its effectiveness in isolating the seismic response. The technique developed in this paper provides access to two previously unattainable signals: the isolated seismic response and the infrasound response uncorrupted by seismic noise. Once separated, these signals open the doors to many new and exciting applications, including augmentation of co-located seismic measurements, more precise calibration of fielded infrasound sensors, and a lowered noise floor for better detection of infrasonic events.

Similarity search is a popular technique for seismic signal processing, with template matching, matched filters and subspace detectors being utilized for a wide variety of tasks, including both signal detection and source discrimination. Traditionally, these techniques rely on the cross-correlation function as the basis for measuring similarity. Unfortunately, seismogram correlation is dominated by path effects, essentially requiring a distinct waveform template along each path of interest. To address this limitation, we propose a novel measure of seismogram similarity that is explicitly invariant to path. Using Earthscope's USArray experiment, a path-rich dataset of 207,291 regional seismograms across 8,452 unique events is constructed, and then employed via the batch-hard triplet loss function, to train a deep convolutional neural network which maps raw seismograms to a low dimensional embedding space, where nearness on the space corresponds to nearness of source function, regardless of path or recording instrumentation. This path-agnostic embedding space forms a new representation for seismograms, characterized by robust, source-specific features, which we show to be useful for performing both pairwise event association as well as template-based source discrimination with a single template.