Stephen Arrowsmith

We compare earthquake wavefield measurements from a ∼95 km urban dark fiber distributed acoustic sensing (DAS) array and a collocated 16 km aperture broadband seismic array in Dallas, Texas. Using data from a regional M 5.2 earthquake, we assess the ability of DAS to recover ground-motion amplitudes and wavefield direction of arrival (DOA). Surface-wave (Rayleigh) amplitude ratios between broadband and DAS range from 0.25 to 22.15, revealing significantly more heterogeneous DAS coupling than previously reported. Although broadband sensors provide consistent amplitude and high-quality DOA estimates across all phases (P, S, and Rayleigh), DAS data show large amplitude variability unrelated to fiber orientation, suggesting coupling and emplacement differences as dominant factors. Signal coherence in DAS is also highly variable, with some segments showing stronger correlations during pre-event noise than during the earthquake signal. Nonetheless, by applying hierarchical clustering to interchannel cross-correlations, we identify a 36-channel DAS subgroup and obtain a DOA estimate for the Rayleigh wave within 4.5° of that derived from the broadband array. These results highlight both the promise and limitations of using urban dark fiber DAS for earthquake monitoring and underscore the need for careful characterization of fiber coupling and geometry to enable reliable wavefield reconstruction.

Infrasound phase identification is challenging because atmospheric variability strongly influences signal propagation on short timescales that are not resolved by standard atmospheric models. While traditional approaches rely on propagation modeling and array‐derived parameters, recent work suggests that waveform structure itself encodes information about atmospheric conditions. Here, we investigate whether explosion‐generated infrasound propagation phases can be identified directly from waveform data using an unsupervised deep representation learning framework. We analyze a densely sampled data set of direct, tropospheric, and stratospheric arrivals and decompose each waveform into complementary representations of energy, amplitude modulation, and wavelet phase. A convolutional‐Transformer autoencoder learns compact latent representations that are fused into a joint embedding and refined through a combined reconstruction and clustering objective. Arrival groupings are identified using a Gaussian Mixture Model applied to the learned latent space. The resulting representations clearly separate the three propagation regimes with high accuracy. Attention analysis shows that the model focuses on physically interpretable temporal features associated with each arrival type. These results demonstrate that infrasound waveforms alone contain sufficient information for phase identification and highlight the potential of data‐driven approaches for atmospheric and geophysical monitoring.

The stratosphere and mesosphere are important regions for the prediction of weather at the Earth's surface for medium‐ and long‐range forecasts. The availability of observations in these layers is lower than that of the troposphere, especially for the dynamics. While seeking new observational sources is important, there are existing infrasound data sets that provide indirect observations of these layers to be exploited. Infrasound waves generated at the Earth's surface travel horizontally and vertically through the atmosphere, and can be detected by sensor arrays at ranges of hundreds or thousands of kilometers. These waves are affected by the atmospheric conditions they encounter during propagation, and the integrated contributions can be observed in the recorded observations. Inverse problem techniques can be readily used to extract information from these integrated observations and provide valuable data related to the atmospheric conditions. We exploit observations from controlled ammunition explosions in Oklahoma, which generate waves traveling to 30–50 km altitude before being refracted back to the surface and detected 256 km from the explosion site. As model background we use the 10‐member European Reanalysis product, valid 1 hr before the explosions. We use the Modulated Ensemble Transform Kalman Filter to combine these two sources of information and obtain updated atmospheric profiles. The assimilated observations bring the atmospheric profiles closer to those obtained by solely interpolating the reanalysis product to the time of the explosions. The most benefited altitudes are those close to the refraction heights of the infrasound waves, 35–55 km.

Infrasound phase identification is challenging because atmospheric variability strongly influences signal propagation on short timescales that are not resolved by standard atmospheric models. While traditional approaches rely on propagation modeling and array-derived parameters, recent work suggests that waveform structure itself encodes information about atmospheric conditions. Here, we investigate whether infrasound phases can be identified directly from waveform data using an unsupervised deep representation learning framework. We analyze a densely sampled dataset of direct, tropospheric, and stratospheric arrivals and decompose each waveform into complementary representations of energy, amplitude modulation, and phase. A convolutional-Transformer autoencoder learns compact latent representations that are fused into a joint embedding and refined through a combined reconstruction and clustering objective. Phase groupings are identified using a Gaussian Mixture Model applied to the learned latent space. The resulting representations clearly separate the three propagation regimes with high accuracy. Attention analysis shows that the model focuses on physically interpretable temporal features associated with each phase. These results demonstrate that infrasound waveforms alone contain sufficient information for phase discrimination and highlight the potential of data-driven approaches for atmospheric and geophysical monitoring.

Data collected via seismic and infrasound array deployments are leveraged in the geosciences to detect and characterize a myriad of natural and anthropogenic sources. These deployments consist of numerous sensors placed in a predetermined configuration to amplify signal strength and improve the efficacy of array processing techniques used to measure signal directionality and waveform coherence. High-fidelity feature extraction is often predicated on interstation distance as well as the frequency content and wavelength of an incident signal. Numerous array processing softwares analyze data in sequential frequency bands to obtain a more detailed characterization of a signal. However, current algorithms are limited in their ability to determine optimal array configuration for each band. We introduce an open-source Python code, called Cardinal, to process seismic and infrasound array data in discretized time–frequency space with the option of applying an adaptive array design to determine optimal subarray configuration for each frequency band. To reduce computational time, the array processing step can be run in parallel using multithreading. Moreover, the software has the capability to aggregate array processing results from different time–frequency pixels to produce separate sets of detections, or families, with added utility via the application of an adaptive semblance threshold, which aids in isolating signals-of-interest from coherent background noise. Upon appropriate configuration, Cardinal exhibits the potential to combine distinct seismic and infrasound phases into separate families.

On 4 March 2020, an accidental explosion occurred at a chemical plant in South Korea, generating both seismic and infrasound signals, which were recorded by multiple sensors at distances as great as 400 km across the southern Korean Peninsula. We used seismic and infrasound data recorded at 5 seismoacoustic arrays, 4 infrasound arrays, 13 single infrasound stations, the KSRS array, and single seismometers from several networks to quantify the explosion characteristics. Seismic, infrasound, and air-to-ground coupled acoustic arrivals from the explosion and coda-like signals from secondary sources are identified based on array processing and analyst review. We estimated the explosion origin time and location using combined seismic Lg and infrasound backprojection methods, demonstrating the importance of the dense network data using both wavefield types. We also found that dense network data can reduce the uncertainty in the location estimation using P arrivals and infrasound back azimuths. Yield estimates using infrasound amplitudes provide a 95% confidence from 1.3 to 3.2 tons with a maximum a posteriori of 1.9 tons, determined using propagation-based, stochastic path geometry models. The methods in this study document a seismoacoustic examination procedure that can be used in forensic investigations to detect, locate, and characterize anthropogenic sources.

Seismic data recorded at industrial sites contain valuable information on anthropogenic activities. With advances in machine learning and computing power, new opportunities have emerged to explore the seismic wavefield in these complex environments. We applied two unsupervised machine learning algorithms to analyze continuous seismic data collected from an industrial facility in Texas, United States. The Uniform Manifold Approximation and Projection for Dimension Reduction algorithm was used to reduce the dimensionality of the data and generate 2D embeddings. Then, the Hierarchical Density-Based Spatial Clustering of Applications with Noise method was employed to automatically group these embeddings into distinct signal clusters. Our analysis of over 1400 hr (around 59 days) of continuous seismic data revealed five and seven signal clusters at two separate stations. At both stations, we identified clusters associated with background noise and vehicle traffic, with the latter’s temporal patterns aligning closely with the facility’s work schedule. Furthermore, the algorithms detected signal clusters from unknown sources and underline the ability of unsupervised machine learning for uncovering previously unrecognized patterns. Our analysis demonstrates the effectiveness of unsupervised approaches in examining continuous seismic data without requiring prior knowledge or pre-existing labels.

SUMMARY: The Korean infrasound catalogue (KIC) covers 1999–2022 and characterizes a rich variety of source types as well as document the effects of the time-varying atmosphere on event detection and location across the Korean Peninsula. The KIC is produced using data from six South Korean infrasound arrays that are cooperatively operated by Southern Methodist University and Korea Institute of Geoscience and Mineral Resources. Signal detection relies on an Adaptive F-Detector that estimates arrival time and backazimuth, which draws a distinction between detection and parameter estimation. Detections and associated parameters are input into a Bayesian Infrasonic Source Location procedure. The resulting KIC contains 38 455 infrasound events and documents repeated events from several locations. The catalogue includes many anthropogenic sources such as an industrial chemical explosion, explosions at limestone open-pit mines and quarries, North Korean underground nuclear explosions and other atmospheric or underwater events of unknown origin. Most events in the KIC occur during working hours and days, suggesting a dominance of human-related signals. The expansion of infrasound arrays over the years in South Korea and the inclusion of data from the International Monitoring System infrasound stations in Russia and Japan increase the number of infrasound events and improve location accuracy because of the increase in azimuthal station coverage. A review of selected events and associated signals at multiple arrays provides a location quality assessment. We quantify infrasound events that have accompanying seismic arrivals (seismoacoustic events) to support the source type assessment. Ray tracing using the Ground-to-Space (G2S) atmospheric model generally predicts observed arrivals when strong stratospheric winds exist, although the predicted arrival times have significant discrepancies. In some cases, local atmospheric data better captures small-scale variations in the wind velocity of the shallow atmosphere and can improve arrival time predictions that are not well matched by the G2S model. The analysis of selected events also illustrates the importance of topographic effects on tropospheric infrasound propagation at local distances. The KIC is the first infrasound catalogue compiled in this region, and it can serve as a valuable data set in developing more robust infrasound source localization and characterization methods.

This study introduces an earthquake detection and location technique that exploits the spatial coherence of the seismic wavefield. The method leverages the signal coherence across clusters of seismic stations to generate characteristic functions that are backprojected (migrated) to detect and locate seismic events. The effectiveness of the technique is assessed using a limited set of stations from the Oklahoma wavefield experiment, with minimal tuning of processing parameters. The technique is then applied to one day of continuous data, leading to the detection of new seismic events compared with an analyst-based catalog. Noteworthy advantages of this method include its independence from prior information or assumptions (such as template waveform) and its ability to operate effectively with a network design for which seismometers are deployed in a relatively small number of clusters rather than distributed throughout a region.

Recent geophysical studies have highlighted the potential utility of integrating both seismic and infrasound data to improve source characterization and event discrimination efforts. However, the influence of each of these data types within an integrated framework is not yet well‐understood by the geophysical community. To help elucidate the role of each data type within a merged structure, we develop a neural network which fuses seismic and infrasound array data via a gated multimodal unit for earthquake‐explosion discrimination within the Korean Peninsula. Model performance is compared before and after adding the infrasound branch. We find that the seismoacoustic model outperforms the seismic model, with the majority of the improvements stemming from the explosions class. The influence of infrasound is quantified by analyzing gated multimodal activations. Results indicate that the model relies comparatively more on the infrasound branch to correct seismic predictions.