Ross Maguire

Deep learning models can help mitigate the need for analyst intervention in seismic classification of local earthquake and explosion signals, but optimizing model performance across multiple localities is challenging. Here, we seek to advance multilocality classification through hyperparameter optimization, evaluation of training data selection, and incorporation of phase pick probabilities from a pre-existing deep learning model into the new source type classification models. One-branch and two-branch convolutional neural network model designs are tested with 16 local data sets, each containing earthquakes and shallow borehole explosions observed by the same receivers. The one-branch model uses a conventional input of time–frequency scalograms and the two-branch model adds vector inputs of P-wave and S-wave probabilities output from a pre-existing phase picking model, which are increasingly available in modern event detection workflows. Hyperparameters for the one- and two-branch models were optimized by a guided search algorithm. For in-distribution data tests spanning all 16 data sets, the two-branch model’s single-station balanced accuracy (95.7%) slightly outperforms the one-branch model (94.7%). In addition, the optimized two-branch model is simpler, with about half as many model parameters and faster convergence in training. Relative performance of the two-branch model increases for out-of-distribution tests (using a locality held out from training), with an average increase in single-station balanced accuracy from 84.4% to 87.6% across 16 held out localities. The two-branch model also substantially outperforms P/S amplitude ratio discrimination accuracy by 14% on average for out-of-distribution tests. The results suggest that the added context from phase detection probabilities, which are a common by-product of event catalog building, can provide an efficient means to improve generalization of event-type classification models when applied to new regions.

Tectonic deformation on stagnant-lid planets reflects the interaction between long-lived lithospheric structures and evolving stress fields. Whether the ancient tectonic fabrics can localise present-day seismic failure under contemporary stress fields remains poorly tested by direct geophysical observations. Here we show that the farside seismic event, S0976a, detected by InSight in the western hemisphere is a marsquake that occurred in Valles Marineris and involved strike-slip faulting in the deep crust. The stress orientation inferred from the focal mechanism is consistent with reactivation of pre-existing fault systems under a present-day stress field shaped primarily by long-term planetary cooling, regional Tharsis-related loading, and surface erosion-related local unloading. These observations provide direct seismic evidence that inherited fault systems on Mars can remain mechanically favourable for reactivation over billion-year timescales, suggesting that slip on deeply rooted faults remains a key mechanism of long-term strain release in a cooling single-lid planet.

The Cascadia subduction zone can produce M 9+ megathrust and M 6.5+ intraplate earthquakes, motivating the need for a detailed 3‐D model of seismic properties at scales relevant to regional geology, basins, and infrastructure. To construct the Generation 0 Cascadia Region Earthquake Science Center Community Velocity Model (CRESCENT CVM), we jointly invert ∼20 years of teleseismic P‐wave radial receiver functions and 6–50 s Rayleigh‐wave phase velocities from ambient‐noise correlations recorded by ∼2,300 onshore and offshore stations. Using a Bayesian framework, we build a pseudo‐3‐D shear‐wave velocity model with associated uncertainties. The model spans central California to southern British Columbia (∼36 to 52N) and the Juan de Fuca–Pacific Plate boundary to eastern limit of major neotectonic deformation (∼131 to 111W), with major boundaries resolved at ∼20 km laterally. Our model recovers previously observed slow velocities of the Franciscan terrane, Central Valley, and Yakima Basin juxtaposed with fast velocities of Klamath and Siletzia terranes, Sierra Nevada Batholith, and Ancestral Cascade Range at upper crustal depths (<10 km). At greater depth, we recover low‐velocity zones in the lower crust beneath the Klamath terrane and Olympic Mountains, high‐velocities extending to the base of the crust below the Siletzia terrane, and pervasive back‐arc low velocity zones. Velocity uncertainties are largest at shallow depths (∼5 km) and near the crust‐mantle transition ‐ regions that could be the focus of improvement in future CVMs. This model can serve as a starting point for future models that incorporate full waveform inversion and higher frequency data sets to better constrain seismic hazard.

The properties of the Martian crust, including its thickness, composition, and porosity, provide invaluable constraints on the origin and geological evolution of Mars. Surface-wave observations from seismic events recorded by InSight’s seismometer SEIS (Seismic Experiment for Internal Structure) have provided the best direct constraints on Martian crustal structure to date. However, surface-wave signals have only been unequivocally identified in three low-frequency family events (S1000a, S1094b, and S1222a), limiting our understanding of crustal structure and its heterogeneity. In this study, we identify fundamental-mode Rayleigh-wave signals in two Cerberus Fossae marsquakes (S0235b and S1133c) and invert their group velocity dispersion measurements for shear wavespeed structure along two source–receiver paths in Elysium Planitia. Surface-wave signals are isolated using frequency-dependent polarization analysis, which allows us to filter seismic energy based on particle motion. R1 signals identified in events S0235b and S1133c exhibit retrograde elliptical particle motion, as expected for Rayleigh waves, and yield back azimuths that are compatible with the propagation direction indicated by body waves. We find that the crustal shear wavespeed structure along both minor arc paths is consistent with existing velocity profiles in the northern lowlands, suggesting a relatively uniform composition in the mid-to-lower crust. In addition, our observations of R1, together with weak or absent Rayleigh-wave overtone signals, suggests hypocentral depths of S0235b and S1133c of less than 30 km, placing them in the crust.

We advance the use of convolutional neural networks (CNNs) for discriminating low-yield seismic events recorded at local distances by evaluating a CNN approach based on time–frequency representations (scalograms) of seismic records from earthquakes, mine blasts, and mining-related seismic events in the Kiruna mining region of northern Sweden to (1) determine if the CNN approach can outperform the P/S amplitude ratio method in classifying these source types, and (2) examine the regional transportability of a CNN model trained on data from the United States. An accuracy of 90% or greater was obtained for the CNN approach for binary source classification between the three source types (earthquakes, mine blasts, and mining-related events), an accuracy level not achieved by the P/S amplitude ratio method, illustrating superior performance of the CNN approach over the amplitude ratio approach. The CNN model trained on explosions and earthquakes in United States yields poor binary classification performance (accuracy < 90%) when applied to earthquakes and mine blasts in the Kiruna mining region, suggesting limited transportability of the U.S.-trained model. However, the poor performance may arise from differences in the blasting style between the two data sets (single-fired borehole explosions in the United States versus ripple-fired blasts into a mine shaft at the Kiruna mine) and source depths (near surface in United States vs. 800–900 m depth in the Kiruna mine), leaving open the question of whether transportability is more limited by differences in local geologic structure or in explosion source processes.

Deep learning models trained to estimate the probability of seismic P and S phases are rapidly expanding the scale of local event detections. Here, we evaluate the potential for deep learning model output phase detection probabilities to contribute to event-type classification, particularly discrimination of single-fired borehole explosions and earthquakes at local distances (<300 km). Motivated by the empirical success of P/S amplitude ratios, we consider the difference between P and S pick probability output from previously developed phase detection models, Pprob−Sprob, as a discriminant. Test data include ML∼1–4 earthquakes and explosions observed by common seismographs in ten geologically diverse localities. Depending on the picking model and training data, binary classification using Pprob−Sprob with at least three stations can achieve approximately equivalent classification accuracy as P/S amplitude ratios without requiring any customization. Joint classification with P/S and Pprob−Sprob improves accuracy for most quality control scenarios. Pick probabilities are an efficient attribute to consider in explosion discrimination because they can be automated byproducts of event detection. They avoid the binary choice of picking or not picking weakly visible S waves common to explosions.

The seismometer onboard InSight NASA Mars mission discovered a seismically active planet. We focused on the strongest event named S1222a (4 May 2022, M _w ∼ 4.7), which was recorded by the Very Broad Band sensors and associated channel ELYSE and is located 37.2° away from InSight. We use two different methods based on a point source approach for an elastic, horizontally layered medium to retrieve source parameters of S1222a. In the first case, the seismic moment tensor inversion of high‐frequency seismogram data is calculated using a matrix method for the direct waves. The process includes the generation of records in displacement using the frequency‐wavenumber integration technique. A method of inversion of the moment tensor of direct P ‐ and S ‐waves, less sensitive to path effects than reflected and transformed waves, is presented, which significantly increases the accuracy and reliability of the method. In the second case, tensors were calculated using common low‐frequency full‐waveform inversion and the tests to verify the plausibility of this solution obtained from the single station calculation were performed and the uncertainty estimations for inversions can be useful in future research.

Although accurately classifying signals from earthquakes and explosions at local distance (P/S ratio (measured in the 10–18 Hz frequency band), even when CNN models are tested on data from entirely unseen regions.

On 4 May 2022 the InSight seismometer SEIS‐VBB recorded the largest marsquake ever observed, S1222a, with an initial magnitude estimate of 4.6. Understanding the depth and source properties of this event has important implications for the nature of tectonic activity on Mars. Located ∼37° to the southeast of InSight, S1222a is one of the few non‐impact marsquakes that exhibits prominent surface waves. We use waveform modeling of body waves (P and S) and surface waves (Rayleigh and Love) to constrain the focal mechanism, assuming a double‐couple source, and find that S1222a likely resulted from reverse faulting in the crust (source depth near 22 km). We estimate the scalar moment to be 2.5 × 10¹⁵–3.5 × 10¹⁵ Nm (magnitude M_W 4.2–4.3). Our results suggest active compressional tectonics near the dichotomy boundary on Mars, likely due to thermal contraction from planetary cooling.

Using seismic recordings of event S1222a, we measure dispersion curves of Rayleigh and Love waves, including their first overtones, and invert these for shear velocity (V_S) and radial anisotropic structure of the Martian crust. The crustal structure along the topographic dichotomy is characterized by a fairly uniform vertically polarized shear velocity (V_SV) of 3.17 km/s between ∼5 and 30 km depth, compatible with the previous study by Kim et al. (2022), https://doi.org/10.1126/science.abq7157. Radial anisotropy as large as 12% (V_SH > V_SV) is required in the crust between 5 and 40 km depth. At greater depths, we observe a large discontinuity near 63 ± 10 km, below which V_SV reaches 4.1 km/s. We interpret this velocity increase as the crust‐mantle boundary along the path. Combined gravimetric modeling suggests that the observed average crustal thickness favors the absence of large‐scale density differences across the topographic dichotomy.