Geophysical Journal International

SummaryFractures in reservoirs are potential conduits for fluid flow. Therefore, it is crucial to know to what extent fluid flowing through a fracture could be lost by seepage to its porous background. For this reason, the hydraulic contact between the porous background and the fracture should be characterized, ideally based on seismic reflections. The representation of a fracture as a thin porous layer can provide insight into this seepage from a dynamic poroelasticity perspective. This is possible because the seismic waves reflected from a fracture are partially converted into the slow P-wave, which is the fluid motion relative to the solid-frame, and are sensitive to hydraulic contact being sealed or leak. It is well known that the P-wave reflectivity of fractures exhibits a marked difference between sealed and leaky cases for small angles of incidence (below twenty degrees) because of the variation in conversion scattering to slow P-wave. Drawing from a recent finding that a vertically polarized shear wave (SV-wave) can also generate a robust slow P-wave, we analyze the SV-wave reflectivity at fractures that can be hydraulically connected or disconnected from the surrounding porous medium, with the aim of advancing fracture characterization. We find that the reflectivity of the SV-wave is sensitive to fluid seepage, particularly at larger incident angles (above thirty degrees) where the amplitude is diminished substantially. Therefore, SV-wave reflectivity can also be used to identify leaky fractures, complementing the information provided by P-wave reflectivity.

SummaryEarthquake early warning systems are designed to provide critical seconds of warning before strong ground shaking, facilitating emergency mitigation efforts. Existing methods, such as neural networks and ground motion prediction equation-based approaches, rely on manually defined parameters and physics-based computations, which introduce human bias and hinder the efficiency of real-time applications. Furthermore, current studies primarily focus on scalar metrics such as peak ground acceleration and peak ground velocity to evaluate earthquake impacts. These metrics are limited to measuring ground shaking intensity and fail to capture the spectral characteristics of ground motion. Therefore, a ground-motion and structural-oriented deep learning-based model is proposed to predict uniform hazard spectral acceleration values across 111 periods ranging from 0.01 to 20 seconds. The framework is initially trained and evaluated on 17,500 ground-motion records from the crustal Next Generation Attenuation West 2 project. Spectral acceleration values are predicted by two subsets: deep learning-based uniform hazard spectral acceleration models 1 and 2. These models effectively utilize feature information from the initial seconds of seismic waveforms, eliminating the need for empirically defined parameters. Two deep learning-based models are developed for two datasets representing two distinct broad geographical regions. Both models utilize a similar deep-learning architecture but vary in input settings and hyperparameters to account for regional seismic characteristics. To assess the model's goodness-of-fit between observed and predicted values, as well as its generalization ability, we rigorously compare the two models with the latest data from the U.S. Geological Survey Earthquake Hazard Toolbox and the Japanese Strong-Motion Earthquake Network, respectively. An explainable artificial intelligence technique has been applied to better understand the framework and analyze how individual input features influence the outputs of the trained models. Integrating cutting-edge deep learning technologies into ground motion and engineering seismology reveals the significant potential of the model in enhancing real-time early warning systems. This integration also provides valuable support to various end-users involved in seismic monitoring, facilitating well-informed decisions in both real-time and near-real-time scenarios.

SummaryWe demonstrate a method for the prediction of seismic discontinuity topography from thermochemical Mantle Circulation Models (MCMs). We find the discontinuity depth by using the peak reflectivity at each location in our mantle transition zone, taking account of compositional as well as thermal variations. We make some comparisons of our predicted topographies with those observed using SS-precursors, developing a simple smoothing filter to capture the distribution of sensitivity of a published topography model – finding that such filtering has a significant impact on the predicted discontinuity topographies. We also consider the significance of lateral variations in reflectivity or reflection amplitude in our predicted datasets and the real Earth. Finally, we consider what aspects of mantle-transition zone discontinuity structure would be matched by the predicted discontinuity structure from an Earth-like MCM – particularly the mean depths of the discontinuities, the amplitude of the topography and the shape of its spherical harmonic spectra.

SummaryA joint tomographic inversion for high-resolution P and S wave velocity models of the crust and uppermost mantle in the Middle East is performed using absolute and differential body wave travel times as well as Rayleigh wave dispersions from earthquakes and ambient noises. Checkerboard tests indicate that the models generally have a resolution of 2° x 2° down to a depth of 100 km and reaches 1° x 1° at a depth of 60 km in areas of high-density data coverage such as the Zagros collision zone. The velocity models reveal that the sedimentary layer in the region is nonuniform with a maximum thickness in the Mesopotamian foreland, Persian Gulf, southern Caspian Sea, and eastern Mediterranean Sea (∼10 km), whereas most of the Arabian Shield has no sedimentary cover. The Moho discontinuity vary considerably beneath the Arabian Plate with its shallowest extent at the Red Sea Rift (∼10 km) and its deepest under the Zagros collision zone (∼50 to 55 km). The Arabian Shield and Arabian Platform have a relatively uniform Moho depth of ∼40 km. Widespread low velocity anomalies in the upper mantle are found along the margins of the Arabian Plate and mountain ranges of the Anatolia and Iran plateaus which coincide with the Quaternary volcanism in the region. Extensive low velocity anomalies are observed in the upper mantle underneath the southern and central Red Sea Rift and the Arabian Shield, which may represent partial melt or upwelling hot asthenosphere material from the Afar plume or East African superplume. The southern Red Sea is in an active rifting stage driven by the upwelling of the asthenosphere, whereas the northern Red Sea is in a hybrid mode of active and passive rifting. The Arabian Plate drift toward the northeast is likely the driving force for the passive rifting. In the Zagros collision zone, crustal thickening with low velocity anomalies in the upper and mid crust is observed. This suggests that the present-day tectonic framework of the Zagros collision zone is the result of oceanic subduction of the Neotethyan Plate under the Eurasian Plate and subsequent continental collision of the Arabian Plate with the Eurasian Plate, during which the lower-velocity felsic upper crust of the Arabian Plate was dragged down under the higher-velocity mafic crust of the Eurasian Plate due to slab pull. The subducted slab has a diversified form with a torn-off central portion. The southern portion slopes steeper than its northern counterpart. The subducted Neotethyan slab likely underwent bending and tearing, and it eventually broke off. The remanent slab underplated to the overriding Eurasian Plate to form a thickened crust under the Zagros orogen. This study corroborates previous findings such as there being different modes of spreading in the northern and southern Red Sea rift and the presence of crustal thickening in the Zagros collision zone, and it unveils more details including asthenosphere material migration along the Red Sea rift and complex suture structure in the Zagros collision zone.

SummaryTimely identification of the triggering mechanism behind the observed seismicity in areas with multiple overlapping human activities is an important research topic that can facilitate effective measures to mitigate the seismic hazard. This task is particularly challenging when dealing with delayed operational data, uncertain focal depths, or uneven seismic monitoring coverage. Here, we propose a deep learning (DL) framework to identify which human activity triggered a certain earthquake in near real-time using only seismic waveforms as input. We use an advanced architecture, the compact convolutional transformer (CCT), to extract high-level abstract features from the three-component seismograms and then use an advanced capsule neural network to link the induced seismicity in West Texas with three potential causal factors, i.e., hydraulic fracturing (HF), shallow saltwater disposal (SWDsh), or deep saltwater disposal (SWDdp). The training data was prepared based on an established probabilistic approach that combined physics-based principles with both real and reshuffled injection data to hindcast past seismicity rates. In the end, each activity was assigned a confidence level for association at the 5 km spatial scale. Even though the training data include only 981 events, we obtain over 90% accuracy for all three causal factors for both the single- and multi-station versions of the model.

AbstractIntradecadal variations in the length-of-day (ΔLOD) can reveal changes in angular velocity interpreted as due to Earth’s core. Previous studies have identified periodic oscillations of around 6 and 8 years. To complement widely used Fourier methods, we investigate the ΔLOD record from 1962-2025 in the time domain, seeking smooth variations using cubic B-splines. We analyse in several ways. A penalised least-squares spline fit allows isolation of coherent variations from analysing the first and second derivatives. Alternatively, a smooth curve fit with least-squares splines allows removal of the long-period behaviour of ΔLOD. From this, we fit the residual with a pure cosine-wave of varying period but examine the data fits carefully in case the signal is non-stationary (for example from impulsive forcing). All approaches show clear evidence of signals with periods around 5.9 — and in the case for the time derivatives — 8.5 years. We find that the pure 5.9-year oscillation breaks down in 2010, with a one-off peak to peak separation of around 4.7 years. After 2014, the variation is once again consistent with an approximate 6 year oscillation. Such a discontinuous, non-stationary effect is not well-characterised by frequency-domain based methods. Seeking to understand this brief interruption of the 6 year oscillation, we extend the study length using a ΔLOD series from lunar occultation data extending back to 1800, and find it suitable to repeat our spline-based analysis from 1830 onwards. From this, we find the 6 year oscillation stable throughout the entire 19th and 20th century, with the exception of 1916–1920, where we observe a similar interruption of the 6 year variation by a single 4 year oscillation. The 2010 disruption to the 6-year oscillation is contemporary with changes in geomagnetic secular variation, modelled core surface flow, and inner core seismic signature. All of these events suggest a step change in core-processes around 2010.

SummaryThe time-variable gravity field obtained from the Gravity Recovery and Climate Experiment/Follow-On (GRACE/GRACE-FO) satellites has been successfully used to detect global water mass changes over the past two decades. However, the north-south striping noise in the GRACE spherical harmonic (SH) solution limits their effectiveness. Efforts to suppress this noise and achieve a higher signal-to-noise ratio (SNR) continue with various product releases, but there is still a great need for improvement. This study presents a new de-striping method called GBVMD, which employs a stepwise enhancing framework combining Gaussian filtering with bi-dimensional variational mode decomposition (BVMD). The methodological breakthrough comes from two innovations: First, it employs adaptive scale decomposition by dynamically adjusting the radius of the Gaussian filter in conjunction with BVMD reconstruction, effectively reducing noise across multiple scales. Second, it features a dual-decision optimization strategy that integrates SNR-driven mode reconstruction and iterative termination, thereby maximizing the SNR while adapting to the specific characteristics of the noise. In simulations, the GBVMD outperforms the five other filters in reducing noise and keeping signals, achieving an improvement in SNR by at least 19%, and reductions in root mean square error and mean absolute error by at least 14% and 11%, respectively. When applied to GRACE/GRACE-FO Level-2 SH solutions, GBVMD led to a higher SNR with an improvement of at least 12% compared to other filters. The GBVMD-filtered SH data showed strong consistency with three Level-3 Mascon solutions across 183 river basins. Comparable results were also found in polar regions, validated by altimetry data. Furthermore, we effectively corrected the leakage errors for two examples in the Caspian Sea and the Great Lakes, demonstrating the advantages of GBVMD-filtered SH over the Mascons for signal reanalysis. We recommend GBVMD for further applications, especially in specific regions such as ocean areas and other satellite missions requiring similar de-striping approaches.

SummaryDue to the extremely destructive characteristics of seismic hazards, and as one of the effective pre-seismic physical signals, abnormal changes in subsurface fluids can provide key precursor information for earthquake prediction. Furthermore, an efficient method for labelling anomalies in seismic monitoring data is urgently needed. Therefore, this paper analyses the change characteristics of subsurface fluid-water level data and proposes an Anomaly Detection Algorithm of Single Variable Time Series Data Based on Dynamic Parameter Tuning (ADSV-DPT) based on three important characteristics (jump, step and steep), which firstly determines the central tendency of the data by calculating the median of the water level data within the initial window, and then utilizes the Median Absolute Deviation (MAD) as a robust dispersion metric to reduce the impact of extreme values on the anomaly detection. The sliding window mechanism is employed to update the median and MAD step by step, thereby ensuring the efficiency and adaptability of the algorithm in processing time series data. Finally, the anomalies in the data are detected by setting dynamic thresholds. A comparison of the anomaly detection efficacy of the proposed ADSV-DPT algorithm with that of three alternative models (namely, K-Nearest Neighbor, KNN; Pruned Exact Linear Time, PELT; and One-Class Support Vector Machine, OC-SVM) was conducted. The experimental results demonstrate that the ADSV-DPT algorithm outperforms the other models in accurately identifying anomalous features. The average precision, recall, and F1-score of the ADSV-DPT algorithm all exceed 85%. The algorithm's capacity for adapting to variations in the data is noteworthy, as is its ability to accurately identify abnormal values that deviate from the established normal range.

SummaryThe quantitative interpretation of geological structures relies on multi-parameter models (MPMs) inversion. However, conventional full waveform inversion that matches simulated seismic data to observed seismic data cannot accurately obtain high-resolution MPMs because of the implicit inter-parameter coupling relations in the multi-parameter wave equation. Additionally, conventional supervised deep learning approaches that require a significant number of annotated labels cannot predict precise MPMs, as only a limited number of sophisticated synthetic MPMs are available as labels. To address this issue, we propose a self-supervised multi-parameter inversion (SS-MPI) to provide high-resolution MPMs from the prior first-arrival-based tomography and reflection-based migration image. SS-MPI creates representative MPMs from the prior information as pseudo-labels to pre-train the deep learning algorithm, which then predicts MPMs as feedback to update these training pseudo-labels iteratively. Synthetic examples of elastic and anisotropic models indicate that SS-MPI outperforms the conventional elastic full waveform inversion (EFWI) and delivers highly accurate and high-resolution MPMs.

SummaryPermafrost degradation in the Arctic is both an indicator of, and contributor to warming global temperatures. In addition to the global impact of thawing permafrost, at the local scale permafrost degradation can result in infrastructure damage, ecosystem changes, chemical and microbial releases, landfill leaks, and river bank erosion. As such, accurate mapping and monitoring of the in-situ permafrost extent is crucial. Differentiation of frozen and thawed arctic soils is commonly achieved through electrical geophysical imaging methods; however, recently a hysteresis-like phenomenon was observed in the electrical conductivity of soils undergoing freezing followed by thawing. This phenomenon can result in over an order of magnitude difference in conductivity when measured on the same sample at precisely the same temperature. Here, we explore this effect in a clean sand with low surface conduction and a frost-susceptible clay-rich soil with relatively higher surface conduction. The unsaturated samples were prepared at a moisture content of 16.5 per cent by weight. Each soil was evaluated under four different NaCl pore-fluid conductivities 10 ppm, 100 ppm, 1000 ppm, 1000 ppm. We then developed a model to simulate the full hysteresis loop for the soil. In both cases the hysteresis effect was clearly observed in the real component of the conductivity. In the frequency dependent imaginary conductivity response only the frost-susceptible soil demonstrated consistent signs of hysteresis along with a soil-specific frequency response.

SummaryDownward continuation is a very interesting approach to enhance the information content of potential field data. However, the calculation of the downward continuation represents a fundamental challenge due to its inherent instability. In this study, a strategy to perform high-order vertical derivatives using the β-VDR method is introduced, called the generalized β-VDR method. Testing on a noisy synthetic model shows that the proposed strategy has the lowest noise compared to other methods. Based on stable vertical derivatives computed by using the generalized β-VDR method, a stable downward continuation method is also presented to enhance the information content of potential field data. The applicability of the generalized β-VDR downward continuation algorithm is demonstrated on both synthetic and real field gravity anomalies and compared to other downward continuation algorithms. In the case of synthetic examples, the proposed method provides sharper images and estimates more accurate amplitudes than other algorithms, even continuing the field to a level close to causative bodies. The real application shows that the proposed algorithm can give a meaningful result that agrees well with seismic data along a profile in the area.

SummaryThere are two fundamental probabilities in the seismic phase picking process – the probability of the existence of a seismic phase (detection probability) and the probability associated with the phase arrival time estimation (timing probability). The nearly ubiquitous approach in developing deep learning phase picking models is to use a kernel, such as a truncated Gaussian, to mask the labeled phase arrival time and train a segmentation model. Once a model is trained, the times of the peaks in the output are taken as phase arrival times (picks), and the height of the peaks are taken as “probability” of the picks. Here, we show that this “probability” represents neither the detection nor the timing probability because this approach forces the output to follow the shape of the kernel. We introduce an approach using two models to estimate these two distinct probabilities. We use a binary classifier with a calibrated confidence to address the detection probability and a multi-class classifier to obtain a probability mass function to address the timing probability. This new approach can make the deep learning-based phase picking process more interpretable and provide options to logically control seismic monitoring workflows.

SummaryThis study introduces a new method for calculating acoustic-gravity waves in a spherically layered atmosphere. The method introduces a model assumption and divides the atmosphere into finely stratified layers to solve the PDE with respect to the radial coordinate. The time-domain synthetic signal is obtained by summing over the orders of the associated Legendre functions and then applying the FFT. The method is applied to numerically simulate wave behaviour, including Earth curvature effects, and compares with the horizontally layered model (HLM). Results show that at near-field distances, our method aligns closely with HLM, but significant differences emerge in the far field, particularly beyond an epicentral distance of 50°, where Earth curvature becomes critical. Our method successfully simulates head waves of seismic phases, and Rayleigh waves, even for waves travelling multiple times around the Earth, which HLM cannot achieve. Simulations using a homogeneous Earth model reveal head wave characteristics consistent with previous studies, with the strongest energy observed in Rayleigh head waves. The application of the AK135 Earth model highlights the visibility of seismic phases through the Earth’s core. We validate our method by comparing synthetic records with actual data from the 1999 Chi-Chi earthquake. The synthetic records show good agreement with observed seismic signals and ionospheric perturbations in terms of arrival time and wave envelope. These results demonstrate the accuracy of our method in simulating acoustic-gravity waves at large epicentral distances.

SummaryThe widespread, multi-year crustal deformation induced by megathrust earthquakes (Mw8+) is primarily controlled by the combined effects of continuous aseismic slip on the fault plane (afterslip) and viscoelastic relaxation driven by coseismic stress perturbations in the upper mantle. However, till today it remains a considerable challenge to separate these two mechanisms in geodetic observations. We derived the first 3-year GNSS observations following the 2021 Chignik Mw8.2 earthquake to investigate the mechanisms of postseismic deformation. We established a model capable of simultaneously simulating afterslip and viscoelastic relaxation, and constrained the upper mantle rheology beneath the Alaska Peninsula. The best-fit model effectively reproduces the GNSS observations and reveals a notable viscosity difference between the mantle wedge and the oceanic asthenosphere, with steady-state viscosities of $3 \times {{10}^{18}}$ Pa s and $4 \times {{10}^{19}}$ Pa s, respectively. The inferred mantle wedge viscosity beneath the Alaska Peninsula is lower than the values reported for south-central and southeastern Alaska, suggesting an eastward increase in viscosity along the subduction zone. Two main patches of afterslip are identified during the first 3 years. The patch of up-dip afterslip overlaps with the 1938 Chignik Mw8.3 earthquake rupture zone, and demonstrates a close spatial correlation with the slow slip event in 2018. The above new results enhance our insights into the spatial variability of regional rheology and slip behavior along the Alaska-Aleutian subduction zone.

SummaryNew Q (1/attenuation) models of the Central Alpine Fault provide unprecedented resolution to 20 km depth by incorporating new t* measurements from dense temporary seismograph deployments in the area. The models reveal significant heterogeneity in the crust, with the main Q features broadly similar along-strike the Alpine Fault but varying at length scales of 10-30 km. Accounting for heterogeneity is an important step towards understanding the seismic cycle of M7+ Alpine Fault earthquakes. Our models show the Alpine Fault as a southeast-dipping zone of very (<300) to moderately (600-900) low Q, contrasting sharply with high Q values (Qp>600, Qs>1000) within the Western Province bedrock and high Q values (Qp∼900, Qs∼1200) associated with uplifted Alpine schists to the east. The wealth of previous geologic and geophysical studies along this section of the Alpine Fault support a detailed interpretation of the observed Q values. We interpret the low Q values along the Alpine Fault as resulting from enhanced fracturing within the brittle crust with a proportion of these fractureslikely filled with fluids, which further enhance seismic attenuation through viscous dissipation. In the ductile crust (below ∼8 km depth), low Q values (<400) are likely predominantly caused by grain-size reduction from very high total shear strain and by small amounts of metamorphic fluids. Low Q values of 200-400 at 20-40 km depth downdip of the Alpine Fault and the generally low Q (<600) within the crustal root farther from the Alpine Fault, suggest increasing role of metamorphic fluids relative to that of grain-size reduction with depth and distance from the fault. The updated model also reveals a newly identified zone of low Q east of the Main Divide, approximately 40 km southeast of the Alpine Fault trace. This zone of low Q indicates significant strain accumulation on faults striking along the eastern flank of the Southern Alps, some of which have produced M6+ earthquakes in recent history. These faults represent a considerable seismic hazard for the South Island. The improved dataset and recent velocity models from temporary deployments also allow us to investigate the influence of the initial velocity model on the resulting t* measurements and Q models.

SummaryThe bulk component of the electrical conductivity of a porous material is related to the (connected) porosity and saturation by power-law functions defining the first and second Archie's laws. Recently, it was shown that for porous materials with fractal characteristics, the power-law exponent of Archie's law could be related to the fractal dimension of such materials. Similarly, the real and imaginary parts of the complex-valued surface conductivity are not just proportional to the specific surface area and saturation of the material but to power law functions of these properties defining two additional “interfacial” Archie's laws, which are called the third (saturated case) and fourth (unsaturated case) Archie's laws in this paper. These new laws have been poorly recognized and studied so far. A number of porous materials and especially clay-rich media are multiscale materials characterized by broad distributions of particle and pore sizes. We extend Archie's laws concept to describe the complex conductivity of such materials. We use both numerical simulations in fractal porous materials as well as published experimental datasets to propose a unified physical interpretation of the exponents entering the four Archie's type power-law relationships, which offer an updated complex conductivity model for natural porous media.

SummeryThe seismic hazard due to higher magnitude Himalayan earthquakes largely depends on the geometry of the underthrusting Indian Plate beneath the Himalayas, i.e., the Main Himalayan Thrust (MHT). For an objective assessment of seismic hazard in the central Himalayan seismic gap, we determine the geometry of the Main Himalayan Thrust (MHT) along 4 ∼SW-NE oriented arc-normal seismic profiles covering the central Himalayan seismic gap. We use teleseismic P- wave coda autocorrelation on waveforms recorded at 117 broadband seismic stations spread along these profiles, with an interstation spacing of 3-5 km. The results show that along these seismic profiles, the MHT is mostly of flat-ramp-flat geometry. However, the mid-crustal ramp of the MHT shows variations in its location, dip angle, and width. We also observe variations in the MHT near the Main Frontal Thrust (MFT) and Main Boundary Thrust (MBT). The observed variations in the MHT geometry within the central Himalayan seismic gap thus suggest the possibility of along-strike segmentation of the Himalayan arc, and different seismic hazard scenarios may be present during any possible higher magnitude earthquake in the central Himalayan seismic gap.

SummaryRayleigh wave is widely used for characterizing shallow subsurface structures. The conventional Rayleigh-wave methods rely on the manual picking of dispersion curves, and the dispersion curves of multi-component data are usually merged manually. The manual processing of multi-component Rayleigh waves reduces the efficiency of the method, especially when the data size and the number of modes are large. To overcome these limitations, we develop an energy-based clustering method, namely the Energy-Density-Based Spatial Clustering of Applications with Noise (E-DBSCAN) algorithm. The E-DBSCAN algorithm extracts energy clusters and dispersion curves from a single dispersion image. It considers the dispersion-energy values of the surface wave and is able to pick the dispersion curve more reliably compared with the conventional DBSCAN algorithm. We propose a two-step clustering approach for the automatic picking of multi-mode dispersion curves from multi-component data: we first extract the energy clusters in the dispersion spectra of horizontal- and vertical-component data using E-DBSCAN, respectively, and combine them in the frequency-velocity domain. Then we extract multi-modal dispersion curves from the combined multi-component energy clusters with E-DBSCAN or DBSCAN. Numerical results show that our proposed method has fairly high accuracy and estimates more abundant multi-modal dispersion curves than the single-component method. Two field examples, including an active-source and an ambient-noise dataset, prove the validity of our method and the outperformance of multi-component results compared with the single-component results. Our proposed method has a relatively low dependence on parameter selection and is also applicable to multi-offset data, which is valuable for picking multi-modal dispersion curves.

SummaryRecently, there has been an increasing interest in employing rotational motion measurements for seismic source inversion, structural imaging and ambient noise analysis. We derive reciprocity and representation theorems for rotational motion. The representations express the rotational motion inside an inhomogeneous anisotropic earth in terms of translational and rotational motion at the surface. The theorems contribute to the theoretical basis for rotational seismology methodology, such as determining the moment tensor of earthquake sources.

SummaryDetermining earthquake focal mechanisms is essential for understanding fault geometry and the stress field in the Earth's crust. When focal mechanisms are estimated using P-wave first-motion polarities, accurate polarity determination is critical. In recent years, deep-learning-based polarity-determination models have been developed. However, the estimation of focal mechanisms using P-wave polarities is often not robust. When automating this process using deep learning models, it is crucial to identify and utilize only those polarity predictions that the model infers with high accuracy and low uncertainty. In this study, we propose a novel deep learning model, PoViT-UQ, that combines a Vision Transformer (ViT) with Monte Carlo Dropout (MCD) to estimate high-precision initial P-wave polarity classification and arrival time detection with uncertainty quantification. Using seismic waveform data sampled at 100 and 250 Hz, the model classifies polarities into three classes (Up, Down, and Noise) and simultaneously estimates P-wave arrival times. The results showed a classification accuracy exceeding 98% and a standard deviation of 0.027 s in the arrival time estimation using the 250 Hz model. By integrating MCD, we evaluate prediction uncertainty and apply an interquartile range threshold of ≤0.15 to improve the accuracy of focal mechanism estimates. Validation using aftershock data from the 2016 Central Tottori Earthquake confirmed that our approach contributes to efficient and high-precision focal mechanism estimates. Our model advances automated initial P-wave polarity determination and enables reliable data selection based on uncertainty quantification.