The source mechanisms of induced microseismic events help understanding underground operations and mitigating hazards associated with induced seismicity. However, the uncertainty in the inverted source mechanisms is not well understood. In this study, we examine the impact of digital filters applied to dense surface monitoring data on the inverted source mechanisms derived from P-wave amplitudes. Ten filters, designed and used to increase signal to noise ratio, were tested. Filtering strongly affects both the shear and non-shear components of the full moment tensor. The differences in shear component orientation can exceed 20 degrees in Kagan angle for some filters, despite the excellent coverage provided by the monitoring network. By constraining the inversion to pure shear mechanisms, the orientation was more stable. The smallest errors were observed with bandpass, interferometry, wavelet (with a well-chosen wavelet), and Wiener filters. On the other hand, the SVD and AGC filters resulted in largest changes in source mechanisms. Our results show that data filtering can lead to significant errors in the source mechanisms, which could potentially be misinterpreted if used to infer stress or other reservoir parameters.

Real-time monitoring of induced seismicity is critical to mitigate operational risks, relying on the rapid and accurate classification of triggered data from continuous data streams. Deep learning models are effective for this purpose but require substantial computational resources, making real-time processing difficult. To address this limitation, a lightweight model based on the Fourier neural operator (FNO) is proposed for the classification of microseismic events, leveraging its inherent resolution-invariance and computational efficiency for waveform processing. In the STanford EArthquake Dataset (STEAD), a global and large-scale database of seismic waveforms, the FNO-based model demonstrates high effectiveness for trigger classification, with an F1 score of 95% even in the scenario of data sparsity in training. The new FNO model greatly decreases the computer power needed relative to current deep learning models without sacrificing the classification success rate measured by the F1 score. A test on a real microseismic dataset shows a classification success rate with an F1 score of 98%, outperforming many traditional deep-learning techniques. The reduced computational cost makes the proposed FNO model well-suited for deployment in resource-constrained, near-real-time seismic monitoring workflows, including traffic-light implementations. The source code for the proposed FNO classifier is available at https://github.com/ayratabd/FNOclass.

We use Phase Neural Operator (PhaseNO) combined with transfer learning (TL) to design robust phase arrival time (picking) algorithm for microseismic events detected by a local sparse surface network of receivers. Unlike traditional single-station methods, PhaseNO employs both Fourier and Graph Neural Operators to exploit spatio-temporal context across the entire seismic network. By fine-tuning the pretrained model on limited microseismic datasets induced by hydraulic fracturing, we achieve higher consistency and further improvements in phase arrival detection. The finetuned algorithm is showing enhanced precision, recall, F1 score, and accuracy, attaining human-level picking performance while avoiding any manual effort. Moreover, while training the network on a large surface array of nine receivers, we are able to achieve improvement in confusion matrix scores on a sparse array of five stations.

Seismic receivers are placed in shallow boreholes to increase recorded seismic waves’ signal-to-noise ratio, especially for detecting small seismicity or microseismicity. This study demonstrates that both the seismic signal and seismic noise at frequencies greater than 1 Hz decay with depth in shallow boreholes. Furthermore, we observe that the seismic noise consists of body and surface waves. The body-wave noise can be modeled as a wave originating from sources at the surface that penetrate depths exceeding one wavelength of the surface waves. We show that seismic noise at the surface and its immediate vicinity decays exponentially because it is dominated by surface waves for depths smaller than one wavelength of surface waves. The specific sources of seismic noise significantly vary between two studied datasets (anthropogenic in Groningen and wind in FORGE), the observed shallow borehole noise levels can be characterized with the same conceptual model.

Seismic phase picking is fundamental for microseismic monitoring and subsurface imaging. Manual processing is impractical for real-time applications and large sensor arrays, motivating the use of deep learning-based pickers trained on extensive earthquake catalogs. On a broader scale, these models are generally tuned to perform optimally in high signal-to-noise and long-duration networks and often fail to perform satisfactorily when applied to campaign-based microseismic datasets, which are characterized by low signal-to-noise ratios, sparse geometries, and limited labeled data. In this study, we present a microseismic adaptation of a network-wide earthquake phase picker, Phase Neural Operator (PhaseNO), using transfer learning and parameter-efficient fine-tuning. Starting from a model pre-trained on more than 57,000 three-component earthquake and noise records, we fine-tune it using only 200 labeled and noisy microseismic recordings from hydraulic fracturing settings. We present a parameter-efficient adaptation of PhaseNO that fine-tunes a small fraction of its parameters (only 3.6 We then evaluate our adapted model on three independent microseismic datasets and compare its performance against the original pre-trained PhaseNO, a STA/LTA-based workflow, and two state-of-the-art deep learning models, PhaseNet and EQTransformer. We demonstrate that our adapted model significantly outperforms the original PhaseNO in F1 and accuracy metrics, achieving up to 30

This study assesses the effectiveness of various seismic monitoring arrays, including surface-based, shallow borehole, and Distributed Acoustic Sensing (DAS) arrays, for detecting microseismic events at a potential sequestration site in Saudi Arabia. The analysis focuses on two key parameters: the sensitivity of the arrays in detecting seismic events and the accuracy in locating detected events. Sensitivity is quantified by determining the minimum detectable moment magnitude at three depth intervals of interest: the seal, reservoir, and underburden layers. Results indicate that surface-based and shallow borehole arrays are more effective at detecting weak (around moment magnitude 0.5 and weaker) seismic events at shallow layers, whereas DAS arrays exhibit significantly reduced sensitivity at greater distances from the monitoring borehole due to increased attenuation. The study also examines location uncertainty caused by variations in seismic wave arrival times and the impact of different array configurations. While DAS arrays can detect seismic events, their ability to accurately locate events is limited by their sensitivity to horizontally propagating waves, particularly at greater distances from the monitoring borehole. For effective microseismic monitoring, the study concludes that DAS arrays should be spaced between 2 km and 4 km, while surface and shallow borehole arrays are preferable for monitoring the top seal formation. Furthermore, surface and near surface arrays are able to differentiate between seismicity from the seal, reservoir, and underburden subject to accurate velocity model.

Velocity models are essential for accurately locating the rapidly increasing seismicity in Texas. The region's limited monitoring infrastructure and extensive sedimentary basins underscore the need for developing both P- and S-wave models, especially for precise depth estimation of seismic events. This study utilizes seismic interferometry and surface wave inversion techniques, along with receiver functions, to construct a three-dimensional velocity model for Western, Central and Southern Texas. Our results indicate that the integration of receiver functions significantly improves the stability of the surface wave inversion process. The resulting inverted model aligns well with known geological structures, revealing lower S-wave velocities in sedimentary basins and higher velocities in areas with bedrock exposure. Notably, the velocity contrasts between the sedimentary basins and bedrock can reach up to 30% at equivalent depths. Furthermore, the S-wave velocities derived from our model are considerably lower than those reported in previous research, suggesting that the use of this revised S-wave model may require a reevaluation of the depths at which seismic events are located.

Advanced seismicity monitoring is needed for CO2 sequestration monitoring. Current regulator practices (so-called traffic light systems-TLS) are limited to mitigate public hazards and associated risks caused by induced seismicity. Such seismicity is often associated with slip on larger faults below the reservoir. We propose an advanced seismic monitoring strategy that not only accounts for felt seismicity but also targets seismicity in the seal and reservoir. This novel concept of tiered seismicity criteria for an advanced seismic monitoring strategy is governed by a storage site's specific geological properties (underburden, reservoir and seal). These observed seismicity criteria can be set by the regulator or operator to develop a corresponding and fit for purpose system that further manages induced seismicity to ensure seal integrity and storage longevity.

Abstract. Induced microseismicity has been detected in the Decatur CO2 sequestration area, providing critical constraints on the stress state at the reservoir. We invert the full stress tensor with two subsets of source mechanisms from the induced microseismic events. To achieve this, we incorporate additional information on the vertical stress gradient and instantaneous shut-in pressure (ISIP) measured in the area. Additionally, our results demonstrate that constraining the intermediate stress tensor to a vertical orientation is essential to achieve a consistent stress inversion. The inverted stress is then used to estimate the minimum activation pressure required to trigger seismicity on fault planes identified by the source mechanisms. The comparison of the minimum activation pressure with injection pressure indicates one of three possibilities: the ISIP pressures are significantly lower than predicted (approximately 28–29 MPa), the maximum horizontal principal stress is extremely high (exceeding 120 MPa), or the coefficient of friction is significantly lower than 0.6 on a large number of activated faults. Our analysis also shows that poorly constrained source mechanisms do not lead to reasonable stress constraint estimates, even when considering alternative input parameters such as ISIP and vertical stress. We conclude that induced microseismicity can effectively be used to estimate the stress field when source mechanisms are also well constrained. For future CO2 sequestration projects, measuring and constraining ISIP pressure and maximum horizontal stress in the reservoir will ensure that more accurate estimates of stress state from moment tensor inversions can be obtained for improved prediction of the long-term reservoir response to injection.

Long-term seismic monitoring arrays are often deployed to shallow boreholes to reduce the seismic noise. We investigate noise level decay in shallow boreholes. A large number of publicly available data with such deployment is available at the seismic monitoring array near the town of Groningen, which allows also characterization of the seismic noise decay in shallow boreholes in urban environments. We study noise distribution at 4 sites from this array. Each site includes 5 receivers deployed in shallow vertical boreholes with 50 meters intervals between the surface and 200 m depth. We show there is no difference between noise levels during the summer and winter at the borehole instruments. However, we observe diurnal variation at all depth levels. We also show there are higher noise levels throughout weekdays and lower during weekends and state holidays. These changes are not only observed at the surface but also at the deepest receivers. This implies that the dominant source of this noise is anthropogenic and it penetrates to depths of 200 meters even at frequencies exceeding 5 Hz. This observation is contradicting the common assumption that the seismic noise consists of the surface waves.

We adopt extreme value theory to estimate the upper limit of the next record-breaking magnitudes of induced seismic events. The methodology is based on order statistics and does not rely on knowledge of the state of the subsurface reservoir or injection strategy. The estimation depends on the history of record-breaking events produced by the anthropogenic activities. We apply the methodology to three different types of industrial operations: natural gas production, saltwater disposal and hydraulic fracturing. We show that the upper limit estimate provides a reliable and realistic upper bound for magnitudes of the record-breaking events in investigated datasets including 15 publicly available datasets. The predicted magnitudes do not overestimate the observed magnitudes by more than 1.0 magnitude unit and underestimation is rare, probably resulting from insufficient sampling of the statistical distribution of the induced seismicity. The richest dataset, sourced from downhole and surface monitoring of the Preston New Road hydraulic fracturing, provides reliable estimates of the magnitudes over three orders of magnitudes with only slight underprediction of the largest observed event. While the detection of weaker events improves the performance of the method, we show that it can be applied even with a few observed record-breaking events to provide reliable estimates of magnitudes. However, care must be taken to ensure that event catalogues are estimated consistently across a range of magnitudes.

Summary We evaluate reliability of the inverted source mechanisms from P-wave amplitudes acquired by a dense surface monitoring array while using filters designed for a noise attenuation in surface dataset. Application of wavelet, Wiener, AGC and interferometry filters results in a significant distortion of the source mechanism of the tested microseismic event. Using filters significantly affects shear and non-shear components of the full moment inverted moment tensor as well as orientation of the shear component of the full moment tensor. The differences in shear components may exceed 20° of Kagan angle. Constraining the inversion to pure shear source and using the same data is providing more consistent results resulting in less than 10° difference in Kagan angle for interferometry, wavelet (with good choice of the wavelet) and Wiener, beside the bandpass filter. We show that data filtering can lead to noteworthy errors in the source mechanisms that might be potentially misinterpreted. These errors can be avoided by careful application of the filters before inverting for source mechanism. We recommend bandpass filtering of surface microseismic data as this filter seems to be the least affecting inverted source mechanism.

Preview this article: How Large should Microseismic Monitoring Networks be for CO2 Injection?, Page 1 of 1 < Previous page | Next page > /docserver/preview/fulltext/fb/42/4/fb2024031-1.gif

We have developed a Recurrent Neural Network (RNN)-based phase picker for data obtained from a local seismic monitoring array specifically designated for induced seismicity analysis. The proposed algorithm was rigorously tested using real-world data from a network encompassing nine three-component stations. The algorithm is designed for multiple monitoring of repeated injection within the permanent array. For such an array, the RNN is initially trained on a foundational dataset, enabling the trained algorithm to accurately identify other induced events even if they occur in different regions of the array. Our RNN-based phase picker achieved an accuracy exceeding 80% for arrival time picking when compared to precise manual picking techniques. However, the event locations (based on the arrival picking) had to be further constrained to avoid false arrival picks. By utilizing these refined arrival times, we were able to locate seismic events and assess their magnitudes. The magnitudes of events processed automatically exhibited a discrepancy of up to 0.3 when juxtaposed with those derived from manual processing. Importantly, the efficacy of our results remains consistent irrespective of the specific training dataset employed, provided that the dataset originates from within the network.

Should we try to improve surface monitoring array performance by deploying the sensors at the shallow borehole? We show a case where the surface sensor recorded higher signal-to-noise than the shallow borehole sensor. This improvement is limited to a frequency band between 20 Hz and 50 Hz and it is not caused by lack of coupling of the sensor in the shallow borehole. Our explanation is the resonance effect in the near-surface layers which improves detection of microseismic events.

Summary We use the unique dataset of the Preston New Road induced seismicity monitoring to evaluate the predictive power of the upper limit magnitude methodology. The upper limit magnitude methodology is based on extreme value theory and provides an upper limit on next record-breaking magnitudes based on observed seismicity. We show that this methodology provides reliable upper bound for the magnitudes during the hydraulic fracturing and we additionally evaluate its performance assuming different values of magnitude completeness corresponding to different induced seismicity monitoring networks ranging from downhole to surface monitoring arrays. We show that a higher magnitude of completeness does not results in reduced performance of the upper limit magnitude methodology. Finally, we show that it is more important to consider prior seismicity than drilling a deep dedicated monitoring borehole for induced seismicity monitoring.

The confluence of our ability to handle big data, significant increases in instrumentation density and quality, and rapid advances in machine learning (ML) algorithms have placed Earth Sciences at the threshold of dramatic progress. ML techniques have been attracting increased attention within the seismic community, and, in particular, in microseismic monitoring where they are now being considered a game-changer due to their real-time processing potential. In our review of the recent developments in microseismic monitoring and characterisation, we find a strong trend in utilising ML methods for enhancing the passive seismic data quality, detecting microseismic events, and locating their hypocenters. Moreover, they are being adopted for advanced event characterisation of induced seismicity, such as source mechanism determination, cluster analysis and forecasting, as well as seismic velocity inversion. These advancements, based on ML, include by-products often ignored in classical methods, like uncertainty analysis and data statistics. In our assessment of future trends in ML utilisation, we also see a strong push toward its application on distributed acoustic sensing (DAS) data and real-time monitoring to handle the large amount of data acquired in these cases.