Deep Learning
of
Seismograms
Seismology is the study of seismic waves to understand the sources of those waves - such as earthquakes, explosions, volcanic eruptions, glaciers, landslides, ocean waves, vehicular traffic, aircraft, trains, wind, air guns, and thunderstorms- and to infer the structure and properties of planetary interiors. The development of increasingly cost-effective sensors and emerging ground-motion sensing technologies, such as fiber optic cable and accelerometers in smart devices, portend a continuing acceleration of seismological data volumes, such that deep learning is likely to become essential to seismology’s future. The availability of large-scale seismic datasets and the suitability of deep-learning techniques for seismic data processing have pushed deep learning to the forefront of fundamental, long-standing research investigations in seismology. My works mainly focus on demystifying deep-learning techniques and their potential use within earthquake seismology. My contributions span from proof-of-concept research all the way to the implementation of production-ready technologies.
Related Papers
Reviews and Opinion Papers
-
Mousavi, S. M., and Beroza, G. C., (2022). Deep-Learning Seismology, Science.
Seismic waves from earthquakes and other sources are used to infer the structure and properties of Earth’s interior. The availability of large-scale seismic datasets and the suitability of deep-learning techniques for seismic data processing have pushed deep learning to the forefront of fundamental, long-standing research investigations in seismology. However, some aspects of applying deep learning to seismology are likely to prove instructive for geosciences, and perhaps other research areas more broadly. Deep learning is a powerful approach, but there are subtleties and nuances in its application. We present a systematic overview of trends, challenges, and opportunities in applications of deep-learning methods in seismology.
-
Mousavi, S. M., and Beroza, G.C., (2023). Machine Learning in Earthquake Seismology, Annual Review of Earth and Planetary Sciences.
ML methods are becoming the dominant approaches for many tasks in seismology. ML and data mining techniques can significantly improve our capability for seismic data processing. In this review, we provide a comprehensive overview of ML applications in earthquake seismology, discuss progress and challenges, and offer suggestions for future work.
▪ Conceptual, algorithmic, and computational advances have enabled rapid progress in the development of machine-learning approaches to earthquake seismology.
▪ The impact of that progress is most clearly evident in earthquake monitoring and is leading to a new generation of much more comprehensive earthquake catalogs.
▪ The application of unsupervised approaches for exploratory analysis of these high-dimensional catalogs may reveal a new understanding of seismicity.
▪ Machine learning methods are proving to be effective across a broad range of other seismological tasks, but systematic benchmarking through open-source frameworks and benchmark data sets is important to ensure continuing progress.
-
Sun, Z., Sandoval, L., Crystal-Ornelas, Mousavi, S. M., et al., (2022). A review of Earth Artificial Intelligence, Computers and Geosciences.
In recent years, Earth system sciences are urgently calling for innovation on improving accuracy, enhancing model intelligence level, scaling up operations, and reducing costs in many subdomains amid the exponentially accumulated datasets and the promising artificial intelligence (AI) revolution in computer science. This paper presents work led by the NASA Earth Science Data Systems Working Groups and ESIP machine learning cluster to give a comprehensive overview of AI in Earth sciences. It holistically introduces the current status, technology, use cases, challenges, and opportunities, and provides all the levels of AI practitioners in geosciences with an overall big picture and to “blow away the fog to get a clearer vision” about the future development of Earth AI. The paper covers all the major spheres in the Earth system and investigates representative AI research in each domain. Widely used AI algorithms and computing cyberinfrastructure are briefly introduced. The mandatory steps in a typical workflow of specializing AI to solve Earth's scientific problems are decomposed and analyzed. Eventually, it concludes with the grand challenges and reveals the opportunities to give some guidance and pre-warnings on allocating resources wisely to achieve the ambitious Earth AI goals in the future.
-
Beroza, G.C., Segou, M., and Mousavi, S. M., (2021). Machine learning and earthquake forecasting—next steps, Nature Communication.
A new generation of earthquake catalogs developed through supervised machine learning illuminates earthquake activity with unprecedented detail. The application of unsupervised machine learning to analyze the more complete expression of seismicity in these catalogs may be the fastest route to improving earthquake forecasting.
Technique Developments
-
Zhu, W., Hou, AB., Yang, R., Mousavi, S. M., Ellsworth, W., and Beroza, G. C., (2022). QuakeFlow: A Scalable Machine-learning-based Earthquake Monitoring Workflow with Cloud Computing, Geophysical Journal International.
Earthquake monitoring workflows are designed to detect earthquake signals and to determine source characteristics from continuous waveform data. Recent developments in deep learning seismology have been used to improve tasks within earthquake monitoring workflows that allow the fast and accurate detection of up to orders of magnitude more small events than are present in conventional catalogs. To facilitate the application of machine-learning algorithms to large-volume seismic records at scale, we developed a cloud-based earthquake monitoring workflow, QuakeFlow, which applies multiple processing steps to generate earthquake catalogs from raw seismic data. QuakeFlow uses a deep learning model, PhaseNet, for picking P/S phases and a machine learning model, GaMMA, for phase association with approximate earthquake location and magnitude. Each component in QuakeFlow is containerized, allowing straightforward updates to the pipeline with new deep learning/machine learning models, as well as the ability to add new components, such as earthquake relocation algorithms. We built QuakeFlow in Kubernetes to make it auto-scale for large data sets and to make it easy to deploy on cloud platforms, which enables large-scale parallel processing. We used QuakeFlow to process three years of continuously archived data from Puerto Rico within a few hours and found more than a factor of ten more events that occurred on much the same structures as previously known seismicity. We applied Quakeflow to monitoring earthquakes in Hawaii and found over an order of magnitude more events than are in the standard catalog, including many events that illuminate the deep structure of the magmatic system. We also added Kafka and Spark streaming to deliver real-time earthquake monitoring results. QuakeFlow is an effective and efficient approach both for improving real-time earthquake monitoring and for mining archived seismic data sets.
-
Zhu, W., McBrearty, I. W., Mousavi, S. M., Ellsworth, W.L., and Beroza, G. C., (2022). Earthquake Phase Association using a Bayesian Gaussian Mixture Model, Journal of Geophysical Research.
Earthquakes are monitored by seismic networks consisting of several to hundreds of seismometers. An earthquake detection workflow usually has two important steps: detecting/picking seismic phases at each seismometer and associating picked phases across multiple seismometers in a network. Deep-learning-based phase pickers have greatly improved phase detection performance and can automatically generate many more seismic phases than conventional algorithms. These massive numbers of automatic phase pick pose a challenge for the phase association task. We have developed a new phase association method using a Bayesian Gaussian Mixture Model. We treat the phase association problem as an unsupervised clustering problem meaning that we aim to cluster detected phases into different groups based on individual earthquakes that produce these phases. The Gaussian mixture model makes it easy to consider multiple phase parameters, such as phase arrival time, phase-amplitude, phase picking quality score, and phase type, to improve phase association. We test our method on both synthetic data and the 2019 Ridgecrest earthquake. The results show that our method can effectively associate phases from a temporally and spatially dense earthquake sequence and generate a more complete earthquake catalog than catalogs created using conventional methods.
-
Zhu, W., Tai, K. S., Mousavi, S. M., Bailis, P., and Beroza, G. C., (2022). An End-to-End Earthquake Detection Method for Joint Phase Picking and Association using Deep Learning, Journal of Geophysical Research.
-
Mousavi, S. M., Ellsworth, W.L., Zhu, W., and Beroza, G.C., (2020). Earthquake Transformer: An Attentive Deep-learning Model for Simultaneous Earthquake Detection and Phase Picking, Nature Communication.
Earthquakes are monitored using multiple seismic stations in a seismic network. A typical earthquake detection workflow consists of two stages: first, earthquake signals are detected at each station; then these detections are associated across multiple stations to determine whether an earthquake occurred. These two stages are independently optimized in conventional algorithms, thus the overall earthquake detection performance can be limited due to information loss between each stage. In this work, we proposed an end-to-end approach to jointly optimize both stages (i.e., signal-station detection and multi-station association) inside one combined neural network architecture. The results of the 2019 Ridgecrest, CA earthquake sequence show that our end-to-end approach achieves earthquake detection accuracy rivaling that of other state-of-the-art approaches. Because our approach preserves information across tasks in the detection pipeline, it has the potential to outperform approaches that do not.
Earthquake signal detection and seismic phase picking are challenging tasks in the processing of noisy data and the monitoring of microearthquakes. Here we present a global deep-learning model for simultaneous earthquake detection and phase picking. Performing the two related tasks in tandem improves model performance in each individual task by combining information in phases and in the full waveform of earthquake signals by using an ahierarchical attention mechanism. We show that our model outperforms previous deep-learning and traditional phase-picking and detection algorithms. Applying our model to5 weeks of continuous data recorded during the 2000 Tottori earthquakes in Japan, we were able to detect and locate two times more earthquakes using only a portion (less than 1/3) of seismic stations. Our model picks P and S phases with precision close to manual picks by human analysts; however, its high efficiency and higher sensitivity can result in detecting and characterizing more and smaller events.
-
Mousavi, S. M., and Beroza, G.C., (2020). Bayesian-Deep-Learning Estimation of Local Earthquake Location from Single-Station Observations, IEEE Transactions on Geoscience and Remote Sensing.
We present a deep-learning method for a single-station earthquake location, which we approach as a regression problem using two separate Bayesian neural networks. We use a multitask temporal convolutional neural network to learn epicentral distance and P travel time from 1-min seismograms. The network estimates epicentral distance and P travel time with mean errors of 0.23 km and 0.03 s and standard deviations of 5.42 km and 0.66 s, respectively, along with their epistemic and aleatory uncertainties. We design a separate multi-input network using standard convolutional layers to estimate the back-azimuth angle and its epistemic uncertainty. This network estimates the direction from which seismic waves arrive at the station with a mean error of 1°. Using this information, we estimate the epicenter, origin time, and depth along with their confidence intervals. We use a global data set of earthquake signals recorded within 1° (~112 km) from the event to build the model and demonstrate its performance. Our model can predict epicenter, origin time, and depth with mean errors of 7.3 km, 0.4 s, and 6.7 km, respectively, at different locations around the world. Our approach can be used for fast earthquake source characterization with a limited number of observations and also for estimating the location of earthquakes that are sparsely recorded-either because they are small or because stations are widely separated.
-
Zhu, W., Mousavi, S. M., and Beroza, G. C., (2020). Seismic Signal Augmentation to Improve Generalization of Deep Neural Networks, Advances in Geophysics.
Deep learning has emerged as an effective approach for seismic data processing in general, and for earthquake monitoring in particular. The ability of deep learning models to generalize beyond the training and validation data is important for comprehensive earthquake monitoring; this ability furthermore depends on the availability of a sufficiently large and complete training dataset. However, this requirement can prove challenging to meet due to significant effort and time for data collection and labeling. Data augmentation provides an efficient and effective approach for increasing the dimension of training samples and improving generalization to unseen samples. In this paper, we present augmentation methods appropriate for seismic waveforms and demonstrate their ability to reduce bias and increase performance. These augmentation methods can be applied to a wide range of deep-learning applications designed for seismic data.
-
Mousavi, S. M., and Beroza, G.C., (2019), A Machine-Learning Approach for Earthquake Magnitude Estimation, Geophysical Research Letters.
Deep learning has emerged as an effective approach for seismic data processing in general, and for earthquake monitoring in particular. The ability of deep learning models to generalize beyond the training and validation data is important for comprehensive earthquake monitoring; this ability furthermore depends on the availability of a sufficiently large and complete training dataset. However, this requirement can prove challenging to meet due to significant effort and time for data collection and labeling. Data augmentation provides an efficient and effective approach for increasing the dimension of training samples and improving generalization to unseen samples. In this paper, we present augmentation methods appropriate for seismic waveforms and demonstrate their ability to reduce bias and increase performance. These augmentation methods can be applied to a wide range of deep-learning applications designed for seismic data.
-
Mousavi, S. M., Sheng, Y., Zhu, W., and Beroza, G.C., (2019), STanford Earthquake Dataset (STEAD): A Global Data Set of Seismic Signal for AI, IEEE Access.
Deep learning has emerged as an effective approach for seismic data processing in general, and for earthquake monitoring in particular. The ability of deep learning models to generalize beyond the training and validation data is important for comprehensive earthquake monitoring; this ability furthermore depends on the availability of a sufficiently large and complete training dataset. However, this requirement can prove challenging to meet due to significant effort and time for data collection and labeling. Data augmentation provides an efficient and effective approach for increasing the dimension of training samples and improving generalization to unseen samples. In this paper, we present augmentation methods appropriate for seismic waveforms and demonstrate their ability to reduce bias and increase performance. These augmentation methods can be applied to a wide range of deep-learning applications designed for seismic data.
-
Mousavi, S. M., and Beroza, G. C., (2019). Unsupervised Clustering of Seismic Signals Using Deep Convolutional Autoencoders, IEEE Geoscience, and Remote Sensing Letters
Deep learning has emerged as an effective approach for seismic data processing in general, and for earthquake monitoring in particular. The ability of deep learning models to generalize beyond the training and validation data is important for comprehensive earthquake monitoring; this ability furthermore depends on the availability of a sufficiently large and complete training dataset. However, this requirement can prove challenging to meet due to significant effort and time for data collection and labeling. Data augmentation provides an efficient and effective approach for increasing the dimension of training samples and improving generalization to unseen samples. In this paper, we present augmentation methods appropriate for seismic waveforms and demonstrate their ability to reduce bias and increase performance. These augmentation methods can be applied to a wide range of deep-learning applications designed for seismic data.
-
Mousavi, S. M., Zhu, W., Sheng, Y., Beroza, G. C., (2018). CRED: A deep residual network of convolutional and recurrent units for earthquake signal detection, Scientific Report.
Deep learning has emerged as an effective approach for seismic data processing in general, and for earthquake monitoring in particular. The ability of deep learning models to generalize beyond the training and validation data is important for comprehensive earthquake monitoring; this ability furthermore depends on the availability of a sufficiently large and complete training dataset. However, this requirement can prove challenging to meet due to significant effort and time for data collection and labeling. Data augmentation provides an efficient and effective approach for increasing the dimension of training samples and improving generalization to unseen samples. In this paper, we present augmentation methods appropriate for seismic waveforms and demonstrate their ability to reduce bias and increase performance. These augmentation methods can be applied to a wide range of deep-learning applications designed for seismic data.
-
Mousavi, S. M., Horton, S. P., Langston, C. A., Samei, B., (2016). Seismic Features and Automatic Discrimination of Deep and Shallow Induced-Microearthquakes Using Neural Network and Logistic Regression, Geophysical Journal International.
Deep learning has emerged as an effective approach for seismic data processing in general, and for earthquake monitoring in particular. The ability of deep learning models to generalize beyond the training and validation data is important for comprehensive earthquake monitoring; this ability furthermore depends on the availability of a sufficiently large and complete training dataset. However, this requirement can prove challenging to meet due to significant effort and time for data collection and labeling. Data augmentation provides an efficient and effective approach for increasing the dimension of training samples and improving generalization to unseen samples. In this paper, we present augmentation methods appropriate for seismic waveforms and demonstrate their ability to reduce bias and increase performance. These augmentation methods can be applied to a wide range of deep learning applications designed for seismic data.
Applications
-
Mohammadigheymasi, H., Tavakolizadeh, N., Matias, L., Mousavi, S. M., Silveira, G., Custódio, S., and Moradichaloshtori, Y. (2022). Application of deep learning for seismicity analysis in Ghana, Geosystems and Geoenvironment.
We present the characterization of regional seismicity in Ghana by processing the Ghana Digital Seismic Network (GHDSN) data set recorded between September 2012 and April 2014, implementing deep learning (DL). Local earthquakes are detected in this dataset using EQTransformer, a DL model with a hierarchical attentive mechanism (HAM) for simultaneous earthquake detection and P- and S-phase picking. A Conservative Strategy (CS) is devised to detect the missing phases and to associate the detected phases to circumvent the false-negative issue of EQTrans-former processing low signal-to-noise ratio (SNR) seismograms. We performed a joint inversion by grid search in 1D velocity model space and simultaneous inversion for the hypocentral parameters, incorporating 559 detected arrival times (292 P and 267 S phases). The results obtained by velocity inversion contain thicknesses of 1, 13, 8, 13, and 10 km, from the surface to a depth of 45 km, with Vp = 5.9, 6.1, 6.3, 6.5, 6.9, and 7.2 km/s, respectively. The updated velocities for the first and last layers are 6% and 26% and the Vp/Vs (=1.70) is 3.03% higher than the previously reported values. A total number of 73 earthquakes with a local magnitude of 2.5 < Ml < 3.9 are located, comprising four main clusters of events, showing a high correlation with the mapped fault zones. The hypocentral depth distribution is mainly in the range of 7-15 km, confined to the upper crust in the region. No specific seismic activity in the eastern branch of the Coastal Boundary Fault (CBF) and the continuation of the Romanche Fracture Zone (RFZ) in the study period was observed, casting further doubt on the activity of this branch and the hypothesis of stress transfer from RFZ to southern Ghana. The results reinforce the intraplate nature of the tectonic activities in the region. Finally, an updated seismic catalog up to April 2022 is presented for Ghana by incorporating all reported catalogs and combining the newly detected events.
-
Park, Y., Mousavi, S. M., Zhu, W., Ellsworth, W.L., and Beroza, G.C., (2020). Machine learning based analysis of the Guy-Greenbrier, Arkansas earthquakes: a tale of two sequences, Geophysical Research Letters.
Finding small earthquake signals from long duration continuous seismic data is a time-consuming task, but machine learning algorithms have the potential to accelerate the workflow and improve the results. We reprocessed the seismic data from the area spanning Guy and Greenbrier in central Arkansas in 2010 and 2011 using a machine learning algorithm to reexamine this well-studied earthquake sequence, which is thought to be caused by injection of wastewater from unconventional hydrocarbon production into deep disposal wells. Even using conservative postprocessing steps, we were able to locate nearly 90,000 earthquake events. The improved catalog illuminates previously unseen aspects of this earthquake sequence that give new insights into its behavior.