Search

Double difference earthquake relocation is an essential component of many earthquake catalog development workflows. This technique produces high-resolution relative relocations between events by minimizing differential measurements of the arrival times of waves from nearby sources, which highlights the resolution of faults and improves interpretation of seismic activity. The inverse problem is typically solved iteratively using conjugate-gradient minimization, however the cost scales significantly with the total number of sources and stations considered. Here we propose a Graph Neural Network (GNN) based earthquake double-difference relocation framework, Graph Double Difference (GraphDD), that is trained to minimize the double-difference residuals of a catalog to locate earthquakes. Through batching and sampling the method can scale to arbitrarily large catalogs. Our architecture uses one graph to represent the stations, a second graph to represent the sources, and creates the Cartesian product graph between the two graphs to capture the relationships between the stations and sources (e.g., the residuals and travel time partial derivatives). This key feature allows a natural architecture that can be used to minimize the double-difference residuals. We implement our model on several distinct test cases including seismicity from northern California, Turkiye, and northern Chile, which have highly variable data quality, and station and source distributions. We obtain high resolution relocations in these tests, and our model shows adaptability to variable types of loss functions and location objectives, including learning station corrections and mapping into the reference frame of a different catalog. Our results suggest that a GNN approach to double-difference relocation is a promising direction for scaling to very large catalogs and gaining new insights into the relocation problem.

Seismic phase association connects earthquake arrival time measurements to their causative sources. Effective association must determine the number of discrete events, their location and origin times, and it must differentiate real arrivals from measurement artifacts. The advent of deep learning pickers, which provide high rates of picks from closely overlapping small magnitude earthquakes, motivates revisiting the phase association problem and approaching it using the methods of deep learning. We have developed a Graph Neural Network associator that simultaneously predicts both source space-time localization, and discrete source-arrival association likelihoods. The method is applicable to arbitrary geometry, time-varying seismic networks of hundreds of stations, and is robust to high rates of sources and input picks with variable noise and quality. Our Graph Earthquake Neural Interpretation Engine (GENIE) uses one graph to represent the station set and another to represent the spatial source region. GENIE learns relationships from data in this combined representation that enable it to determine robust source and source-arrival associations. We train on synthetic data, and test our method on real data from the Northern California (NC) seismic network using input generated by the PhaseNet deep learning phase picker. We successfully re-detect ~96% of all events M>1 reported by the USGS during 500 random days between 2000$\unicode{x2013}$2022. Over a 100-day continuous interval of processing in 2017$\unicode{x2013}$2018, we detect ~4.2x the number of events reported by the USGS. Our new events have small magnitude estimates below the magnitude of completeness of the USGS catalog, and are located close to the active faults and quarries in the region. Our results demonstrate that GENIE can effectively solve the association problem under complex seismic monitoring conditions.

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, we developed a cloud-based earthquake monitoring workflow, QuakeFlow, that 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 datasets and to make it easy to deploy on cloud platforms, which enables large-scale parallel processing. We used QuakeFlow to process three years of continuous archived data from Puerto Rico, 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 frequent 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 realtime earthquake monitoring and for mining archived seismic data sets.

We solve the traditional problems of earthquake location and magnitude estimation through a supervised learning approach, where we train a Graph Neural Network to predict estimates directly from input pick data, and each input allows a distinct seismic network with variable number of stations and positions. We train the model using synthetic simulations from assumed travel-time and amplitude-distance attenuation models. The architecture uses one graph to represent the station set, and another to represent the model space. The input includes theoretical predictions of data, given model parameters, and the adjacency matrices of the graphs defined link spatially local elements. As we show, graph convolutions on this combined representation are highly effective at inference, data fusion, and outlier suppression. We compare our results with traditional methods and observe favorable performance.

Earthquake monitoring by seismic networks typically involves a workflow consisting of phase detection/picking, association, and location tasks. In recent years, the accuracy of these individual stages has been improved through the use of machine learning techniques. In this study, we introduce a new, end-to-end approach that improves overall earthquake detection accuracy by jointly optimizing each stage of the detection pipeline. We propose a neural network architecture for the task of multi-station processing of seismic waveforms recorded over a seismic network. This end-to-end architecture consists of three sub-networks: a backbone network that extracts features from raw waveforms, a phase picking sub-network that picks P- and S-wave arrivals based on these features, and an event detection sub-network that aggregates the features from multiple stations and detects earthquakes. We use these sub-networks in conjunction with a shift-and-stack module based on back-projection that introduces kinematic constraints on arrival times, allowing the model to generalize to different velocity models and to variable station geometry in seismic networks. We evaluate our proposed method on the STanford EArthquake Dataset (STEAD) and on the 2019 Ridgecrest, CA earthquake sequence. The results demonstrate that our end-to-end approach can effectively pick P- and S-wave arrivals and achieve earthquake detection accuracy rivaling that of other state-of-the-art approaches.

Earthquake phase association algorithms aggregate picked seismic phases from a network of seismometers into individual earthquakes and play an important role in earthquake monitoring. Dense seismic networks and improved phase picking methods produce massive earthquake phase data sets, particularly for earthquake swarms and aftershocks occurring closely in time and space, making phase association a challenging problem. We present a new association method, the Gaussian Mixture Model Association (GaMMA), that combines the Gaussian mixture model for phase measurements (both time and amplitude), with earthquake location, origin time, and magnitude estimation. We treat earthquake phase association as an unsupervised clustering problem in a probabilistic framework, where each earthquake corresponds to a cluster of P and S phases with hyperbolic moveout of arrival times and a decay of amplitude with distance. We use a multivariate Gaussian distribution to model the collection of phase picks for an event, the mean of which is given by the predicted arrival time and amplitude from the causative event. We carry out the pick assignment for each earthquake and determine earthquake parameters (i.e., earthquake location, origin time, and magnitude) under the maximum likelihood criterion using the Expectation-Maximization (EM) algorithm. The GaMMA method does not require the typical association steps of other algorithms, such as grid-search or supervised training. The results on both synthetic test and the 2019 Ridgecrest earthquake sequence show that GaMMA effectively associates phases from a temporally and spatially dense earthquake sequence while producing useful estimates of earthquake location and magnitude.

Full-waveform inversion (FWI) is an accurate imaging approach for modeling velocity structure by minimizing the misfit between recorded and predicted seismic waveforms. However, the strong non-linearity of FWI resulting from fitting oscillatory waveforms can trap the optimization in local minima. We propose a neural-network-based full waveform inversion method (NNFWI) that integrates deep neural networks with FWI by representing the velocity model with a generative neural network. Neural networks can naturally introduce spatial correlations as regularization to the generated velocity model, which suppresses noise in the gradients and mitigates local minima. The velocity model generated by neural networks is input to the same partial differential equation (PDE) solvers used in conventional FWI. The gradients of both the neural networks and PDEs are calculated using automatic differentiation, which back-propagates gradients through the acoustic PDEs and neural network layers to update the weights of the generative neural network. Experiments on 1D velocity models, the Marmousi model, and the 2004 BP model demonstrate that NNFWI can mitigate local minima, especially for imaging high-contrast features like salt bodies, and significantly improves the inversion in the presence of noise. Adding dropout layers to the neural network model also allows analyzing the uncertainty of the inversion results through Monte Carlo dropout. NNFWI opens a new pathway to combine deep learning and FWI for exploiting both the characteristics of deep neural networks and the high accuracy of PDE solvers. Because NNFWI does not require extra training data and optimization loops, it provides an attractive and straightforward alternative to conventional FWI.

Imaging Earth structure or seismic sources from seismic data involves minimizing a target misfit function, and is commonly solved through gradient-based optimization. The adjoint-state method has been developed to compute the gradient efficiently; however, its implementation can be time-consuming and difficult. We develop a general seismic inversion framework to calculate gradients using reverse-mode automatic differentiation. The central idea is that adjoint-state methods and reverse-mode automatic differentiation are mathematically equivalent. The mapping between numerical PDE simulation and deep learning allows us to build a seismic inverse modeling library, ADSeismic, based on deep learning frameworks, which supports high performance reverse-mode automatic differentiation on CPUs and GPUs. We demonstrate the performance of ADSeismic on inverse problems related to velocity model estimation, rupture imaging, earthquake location, and source time function retrieval. ADSeismic has the potential to solve a wide variety of inverse modeling applications within a unified framework.

We present a deep learning method for single-station earthquake location, which we approach as a regression problem using two separate Bayesian neural networks. We use a multi-task temporal-convolutional neural network to learn epicentral distance and P travel time from 1-minute seismograms. The network estimates epicentral distance and P travel time with absolute mean errors of 0.23 km and 0.03 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 to the station with a mean error of 1 degree. Using this information, we estimate the epicenter, origin time, and depth along with their confidence intervals. We use a global dataset of earthquake signals recorded within 1 degree (~112 km) from the event to build the model and to demonstrate its performance. Our model can predict epicenter, origin time, and depth with mean errors of 7.3 km, 0.4 second, 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 location of earthquakes that are sparsely recorded -- either because they are small or because stations are widely separated.

In this study we develop a single-station deep-learning approach for fast and reliable estimation of earthquake magnitude directly from raw waveforms. We design a regressor composed of convolutional and recurrent neural networks that is not sensitive to the data normalization, hence waveform amplitude information can be utilized during the training. Our network can predict earthquake magnitudes with an average error close to zero and standard deviation of ~0.2 based on single-station waveforms without instrument response correction. We test the network for both local and duration magnitude scales and show a station-based learning can be an effective approach for improving the performance. The proposed approach has a variety of potential applications from routine earthquake monitoring to early warning systems.

Denoising and filtering are widely used in routine seismic-data-processing to improve the signal-to-noise ratio (SNR) of recorded signals and by doing so to improve subsequent analyses. In this paper we develop a new denoising/decomposition method, DeepDenoiser, based on a deep neural network. This network is able to learn simultaneously a sparse representation of data in the time-frequency domain and a non-linear function that maps this representation into masks that decompose input data into a signal of interest and noise (defined as any non-seismic signal). We show that DeepDenoiser achieves impressive denoising of seismic signals even when the signal and noise share a common frequency band. Our method properly handles a variety of colored noise and non-earthquake signals. DeepDenoiser can significantly improve the SNR with minimal changes in the waveform shape of interest, even in presence of high noise levels. We demonstrate the effect of our method on improving earthquake detection. There are clear applications of DeepDenoiser to seismic imaging, micro-seismic monitoring, and preprocessing of ambient noise data. We also note that potential applications of our approach are not limited to these applications or even to earthquake data, and that our approach can be adapted to diverse signals and applications in other settings.

Earthquake signal detection is at the core of observational seismology. A good detection algorithm should be sensitive to small and weak events with a variety of waveform shapes, robust to background noise and non-earthquake signals, and efficient for processing large data volumes. Here, we introduce the Cnn-Rnn Earthquake Detector (CRED), a detector based on deep neural networks. The network uses a combination of convolutional layers and bi-directional long-short-term memory units in a residual structure. It learns the time-frequency characteristics of the dominant phases in an earthquake signal from three component data recorded on a single station. We train the network using 500,000 seismograms (250k associated with tectonic earthquakes and 250k identified as noise) recorded in Northern California and tested it with an F-score of 99.95. The robustness of the trained model with respect to the noise level and non-earthquake signals is shown by applying it to a set of semi-synthetic signals. The model is applied to one month of continuous data recorded at Central Arkansas to demonstrate its efficiency, generalization, and sensitivity. Our model is able to detect more than 700 microearthquakes as small as -1.3 ML induced during hydraulic fracturing far away than the training region. The performance of the model is compared with STA/LTA, template matching, and FAST algorithms. Our results indicate an efficient and reliable performance of CRED. This framework holds great promise in lowering the detection threshold while minimizing false positive detection rates.

In this work, we report on a novel application of Locality Sensitive Hashing (LSH) to seismic data at scale. Based on the high waveform similarity between reoccurring earthquakes, our application identifies potential earthquakes by searching for similar time series segments via LSH. However, a straightforward implementation of this LSH-enabled application has difficulty scaling beyond 3 months of continuous time series data measured at a single seismic station. As a case study of a data-driven science workflow, we illustrate how domain knowledge can be incorporated into the workload to improve both the efficiency and result quality. We describe several end-to-end optimizations of the analysis pipeline from pre-processing to post-processing, which allow the application to scale to time series data measured at multiple seismic stations. Our optimizations enable an over 100$\times$ speedup in the end-to-end analysis pipeline. This improved scalability enabled seismologists to perform seismic analysis on more than ten years of continuous time series data from over ten seismic stations, and has directly enabled the discovery of 597 new earthquakes near the Diablo Canyon nuclear power plant in California and 6123 new earthquakes in New Zealand.

As the number of seismic sensors grows, it is becoming increasingly difficult for analysts to pick seismic phases manually and comprehensively, yet such efforts are fundamental to earthquake monitoring. Despite years of improvements in automatic phase picking, it is difficult to match the performance of experienced analysts. A more subtle issue is that different seismic analysts may pick phases differently, which can introduce bias into earthquake locations. We present a deep-neural-network-based arrival-time picking method called "PhaseNet" that picks the arrival times of both P and S waves. Deep neural networks have recently made rapid progress in feature learning, and with sufficient training, have achieved super-human performance in many applications. PhaseNet uses three-component seismic waveforms as input and generates probability distributions of P arrivals, S arrivals, and noise as output. We engineer PhaseNet such that peaks in probability provide accurate arrival times for both P and S waves, and have the potential to increase the number of S-wave observations dramatically over what is currently available. This will enable both improved locations and improved shear wave velocity models. PhaseNet is trained on the prodigious available data set provided by analyst-labeled P and S arrival times from the Northern California Earthquake Data Center. The dataset we use contains more than seven million waveform samples extracted from over thirty years of earthquake recordings. We demonstrate that PhaseNet achieves much higher picking accuracy and recall rate than existing methods.