Dr. Jesse Williams will be attending this year’s AGU conference in New Orleans, LA (Dec. 13-17, 2021). Please attend our posters, which will be available both in-person and online.
Towards a Rapid Integrated Associator for Seismic Events
Authors: Louisa Barama, Jesse Williams, Lindsay Yuling Chuang, Zhigang Peng, Andrew Vern Newman
Abstract: It is increasingly important to develop automatic and robust systems that can detect, associate, locate, and discriminate all significant seismic events from available global seismometers. Association can be the most challenging step in seismic monitoring, as signals may include other seismic sources and can be inundated with false detection due to natural or anthropogenic noise. Recently, deep neural networks were used to perform phase association with promising results for local or regional earthquakes. However, their performance on diverse global events and network information is unclear. This study focuses on the development of a Deep Neural Network (DNN) seismic associator, a form of an advanced multilateration algorithm using phase timing as well as anticipated event and path pattern recognition, that can operate during network operations. First, we developed a testing dataset, assembled using a catalog of all earthquakes globally between magnitudes 5.5 and 6.5 and years 2000 and 2016 from the ISC Bulletin and using the ISC analyst P-phase determinations, avoiding events with overlapping arrivals. In order to provide the DNN with as much information as possible, we use an autoencoder to identify the principal components of the seismograms in an unsupervised and structured format, in addition to using direct (manually extracted) features that were selected for their relevance to the seismic detection/association problem. Initial results based solely on P-wave detections are promising, and we expect significant improvements with the utilization of the full feature vector (encoded phases generated from an autoencoder and directly extracted features from the respective phases at multiple stations) from the feature extraction. See more about the poster here.
A supervised autoencoder (SAE) for tele-seismic event distance prediction and waveform compression
Authors: Lindsay Yuling Chuang, Jesse Williams, Louisa Barama, Zhigang Peng, Andrew Vern Newman
Abstract: The success of a deep learning model depends on its ability to solve specific tasks while remaining general. To keep a generic model, one could simply pose regularizations, or strategically integrate auxiliary tasks into model training. Supervised AutoEncoder (SAE) is a type of model that can adopt both approaches. An SAE achieves good performance and generalizability by jointly reconstructing the original inputs and solving tasks on the encoding layer. In this study, we aim to train an SAE to extract features that is informative of tele-seismic event distance while preserving rich information from the original waveforms. We do so by adding a station-event distance predictor on the encoding layer and jointly train the predictor with the autoencoder. We model the training labels as a narrow Gaussian probability distribution centered at the apriori event-station distance. The regression loss of the predictor is then defined as the mean square error between the prediction and the label. We trained the SAE on approximately 150,000 three-component teleseismic P-waves that were recorded by more than 500 broadband stations globally. These P-waves were generated by M5.5 to M6.5 earthquakes between the years 2000 and 2016. Our preliminary results suggest the SAE is capable of capturing low-frequency contents of tele-seismic P-waves with at least 10 times compression without losing its generalizability. The dimension-reduced salient features extracted by the SAE can be further used for a wide range of machine learning applications, such as single station location, seismic event association, and waveform quality control. See more about the poster here.
P-wave first-motion polarity determination using deep learning
Authors: Qiushi Zhai, Zhigang Peng, Roohollah Heidary, Jesse Williams, Savannah Howard, Sheng Dai, Lijun Zhu, Yih-Min Wu
Abstract: Subsurface stress state is the key to understanding natural tectonic environments. It also facilitates an efficient and safe exploration of geothermal extraction and hydraulic fracturing. In many cases, focal mechanisms of microseismic events collected in the target region, which is inferred from their P-wave first-motion polarities, are usually used for rapid and robust estimation of the subsurface stress state. Conventional algorithms for determining polarities remain inferior to human experts, who can focus their attention on a narrow window around the phase arrivals at high resolution while keeping track of the surrounding waveforms at low resolution. Deep-learning models have been recently developed to classify polarities, however, mostly based on convolutional neural networks (CNNs) using a specific regional dataset for both network training and performance validation. In this study, we develop a generic deep learning model to determine the P-wave first-motion polarity. We train the model with global datasets and explore its generalization by testing its ability to adapt to a dataset from an unseen region. Our preliminary results are based on a CNN model trained with a test dataset in Taiwan, which includes 43,119 polarity labels manually assigned by analysts at the Central Weather Bureau Seismic Network of Taiwan. The accuracy of our model on the test dataset is 95%. Using this model, we have built a more complete earthquake focal mechanism catalog in Taiwan from 2009 to 2010, with a seven-fold increase in the number of events compared to the local standard catalog. Our next step is to incorporate the aforementioned attention mechanism into our deep-learning model, and then, train the updated model with a global dataset and examine its generalization. We expect that this tool will be advantageous for determining polarity and focal mechanisms of diverse seismic events in different tectonic settings and even acoustic-emission events in the lab experiments. See more about the poster here.