Improving earthquake detection and localization with deep learning

Project Description The growing amount of seismic data necessitates efficient and effective methods to monitor earthquakes. Current methods are computationally expensive, ineffective under noisy environments, or labor intensive. We will leverage advances in deep learning to develop an improved solution, ArrayConvNet—a convolutional neural network that uses continuous array data from a seismic network to seamlessly detect and localize events, without the intermediate steps of phase detection, association, travel-time calculation, and inversion. In our initial work testing this methodology with events at Hawai‘i, we achieve 99.4% accuracy and predict hypocenter locations within a few kilometers of the U.S. Geological Survey catalog. We demonstrate that training with relocated earthquakes reduces localization errors significantly. Application to continuous records shows that our algorithm detects 690% as many earthquakes as the published catalog. To further improve the deep learning model, we will include enhanced data augmentation, use of relocated offshore earthquakes recorded by ocean-bottom seismometers, test and apply the model to other tectonically active regions (e.g., Alaska and southern California). Because of the enhanced detection sensitivity, localization granularity, and minimal computation costs, our solution is valuable, particularly for real-time earthquake monitoring.

To date, the development of the model has been carried out on a workstation. To bring the ArrayConvNet model close to the practical and operational levels, the training of the deep learning model may involve millions of seismic events and a huge amount of continuous data. The goal of this CAREERS project is to develop this workflow for execution in an HPC environment like UMass-URI UNITY located at the MGHPCC.