Using a Deep Neural Network and Transfer Learning to Bridge Scales for Seismic Phase Picking

Abstract

The important task of tracking seismic activity requires both sensitive detection and accurate earthquake location. Approximate earthquake locations can be estimated promptly and automatically; however, accurate locations depend on precise seismic phase picking, which is a laborious and time‐consuming task. We adapted a deep neural network (DNN) phase picker trained on local seismic data to mesoscale hydraulic fracturing experiments. We designed a novel workflow, transfer learning‐aided double‐difference tomography, to overcome the 3 orders of magnitude difference in both spatial and temporal scales between our data and data used to train the original DNN. Only 3,500 seismograms (0.45% of the original DNN data) were needed to retrain the original DNN model successfully. The phase picks obtained with transfer‐learned model are at least as accurate as the analyst's and lead to improved event locations. Moreover, the effort required for picking once the DNN is trained is a small fraction of the analyst's.

Creation of a Mixed‐Mode Fracture Network at Mesoscale Through Hydraulic Fracturing and Shear Stimulation

“…The phyllitic rock mass with strong foliation planes and ubiquitous fractures does however show anisotropic velocity behavior as well as lateral heterogeneity. In order to characterize the heterogeneity, Chai et al (2020) built on our analysis of the seismic events and applied the PhaseNet software (Zhu & Beroza, 2019) and transfer learning to measure additional arrival times and performed a 3‐D tomographic study using the tomoDD package (Zhang & Thurber, 2003). While their relative relocations provide some higher‐resolution images of the identified fractures, the differences to the absolute locations presented here are fairly minor.…”

A Comparative Analysis of Machine Learning Models for First-break Arrival Picking

“…Their study showed that CNN in combination with RNN is more promising for P-and S-detection when there are not enough training data available. To mitigate high intensive labour and thus high cost of manual seismic picking, the study [30] transferred the PhaseNet model and incorporate it with double-difference tomography. The results showed that the model's prediction was nearly as accurate as the result of a human expert with very low time and cost.…”

In-situ observation of pre-, co- and post-seismic shear slip at 1.5 km depth

“…However, additional seismic events could be located close to the injection interval. These were relocated by Chai et al (2020) using tomoDD for joint inversion for relative locations and a 3-D velocity model. Figure 4 shows the seismicity located during the entire test together with our absolute location of the discrete shear event.…”

“…The slip vectors also coincide with the plane defined by the machined notch, indicating that it could have helped to initiate the slip event. (Chai et al, 2020) with an average plane and slip vectors for the shear event measured by the SIMFIP.…”

“…Figure 4: (a-b) Three-dimensional view of SIMFIP assembly, the absolute location of the shear event and relocated seismicity during the injection step. (c) Stereoplot of structures identified on image logs, possible planes derived from hypocenters of relocated seismicity(Chai et al, 2020) with an average…”