GNGTS 2023 - Atti del 41° Convegno Nazionale

Session 1.1 GNGTS 2023 A Workflow for Microseismic Detection Using Deep Learning M. De Solda, F. Grigoli Department of Earth Sciences, University of Pisa, Italy In the last years, the number of dense seismic networks deployed around the world has grown exponentially and will continue to grow in the next years, producing larger and larger datasets. Furthermore, the rising popularity of new technologies for seismic data acquisition such as fiber optics and nodal seismometers, are making seismological datasets grow in size and variety at an exceptionally fast rate, pushing the limit of current data storage infrastructures and, in particular, of data analysis procedures. This data explosion lead seismology to enter in the so-called BIGDATA era. Among the different seismological applications where these massive datasets are usually collected, fluid-induced seismicity studies (both natural and induced) are certainly the most relevant and are a perfect playground for data intensive techniques. In these applications we generally deal with seismic sequences characterized by a large number of weak earthquakes  overlapping each other or with short inter-event times; in these cases, pick-based detection and location methods may struggle to correctly assign picks to phases and events, and errors can lead to missed detections and/or reduced location resolution, which can have significant consequences if real-time seismicity information are used for risk assessment frameworks and, more in general, in the interpretation of the evolution of seismicity in the space-time-magnitude domain. The availability of increasingly powerful machines enabled the use of sophisticated data analysis methods that can digest massive seismicity datasets and have the potential to outperform traditional data analysis algorithms and even experienced seismologists (Mousavi and Beroza 2021). Among the seismological data analysis methods recently developed, waveform-based approaches have gained popularity due to their ability to detect and locate earthquakes without the phase picking and association steps (Li et al 2020). These approaches exploit the information of the entire network (unlike the traditional approaches that use station-wise information for event detection) to simultaneously detect and locate seismic events, producing coherence matrices whose maximum corresponds to the hypocentral coordinates of the seismic event. These methods are particularly powerful at locating microseismic events strongly noise contaminated, and have been successfully applied in induced seismicity analyses. Despite the excellent performance as locators, waveform-based methods still shows sever disadvantages when used as earthquake detectors. Waveform-based earthquake detectors strongly depend on the threshold selected for a certain application. If it is too high, small events may be missed; if it is too low, false events might be detected. To solve this problem, deep learning techniques used for classification of images and/or speech signals can be used within full waveform methods to remove the dependence on threshold levels during the detection process, making these approaches much more robust and