Earthquake Location Imaging (ELI) for single-well Distributed Acoustic Sensing using Wavefield Classification

This study investigates an innovative approach to the earthquake location problem by simulating the use of Distributed Acoustic Sensing (DAS) technology deployed in a single vertical borehole. Traditional methods typically rely on extensive networks of seismometers distributed horizontally on the surface to accurately determine earthquake hypocenters. In contrast, this work examines the feasibility of deriving earthquake locations from DAS seismogram images recorded by 700 virtual receivers along a 3.5 km vertical cable in a well.

We evaluated multiple methodologies, including cross-correlation-based matching with a database of synthetic waveforms and advanced machine learning (ML) techniques such as convolutional neural networks (CNNs) and autoencoders. While the cross-correlation approach produced promising results for simple velocity models, it faced limitations when applied to more complex, realistic subsurface structures. To overcome these challenges, CNNs were employed to classify earthquake locations within a grid framework, and autoencoders were utilized to enhance the resolution of derived location images. The methodology was tested against two benchmark velocity models: the Marmousi model and a region-specific model derived from seismic exploration at a geothermal energy site.

Our findings highlight the potential of integrating DAS technology with ML for earthquake location imaging, particularly in environments with sparse seismic instrumentation. Our approach demonstrates promise in improving the efficiency and accuracy of seismic monitoring, especially in regions characterized by lateral velocity heterogeneities.