Data-Driven Earthquake Detection, Localization and Source Mechanism Estimation Based on Wavefield Extrapolation and 2D Deconvolution in High-Noise Environments

Date: 4/25/2019

Time: 06:00 PM

Room: Fifth Avenue

Near real-time detection of weak induced earthquakes is a topic of active research in the field of seismic monitoring with the purpose of reducing the seismic risk and increasing the productivity of a reservoir. With low signal-to-noise ratios standard earthquake detection methods that work on a single trace time signal fail to detect earthquakes. Over the past few decades seismologists started to use data-driven methods to detect and image earthquakes recorded with spatially Nyquist-sampled arrays. By exploiting the time-invariance of the wave equation in a lossless medium, the wavefields can be extrapolated backward towards the source they originated from, thus increasing the signal-to-noise ratio, which results in an increased detection threshold.

In this study the wavefields are backward extrapolated in depth with the weighted-least squares one-way acoustic wavefield extrapolation operator in the space-frequency domain. These operators were designed to work efficiently and with high accuracy in heterogeneous media. Ideally, the amplitude of the extrapolated wavefields reaches a maximum at the location of the source, which can be used as an imaging condition to determine the hypocenter and origin time. However, the polarity changes in the P- and S-phases along the array lead to destructive interference of the extrapolated data at the source location. We propose to maximize the focused signal by computing the 2D deconvolution between the extrapolated data and a pre-computed “ideal” filter. This filter is created from a synthetic wavefield measured at the surface for a specific source mechanism that is backward extrapolated to the source location. Since the source mechanism is unknown, a database of such filters for different source mechanisms, given the P- and S-wave velocity model and the acquisition geometry as input is pre-computed. The maximum amplitude is reached when the source mechanism of the data optimally matches the filter’s source mechanism. With this procedure the hypocenter, source origin time and source mechanism can be estimated simultaneously.

Presenting Author: Nicolas A. Vinard

Authors