Dept. of Geosciences Colloquium: Inverse Methods and Their Applications in Dense Array Seismology

Zoom: https://tau-ac-il.zoom.us/j/87073901428?pwd=Zay3aB8aBEigEH9SC4iEU9pByK9FCN.1

Abstract:

Accurate estimation of seismic source and structural parameters in seismically active regions is a fundamental task in seismology. Traditional seismic measurements are often sparse, and the large number of model parameters influencing seismic wave propagation can lead to non-unique, and sometimes biased, estimations depending on the inversion strategy used. However, recent advances in dense seismic measurements, driven by the broader use of seismic arrays and the introduction of Distributed Acoustic Sensing (DAS) technology, offer new opportunities to create unique and unbiased estimates of both source and structural parameters.

To maximize the data space, waveform-based techniques are employed, which utilize the full range of recorded data. Since the relationship between the observables and the model parameters is nonlinear, Full Waveform Inversion (FWI) is employed as a general solution framework. As a local optimization technique, FWI requires an initial model for both the source and structural parameters.

A joint location-origin time inversion method, designed to work with dense arrays, is developed. This method builds upon the source-scanning algorithm and incorporates both P- and S-waves while considering spatial coherence. The robustness of this method to errors in the velocity model is demonstrated using synthetic data. It is applied to aftershocks recorded by the dense array deployed in Ridgecrest following the 2019 earthquake. A variant of this method is also applied to data from a borehole DAS array, revealing that source locations are not resolvable under reasonable velocity model uncertainty. Further analysis identifies the array geometries needed to achieve source location resolvability.

A waveform-based moment tensor inversion method suitable for unidirectional strain data is developed and applied to microseismic data collected as part of the HFTS-2 project. Despite the unidirectional nature of the measurements, a well-resolved moment tensor is successfully estimated, with low uncertainty for all components, provided that the signal-to-noise ratio (SNR) is maintained above a reasonable threshold. An analysis is performed to assess the sensitivity of each moment tensor component to noise and to evaluate the contribution of different parts of the array in resolving each component.
Existing FWI algorithms are adapted for the joint inversion of both source and structural parameters. A normalized envelope misfit is introduced to enable the incorporation of seismic amplitudes into the inversion process. This approach expands the data space while ensuring that the misfit remains more linearly related to the model parameters. Synthetic tests show that this modification is essential for accurately resolving source locations. Furthermore, a modification of the commonly used L-BFGS optimization algorithm is made to improve pre-conditioning by two orders of magnitude, thereby enabling effective joint inversion of parameters with different physical natures.
 

Event Organizer: Dr. Ariel Lellouch