Source Location Estimation and Uncertainty Analysis for Induced Earthquakes Using Large-N Seismic Arrays: A Comparison of 3D Traveltime, Kirchhoff, and Wave Equation Methods

Abstract:

Reactivation of faults by water injection and the increased occurrence of M_w ≥ 3 earthquakes has raised public concern in recent years. Accurate methods for estimating earthquake locations with quantitative uncertainty analysis is important to help distinguish induced from natural seismicity. Traveltime and Kirchhoff methods are the primary techniques to locate the earthquake sources, but the event location uncertainties are sometimes very large even when the velocity model is known accurately. Wave equation methods have the potential to improve event location accuracy and resolution, but require large-N (dense) receiver array deployment.

We present the Traveltime and Kirchhoff methods, and compare to our proposed Wave equation approach to estimate earthquake location uncertainties. For the Traveltime inversion, we use a probabilistic objective function with the Oct-tree optimization algorithm to compute the event location. The location error can be large for a deep basement event because a small traveltime error causes large uncertainty in high velocity zones. The Kirchhoff method does not require traveltime picking, but instead requires a time window that contains the first arrival waveforms used to back-project along wavefront surfaces. The imaging result shows a distribution of coherent source energy that is used to estimate the location uncertainty. Wave equation methods do not use the arrival traveltimes, or partial waveforms, but sum over the full propagated wavefield. This is helpful in the presence of strong data noise. We present an imaging condition that uses the full elastic wavefield to focus the P and S wave energy at the source location with associated uncertainty estimates.

We use two data examples to compare the three methods. First, we perform a synthetic experiment using 3D elastic microseismic data with the SEAM4D reservoir model. Second, we use the LArge-n Seismic Survey in Oklahoma (LASSO) and recordings of injection-induced earthquakes to estimate event locations. We select two M_w 3.0 earthquakes that occurred in Grant County, north Oklahoma. The data is recorded by 1,826 seismic nodes deployed in a 25 km × 32 km area. Our results show that Wave equation methods can significantly reduce event location uncertainty with increased resolution compared to the Traveltime and Kirchhoff methods.