Joint location of microseismic events in the presence of velocity uncertainty

Abstract

The locations of seismic events are used to infer reservoir properties and to guide future production activity, as well as to determine and understand the stress field. Thus, locating seismic events with uncertainty quantification remains an important problem. Using Bayesian analysis, a joint probability density function of all event locations was constructed from prior information about picking errors in kinematic data and explicitly quantified velocity model uncertainty. Simultaneous location of all seismic events captured the absolute event locations and the relative locations of some events with respect to others, along with their associated uncertainties. We found that the influence of an uncertain velocity model on location uncertainty under many realistic scenarios can be significantly reduced by jointly locating events. Many quantities of interest that are estimated from multiple event locations, such as fault sizes and fracture spacing or orientation, can be better estimated in practice using the proposed approach.

Fast GPU-Based Seismogram Simulation From Microseismic Events in Marine Environments Using Heterogeneous Velocity Models

“…This is even more justified in the presence of complicated noise characteristics in the receivers in a real field survey. Previously Poliannikov et al [52]- [54] used a travel time based Bayesian inference, although the influence of background seismic noise on the travel time calculation may not have a simple linear and Gaussian nature. However this condition can be relaxed if the Bayesian inference can be framed rather in the real measurement i.e.…”

What moved where?: The impact of velocity uncertainty on microseismic location and moment-tensor inversion

“…We formulate the problem by defining a likelihood function, p (t|x,v), which is read as "the likelihood of observing arrival time t, given the input distributions on x and v." This is a modeled range of arrival times for a given range of locations and velocities. In other words, we calculate p (t|x,v) by computing the arrival times that would be recorded for a microseism at x with velocity model v. For simplicity, we are fixing the origin time; it can be incorporated easily (Poliannikov et al, 2014) but adds to the notational clutter. From this, we would then like to compute the posterior, p(x|t), or the probability of the event being located at x given the distributions of arrival times with the effects of the velocity uncertainty folded in.…”

“…Here, we apply this methodology to determine what the uncertainties are on both locations and focal mechanisms for given uncertainties in velocity. The location part builds on our past work (Poliannikov et al, 2011a(Poliannikov et al, , 2011b(Poliannikov et al, , 2012(Poliannikov et al, , 2014(Poliannikov et al, , 2016Poliannikov and Malcolm, 2016), which has focused primarily on the location problem and highlights how different location algorithms perform under different assumptions on the uncertainties in both arrival times and velocity. That work shares some similarities with that of Eisner et al (2009).…”

Bayesian inversion of pressure diffusivity from microseismicity

“…Microseismic event locations and origin times are not observed directly. Instead, they are estimated with some uncertainty from recorded microseismic data (Michaud et al, 2004;Bennett et al, 2005;Huang et al, 2006;Poliannikov et al, 2013Poliannikov et al, , 2014.…”