Towards fast machine-learning-assisted Bayesian posterior inference of microseismic event location and source mechanism

Aug 4, 2026

Bayesian inference for microseismic monitoring enables accurate event location from recorded seismograms and quantifies associated uncertainties. The computational cost of forward modelling required for Bayesian source inversion can be prohibitive, motivating surrogate modelling. This work trains a machine-learning emulator on the power spectrum of recorded pressure waves to accelerate inference for arbitrary source mechanisms, achieving complete fast event locations in under 1 hour on a laptop using fewer than 10^4 training seismograms. It also infers source mechanisms via Bayesian evidence and remains robust under realistic field noise.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Summary

Introduction

Bayesian inversion and computational bottlenecks

Machine-learning emulator approach

Evidence-based source mechanism estimation

Robustness to field noise

Conclusions and outlook

Geophys. J. Int. (2023) 232, 1219–1235 [https://doi.org/10.1093/gji/ggac385](https://doi.org/10.1093/gji/ggac385) Advance Access publication 2022 October 06 GJI Seismology Towards fast machine-learning-assisted Bayesian posterior inference of microseismic event location and source mechanism D. Piras , 1 A. Spurio Mancini , 1,2,3 A. M. G. Ferreira,4 B. Joachimi 1 and M. P. Hobson3 1 Department of Physics and Astronomy, University College London, Gower Street, London WC1E 6BT, [UK. E-mail: dr.davide.piras@gmail.com](UK. E-mail: dr.davide.piras@gmail.com) [2](2 Mullard Space Science Laboratory)[ Mullard Space Science Laboratory](2 Mullard Space Science Laboratory), [University College London](University College London), [Holmbury St](Holmbury St). Mary, Dorking, Surrey RH5 6NT, UK 3 Astrophysics Group, Cavendish Laboratory, J. J. Thomson Avenue, Cambridge CB3 0HE, UK 4 Deptartment of Earth Sciences, Faculty of Mathematical & Physical Sciences, University College London, London WC 1E 6BT, UK Accepted 2022 October 3 . Received 2022 August 8; in original form 2021 January 14 SUMMARY Bayesian inference applied to microseismic activity monitoring allows the accurate location of microseismic events from recorded seismograms and the estimation of the associated uncertainties. However, the forward modelling of these microseismic events, which is necessary to perform Bayesian source inversion, can be prohibitively expensive in terms of computational resources. A viable solution is to train a surrogate model based on machine learning techniques to emulate the forward model and thus accelerate Bayesian inference. In this paper, we substantially enhance previous work, which considered only sources with isotropic moment tensors. We train a machine learning algorithm on the power spectrum of the recorded pressure wave and show that the trained emulator allows complete and fast event locations for any source mechanism. Moreover, we show that our approach is computationally inexpensive, as it can be run in less than 1 hr on a commercial laptop, while yielding accurate results using less than 104 training seismograms. We additionally demonstrate how the trained emulators can be used to identify the source mechanism through the estimation of the Bayesian evidence. Finally, we demonstrate that our approach is robust to real noise as measured in ﬁeld data. This work lays the foundations for efﬁcient, accurate future joint determinations of event location and moment tensor, and associated uncertainties, which are ultimately key for accurately characterizing human-induced and natural earthquakes, and for enhanced quantitative seismic hazard assessments. Key words: Machine learning; Statistical methods; Induced seismicity; Waveform inversion. 1 INTRODUCTION Underground human activity, including ﬂuid injection in rocks and mining operations, can cause microseismic events (Majer et al. 2007; Ellsworth 2013) . The monitoring of both human-induced and natural microseismicity is critical for understanding seismic hazard (Brueckl et al. 2008; Shapiro et al. 2010; Mukuhira et al. 2016; Das et al. 2017, and references therein) . Accurate seismic event locations in space and in time are of paramount importance for reliable seismic monitoring efforts, and are mainly obtained from seismograms recorded on land and/or at the seaﬂoor. Various methods for locating seismic events are available in the literature, dating back to the work of Geiger (1910), and up to today (see e.g. Vasco et al. 2019, and references therein, for a recent review) . One of the most common approaches relies on using the Eikonal equation to determine the theoretical travel time of ﬁrst seismic arrivals (see e.g. Noack & Clark 2017; Smith et al. 2020), which is compared with real travel times through a direct grid search or more sophisticated inverse modelling techniques (Wuestefeld et al. 2018) . More accurate source location estimations can be obtained exploit

FAQs

What problem does the work address in Bayesian microseismic inversion?

The forward modelling needed for Bayesian source inversion is computationally expensive, which makes traditional approaches difficult to apply efficiently.

How is the surrogate model constructed and what does it emulate?

A machine-learning emulator is trained on the power spectrum of recorded pressure waves to emulate the forward model and produce fast event locations for any source mechanism.

How can the source mechanism be estimated in this framework?

The trained emulators are used to estimate Bayesian evidence, enabling identification of the source mechanism in addition to event location.

Aug 4, 2026

Upload by Lucas Martin