Full 4D Trans-dimensional Local Earthquake Tomography

Seismic tomography has been routinely applied to seismic data to construct 3D models of the subsurface elastic properties, through the solution of a linearized inverse problem [Ulm and Thurber, BSSA, 1987]. In previous attempts to determine temporal variations of the elastic properties of the crust, they have been inferred via simple repetition of a 3D seismic tomography on ensembles of events occurring in different time-windows [Calò et al., GJI, 2011]. The main drawbacks of the 3D linearized approach are: (1) the “not-adaptive” characteristic of such inversion schemes, i.e. a fixed subsurface parameterization, (2) a minimum number of rays needed to reach numerically stable results and (3) a “starting model” may affect the solution. Moreover, strong seismic-velocity variations tend to be suppressed by linearization and regularization.

In this project, we developed a new tool for a full 4D local earthquake tomography based on a reversible jump Markov chain Monte-Carlo (RjMcMC) algorithm [Gallagher et al., Marine and Petroleum Geology, 2009]. This approach avoids the pitfalls of classical 3D seismic tomography, and the extension from 3D to full 4D tomography is straightforward. The new algorithm has been applied (initially, in 3D) to the natural seismicity occurred in an oil field in Italy, to test the reliability of the proposed approach to LET. Subsequently, trans-D LET has been applied to Ireland, to improve our knowledge of the subsurface structure in the geothermally promising Donegal Granite. Application of the 4D LET to the human-induced seismicity recorded at the Soultz-sous–Forets EGS site (Rhine Graben, France), has been presented, to improve our knowledge of the fluid-rocks interaction during micro-earthquake nucleation.

**Comparison of differences between the RJMCMC and SimulPs inversions** at 3 km depth. (a) Difference between the posterior mean of P-wave velocity from RJMCMC sampling and the velocity inverted by SimulPs. (b) Absolute velocity differences in (a) divided by the posterior standard deviation of P-wave velocity from RJMCMC sampling, showing that most differences are within one posterior standard deviation (from Piana Agostinetti et al., GJI, 2015).

ACKNOWLEDGMENTS

NPA thanks Alberto Malinverno and Marco Calo' for their fundamental contributions to the development and testing of the 4D LET algorithm. NPA is deeply in debt with Daniele Melini at INGV-Rome for his continuous (24/7!) support at the INGV HPC facilities.

PUBLICATIONS and MEETING ABSTRACTS

**Piana Agostinetti, N.** (2016) Elastic structure and seismicity of Donegal (Ireland): insights from passive seismic analysis, Geophysical Research Abstracts, 18, EGU2016-11496, EGU general assembly, Vienna

**Piana Agostinetti, N., ** G. Giacomuzzi and A. Malinverno (2015) Local 3D earthquake tomography by trans-dimensional Monte Carlo sampling, Geophys J. Int., 201, 1598–1617, doi: 10.1093/gji/ggv084

**Piana Agostinetti, N.** (2015) Trans-dimensional MC sampling of geophysical inverse problems and the dataspace. Societa' Italiana di Fisica, 2015 General Assembly, Rome.

**Piana Agostinetti, N.,** and M. Calo' (2015) Refined event location and lowered uncertainties in temporal variation of seismic velocity: tackling down the spatial and temporal correlation between elastic and hypocentral parameters in full 4D local earthquake tomography of a geothermal site, Geophysical Research Abstracts, 17, EGU2015-8454

**Piana Agostinetti, N.,** and M. Calo' (2014) Close-up to the stimulation phase of a EGS geothermal site: mapping the time-evolution of the subsurface elastic parameters using a trans-dimensional Monte Carlo approach, Geophysical Research Abstracts, 16, EGU2014-10769.