Physics and Application of Seismic Emission

8th Annual Report

2012
2

Copyright 2013
c by

Freie Universität Berlin

Institute for Geological Sciences
Department of Geophysics
Malteserstr. 74-100, Bldg. D
D 12249 Berlin, Germany

Disclaimer: Please note that there is no review process concerning the individual papers within this
annual report. The authors are responsible for their articles.

Permission is granted to make and distribute verbatim copies of this report, or parts of it, for internal
purposes of the sponsors of the PHASE consortium, provided that the PHASE consortium is acknowl-
edged as the source on all reproductions and copies. All copies must contain the above Disclaimer and
Copyright Notice.
Annual Report 2012 3

Contents

Introduction to the 8th Annual Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

The Physics and Applications of Seismic Emission (PHASE) Project . . . . . . . . . . . . . . 9

Microseismic Monitoring: Imaging and Analysis of Microseismic Wavefields 13

Joint Arrival Time Optimization for Microseismic Events Recorded by Seismic Borehole
Arrays
J. Kummerow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Precise Seismic Event Relocation Using Waveform Cross-correlation and Source Specific
Station Terms
N.W. Bloch, J. Kummerow and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Fault Plane Solutions for Multiplet Events

A. Rusu, J. Kummerow, A. Reshetnikov and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . 27

Rupture Propagation Imaging at Microseismic Scale

J. Folesky, J. Kummerow and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location:

Synthetic and Real Data Examples
C. P. Yu, A. Reshetnikov and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

Microseismic Reflection Imaging: Methodology and Applications

A. Reshetnikov, J. Kummerow and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . 53

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

A. Oelke, O.S. Krüger and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas
Reservoir
S. Franke, N. Hummel and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

Microseismic Monitoring: Physical Fundamentals and Interpretation 95

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter
b-value of Earthquakes
C. Langenbruch and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

Triggered versus Induced Events and Bounds of the Magnitude Probability

S. A. Shapiro, O. S. Krüger and C. Dinske . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

J. Galindo Guerreros, C. Dinske, C. Langenbruch and S. A. Shapiro . . . . . . . . . . . . . . . 131
4

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir

Stimulations
C. Dinske, C. Langenbruch, J. Galindo Guerreros and S. A. Shapiro . . . . . . . . . . . . . . . 139

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its
Stress Dependence
S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

N. Hummel and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

Permeability Tensor Characterization from the Back Front of Seismicity

N. Hummel and S. A. Shapiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

Members and Sponsors 187

PHASE Research Team . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

List of Sponsors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

Annual Report 2012 5

Introduction to the 8th Annual Report

This volume is the eighth annual report of the PHASE project. In the year 2012 we continued our research
on the physics, interpretation and processing of microseismic data. The project provides new original in-
sights to all these three aspects of the microseismic technology.

The first part of the report summarizes our research on event location, fault plane solution, rupture prop-
agation and microseismic reflection imaging.

In the first contribution we propose an automatic approach of enhancing consistency in determining ar-
rival times of direct seismic P- and S- waves. It is based on processing of multiplet events (similar
microearthquakes) by several receivers. We cross- correlate microseismic borehole array data for such
events. This leads to a cross-linking of waveforms of different events in a multiplet recorded by closely
spaced receivers. Improved traveltimes lead finally to better event locations. The method has been ap-
plied to microseismic data recorded by two inclined borehole arrays consisting of in total 30 receivers.
Relocation of the events using the new time picks results in a reduction of the location RMS by about 20%.

Also the next contribution proposes a workflow to precisely relocate seismic events. The workflow in-
corporates a correction for arrival-time picking uncertainties and for incorrectness of the velocity model,
respectively. We have applied this type of processing to a large-scale dataset of tectonically induced seis-
micity from the northern Chilean subduction zone. The resulting locations reveal the previously poorly
resolved phenomenon of a triple seismic zone in the subducting slab.

Further we study a relation between the fault plane solutions obtained using surface and borehole arrays.
We use the example of the Basel geothermal reservoir. Firstly we show that for large-magnitude events
the borehole based solutions are in a very good agreement with the solutions obtained using surface ar-
rays. We show then that individual solutions are representative for events of multiplets.

In the next contribution we start to investigate the process of rupture propagation in microseismic events.
We implement the back projection approach combined with the P-wave polarization analysis to image
the rupture propagation. We apply the algorithm to a synthetic data set obtained by a finite-difference
simulation. Our modeling shows that we can retrieve rupture direction, duration and speed for a 200m,
horizontal, unilateral rupture approximately corresponding to a ML =3.4 event.

In the following paper we are developing an inversion-based approach to event locating and a macro-
model constructing in transversely-isotropic layered media. The velocity anisotropy is taken into account
in an additional step after a conventional event locating. In this step, the isotropic velocity model and
corresponding locations serve as an initial model. Then the anisotropy parameters as well as resulting
locations are obtained simultaneously. The method is ray theory based. The nonlinear inversion problem
is solved by using Gauss-Newton method.

Further we give a detailed overview of technological steps of our microseismic reflection imaging ap-
proach. We show its applications for tectonic and geothermic targets. We conclude this contribution by
showing a microseismic reflected wavefield of hydraulic-fracture induced events in a gas shale.

In the next contribution we continue our research on reflection coefficients at thin fluid layers represent-
ing hydraulic fractures. We consider also double-couple sources. We show that under a realistic choice
of parameters the viscosity can be neglected and water is adequate as a fluid filler. High reflection co-
efficients (0.2-0.7) may occur at layers of thicknesses in mm-range and at frequencies of 50-400Hz, like
one comes across in microseismicity. Thus, data recording microseismic events can, on the one hand,
be treated as reflection seismic data, and small-scale subsurface images can be derived, and, on the other
hand, reflection coefficients can be interpreted in terms of reservoir properties.
6

We conclude the first part of the volume by a short review of microseismic data recorded during a hy-
draulic fracturing treatment in a shale gas reservoir. This data was provided by one of the sponsors of the
PHASE research project. The purpose of this analysis is to get an overview of the data and image the first
results. This data set will play important role in our future research.

The second part of this report volume is dedicated to the physics of the microseismicity and to interpre-
tation of microseismic data.

In the first contribution we try to derive geomechanical relations between the microseismicity and the
elastic heterogeneity in rocks. Statistical analysis of logging data from various sites suggests that elastic
parameters exhibit a heterogeneous distribution characterized by broad band fractal scaling with a uni-
versal fractal dimension of d ≈ 1.9. Independently, it is known that power law scaling of the number
of earthquake with magnitude, expressed in the Gutenberg-Richter relation, can be explained by frac-
tal distributed stress. However, a direct reason for the fractal distribution of stress is still lacking. We
establish relations between the fractal distribution of elastic parameters, stress field variations and the
Gutenberg-Richter b-value of earthquakes. Using the finite element modeling we determine the distribu-
tion of Coulomb Failure Stress (CFS) in media with fractal heterogeneity. The CFS is as a measure of
the vicinity to failure. It can be correlated with the brittleness of the rock. The resulting CFS is highly
heterogeneous and of fractal nature. It is strongly related to the initially assigned elastic parameters.
The volumes of critical CFS represent fault systems. We show that the distribution of fault sizes and
correspondingly magnitudes exhibits power law scaling according to the Gutenberg Richter law. The
application of our method to well log data measured along the Continental Deep Drilling Site (KTB, Ger-
many) main hole results in a realistic b-value of b = 0.95. The universal degree of elastic heterogeneity
measured at various sites could hence explain a universal b-value of b ≈ 1.

Previously we observed that fluid-induced large-magnitude events at geothermal and hydrocarbon reser-
voirs are frequently underrepresented in comparison with the Gutenberg-Richter law. This is an indication
that the events are much more probable on rupture surfaces contained within the stimulated volume. Here
we theoretically and numerically analyze this effect. We consider different possible scenarios of event
triggering: rupture surfaces located completely within or intersecting only the stimulated volume. We
approximate a stimulated volume by an ellipsoid or cuboid and derive the statistics of induced events
from the statistics of random thin flat discs modeling rupture surfaces. We derive lower and upper bounds
of the probability to induce a given- magnitude event. The bounds depend on the characteristic scales
of the stimulated volume. Its minimum principal axis is the most influential geometric parameter. We
compare our analytical results with data on seismicity induced by fluid injections in boreholes. Fitting
the bounds to the frequency-magnitude distribution can provide an estimate of a largest expected induced
magnitude and a characteristic stress drop, in addition to improved estimates of the Gutenberg-Richter a-
and b- parameters. The bounds help also to distinguish between triggered and induced events. It seems
that the statistics of induced events follows the lower bound of the magnitude probability. The statistics
of triggered events is represented by the reconstructed Gutenberg-Richter law.

In the next contribution we analyze the influence of log-normal hydraulically heterogeneous fractal ran-
dom media on pore pressure diffusion. For this purposes, time dependent numerical simulations of fluid
injection were computed for a reservoir rock described as a two-dimensional fractal random media using
a finite element code. A general finding of this study is that an increasing complexity of heterogeneity
with increasing fractal dimension does not significantly influence the average distribution of pore pres-
sure. However, the maximum decay of pressure seems to be stronger in fractal media. Diffusion flux
magnitudes for all hydraulically heterogeneous models show interconnected porosity ’channels’. We ob-
serve that the channel widths decrease and the flux magnitudes increase with increasing fractal dimension.

Further, we interpret borehole logging measurements from the German Continental Deep Drilling Pro-
gram (KTB) to construct a hydraulic diffusivity profile for the central part of the main borehole. The
profile indicates strong fluctuations of diffusivity ranging over three orders of magnitude. We find that
Annual Report 2012 7

the variations can statistically be described by a log-normal distribution and by a power-law scaling in the
corresponding power spectrum. We generate 2D fractal random media which obey the same statistical
properties of the diffusivity, and numerically simulate the diffusion of pore pressure. It seems that the
magnitude of pore pressure changes is affected, in particular in the vicinity of the injection bore- hole
where regions of strong enhancement as well as reduction exist. Another feature is the development of
flow paths along which the injection pressure is transported. We assume that these flow paths reflect
higher permeable fractures embedded in the low porosity rock matrix.

In the next contribution we compare the behavior of permeability and elastic properties of rocks as func-
tions of the tectonic stress and pore pressure. We propose that different parts of the pore and fracture
space differently contribute to a control of these two types of physical properties of rocks. The stress
dependence of elastic moduli of drained rocks is mainly controlled by deformations of the compliant
part of the pore (including fractures) space. This leads to saturating exponential functions expressing
load dependences of seismic velocities and of related seismic parameters. On the other hand a temporal
evolution of seismicity induced by fluid stimulations (including hydraulic fracturing) of rocks sometimes
implies power-law dependences of the permeability on the pore pressure. It seems that in some rocks
permeability is controlled by the stiff part of the pore space. This leads to a power-law pressure depen-
dence of the permeability. Sometimes permeability can be controlled by compliant pores (e.g., cracks).
Then exponential type of permeability as function of pressure can be observed. We propose an analytic
permeability model taking these two parts of the porosity into account.

The last two contributions address the so-called back front of induced seismicity. The back front de-
scribes the distance from the injection point at which seismicity is terminated at a given time after the
end of the fluid injection. Assuming a constant with time and pressure hydraulic transport the back front
allows to estimate hydraulic diffusivity of the rock from the spatio-temporal evolution of post-injection
induced seismicity. However, fluid injections can also significantly change hydraulic transport properties.
For instance, hydraulic fracturing treatments can considerably enhance the permeability. In this case the
strongly increasing permeability becomes a function of the pressure perturbation. For such situations we
explore microseismic signatures of the back front. We numerically consider a power-law as well as an
exponential dependence of the diffusivity on the pressure perturbation. We numerically compute solu-
tions of corresponding nonlinear diffusion equations. We introduce a medium realization which takes into
account a post-injection enhanced permeability. For different locations this model changes the nonlinear
hydraulic transport into a constant one. This hydraulic modification is done as soon as the local pressure
perturbation has reached its maximum. For such a medium we compute pressure profiles which are used
to generate synthetic microseismicity. We analyze the spatio-temporal characteristics of post-injection
induced seismicity to identify the character of the back front. Our results show that nonlinear fluid-rock
interaction still leads to a distinct and therefore detectable back front. However, the time dependence of
the back front can significantly differ from a square root.

Further, we explore what kind of diffusivity estimates are provided by the back front in real situations.
For this we consider seismicity recorded during the borehole fluid injection experiment at Ogachi (Japan)
and Fenton Hill (New Mexico). We apply a scaling approach which is based on a model of a factor-
ized anisotropic pressure dependence of permeability. With this scaling approach we transform clouds of
hypocenters of events obtained in a hydraulically anisotropic medium into a cloud which would be ob-
tained in an equivalent isotropic but still nonlinear medium. We then apply the back front concept which
provides a certain diffusivity estimate. This diffusivity estimate is used together with scaling factors from
the transformation approach to reconstruct the corresponding diffusivity tensor. Our results show that the
back front provides a diffusivity value which corresponds to the smallest diffusivity tensor element.

The PHASE team is greatly indebted to our sponsors for their generous support making our research
possible. Thank you!

Serge Shapiro
8
Annual Report 2012 9

The Physics and Applications of Seismic Emission (PHASE) Project

The goals of the PHASE project are to develop a better understanding of the physics of fluid induced
microseismicity, to establish physical fundamentals for microseismic monitoring and to further develop
the rock-physics based reservoir characterization approach.
The project includes numerical, theoretical and real-data based studies.

The main research directions of the PHASE project are:

• Triggering mechanisms of induced seismicity
• Modeling and imaging of induced seismicity
• Interpretation and visualization of microseismic data combined with 3-D reflection seismic images
• Seismicity-based methods for estimating hydraulic properties of rocks
• Combining microseismic data and reservoir simulations
• Estimating seismic criticality of rocks
• Physics and monitoring of hydraulic fracturing
• Physics and monitoring of poroelastic stress relaxation
• Stress and pore pressure related numerical and theoretical rock physics
• Effective properties of fractured reservoirs

The Physics and Applications of Seismic Emission (PHASE) Project

10

Selected publications (in reverse chronological order)

Seismotectonic state of reservoirs inferred from magnitude distributions of fluid-induced seismic-

ity. Dinske, C., and Shapiro, S.A., J. Seis., Vol.17, doi: 10.1007/s10950-012-9292-9,2013.

Comment on "Role of seepage forces on seismicity triggering" by Alexander Y. Rozhko. Shapiro,

S.A., J. Geophys. Res., Vol. 117, B03305, doi:10.1029/2011JB008346, 4 pp., 2012.

Microseismic estimates of hydraulic diffusivity in case of non-linear fluid-rock interaction. Hum-

mel, N., and Shapiro, S.A., Geophysical J. Int., Vol. 188, 3, pp. 1441-1453, doi: 10.1111/j.1365-
246X.2011.05346.x, 2012.

Magnitudes of induced earthquakes and geometric scales of fluid-stimulated rock volumes. Shapiro
S., A., Krüger O., Dinske C., and Langenbruch, C., Geophysics, Vol.76, 6, pp. WC55-WC63, doi:
10.1190/GEO2010-0349.1,2011.

Inter event times of fluid induced earthquakes suggest their Poisson nature. Langenbruch, C.,
Dinske, C., and Shapiro, S.A., Geophysical Research Letters, Vol. 38, L21302, 6 PP.,
doi:10.1029/2011GL049474, 2011.

Acoustic emission induced by pore pressure changes in saturated rocks. Mayr, S., Stanchits, S., Lan-
genbruch, C., Dresen, G., and Shapiro, S.A., Geophysics, Vol. 76, pp. MA21, doi:10.1190/1.3569579,
2011.

Fracturing of porous rock induced by fluid injection. Stanchits, S., Mayr, S., Shapiro, S.A., and
Dresen, G., Tectonophysics, Vol. 503, Issues 1-2, pp 129-145, doi:10.1016/j.tecto.2010.09.022, 2011.

Magnitude estimation for microseismicity induced during the KTB 2004/2005 injection experiment.
Haney, F., Kummerow, J., Langenbruch, C., Dinske, C., Shapiro, S.A., and Scherbaum F., Geophysics,
Vol. 76, pp. WC45, 2011.

Seismogenic index and magnitude probability of earthquakes induced during reservoir fluid stim-
ulations. Shapiro, S.A., Dinske, C., Langenbruch, C., and Wenzel, F., The Leading Edge, pp. 936-939,
2010.

Predicting permeability and gas production of hydraulically fractured tight sands from microseis-
mic data. Grechka, V., Mazumdar, P., and Shapiro, S.A., Geophysics, Vol. 75, pp. B1-B10, 2010.

Decay rate of fluid-induced seismicity after termination of reservoir stimulations. Langenbruch, C.,
and Shapiro, S.A., Geophysics, Vol. 75, p. MA53, doi:10.1190/1.3506005, 2010.

Seismic imaging using microseismic events: Results from the San Andreas Fault System at SAFOD.
Reshetnikov, A., Buske, S., and Shapiro, S.A., J. Geophys. Res., Vol. 115, B12324, 9 pp.,
doi:10.1029/2009JB007049, 2010.

Using the value of the crosscorrelation coefficient to locate microseismic events. Kummerow, J., Geo-
physics, Vol. 75 (4), pp. MA47-MA52, DOI: 10.1190/1.3463713, 2010.

Interpretation of microseismicity resulting from gel and water fracturing of tight gas reservoirs.
Dinske, C., Shapiro, S.A., and Rutledge, J.T., PAGEOPH, Vol. 167 (1-2), pp. 169-182, 2010.

The Physics and Applications of Seismic Emission (PHASE) Project

Annual Report 2012 11

Migration-based location of seismicity recorded with an array installed in the main hole of the San
Andreas Fault Observatory at Depth (SAFOD). Rentsch, S., Buske, S., Gutjahr, S.,Kummerow, J., and
Shapiro, S.A., Geophys J. Int., Vol. 182, pp. 477-492, 2010.

Stress triggering and stress memory observed from acoustic emission records in a salt mine. Becker,
D., Cailleau, B., Dahm, T., Shapiro, S.A., and Kaiser, D., Geophys. J. Int., Vol. 182 (2), pp. 933-948,
DOI: 10.1111/j.1365-246X.2010.04642.x, 2010.

Scaling of seismicity induced by nonlinear fluid-rock interaction. Shapiro, S. A., and Dinske, C., J.
Geophys. Res., Vol. 114, B09307, doi:10.1029/2008JB006145, pp. 1-14, 2009.

Stress-dependent anisotropy in transversely isotropic rocks: Comparison between theory and lab-
oratory experiment on shale. Ciz, R., and Shapiro, S.A., Geophysics, Vol. 74 (1), pp. D7-D12, 2009.

Fluid-induced seismicity: Pressure diffusion and hydraulic fracturing. Shapiro, S. A., and Dinske,
C., Geophysical Prospecting, Vol. 57, pp. 301-310, 2009.

Microseismic signatures of non-linear pore-fluid pressure diffusion. Hummel, N., and Müller, T.M.,
Geophysical J. Int., Vol. 179 (3), pp. 1558-1565, 2009.

Microseismicity - a tool for reservoir characterization. Shapiro S.A., EAGE Publications, 67 pp.,
2008.

Statistics of fracture strength and fluid-induced microseismicity. Rothert, E., and Shapiro, S.A., J.
Geophys. Res., Vol. 112, B04309, doi:10.1029/2005JB003959, pages 1-16, 2007.

A numerical study on reflection coefficients of fractured media. Krüger, O.S., Saenger, E.H., Oates,
S.J., and Shapiro, S.A., Geophysics, Vol. 72 (4), pages D61-D67, 2007.

Probability of a given-magnitude earthquake induced by a fluid injection. Shapiro, S. A., Dinske,

C., and Kummerow, J., Geophys. Res. Lett., Vol. 34, L22314, doi:10.1029/2007GL031615, pages 1-5,
2007.

Violation of the Kaiser effect by hydraulic-fracturing-related microseismicity. Shapiro, S. A., and

Dinske, C., J. Geophys. Eng., Vol. 4 doi:10.1088/1742-2132/4/4/003, pages 378-383, 2007.

Generalization of Gassmann equations for porous media saturated with a solid material. Ciz, R.,
and Shapiro, S.A., Geophysics, Vol. 72 (6), pages A75-A79, 2007.

Fast location of seismicity: A migration-type approach with application to hydraulic-fracturing

data. Rentsch, S., Buske, S., Lüth, S., and Shapiro, S. A., Geophysics, Vol. 72, No. 1, pages S33 - S40,
2007.

Fluid induced seismicity guided by a continental fault: Injection experiment of 2004/2005 at the
German Deep Drilling Site (KTB). Shapiro, S. A., Kummerow, J., Dinske, C., Asch, G., Rothert, E.,
Erzinger, J., Kümpel, H.-J., and Kind, R., Geophys. Res. Lett., Vol. 33, No. 1, L01309, doi:10.1029/
2005GL024659, 2006.

Hydraulic-fracturing controlled dynamics of microseismic clouds. Shapiro, S.A., Dinske, C. and,

Rothert, E., Geophys. Res. Lett., Vol. 33, L14312, doi:10.1029/2006GL026365, 1-5, 2006.

Effective elastic properties of fractured rocks: dynamic vs. static considerations. Saenger, E.H.,
Krüger, O.S., and Shapiro, S.A., International Journal of Fractures, Vol. 139, doi:10.1007/s10704-006-
0105-4, 569-576, 2006.

The Physics and Applications of Seismic Emission (PHASE) Project

12

Porosity and elastic anisotropy of rocks under tectonic stress and pore-pressure changes.
Shapiro, S.A., and Kaselow, A., Geophysics, Vol. 70, pages N27 - N38, 2005.

Fluid induced seismicity: theory, modelling and applications. Shapiro, S.A., Rentsch, S., and
Rothert, E., J. Eng. Mech., ASCE, Vol. 131, pages 947-952, 2005.

Evidence for triggering of the Vogtland swarms 2000 by pore pressure diffusion. Parotidis, M.,
Shapiro, S. A., Rothert, E., J. Geophys. Res., Vol. 110, B05S10, doi:10.1029/ 2004JB003267, pages 1 - 12,
2005.

Characterization of hydraulic properties of rocks using probability of fluid-induced microearth-

quakes. Shapiro, S.A., Rentsch, S., and Rothert, E., Geophysics, Vol. 70, pages F27 - F34, 2005.

A statistical model for the seismicity rate of fluid-injection-induced earthquakes. Parotidis, M., and
Shapiro, S.A., Geophys. Res. Lett., Vol. 31, L17609, doi:10.1029/2004GL020421, 2004.

Effective elastic properties of randomly fractured soils: 3D numerical experiments. Saenger, E.H.,
Krüger, O.S., and Shapiro, S.A., Geophysical Prospecting, Vol. 52, pages 183 - 196, 2004.

Characterization of fluid transport properties of the Hot Dry Rock reservoir Soultz-2000 using in-
duced microseismicity. Delépine, N., Cuenot, N., Rothert, E., Parotidis, M., Rentsch, S., and
Shapiro, S.A., J. Geophys. Eng., Vol .1, pages 77 - 83, 2004.

Stress sensitivity of elastic moduli and electrical resistivity in porous rocks. Kaselow, A., and
Shapiro, S.A., J. Geophys. Eng., Vol. 1, pages 1 - 11, 2004.

Back front of seismicity induced after termination of borehole fluid injection. Parotidis, M.,
Shapiro, S.A., and Rothert, E., Geophys. Res. Lett., Vol. 31, L02612, doi:10.1029/ 2003GL018987,
2004.

Mutual relationship between microseismicity and seismic reflectivity: Case study at the German
Continental Deep Drilling Site (KTB). Rothert, E., Shapiro, S.A., Buske, S., and Bohnhoff, M.,
Geophys. Res. Lett., Vol. 30, No. 17, 1893, doi:10.1029/2003GL017848, 2003.

Microseismic monitoring of borehole fluid injections: data modeling and inversion for hydraulic
properties of rocks. Rothert, E., and Shapiro, S.A., Geophysics, Vol. 68, pages 685-689, 2003.

Characterization of fluid transport properties of reservoirs using induced microseismicity.

Shapiro, S.A., Rothert, E., Rath, V., and Rindschwentner, J., Geophysics, Vol. 67, pages 212-220, 2002.

An inversion for fluid transport properties of three-dimensionally heterogeneous rocks using in-
duced microseismicity. Shapiro, S.A., Geophys. J. Int., Vol. 143, pages 931-936, 2000.

Large-scale in-situ permeability tensor of rocks from induced microseismicity.

Shapiro, S. A., Audigane, P. and Royer, J.-J., Geoph. J. Int., Vol. 137, pages 207-213, 1999.

Estimating the crust permeability from fluid-injection-induced seismic emission at the KTB site.
Shapiro, S.A., Huenges, E., and Borm, G.,Geoph. J. Int., Vol. 131, pages 15-18, 1997.

The Physics and Applications of Seismic Emission (PHASE) Project

Annual Report 2012 13

Microseismic Monitoring: Imaging

and Analysis of Microseismic
Wavefields
14
Annual Report 2012 15

Joint Arrival Time Optimization for Microseismic Events Recorded by

Seismic Borehole Arrays

J. Kummerow1
email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary

We cross- correlate microseismic borehole array data for similar events (multiplet events) in order to
determine accurate P & S arrival times. The proposed method extends existing concepts by cross-linking
waveforms of different events in a multiplet recorded by closely spaced receivers, and the augmented
interconnectivity of waveforms increases the consistency of the arrival time data. The method is applied
to microseismic data recorded by two inclined borehole arrays consisting of in total 30 receivers. It
is shown that the picking accuracy is significantly improved compared to the original picks and also
compared to the adjusted picks obtained from single receiver based processing. Relocation of the events
using the new picks results in a reduction of the location RMS by about 20%.
An advantage of the method is that it works fully automatically and avoids time consuming interactive,
semi-manual repicking of seismogram traces.

Introduction

Along with a realistic velocity model and a well-desgined seismic monitoring system, the quality of ar-
rival time measurements is one of the most important factors to limit the uncertainty in the earthquake
location problem (e.g., Pavlis (1992)). The accuracy of time picks is generally increased by exloiting the
waveform similarity, either in time or in frequency domain. This allows to calculate more precise dif-
ferential arrival times which replace, supplement or adjust the original arrival time picks (e..g, VanDecar
and Crosson (1990), Got et al. (1994), Shearer (1997), Waldhauser and Ellsworth (2000), Rowe et al.
(2002)). The data are processed either event-based or receiver-based. VanDecar and Crosson (1990) use
the waveform cross correlation of single teleseismic events recorded at stations of a regional network to
determine accurate relative phase arrival times between the network stations. In most cases, the process-
ing is applied to events with highly similar waveforms recorded by a single receiver (e..g, Shearer (1997),
Rowe et al. (2002)). DeMeersman et al. (2009) apply the two cross correlation procedures one after the
other.
In this paper, we propose to combine the receiver- oriented and the event- oriented approaches to opti-
mize jointly arrival time picks for microseismic events recorded by multi-level borehole arrays. For this
geometry, which is one of the prevalent layouts in the seismic monitoring of hydaulic fracturing experi-
ments, the separation of neighbouring receivers is of the same order as the typical inter-event distances
within a multiplet (typically meters to tens of meters). Assuming similar source mechanisms for the
multiplet events, waveform similarity depends primarily on inter-event or inter-receiver distance (Menke
et al. (1990), Menke (1999), Kummerow (2010)), and we expect well- correlated seismograms for any
combinations of multiplet events and adjacent receivers. Fig. 1 illustrates the high wavefomr similarity
for neighbouring receivers and the general tendency of decreasing similarity with increasing station sep-
aration.

Joint Arrival Time Optimization for Microseismic Events Recorded by Seismic Borehole Arrays
16

Figure 1: The cross correlation matrix for a single event recorded by neighbouring receivers shows the decrease of
waveform similarity with increasing receiver distance.

Arrival time optimization

We assume that we have identified groups of closely spaced similar events (multiplets), e.g. by using
a threshold value for the cross correlation coefficient CC that seismograms of event pairs within the
multiplet must exceed for a certain number of stations (typically CC > 0.85, depending on window
length and filter parameters). This pre-selection greatly reduces the number of event pairs, which scales
approximately as the square of the number of events.
For each multiplet, we then cross-link the seismograms of different events and receivers. Let tik be
the preliminary arrival time pick (P or S) for event k, recorded at receiver i, and tjl the corresponding
preliminary arrival time pick for event l, recorded at receiver j. ∆tikjl is the cross- correlation based time
difference. It is determined as the time lag for which the normalized cross correlation function ccikjl (t)
between the two seismogram traces uik and ujl is maximum:

R
uik (τ ) ujl (τ − t) dτ
ccikjl (t) = qR
u2ik (τ )dτ · u2jl (τ )dτ
R

The maximum value of ccikjl (t) is referred to as the cross correlation coefficient CCikjl .
If the two independent measurements, i.e. the arrival time picks and the cross- correlation derived relative
times, are perfectly consistent, then

tik − tjl = ∆tikjl (1)

Generally, the measurements are inconsistent, and Eq. 1 is not exactly valid. We can, however, write Eq.
( 1) as an overdetermined system of linear equations. The following equation system illustrates this for
the case of two receivers and three events:

Joint Arrival Time Optimization for Microseismic Events Recorded by Seismic Borehole Arrays
Annual Report 2012 17

   
t11 1 0 0 0 0 0

 t12  
  0 1 0 0 0 0 


 t13  
  0 0 1 0 0 0 


 t21  
  0 0 0 1 0 0 


 t22  
  0 0 0 0 1 0 


 t23  
  0 0 0 0 0 1 


 ∆t1112  
  −1 1 0 0 0 0 
 

 ∆t1113  
  −1 0 1 0 0 0 
 T11

 ∆t1121  
  −1 0 0 1 0 0 
 T12 


 ∆t1122  
  −1 0 0 0 1 0 
 T13 


 ∆t1123 =
  −1 0 0 0 0 1 
 T21 


 ∆t1213  
  0 −1 1 0 0 0 
 T22 


 ∆t1221  
  0 −1 0 1 0 0 
 T23 

 ∆t1222  
  0 −1 0 0 1 0 


 ∆t1223  
  0 −1 0 0 0 1 


 ∆t1321  
  0 0 −1 1 0 0 


 ∆t1322  
  0 0 −1 0 1 0 


 ∆t1323  
  0 0 −1 0 0 1 


 ∆t2122  
  0 0 0 −1 1 0 

 ∆t2123   0 0 0 −1 0 1 
∆t2223 0 0 0 0 −1 1
Tij are the 6 = 2 (receivers) ·3 (events) optimized arrival time picks, tij are the original arrival time
picks.
For the general case of M receivers and N events, we can write an overdetermined system of N · M +
(N · M )(N · M − 1)/2 linear equations. Each optimized pick time Tij is constrained by up to N · M
measurements: the preliminary pick time tij and (N · M − 1) relative arrival times ∆tikjl . This implies
an increase of correlations per trace compared to single receiver processing (N · (N − 1)/2 correlations)
and compared to single event processing (M · (M − 1)/2 correlations).
In practice, it is reasonable to account for the general decrease of waveform similarity with increasing
inter-event and inter-receiver distance by weighting the linear equations according to the cross correlation
coefficients and to omit pairs for which the waveform similarity falls below a threshold value. The
equation system can be solved iteratively, with the updated arrival times substituting the original picks.
Optionally, it may be a good choice to reduce successively the window size for the waveform correlation
in later iterations due to the increasing accuracy of the onset times, which in turn further improves the
resolution of the correlation- derived time lag values.
Although the arrival time optimization is a sparse matrix problem and could be solved more efficiently by
using e.g. the conjugate gradient method, we have firstly implemented a routine based on singular value
decomposition (SVD) to solve the problem.
Fig. 2 shows exemplarily the application to 40 P wave- windowed seismograms, recorded by eight
receivers for five multiplet events. All waveforms are similar with correlation coefficients 0.70 < CC <
0.97. Some of the automatically determined original P arrival time picks are not accurate (top row (I)
in Fig. 2; see e.g. the untimely pick for event ev01 at receivers R01 and R02 in (Ib)). Adjustment of
arrival time picks using only events recorded by the same receiver results in more consistent arrival times
(middle row (II) in Fig. 2), but a substantial improvement is only achieved by cross-linking all receiver-
event- combinations (bottom row (III) in Fig. 2). In particular it is seen, that the joint optimization is less
sensitive to large initial arrival time errors. For the data shown, a single iteration is necessary to obtain
fully consistent arrival time picks.

Application to field data

We apply the arrival time optimization to microseismic data recorded by two inclined borehole arrays
during a hydraulic fracturing experiment (s. Fig. 3). The first array consisted of 10 3-comp. receivers
with a receiver spacing of 15 m, and the second array of 20 receivers with a receiver spacing of 10 m.

Joint Arrival Time Optimization for Microseismic Events Recorded by Seismic Borehole Arrays
18

(a) (b) (c) (d) (e) (f)

Receiver R02 Event ev01 Event ev02 Event ev03 Event ev04 Event ev05
01
02

03
04

05

1
2
3
4
5
6
7
8

1
2
3
4
5
6
7
8

1
2
3
4
5
6
7
8

1
2
3
4
5
6
7
8

1
2
3
4
5
6
7
8
R0
R0
R0
R0
R0
R0
R0
R0

R0
R0
R0
R0
R0
R0
R0
R0

R0
R0
R0
R0
R0
R0
R0
R0

R0
R0
R0
R0
R0
R0
R0
R0

R0
R0
R0
R0
R0
R0
R0
R0
ev
ev

ev
ev

ev
−5.0

(I)
0.0
time [ms]
5.0
10.0

Original picks
−5.0

(II)
0.0
time [ms]
5.0
10.0

Corrected picks (no cross−linking)

−5.0

(III)
0.0
time [ms]
5.0
10.0

Corrected picks (with cross−linking)

Figure 2: Illustration of cross-linked P arrival times for 8 receivers and 5 microseismic events. Column (a) shows a
receiver gather (receiver R02) for the five different events (vertical component), columns (b)-(f) show the five event
gathers for receivers R01 to R08. Top (I): Traces aligned after original P pick times (i.e. the window starts 5 ms
before the P pick). Middle (II): Traces aligned after adjusted picks, without cross-linking waveforms recorded by
different receivers (’standard processing’). Bottom (III): Traces aligned after adjusted pick times, using the new
method which cross-links event- and receiver- combinations. Results shown in (II) and (III) are obtained after a
single iteration. Note the greatly increased consistency of arrival time picks in (III) compared to (I) and (II). The
receiver spacing is 15 m.

We identify multiplets as groups of recorded microseismic events for which the average cross correlation
coefficient with at least one other event in the multiplet is CCav > 0.85. For each multiplet, the arrival
times are then corrected by the method proposed in the previous section, including the value of the
cross correlation coefficient as a weighting factor. Due to the large distance between the two boreholes
(> 200 m), the data are processed separately for the two arrays.
We use the NonLinLoc software (Lomax et al. (2000)) to locate the events. Fig. 3a,b show the refined
locations for 213 multiplet events based on the optimized arrival times. The comparison of the RMS
location misfit with the misfit obtained for the original pick times (Fig. 3c) demonstrates the improved
location quality.

Conclusions

We have formulated a method that exploits the waveform similarity of cross-linked waveforms of com-
binations of multiplet events recorded by neighbouring receivers in order to optimize the original arrival
time picks. The method has been tested on microseismic data, resulting in a greatly improved picking
consistency. Due to the increased inter- connectivity it outperforms earlier approaches, which involve ei-
ther only receiver- based or only event- based arrival time corrections. Application to data of a case study
with two multi-level borehole arrays shows the potential of the method to obtain refined microseismic
event locations.

Joint Arrival Time Optimization for Microseismic Events Recorded by Seismic Borehole Arrays
Annual Report 2012 19

(a) (b) (c)

Figure 3: Refined locations for multiplet events in plane view (a) and side view (b). Also shown is in subfigure (c)
the distribution of the RMS location misfit for the relocated events (blue) compared to the poorer misfit for locations
using the original pick times (gray).

Acknowledgements
We thank one of the sponsors of the PHASE consortium for providing the presented waveforms and
original arrival time data and all PHASE sponsors for supporting the research presented in this paper.

References
DeMeersman, K., Kendall, J.-M., and van der Baan, M. (2009). The 1998 Valhall microseismic data
set: An integrated study of relocated sources, seismic multiplets, and S-wave splitting. Geophysics,
74:B183–B195.
Got, J.-L., Frechet, J., and Klein, F. (1994). Deep fault plane geometry inferred from multiple relative
relocation beneath the south flank of Kilauea. J. Geophys. Res., 15:15375–15386.
Kummerow, J. (2010). Using the value of the crosscorrelation coefficient to locate microseismic events.
Geophysics, 75:DOI:10.1190/1.3463713.
Lomax, A., Virieux, J., Volant, P., and Berge, C. (2000). Probabilistic earthquake location in 3D and
layered models: Introduction of a Metropolis-Gibbs method and comparison with linear locations. In
Advances in Seismic Event Location. Thurber and Rabinowitz (eds.), pages 101–134.
Menke, W. (1999). Using waveform similarity to constrain earthquake locations. Bull. Seism. Soc. Am.,
89(4):1143–1146.
Menke, W., Lerner-Lam, A., Dubendorff, B., and Pacheco, J. (1990). Polarization and coherence of
5 − 30 Hz seismic wave fields at a hard rock site and their relevance to velocity heterogeneities on the
crust. Bull. Seism. Soc. Am., 80:430–449.
Pavlis, G. (1992). Appraising relative earthquake location errors. Bull. Seism. Soc. Am., 82(2):836–859.
Rowe, C., Aster, R., Borchers, B., and Young, C. (2002). An automatic, adaptive algorithm for refining
phase picks in large seismic data sets. Bull. Seism. Soc. Am., 92:1660–1674.
Shearer, P. (1997). Improving local earthquake locations using the L1 norm and waveform cross cor-
relation: Application to the Whittier Narrows, California, after shock sequence. J. Geophys. Res.,
102:8269–8283.
VanDecar, J. and Crosson, R. (1990). Determination of teleseismic relative phase arrival times using
multi-channel cross-correlation and least squares. Bull. Seism. Soc. Am., 80(1):150–169.

Joint Arrival Time Optimization for Microseismic Events Recorded by Seismic Borehole Arrays
20

Waldhauser, F. and Ellsworth, W. (2000). A double-difference earthquake location algorithm: Method

and application to the northern Hayward fault, California. Bull. Seism. Soc. Am., 90:1353–1368.

Joint Arrival Time Optimization for Microseismic Events Recorded by Seismic Borehole Arrays
Annual Report 2012 21

Precise Seismic Event Relocation Using Waveform Cross-correlation

and Source Specific Station Terms

N.W. Bloch1 , J. Kummerow1 , and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
We present a workflow to precisely relocate seismic events, and we demonstrate an application to a
dataset of tectonically induced seismicity from the northern Chilean subduction zone. The workflow
incorporates a correction for arrival-time picking uncertainties and for incorrectnesses of the velocity
model, respectively. The resulting locations reveal details of the deformation pattern of the overriding
plate. In the subducting slab, the previously poorly resolved phenomenon of a triple seismic zone becomes
evident.

Introduction
The Chile dataset
We performed a passive seismic experiment in the forearc of the northern Chilean subduction zone. The
aim of the experiment was to explore the transient phenomena that occur due to the deformation of the
overriding continental lithosphere in a subduction regime. The data used here origin from two networks
that were designed to record seismic events in the vicinity of two major crustal scale fault zones. Never-
theless, they also delivered valuable information about processes taking place in the subducting oceanic
lithosphere. The Tarapacá- (T-) network in the east was in operation from 2005 to 2012 (Salazar (2011));
the Atacama Fault- (AF-) network in the west from 2010 to 2012. We detected 7,165 events from which
we localized 5,456 (Fig. 1, left).

Velocity model
For seismic event location, we used the velocity model by Lüth (2000). It was built on the basis of
data from a wide-angle seismic experiment conducted within the framework of the ANCORP project
(ANCORP Working Group (2003)). The E-W-oriented recording line of this experiment transected the
locations of both networks. It ended at the coast near 70.2◦ W. Offshore we extrapolated the velocity
model, assuming a constant subduction angle. For the mantle below the oceanic crust, we assumed a
uniform velocity of 8 km/s.

Methodology
Here, we present a workflow to precisely relocate seismic events and determine double couple source
mechanisms. The localization procedure explicitly takes into account and corrects for uncertainties from
arrival-time picking and inconsistencies in the velocity model (Fig. 1, right).
From the continuous data, we manually detected events and picked arrival-times. To correct for picking
uncertainties, we grouped events by means of their waveform similarity into multiplets. Inside each multi-
plet, we refined the P- and S-wave arrival-time picks by evaluating the cross-correlation function between
the respective event waveforms. We localized the events with NonLinLoc (Lomax et al. (2000)). From
the arrival-time residuals, we computed source-specific station terms (SSST) to correct space-dependently

Precise Seismic Event Relocation Using Waveform Cross-correlation and Source Specific Station Terms
22

−70˚30' −70˚00' −69˚30' −69˚00'

−20˚30'
Continuous Data
pql detection
Seisgram2K cross-
Event waveforms Multiplets
correlation
−21˚00' Seisgram2K picking Time lags

P- & S- arrival times arrival-time

refinement
Velocity Model

km SSST
−21˚30'
0 50 NonLinLoc arrival-time inversion

Locations arrival-time
0 residuals

20
P- polarities
Depth [km]

40 S/P- amplitude ratios

Hash grid-search
60
Source mechanism
80

100

120
−70.5 −70.0 −69.5 −69.0
Longitude
Seismic event AF−network
4 5 6 7 8 Mapped fault T−network
vP [km/s]

Figure 1: Left: Refined event locations with seismicity of the continental South American plate colored gray and
seismicity of the oceanic Nazca plate colored black. Right: Workflow used to obtain these locations.

for inconsistencies in the velocity model. We relocalized the events and iteratively optimized SSST. The
result of this procedure are event locations that suffer less from picking uncertainties and an imprecise
velocity model. From manual readings of first arrival P-wave polarities and S- to P-wave amplitude ratios,
we computed double couple focal mechanisms using the software HASH.

Arrival-time refinement
To correct for picking uncertainties, we define multiplets, i.e. groups of events that have a similar wave-
form by means of their mutual cross-correlation coefficient. Inside each multiplet, we compare the relative
positioning of arrival-time picks on similar waveforms and use the respective time-lags in combination
with the original picks to invert for a set of refined arrival-time picks. An example of pick-refinement by
means of this method is shown in Fig. 2.
The refined picks are situated at the first break of the respective wave train; the comparison of different
picks on similar waveforms yielded a consistent set of refined picks. This method is adapted from Shearer
(1997) and was previously applied to datasets of induced seismicity (e.g. Kummerow et al. (2010)).

Source-specific station terms

After a first event localization, a brief analysis showed that arrival-time residuals are systematically biased
as a function of their respective source location. This is due to weak but large-scale incorrectnesses in the
velocity model. To correct for them, we adapted the concept of source-specific station terms (SSST) by
Richards-Dinger and Shearer (2000). We first performed a Delaunay triangulation on the seismic event
cloud. We used the resulting graph as a distance measure to determine the 300 natural neighbors of each
event. In this manner, we defined a subset of neighboring events for each source location (Fig. 3). We
then applied the mean of the arrival-time residuals of the subset of neighboring events as a station term
for a subsequent relocalization, and optimized SSST iteratively.

Precise Seismic Event Relocation Using Waveform Cross-correlation and Source Specific Station Terms
Annual Report 2012 23

P-wave trains on Station AF10 for Multiplet 007 P-wave trains on Station AF10 for Multiplet 007
0 0
Time [s]

1 1

Figure 2: Waveforms of the 1.1 s P-wave time window of a multiplet consisting of 60 events as registered on one
station. Traces are bandpass filtered 3-15 Hz and normalized to the maximal amplitude. Left: traces aligned by the
original P-wave arrival time pick (red line). Note, that picks are not always situated at the first break of the arriving
P-wave train, and that some picks are erroneous. Right: traces aligned by the refined P-pick. Here, picks are much
more consistent and the overall waveform coherency becomes visible.

0 0 0
Depth [km]

Depth [km]

Depth [km]
−50 −50 −50

−100 −100 −100

−150 −150 −150

0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250
Easting [km] Easting [km] Easting [km]

Figure 3: For each event (green), a subset of 300 neighbors (red) was chosen from the entire dataset (black) for
the determination of SSST. Note, that the Delaunay triangulation bundles nearby events and emphasizes intrinsic
structures naturally.

1000 1000
AF10, uncorr. AF10, SSST
750 750
Number

Number

500 500

250 250 0

0 0 −20
−0.5 0.0 0.5 −0.5 0.0 0.5
Residual Time [s] Residual Time [s] −40
400 400
Depth [km]

T24, uncorr. T24, SSST

−60
300 300
Number

Number

−80
200 200
−100
100 100
−120
0 0
−0.5 0.0 0.5 −0.5 0.0 0.5 −140 AF10, SSST
Residual Time [s] Residual Time [s]
2000 2000 150
50
1500 100 100
1500
Number

Number

150 50 Northing [km]

1000 Easting [km] 200
1000
250
500 500

0 0
0.0 0.1 0.2 −0.1 −0.05 0 0.05 0.1
0.0 0.1 0.2
RMS [s] Station Term [s]
RMS [s]

Figure 4: Left: Residual times of P-phases (red) and S-Phases (green) on selected stations, and RMS misfit of the
entire dataset (brown), without any station terms (left column), and after the application of SSST (right column).
Right: Distribution of SSST attributed to the events for Station AF10.

Precise Seismic Event Relocation Using Waveform Cross-correlation and Source Specific Station Terms
24

The application of SSST allowed the correction of arrival-time residual distributions that were skew or
even bimodal. The resulting distributions are near-gaussian and suffer less from biases introduced by
the velocity-model (Fig. 4, left). The spacial distribution of SSST (Fig. 4, right) allows to evaluate
unresolved heterogeneities of the velocity model and may be interpreted geologically.

Fault-plane solutions
We implemented the software HASH provided by Hardebeck and Shearer (2002, 2003) to our workflow
to reliably determine double couple fault-plane solutions. The software uses take-off angles, first-motion
P-wave polarities and S- to P-wave amplitude ratios to find a minimum-misfit set of strike-, dip-, and
rake-values for a seismic event. It takes into account uncertainties in the take-off angle to find a robust
solution. Examples are shown in Fig. 5. Take-off angles were extracted from the raytracing routine
that was previously used for event localization. P-wave polarities and S- to P-wave amplitude ratios were
determined manually, an automated procedure is currently under development. A large-scale application
to the dataset is a current effort of our work.
The resulting focal mechanisms are stable, even with few or inconsistent observations. The perturbation
of take-off angles gives an intuitive uncertainty measure (Fig. 5).

Results
Seismicity occurs in the continental crust, in the oceanic crust and in the oceanic mantle (Fig. 1 and 6).

Seismicity of the continental crust

In the continental crust west of 69.7◦ W, seismicity traces patterns of large, approximately E70◦ S striking
fault zones (Fig. 1). It vanishes in the east of 69.1◦ W in the middle and lower crust, where P-T-
conditions hinder brittle failure. Seismic activity is higher below the AF-network in the west than below
the T-network in the east; between both networks a lack of instrumental coverage results in fewer detected
events.

Seismicity of the subducting slab

In the subducting slab, seismicity occurs in three distinct slab-parallel bands (see also Fig. 7): One at
the contact between the oceanic- and the continental plate, a second near the oceanic Moho and a third
deeper within the oceanic mantle. On the plate interface the main fraction of seismic activity is restricted
to depths shallower than 50 km and occurs on patches of enhanced seismicity around depths of 35 and
45 km, respectively. A part of the relative displacement between the continental- and the oceanic plate is
accommodated here and released as seismic energy. Immediately above the plate interface, seismicity is
sparse. Near the oceanic Moho seismicity occurs in a band parallel to the former one. A slight increase
in event density can be observed here below a depth of around 50 km. In the oceanic mantle, seismicity

Figure 5: Examples of double-couple fault-plane solutions provided by HASH. Thick lines indicate preferred solu-
q angles. The +- and −-signs indicate first-motion
tions. Thin lines indicate alternative solutions for perturbed take-off
AP
P-wave polarity. The respective symbol size is proportional to AS
, where AP,S is the maximal absolute displace-
ment amplitude of the P- and the S-wave train, respectively.

Precise Seismic Event Relocation Using Waveform Cross-correlation and Source Specific Station Terms
Annual Report 2012 25

20

40
Depth [km]

60

80

100

120
−70.5 −70.0 −69.5 −69.0 −68.5
Longitude

Figure 6: Seismicity from this study superimposed on the ANCORP reflectivity image (ANCORP Working Group
(2003)). Note the correlation between areas of high seismicity and high reflectivity, especially at the plate interface.

x‘
40 80 120 160 200 240 280

0 0
z‘

z‘

20 20

40 40
200 400
# Events
0
z‘

20

40
0 100 2000 100 2000 100 2000 100 2000 100 2000 100 200
# Events # Events # Events # Events # Events # Events
40−80km 80−120km 120−160km 160−200km 200−240km 240−280km

Figure 7: Detail of the seismic event cloud, rotated in the slab-parallel coordinate system. x′ -axis is slab-parallel
in direction of the convergence vector (E13◦ N). z ′ -axis is slab-normal, z ′ = 0 is the approximate plate interface.
Histograms indicate event densities as a function of depth in 2 km bins along the complete profile (top right), and in
40 km wide bins along x′ (bottom row). Note, that seismicity is concentrated in three distinct slab-parallel layers.

Precise Seismic Event Relocation Using Waveform Cross-correlation and Source Specific Station Terms
26

occurs in a third slab-parallel band that exhibits a greater event density in shallower depths around 60 km.
This band may be related to the unbending of the subducting slab and/or to mineral dehydration reactions.
East of 69.5◦ W below a depth of 80 km, the three bands are indistinguishable. Here, the transformation
from blueschist- to eclogite-facies occurs.
A comparison with reflectivity data from the ANCORP experiment (ANCORP Working Group (2003))
shows that reflectivity and seismicity correlate tightly along the plate interface (Fig. 6).

Conclusions
We successfully implemented the concepts of waveform based arrival-time pick refinement and source-
specific station terms to our localization workflow. In this manner, we reduced the disadvantageous
effects of uncertainties from arrival-time picking and incorrectnesses in the velocity model. The resulting
locations reveal details of the subduction process, such as deformation patterns in the overriding plate,
seismic heterogeneity along the plate interface, and the previously poorly resolved phenomenon of a triple
seismic zone.

Acknowledgments
We thank the Deutsche Forschungsgemeinschaft (PAK 262, Sh55/10-1.1), the Freie Universität Berlin,
and the sponsors of the PHASE consortium for supporting the research presented in this paper. A part of
the instruments were provided by the GIPP of the Geoforschungszentrum Potsdam. W.B. thanks open-
source developers for providing free software.

References
ANCORP Working Group (2003). Seismic imaging of a convergent continental margin and plateau in
the central Andes (Andean Continental Research Project 1996 (ANCORP’96). Journal of Geophysical
Research, 108.
Hardebeck, J. L. and Shearer, P. M. (2002). A new method for determining first- motion focal mecha-
nisms. Bulletin of the Seismological Society of America, 92:2264–2276.
Hardebeck, J. L. and Shearer, P. M. (2003). Using S/P Amplitude Ratios to Constrain the Focal Mecha-
nisms of Small Earthquakes. Bulletin of the Seismological Society of America, 93:2434–2444.
Kummerow, J., Shapiro, S. A., Häring, M., and Asanuma, H. (2010). Application of a Waveform
Similarity- Based Location Algorithm to the Basel 1 Microseismic Data. PHASE Annual Report,
pages 15–21.
Lomax, A., Virieux, J., Volant, P., and Berge, C. (2000). Probabilistic earthquake location in 3D and
layered models: Introduction of a Metropolis-Gibbs method and comparison with linear locations.
In Thurber, C. H. and Rabinowitz, N., editors, Advances in Seismic Event Location, pages 101–134.
Springer, Amsterdam, The Netherlands.
Lüth, S. (2000). Ergebnisse weitwinkelseismischer Untersuchungen und die Struktur der Kruste auf einer
Traverse über die Zentralen Anden bei 21◦ S. Berliner Geowissenschaftliche Abhandlungen, 37.
Richards-Dinger, K. B. and Shearer, P. M. (2000). Earthquake locations in southern California obtained
using source-specific station terms. Journal of Geophysical Research, 105:10,939–10,960.
Salazar, P. (2011). The upper crustal microseismicity image from the North Chilean subduction zone:
implications for tectonics and fluid migration. PhD thesis, Freie Universität Berlin, Berlin, Germany.
Shearer, P. M. (1997). Improving local earthquake locations using the L1 norm and waveform cross cor-
relation: Application to the Whittier Narrows, California, aftershock sequence. Journal of Geophysical
Research, 102:8,269–8,283.

Precise Seismic Event Relocation Using Waveform Cross-correlation and Source Specific Station Terms
Annual Report 2012 27

Fault Plane Solutions for Multiplet Events

A. Rusu, J. Kummerow, A. Reshetnikov and S. A. Shapiro

email: [email protected]
Freie Universität Berlin, FR Geophysik

Summary
The focus of this study is to determine fault plane solutions using both first motion polarities and S to
P wave amplitude ratios. While using a limited number of just 6 downhole geophones, the results of
events with magnitudes Mw>1.7 are consistent with those that have already been published using data by
a dense regional seismic network. This justifies the application of the method to weaker (Mw<1) multiple
events with hitherto unknown source mechanisms. It is further shown that individual solutions can be
regarded as representative of the whole multiplet. The results show mostly strike-slip and normal fault
mechanisms, which is in agreement with the stress-regime of the region.

Introduction
The Deep Heat Mining Project in Basel (Switzerland) was instigated to deliver both electrical and geother-
mal energy from a co-generation plant. In order to succeed, the reservoir had to be hydraulically stimu-
lated, but the gradual increase in flow rate and well-head pressure lead to seismic activity. The in-place
monitoring system recorded more than 10000 events, for which about 3000 hypocenters could be de-
termined. This work has the purpose of calculating the fault plane solutions from the Basel dataset by
using P-wave first motion polarities and S- to P-wave amplitude ratios. It is shown that this technique,
while using just 6 borehole geophones, is able to reproduce the results of Deichmann and Ernst (2009)
who used a dense network of local and regional surface stations. The good agreement of the results for
the stronger events justifies the calculation of further focal mechanisms for weaker multiplet events. It is
further shown that individual solutions can be regarded as representative for the whole multiplet. Thus
one solution is enough to represent a whole group of similar events. The results for both the strong and
weak events agree with previous studies and show strike-slip and normal fault mechanisms with P-axes
oriented in NW-SE and T-axes in NE-SW.

The seismic monitoring system

To better understand and monitor the seismic activity of the target area, a sensor network had to be set in
place (see Figure 1). It consisted of 6 downhole geophones and was operated by the Geothermal Explorers
Ltd. (GEL). The geophones comprise 3 components and were strategically installed at several locations
with the following depth: St. Johann-317m, Otterbach1-500m, Haltingen-542m, Schuetzenmatte-553m,
Riehen2-1178m and Otterbach2-2740m Deichmann and Ernst (2009). Apart from the GEL stations,
there is a vast network of near-surface seismometers and accelerometers spread across the area, which are
operated by the SED (Schweizerischer Erdbebendienst, Switzerland) and LED (Landeserdbebendienst
Baden-Wuerttemberg, Germany). The SED and LED data were used in several studies, results of which
will be compared with this work that used only data from GEL.

Tectonic setting
The geothermal reservoir is situated in northwestern Switzerland, in the Kleinhüningen Quartier of Basel.
The region is at the south-eastern margin of the Upper Rhine Graben, which forms part of the Cenozoic
rift system. The here prevailing faults relate to the Rhine-Bresse transfer Zone and the Variscan Orogeny.

Fault Plane Solutions for Multiplet Events

28

Figure 1: Regional map, showing the monitoring system locations.

The research of Plenefisch and Bonjer (1997) , Evans and Roth (1998) and Kastrup et al. (2004) show a
strike-slip dominated stress regime, oriented NNE-SSW and a small normal faulting component.

Hydraulic stimulation
In order to improve permeability, the crystalline rock had to be stimulated. This was done by injecting
water at high pressures into the well. The main stimulation started on December 2nd. Between then and
December 8th of 2006, more than 11.500m3 of water were injected. The flow rate and pressure were
steadily increased, reaching values of 50 l/s and 29.6 MPa respectively. In the morning of December
8th, the first event that exceeded the allowed safety threshold was registered. Its magnitude was ML
2.6 and lead to a precocious stoppage of the stimulation. Later that day two even stronger events, with
magnitudes of ML 2.7 and 3.4 lead to the opening of the well for bleed-out. During the 6 days of injection
and the following 4 days depressurizing the well, a total of 13500 events were detected Dyer et al. (2008).
December 13th marks the beginning of the post-stimulation phase, during which further events occurred,
but at a much lower rate.

Method
Prior to this work, high resolution locations for almost 3000 events were determined by Kummerow et al.
(2011). The essential idea behind the current study is to produce a set of fault plane solutions by taking
into account first motion polarities and the S- to P-wave amplitude ratios. The ray azimuth and take-

Fault Plane Solutions for Multiplet Events

Annual Report 2012 29

off angles are needed for computing fault plane solutions and were therefore extracted for the chosen
events. The amplitude ratios and the first motion polarities were determined from seismograms. We use
the HASH software (Hardebeck and Shearer (2002)) to calculate a set of acceptable solutions and the
preferred solution for each event in form of three parameters: strike, dip and rake.

Results and Analysis

During the time-period between December 3rd 2006 and May 6th 2007, almost 3000 locatable events
were detected, 28 of which with magnitudes between 1.7 and 3.4 ML. The fault plane solutions of the 28
events were published by Deichmann and Ernst (2009). For their calculations, the data recorded by the
dense network of near surface seismometers and accelerometers operated by the SED and LED, as well
as the data from the GEL downhole geophones were used. The fault plane solutions were determined
based on first motion polarities. In order to test the applicability of the method implemented in this work,
it was assumed that Deichmann and Ernst results could be reproduced. For this, the dataset provided
by GEL was processed and the solutions for 19 of the 28 biggest events calculated. Even though the
fault plane solutions were determined with data from just 6 stations, the results are consistent and the
difference is minor for most calculations. For a better visualization, the discrepancies in strike, dip and
rake are plotted in Figure 2 on three graphs. The corresponding fault plane solutions in form of equal

Figure 2: Graphical plots of the absolute differences to the results of Deichmann and Ernst (2009). The event numbers
are plotted on the x-axis and are consistent with those of Deichmann and Ernst (2009). The absolute difference in
degrees is plotted on the y-axis. For 16 of the strike results, the absolute variation is <12.2◦ . For the dip, the situation
is quite similar, 16 results have a difference of 11.9◦ or less.15 of the rake events show a discrepancy of < 15.9◦ .

Fault Plane Solutions for Multiplet Events

30

are projections shown in Figure 2 reproduce those obtained by Deichmann and Ernst (2009). A closer
look at the events with a slightly higher discrepancy shows that even by using merely 6 stations the
focal mechanisms calculated from both first motion polarities and S- to P-wave amplitude ratios are very
stable. The acceptable mechanisms are nicely grouped together with almost no outliers. The preferred
solution is as expected in the midst of the acceptable solutions and reproduces their mean value. Although
the preferred mechanisms don’t exactly match those of Deichmann and Ernst (2009), the acceptable set
contains their solution. While having a higher number of recording stations would surely be useful, the

Event 5 Event 39 Event 43 Event 93 Event 94

Event 98 Event 102 Event 103 Event 108 Event 112

Event 147 Event 159 Event 162 Event 176 Event 184

Figure 3: Fault plane solutions showing the lower hemisphere for events with magnitudes ML> 1.7. The pluses and
the solid black circles correspond to compressive motions (positive polarities/Up); the minuses and the empty circles
correspond to dilatational motions (negative polarity/Down).

Fault Plane Solutions for Multiplet Events

Annual Report 2012 31

implemented method achieves high quality results and can therefore be used to calculate further solutions
for the current dataset. The multiplet analysis done by Kummerow et al. (2011) allowed the identification
of 330 multiplets, also referred to as clusters. Since events belonging to the same multiplet possess similar
waveforms, its fair to say that they should have the same or very similar focal mechanism. Thus, a single
event could represent the whole cluster. To test the hypothesis, random events (occurred at different
times) from the same cluster were selected and the fault planes solutions calculated. This was done for 5
multiplets. As an example the fault plane solutions for 6 events of the largest multiplet (nr. 22) have been
calculated and compared. The median strike is 152.7◦ from North with a standard deviation of ±4.6◦.
For the dip the median is 48.1◦ ±1.6◦ and for rake -68.9◦±1.9◦. Comparable deviations were calculated
for the 4 other clusters. Consequently the premise has been confirmed and one representative event has
been chosen for 22 further clusters. The calculated fault plane solutions are shown in Table 1 and Figure
4.

Table 1: Table 1 Fault plane solutions in form of strike, dip and rake for 27 multiplets
ClusterID No. of events Event date and time Strike [◦ ] Dip [◦ ] Rake [◦ ]
Cluster 005 9 06.12.06 01:21:28 101.8 76.1 178.4
Cluster 006 9 06.12.06 18:12:45 161.7 4.4 2.8
Cluster 007 13 06.12.06 19:31:50 85.7 51 176.1
Cluster 010 9 06.12.05 02:23:16 156.3 46.3 -73.5
Cluster 011 9 06.12.07 05:48:29 309 44.8 110.4
Cluster 015 11 06.12.06 15:05:22 91 77.7 -167.8
Cluster 016 17 06.12.06 03:10:41 287.3 56.7 -123.9
Cluster 022 84 06.12.08 13:55:14 138.7 49.3 -66.7
Cluster 023 17 06.12.06 17:00:32 228.7 35.9 99.1
Cluster 024 9 06.12.07 02:57:52 81.2 81.8 129.6
Cluster 030 9 06.12.03 20:58:34 272.7 52.4 -116.6
Cluster 033 15 06.12.08 09:55:58 12.3 42.6 -51.9
Cluster 037 9 06.12.05 23:09:12 147.7 47.4 -77.2
Cluster 038 11 06.12.03 22:30:06 199.6 25.9 24.9
Cluster 040 9 07.03.10 20:39:25 70.2 70.7 -164.1
Cluster 043 12 06.12.08 00:09:14 139.1 15.8 -118.2
Cluster 044 11 06.12.05 11:36:49 115.7 75 175
Cluster 053 16 06.12.06 04:22:59 152 44.7 -80.4
Cluster 054 11 06.12.06 18:04:58 323.6 85.1 -120.7
Cluster 056 24 06.12.07 03:58:37 301.3 48.5 -124.1
Cluster 058 19 06.12.07 09:57:16 155 42.8 -83.9
Cluster 071 11 07.03.10 03:03:18 86.7 74.8 -162
Cluster 074 12 06.12.08 04:21:12 18 41.3 -52
Cluster 075 10 06.12.07 12:44:16 223.7 20.8 -19.5
Cluster 088 9 06.12.07 01:50:10 27.4 43.2 -153.5
Cluster 133 10 07.02.18 15:35:27 82.8 77.5 -166.1
Cluster 212 10 06.12.08 23:53:04 139.9 18.5 -109.8

Discussion
As shown in Figure 3, the calculated solutions for the stronger events generally agree with those of Deich-
mann and Ernst (2009). Events 5 and 113 with rather uncharacteristic differences have been proven to be
stable and the acceptable solutions include those of Deichmann and Ernst (2009). Another inconsistency
calculated for event 185 can be explained by carefully looking at Figure 3. There is namely an inconsis-
tent polarity for event 174, 176 and 185 in the upper left quadrant, which is argued to be an indication of
a non-double-couple component, that cannot be accounted for in by a standard double-couple fault plane
solution Deichmann and Ernst (2009). For the multiplet focal mechanisms a significant increase in the

Fault Plane Solutions for Multiplet Events

32

number of normal faults can be noted; 9 of the 27 solutions are normal faults. Whether this is evidence
of a stronger normal component in the smaller events or a statistically irrelevant factor would have to be
tested by determining the solutions of further multiplets. There have been several studies on the regional
field stress of the southern Rhinegraben, northern Switzerland and the Basel region (Plenefisch and Bon-
jer (1997), Evans and Roth (1998), Deichmann et al. (2000), Deichmann et al. (2002),Deichmann et al.

cluster05 cluster06 cluster07 cluster10 cluster11

cluster15 cluster16 cluster22 cluster23 cluster24

cluster30 cluster33 cluster37 cluster38 cluster40

cluster43 cluster44 cluster53 cluster54 cluster56

cluster58 cluster71 cluster74 cluster75 cluster88

cluster133 cluster133 cluster212

Figure 4: Fault plane solutions for 27 events, each representing a multiplet

Fault Plane Solutions for Multiplet Events

Annual Report 2012 33

(2004), Baer et al. (2001), Baer et al. (2005), Kastrup et al. (2004), Valley and Evans (2006), Valley and
Evans (2009)). All of them agree on a strike-slip dominated stress-regime and normal faults. The fracture
evolution in the Basel reservoir might therefore have evolved on a pre-existing fault zone. Kastrup et al.
(2004) studied the southern Rhinegraben region and calculated about 144◦ for the maximum compressive
stress of naturally occurring events. Valley and Evans (2006) documented the same value for the stress
in the crystalline basement of the Basel reservoir. The paper of Deichmann and Ernst (2009) provides
a value of about 138◦ for the induced seismicity during the DHM Project. The 145◦ calculated for the
higher magnitude events from the dataset provided for this thesis are therefore consistent with the above.
The same is true for the P-axis of the multiplet calculations-about 135◦ for the strike-slips and 144◦ for
the normal faults. The values for the regional least compressive stress published by Kastrup et al. (2004)
for the southern Rhine Graben and Valley and Evans (2006) for the crystalline basement, is about 54◦ .
Deichmann and Ernst (2009) derive an azimuth of about 46◦ . In this work the stronger events have a
51.5◦ oriented T-axis. The representatives of the multiplets show a value of about 45◦ for the strike-slips
and 55◦ for the normal-faults. Almost all nodal planes (for both the strong events and the multiplets) of
the strike-slip events are oriented in NS-EW direction, which implies that the shearing process might have
taken place in the same direction. Häring et al. (2008) propose the hypothesis that the fractures caused
by the injection of water increased the pore pressure in the pre-existing cataclastic fracture zone, which
produced a pattern of strike-slip and normal faults. This is supported by the overall reservoir develop-
ment and the orientation of the nodal planes. The same pattern of strike slip and normal faults is also
documented for the lower magnitude representative events of multiplets. Consequently, the generation of
a good distributed network of fractures by fluid injection failed. Due to a pre-existing stress distribution
the fractures are concentrated in a narrow plane.

Conclusion
By processing the provided dataset, two sets of fault plane solutions have been calculated by using a small
network of six downhole geophones. Nonetheless, for the stronger events (ML>1.7) the results have been
compared to those of Deichmann and Ernst (2009) and proved to be consistent. This justified the further
implementation of the method by calculating the fault plane solutions for weaker multiplet events. It
has been shown that a single event can represent a multiplet. Consequently, for 27 multiplets the focal
mechanisms and the corresponding P- and T-axes have been calculated, which represent a total of 395
events. The orientation of the fault plane solutions for both the strong events and those of the weaker
multiplets agree with the regional field stress and show a Basel region and the geothermal reservoir. In
accordance to this, the presented work provides valuable information and makes an important contribution
to the future visualization and reconstruction of the pattern of fault mechanisms in the Basel geothermal
reservoir.

Acknowledgments
We thank the sponsors of the PHASE consortium for supporting the research presented in this paper.

References
Baer, M., Deichmann, N., Braunmiller, J., Ballarin Dolfin, D., Bay, F., Bernardi, F., Delouis, B., Fäh, D.,
Gerstenberger, M., Giardini, D., Huber, S., Kastrup, U., Kind, F., Kradolfer, U., Maraini, S., Mattle, B.,
Schler, T., Salichon, J., Sellami, S., Steimen, and S., Wiemer, S. (2001). Earthquakes in Switzerland
and surrounding regions during 2000. Technical report.

Baer, M., Deichmann, N., Braunmiller, J., Husen, S., Fäh, D., Giardini, D., a. K. P., Kradolfer, U., and
Wiemer, S. (2005). Earthquakes in Switzerland and surrounding regions during 2004. Technical report.

Deichmann, N., Baer, M., Braunmiller, J., Cornou, C., Fäh, D., Giardini, D., Gisler, M., Huber, S., Husen,
S., Kästli, P., Kradolfer, U., Mai, M., Maraini, S., Oprsal, I., Schler, T., Schorlemmer, D., Wiemer, S.,

Fault Plane Solutions for Multiplet Events

34

Wössner, J., and Wyss, A. (2004). Earthquakes in Switzerland and surrounding regions during 2003.
Technical report.
Deichmann, N., Baer, M., J., B., Ballarin Dolfin, D., Bay, F., Bernardi, F., Delouis, B., Fäh, D., a. G. M.,
Giardini, D., Huber, S., Kradolfer, U., Maraini, S., Oprsal, I., Schibler, R., Schler, T., a. S. S., Steimen,
S., Wiemer, S., Wössner, J., and Wyss, A. (2002). Earthquakes in Switzerland and surrounding regions
during 2001. Technical report.
Deichmann, N., Ballarin Dolfin, D., and Kastrup, U. (2000). Seismizität der Nord-und Zentralschweiz.
Technical report.
Deichmann, N. and Ernst, J. (2009). Earthquake earthquake focal mechanisms of the induced seismicity
in 2006 and 2007 below Basel (Switzerland). Technical report.
Dyer, B., Schanz, U., Spillmann, T., Ladner, F., and Häring, M. (2008). Microseismic imaging of a
geothermal reservoir stimulation. Technical Report 7.
Evans, K. F. and Roth, P. (1998). The state of stress in Northern Switzerland inferred from earthquake
seismological data and in-situ stress measurements. Technical report.
Hardebeck, J. L. and Shearer, P. M. (2002). A new method for determining first- motion focal mecha-
nisms. Technical report.
Häring, M., Schanz, U., Ladner, F., and Dyer, B. C. (2008). Characterisation of the Basel 1 enhanced
geothermal system. Technical report.
Kastrup, U., Zoback, M., Deichmann, N., Evans, K., Giardini, D., and Michael, A. (2004). Stress field
variations in the Swiss Alps and the northern Alpine foreland derived from inversion of fault plane
solutions. Technical report.
Kummerow, J., Shapiro, S., Asanuma, H., and Häring, M. (2011). Application of an arrival time and
cross correlation value-based location algorithm to the Basel 1 microseismic data. Expanded abstracts,
EAGE 73nd annual meeting and technical exhibition, Vienna.
Plenefisch, T. and Bonjer, K. (1997). The stress field in the Rhine Graben area inferred from earthquake
focal mechanisms and estimation of frictional parameters. Technical report.
Valley, B. and Evans, K. F. (2006). Stress orientation at the Basel geothermal site from wellbore failure
analysis in bs1. Technical report.
Valley, B. and Evans, K. F. (2009). Stress orientation to 5 km depth in the basement below Basel (Switzer-
land) from borehole failure analysis. Technical report.

Fault Plane Solutions for Multiplet Events

Annual Report 2012 35

Rupture Propagation Imaging at Microseismic Scale

J. Folesky1 , J. Kummerow1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
In this paper we adopt two techniques from global seismology for imaging the rupture process of a
synthetic rupture within a reservoir model of microseismic scale. We implement the Back Projection
Imaging and the P wave Polarization Analysis methods and apply them to a synthetic setup similar to the
Basel 1 Geothermal Reservoir. We use a finite-difference code to model wave field and seismograms. Our
setups show that we can retrieve rupture direction, duration and speed for a 200m, horizontal, unilateral
rupture (corresponding to a ML =3.4 event), for an azimuthal well distributed seismic network and with
reduced resolution for a poorer coverage alike.

Introduction
A number of recent publications in global seismology deal with the tracking of the rupture front of
large (M >7) to megathrust (M >8) earthquakes. Different techniques are permanently improved using
multiple seismic phases to increase the spatio-temporal resolution and accuracy. Applications were e.g.
the Sumatra-Andaman earthquake 2004 (e.g. Ishii et al. (2007)), the Maule, Chile earthquake 2010 or the
Tohoku, Japan earthquake 2011. One technique is to back project the seismograms recorded at an array or
at a seismic network to a grid of possible source locations. This method is called Back Projection Imaging
(e.g. Ishii et al. (2007), Walker and Shearer (2009)) or Source-Scanning Algorithm (Kao and Shan,
2007), and it can provide information on the energy release and yields estimates of rupture properties like
direction, speed or duration. A different approach is to use P-wave polarity estimates for moving time
windows to map the zone of maximum energy release (e.g. P-wave Polarization Analysis (Bayer et al.
(2012))).
In this paper we test the application of the Back Projection Imaging (BPI) and P-wave Polarization
Analysis (PPA) methods at a reservoir scale. Our motivation is the occurrence of relatively large, induced
seismic events at a number of geothermal experiments, stimulated reservoirs or waste disposal sites,
having a magnitudes M =3 and greater and yielding rupture lengths of several hundred meters. The
largest event observed during the stimulation of the Basel 1 Geothermal Reservoir was a ML =3.4 event,
corresponding to a rupture length of about 200m (Häring et al. (2008)). We use the configuration of the
Basel Geothermal Experiment seismic network and reservoir properties to build a synthetic model of a
rupture by modeling the wave field of multiple spatio-temporally separated single sources. We then apply
the BPI and PPA techniques to investigate their applicability in a microseismic context.
In this work we will first present the rupture tracking techniques and their differences and then explain our
modeling approach with the finite-difference program Evamod (Saenger et al. (1999)). With a synthetic
rupture and station distribution scheme we will show the possibility of gaining information like rupture
direction, length and rupture speed in this exemplary setup after which we discuss inaccuracies and give
an outlook on the work in progress.

Rupture Propagation Imaging at Microseismic Scale

36

Rupture Propagation Imaging

In the following we describe two different approaches for tracking a rupture front. One is based on the
BPI technique (e.g. Ishii et al. (2007)). The other one is adapted from the PPA technique (e.g. Bayer
et al. (2012)).

Figure 1: Back Projection Imaging Scheme. (1) Grid of possible source locations and rupture (red arrow). (2) Wave
fronts arriving at seismic network. (3) Seismic records at receivers. Marked is theoretical P-wave onset for the two
colour-coded source grid points. (4) Seismograms get stacked for each grid point with individual time correction
(e.g. blue or green) yielding a back projection energy record per grid point.

Back Projection Technique

Back projecting an entire wave train from all receivers to a grid of possible source locations is a simple
and very straight forward approach to unravel some of the sources properties. The mathematical idea is
to stack all seismograms u(t) received at n receivers as a function of time for every ith potential source
grid point, corrected by the source-receiver travel time, i.e.,
n
ωk u2k (t − tpik ).
X
si (t) = (1)
k=1

Here we have the stack si (t), ωk , a possible weighting factor for each seismogram at the kth station and
tpik , the travel time, calculated from the ith potential grid point location to the kth receiver location.
By iterating this stacking procedure for all points of our initial grid of possible source points we obtain a
seismogram-like trace for every point, containing a stack value, which is related to the energy radiation
of the source, but due to normalization and weighting at the stations not precisely the radiated energy.
We call it back projected energy. For each time step and an unilateral rupture we ideally expect to have
one maximum of back projected energy, representing the center of the back projected energy, the rupture
front. The stack’s peak location will migrate in time, as the rupture progresses in one direction. The BPI
energy distribution potentially provides estimates for rupture orientation, length and speed.

Rupture Propagation Imaging at Microseismic Scale

Annual Report 2012 37

P-wave Polarization Analysis

Generally speaking, polarization analysis is a vector processing technique that can be performed graphi-
cally or mathematically from particle motion hodograms. We use an algorithm adapted by Rentsch (2007)
after Jurkevic (1988).
The analysis yields the three parameters rectilinearity (L), dip (φ) and azimuth (θ), which provide us with
an estimate of the direction of the incoming wave and a measure of quality. Analysing a time interval of
a three component seismogram, auto- and cross- variances can be obtained from:
n
1X
Cij (t) = ui (t)uj (t), (2)
n t=1

where i and j are the component indices x, y, z and t is a time sample. Cij are the entries of the covari-
ance matrix, that is symmetric and contains real values representing the polarization ellipsoid with best
data fit. Solving for eigenvalues (λ) will give us the principal axis or eigenvectors (p).
Rectilinearity relates the magnitudes of the largest and the smaller eigenvalues as a measure of the linear-
ity of the wave field. For perfectly linear polarized waves it will be close to one, otherwise it will indicate
the ellipsoidal polarization character of the incoming wave:
 
λ2 + λ3
L=1− . (3)
2λ1

With p1 = (p1 (x), p1 (y), p1 (z)) as eigenvector for the largest eigenvalue, the dip and azimuth are:
!  
p1 (z) p1 (y)
φ = arctan p 2 , θ = arctan . (4)
p1 (x)p21 (y) p1 (x)

To get a reliable measure of azimuth quality S. Rentsch Rentsch (2007) proposes a combination of dip
and rectilinearity R = L cos φ, which is called azimuth reliability.

Figure 2: P-wave Polarization Analysis Scheme. (1) Rupture with t1 before t2 . (2) Receivers with three component
seismograms. (3) Polarization analysis for every time step and every station. (4) Stack of deviations to the back
azimuth value calculated for each station grid point pair, so that there is close to zero deviation for the origin of the
wavefront at a given time.

Rupture Propagation Imaging at Microseismic Scale

38

Having determined a way to analyse a three component seismogram at one station, it is now necessary
to combine the obtained back azimuths from multiple stations for the same event to get an estimate of
the rupture location and with progress in time also with the rupture process (Figure 2). Here we follow
the approach from Bayer et al. (2012). We define a function si (t) for every ith possible source grid point
representing the summation of the deviations from the back azimuth for each station. For each station,
each source grid point is assigned a value, that represents the distance to the calculated back azimuths.
The stack for all n stations for one time step and for every gird point ideally yields one minimum with a
value close to zero, being the actual radiation source.
n
X
si (t) = d2ki (t)Rk2 (t). (5)
k=1

Here we have the sum of deviations from the back azimuth direction of the kth station to the ith grid point
si (t), the deviation from the back azimuth direction of the kth station to the ith grid point, dki (t), and the
azimuth reliability factor Rk (t) for the kth station. Iterating the procedure for different time steps yields
in different minimum positions, indicating the rupture and its properties.

Modeling
For the purpose of testing the two methods on synthetic data we use the finite-difference code Evamod
(Saenger et al. (1999)). The program is capable of modeling elastic-wave propagation in complex media.
In our case it is expedient to start working with the simplest medium, a homogeneous one, having in mind
to model data in realistic reservoir applications in the future.
Evamod allows us to place different sources at different times in a defined model space. To simulate a
unilateral propagating rupture we put a straight line of explosion sources in short distances next to each
other, firing one after another. To achieve a positive stack of radiated energy each source time function is
chosen to be a simple gauss function. Examples for seismograms and a snapshot of the propagating wave
field are shown in Figure 3.
Our model grid consists of 8003 points and a grid spacing of 10m. We use the central frequency of 20Hz
for our single source wavelets. Since we work with explosion type sources we lack an S-wave leaving us
with the P-wave train for the analysing process. Multiple geophones can be placed in any location within
the grid.

16 zcomp
ycomp
xcomp
14

12
station number

10

0
150 200 250 300 350 400 450
time in ms

(a) modeled seismograms (b) 3D-wavefield

Figure 3: (a) Modeled recordings at eight receivers at source depth. The variation of record length depending on
azimuthal position (directivity effect) is clearly visible. Station 16 lies in rupture direction (0◦ ), station 8 at the
opposite (180◦ ). Station numbers correspond to Setup1 and Figure 4(a). (b) Snapshot of the x-component of the
wave field for the entire 3D model. The directivity effect is seen in higher amplitudes as well as shorter phases in
rupture direction (positive x-direction).

Rupture Propagation Imaging at Microseismic Scale

Annual Report 2012 39

Microseismic Reservoir Model - Setup 1

16

14

12

station number
10

4
zcomp
2 ycomp
xcomp
0
500 600 700 800 900 1000
time in ms

(a) station distribution (b) seismograms

Figure 4: Seismograms and station distribution for Setup1. The rupture is simulated with a series of explosion
sources firing one after another with an 20m horizontal offset in x-direction. Station numbers in (a) correspond to
the seismograms in (b). White gaussian noise is added to the traces with an SNR of 2. Notice the large directivity
effect.

Our first synthetic setup is shown in Figure 4. We use a homogeneous, isotropic model with pv =5940m/s,
ps =3450m/s and ρ=3000kg/m3, which are averaged values for the geothermal reservoir at Basel, Switzer-
land (Häring et al. (2008) ). We use 11 explosion type sources which are horizontally offset by 20m,
simulating a rupture of a total length of 200m within 60ms, yielding an estimated rupture speed of about
vr = 3300m/s. They are placed at a depth of 4000m below surface. For simplicity, we work with a
fixed source depth in our analysis. Here we show the results for a circle of 16 seismometers at 1000m
depth. The initial source-receiver distance is 6250m. The ruptures initiation and ending grid points are
I(390,400) and E(410,400). The seismograms show a very strong directivity effect, indicating the rupture
to propagate in positive x-direction, towards station 16 (Figure 4(b)).

For the BPI we smooth the seismograms by an envelope time function and stack for all grid points the
time corrected seismograms for all stations for the time steps of interest as described in Equation ( 1).
As a result we obtain a distribution of back projected energy. To get an estimate of the rupture front, we
track the maximum position of the back projected energy stacks for each projection time step. Figure 5
shows a map view of the back projected energy distribution for six time steps together with the point of
maximum energy stacked. One can clearly see the rupture moving in positive x-direction. Due to the
normalization of all seismograms, the maximum’s value is a good index of the quality of the stack itself
and thus we can choose a specific threshold value to estimate the rupture initiation and ending points. In
this case we chose 70% of the maximal achievable stack value. All points of higher value are defined to
belong to the rupture and are color coded in red (Figure 6). The map view gives a good impression of the
rupture dimensions and orientation. Taking a look at the stack maxima positions against time we find the
mayor movement in positive x-direction. Although the y-position of the BPI maximum moves too, there
is no consistent trend. The rupture length we imaged is of about 200m in x-direction and does not ex-
tend in y-direction. The rupture duration is 40ms, yielding an approximate rupture speed of 5000ms−1.
Compared to the modeled values, we find the rupture length and orientation as modeled 200m in positive
x-direction. Simultaneously the rupture is about 50m shifted to the west. We underestimate the duration
by 20ms, giving us an overestimate of 50% for the rupture speed. This, however, is independent of the
chosen threshold value since even the stack points of lower energy maintain the imaged rupture trend in
speed an direction.

Rupture Propagation Imaging at Microseismic Scale

40

Figure 5: Back projected energy distribution for 6 time steps from t = 30ms to t = 80ms. Time t = 0 represents
theoretical first P-wave arrival for every station. The blue cross is the maximum stack value for that time step. It
clearly moves towards positive x-direction, indicating a progressing rupture.

Figure 6: BPI energy stack maxima points, color coded in red as belonging to the rupture and in black as pre- and
after- rupture radiation energy. The modeled rupture length, orientation and speed are shown by the solid black lines.
Notice the clear trend in (c) whereas there is no consistent trend in (d).

Rupture Propagation Imaging at Microseismic Scale

Annual Report 2012 41

In contrast to BPI, the PPA yields a map view for each time step representing the medium deviation of
all stations back azimuths for each grid point (Figure 7). The track of the best back azimuth fit for all
stations, will give us an estimate of the rupture front’s location (Figure 8). The best stack value is now

Figure 7: Stacked deviation from back azimuth from P-wave polarization analysis for the time steps t = 40ms, t =
60ms, t = 80ms. Time t = 0 represents theoretical first P-wave arrival. The localization of the minimum deviation,
representing the rupture front, moves clearly to the right.

to be as low as possible for in this case we have all stations back azimuths pointing at the precise same
spot. For our analysis approach, we chose a threshold of 0.01 and defined all points below that value as
belonging to the rupture (Figure 8(a)). A clear trend to the east is visible whereas the y-position values
are very stable. Here we estimate a rupture length of 150m within a period of 38ms, giving us a rupture
speed of 3900ms−1. The localization is accurate and the orientation fits with the modeled parameters.
Again we underestimate the rupture duration by 20ms and in this way overestimate the rupture velocity
by 20%.

Figure 8: Minimum distance from back azimuth, stacked for all stations for different time steps. All points of a value
lower than 0.01 belong to the rupture. Notice the clear movement towards the east direction in contrast to the stable
y-position values. The modeled rupture is shown by the solid lines.

Rupture Propagation Imaging at Microseismic Scale

42

Discussion
Comparing BPI and PPA for Setup1 we obtained quite similar results. Orientation and dimension fit with
the modeled rupture, but both methods show inaccuracies regarding the rupture duration and thus the
rupture speed. BPI shifts the rupture a little to the earlier time steps. One can clearly see the gap in the
trend in Figure 6 which is caused by the back projected wave fronts pulling apart from each other towards
their station directions and therefore loosing coherency. Stations in rupture direction will still image the
rupture but those opposite to it will not contribute and for the time window of the separation into two
wave fronts the maximum is "stuck" at around (404,400) till it resumes the trend. PPA does not show
this particular problem. The image here is better, however the overall trend at the rupture endings is not
continued, which could lead to irritations with the imaging of small ruptures. In this particular example
we underestimate the rupture length of 200m with PPA by about 50m. With BPI the rupture length is a
question of calibration, since the overall trend does not differ for marginal time steps. Since we are still in
the process of optimizing our routines, we hope to improve both techniques to reduce those inaccuracies.
The data we modeled and presented, so far, is still idealised in many aspects. We worked with perfect
azimuthal coverage and a homogeneous, isotropic reservoir model, as well as with a fixed source depth.
Even though we adopted the scale and parameters, like station numbers, receiver-station distances and
averaged rock properties of a real geothermal reservoir, we idealised in many ways.
However, our approaches also work for more realistic setups e.g. for the station distribution of the Basel
1 Geothermal Experiment. Using only six stations for the analysis, Figure 9 shows that we can image the
modeled 200m unilateral eastwards rupture using e.g. BPI. Due to the irregular station distribution the
localization for the y-direction is poor, but the movement of the rupture towards the east is still clearly
resolved. The resolution with PPA is likewise poorer than for Setup1. One of the main questions is now
if we can achieve similar results for more complex realistic settings.

Figure 9: BPI energy stack maxima points for the Synthetic Basel 1 Geothermal Experiment Model, color coded in
red as belonging to the rupture and in black as pre- and after- rupture radiation energy. The modeled rupture length,
orientation and speed are shown by the solid black lines. Six irregular distributed stations are used for the BPI.
Rupture propagation in positive x-direction is clearly resolved and consistent but poor in y-direction.

Conclusions
We show that both methods BPI and PPA are capable of tracking a small unilateral rupture of 200m (cor-
responding to a ML =3.4 event) in our microseismic reservoir model, using azimuthally well distributed
stations. The rupture length and direction can be retrieved with an error of <25%. The localization is
acceptable and covers the rupture nucleation point, especially the localization orthogonal to the rupture
direction is precise. Rupture duration and rupture speed show larger inaccuracies and all results depend

Rupture Propagation Imaging at Microseismic Scale

Annual Report 2012 43

on station coverage and distribution. A first application to a realistic monitoring design adopted from the
Basel geothermal reservoir shows, however, that the rupture direction can still be resolved.

Outlook
We see many possibilities to extend our research with the rupture propagation imaging at microseismic
scale. These can be divided into at least four different aspects and are partly worked on already: (I) The
modelling of more complex setups with, e.g., velocity layers and different rupture orientations, (II) the
inhomogeneous station distribution and station weighting, (III) the possible combination of both tech-
niques to increase the accuracy due to two quality measures and, finally, (IV) the application to real data
from, e.g., the Basel Geothermal Experiment.

Acknowledgments
We thank the sponsors of the PHASE consortium for supporting the research presented in this paper.

References
Bayer, B., Kind, R., Hoffmann, M., Yuan, X., and Meier, T. (2012). Tracking unilateral earthquake
rupture by P-wave polarization analysis. GJI, 188:1141–1153.
Häring, M., Schanz, U., Ladner, F., and Dyer, B. C. (2008). Characterisation of the Basel 1 enhanced
geothermal system. Geothermics, page doi:10.1016/j.geothermics.2008.06.002.
Ishii, M., Shearer, P., Houston, H., and Vidale, J. (2007). Telseseismic P wave imaging of the 26 Decem-
ber 2004 Sumatra-Andaman and 28 March 2005 Sumatra earthquake ruptures using the Hi-net array.
JGR, 112.
Jurkevic, A. (1988). Polarization analysis of three-component array data. BSSA, 78:1725–1743.
Kao, H. and Shan, S.-J. (2007). Rapid identification of earthquanek rupture plane using Source-Scanning
Algorithm. GJI, 168:1011–1020.
Rentsch, S. (2007). A Migration-Type Approach for the Fast Location of Seismicity: Theory and Apllica-
tions. PhD thesis, Freie Universität Berlin.
Saenger, E., Gold, N., and Shapiro, S. (1999). Modeling the propagation of elastic waves using a modified
finite-difference grid. Wave Motion, 31:77–92.
Walker, K. and Shearer, P. (2009). Illuminating the near-sonic ruptur velocities of the intracontinental
Kokoxili Mw 7.8 and Denali fault Mw 7.9 strike-slip earthquakes with global P wave back projection
imaging. JGR, 114.

Rupture Propagation Imaging at Microseismic Scale

44

Rupture Propagation Imaging at Microseismic Scale

Annual Report 2012 45

Simultaneous Inversion of Anisotropic Velocity Model and

Microseismic Event Location: Synthetic and Real Data Examples

C. P. Yu1 , A. Reshetnikov1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
In this study we develop a practical method for the location of microseismic events in anisotropic me-
dia. The velocity anisotropy is taken into account in the additional step after the conventional location
procedure. In this step, the isotropic velocity model and corresponding estimated locations serve as an
initial model, and the anisotropy parameters as well as resulting location misfits are inverted simultane-
ously due to the coupling of velocity model and hypocenter parameters in calculation. After obtaining the
anisotropic velocity model, one can relocate all the events around their isotropic velocity based locations.
The method is ray theory based and only consider transversely isotropic layered model. The nonlinear
inversion problem is solved by using Gauss-Newton method. The results of synthetic example closely fit
with the true values and event depths are better determined than radial distances. Real data test shows that
the estimated Thomsen parameters are consistent with previous study, and the isotropic velocity based
locations are significantly corrected with large decrease of average misfit from 52m to 6m.

Introduction
As shown in Yu et al. (2012), the influence of anisotropy on the locations of microseismic events can be
significant, even for weak anisotropy. Also we found that in one data set provided by one of our sponsors,
the velocity anisotropy can be up to 35% in shale formations. So working with anisotropy can not be
neglected in microseismic studies.
In Reshetnikov et al. (2012), the anisotropic velocity model is estimated using global searching method
by fitting the perforation shots locations. Now we extend this work to inversion based method, which
is able to handle complex model geometry and use microseismic event data in addition to perforation
shot. The unknowns include event locations and anisotropic velocity model, which need to be updated
simultaneously because of the coupling between them (Thurber, 1993).
Grechka et al. (2011) attempt to simultaneously invert elastic coefficients along with microseismic event
location for arbitrary anisotropy media. Yaskevich and Duchkov (2012) perform a simultaneous event
location and velocity model building in HTI media for synthetic data, intended to evaluate the influence
of anisotropy on fracture imaging. The aim of this study is to use the simultaneous inversion strategy
to build a practical procedure for microseismic event location in anisotropic media, and then test the
feasibility of this procedure using synthetic and real data.

Multi-Step Location Strategy

The conventional location procedure usually consists of event picking, perforation shot picking and
isotropic velocity model building using sonic logs and perforation shots data. Then the events can be
located and origin times are estimated. After this step one already have the optimal isotropic velocity
model and event locations, although there may exist large misfits due to the absence of anisotropy in
calculation.

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location: Synthetic and Real Data
Examples
46

When considering anisotropy, we need to estimate anisotropy parameters and then to estimate the result-
ing perturbations of location and origin time. To obtain the additional anisotropy parameters there are
two possibilities by using different types of data:
a) By only using perforation shots data we can perform anisotropic tomography like in cross-well case.
But the result may suffer from bad ray coverage if there are not enough perforation shots, and also only
P-wave related anisotropy parameters can be inverted from perforation shot data.
b) we can use microseismic events data in addition to perforation shots to improve the spatial coverage
and estimate the S-wave related anisotropy parameters. As a result, we have to add the event locations to
model vector and update them simultaneously with anisotropy parameters. The optimal isotropic velocity
model and corresponding event locations can be used as initial model of the simultaneous inversion.
After obtaining the anisotropy parameters, microseismic events can be relocated using the same method as
in conventional location step. Although event locations can be estimated simultaneously with anisotropy
parameters as discussed above, the inversion will be unstable if there are too many unknowns. So we
choose the relocation strategy rather than putting all the events into simultaneous inversion.
Finally we propose the following multi-step location procedure in anisotropic media (see Figure 1). Since
the approaches used in conventional location and relocation steps have been well developed (see Kum-
merow et al., 2011; Kummerow and Shapiro, 2010), we only focus on simultaneous inversion step in the
following.

Figure 1: Flow-chart of the multi-step location procedure for anisotropic media.

Methodology

Model Representation

Model defined on a grid is used in forward modeling which can represent lateral inhomogeneity. But in
the following synthetic and real data tests, we only consider a layer model, either anisotropic or isotropic.
Here we only consider transverse isotropy, because it is geologically reasonable in most of our study
cases.
The anisotropic velocity model is described by Thomsen parameters VT =[VP 0 , VS0 , ǫ, δ, γ] or density-
normalized elastic-moduli VC =[C33 , C55 , C11 , C13 , C66 ]. The phase velocity of different wave types
have different sensitivities to these two types of model parameters over different wavefront-normal direc-
tion range (see Figure 2). One must properly choose model parameters based on recording geometry and
the wave types used.

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location: Synthetic and Real Data
Examples
Annual Report 2012 47

qP Wave qP Wave
3 0.2
VP0 c33
VS0 c55
2 0.15 c11
ε c13
δ
1 0.1

0 0.05

−1 0
−90 −45 0 45 90 −90 −45 0 45 90

qSV Wave qSV Wave

Derivatives of phase velocity

Derivatives of phase velocity

2 0.6
V c33
P0
VS0 c55
1 0.4 c11
ε c13
0 δ 0.2

−1 0

−2 −0.2
−90 −45 0 45 90 −90 −45 0 45 90

qSH Wave qSH Wave

1.5 0.4
VS0 c55
c66
γ 0.3
1

0.2

0.5
0.1

0 0
−90 −45 0 45 90 −90 −45 0 45 90
Slowness vector direction( ° ) Slowness vector direction( ° )

Figure 2: Derivatives of phase velocities with respect to Thomsen parameters and elastic-moduli over the angle range
0◦ ∼180◦ from the vertical direction in homogeneous VTI media(VP =2.755km/s, VS =1.290km/s, ǫ=0.125, δ=-0.075,
γ=0.100 or equivalent C33 =7.59, C55 =1.66, C11 =9.49, C13 =3.66, C66 =2.00).

Inversion Strategy
The simultaneous inversion of anisotropic velocity model and event location can be expressed as nonlinear
least-square problem with the misfit function
 2
XNe N
X t (i)

tcal obs 


f (m) =  i,j (m) − ti,j , (1)
i=1 j=1

where tobs are picked travel times; tcal are calculated travel times; Ne is the number of events; Nt (i) is
the number of time picks of the ith event.
The model vector is m = [Vl , xe , τe ], where Vl is the velocity parameters of the lth layer, and xe and τe
are the location and origin time of the eth event. The Frechét derivatives of travel-times with respect to
model parameters are derived from ray path integral equation
Z xr
cal
T (m) = τ + p(m)ds, (2)
xs

where xs and xr are source and receiver locations; p is slowness vector; s is ray path vector.
We solve the nonlinear least-square problem by using Gauss-Newton method

mn+1 = mn − αH−1
a ∇C(mn ), (3)

where H−1a is the approximated Hessian matrix, and ∇C(mn ) is the gradient vector of misfit function,
and α is the updating step length. The approximated Hessian matrix is used to precondition the steepest
descent direction, which not only significantly accelerate the convergence of misfit function but also
balance the contributions of different types of unknowns in model vector.

Travel-Time Calculation
Travel time is calculated by anisotropic ray-tracing algorithm, adapted from Gajewski and Pšeník (1987).
Since the algorithm is based on Christoffel equations, it does not work properly in the singular region
where the two qS waves propagate with nearly the same phase velocities. The situation can become
worse if the anisotropic media is heterogeneous (see Yu et al., 2012).

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location: Synthetic and Real Data
Examples
48

To overcome this problem, we developed a "polarity correction" method to make sure the two qS waves
have correct phase velocities when crossing the singular region. The analyses show the adapted program
works well in highly heterogeneous anisotropic media.

Synthetic Example

To test the feasibility of multi-step location procedure, we conduct a synthetic data test using the real
recording geometry of a hydraulic fracturing (see Figure 3). Eight pseudo events are placed at the posi-
tions of eight perforation shots. It is assumed the conventional location step has been completed and the
locations obtained are chosen randomly within 50m of the true locations (see Figure 4). We invert the
true event locations simultaneously with anisotropy parameters from the initial locations and isotropic
velocity model.

Distance scale(m)
0 50
Receivers
Sources

S1

R1
X scale(m)
R2 S2
0 50
0
Depth scale(m)

R3
S3

50 R4 S4

R5
S5
Y scale(m)

100 R6 0

S6
R7
50
R8 S7

S8 Receivers
Sources

Figure 3: The recording geometry of synthetic example, section view (left), top view (right).

Since it is a single vertical well monitoring system, we have to locate the events in cylindrical coordinate
system and the event azimuth can be determined by hodogram analysis. So the model vector is

m = [Vl , re , ze , τe ],

where re and ze are radial distances and depths of the events.

The anisotropic velocity model used in synthetic test is homogeneous and characterized by five density-
normalized elastic moduli. To increase the complexity of synthetic test, we also add random perturba-
tion(<10%) to the isotropic velocities and invert all the five velocity parameters. The initial velocity
model is V=[C33 +Error, C55 +Error, 0, 0, 0].
As shown in Figure 4, the inverted event locations and anisotropic velocity model closely fit with the true
values. To make it morn explicit, the elastic moduli are converted into velocity and Thomsen parameters.
Figure 5 are the convergence curves of model parameters and data misfit. We can see the rapid decrease of
data misfit in the first iterations due to the preconditioning of approximated Hessian matrix. The similar
oscillations occurring in the convergence curves of three type of model parameters (e.g. peak in the 4th
iteration) clearly indicate the underlying coupling of them in calculation. However the oscillations of
z-coordinate’s convergence curve is much weaker. The reason is that the events mainly distribute along
z direction and their depths vary a lot with respect to receiver array, which means the ze can be better
constrained than re in such recording system.

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location: Synthetic and Real Data
Examples
Annual Report 2012 49

Distance scale(m)
0 50 100
Receivers Initial Inverted True
True Event Locations
Initial Event Locations VP (m/s) 2731 2755.224 2755
Invertd Event Locations VS (m/s) 1387 1290.058 1290
ǫ 0 0.125263 0.125
δ 0 -0.074797 -0.075
0
γ 0 0.100021 0.100
Depth scale(m)

50

100

Figure 4: The inverted event locations and anisotropic velocity model (converted from elastic moduli) for synthetic
data.
1.4 40
C33 r−coordinate
C55 35 z−coordinate
1.2 C11 Distance
C13
Average Locations Misfits (m)
Elastic Moduli Misfits(km2/s2)

C66 30
1

25
0.8
20
0.6
15

0.4
10

0.2 5

0 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

25 30
T0 qP wave
qSV wave
25 qSH wave
Average Origin Times Misfits (ms)

20

20
Data Misfits (ms)

15

15

10
10

5
5

0 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Iteration Times Iteration Times

Figure 5: Convergence curves of model parameters and data misfit for synthetic data.

Real example
Now we test the multi-step location procedure using the same real dataset with Reshetnikov et al. (2012).
Reshetnikov et al. (2012) built an optimal isotropic velocity model using scaled sonic logs, but the per-
foration shots locations with such velocity model shown average misfit of 54m (see Figure 6). Then they
performed a global searching to estimate the anisotropy parameters by fitting the perforation shots loca-
tions and significantly improved the results. We use their results including the optimal isotropic velocity
model and corresponding perforation shots locations with large misfit (see Figure 6) as the initial model
to invert the anisotropy parameters and the true locations of perforation shots.

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location: Synthetic and Real Data
Examples
50

The recording system is exactly the same as used in synthetic test (see Figure 3). We select 16 perforation
shots as our targets. Even though the isotropic velocity is highly heterogeneous in vertical direction (see
Figure 6), we do not include any heterogeneity in anisotropy parameters. Because as demonstrated in
Reshetnikov et al. (2012), the anisotropy of target depth interval is weak. So the model vector becomes
m = [ǫ, δ, re , ze , τe ]. The third Thomsen parameter γ cannot be determined only using P-wave from
perforation shot data.
Velocity(m/s) Distance scale(m)
500 1000 1500 2000 2500 3000 3500 0 50 100
Vp Receivers
Vs True Locations
Initial Locations

0 0
Depth Scale(m)

Depth scale(m)
50 50

100 100

Figure 6: The optimal isotropic velocity model and corresponding estimated locations with large misfits.

First we remove re and ze from model vector and conduct a cross-well like tomography with perforation
shots fixed in their true locations (see Figure 7). The optimal values of Thomsen parameters are ǫ=0.0235,
δ=-0.0902, which exactly lie in the optimal range obtained in Reshetnikov et al. (2012). This result can
be used as a reference for the following simultaneous inversion.

0.04 7.215
Epsilon T0
Delta
0.02
7.21
Average Origin Time (ms)

0
Thomsen Parameters

7.205

−0.02
7.2
−0.04

7.195
−0.06

7.19
−0.08

−0.1 7.185
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Iteration Times Iteration Times

Figure 7: Thomsen parameters and average origin time obtained by tomography with perforation shots fixed on their
true locations.

Then we perform a simultaneous inversion of Thomsen parameters and the true locations of 16 perforation
shots. The misfit function reaches the minimum when ǫ=0.0113, δ=-0.1135 (see Figure 8). Despite the
small differences with reference values, the inverted Thomsen parameters are still in the optimal range
obtained in Reshetnikov et al. (2012) and are acceptable in weakly anisotropic case. On the other hand,
the locations of perforation shots are significantly shifted from isotropic velocity based result towards the
true locations (see Figure 9), with a large decrease of average misfit from 52m to 6m (see Figure 8).

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location: Synthetic and Real Data
Examples
Annual Report 2012 51

0.02 60
r−coordinate
Epsilon z−coordinate
0 Distance
50
Delta

Average Locations Misfits (m)

−0.02
Thomsen Parameters

40

−0.04
30
−0.06

20
−0.08

10
−0.1

−0.12 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Iteration Times Iteration Times

Figure 8: Thomsen parameters and average location misfits obtained by simultaneous inversion

Distance scale(m)
0 50 100
Receivers
True Locations
Initial Locations
Invertd Locations

0
Depth scale(m)

50

100

Figure 9: The perforation shots location result of real data test.

Conclusions

We built a multi-step procedure for the location of microseismic events in anisotropic media. The pro-
cedure consists of conventional location, simultaneous inversion and relocation, among which the simul-
taneous inversion is mainly discussed in this study because the rest two parts use the similar approaches
that have been well developed. The method is based on transversely isotropic layered model and ray the-
ory, which satisfy most of our study cases. Gauss-newton method is used to solve the nonlinear inversion
problem.
We used synthetic data from a real recording system to test the simultaneous inversion of anisotropic
velocity model and event locations. All the results fit nicely with the true values. The event depths are
better constrained than radial distances. Because the pseudo events mainly distribute along z direction and
event depths vary a lot with respect to receiver array. So the recording geometry can effect the inversion
behavior in this method.
We also conducted a real data test for weak anisotropy case. The estimated Thomsen parameters are con-
sistent with the result of previous study using the same dataset. The incorrect locations are significantly
shifted from isotropic velocity based result towards the true locations, with a large decrease of average
misfit from 52m to 6m.

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location: Synthetic and Real Data
Examples
52

Acknowledgments
We thank the sponsors of the PHASE consortium for supporting the research presented in this paper.

References
Gajewski, D. and Pšeník, I. (1987). Computation of high-frequency seismic wavefields in 3-d laterally
inhomogeneous anisotropic media. Geophysical Journal of the Royal Astronomical Society, 91(2):383–
411.
Grechka, V., Singh, P., and Das, I. (2011). Estimation of effective anisotropy simultaneously with loca-
tions of microseismic events. Geophysics, 76(6):WC143–WC155.
Kummerow, J. and Shapiro, S. (2010). Microseismic event relocation using arrival times and cross-
correlation coefficients: A synthetic study. Expanded abstracts, EAGE 72nd annual meeting and tech-
nical exhibition, Barcelona.
Kummerow, J., Shapiro, S., Asanuma, H., and Häring, M. (2011). Application of an arrival time and
cross correlation value-based location algorithm to the basel 1 microseismic data. Expanded abstracts,
EAGE 73rd annual meeting and technical exhibition, Vienna.
Reshetnikov, A., Kummerow, J., and Shapiro, S. A. (2012). Location of induced microseismic events
using single vertical borehole array. 7th Annual PHASE Report.
Thurber, C. H. (1993). Local earthquake tomography: velocity and Vp/Vs-theroy, in Seismic tomography,
Theory and practice: Iyer, H. M. and Hirahara K. Chapman and Hall, London.
Yaskevich, S. V. and Duchkov, A. A. (2012). Simultaneous microseismic event location and anisotropic
velocity model building in hti media. Expanded abstracts, EAGE 74th annual meeting and technical
exhibition, Copenhagen.
Yu, C. P., Reshetnikov, A., and Shapiro, S. A. (2012). The influence of anisotropy on the location of
microseismic events. 7th Annual PHASE Report.

Simultaneous Inversion of Anisotropic Velocity Model and Microseismic Event Location: Synthetic and Real Data
Examples
Annual Report 2012 53

Microseismic Reflection Imaging: Methodology and Applications

A. Reshetnikov1 , J. Kummerow1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
In this work we present an approach which uses waveforms from induced seismicity to build a detailed
high resolution image of a stimulated reservoir. The idea of Microseismic Reflection Imaging (MRI) is to
treat a located microseismic event as an active seismic source and to apply migration techniques adapted
from reflection seismics. For imaging we use the Fresnel-Volume-Migration method. This is a directional
migration approach which takes advantage of wavefield polarization to restrict the migration operator to a
region around the actual reflection point. It helps to reduce migration artefacts caused by limited aperture.
To obtain polarization of the wavefield, we consider auto- and cross-variances of time samples within a
time window, including several dominant periods of the P- or S-wave. We also apply a hodogram linearity
threshold value to exclude parts of the seismograms with unreliable polarity estimates.
We apply the MRI approach to two datasets where microseismic waveforms were recorded at a borehole
array and at 6 of shallow (about 200 m) and deep (2.5 km) receivers. For first dataset we compare the
images of the same site obtained by a conventional surface seismic migration method and by the MRI
approach. For the richest microseismic dataset from the stimulation of Enhanced Geothermal System at
Basel, Switzerland we demonstrate separate images of PP and SS reflections for 5 different receivers. We
show how the image quality degrades with increasing distance between stimulated volume and receiver.
Finally, we provide a joint interpretation of obtained images and event locations. We discuss how micro-
seismic reflection imaging can complement to surface seismic imaging, what the reflectors in microseis-
mic reflection image can tell us about the stimulated reservoir and how these images can contribute to the
reservoir characterization.

Introduction
The focus of this work is to elaborate the method which allows one to image reflections within seis-
mograms of microseismic events. The reflection-seismic method is well established approach widely
applied in exploration seismology since 1930s (Bednar, 2005). Since that time there were developed a
numerous amount of migration algorithms based on different physical approximations. Unfortunately,
conventional migration techniques aimed to imaging land, marine or VSP data are not directly applicable
for microseismic reflection imaging because of unique particularities of microseismic data.
Particularly, in contrast to active seismic, microseismic events are not controlled sources, though their
location in a case of proper acquisition system can be determined precisely. Moreover, the area of interest
is likely to be the same seismically active zone where microseismicity occurs. This means that the sources
for MRI are located in the direct vicinity of a target and at a distance from seismic sensors.
Acquisition geometries used for microseismic monitoring in many cases are different from geometries
of land seismic or VSP. Since the main purpose of microseismic monitoring systems is to register arrival
times of events, the number of receivers is usually much smaller and they are sparsely distributed which
limits possibility of ant post-stack imaging. In the case of induced seismicity, number of events can reach
hundreds or even thousands which is usually one or two orders of magnitude larger than the number of
sensors.
The sources used in active seismic are different kinds of explosions, vibrators or air guns with well known
and in most cases spherical radiation patterns. Microseismic events are small earthquakes with complex
focal mechanisms which are initially unknown. Moreover, focal mechanisms as well as source energies
can vary significantly even for events located close to each other.

Microseismic Reflection Imaging: Methodology and Applications

54

Methodology
Previous studies
To our knowledge, the first attempt to use reflected waves in an earthquake seismogram to characterise
a subsurface was performed by Sanford et al. (1973). Authors analysed microearthquake seismic data
recorded by sensors located near Rio Grande rift, New Mexico. Having event locations, they identified
SP and SS reflected phases and mapped a sharp dipping discontinuity at a depth of 18 to 30 km. The
mapping process didn’t involve any sophisticated migration procedure and was based on the reflected
wave travel time analysis. Nevertheless, authors have demonstrated the presence of the reflected waves
in microseismic coda and therefore a possibility to image microseismic reflection data.
Soma and Niitsuma (1997) and Soma et al. (1997) firstly proposed the technique dedicated to imaging
reflections in microseismic data. The technique was called Acoustic Emission (AE) reflection method.
To identify reflected phases they used the analysis of 3D hodogram in time domain. More precisely
they compute global polarisation coefficient (Samson (1977)) as a function of time for each 3-component
seismogram from a microseismic event. Authors assumed that time periods of coherent wave arrivals
such as direct or reflected waves were highly linear as opposed to non-linear incoherent random noise. To
find a position of a reflector they performed diffraction stack migration, but instead of stacking reflected
wave amplitudes or energies, they stacked polarisation coefficient (further denoted as linearity) values.
These values were stacked on the iso-delay ellipsoid corresponding to the reflected wave travel time.
Moreover, stacking were restricted to a narrow band on the iso-delay ellipsoid by taking into account
polarisation of the reflected wave. The final image was obtained by stacking the linearity over all possible
time delays and for all microseismic events. Soma et al. (2002a) further updated the AE reflection method
by proposing to construct a spectral matrix of 3D particle motion and considered linearity in both time
and frequency domains. To get the image in the extended version of AE reflection method linearity
values were stacked over all possible time delays and frequency components. It was done to detect more
reflectors and improve the resolution of the final image. Later Tamakawa et al. (2010) and Asanuma et al.
(2011) proposed to use microseismic multiplets as sources for AE reflection method to image reflectors
more accurately using additional information about relative times between multiplet events.
AE reflection method was applied to the data from several geothermal projects such as Kokkonda geother-
mal field (Soma and Niitsuma, 1997) and Soultz Hot Dry Rock experiment (Soma et al., 1997, 2002a,
2007). The most complete results are presented in Soma et al. (2007) where authors integrated the refec-
tion images obtained by the AE reflection method using the Soultz data from 1993 to 2003. The authors
showed that it is possible to get an the image of the reflectors in the vicinity of microseismic sources,
although the image appeared to be piecewise spots of reflectivity which did not amalgamate into the lin-
ear structures, while the reflectors below the seismically active zone are much more solid. The authors
reported that the method was not efficient for imaging structures close to the sources.
In general, there are several issues which limit applicability of the AE reflection method: the AE reflection
method assumes a homogeneous and isotropic velocity model. Besides that, in the published papers, the
method is only applied to S- reflected waves which is explained by much stronger energy of S-waves
which dominates in the observations and can cover the P reflected waves (i.e. Soma et al. (2002b)).
Though PP reflections potentially could provide much more accurate polarisation estimates and in some
cases can be observed in a coda between direct P- and S-waves. The most serious drawback of the AE
reflection method comes from the assumption that high linearity at all frequencies directly indicates the
presence of P– or S-wave for which one can estimate correct polarisation. While coherent reflected waves
in a 3-component seismogram are indeed expected to be highly linear, in general case high linearity of a
signal does not necessarily implies the presence of a coherent wave. For example, it is quite usual that the
coda for a microseismic event is recorded on a background of coherent artificial noise with energies of
the same order of magnitude as the useful signal. Additionally, the frequency bands of a noise and useful
signal are most likely to be overlapping which means that the polarity estimates of linear hodogram
interpreted as a reflected wave are not equally correct for different parts of frequency spectrum. All these
lead to potential misinterpretation of the coda and inaccurate polarity estimates in the case of correctly
detected reflected wave at frequencies where an artificial noise is strong enough.
There are also several publications where authors apply conventional imaging techniques to the micro-
seismic reflection data without adaptation. For instance, Chavarria et al. (2003) presented their results

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 55

of the imaging of data from a number of microearthquakes recorded at SAFOD pilot borehole. Similar
approach has been described in Dyer et al. (2008) for the area around the Upper Rhine Graben near Basel,
where the velocity model is derived directly from the microseismic data, demonstrating the imaging of
potential fractures by migration of the corresponding microseismic data. These results have shown that in
some cases it is possible to get an image of large scale geological structures nearby of seismically active
zone. Though, it is problematic to resolve the internal structure of seismically active zone.

Concept of Microseismic Reflection Imaging

It is clear that small microseismic sources localised in relatively small area cannot compete with active
seismic sources in imaging of large scale structures. However, microseismicity has the unique advan-
tage: it is located directly in the seismically active zone, for instance inside of a stimulated reservoir.
Therefore, in the case of proper acquisition, microseismic data potentially can contain high-frequency
reflected waves scattered in the local heterogeneities of the reservoir. This means that by applying spe-
cialised imaging technique one can get the image of a reservoir interior with resolutions unachievable by
conventional surface seismic.
In order to take into account the specificity of microseismic data and microseismic monitoring systems
the Microseismic Reflection Imaging is based on the following principles:

• use only microseismic events with small location misfits and high signal to noise ratio;

• use a directional migration approach to solve the problem of limited aperture;

• determine the frequency range for which the signal provides the most accurate estimates of the
polarisation;

• use only those parts of seismogram for which one can estimate accurate polarisation information;

• use restricted imaging volumes in order to reduce the contribution of misinterpreted waves in to the
final image.

Event selection and sorting

The Microseismic Reflection Imaging approach employs a directional imaging algorithm, which means
necessity of polarisation estimation. In order to estimate polarity of a reflected wave, it is required that
the corresponding wavelet is above the noise level, while in order to locate an event, one has to just
detect the first break times (in the case of travel time based location algorithms). This makes the imaging
approach much more sensitive to the quality of seismic data compared to the applied microseismic event
location procedure. Therefore, not all locatable events are suitable to be taken as sources for the imaging
procedure. It is necessary to sort out events with large RMS misfits and weak signals.
In most cases the number of considered microsesmic events is much larger than the number of sensors.
In addition to that, the quality of recorded signal can vary at different receivers. Thus, it makes sense to
process seismic traces recorded at different sensors individually.
To facilitate further processing and imaging, the selected traces are grouped into common-receiver gath-
ers. Since the imaging procedure works with each trace separately by stacking the final result from the
images of individual traces, the order of the traces in a gather doesn’t play any role. But for the sake of
visual assessment and interpretation of the data it makes sense to sort traces in a way, that makes it easy to
identify possible coherent reflections within a gather. At the beginning all the traces are shifted in a way
that time zero coincides with event triggering time. Then we apply the following sorting algorithm: the
first trace is taken for the event at the edge of the cloud of events selected for a given receiver. Each next
trace is taken for the event which is the closest one to the event corresponding to the previous seismo-
gram. A gather sorted this way consists of traces produced by groups of events which are located close to
each other. Two neighbour traces are most likely to have similar travel times and the coda. An example
of common receiver gather produced from induced microseismic events recorded at one sensor is shown
in Figure 1.

Microseismic Reflection Imaging: Methodology and Applications

56

Event number
0 10 20 30 40 50 60 70 80 90 100
0.8

1.0

1.2

1.4
Time (sec)

1.6

1.8

2.0

2.2

2.4
STJ

Figure 1: An example of common receiver gather produced from hydraulically induced seismicity.

Polarisation estimation
In most cases the spatial dimensions of a microseismic cloud are small or comparable to the distances
from events to a monitoring system. Accordingly, the aperture of scattered waves for such system is low
in comparison to wide-angle active seismic data. In this situation the only way to get a resolved image of
the cloud interior is to take into account polarity of the reflected waves.
To obtain a wave propagation direction of a 3-component seismic trace, we use auto- and cross-variances
of time samples within a time window which have to include several dominant periods of P- or S-wave
(Jurkevics, 1988)
T + 12 N
1 X
Cij = ui (n)uj (n) i, j = {1, 2, 3}
N
n=T − 12 N

where i and j are seismic trace component indexes, ui (n) is n-th sample value of i-th component of a
trace, T is a time sample number for which the direction is being estimated, N is a time window length.
By solving the eigenvalue problem

(C − λI)p = 0 λ1 ≤ λ2 ≤ λ3
one can get corresponding eigenvectors and eigenvalues. Then, the eigenvector − →
p3 corresponding to
the largest eigenvalue specifies the direction of linear component of hodogram and therefore represents
P-wave polarisation at the current time. To specify S-wave propagation direction we use eigenvector
−
→
p1 corresponding to the smallest eigenvalue which by definition is orthogonal to the particle motion
hodogram (see Figure 2).

P-wave S-wave

Figure 2: Hodograms of P- and S-waves for a three component seismic data and corresponding wave propagation
directions.

One can also formulate the quality control functions for estimates of wave propagation direction. For
P-waves, it is clear that the more linear shape of hodogram leads to the more accurate estimate of P-wave
polarity. The linearity of hodogram can be defined as follows (adapted from Samson (1977))

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 57

(λ1 − λ2 )2 + (λ2 − λ3 )2 + (λ3 − λ1 )2

R= .
2(λ1 + λ2 + λ3 )2
Linearity has small positive values for non-coherent noise and tends to 1 for the case of perfectly linear
signal (λ1 ≪ λ3 and λ2 ≪ λ3 ).
We also introduce the quality control function for S-wave propagation direction estimates. Direction of
S-wave can be accurately estimated if the hodogram is flat and circular. This condition is satisfied in the
case when λ2 ≫ λ1 (then λ3 ≫ λ1 by definition). Therefore, hodogram “flatness” can be defined as

λ22 + (λ2 − λ1 )2
C= .
2(λ2 + λ1 )2
Hodogram “flatness” values vary from 12 which indicates linearly polarized or non-coherent signal and
1 which indicates that hodogram is flat and circular and hence the orthogonal direction can be estimated
accurately.
Estimated values of linearity and “flatness” as function of time for each seismogram are then used to
filter out parts of seismic data for which it is not possible to estimate reliable wave propagation direction.
Figure 3 illustrates this procedure. Only parts of the seismogram which have linearity larger than a given
threshold (marked with red bars in Figure 3) are taken for further imaging. It helps to reduce the noise
and migration artefacts in the final image.

Figure 3: Seismograms from three microseismic events and corresponding P-wave linearity values as function of
time.

Analysis of linearity and flatness of a hodogram is helpful for detecting consistent signals. The fact that
at some moment of time and at some frequency linearity of hodogram is high likely indicates that there
is a coherent signal at this time. But it doesn’t guarantee the correctness of polarity estimation for this
frequency. For example, usually signals are more linear at lower frequencies, but polarities estimated at
lower frequencies could be wrong.
In order to get accurate directivity estimates we propose to make an additional processing step. The
goal of this procedure is to estimate frequency band which provides the most reliable reflected wave
propagation direction estimates. But there is no a priori information about the position of reflectors and
therefore information about reflected wave incident angles. Though, coordinates of sources and receivers
are known and therefore theoretical polarity of direct waves can be calculated, for example by tracing
rays from microseismic sources to a receiver. Assuming that the frequency bands of direct and reflected

Microseismic Reflection Imaging: Methodology and Applications

58

waves of the same type are close to each other, one can build an optimisation procedure where variables
are parameters of the filter (see Figure 4) and the criterion is the difference between the empirical and
theoretical polarity estimates of the direct wave.

A
1

F
X-S X Y Y+S

A
1

F
F
F-S-W/2 F-W/2 F+W/2 F+S+W/2

Figure 4: Parameters of band-pass and notch frequency filters used for the optimisation.

The complete workflow of wave propagation direction estimation applied to each receiver separately and
consists of 4 steps:

1. Use optimisation procedure to find a frequency filter which provides the most accurate polarity
estimates by minimisation of the misfit between empirical and predicted polarisations of the direct
wave. It is done for P- and S-waves independently. The by-product of this optimisation procedure
is the sensor azimuthal orientation (in the case when it is unknown).
2. Use obtained angle to rotate seismic data recorded at particular sensor from local XYZ to the
geographical ENZ coordinates and apply the frequency filters to get two filtered gathers optimised
for P- and for S-wave polarity estimation.
3. Use filtered data to estimate polarity of each time sample of each 3-component seismogram for
both wave-types.
4. Use linearity (for P-waves) and hodogram flatness (for S-wave) threshold to exclude parts of traces
which have unreliable polarity estimates (see Chapter 2 for more details).

As a result of this workflow for each coherent part of a seismogram we get two direction estimates: one
from the assumption that it is the P-wave and another one from the assumption that it is the S-wave.
While, unreliable noisy parts of seismogram are sorted out by linearity/flatness analysis.

Directional Mirgation Approach

To construct images we employ the Fresnel-Volume-Migration method (Lüth et al., 2005; Buske et al.,
2006). This is an extension of the pre-stack Kirchhoff depth migration which uses the wavefield po-
larisation information to restrict the migration operator to a region around the actual reflection point.
The procedure itself resolves spatial ambiguities caused by, for instance, limited aperture and results in
strongly reduced migration artefacts, particularly for data sets with low coverage.

Reflection imaging using natural seismicity

The SAFOD boreholes are located near Parkfield (California, USA) approximately two kilometers away
from the known surface trace of the San Andreas Fault (Figure 5). The aim of the SAFOD project is
to study the physical and chemical processes that control deformation and earthquake generation within

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 59

SW NE
SAFOD
faults 0
San Andreas
Fault

Sea ?
level
1000
San Andreas

TVD (m)
fault zone

36° Receiver
array
2000 Pilot Hole ?

?
SAFOD Main Hole
drill holes
3000
N

35° 57' 30''

1 km
120° 33' 30'' 120° 30' 0 1000 2000 3000
Distance from SAFOD wellhead (m)
(a) (b)

Figure 5: (a) Faults map of the area in the vicinity of the SAFOD drill holes (geometry of faults is taken from Brad-
bury et al. (2007)), the blue line represents direction of the cross-section. (b) Fault perpendicular cross-section around
SAFOD boreholes (geologic interpretation is taken from Zoback et al. (2010)) and location of the 6 earthquakes an-
alyzed further.

2 3 4 6
1 5

5 0 0 m

Figure 6: Images from the six selected earthquakes and borehole with color-coded lithology information (east (X),
north (Y) and vertical (Z) components), sorted by a recorded signal strength from high to low.

an active plate-bounding fault zone (Zoback et al., 1998). In this area the movement of the SAF is
characterized by a combination of repeating microseismicity clusters and aseismic creep. In order to
investigate these processes in situ, a 2.2 km deep vertical pilot hole was drilled in 2002, followed by
the main borehole in 2004–2005, which was drilled vertically down to a depth of 0.9 km where it was
deviated towards the San-Andreas-Fault reaching a final depth of 2.4 km. During one of the several
drilling phases in 2005, the main borehole was equipped with an 80-level-3C-borehole-receiver array by
Paulsson Geophysical Services (PGS/I). The array was deployed in the inclined part of the borehole at a
depth interval of 878 – 1703 m below sea level with 15 m receiver spacing.
We apply our approach to 6 events, recorded between May 1 and May 8, 2005, in the SAFOD main
borehole. For selected data sets we performe the event detection procedure, locate these events and then
migrate the PP reflection data for each of them. For the location and imaging we used a known P-velocity
model of Thurber et al. (2004), to get the S-velocity model an effective Vp-Vs ratio of 1.83 was used.
We migrate the PP reflections of these microearthquakes. The Fresnel-Volume-Migration images for the
six events obtained from the east, north and vertical components, sorted by a recorded signal strength from
high to low, are shown in Figure 6. These images reveal a network of reflectors mutually complementing

Microseismic Reflection Imaging: Methodology and Applications

60

X component Y component Z component

1000 1000

1500 1500

2000 2000

2500 2500

0 500 1000 0 500 1000 5 500 1000

0
Figure 7: Cross section of the stacked image obtained from the east (X), north (Y) and vertical (Z) components for
the six events. Thin solid black lines - obtained interface map.

each other in the vicinity of the SAFOD main borehole.

Six obtained images were then stacked to the final image for three components separately. A cross section
of stacked images is shown in Figure 7. One can see that the resulting images are partially similar to each
other, while in some other parts they are clearly complementary. In addition to the strongest reflectors
presented in Figure 6, several less pronounced reflectors appeare at larger distances from the borehole.
The solid black lines in Figure 7 summarize these reflectors as an interface map obtained from the stacked
sections of the three migrated wavefield components.
The image shows a complex reflector system in the direct vicinity of the SAFOD main borehole. This is
a high resolution reflection seismic image of the small-scale environment of the San-Andreas-Fault.

Comparison with other imaging results

Several active- and passive-source seismic imaging investigations around the SAFOD site have been
presented earlier in the literature. Here we compare our findings to the following studies. We use result
of Chavarria et al. (2003) who applied a similar microseismic imaging method to VSP seismograms
recorded at SAFOD pilot borehole. As examples of active seismic studies we take imaging results of
surface reflection seismic obtained by Buske et al. (2007) and Bleibinhaus et al. (2007).
In Figure 8 one can see the four panels containing the vertical component of the stacked image and
results obtained by Chavarria et al. (2003), Buske et al. (2007) and Bleibinhaus et al. (2007) presented in
the same scale in the direct vicinity of the SAFOD main borehole. Obtained interface map (see Figure 7)
is presented over each of these panels.
Chavarria et al. (2003) apply the Kirchhoff migration to the secondary signals recorded from numerous
microearthquakes at the SAFOD pilot borehole (Figure 8b). Microearthquakes are located within the San
Andreas Fault zone with the depths varying roughly between 1 to 7 km below sea level. For comparison
we use their PP reflections migration result for the vertical component seismograms from hypocenters
located ≤ 2 km from the pilot hole. This was motivated by the fact of geometric similarity to our study.
Although the dip and shape of some sub-horizontal reflectors demonstrates a similar trend, the matching
reveals significant differences. One possible reason for this is inaccuracy in the events location and limited
aperture because of use of the vertical pilot hole data in Chavarria et al. (2003). Despite the fact that the
same velocity model (Thurber et al. (2004)) was used in both studies, the divergence can also take place
due to some inconsistencies in different parts of this model because these two imaging methods used
different acquisition systems.
Buske et al. (2007) apply the same Fresnel-Volume-Migration technique to the SAFOD2003 surface re-
flection seismic data set. In their image red- and yellow- colored areas correspond to increased reflectivity
(Figure 8c). In general, the image represents a low resolution (large-scale) structure of the fault system
consisting of several fault branches with different dips and reflection strengths. Comparison of the images
presented in Figures 8a and 8c show some differences. However, both images demonstrate a similar trend

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 61

0 500 1000 0 500 1000

1000 1000
(a) (b)
1500 1500

2000 2000

2500 2500

3000 3000

1000 1000
(c) (d)
1500 1500

2000 2000

2500 2500

3000 3000

0 500 1000 0 500 1000

Figure 8: (a) Vertical component of stacked image. (b) Image from Chavarria et al. [2003]. Reprinted with permission
from AAAS. (c) Image by Buske et al. [2007]. (d) Image by Bleibinhaus et al. [2007]. Thin solid black lines -
obtained fault map (see Figure 7).

Microseismic Reflection Imaging: Methodology and Applications

62

1000 1400 Geology

Granite 1200
1600

Measured depth along borehole (m)

Fault 1400
1800
2000 1600
1500
2200 Granodiorite
Depth (m)
2400 Fault 1800
Bredicated granite
2600 Sandstone with shale
Sandstone 2000
2800
Shale
3000 Unbedded siltstone
2000 Sandstone 2200

Faulted sandstone
2400
Sandstone with shale
Fault
Sandstone with shale 2600
Shale with sanstone
Sandstone with shale
2500 Shale 2800
Sandstone

3000

0 500 1000

Figure 9: Vertical component of stacked image and lithology correlation. Solid lines represent supposed reflectors
map (see Figure 7).

in the transition from a sub-vertical to a sub-horizontal direction, inclined at a similar angle. Moreover,
our results may provide much higher image resolution, and instead of a smooth zone of high reflectivity
shows the fine-scaled structure inside of the fault zone, probably related to second-order faults encoun-
tered along the borehole.
Bleibinhaus et al. (2007) perform the steep-dip pre-stack migration with the same SAFOD2003 data set.
The major difference between the images is the absence of the sub-vertical reflectors in Figure 8d. The
other inclined parts of the reflectors are in a good agreement.
Passive and active imaging results may supplement each other at different scales: the surface seismic
images obtained from lower frequency signals provide the information about the large scale structure of
the fault system, whereas the images obtained by using the high frequency microseismic events support
these findings and help to understand the complex fine-scale structure of these reflectors. In particular,
this holds for highly reflective parts where separate reflectors cannot be easily resolved from the surface
seismic image. Furthermore, the consistency coincidence of these images proves the accuracy of the
microseismic event location procedure and the corresponding polarization estimation.

Correlation with geology

We also compare the position of obtained reflectors with lithology along the SAFOD main borehole
(Boness and Zoback (2006)). Figure 9 shows the image obtained for the vertical component of the stacked
image and supposed interface map (see Figure 7) together with the SAFOD main borehole trajectory
and corresponding lithology. There are several evident reflectors in the vicinity of the borehole. The
first reflector (from the left to the right) crosses the borehole around the 1900 m measured depth (MD)
along the borehole trajectory, and likely corresponds to the interface between granodiorite and bredicated
granite which is represented by a thin fault layer. The second reflector intersects the borehole at level 2280
m MD. This level does not correspond to any of the interfaces, but it is located inside of the fractured
sandstone layer. Probably this reflector can be interpreted as a second-order fault inside the fractured
sandstone bedding. The last high reflective zone in the image, in the vicinity of the borehole at 2620 m
measured depth, may be related to a thin fault layer at this level of the lithology log.

Reflection imaging using fluid injection induced seismicity

In this part we present new results for the injection-induced microsesmic data from the EGS project in
Basel, Switzerland.

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 63

Microseismic monitoring system

The microseismic waveforms produced during and after reservoir stimulation were recorded by several
networks operated by independent institutions. Here we consider network operated by Geothermal Ex-
plorers Ltd.. This network includes six three-component downhole sensors in long-term operation shown
in Figure 10.

13 HAL Injection
well
0
12 MAT STJ
RI2 OT1 HAL
1
11 Injection RI2
well
Northing [km]

OT2 2 sediments

Depth [km]
10
granitic basement
STJ OT1 OT2
9 3

8 4

7 5
MAT 7 8 9 10 11 12 13
Northing [km]
6
10 11 12 13 14 15 16 17
Easting [km]

Figure 10: Distribution of downhole sensors, left: map view; right: side view. Black dots – borehole sensor positions,
black solid lines – boreholes, red solid line – injection well open hole interval.

Located microseismic cloud

In total, 2138 microseismic events were located (see Figure 11) including 1231 clustered events which
form 260 multiplets. The maximal cluster includes 109 events, the minimal – 3 events. The microseismic
cloud has narrow and nearly vertical shape with a main trend oriented in the SSE direction for about 1100
and 1000 m in a vertical plane. In the map view, the cloud splits up into two well pronounced planar
nearly vertical structures. The major one has an azimuth of approximately N 156E which is close to
the orientation of the maximal horizontal stress of 144 ± 14◦ (Valley and Evans (2006)). The second
structure, branches from the main one in the vicinity of the injection interval and propagates in the SE
direction at an azimuth of approximately N 105E.
There are also less pronounced but noticeable features in the vertical section along the main cloud trend.
Multiplets assemble into a system of quasi linear structures which together form a “skeleton” of the cloud,
while non-clustered events are sparsely distributed around it.
There is also a number of steeply dipping structures in the both sides from the treatment interval with
dips of about 60◦ . This is consistent with the dominant dips of natural fractures detected at this interval
of the treatment borehole (Häring et al. (2008)).
Additionally to the event locations, the location procedure also provides statistical characteristics of the
obtained locations, such as root-mean-square of residuals at maximum likelihood hypocenter and the con-
fidence ellipsoids (the confidence ellipsoid represents the 68% confidence region of the located hypocen-
ter for 3 degrees of freedom). In Figure 12 one can see error ellipsoids for the first 200 located events
after the start of the injection. The average error ellipsoid semi-major axis length for clustered events is
approximately 25 m.
In Figure 13 one can see histograms of moment magnitudes and RMS residuals obtained for clustered and
non-clustered events. The figure shows that multiplets significantly outperform the other events in terms
of location residuals. The average misfit for multiplets is about 0.5 ms and approximately 1.5 ms for other
events. One might hypothesise that this can be explained by smaller magnitudes of non-clustered events:

Microseismic Reflection Imaging: Methodology and Applications

64

3700

3800

3900

4000

4100

4200

4300
Depth [m]

4400

4500

4600

4700

4800

4900

5000
9900 10000 10100 10200 10300 10400 10500 10600 10700 10800 10900 11000 11100 11200
Northing [m]

Figure 11: Location of 2138 microseismic events. Black line – borehole, red line – open hole section, red cross –
borehole position, black dots – clustered events, gray dots – non-clustered events, green dots – events with magnitudes
ML > 1.7.

4

4

4


 4



 44


D

44

4

4
11


11

11


1


1

11 

1


1

11


1


1


E
 
N 
 

Figure 12: 68% confidence error ellipsoids corresponding to first 200 located events.

some events would not be assigned to any cluster because of lower signal to noise ratio and not because
of their specific waveforms. The magnitude distribution of clustered and non-clustered events, however,
shows that it is not the case. Histograms of the magnitudes for clustered and non-clustered events almost

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 65

400 400

300 300
Frequency

Frequency
Events
200 200
clustered
non-clustered

100 100

0 0

0.0 0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.2 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Magnitude RMS [ms]

Figure 13: Histograms of moment magnitudes (left) and root-mean-square of residuals at maximum likelihood
hypocenter (right) for the located clustered and non-clustered events

coincide. This indicates that there is a factor other than the size of an event which controls whether this
event is included or not included in a cluster.

Imaging results
We apply the imaging procedure to the data recorded at six receivers (see seismic data example in Figure
1). P-waves were migrated assuming that they are PP reflections, S-wave were imaged as SS reflections.
As a result, We’ve obtained PP and SS reflection images from the data recorded at OT2, OT1, STJ
stations and the image of SS reflections for HAL station. Migration of HAL PP reflection data and
the data recorded at MAT sensors do not provide any focused images in the vicinity of the stimulated
area. Seismograms recorded at RI2 station are not migrated because it is not possible to estimate the
polarisation with reasonable accuracy.
For the given geometry of acquisition system and microseismic sources location one can expect that
reflected waves recorded at different sensors may illuminate different parts of the reservoir. Therefore
location of most reflective areas in images obtained for various receivers may differ significantly. Assum-
ing that the velocity model and estimated static corrections are precise, images of PP and SS reflections
recorded at the same receiver are expected to reveal similar structures. Though the wave path of P and S
waves can be different and corresponding illuminated areas do not have to necessarily coincide. Polarity
estimates for P waves are generally significantly more stable than the estimates for shear waves, so in
most cases PP reflection images are more reliable than SS reflections images for the same receiver.
Obtained images are 3D grids containing stacked reflectivity values at discrete points of the reservoir.
Migration results are obtained by stacking images from single event-receiver pairs, so reflectivity is higher
in the areas which are illuminated by many sources. For this reason, obtained high reflective areas are
located mostly within the microseismic cloud. There are several distinct reflectors in the images. These
reflectors have a planar shape mostly horizontally oriented in the West-East plane with diverse inclinations
in the South-North direction.
Since obtained reflective spots are located mostly inside of the cloud, the vertical sections to the North
direction which is close to the main horizontal trend of the cloud are of the greatest interest. In the
following we present vertical sections along the main trend of the cloud for obtained images in the vicinity
of the open hole section.
In total, 7 images were obtained using PP and SS reflection data from 4 borehole sensors. The PP reflec-
tions image for HAL sensor reveals least distinct picture compared with the images from other stations
and does not considered further. For the data recorded at OT2, OT1 and STJ sensors we have produced
focused images for both wave types. These 6 images can be treated as independent results. All micro-
seismic waveforms recorded at different stations were processed individually: for each sensor we have
estimated azimuthal orientation and selected its own subset of suitable events. Since the directional mi-

Microseismic Reflection Imaging: Methodology and Applications

66

OT2 pp STJ pp

OT1 ss STJ ss

Figure 14: Vertical sections of the obtained images for OT2, OT1, STJ sensors (from left to right), for PP and SS
reflected waves (top, bottom).

gration approach is applied, PP and SS reflections images obtained for the same receiver do not replicate
each other. There are different polarity estimation procedures applied for pressure and shear waves, ad-
ditionally thresholds used to select the most reliable parts of seismograms suitable for imaging defined
differently. Vertical SN sections of the 6 images are shown in Figure 14. One can see that, as it was
expected, PP images are more reliable and show more details compared with SS images. One can also
observe that quality of images degrades with increasing distance from the injection point to a receiver (in
Figure 14 images are sorted by the distance from receiver to injection point). Nevertheless, generally all
images reveal similar structure of reflectors within and outside of the cloud. The most of information is
contained in OT2 PP and OT1 PP images. Other images can serve as an independent confirmation of the
main results.

3D representation of the reflectors

An alternative way to present results of imaging is to show reflectors in a form of 3D objects. Since
migrated images are 3D grids with reflectivity values in each cell, it is possible to plot the cells with have
reflectivity higher than some predefined threshold to show revealed 3D structure of reflectors.

Figure 15 shows 3D iso-surfaces corresponding to the most reflective parts of the OT2 PP and OT1 PP
images. To make it more prominent, we plot here only part of 3D reflectors structure belonging to the
volume in direct vicinity of the microseismic cloud. The most reflective part of the OT2 PP image, as
seen in Figure 14 is located in the southern part, while the OT1 PP image has the strongest reflectors
in the northern part of the volume. Revealed reflectors have a planar shape. In some cases it is slightly
curved due to limited aperture. Vertical sections of the images show more details and demonstrate more
consistency compared to 3D plot. Nevertheless, in the area where most reflective parts are overlapping
the reflectors almost coincide by depth and dip.
Altogether the 3D plots of reflectors repeat shape of the seismically active area. There are several sub-
parallel structures in the left and right hand side of the volume which coincide with respective branches
of the cloud. There is also a strong horizontal reflector in the bottom part of the reservoir. In the upper
part of the cloud any prominent reflectors are not presented. This can be explained by disposition of the

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 67

OT2 PP OT1 PP

E N
N
E 500 m

Figure 15: 3D representation of the most reflective areas in OT2 PP (blue) and OT1 PP image (red). Black solid line
– borehole, black dots – multiple events, grey dots – non clustered events, green dots – big events, red solid line –
borehole open hole section.

sources and receivers. There are just not enough microseismic events to illuminate the upper part of the
reservoir for the given configuration of seismic sensors.

Correlation with borehole data

Highly reflective parts on the obtained images may indicate some local heterogeneities such as inclusions
or fault zones. Geological data from the BS1 stimulation borehole logs is an independent source of
information about such heterogeneities. In Häring et al. (2008) it is noted that the bottom part of the well
including open hole section is characterised by a natural set of fractures trending NW-SE to NNW-SSE,
with steep dips exceeding 60◦ . Additionally, they identify two fracture zones at 4700 m and 4835 m depth
measured along the borehole.
Valley and Evans (2009) investigate the SHmax direction beneath city of Basel from observations of
wellbore failure derived from ultrasonic televiewer images obtained in two wells: the exploration well
where OT2 sensor was installed and the deep borehole BS1 where the injection took place. BS1 borehole
was imaged entirely within the granite basement from 2569 m to 4992 m measured depth. The acoustic
reflectivity image provides information about the presence of fractures and changes in rock properties.
Figure 16 shows the fragments of two most reliable microseimic reflection images (OT2 PP and OT1 PP)
in the vicinity of BS1 well and the bottom part of the ultrasonic reflectivity image for the same well by
Valley and Evans (2009). The upper reflector which intersects the borehole is the reflector from OT2 PP
image, which crosses the borehole at 4275 m true vertical depth (TVD) or 4540 m depth measured along
borehole (MD). This depth corresponds to local reflectivity maximum at the televiewer image. The next
relatively small reflector at OT2 PP image crosses the borehole at approximately 4420 m TVD which is
slightly higher than 4430 m TVD (4700 m MD) where the local maximum of ultrasonic image is located.
This is also consistent with the fault zone at 4700 m MD identified by Häring et al. (2008). The first
distinct reflector in OT1 PP image crosses the borehole at approximately 4470 m TVD (4740 m MD)
which corresponds to the area of elevated reflectivity surrounded by a non reflective area on the top and
bottom in the televiewer image. The strongest reflector on both OT2 and OT1 images intersects the well
at the interval 4530 – 4570 m TVD (4800 – 4840 m MD) which correlates with the highly reflective

Microseismic Reflection Imaging: Methodology and Applications

68

SSE NNW SSE NNW

3800
OT2 pp OT1 pp

4200

True vertical depth [m] 4000

Measured depth [m]

4400
4200

4600

4400

4800
4600

5000
N E S W N
Figure 16: Correlation of obtained images and ultrasonic reflectivity acquired in the injection borehole. Left: OT2 PP
image vertical section, center: OT1 PP image vertical section, right: acoustic televiewer image by Valley and Evans
(2009). White solid line indicates the geometry of BS1 borehole, red solid line shows the well open hole section.
Distance between horizontal and vertical guidelines is 200 m.

areas in ultrasonic images as well as the fault zone at 4835 m MD mentioned in Häring et al. (2008).
Sub-horizontal reflector at 4640 m TVD (4910 m MD) in OT2 PP image does not coincide with any
distinct reflectivity zone, but lays within the highly reflective interval in the bottom part of the borehole in
the televiewer image. The ultrasonic reflectivity image provides also information about the direction of
heterogeneities. According to Valley and Evans (2009) most of reflectivity is oriented in NE-SW direction
(see Figure 16). Unfortunately, it is quite complicated to determine the azimuthal direction of reflectors
from microseismic images because it is controlled by the geometry of the cloud and the area which can
be illuminated.
Generally, OT2 PP and OT1 PP are in a good agreement with the televiewer image. Its most reflective
parts in the bottom part of the BS1 borehole coincide quite well with the reflectors revealed in micro-
seismic reflectivity images. Though there are some high reflectivity areas in ultrasonic images which are
not presented in the microseismic images. First, this can be explained by limited area of illumination by
reflected waves in the microseismic data. Second, not all the reflectivity in televiewer image necessarily
has to be heterogeneities penetrating far away from the borehole and therefore could be non imaginable.
Obtained images are also confirmed by two major fault zones in the open hole section identified by Häring
et al. (2008).

Correlation with Vp/Vs Ratio Distribution

Kummerow et al. (2012) estimate the distribution of Vp/Vs ratio within the Basel 1 geothermal reser-
voir from differential arrival time measurements of microseismic data. The study is based on the same
microseismic event location result as we use in this paper.
In Figure 17 one can see the 3D representation of Vp/Vs ratio distribution and the strongest reflectors
from the two most reliable microseismic images (the same as in Figure 15). Estimated Vp/Vs ratio is
defined on 3D grid with cell size of 50 m. In the figure one can see only the cells for which the local
Vp/Vs value was determined. The values vary from 1.42 to 2.09 with the average about 1.70 which is
close to the Vp/Vs value computed from average reservoir velocities (1.72). Additionally, authors have
verified that there is no significant bias of VP/Vs values related to the density of events.
Though the resolution of the Vp/Vs ratio distribution is relatively low, one can identify several distinct

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 69

N
Vp/ Vs
1.9
Z

1.8

1.7

1.6

1.5

Figure 17: Vp/Vs ratio distribution and the strongest reflectors from OT2 PP and OT1 PP images. Black solid line
– borehole, black dots – multiple events, grey dots – non clustered events, green dots – big events, red solid line –
borehole open hole section, grey box – volume in which Vp/Vs ratio distribution is defined and the reflectors are
plotted.

anomaly zones where the ratios are much bigger or much smaller than the average. These anomaly zones
spatially correlate with the strongest reflectors in two most reliable microseismic reflection images (OT2
PP and OT1 PP) rather well. Particularly, a group of sub-parallel reflectors in the southern part of the
image fits to the area where Vp/Vs close to 1.5. In turn, the reflector in the northern part of the region
coincides with the area of high Vp/Vs values in the deeper part of the reservoir.

Discussion

For the interpretation of conventional 3D seismic images, reflectors are used to map large scale horizons
and faults. In this study the illuminated area is substantially restricted and the obtained reflectivity shows
local heterogeneities within the stimulated part of granitic basement. The key question here is what kind
of geological objects can be imaged using injection induced seismicity and therefore how can it contribute
to the reservoir characterization.
The first possible interpretation is that the reflectors represent fault planes where microseismicity nucle-
ates, but this is in a contradiction with other observations. Particularly, reflectors in the images are flat and
have dips less than 60◦ , while the focal mechanisms of big events which were estimated by Deichmann
and Ernst (2009) indicate that the corresponding fault planes oriented sub-vertically with azimuths close
to the principal horizontal stress. It is also confirmed by the overall shape of the cloud which has a planar
vertical shape with azimuth about N150E. There is also one more reason to decline this interpretation. In
order to produce visible reflections, which can be migrated by the imaging procedure, a reflector must
have a significant contrast of physical properties in comparison with the background medium. Therefore,
the second order cracks produced by simple slip, where most of multiplets are expected to occur due to
their waveform similarity, can unlikely be seen in the image.
At the other hand, the number of independent observations confirms the presence of horizontally oriented
heterogeneities in the reservoir. Particularly, the spatial distribution of microseismic events shows that in
the vertical projection, there are some areas in which most of the microseismicity occurs and there are
some aseismic zones, even in the direct vicinity of the injection interval. Borehole logging data analysis
also shows some relatively thin horizontally oriented fractures at several depth intervals (see Figure 16).
Apart from it, in the Vp/Vs ratio distribution one can see significant variations of the values in a vertical
plane (Figure 17).

Microseismic Reflection Imaging: Methodology and Applications

70

E
E
N N

Figure 18: Focal mechanisms of 27 biggest clusters and corresponding fault planes (cyan disks); combined 3D
representation of the most reflective areas in OT2 PP and OT1 PP image (grey planar structures). Black solid line –
borehole, black dots – multiple events, green dots – big events, red solid line – borehole open hole section, grey box
– volume in which reflectors are plotted.

Additionally, we analysed the fault plane solutions of multiple events obtained by Rusu (2012). The
author used the same microseismic data as used in this study to compute earthquake focal mechanisms
for several events in each cluster. He revealed that events, as expected, have quite similar fault plane
solutions, and produced cumulative solutions for the 27 biggest clusters. Each solution specifies two
possible fault planes which satisfy the data. In order to get a single solution for each cluster we computed
the average distance between both planes passing through the centre of the cluster and events form this
cluster. The plane which provides the smallest misfit we consider as a fault plane of the cluster. Obtained
fault planes as well as focal mechanisms and the reflectors are shown in Figure 18. Each plane is presented
in form of a disk. The radius of a disk is the maximal distance between an event and the centre of the
cluster. One can observe that the fault plane solutions for the clusters are generally consistent with the
results of Deichmann and Ernst (2009) for the big events. Estimated fault planes are mostly sub-vertical
and oriented along the main cloud trend direction. Estimated planes clearly do not coincide with the
reflectors, moreover in most cases they intersect or touch the reflectors with an angle close to 90◦ . A
probable interpretation which explains such behaviour is that the imaged reflectors are cracks or weak
rock zones with a higher permeability compared to the background rock. These highly permeable zones
serve as pathways for the fluid and the elevated pressure to the medium and therefore control the shape
of the cloud which is confirmed by the observation. The pathways are aseismic, while the majority of
microseismicity occurs in a surrounding rock. Clustered events could occur in the second order cracks
branching from the main pathways. Waveform similarity of events in the clusters can be explained by the
fact that they occur at the same or at parallel systems of the second-order fractures. Non-clustered events
are have a specific source mechanism and possible can occur as a result of pore pressure perturbation in
the background rock.

Outline: application to hydraulic fracturing data

Recently, we started with processing of hydraulic fracturing dataset which contains a number of stimula-
tion stages recorded at two borehole arrays at the distance of few hundreds of meters. In Figure 19 one can
see that despite of much smaller magnitudes of events and closer distances, the common receiver gather
produced from hydraulic fracturing induced events demonstrates similar features as the seismicity in-

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 71

duced by stimulation of the geothermal system. Moreover, in some neighboring traces between P- and S-
direct waves one can see some correlated wavelets which may indicate the presence of reflected waves.
Considering the travel times, this reflection could come from the fluid filled fractures which remained
open after stop of the stimulation during the previous injection stage.

Event number
0 5 10 15 20 25 30 35 40 45 50 55
0

0.02

0.04

0.06

0.08

0.10
Time (sec)

0.12

0.14

0.16

0.18

0.20

0.22

0.24

Figure 19: Common receiver gather of a hydraulic fracturing induced seismicity recorded at a borehole instrument
(vertical component).

Acknowledgments
We thank the sponsors of the PHASE consortium for supporting the research presented in this paper.

References
Asanuma, H., Tamakwa, K., Soma, N., Niitsuma, H., Baria, R., and Häring, M. (2011). Reflection
imaging of EGS reservoirs using microseismicity as a source. In Thirty-Sixth Workshop on Geothermal
Reservoir Engineering, Stanford, California. Stanford University.

Bednar, J. B. (2005). A brief history of seismic migration. Geophysics, 70(3):3MJ–20MJ.

Bleibinhaus, F., Hole, J. A., Ryberg, T., and Fuis, G. S. (2007). Structure of the california coast ranges
and san andreas fault at safod from seismic waveform inversion and reflection imaging. J. Geophys.
Res., 112(B6). doi:10.1029/2006JB004611.

Boness, N. and Zoback, M. (2006). A multi-scale study of the mechanisms controlling shear ve-
locity anisotropy in the san andreas fault observatory at depth. Geophysics, 71(5):F131–F136.
doi:10.1190/1.2231107.

Microseismic Reflection Imaging: Methodology and Applications

72

Bradbury, K., Barton, D., Solum, J., Draper, S., and Evans, J. (2007). Mineralogic and textural analyses
of drill cuttings from the san andreas fault observatory at depth (safod) boreholes: Initial interpretations
of fault zone composition and constraints on geologic models. Geosphere, 3(5):299–318.

Buske, S., Gutjahr, S., Rentsch, S., and Shapiro, S. (2007). Application of fresnel-volume-migration
to the safod2003 data set. EAGE 69th annual meeting and technical exhibition, London, Expanded
Abstracts, P335.

Buske, S., Heigel, M., and Lüth, S. (2006). Fresnel-volume-migration of single-component seismic data.
EAGE 68th annual meeting and technical exhibition, Vienna, Expanded Abstracts, G044.

Chavarria, J. A., Malin, P., Catchings, R. D., and Shalev, E. (2003). A look inside the san andreas fault at
parkfield through vertical seismic profiling. Science, 302(5651):1746–1748.

Deichmann, N. and Ernst, J. (2009). Earthquake focal mechanisms of the induced seismicity in 2006 and
2007 below Basel (Switzerland). Swiss Journal of Geosciences, 102:457–466. 10.1007/s00015-009-
1336-y.

Dyer, B. C., Schanz, U., Ladner, F., Häring, M. O., and Spillman, T. (2008). Microseismic imaging of a
geothermal reservoir stimulation. The Leading Edge, 27(7):856–869.

Häring, M., Schanz, U., Ladner, F., and Dyer, B. (2008). Characterisation of the Basel 1 enhanced
geothermal system. Geothermics, 37(5):469–495.

Jurkevics, A. (1988). Polarization analysis of three-component array data. Bulletin of the Seismological
Society of America, 78(5):1725–1743.

Kummerow, J., A. Reshetnikov, M. H., and Asanuma, H. (2012). Distribution of the Vp/Vs Ratio within
the Basel 1 Geothermal Reservoir from Microseismic Data. Expanded abstracts, EAGE 74th annual
meeting and technical exhibition, Copenhagen.

Lüth, S., Buske, S., Görtz, A., and Giese, R. (2005). Fresnel-volume-migration of multicomponent data.
Geophysics, 70(6):S121–S129.

Rusu, A. (2012). Untitled - Bachelor’s thesis, FR Geophysik, Freie Universität Berlin, Germany.

Samson, J. C. (1977). Matrix and stokes vector representations of detectors for polarized waveforms:
theory, with some applications to teleseismic waves. Geophysical Journal of the Royal Astronomical
Society, 51(3):583–603.

Sanford, A., Alptekin, Ö., and Toppozada, T. (1973). Use of reflection phases on microearthquake seis-
mograms to map an unusual discontinuity beneath the Rio Grande rift. Bulletin of the Seismological
Society of America, 63(6-1):2021–2034.

Soma, H., Niitsuma, H., and Baria, R. (1997). Estimation of deeper structure at the Soultz Hot Dry Rock
field by means of reflection method using 3c ae as wave source. Pure Appl. Geophys., 150:661–676.

Soma, H., Niitsuma, H., and Baria, R. (2002a). Reflection technique in time-frequency domain us-
ing multicomponent acoustic emission signals and application to geothermal reservoirs. Geophysics,
67(3):928–938. doi:10.1190/1.1484535.

Soma, N. and Niitsuma, H. (1997). Identification of structures within the deep geothermal reservoir of the
kakkonda field, japan, by a reflection method using acoustic emission as a wave source. Geothermics,
26:43–64.

Soma, N., Niitsuma, H., and Baria, R. (2002b). Deep sub-vertical structure at Soultz Hot Dry Rock site
estimated by reflection technique using multicomponent acoustic emission events. Trans. Geotherm.
Resour. Counc., 26:255–266.

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 73

Soma, N., Niitsuma, H., and Baria, R. (2007). Reflection imaging of deep reservoir structure based on
three-dimensional hodogram analysis of mumicroseismic waveforms. J. Geophys. Res., 112:B1103.
doi:10.1029/2005JB004216.
Tamakawa, K., Asanuma, H., Niitsuma, H., and Soma, N. (2010). Principles of a seismic reflection
method using AE multiplets as a source. In Renewable Eenergy 2010 Proceedings.
Thurber, C., Roecker, S., Zhang, H., Baher, S., and Ellsworth, W. (2004). Fine-scale structure of the
san andreas fault zone and location of the safod target earthquakes. Geophysical Research letters,
31(L12S02). doi:10.1029/2003GL019398.
Valley, B. and Evans, K. F. (2006). Stress orientation at the Basel geothermal site from wellbore failure
analysis in BS1. Technical Report ETH 3465/56, ETH Zürich.
Valley, B. and Evans, K. F. (2009). Stress orientation to 5Âăkm depth in the basement below Basel
(Switzerland) from borehole failure analysis. Swiss Journal of Geosciences, 102:467–480.
Zoback, M., Hickman, S., and Ellsworth, W. (2010). Scientific drilling into the san andreas fault zone.
Eos Trans., 91(22):197–199.
Zoback, M., Hickman, S. H., and Ellsworth, W. L. (1998). Injection induced earthquakes and the crustal
stress at 9 km depth at the ktb deep drilling site. San Andreas Fault Observatory at Depth, Proposal,
Scientific Drilling into the San Andreas Fault at Parkfield, CA, Project Overview and Operational Plan.

Microseismic Reflection Imaging: Methodology and Applications

74

Microseismic Reflection Imaging: Methodology and Applications

Annual Report 2012 75

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a

Hydraulic Fracture

A. Oelke1 , O.S. Krüger1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
Analytical solutions for reflection coefficients at thin fluid layers are compared to reflection coefficients
derived from synthetic data in 3-D. Thin fluid layers represent hydraulic fractures and explosive or double
couple sources represent microseismic events. We show, that in our case viscosity can be neglected and
water is adequate as a fluid filler. High reflection coefficients (0.2-0.7) may occur at layers of thicknesses
in mm-range and at frequencies of 50-400Hz, like one comes across in microseismicity. Thus, data
recording microseismic events can, on the one hand, be treated as reflection seismic data, and small-scale
subsurface images can be derived, and, on the other hand, reflection coefficients can be interpreted in
terms of reservoir properties.

Introduction
In the frame of wave propagation at a thin layer numerous publications already exist. Krauklis (1962)
for example introduced the slow fluid wave, recently referred to as Krauklis wave (Korneev et al., 2012),
which has a large amplitude, high dispersion and propagates with low velocity at low frequencies. Ko-
rneev (2010) explore the Krauklis wave within a viscous fluid crack in an elastic medium. Quintal et al.
(2009) investigate the reflectivity of a viscoelastic layer for the 1-D case and confirm theoretical predic-
tion with numerical results. For this case, they show, that the contrast in attenuation yields significant
reflections. In their paper, Fehler (1982) assume a thin fluid layer to be filled with a viscous fluid and
analyze the interaction of seismic waves with this layer.
In our previous work we have introduced analytical solutions for reflection coefficients at thin layers
filled with water (Oelke et al., 2011). Reflection coefficients derived from synthetic data in 1-D and 2-D
confirm our analytical solution (Oelke et al., 2012).
In this paper we investigate a thin fluid layer that is embedded in an elastic isotropic rock. The fluid filler
is water and represents the simplest model of a hydraulic fracture. We find, that microseismic events can
be reflected at hydraulic fractures and reflection coefficients are remarkably high (e.g. 0.4) so that these
reflections can be used to obtain small-scale subsurface images or can be interpreted in terms of fracture
properties, such as width.
We will first recall theoretical background and comment on why we can neglect viscosity in our case and
restrict the domain in which our analytical solution is valid. In the section ’Modeling’We will sketch
the 2-D numerical verification. Then we will introduce the 3-D modelling and its resulting comparison
of reflection coefficients to the analytical solution. In a further step, the explosive source is replaced by
a double couple source. Finally we will show examples of microseismic events from which reflection
coefficients will be extracted and interpreted in future work.

Theoretical Background
The most simple model of a hydraulic is a thin fluid layer embedded in an isotropic, elastic solid. The
fluid filler is water and viscosity of the fluid is neglected. In order to obtain an analytical solution for

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

76

the reflection coefficient at the fluid layer we rewrite the reflected wave as a superposition of reflected
and multiply reflected waves. The full reflected P-wavefield of an incident P-wave, thus containing the
reflection coefficient Rpp , can be written as a sum of P-waves which are multiply reflected within the
fluid layer. The resulting expression depends on the solid P-wave velocity vp , the solid S-wave velocity
vs , and the solid density ρs , the velocity within the fluid vf , and the density ρf :

(βs2 − ζ 2 )2 r(1 − e2iωhβf )

 
Rpp = 1 − 1+ , (1)
R 1 − r2 e2iωhβf

ζ denotes the horizontal slowness which is constant on the contrary to the vertical slowness
where q
1
βi = v2
− ζ 2 . R is the Rayleigh function:
i

r = (a − b)(a + b)−1 , a = ρs vs4 βf R, b = ρ f βp , (2)

 2
1
R= − 2ζ 2 + 4βp βs ζ 2 . (3)
vs2

Validity
In principle the above derived formula 1 is valid for all frequencies. However, since the main goal is
to investigate reflections from microseismic events, we restrict ourselves to frequencies between f =
50 − 400Hz.
This frequency restriction also becomes crucial when justifying the assumption of a non-viscous fluid
filler. In our model, we assume an ideal water fluid layer to describe a hydraulic fracture. Of course,
in nature fillers are viscous. However, we then would have to consider different wave types within the
fluid. Following Biot (1956), we can back up, why the ideal fluid is justified though. Considering our
restriction in frequency range, the fracture width h is still big enough so that is in the order of the quarter
wavelength of the boundary layer and Poiseuille flow breaks down.

Modeling
The accuracy of the Rotated Staggered Grid (RSG) Finite Difference (FD) algorithm, introduced by
Saenger et al. (2000), for thin fractures has been shown before (Krüger et al., 2005). We make use of this
algorithm and calculate synthetic wavefields in order to extract reflection coefficients and compare them
to the analytical solution.

2-D Modeling
The experimental setup for the 2-D modeling is sketched in figure 1(a), the modeling conditions are listed
in table 1. Figure 1(b) shows the recorded wavefield. Direct, reflected P-wave and reflected S-wave can

Source type Point source Sourcetype Double Couple

Rock P-wave velocity 4500m/s Rock P-wave velocity 5940m/s
Rock S-wave velocity 2000m/s Rock S-wave velocity 3450m/s
Fluid P-wave velocity 1500m/s Fluid P-wave velocity 1500m/s
Dominant frequency 171Hz Dominant Frequency 168Hz
Wavelength in fluid 8.8m Wavelength in fluid 8.9m
Distance source - reflector 65m Distance source - reflector 100m
Fracture width 0.5m Fracture width 2.5m
Grid spacing 0.05m Grid spacing 0.5m

Table 1: Properties of the 2-D numerical experiment. Table 2: Properties of the 3-D numerical experiment.

be clearly distinguished. Then the direct and reflected P-wave are Fourier transformed and the ratio of the
amplitude at a given frequency gives the reflection coefficient. The resulting angle-dependent reflection

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

Annual Report 2012 77

coefficient is plotted in figure 1(c) and compared to the analytical solution. Both results are in a good
agreement. Finally, in figure 1(d), the angle of incidence is kept fixed and the fracture width is varied. The
result is compared again to the analytical solution and as well is in a good agreement. After exceeding the
range of validity, which means the fracture width is big enough so that the directly reflected and within
the layer reflected wave are clearly separated, the measured reflection coefficient equals a reflection at a
single interface. We assume here, that the source-receiver distance is big enough, so that the cylindrical
wave emitted by the explosive source can be treated as a plane wave at the receiver and fluid layer.

0
direct wave
reflected P−wave
0.02 reflected S−wave
Receivers
Source ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ 0.04

time (s)
* 0.06
65m
solid
0.08
fluid h
solid 0 50 100 150 200
offset (m)

(a) 2-D experimental Setup. (b) Recorded wavefield.

1
Analytical Solution
FD Solution
0.8
0.5
Rpp(θ=31.6 )
°

0.6
Rpp

0.4
0.4

0.3
0.2
Analytical Solution
FD Solution
0 0.2
0 10 20 30 40 50 60 70 80 90 0 0.1 0.2 0.3 0.4 0.5
incidence angle (degrees) h / λf

(c) The angle-dependent P-P reflection coefficient. (d) At a fixed angle of incidence (θ = 31.6◦ ), the layer width h
is varied. Reflection coefficients are compared to the analytical
solution.

Figure 1: Processing steps of the comparison between analytical solution and reflection coefficients extracted from
synthetic data.

3-D Modeling
The confirmation of the analytical solution in 3-D is a balance between required resolution and compu-
tational facilities. A first step is the implementation of an explosion source. The experimental setup is
sketched in figure 2. Properties of the experiment are listed in table 2. The recorded wavefield is plotted in

Figure 2: Aquisition Geometry of the 3-D numerical experiment, the fluid layer is at 200m depth.

figure 3(a), geometrical spreading is considered and calculated traveltimes are marked in order to identify

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

78

the signals. Reflection coefficients for different angles of incidence are determined and compared to the
analytical solution in figure 3(b). Both values are in a good agreement for small angles of incidence. For
bigger angles of incidence the direct and reflected P-wave can not be distinguished and thus the resulting
reflection coefficients are distorted. However, the suggested experimental setup gives reasonables results
and in a next step a more realistic source is deployed.

1
Theoretical PP reflection
0.9 FD Model
0.02 0.8

0.7
0.04 0.6

magnitude
←P
t(s)

0.5
0.06 ← P−P
0.4

0.3
← P−S
0.08
0.2

0.1
0.1
0
100 200 300 0 10 20 30 40 50 60 70 80 90
offset(m) angle (deg)

(a) Recorded wavefield, spherical corrected. Calculated travel- (b) Comparison of modelled reflection coefficients to analytical
times are marked. solution.

Figure 3: The 3-D numerical results using an explosive source.

Double Couple
The same experimental setup as sketched in figure 2 and properties as listed in table 2 are deployed to
calculate the synthetic wavefield. The focal mechanism is defined by a strike, dip and rake of 73.5◦, 90◦
and 0◦ , respectively.
A snapshot of the wavefield after 60ms is sketched in figure 4. Because of the radiation pattern, the
dominant visible wavefront is the direct S-wave and the S-S-reflection. The direct P-wave and the P-S
reflection are rather weakly visible. Only suspected, at least from the snapshot, can be the P-P reflection.
The recorded wavefield with indicated expected arrival times is sketched in figure 5.
The signals of direct P-wave, and reflected P- and S-wave tend to overlap. Additionally there is the direct
S-wave and its reflected modes, which overlap with the signals from the incident P-wave. Here a polarity
analysis should be deployed in order to separate the respective signal and calculate reflection coefficients.
In addition, spherical spreading and the source function have to be taken into account.

S
S−S S−P P
P−S

Figure 4: A snapshot of the wavefield after 60ms. The fluid layer at 200m depth is indicated on the z-axis.

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

Annual Report 2012 79

x y z

0.02 0.02 0.02

0.04 0.04 0.04

←P
t(s)

0.06 0.06 0.06 ← P−P

0.08 0.08 0.08 ← P−S

←S
← S−P
0.1 0.1 0.1
← S−S
100 200 300 100 200 300 100 200 300
offset(m) offset(m) offset(m)

Figure 5: Recorded seismogram using a double couple source, showing the x-, y-, and z-component. Marked are
calculated P-, and S-wave arrivals, as well as its reflections. Signals of each wave type overlap and can not be
distinguished clearly.

Real Data
In this section examples of hydraulic fracturing data, in which we see signatures of fractures, are given.
The first example is taken and modified from Rentsch (2007). A microseismic event recorded at 20
receivers (East-, North- and vertical component) at Cotton Valley in 1997 is shown in figure 6. The
dominant P-wave frequency of this event is at approximately 100Hz. A quite strong signal between first
arrival and S-wave can be seen in several events and is marked here. We interpret this signal to occur
from a hydraulic fracture that was opened in a previous fracking stage.

Figure 6: A seismic shot gather from a microseismic event recorded at Cotton Valley. East-, North- and vertical
component are shown.

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

80

The second example we show here is from a shale gas hydraulic fracturing data set, provided by ESG. It
has been recorded at Horn River Basin, Canada. In figure 7 the horizontal component of different events
are shown. The signals were detected at 2 different receiver arrays containing 9 and 21 receivers, and are
P-wave aligned. Again between first arrival and S-wave arrival some signature from a potential fracture
that was opened in a previous stage can be seen.

1580 1590 1600 1580 1590 1600 1610 1620 1630

0.45 0.45 0.45

0.50 0.50 0.50

0.55 0.55 0.55

0.60 0.60 0.60

0.65 0.65 0.65

0.70

(a) Event 1576. (b) Event 1577. (c) Event 1601.

1630 1640 1650 1660 1670 1680 1730 1740 1750

0.45 0.45 0.45

0.50 0.50 0.50

0.55 0.55 0.55

0.60 0.60 0.60

0.65 0.65 0.65

0.70 0.70

(d) Event 1625. (e) Event 1651. (f) Event 1728.

Figure 7: Microseismic events recorded in shale gas reservoir show signatures of reflections at fractures.

Conclusion

In this paper we show that reflections which occur in microseismic data may be caused by reflections at
hydraulic fractures.
We investigate reflection coefficients at an idealized hydraulic fracture. The fracture is modelled by a
thin fluid layer which is embedded between identical solids. We give the exact analytical solution for the
reflection coefficient of an incident plane wave and compare it to numerical modeling results in 2-D and
3-D. In 3-D, a double couple is deployed, thus to obtain reflection coefficients, the spherical spreading
and the source function has to be taken into account. Finally, we have shown some real data examples in
which we see such kind of reflections from hydraulic fractures and we plan to further investigate that data
with respect to reflection coefficients.
However, the main conclusion of our work is, that microseismic data can be used in order to create
small-scale subsurface images and that it can be evaluated for reflection coefficients.

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

Annual Report 2012 81

Acknowledgements
We thank the sponsors of the PHASE consortium and the Helmholtz graduate research school GeoSim
for supporting the research presented in this paper.

References
Biot, M. (1956). Theory of propagation of elastic waves in a fluid-saturated porous solid. i low-frequency
range. THE JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA, 28(2):186–178.
Fehler, M. (1982). Interaction of seismic waves with a viscous liquid layer. Bulletin of the Seismological
Society of America, 72(1):55–72.
Korneev, V. (2010). Low-frequency fluid waves in fractures and pipes. Geophysics, 75(6):N97–N107.
Korneev, V., Goloshubin, G., Bakulin, A., Troyan, V., Maximov, G., Molotkov, L., Frehner, M., Shapiro,
S., and Shigapov, R. (2012). Krauklis wave - half a century after. In 5th Saint-Petersburg International
Conference & Exhibition 2012 Saint-Petersburg, Extended Abstract.
Krauklis, P. (1962). About some low frequency oscillations of a liquid layer in elastic medium. PMM,
26(6):1111–1115. In Russian.
Krüger, O., Saenger, E., and Shapiro, S. (2005). Scattering and diffraction by a single crack: an accuracy
analysis of the rotated staggered grid. Geophysical Journal International, 162:25 – 31.
Oelke, A., Alexandrov, D., Abakumov, I., Troyan, V. N., Kashtan, B. M., and Shapiro, S. A. (2011).
Reflection Coefficients at a Thin Fluid Layer as a Model of a Hydraulic Fracture. In 73rd EAGE
Conference & Exhibition incorporating SPE EUROPEC 2011 Vienna, Extended abstract,P353.
Oelke, A., Krüger, O., Shapiro, S., and Abakumov, I. (2012). Modeling reflection coefficients at a thin
fluid layer representing a hydraulic fracture. In SEG Technical Program Expanded Abstracts 2012: pp.
1-5.
Quintal, B., Schmalholz, S. M., and Podladchikov, Y. Y. (2009). Low-frequency reflections from a thin
layer with high attenuation caused by interlayer flow. Geophysics, 74(1):N15–N23.
Rentsch, S. (2007). A Migration-Type Approach for the Fast Location of Seismicity: Theory and Appli-
cations. PhD thesis, Freie Universität Berlin.
Saenger, E. H., Gold, N., and Shapiro, S. A. (2000). Modeling the propagation of elastic waves using a
modified finite-difference grid. Wave Motion, 31(1):77–92.

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

82

Modeling Reflection Coefficients at a Thin Fluid Layer Representing a Hydraulic Fracture

Annual Report 2012 83

Overview of a Data Set of a three Stage Hydraulic Fracturing

Treatment in a Shale Gas Reservoir

S. Franke1 , N. Hummel1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
We consider microseismic data recorded during a hydraulic fracturing treatment in a shale gas reservoir.
This data was provided by one of the sponsors of the PHASE research project. The purpose of this
analysis is to get an overview of the data and image first results. Therefore, we describe properties of the
microseismic event distributions and analyse their behaviour in correlation with engineering parameters.
Furthermore we take into account magnitude data for a characterisation of the source receiver properties.

Introduction
In the oil and gas industry hydraulic fracturing treatments are used to explore so called unconventional
reservoirs. These unconventional reservoirs are characterised by extremely low formation permeabili-
ties. By fracturing the producing formation the well drainage area and therefore well productivity is
increased (Economides and Nolte, 2000). In such a way the permeability of tight reservoir rocks can
be significantly enhanced. Usually hydraulic fracturing treatments induce an elongated narrow cloud
of microseismic event locations (Albright and F., 1982). A technique that is being used increasingly to
understand complex fracture growth and geometry is microseismic mapping (Rutledge, 1998). In such
a way detailed information in terms of height, length and complexity of induced fractures can be obtained.

Here we consider a microseismic data set which has been provided by one of the sponsors of the PHASE
research project. This data set includes more than 3500 microseismic events which have been monitored
and located during a three stage hydraulic fracturing treatment in a tight shale gas reservoir. We first
present an overview of the operational setting. We then analyse the event locations and compare their
pattern among all these three stages. Additionally we correlate the spatio-temporal behaviour of events
with corresponding engineering data. We also consider magnitudes of induced events which we analyse
in terms of their distribution in the event cloud as well as source-receiver-distance.

Overview of microseismic data

The data set under consideration contains three different stages. Stage A contains 1376 located events
which have been induced during a five hour long hydraulic fracturing treatment. Stage B reveals the
smallest number of located hypocenters with 945 events and an injection period of roughly three and a
half hours. In comparison stage C contains 1325 located events which have been induced during a four
hour treatment. The monitoring system consisted of 30 borehole sensors which have been installed in two
monitoring wells.

More detailed information about the time of treatment and the location of the injection points is listed
in Table 1. In order to analyse event locations including their spatio-temporal characteristics as well as
magnitudes we consider the information given in Table 1 as well as the following data:

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
84

• event coordinates of the hypocenter (Northing, Easting, Depth)

• location of seismic sensors and injection point

• occurrence time t of events

• engineering data (injection rate, injection pressure, proppant flow rate)

• event moment magnitudes

The coordinates of the event locations Northing, Easting, Depth are used to image the spatial orientation
of the event clouds. The coordinates of the injection points are used to compute the radial distance
of events from the well bore. To determine correlation of injection rates and pressures with the spatio-
temporal evolution of the events, we show engineering data together with the so called r − t plot. Here the
radial distance of events r is plotted versus the occurrence time t of events. For all three stages Figure 1
shows the location and orientation of the monitoring system, injection points and the hypocenter of events.

Stage A Stage B Stage C

♯ of events 1376 945 1325

Date 06/08/2010 07/08/2010 09/08/2010

Frac Start 11:13 00:13 07:50
Frac Stop 15:54 03:3 6 11:50
Injection Point

Northing [m] -442.6 / -441.0 -359.9 / -361.6 -280.7 / -279.2

Easting [m] 108.1 / 106.9 47.2 / 48.3 4.5 / 3.8
Depth [m] -1818.7 / -1818.7 -1816.2 / -1816.5 -1779.6 / -1778.5

Table 1: Overview of the operational setup for the three stage hydraulic fracturing treatment.

From the map view projection one can see three nearly elliptically formed narrow event clouds striking
in North-East direction. In the northern part both installed geophone arrays with 9 and 21 sensors are
installed. The dimension of the event clouds in strike direction varies from ∼ 660 m (stage A) to ∼ 550 m
(stage B) and ∼ 575 m (stage C). Considering the depth view projections we see a curved shape of the
event clouds which are quite limited to a distinct depth. Additionally one can see that all three event
clouds are characterised by an asymmetrical distribution of events around the injection point. Facing
the injection point, the number of events is significantly larger compared to the number of events being
induced South-West of the injection point. However this effect is possibly due to the detection footprint.
In terms of striking extend and the injection point position, microseismic events related to the first treat-
ment (stage A) are located furthermost to both geophone arrays. As the distance to the monitoring system
decreases, microseismic events from stage B and C follow the pattern of events from stage A. However,
we note that depending on the geometry particular events from stage A are possibly closer to the moni-
toring system compared to events from stage C.

In the following, we analyse event locations and their spatio-temporal evolution. For this we include
information of injection rates and pressures from the engineering data. We then determine the distribution
of magnitudes of events from corresponding stage A, B and C. Furthermore we evaluate the magnitude
data to get a clue about the detection range of the receiver arrays. First we map the distribution of
magnitudes in the event clouds. We then plot magnitudes of events versus their source-receiver distance.
Therefore it is necessary to determine a reference point of each sensor array and relate it with its distance
to the injection point and the magnitude value. In this case the reference point corresponds to the average
sensor position of either Array-1 or Array-2.

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
Annual Report 2012 85

Map View Depth View I

400 400
Sensors
Injection Points
200 200

0 0

Northing [m]
Northing [m]

−200 −200
C
B
−400 −400
A

Stage A
−600 Stage B −600
Stage C

−800 −800
−200 0 200 400 600 −1400 −1600 −1800 −2000
Easting [m] Depth [m]

3 Dimensional View
−1400
Depth View II
−1400
−1600
Depth [m]

−1600
Depth [m]

−1800
C
C
−1800
−2000
B A
1000 B A
500
Nor

−2000 0
thin

−200 0 200 400 600

−500
g

Easting [m] 400 600

[m]

−1000 0 200
−200
Easting [m]

Figure 1: Overview of microseismic data recorded during a three stage hydraulic fracturing treatment. In addition to
corresponding injection points the monitoring system is also included. It consists of two borehole geophone arrays
with 9 and 21 three component receivers. From the map view projection one can see that the orientation of all three
event clouds strikes North-East. All three stages reveal a similar depth of corresponding microseismic events which
is in a range of -1700 m to -1850 m.

Stage A

During the first stage of the hydraulic fracturing treatment 1376 events have been detected and located
(Figure 2). In comparison to the other stages, stage A is the most Southern one. From the map view
projection it can be seen that the nearly elliptically formed event cloud strikes in North-East direction.
Most of the events are located in the North-Eastern part of the cloud. In comparison only a few events
are distributed South-West of the injection point. Both depth views indicate that the seismicity is fairly
constrained to a depth interval of -1700 m to -1850 m.
The engineering data shows an injection period of almost 5 h (Figure 3). Flow rates and injection pres-
sures increase up to 245 l/s and 52 M P a. A total volume of 3206 m3 of fluid has been injected. In the

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
86

Figure 2: Overview of 1376 hypocenter locations of stage A presented in a map view and two depth view projections.
Furthermore, the injection point and the monitoring system consisting of two sensor arrays are shown. Stage A
demonstrates the largest distribution of events in terms of length and width. Additionally one can see that in the most
North-Eastern part of the cloud there is a separated area of events.

beginning of this stage there is a small pre-injection test. Taking into account the r – t plot of induced
events there is a clear correlation with corresponding flow rates and the onset of seismicity. At the be-
ginning the flow rate of nearly 10 l/s was kept constant for about 15 minutes. During this period the
injection pressure rapidly increased to approximately 40 M P a and starts to decrease thereafter. With the
onset of seismicity this shows that a fracture has been initiated. Following this pre-injection test, the flow
rate was stepwise increased to a maximum of 245 l/s. During this period the seismicity cloud has an
increased rate of growth. In a range of two till four hours of injection a small cluster of events is located
in an isolated area on the base of the r − t plot. However, this cluster of events does not correlate with the
beginning of adding proppant. After termination of fluid injection there is a decrease in the event density.
For radial distances smaller then 300 m there are only very few events.

In what follows we consider magnitudes of induced events. Figure 4 shows a map view projection of the
seismicity cloud where the locations of events are color coded according to their moment magnitude. In
terms of the distribution of magnitudes in the event cloud we detect two patterns. One is located in the
middle of the seismicity cloud. There, many events are characterised with low magnitudes. North-East as
well as South-West of this area events are characterised with higher magnitudes. Additionally Figure 4
shows two plots including the distribution of magnitudes as function of the source-receiver-distance. This

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
Annual Report 2012 87

Figure 3: Engineering data and r – t plot of induced events from stage A. The dashed line represents the termination of
the fluid injection. The onset of seismicity correlates well with the beginning of the treatment. The first microseismic
events are triggered during the pre-injection test. Additionally between two and four hours there is a cluster of events
close to the injection point. However the occurrence of corresponding events does not correlate with the onset of
adding proppant.

400
Sensors 0 Mom. Magnitude
Injection Point

−0.5
Mom. Magnitude

Array-2
200
−1

Array-1
−1.5

−2

300 400 500 600 700 800 900

Northing [m]

Source−to−Array−1 Distance [m]

−200

0
0
−400
−0.5
Moment Magnitudes

Mom. Magnitude

−0.5

−1 −1

−600
−1.5
−1.5

−2
−2
−800
−200 −100 0 100 200 300 400 500 600 500 600 700 800 900 1000 1100 1200
Easting [m] Source−to−Array−2 Distance [m]

Figure 4: Left: Map view of event moment magnitudes, sensor arrays and injection point of stage A. Right: Moment
magnitudes versus source-receiver distance for Array-1 and Array-2. The sloping line represents an approximate
event detection threshold.

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
88

source-receiver-distance corresponds to the distance of each event from the averaged receiver location of
either sensor Array-1 or Array-2. For Array-1 the closest events are detected at distances of about 300 m.
For this particular source-receiver-distance moment magnitudes in a range of -2.5 to 0 are included. A
fitted sloping line in the bottom of the magnitude versus distance plots represents the approximate event
detection threshold. Events falling under this line would not be detected with a sufficient signal strength
to be located (Rutledge and Phillips, 2003). This detection threshold increases with increasing source-
receiver-distance. Furthest located events feature average magnitude values at source-receiver-distances
of approximately 650 m. A similar pattern is shown for the magnitude versus source-receiver-distance
plot of Array-2.

Stage B
During the second stage of hydraulic fracturing treatment 945 events have been recorded and located.
Compared to stage A, the cloud of event locations is located more the North-West. From the map view
projection one can see a similar formed event cloud to stage A with the same striking direction (Figure
5). Additionally the seismicity is constrained in an equal depth interval of -1700 m to 1850 m. Most of
the hypocenters are located in the North-Eastern part of the event cloud. In comparison to stage A, even
less events are distributed South-West of the injection point.

Figure 5: Overview of 945 hypocenter locations of stage B presented in a map view and two depth view projections.
Furthermore the injection point and the monitoring system consisting of two sensor arrays are shown. Stage B
demonstrates the smallest distribution of events in terms of length and width.

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
Annual Report 2012 89

The engineering data shows an injection period of 3.3 h (Figure 6). Injection pressure increases up to
61 M P a and flow rates up to 250 l/s. The total injected fluid volume amounts 2300 m3 . Similar to
Stage A, there is a small pre-injection test at the beginning. Taking into account the r – t plot of induced
events there is a correlation with corresponding flow rates an equivalent onset of seismicity. In this pre-
injection test the flow rate of approximately 10 l/s was kept constant for about 12 minutes. During this
time period the injection pressure increases rapidly to 35 M P a and starts to decrease lightly thereafter.
Corresponding to the similarity with stage A of this first period of injection, seismicity shows that a
fracture has been initiated. After this pre-injection test the flow rate was increased stepwise up to almost
almost 150 l/s. This caused injections pressures of approximately 50 M P a. At 40 minutes of fluid
injection the flow rate and treating pressure suddenly decrease. Thereafter one can see a stepwise increase
of both parameters up to their maximum of 250 l/s and 61 M P a. During this period the seismicity cloud
does not have an increased rate of growth. After 1.2 h proppant has been added with a stepwise increase
up to concentrations of approximately 150 kg/m3 . Simultaneously one can see a slight decrease of
the treating pressure. Taking into account the r – t plot of induced events at this point, there is a clear
correlation of the beginning of proppant addition and a steep increase of seismicity. After termination of
fluid injection there is a decrease in the event density. In a time range of 3.8 h to 4.3 h there appears a
denser distribution of microseismic events in the r – t plot.

Engineering and r−t sata stage B

400 80
Proppant Flow Rate [kg/m3]

Treating Pressure

Treating Pressure [MPa]

Injection Rate
Injection Rate [l/s]

300 60
Proppant Flow Rate

200 40

100 20

0 0
0 1 2 3 4 5 6 7

700

600

500
Distance [m]

400

300

200

100

0
0 1 2 3 4 5 6 7
Time [h]

Figure 6: Engineering data and r – t plot of induced events from stage B. The dashed line represents the termination of
the fluid injection. The onset of seismicity correlates well with the beginning of the treatment. The first microseismic
events are triggered during the pre-injection test. Additionally after one hour the beginning of proppant injection
correlates with a sudden onset of seismicity and a decreasing treating pressure.

In the following we consider magnitudes of induced events (Figure 7). Similar to stage A the map view
projection of the microseismic cloud shows the locations of events color coded according to their moment
magnitude. In terms of the distribution of magnitudes in the event cloud we detect similar patterns
compared to stage A. The middle part of the seismic cloud is characterised with many events of low
magnitudes. North-East and South-West of this area events are characterised with higher magnitudes.
Figure 7 also shows the plots concerning magnitudes as function of the source-receiver-distance. For
Array-1 closest events are detected at distances of about 200 m. A fitted sloping line is presented as well
which increases with increasing source-receiver distance. Compared to the corresponding plots in stage
A we determine a steeper slope of the detection threshold. The moment magnitudes for both source-
receiver-distances show a range of -2.1 - 0. Array-2 features closest events detected at distances of about
400 m. Furthest located events of Array-1 and Array-2 feature average to high magnitude values at
source-receiver distances of approximately 650 m and 850 m.

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
90

400
0
Mom. Magnitude
Sensors
Injection Point

−0.5

Mom. Magnitude
Array-2
200
−1

Array-1
−1.5
0

−2
200 300 400 500 600 700 800 900
Northing [m]

Source−to−Array−1 Distance [m]

−200

0
−400
−0.5
Moment Magnitude

Mom. Magnitude
−0.5

−1 −1

−600
−1.5
−1.5

−2
−2
−800
−200 −100 0 100 200 300 400 500 600 400 500 600 700 800 900 1000 1100
Easting [m] Source−to−Array−2 Distance [m]

Figure 7: Left: Map view of event moment magnitudes, sensor arrays and injection point of stage B. Right: Moment
magnitudes versus source-receiver distance for Array-1 and Array-2. The sloping line represents an approximate
event detection threshold.

Stage C

In the third stage of the hydraulic fracturing treatment 1325 events have been detected and located. In
comparison to the other stages, stage C is the most Northern one. From the map view projection in Figure
8 one can see that the majority of the events are located in the North-Eastern part. In comparison to stage
A and B, stage C demonstrates the smallest distribution of events in terms of width. Both depth view
projections indicate seismicity distributed in a range of -1700 m to 1830 m.

The engineering data demonstrates an injection period of exactly 4 h (Figure 9). A total volume of
2799 m3 has been injected in this stage. Flow rates and injection pressures increase up to 252 l/s and
59 M P a. Similar to stage A and B there is also a pre-injection test. In this period the flow rates of
nearly 10 l/s was kept constant for about 15 minutes. Taking into account the r – t plot of induced
events there is a clear correlation with the onset of seismicity and corresponding flow rates. During
this period the treating pressure rapidly increases up to 46 M P a and decreases thereafter. After this
pre-injection test, the seismicity increases constantly in its rate of growth. Fluid flow rate and treating
pressure increase stepwise up to their maximum of 59 M P a and 252 l/s. After 1.2 h proppant has been
added. Although the treatment pressure decreases there is no change in the spatio-temporal pattern of the
seismicity distribution. With ongoing fluid injection proppant has been step wised increased up to 150
kg/m3 . At 2.5 h the concentration decreases suddenly and a second stepwise increase up to 255 kg/m3
follows thereafter. After termination of the fluid injection there is a decrease in the event density. From
4.7 h to 6.3 h one can see a number of events which back propagate towards the injection point.
Figure 10 shows a map view projection of the seismicity cloud where events are colour coded according
to their moment magnitude. In contrast to stage A and B, event cloud shows no clear pattern in the
distribution of magnitudes. High and low magnitudes are distributed in the entire cloud. Furthermore,
one can see that the range of magnitude strength is tighter compared to the range of stage A and B. Figure

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
Annual Report 2012 91

Figure 8: Overview of 1325 hypocenter locations of stage C presented in a map view and two depth view projections.
Furthermore the injection point and the monitoring system consisting of two sensor arrays are shown. Stage C
demonstrates the smallest distribution of events in terms of width.

10 shows magnitudes from ∼ -1.8 to ∼ -0.4. The figure also includes two plots in terms of magnitudes
as function of the source-receiver-distance. If we take into account the remote sensor Array-1 one can
see that closest events are detected in distances of approximately 200 m. For Array-2 the closest events
are registered at distances of about 350 m. A sloping line has been fitted on the bottom of the magnitude
versus source-receiver-distance plot for both arrays. In contrast to stage A and B this detection threshold
has the strongest increases. Furthest located events for Array-1 and Array-2 feature average magnitude
values at source-receiver-distances of approximately 600 m and 750m.

Conclusion

In this study we considered hydraulic fracturing induced data provided by one of the sponsors of the
PHASE research project. We analysed the distribution of induced microseismic event locations, spatio-
temporal behaviour in correspondence to engineering data and considered magnitudes of induced events.
One can see that stage A, B and C feature very similar patterns in terms of the distribution of hypocenter
locations in the event cloud. However, stage A and B show an irregular pattern of the rate of growth in
the r – t plots. Considering magnitudes of induced events, a limited detection range of both sensor arrays
is likely.

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
92

Engineering and r−t data stage C

400 80

Proppant Flow Rate [kg/m3] Treating Pressure

Treating Pressure [MPa]

Injection Rate
Injection Rate [l/s]

300 60
Proppant Flow Rate

200 40

100 20

0 0
0 1 2 3 4 5 6 7

700

600

500
Distance [m]

400

300

200

100

0
0 1 2 3 4 5 6 7
Time [h]

Figure 9: Engineering data and r – t plot of induced events from stage B. The dashed line represents the termination of
the fluid injection. The onset of seismicity correlates well with the beginning of the treatment. The first microseismic
events are triggered during the pre-injection test.

400
−0.4 Mom. Magnitude
Sensors
Injection Point
−0.6
Mom. Magnitude

Array-2 −0.8
200
−1

−1.2

Array-1 −1.4

0 −1.6

−1.8

200 400 600 800 1000

Northing [m]

Source−to−Array−1 Distance [m]

−200

−0.4

−400 −0.4 −0.6

−0.6
Mom. Magnitude

−0.8
Moment Magnitude

−0.8
−1
−1

−1.2 −1.2
−600
−1.4 −1.4
−1.6
−1.6
−1.8
−1.8
−800
−200 −100 0 100 200 300 400 500 600 400 600 800 1000
Easting [m] Source−to−Array−2 Distance [m]

Figure 10: Left: Map view of event moment magnitudes, sensor arrays and injection point of stage C. Right: Moment
magnitudes versus source-receiver distance for Array-1 and Array-2. The sloping line represents an approximate
event detection threshold.

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
Annual Report 2012 93

Outlook
The data might be appropriate in correspondence with a future fracture geometry and fluid transport
analysis according to the SBRC (Seismicity Based Reservoir Characterisation) approach introduced by
Shapiro et al. (1997, 1999, 2002). A prove for a limited detection of the sensors in South-West direction
of the operational setting in terms of the location of the injection point could not be provided so far. Since
there is a possibility of a symmetric expansion of the cloud, we continue to further analyse magnitudes in
terms of their source-receiver distances.

Acknowledgments
We thank the sponsors of the PHASE consortium for supporting the research presented in this paper. We
also thank the anonymous sponsor for providing us the data as well as the permission for publication.

References
Albright, J. N. and F., P. C. (1982). Acoustic emission as a tool for hydraulic fracture location: Experience
at the Fenton Hill hot dry rock site. Soc. Petro. Eng. J., 22:523–530.
Economides, M. J. and Nolte, K. G. (2000). Reservoir stimulation. John Wiley & Sons, Inc, New York.
Rutledge, J. T. (1998). Microseismic mapping of a cotton valley hydraulic fracture using decimated
downhole arrays. Society of Exploration Geophysicists.
Rutledge, J. T. and Phillips, W. S. (2003). Hydraulic stimulation on natural fractures as revealed by
induced microearthquakes, carthage cotton valley gas field, east texas. Geophysics, 68:441–452.

Shapiro, S. A., Audigane, P., and Royer, J.-J. (1999). Large-scale in situ permeability tensor of rocks
from induced microseismicity. Geophysical Journal International, 137:207–213.
Shapiro, S. A., Huenges, E., and Borm, G. (1997). Estimating the crust permeability from fluid-injection-
induced seismic emission at the KTB site. Geophysical Journal International, 131:F15–F18.
Shapiro, S. A., Rothert, E., Rath, V., and Rindschwentner, J. (2002). Characterization of fluid transport
properties of reservoirs using induced microseismicity. Geophysics, 114:212–220.

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
94

Overview of a Data Set of a three Stage Hydraulic Fracturing Treatment in a Shale Gas Reservoir
Annual Report 2012 95

Microseismic Monitoring: Physical

Fundamentals and Interpretation
96
Annual Report 2012 97

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations

and the Gutenberg Richter b-value of Earthquakes

C. Langenbruch1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, Fachrichtung Geophysik

Summary
The analysis of well log data from various sites suggests that the distribution of elastic properties in the
Earth’s crust is highly heterogeneous and of fractal nature. We extract statistical parameters of elastic
rock heterogeneity from sonic logs along the KTB-1 main hole and create an elastically heterogeneous
3D random medium representing the surrounding rock. Since it has been shown that the scaling behavior
of physical property heterogeneity derived from well logs is universal the scaling of elastic rock hetero-
geneity in our model can be considered as universally valid. Using the finite element program Abaqus
we apply an externally homogeneous stress (far) field and compute resulting stress fluctuations inside
the medium. We find that elastic heterogeneity causes significant stress fluctuations up to ±20% of the
externally applied stresses. Furthermore, we observe rotations of principle stress directions of up to ±6◦ .
We use geomechanical considerations to determine the distribution of Coulomb Failure Stress (CF S)
as a measure of fracture strength in rocks. The resulting CF S distribution is highly heterogeneous and
strongly related to initially assigned elastic parameters. We relate our results to the concept of brittle-
ness, which is used to detect sweet spots for hydraulic stimulation inside a reservoir. Our results show
that brittleness is a physically reasonable indicator, since the ease of fracture initiation and reactivation
increases with brittleness in our model. Fluid induced seismicity will hence most probably occur in and
migrate along brittle parts of a rock. We assume that rupturing, once it is initiated, takes place inside
isovolumes of CF S and determine the resulting size distribution of fractures. The distribution of fracture
sizes and correspondingly earthquake magnitudes exhibits power law scaling according to the Gutenberg
Richter law with b-value of b ≈ 1. Our result proofs that elastic heterogeneity in rocks cause significant
stress fluctuations of power law type. The scaling of fracture sizes resulting from the fluctuations suggest
that elastic rock heterogeneity is the origin of the Gutenberg Richter relation. Furthermore, we show that
the b-value is related to the degree of elastic rock heterogeneity, expressed in the fractal dimension of
elastic parameter distribution. Additionally, we find an inverse relation between differential stress level
and b-value. Our results coincide with findings of laboratory studies and observations. Finally, we discuss
that a relation between b-value and stress, stands for a break in self-similarity, which can only occur, if
naturally site specific characteristic length scales are existing.

Introduction
Sonic well logs provide in situ measurements of physical properties and their fluctuations in the Earth’s
crust. The analysis of collected log data has contributed to the characterization of elastic rock hetero-
geneity, since it proofed its fractal nature [see e.g. Leary (1997), Dolan et al. (1998), Goff and Holliger
(1999)]. Even though it is evident that elastic rock heterogeneity should result in a fluctuating stress
field, its influence on the distribution of stress in rocks has not yet been characterized. Has elastic hetero-
geneity a significant influence on stress at all? If so, does a power law distribution of elastic parameters
cause fractal stress variations? This would be of great importance, since it has been shown that fractal
stress fluctuations result in power law distribution of earthquake magnitudes, expressed in the Gutenberg
Richter Relation [Gutenberg and Richter (1954), Huang and Turcotte (1988)].

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
98

In this paper we analyze relations between elastic rock heterogeneity, stress fluctuations and the Guten-
berg Richter b-value of earthquakes. We start with a characterization of elastic rock heterogeneity by
analyzing sonic logs along the Continental Deep Drilling Site (KTB) main hole. Using derived param-
eters of elastic property distribution we simulate an elastically heterogeneous 3D random medium with
spatial correlation function of power law type. Day-Lewis et al. (2010) observe that the scaling behavior
of physical property heterogeneity derived from well logs is universal. Therefore, the scaling of elastic
rock heterogeneity in our model can be considered as universally valid. We apply an externally homo-
geneous stress (far) field and compute stress fluctuations inside the model. The externally applied stress
field is determined from smoothed stress profiles along the KTB main hole [see Ito and Zoback (2000)].
In the next section we interpret the occurring stress fluctuations in terms of fracture strength variations by
determining the distribution of Coulomb Failure Stress (CF S) as a measure of fracture strength in rocks.
In general, fracture behavior and complexity is controlled by directions and magnitudes of principle
stresses. These parameters are not directly accessible. Therefore, sweet spots inside a reservoir, which
should respond most effectively to hydraulic stimulation, have been defined in terms of brittleness [see
Rickman et al. (2008)], combining Poisson ratio and Young’s modulus. The obvious advantage of brittle-
ness is the possibility of in-situ measurements of elastic parameters by well logging. We analyze, if stress
conditions in brittle parts of a rock are preferable for development of complex fracture networks.
In the final part we analyze the scaling behavior of fracture strength distribution represented by the CF S.
Aki (1981) describes the assemblage of faults by the concept of fractals and shows that the Gutenberg
Richter relation is equivalent to a fractal distribution of fracture sizes. The power law exponent of fault
size distribution is equivalent to the b-value of earthquakes, describing the ratio between small to large
magnitude earthquakes. Assuming that rupturing, once it is initiated (e.g. by an increase of stress or pore
pressure), takes place inside isovolumes of CF S we determine the scaling exponent of resulting fracture
sizes and compare it to the b-value.
Laboratory studies [Mogi (1962) , Scholz (1962), Amitrano (2003)], observations [Schorlemmer et al.
(2005)] and modeling [Huang and Turcotte (1988)] suggest that the b-value is not universal. Generally,
a positive relation to the degree of rock heterogeneity and an inverse relation to the level of differen-
tial stress is observed. The fractal dimension of elastic parameter distribution is related to the degree of
material heterogeneity and describes the relation between large to small scale structures. Therefore, we
determine if the degree of elastic heterogeneity can explain observed b-value variations in laboratory. We
discuss the importance of the finding, that the degree of elastic heterogeneity measured from well logs at
various drilling sites seems to be universal. Furthermore, we analyze relations between differential stress
level and b-value resulting from our model. Because for formally defined fractals stress cannot have any
influence on the b-value, since variability of b stands for a break in self-similarity, we finally discuss the
influence of naturally existing characteristic length scales on the scaling of earthquake magnitudes.

Characterization of elastic heterogeneity of the Earth’s crust

In this section we characterize elastic heterogeneity of the Earth’s crust by analyzing sonic logging data
collected along the Continental Deep Drilling site KTB-1 main hole located in South-Eastern Germany
[see e.g. Pechnig et al. (1997)]. We analyze the depth section between 4450m and 6017m in the crys-
talline basement. We chose this section since fluid injected during an injection operation in the year
2000 stimulated the rock formation at this depth and induced numerous seismic events [see Baisch et al.
(2002)]. The occurrence of seismic events shows that the state of stress in this depth section is close to a
critical one causing brittle failure of rocks.
Figures 1a and 1b show shear (µ) and bulk (K) modulus computed from 4m averaged P- and S-wave
travel time and density logs along the KTB-1 main hole. We averaged the raw data (sampled at 0.152m)
in 4m intervals to eliminate inherent averaging of the logging process over the active length of the tool
(≈ 1m) and to reduce high frequency noise. The distributions of µ and K with depth are highly hetero-
geneous. Two narrow layers of paragneisses cut the metabasitic basement in depth sections 5210-5310m
and 5540-5640m [Pechnig et al. (1997)]. The moduli in these depth sections are characterized by lower
mean values, resulting in non stationary statistics of the distribution. Therefore, we exclude these layers
from our analysis to ensure statistical stationarity of elastic parameter distribution.
We now determine the statistical parameters describing the distribution of shear and bulk modulus to cre-

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
Annual Report 2012 99

Figure 1: a, b): Shear and bulk modulus computed from 4m averaged P- and S-wave travel time and density logs
along the KTB-1 main hole. Grey shaded areas indicate gneiss layers [see Pechnig et al. (1997)], which we exclude
from our analysis in c-g). c, d): Probability Density Function (PDF) of µ and K obtained from a) and b). e): Cross
plot of µ and K values. f): Cross plot of simulated random number pairs representing shear and bulk modulus. g, h):
Power Spectral Density Function (PSDF) of µ and K log data.

ate a random medium representing the rock surrounding the borehole. Since we exclude the paragneiss
layers from our analysis, the model will represent elastic heterogeneity occurring in one single type of
rock. Fig. 1c and 1d show the Probability Density Functions (PDF) of shear and bulk modulus. We
determine best fitting Gaussian distributions to obtain mean values (< µ >, < K >) and standard devi-
ations (σµ , σK ). Similarities between the distributions of µ and K with depth are visible in the logging
data. This indicates a positive relation between both moduli. To analyze this relation in more detail, we
present a cross plot of µ and K in Fig. 1 e. The moduli indeed show a positive relation, which we take
into account by determining the best fit linear relation and the deviation normal to it. We then apply
the obtained mean values, standard deviations and the relation between the moduli to simulate correlated
random number pairs representing bulk and shear modulus. A cross plot of simulated moduli is presented
in Fig. 1f. Random assignment to a 3D medium consisting of 100x100x100 equally sized cells, results in
two related spatially uncorrelated media (see Fig. 2a).

However, real rocks show spatial correlations of elastic properties. Data analysis from various drilling
sites suggest unbounded fractal scaling of the Earth’s crust heterogeneity. This is expressed in power
law (k −β ) dependence of the log data’s power spectra on the wavenumber k [Leary (1997)]. The scaling
exponent β is given by: β = 2H + E, where E is the Euclidean dimension and H is the Hurst exponent,
which can be related to the fractal dimension D of the log data by: D = (E + 1) − H [see e.g. Dolan
et al. (1998)]. The power spectral densities of shear and bulk modulus logs at the KTB (see Fig. 1g and
1h) possess power law dependence on the wave number in the complete range. Power law exponents β
are given by β = 1.29 (H = 0.145) for shear and β = 1.24 (H = 0.12) for bulk modulus. Therefore,
we will apply fractal scaling according to a Hurst exponent H = 0.13 to to the delta correlated media
shown in Fig. 2a. This is realized by filtering the medium in the wave number domain. More precisely,
the Fourier transforms of the simulated shear and bulk modulus distributions are multiplied by the square
root of the fractal Power Spectral Density Function P SDF (k) = k −β , in the wave number domain,

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
100

Figure 2: Random media realizations. a) Related spatially uncorrelated media representing shear (top) and bulk
modulus (bottom). The media result from a random assignment of simulated shear and bulk moduli pairs shown
in Fig. 1 f). b-c) show the same media after spatial correlation according to PSDF of power law type. The spatial
correlation results in media with fractal dimension of b) DE = 3.1, c) DE = 3.5 and d) DE = 3.87. The media
shown in column d) represent the rock surrounding the KTB-1 borehole and is used for further modeling. The figure
shows that the fractal dimension D is a measure for the degree of rock heterogeneity.

q
where k = kx2 + ky2 + kz2 and β = 3.26 (E = 3). By taking the inverse Fourier transform we obtain
the fractal random media in the spatial domain. The resulting distributions of µ and K (see Fig. 2d) are
characterized by a fractal dimension of DE = 3.87. Fig. 2b and c show fractal media characterized by
lower fractal dimensions of DE = 3.1 and DE = 3.5 respectively. The figure illustrates that the fractal
dimension describes the relation of small to large scale structures and can be seen as a measure of rock
heterogeneity.

Stress fluctuations in elastically heterogeneous rocks.

In this section we analyze stress fluctuations resulting from elastic rock heterogeneity and establish re-
lations between stress fluctuations and elastic properties. Therefore, the medium realization of elastic
moduli (Fig. 2d) is used as input to an ABAQUS Finite Element stress analysis model. An externally
homogeneous stress (far) field is applied to determine the evolution of stress inside the elastically het-
erogeneous rock model. Compressive stresses are always defined positive in the following. We consider
smoothed stress profiles reported by Ito and Zoback (2000): SH = 0.045 MP a
m , SV = 0.028 m ,
MP a

Sh = 0.02 MP a
m . The stress profiles are derived from hydraulic fracturing tests, breakouts and drilling
induced fractures along the KTB main hole. A strike-slip stress regime is prevailing and the differential
stress is increasing with depth [Brudy et al. (1997)]. We calculate effective principle stresses, considering
a hydrostatic pore pressure gradient of 0.0115 MP m
a
reported by Huenges et al. (1997) at the KTB up
to a depth of 9.1km. The principle stress components of the externally applied stress field are defined
at the corresponding boundary surfaces of the model medium as: σ1e = 180.9 M P a = σyy , σ2e =
89.1 M P a = σzz , σ3e = 45.9 M P a = σxx . These values correspond to maximum horizontal, vertical
and minimum horizontal effective principle stresses in a depth of 5.4km. Again we note that we consider
this depth since a fluid injection in the year 2000 induced seismic events in this depth range. By evalua-
tion of the ABAQUS model we obtain the full stress tensor in each model cell. Fig. 3 shows the resulting

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
Annual Report 2012 101

normal stress components in y(σ1e ), z(σ2e ) and x(σ3e ) directions.

σ σ σ
1e 1e 1e

σ σ2e
205 σ σ2e σ σ2e 54
3e 3e 3e
100
200 200
200 200 52
195

σxx
50
σyy 190 σzz 95

48
185
y[m]

y[m]
0

y[m]
0 0 90 46
180

175 44
85
170
42
−200 −200
165 −200 200
200 200 40
80 200
200 160 200
0 0 38
0 155 0 0
0 x[m]
x[m] z[m] MPa x[m] MPa z[m] MPa
−200 −200 z[m] −200 −200
−200 −200

Figure 3: Stress modeling results: Normal stress component in y(σ1e ), z(σ2e ) and x(σ3e ) directions. The color bars
are scaled to ±20% of the externally applied stresses. The directions of externally applied principle stresses are
shown on top of the figures.

In the case of an elastically homogeneous medium the stress components would coincide with σ1e , σ2e
and σ3e in all cells of the model. We find that elastic heterogeneity has a significant influence on stress
magnitudes, which vary by up to more than ±20% of the externally applied stresses. We note that the
directions of normal stresses shown in Fig. 3 do not correspond to principle stress directions. We find
that due to elastic heterogeneity the orientation of principle stresses vary by up to ±6◦ .
Since we are interested in analyzing the importance of elastic heterogeneity in relation to the occurrence
of seismicity, we now analyze its influence on fracture strength of rocks. The strength of fractures can
be described by the Coulomb Failure Stress (CF S) associated with the Mohr Coulomb failure criterion
according to [see e.g. Jaeger et al. (2007)]:
CF S = −τ + µ(σn − pp ) + S0 , (1)
where τ and σn are shear and normal stress acting on the fault plane. S0 is the cohesion, that is, the
shear stress necessary to initiate failure in absence of normal stress, pp is the pore pressure and µ is the
coefficient of internal friction. A fault is in a stable state, if CF S > 0. A fault is in an unstable state, if
CF S ≤ 0 and failure occurs if CF S = 0. We assume that the rock is able to fail in optimal orientation
for failure in each model cell. It means that each cell can be considered as the location of a fracture
optimally oriented for failure. Furthermore, we assume a friction coefficient of µ = 0.9 and cohesion-
less fractures (S0 = 0). Since we want to analyze only the influence of elastic heterogeneity we keep this
value constant in all model cells. Using the principle stresses in the individual model cells we compute
the shear and normal stress acting on optimal oriented fractures and the resulting CF S according to Eq.
1. We note that due to the rotations of principle stress components in the medium the optimal orientation
for failure in the coordinate system of externally applied stresses is not the same in all model cells.
a) b)
20 4
x 10
200 6 Figure 4: a):
Spatial dis-
15 5 tribution of
number of cells

CFS 4 Coulomb Failure

y[m] Stress resulting
10 3 from the stress
distribution in
2
the elastically
−200 5
200 1 heterogeneous
model. b)
0 0 Histogram of
x[m] 0 0 5 10 15 20 25
z[m]
MPa CFS [MPa] CF S.
−200 −200
Fig. 4 shows the resulting spatial distribution and a histogram of computed CF S. We find that elastic
heterogeneity causes significant CF S variations. Some parts of the medium are in direct vicinity to

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
102

failure, suggesting that already very small stress or pore pressure perturbations may cause brittle failure
in rocks. This finding is confirmed by observed fluid injection induced seismicity [see Baisch et al.
(2002)] and natural seismic activity in the vicinity of the KTB [see Dahlheim et al. (1997)]. Nevertheless,
in the most stable parts of the model CF S perturbations of up to 20 MPa are needed to initiate failure.
This range of fracture strength is in agreement with rock strength reconstructed from microseismicity in
sedimentary and crystalline rocks [Rothert and Shapiro (2007)]. Fig. 5 shows a cross plot of bulk and
shear modulus pairs that where initially assigned to the model cells. The input is color coded according to
the resulting CF S in the corresponding cells. The figure illustrates that elastic moduli, stress and fracture
strength fluctuations are strongly related. Cells characterized by a low bulk and a high shear modulus are
in a stress state close to failure. The fracture strength represented by the CF S is systematically increasing
with increasing bulk and decreasing shear modulus. Failure will hence most likely occur in parts of a rock
characterized by low bulk and high shear modulus, while parts of high bulk and low shear modulus act
as barriers for fracture propagation. The direct relation between elastic properties and CF S suggests that
elastic heterogeneity causes fractal stress variations. We will analyze the scaling behavior of CF S and
its implications for earthquake magnitudes scaling after the next section in which we concentrate on the
relations between stress and elastic property distribution in terms of rock brittleness.

Figure 5: Cross plot

of elastic input param-
eters of stress mod-
eling. The input is
color coded according
to the CF S in the cor-
responding cells. A
clear relation between
elastic input parame-
ters and the fracture
strength represented by
the CF S is visible.

Relations between rock brittleness, stress fluctuations and fluid induced seismicity
Rickman et al. (2008) define the Brittleness Index BI of a rock as a combination of Poisson ratio ν and
Young’s modulus E:
 
E − Emin ν − νmax 100
BI = + ∗ , (2)
Emax − Emin νmin − νmax 2
where Emin , νmin , Emax and νmax are minimum and maximum Young’s modulus and Poisson ratio in
the reservoir of interest. Brittleness is a relative quantity of a reservoir and according to Rickman et al.
(2008) it describes the rocks ability to fail under stress (Poisson ratio) and to maintain a fracture (Young’s
modulus) once the rock fractures. Brittle zones of the reservoir, characterized by high E and low ν val-
ues, seem to be best suitable for hydraulic stimulation since they allow the development of a complex
fracture network. However, opening and grows direction of hydraulic fractures and the reactivation of
pre-existing fractures define the complexity of the fractured zone. In general, fracture opening, growth,
and reactivation is controlled by magnitudes and directions of stresses inside a reservoir rock. Obviously,
brittleness and stress distribution inside a reservoir should be related, if brittleness is a physically reason-
able indicator. We now analyze relations between brittleness, stress and reactivation pressure distribution
to analyze, if stress conditions in more brittle parts of a rock are more beneficial for the development of a
complex fracture network.
In the frame of developing Enhanced Geothermal Systems (EGS) and hydrocarbon reservoirs high pres-

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
Annual Report 2012 103

surized fluids are injected. Although, the KTB is not designed to be utilized for heat extraction the basic
conditions are quite similar to EGS sites. However, fluid injection operations into boreholes frequently
induce detectable sometimes also perceptible seismicity. The main factor controlling the triggering of
these seismic events is the relaxation of pore pressure perturbation initially create around the open hole
section [see Langenbruch et al. (2011)]. Therefore, we analyze the amount of additional pore pressure
C above which reactivation of pre-existing fractures is possible. These pressures can be seen as a mea-
sure for fracture strength. The value of C corresponds to the pore pressure perturbation pp resulting in
CF S = 0. After rewriting Eq. 1 we compute fracture reactivation pressures C according to:
τ
C=− + σn , (3)
µ

Fig. 6a shows a cross plot of elastic input parameters to stress modeling. The coloring corresponds
to brittleness calculated after Eq. 2. E and ν values are computed from K and µ distributions shown
in Fig. 2d. The color coding in Fig. 6b and c correspond to fracture reactivation pressures (C) and
minimum principle stress (σ3 ) resulting from stress modeling. The inverse relation between BI and
C, which is shown in panel d), suggests that fracture reactivation and correspondingly seismicity will
preferentially occur in more brittle parts of a rock, because fracture reactivation pressures are lower. The
inverse relation between BI and σ3 (Fig. 6f) shows that hydraulic fractures will also open more easily
in brittle parts of a reservoir, since the injection pressure has to exceed the magnitude of σ3 to allow a
new hydraulic fracture to form in the reservoir. Taking also into account the positive relation between BI
and the maximum principle stress component (σ1 ), we find that differential stress (∆σ = σ1 − σ3 ) will
increase with brittleness of the rock. In summary, the ease of fracture initiation and reactivation increase
with brittleness of a reservoir rock. This suggests that also the complexity of the stimulated fracture zone
will increase with brittleness. Thus, seismicity will most probably occur in and migrate along brittle parts
of a rock characterized by high values of E and low values of ν. Less brittle parts of a rock are good
reservoir seals, because fractures will very unlikely propagate into these parts of a rock.

Figure 6: a-c): Elastic input parameters of stress modeling color coded according to brittleness (a), reactivation
pressures (b), minimum principle stress (c). d-f): Relations between brittleness and fracture reactivation pressures
(d), minimum principle stress (e) and maximum principle stress (f).]

In the case of unconventional shale reservoirs elastic anisotropy may play an important role for stress fluc-
tuations and their relation to brittleness. Modern logging technologies and laboratory core measurements
allow to characterize elastic anisotropy. The modeling of stress and fracture strength in shale reservoirs as
Vertical Transverse Isotropic (VTI) fractal media is part of our current research and will be further devel-
oped. The influence of anisotropy on stress distribution will be characterized and the relations between
brittleness, reactivation and initiation pressures will be extended to the anisotropic case.

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
104

Earthquake magnitude scaling

Considering the fact that zones of failure in rocks are not continuous planes but exhibit self similar
fragmentation at all scales, Aki (1981) describes the assemblage of faults by the concept of fractals.
3
Assuming that: M0 ∝ A 2 and log(M0 ) ∝ 32 M [see Kanamori and Anderson (1975)], where M0 is
the seismic moment, A is the fault area and M is the magnitude, the Gutenberg Richter relation [see
Gutenberg and Richter (1954)]:
log(NM ) = a − bM, (4)
can be expressed in the following form [see Aki (1981)]:

log(NM ) = a − b log(A). (5)

The constants a and b characterize earthquake productivity and the ratio between small to large magnitude
earthquakes, respectively. The Gutenberg Richter relation in the form of Eq. 5 suggests that the power
law exponent of fault size distribution is equivalent to the b-value of earthquakes. We now analyze, if
fracture sizes resulting from our model of elastic rock heterogeneity possess power law scaling and an
scaling exponent similar to the b-value of earthquake.
If stress transfer is neglected a decrease of CF S will initiate failure in all critically stressed cells char-
acterized by: CF S ≤ 0. Causes of a CF S decrease are for instance tectonic loading or an increase
of pore pressure, resulting from a fluid injection or ascending crustal fluids. In the most simple case of
a constant CF S decrease ∆CF S in the complete medium the isoset CF S(x, y, z) ≤ ∆CF S defines
the number size distribution of faults and accordingly magnitude scaling. Each single closed cluster of
interconnected critically stressed cells represents one fault in our model. In the previous section we have
shown that the CF S is strongly related to the fractal distribution of elastic properties. Therefore, the
fluctuations of CF S should also possess fractal scaling. Consequently, scaling of fault sizes is explicitly
defined by the fractal dimension Di of the isoset: CF S(x, y, z) ≤ ∆CF S. Isichenko and Kalda (1991)
show that an isoset of a fractal distribution again is a fractal with fractal dimension of one unity less than
the fractal distribution itself. Thus, the number of clusters Nj characterized by a volume equal or larger
than a given volume Vj , that is, the number of clusters consisting of j or more interconnected critically
stressed cells, is given by:
c
Nj = Di , (6)
Vj 3
where Di = DC − 1 is the fractal dimension of the isoset: CF S(x, y, z) ≤ ∆CF S, DC is the fractal
dimension of the CF S distribution. The constant c is given by the number of all clusters and is related to
the a-value of the Gutenberg Richter relation. If we consider that failure takes place in all cells of a cluster
the fault area is proportional to the number of cells. Correspondingly, the magnitude is proportional to
the logarithm of the number of cells. Thus, the b-value is given by:
Di DC − 1
b= = . (7)
3 3
Accordingly, the b-value will be in the range of: 23 ≤ b ≤ 1, since 3 < DC < 4. If the CF S distribution
is characterized by a fractal dimension close to the fractal dimension of elastic property distribution
(DE = 3.85), fault sizes should scale with a b-value of: b = DE3−1 = 0.95.
Fig. 7 presents CF S histogram, fracture assemblage and cumulative size distribution of critically stressed
clusters resulting from an uniform CF S decrease of ∆CF S = 5M P a in the complete model medium.
A decrease of 5M P a is applied to assure a statistically significant number of failing cells. All cells char-
acterized by CF S ≤ 0 contribute to the fracture assemblage shown in the middle panel. As discussed
above the logarithm of the number of cells (x-axis of Fig. 7c) represents earthquake magnitudes. The
modeled fault sizes exhibit power law scaling. Furthermore, the power law exponent of fault size scaling
is remarkably similar to the b-value of earthquakes, which is usually given by b ≈ 1. A b-value of b ≈ 1
is also observed for magnitude scaling of microseismicity induced by a fluid injection operation at the
KTB [see Dinske and Shapiro (2012)].

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
Annual Report 2012 105

4
modeling results: scaling of fault sizes
3.5 scaling of earthquake magnitudes: b=1

log10(cumulative number)
3

2.5

1.5

0.5

0
0 1 2 3 4
log10(number of cells)

(a) (b) (c)

Figure 7: Statistics of critically stressed clusters after an uniform CF S decrease of 5M P a in the complete model
medium. Left: Histogram of CF S. Middle: Fracture assemblage resulting from all cells characterized by:
CF S ≤ 0. Right: Cumulative size distribution of critically stressed clusters. As discussed above the logarithm
of the number of cells (x-axis) represents earthquake magnitudes. The modeled magnitudes scale according to the
Gutenberg Richter relation with a b-value of b ≈ 1.

Our results proof that fractal fluctuations of elastic properties in the Earth’s crust cause significant fractal
stress and fracture strength fluctuations. Since magnitudes, determined by the resulting fault size distribu-
tion, scale according to a power law with exponent close to one, elastic heterogeneity can be considered
as the origin of the Gutenberg Richter relation of earthquake magnitude scaling. The assumption of a
fault area A proportional to the number of cells presumes very complex fracture surfaces. Resulting b-
values can be interpreted as a lower limit. The upper bound of b-values should be given for even fracture
2 2
surfaces resulting in A ∝ Nj3 , where Nj3 corresponds to the characteristic area of a critically stressed
cluster. In this case the b-value is given by: b = D2i , and accordingly in the range of 1 ≤ b ≤ 1.5. The
b-value resulting from our model of elastic rock heterogeneity would be given by b = 1.43 in this case.
However, we will assume complex fracture surfaces A ∝ Nj in the following and keep in mind that the
upper bound of b-values is given by b = 1.5, which is still in the range of observed values.

b-value variability
Laboratory studies [Mogi (1962) , Scholz (1962), Amitrano (2003)], observations [Schorlemmer et al.
(2005)] and modeling [Huang and Turcotte (1988)] suggest that the b-value is not universal. One ob-
servation in laboratory is a positive relation between the degree of specimen heterogeneity and b-value.
Fig. 2 illustrates that the fractal dimension of elastic parameter distribution DE is related to the degree
of rock heterogeneity. DE is directly linked to the power law exponent of the P SDF and describes the
ratio between small to large scale structures. Similarly, the b-value describes the ratio between small
and to large magnitude earthquakes. We now determine if the degree of elastic heterogeneity (DE ) can
explain the observed b-value variations in laboratory. Therefore, we apply P SDF of power law type
with different exponents to the media shown in Fig. 2a to simulate fractal media characterized by fractal
dimensions DE in the range from 3.1 to 3.9. Stress modeling is performed in the same way as in the
previous sections. Fig. 8 shows how scaling of CF S fluctuations depends on the fractal dimension DE
of elastic property scaling. The obvious positive relation of b-value and fractal dimension DE confirms
the observed increase of b-value with specimen heterogeneity in laboratory.
One might think about the possibility that observed b-value variations of earthquakes are caused by the
variations of the fractal dimension of elastic property distribution. Although type and composition of
rocks vary from one drilling site to another, it has been observed that the scaling behavior of physical
property heterogeneity is similar [see Day-Lewis et al. (2010) and references within]. More precisely,
scaling according to a power law exponent of β ≈ 1 is typical for physical property scaling derived from
sonic well logs. Therefore, it is not possible to estimate b-values variations from sonic logging data by
determining the fractal dimension of elastic property distribution. However, it shows that the scaling

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
106

of our model medium can be considered as universally valid. Elastic heterogeneity of the Earth’s crust
should hence results in an universal b-value close to b = 1. Although there are regional fluctuations of
the b-value, the b-value for worldwide earthquake catalogs is given by b = 1.

DE=3.1 DE=3.2 DE=3.3 DE=3.4 D =3.5

E E
D =3.6 DE=3.7 DE=3.8 DE=3.9
increasing degree of heterogeneity

4 1

3.5
0.9
3
log10(cumulative number)

0.8
2.5

b−value
2 0.7

1.5
0.6
1

0.5
0.5

0 0.4
0 0.5 1 1.5 2 2.5 3 3.5 4 3.2 3.4 3.6 3.8
log (number of cells) DE
10

Figure 8: Influence of the fractal dimension of elastic parameter distribution (DE ) on fault size scaling. Left: Cumu-
lative number size distributions resulting from stress modeling with input fractal dimension DE ranging from 3.1 to
3.9. The figure shows fault sizes computed for CF S decrease of ∆CF S = 5M P a. Right: Relation between input
fractal dimension DE and resulting b-value.

A second observation [Scholz (1962)] is a b-values decrease with increasing differential stress applied in
laboratory. Schorlemmer et al. (2005) connect the degree of differential stress to different tectonic stress
regimes and find an inverse relation between stress and b-value. Thus, high differential stress should
results in a low b-value, and vice versa. We now analyze the influence of the stress level on the b-value
by varying the amount of CF S perturbation ∆CF S. The analysis performed in the following is always
related to elastic property scaling derived from the KTB-1 well logs (DE = 3.85). Fig. 9 shows b-values
computed for CF S perturbations in the range from 1.5 to 13.5M P a. Higher values of ∆CF S represents
higher levels of differential stress, since an increase of differential stress results in a decrease of CF S,
that is, it brings fractures closer to failure.
The results presented in Fig. 9 confirm the observed inverse relation between differential stress level
and b-value. Precisely, we observe a decrease of the b-value with increasing stress level until the per-
colation threshold is reached, that is, until a cluster of interconnected critically stressed cells connects
the boundaries of the model. The percolation threshold represents a critical point in the system and may
correspond to the point of macroscopic sample failure in laboratory. Usually, a decrease of the b-value
is observed before large slip events in samples. After occurrence of a percolation cluster the b-value is
increasing with differential stress, since more and more larger clusters get connected, while small clus-
ters still can develop. Huang and Turcotte (1988) analyzed 2D fractal fields representing the difference
between stress and strength. They find an inverse relation between stress level and b-value in a larger
range of stress levels. Here it is important to note that the percolation threshold in 2D is larger than the
percolation threshold in 3D. It also seems that Huang and Turcotte (1988) fit the cumulative number size
distributions in the complete range, while we exclude the largest cluster from the determination of the
b-value. Our analysis corresponds to simple cubic site percolation. In the case of percolation of spatially
uncorrelated fields (e.g. Fig. 2a) the critical site occupancy probability is given by: pc = 0.53 − 0.59 for
2D and pc = 0.3116077 for 3D [see e.g. Lorenz and Ziff (1998)]. It means, that as soon as 31.16% of
sites are occupied (CF S ≤ 0) a percolation cluster will occur for 3D uncorrelated random fields. Due to

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
Annual Report 2012 107

Figure 9: Influence of the CF S perturbation on scaling of critically stressed clusters. Left: Cumulative number size
distributions resulting from CF S decrease ∆CF S in the range from 1.5 to 13.5M P a. Higher values of ∆CF S
represent higher differential stress levels.

the power law correlation of elastic properties and correspondingly CF S, we find a much lower threshold
of pc = 0.11.
However, for formally defined fractals stress cannot have any influence on the power law exponent (b-
value), since variability of the exponent stands for a break in self-similarity. Isichenko (1992) notes that
no physical object in real space qualifies for the formal definition of a fractal, because each physical
model has certain limits of applicability expressed in characteristic length scales involved. As discussed
above, one characteristic scale in our model is the percolation threshold. However, the most important
limitation of our model is given by finite model and cell size. Clusters touching the boundaries of the
model are counted as smaller than real, since they actually may continue outside of the model medium.
Therefore, the size distribution is biased at all scales and deviates from a precisely self-similar behavior.
As the number and correspondingly also sizes of clusters increase, if more cells fail due to a higher stress
level, the bias in the size distribution is getting stronger. This strength dependency of boundary effect on
the number of failing cells explains the observable relation between b-values and stress level. The limita-
tions of our model, that is, a finite size and a constant stress level at the boundary surfaces are comparable
to studies at rock samples in laboratory. The observations in laboratory coincide with the results of our
modeling study.
Also in nature characteristic length scales are present and cause deviations from a strict power law be-
havior of earthquake magnitude scaling. For instance layering in rocks, that is, changes of rock type
and composition represents a characteristic scale. During the analysis of log data at the KTB we have
excluded narrow layers of paragneisses, because they caused non-stationary statistics of elastic property
distribution (see Fig. 1). Including these layers in our model would introduce characteristic scales in
the distribution of elastic properties and correspondingly in the resulting CF S distribution. A further
example of a naturally existing characteristic length scale is provided by the observation of a break in
self-similarity, from small to large earthquakes, at a point where the dimension of the event equals the
down dip width of the seismogenic layer [Pacheco et al. (1992)]. The finiteness and geometry of the CF S
perturbed volume introduces a further characteristic scale. In our analysis we have assumed that CF S is
decreased by a constant amount in the complete model medium. This of course is a strong simplification
of perturbations occurring in nature, since it is clear that natural perturbations always are limited to finite
volumes and may show complex geometries and heterogeneities in the level of perturbation. For instance,
Shapiro et al. (2011) analyze the influence of the finiteness and geometry of the perturbed zone resulting
from fluid injections into geothermal and hydrocarbon reservoirs. Their results show, that finite size and
geometry cause strong deviations of a and b-values compared to values resulting from a strict power law
distribution of fault sizes. In summary, the observed variability and stress dependency of the b-value
results from naturally existing site specific characteristic length scales.

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
108

Conclusions
We have shown that elastic rock heterogeneity has significant impact on stress distribution in the Earth’s
crust. The proven fractal nature of heterogeneity results in fractal stress fluctuations with significant mag-
nitudes. As a consequence, fracture size distribution of earthquakes exhibit power law scaling according
to the Gutenberg Richter relation. Consistent with observations in laboratory, we find that in theory the b-
value is increasing with the degree of elastic rock heterogeneity. Lower and upper bound of the b-values,
which solely can be explained by the elastic rock heterogeneity, are given by b = 0.6 and b = 1.5, respec-
tively. However, we show that due to the universal scaling behaviour of elastic heterogeneity, measured
from log data at various drilling sites around the world, the b-value should be universal and close to b = 1.
This statement holds only for formally defined fractals, which only can exist if no characteristic length
scales are introduced. Since in all physical models and ongoing seismogenic processes in nature char-
acteristic length scales are involved, deviations from a strictly self-similar earthquake magnitude scaling
occur. Because involved characteristic length scales, for instance the finite size and geometry of the stress
perturbed volume or changes in rock composition, are site and in some cases also time dependent quanti-
ties, b-values determined for different regions or at different times are diverse. We find an inverse relation
between differential stress and b, which coincides with results of laboratory studies and observations.
A relation between stress level and b-value stands for a break in self-similarity and can only occur, if
characteristic length scales are present. In summary, our stress modeling analysis according to a model
of elastic rock heterogeneity, derived from sonic logs along the KTB-1 main hole, explicitly shows that
elastic rock heterogeneity can be considered as the origin of the Gutenberg Richter relation of earthquake
magnitude scaling.
The obvious relations between elastic properties and stress fluctuations, established in this paper, proof
the physical meaning of rock brittleness as a sweet spot indicator for hydraulic stimulation. Fractures will
be initiated and reactivated at lower pressure perturbations in brittle parts of a reservoir rock. Therefore,
brittle rock failure and accordingly seismicity will most probably occur in and migrate along brittle rock
sections. Consequently, effectiveness of hydraulic reservoir stimulations and resulting fracture complex-
ity increases with brittleness.

Acknowledgments
We thank the sponsors of the PHASE consortium project for supporting the research presented in this
paper.

References
Aki, K. (1981). A probabilistic synthesis of precursory phenomena, in earthquake prediction: An inter-
national review. Maurice Ewing Ser., 4:doi:10.1029/ME004p0566.

Amitrano, D. (2003). Brittle-ductile transition and associated seismicity: Experimental and numerical
studies and relationship with the b value. JGR, 108(B1):doi:10.1029/2001JB000680.

Baisch, S., Bohnhoff, M., Ceranna, L., Tu, Y., and Harjes, H.-P. (2002). Probing the crust to 9-km depth:
Fluid-injection experiments and induced seismicity at the ktb superdeep drilling hole, germany. BSSA,
92:doi: 10.1785/?0120010236.

Brudy, M., Zoback, M., Fuchs, K., Rummel, F., and Baumgartner, J. (1997). Estimation of the complete
stress tensor to 8 km depth in the KTB scientific drill holes: Implications for crustal strength. J.
Geophys. Res., 102:doi:10.1029/96JB02942.

Dahlheim, H., Gebrande, H., Schmedes, E., and Soffel, H. (1997). Seismicity and stress field in the
vincinity of the KTB location. J. Geophys. Res., 102:doi:10.1029/96JB02812.

Day-Lewis, A., Zoback, M., and Hickman, S. (2010). Scale-invariant stress orientations and seismicity
rates near the san andreas fault. GRL, 37:doi:10.1029/2010GL045025.

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
Annual Report 2012 109

Dinske, C. and Shapiro, S. (2012). Seismotectonic state of reservoirs inferred from magnitude distribu-
tions of fluid-induced seismicity. Journal of Seismology, pages doi:10.1007/s10950–012–9292–9.

Dolan, S., Bean, C., and Riollet, B. (1998). The broad-band fractal nature of heterogeneity in the upper
crust from petrophysical logs. GJI, 132:doi:10.1046/j.1365–246X.1998.00410.x.

Goff, J. and Holliger, K. (1999). Nature and origin of upper crustal seismic velocity fluctuations and as-
sociated scaling properties- combined stochastic analyses of KTB velocity and lithology form, chance,
and dimensions. W. H. Freeman and Company, 104.

Gutenberg, B. and Richter, C. F. (1954). Seismicity of earth and associated phenomenon, 2nd ed. Princton
Univ. Press, Princton, N. J.

Huang, J. and Turcotte, D. (1988). Fractal distributions of stress and strength and variations of b-value.
Earth and Planetary Science Letters, 91:doi:10.1016/0012–821X(88)90164–1.

Huenges, E., Erzinger, J., Kueck, J., Engeser, B., and Kessels, W. (1997). The permeable crust: Geohy-
draulic properties down to 9101 m depth. J. Geophys. Res., 102:doi:10.1029/96JB03442.

Isichenko, M. (1992). Percolation, statistical topography, and transport in random media. Reviews of
Modern Physics, 64.

Isichenko, M. B. and Kalda, J. (1991). Statistical topography. i. fractal dimension of coastlines and
number-area rule for islands. Journal of Nonlinear Science, 1:doi:10.1007/BF01238814.

Ito, T. and Zoback, M. (2000). Fracture permeability and in situ stress to 7 km depth in the KTB scientific
drillhole. Geophys. Res. Lett., 27:doi:10.1029/1999GL011068.

Jaeger, J., Cook, N., and Zimmerman, R. (2007). Fundamentals of rock mechanics 4th edition. Blackwell
Publishing.

Kanamori, H. and Anderson, D. (1975). Theoretical basis of some empirical relations in seismology.
BSSA, 65:1073–1095.

Langenbruch, C., Dinske, C., and Shapiro, S. (2011). Inter event times of fluid induced earthquakes
suggest their Poisson nature. GRL, 38:doi:10.1029/2011GL049474.

Leary, P. C. (1997). Rock as a critical-point system and the inherent implausibility of reliable earthquake
prediction. GJI, 131:doi:10.1111/j.1365–246X.1997.tb06589.x.

Lorenz, D. and Ziff, R. (1998). Universality of the excess number of clusters and the crossing prob-
ability function in three-dimensional percolation. J. Phys. A: Math. Gen., 31:doi:10.1088/0305–
4470/31/40/009.

Mogi, K. (1962). Magnitude-frequency relation for elastic shocks accompanying fractures of various
materials and some related problems in earthquakes (2nd paper). Bull. Earthquake Res. Inst., 40.

Pacheco, J., Scholz, C., and Sykes, L. (1992). Changes in frequency-size relationship from small to large
earthquakes. Nature, 355:doi:10.1038/355071a0.

Pechnig, R., Haverkamp, S., Wohlenberg, J., Zimmermann, G., and Burkhardt, H. (1997). Integrated log
interpretation in the german continental deep drilling program: Lithology, porosity, and fracture zones.
JGR, 102:doi:10.1029/96JB03802.

Rickman, R., Mullen, M., Petre, E., Grieser, B., and Kundert, D. (2008). A practical use of shale
petrophysics for stimulation design optimization: All shale plays are not clones of the barnett shale.
SPE:115258, pages doi:10.2118/115258–MS.

Rothert, E. and Shapiro, S. A. (2007). Statistics of fracture strength and fluid-induced microseismicity.
JGR, 112:doi:10.1029/2005JB003959.

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
110

Scholz, C. (1962). The frequency-magnitude relation of microfracturing in rock and its relation to earth-
quakes. BSSA, 58:399–415.
Schorlemmer, D., Wiemer, S., and Wyss, M. (2005). Variations in earthquake-size distribution across
different stress regimes. Nature, 437:doi:10.1038/nature04094.
Shapiro, S., Krueger, O., Dinske, C., and Langenbruch, C. (2011). Magnitudes of induced earthquakes
and geometric scales of fluid-stimulated rock volumes. Geophysics, 76:doi:10.1190/geo2010–0349.1.

Relations Between Elastic Rock Heterogeneity, Stress Fluctuations and the Gutenberg Richter b-value of
Earthquakes
Annual Report 2012 111

Triggered versus Induced Events and Bounds of the Magnitude

Probability

S. A. Shapiro1 , O. S. Krüger1 and C. Dinske1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
Fluid-induced microearthquakes in reservoirs, aftershocks or seismic emission in rock samples are ex-
amples of the seismicity resulting from an activation of finite rock volumes. Such a finiteness of per-
turbed volumes influences frequency-magnitude statistics. Previously we observed that fluid-induced
large-magnitude events at geothermal and hydrocarbon reservoirs are frequently underrepresented in
comparison with the Gutenberg-Richter law. This is an indication that the events are much more proba-
ble on rupture surfaces contained within the stimulated volume. Here we theoretically and numerically
analyze this effect. We consider different possible scenarios of event triggering: rupture surfaces located
completely within or intersecting only the stimulated volume. We approximate a stimulated volume by
an ellipsoid or cuboid and derive the statistics of induced events from the statistics of random thin flat
discs modeling rupture surfaces. We derive lower and upper bounds of the probability to induce a given-
magnitude event. The bounds depend on the characteristic scales of the stimulated volume. Its minimum
principal axis is the most influential geometric parameter. We compare our analytical results with data
on seismicity induced by fluid injections in boreholes. Fitting the bounds to the frequency-magnitude
distribution can provide an estimate of a largest expected induced magnitude and a characteristic stress
drop, in addition to improved estimates of the Gutenberg-Richter a- and b- parameters. The bounds help
also to distinguish between triggered and induced events. It seems that the statistics of induced events
follows the lower bound of the magnitude probability. The statistics of triggered events is represented by
the reconstructed Gutenberg-Richter law. Our analysis is a phenomenological one. It attempts to explain
observable statistical features of induced seismicity from geometrical constraints only. The results of
the analysis are possibly applicable to different physics of the seismicity triggering like a triggering by
pore-pressure perturbations, stress perturbations, or triggering by friction alterations.

Introduction
Fluid-induced microearthquakes in hydrocarbon or geothermal reservoirs, aftershocks of tectonic earth-
quakes or seismic emission in rock samples are examples of seismicity resulting from a seismogenic
activation of finite volumes of rocks. Fluid-induced earthquakes of magnitudes 3 to 4 occurred at several
Enhanced Geothermal Systems (EGS) like those of Basel, Cooper Basin, The Geysers field and Soultz
(Giardini, 2009; Majer et al., 2007; Häring et al., 2008; Dyer et al., 2008; Baisch et al., 2009). It seems
that smaller but still perceptible events can be also observed by hydraulic fracturing of hydrocarbon
reservoirs (see http://www.cuadrillaresources.com/news/). Induced seismic hazard becomes a topic of
significance in the shale-gas industry (see http://www.energy.senate.gov/public/index.cfm /hearings-and-
business-meetings?ID=2c908340-a9bb-40b4-bf7f-8308b272893d). Its understanding is of a considerable
importance for mining of deep geothermic energy (Giardini, 2009; Majer et al., 2007; Cornet at al., 2007;
Häring et al., 2008). It is of significance for CO2 underground storage (see Zoback and Gorelick, 2012)
and possibly also for other types of geo-technological activity (see Avouac, 2012).
Similarly to the tectonic seismicity, statistics of the induced seismicity can be rather good described by
the Gutenberg-Richter frequency-magnitude distribution (Shapiro et al, 2007, 2010, 2011, Shapiro and
Dinske, 2009, Dinske and Shapiro, 2012). However, large-magnitude events deviate from this statistics

Triggered versus Induced Events and Bounds of the Magnitude Probability

112

(Shapiro et al, 2011). In this paper we theoretically and numerically analyze the influence of the finiteness
of a perturbed volume on the frequency-magnitude statistics of induced events. In contrast to exact
specific mechanical models of rupturing a given pressurized fault (see the recent detailed analysis by
Garagash and Germanovich, 2012) our analysis is a phenomenological one. It considers many faults
and attempts to relate geometry of the stimulated volume to observable statistical features of the induced
seismicity using rather general heuristic assumptions. It is possibly applicable to different types of the
seismicity triggering physics like a triggering by pore-pressure perturbations or stress perturbations, or
a triggering by rate-and-state processes modifying the friction. On the other hand we describe different
statistical scenarios of the triggering process. A tendency of real statistics of the induced seismicity to
follow one or other scenario may be indicative for the physics of event nucleation.
We start our analysis with a brief review of a statistical model of induced seismicity neglecting the fact that
rupture surfaces and stimulated volumes are finite. Then we shortly review data-based indications that the
large-magnitude events have different statistical features than small events. Then we propose lower and
upper bounds of the occurrence probability of given-magnitude induced events taking into account the
finiteness of the rupture surfaces and of the stimulated volumes. Our consideration is based on computing
the statistics of arbitrary-size arbitrary-oriented penny-shaped inclusions (representing potential rupture
planes) intersecting a finite stimulated rock volume. We show how the finiteness of rupture surfaces and
of the stimulated volume influences the frequency-magnitude relation of induced seismicity. We also
show how an estimate of an averaged stress drop can be obtained from the statistics of seismicity. Further
we discuss applications of our results to some borehole-injection based case histories. The bounds of the
magnitude probability will help to distinguish between triggered and induced events. The statistics of
induced events follows the lower bound of the magnitude probability. The statistics of triggered events is
represented by the Gutenberg-Richter law.

A model of point-like defects and observations

In the following we review a model of magnitude distributions of fluid-induced earthquakes based on
an assumption that their rupture surfaces are very small in respect to the size of the stimulated volume
(Shapiro and Dinske, 2009, Shapiro et al, 2011, Dinske and Shapiro, 2012). In spite of its simplicity and
schematic character this model is able to well explain large-number statistics of induced events. In the
next chapters our theoretical analysis will not be any more based on the fluid-induced mechanism of event
triggering. However, we will need this model as a reference point for understanding our observations.
We consider a point-like pressure source at the origin of an infinite poroelastic medium with a uniaxial
storage coefficient S. Qc (t) denotes a cumulative volume of the fluid injected during the an elapsed injec-
tion time t. Preexisting cracks (defects) are assumed to be randomly distributed with a bulk concentration
N . Each of these cracks is characterized by an individual critical value C of the pore pressure necessary
in accordance with the Coulomb failure criterion for occurrence of an earthquake along such a defect.
Cmax denotes a maximum value of the critical pore pressure characterizing the most stable defect in the
medium.
Previously we have derived a reflation between an event probability Wev and a pore pressure perturbation
during a monotonous non-decreasing fluid injection. We shown that the total event number Nev (t) is
given by:
N Qc (t)
Nev (t) ≈ . (1)
Cmax S
To compute an amount of events with magnitude larger than a given one we must introduce some assump-
tions about magnitude statistics. We assume further that the induced seismicity obeys the Gutenberg-
Richter statistics (see Shearer, 1999, p. 288). This means that the probability W≥M of events with the
magnitude larger than M is given by log10 W≥M = a − bM , where a and b are regional seismicity
constants. Thus, we obtain
log10 NM ≈ log10 Qc (t) + Σ − bM, (2)
where Σ = a − log10 (Cmax S) + log10 N . This constant is completely defined by seismotectonic features
of a given location. We call it a "seismogenic index". The larger this index is the larger is the probability
of events of significant magnitudes. It is difficult to theoretically calculate Σ because of some unknown

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 113

parameters (e.g., N ). However, Σ can be estimated using equation ( 2) and parameters of seismicity
induced by an injection experiment at a given location (see also Shapiro et al, 2007, 2010, 2011; Shapiro
and Dinske, 2009, Dinske and Shapiro, 2011).
Equation ( 2) can be further rewritten in a more conventional form of the Gutenberg-Richter law but with
the a value being time-dependent:

log10 NM (t) ≈ at (t) − bM, (3)

with
at (t) = log10 Qc (t) + Σ. (4)
Thus, the a-value depends on the seismogenic index and on the cumulative volume of the injected fluid.
Finally, one more useful reformulation of equation ( 2) can be proposed. Let us introduce a characteristic
magnitude MΣ such that:
Σ
MΣ = . (5)
b
Note that similarly to the seismogenic index the characteristic magnitude is also completely defined by
seismotectonic features of the injection site. The larger is the characteristic magnitude the larger is prob-
ability of significant induced events. Using the characteristic magnitude the modified Gutenberg-Richter
law ( 2) can be written in the following form:

log10 NMΣ ≈ log10 Qc (t). (6)

Recently Dinske and Shapiro (2012) observed this type of behavior in real data.
The statistical model of induced seismicity we summarized above can be equally well applied for non-
monotonous injections. In such a case the probability Wev is given by a minimum monotonous majorant
of the pore pressure (see Parotidis and Shapiro, 2004) and numerical computations are then required to
estimate log10 NM (t) for each particular situation.
Our practical experience show that during an active fluid injection with non-decreasing injection pressure
equation ( 2) can be applied to approximately describe a large number of induced earthquakes. However,
we observe systematic deviations from the equation ( 2) for large-magnitude events. Their number is
significantly smaller than expected, especially for short injection times (Shapiro et al, 2011).
Large-magnitude events correspond to large-scale ruptures. Such events are less common than small-
magnitude events. The statistics of potential rupture surfaces in rocks must correspond to a classical
Gutenberg-Richter distribution of earthquakes produced on them. For a rupture surface a probability
to intersect a stimulated volume of rocks will be the higher the larger the scale of this surface is. We
will accept the following evident (but still heuristic) assumption: the probability of an earthquake on the
corresponding rupture surface depends on the geometric relation between this surface and the stimulated
volume. We will consider then two possible scenarios.
In the first scenario we assume that to induce an event with a given rupture surface it is enough to stim-
ulate a small spot of this surface. Then, in the induced seismicity the portion of large-magnitude events
in respect to the portion of small-magnitude events should be higher than the portion of potential large-
scale rupture surfaces in the medium is in respect to the portion of small ones. Therefore, large-magnitude
events should be overrepresented in comparison to the expectations based on the Gutenberg-Richter statis-
tics. This scenario seems to contradict with the observations.
In the second scenario we assume that to induce an earthquake on a fault patch a significant part of
this patch must be stimulated. This is in agreement with the following formulation of the Coulomb
failure criterion: to enable an earthquake along a given interface an interface-integrated tangential stress
must overcome a total friction force. As soon as a largest part of a potential rupture surface remains
unperturbed, a probability of an earthquake remains low. Therefore, to enable an earthquake, a significant
part of the corresponding rupture surface should belong to a stimulated volume. This scenario seems to
agree with our observations.
We will try to quantify the both scenarios above and compare them with the frequency-magnitude statis-
tics of induced events. Thus, we must modify the probability W≥M from the previous section to take into
account the effect of the finiteness of the stimulated volume and of rupture surfaces.

Triggered versus Induced Events and Bounds of the Magnitude Probability

114

Note that the both scenarios are not directly related to the physics of the stimulation of potential rupture
surfaces. They just describe two different possible statistical patterns of the phenomenon. Therefore, they
may be applicable to the seismicity induced by elastic stress- or pore-pressure perturbations as well as to
the seismicity induced by other processes like rate- and state-dependent friction alterations. On the other
hand, a clear preference of the seismogenic process to follow one of these scenarios can provide us with
a useful seismo-tectonic information.

Statistics of potential rupture surfaces and seismic events

Let us assume that a finite volume of rocks V (the stimulated volume) has been somehow impacted so that
seismogenic conditions in it have been changed sufficiently to produce seismicity. In praxis we assume
that the stimulated volume is approximately defined by an outer envelope of a cloud of hypocenters of
induced seismicity. We consider the following simplified abstract model. A stimulated volume is an ellip-
soid or a rectangular cuboid (which can grow with time in the case of a fluid injection). Rupture surfaces
are randomly (or preferentially) oriented planar circular discs (penny shaped inclusions with vanishing
thickness). The spatial distribution of centers of the discs is random and statistically homogeneous with
the bulk concentration N .
We introduce firstly the probability Wf = fX (X)dX , which is is a probability of a given potential
rupture surface in the unlimited medium to have a diameter X. fX (X) is a PDF of a rupture surface of
the size X. For example, later, we will assume a power–law fX (X). Further, let Wc (X) to denote a
probability that the center of the rupture surface (of the diameter X) belongs to the stimulated volume
under the condition that this surface intersects the volume. Let us also for the moment assume that all
ruptures have the size X. It is clear then that the product of Wc (X) with the number of all rupture
surfaces intersecting the stimulated volume gives the number of rupture surfaces having their centers in
the volume. On the other hand this number is equal to N V . Thus, the ratio N V /Wc (X) is the number of
the all rupture surfaces intersecting the volume. Recalling now that the rupture surfaces are distributed on
their size we conclude that the ratio Wf (X)/Wc (X) gives the probability of a rupture surface intersecting
the stimulated volume to have the diameter X.
Now we can formulate a probability WE (X) of an induced seismic event with a rupture surface of diam-
eter X:
WE (X) = Ws (X)Wf (X)/Wc (X). (7)
Ws denotes a probability of this rupture surface to be sufficiently stimulated to produce the event. This
probability is a conditional one. It implies that the potential rupture surface has something to do with the
stimulated volume, i.e., at least intersects it (intersecting includes also touching). The presence of Wc in
equation ( 7) shows that there are much more rupture surfaces intersecting the stimulated volume than
just the product N V .
The quantity Wg (X) = Ws (X)/Wc (X) describes the influence of the geometry of the stimulated vol-
ume. Let us now concentrate on its part, the probability Ws (X). Shapiro et al, (2011) assumed that
seismic events will be possible only, if their rupture surfaces are located completely within the stimulated
volume. Consequently, they investigated the probability Wvol (X) that a disc of a diameter X is com-
pletely contained within a given stimulated volume under the condition that a center of this disc is located
within the stimulated volume. They have neglected a possibility of seismic events with rupture surfaces
intersecting the stimulated volume partially only. Here we will further develop their approach. Wg will
also take into account rupture surfaces intersecting the stimulated volume.
Let us consider all potential ruptures intersecting or located within the stimulated volume. We recall
that Wc (X) is a probability of such a rupture of diameter X to have its center within the stimulated
volume. Let us also define a probability of a seismic event along this whole rupture, We1 (X), under
the condition that the rupture center is located within the stimulated volume. Respectively, We2 (X)
will denote the event probability for such a rupture under the condition that the rupture center is outside
of the stimulated volume. We compute further the probability Ws (X) of a seismic event on a rupture
of diameter X. Such an event occurs on a rupture located completely within the stimulated volume.
Its probability is given by the product Wvol (X)Wc (X). Such an event can also occur on a rupture
with the center inside the volume but rather intersecting the volume only. Corresponding probability is
given by (1 − Wvol (X))Wc (X)We1 (X). Finally, such an event can occur on a rupture intersecting the

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 115

stimulated volume and having its center outside of the volume. Corresponding probability is given by
(1 − Wc (X))We2 (X). Therefore, the probability of a seismic event along a rupture of diameter X is
given by the sum:

Ws (X) = Wvol (X)Wc (X) + (1 − Wvol (X))Wc (X)We1 (X) + (1 − Wc (X))We2 (X). (8)

Note that under assumptions of Shapiro et al, (2011) , Wc (X) = 1 and We1 (X) = We2 (X) = 0, and
we obtain Ws (X) = Wvol (X). This corresponds to the lower bound of Ws under the condition that all
potential ruptures have their centers inside the stimulated volume. In a general case of accounting for
any potential rupture surfaces intersecting the stimulated volume (i.e., no any limitation for locations of
rupture centers), the lower bound of Ws will be given by an arbitrary Wc and We1 (X) = We2 (X) = 0:

Wsl (X) = Wc (X)Wvol (X). (9)

For the upper bound estimate several alternatives can be considered. The first and simplest one is
We1 (X) = We2 (X) = 1, and thus,
Wsu (X) = 1. (10)

This corresponds to the situation discussed in the previous section, where stimulation of an arbitrary small
spot of a potential rupture surface is enough for a corresponding seismic event. This would mean an over-
representation of the large-magnitude events in respect to the standard Gutenberg-Richter distribution. It
can be seen from equation ( 7), where Wc (X) becomes especially small for large X.
The next simple assumption would be We1 (X) = 1 and We2 (X) = 0. This assumption means that for
triggering an event, the centrum of its potential rupture surface must be within the stimulated volume.
Such a restriction is a reasonable formalization of the intuitive requirement that a "significant part" or a
"nucleation" spot of the rupture surface must be within the stimulated volume (note also a topological
equivalence between a disc centrum and any other "nucleation" centrum placed inside the rupture). It
leads to the following estimate
Wsu0 (X) = Wc (X). (11)

Corresponding to equation ( 7) this would mean that the statistics of induced events should be given by
Wf (X), i.e. given by a standard Gutenberg-Richter distribution.
Other estimates involving more assumptions on We1 and We2 are possible. However, we will restrict
our consideration to the three bounds ( 9)-( 11) as more natural ones. These three bounds represent three
different scenarios of the development of induced seismicity. Their comparison with real data will clearly
show which of the scenarios is more preferable for the induced seismogenesis.
The relation between the different estimates of Ws is:

Wsl < Wsu0 < Wsu . (12)

Correspondingly with the bounds of the probability Ws we obtain the bounds for the quantity Wg =
Ws /Wc :
Wgl < Wgu0 < Wgu , (13)

where
Wgl (X) = Wvol (X), (14)

Wgu0 (X) = 1, (15)

and, finally,
Wgu (X) = 1/Wc (X). (16)

In the following we summarize estimations of probabilities Wc (X) and Wvol (X).

Triggered versus Induced Events and Bounds of the Magnitude Probability

116

Probability of a given-size rupture surface in a stimulated volume

Shapiro et al, 2011 investigated the probability Wvol (X) that a disc of a diameter X is completely con-
tained within a given stimulated volume. They found an exact expression of the function Wvol (X) for a
spherical stimulated volume of the diameter L :
X 1 X X 3π X 3X X
Wvol (X) = Wsp ( ) ≡ (1 + ( )2 )(1 − ( )2 )1/2 − + arcsin( ). (17)
L 2 L L 4 L 2L L
This function (the subscript "sp" indicates a spherical stimulated volume) is quickly decreasing with
increasing XL . Shapiro et al., (2011) have numerically computed Wvol (X) for ellipsoidal volumes with
principal axes Lmin < Lint < Lmax . They found that
1
γ = [ (1/L3min + 1/L3int + 1/L3max)]−1/3 (18)
3
often provides a good estimate of a characteristic scale Y such that Wvol (Y ) becomes nearly vanishing
(<< 0.1). If Lmin is sufficiently small, then corresponding to equation ( 18) it will provide a dominant
contribution to γ. The function Wsp (X/γ) will give a good approximation, if the axes of the stimulated
ellipsoid are close to each other. Frequently, (especially in the case of a hydraulic fracturing) one of
the axis is extremely small: Lmin << Lint < Lmax . In this case further corrections are required to
approximate Wvol (X). In the appendix equation ( 45) defines a function Wel (X) providing such an
approximation (see also Figure 1).

geothermal case
0 hydrofrac case
10 0
10
Wel Wel
Wvol Wvol
data data

−1
10
−1
10
probability

probability

−2 Lmin = 10
10
Lmin = 150 −2
10 Linter = 50

Linter = 250 Lmax = 400

Lmax = 500
γ = 14.3842
−3
10 γ = 201.2062

−3
10
0 0.5 1 1.5 0 0.5 1 1.5 2 2.5 3
diameter / Lmin diameter / Lmin

Figure 1: A comparison of numerically computed (crosses) and theoretically estimated (lines) probabilities Wvol as
functions of disc diameters normalized to the minimum principal axes of the volume. To the left: for an ellipsoidal
Soultz93–like stimulated volume The line shows the result of equation ( 45). To the right - the same, but for a
hydraulic-fracture like ellipsoid. The parameters of the ellipsoids and resulting size γ are given on the plots

Until now we have considered chaotically oriented potential rupture surfaces. Let us assume that such
surfaces tend to be inclined by an angle ±φ (defined by the friction coefficient) to a plane of the maxi-
mum and intermediate tectonic stresses. Usually this angle is close to ±30◦ . Thus, the rupture surfaces
have a larger angle to the minimum stress axis. To simplify the consideration we assume that the all
potential rupture surfaces have these inclinations. Further, we will approximate the stimulated volume by
an rectangular cuboid with sides Lmin , Lint , and Lmax , rather than by an ellipsoid. Then, it is simple to
show that the sought-after probability is given by:
X X X
Wvol (X) = Wcub (X) ≡ (1 − | sin φ|)(1 − | cos φ|)(1 − ). (19)
Lmin Lmax Lint

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 117

For example, if φ = 0 all potential rupture surfaces will belong to the same plane. Such a geometry seems
to be less relevant for seismicity induced by fluid stimulations of rocks. However, it is more adequate for
aftershocks of earthquakes in subduction zones.

Probability of a given-size rupture surface to have its center within a stimulated volume
Let us firstly consider chaotically oriented rupture surfaces and a spherical stimulated volume of the
diameter L. The sought-after probability Wc (X) is given by the ratio of the total number of rupture
surfaces with the centers within the stimulated volume to the total number of all rupture surfaces having
any intersections with (or completely located within) this volume (see the PHASE Report, 2011, p. 73):

3 X2 3π X −1
Wc (X) = (1 + + ) . (20)
2 L2 4 L
We have numerically investigated this probability for ellipsoidal volumes with principal axes Lmin <
Lint < Lmax . Numerical results show that substituting into equation ( 20) instead of L the following
quantity
1 3/2 3/2
γc = [ (1/Lmin + 1/Lint + 1/L3/2 max )]
−2/3
(21)
3
often provides a good estimate of Wc . Note again, that if Lmin is sufficiently small, then corresponding
to equation ( 21) it will provide a dominant contribution to γc .
For rupture surfaces inclined by an angle ±φ to a plane of the maximum and intermediate tectonic stresses
we will approximate the stimulated volume by an rectangular cuboid with sides Lmin , Lint , and Lmax ,
rather than by an ellipsoid. Then, it is simple to show that the sought-after probability is given by:
  −1
X X X
Wc (X) = Wcc (X) ≡ (1 + | sin φ|) 1 + | cos φ| + . (22)
Lmin Lmax Lint

Also here, if φ = 0 then all potential rupture surfaces belong to the same plane. Equation ( 22) provides
then probability Wc for a stimulated area of a rectangular form. Such a geometry may be relevant for
aftershocks of tectonic earthquakes.

Magnitude-length scaling and frequency-magnitude distributions.

A spatial scale of the rupture surface controls a magnitude of the corresponding earthquake. A relation-
ship between a rupture size X and an earthquake magnitude M can be found using a standard relationship
for seismic moment magnitudes (Shearer, 1999; Lay and Wallace, 1995; Kanamori and Brodsky, 2004).
A seismic moment is given by a product of the shear modulus, G, of a slip displacement, D, and of an
involved rupture surface area, S. The moment magnitude is given by (Lay and Wallace, 1995; Shearer,
1999; Kanamori and Brodsky, 2004):

M = [log10 (GDS)]/1.5 − 6.07 = 2 log10 X + [log10 ∆σ − log10 C]/1.5 − 6.07 (23)

for a seismic moments measured in N m. In the last part of the equation we conventionally assume that
the slip displacement D scales as a characteristic length X of the slipping surface (see also equation 9.26
and Table 9.1 from Lay and Wallace, 1995). The quantity ∆σ is usually defined as a static stress drop,
and C is a geometric constant of the order of one. We will use a shorter form of equation ( 23):

M = 2 log10 (X/Cσ ), (24)

where we introduced the notation Cσ = 1084C 1/3 /∆σ 1/3 ≈ 103 ∆σ −1/3 . Equation ( 24) defines the
magnitude M as a function of two random variables, X and Cσ . It can be also written in the following
form:
X = Cσ 10M/2 . (25)
This equation defines the rupture length X as a function of two random variables, M and Cσ . There
are two transformation equations relating the pair of random variables (M ; Cσ ) to another pair, (X; Cσ ).

Triggered versus Induced Events and Bounds of the Magnitude Probability

118

The first relation, X(M, Cσ ), is given by equation ( 25). The second relation is just: Cσ = Cσ . These
relations along with their Jacobian (equal to ∂X(M, Cσ )/∂M ) yield the PDF of magnitudes, fM :

fM (M ) = 1, 151fX (Cσ 10M/2 )fC (Cσ )Cσ 10M/2 , (26)

where we accepted ln 101/2 ≈ 1, 151 and introduced the following notations: fX (X) is a PDF of the
rupture length, fC (Cσ ) is a PDF of Cσ , and the random variables X and Cσ are assumed to be statistically
independent. Thus, a probability W≥M of events with the magnitude larger than an arbitrary M is equal
to: Z ∞Z ∞
W≥M = 1, 151 fX (Cσ 10M/2 )fC (Cσ )Cσ 10M/2 dM dCσ . (27)
0 M

Let us firstly assume that the following factorization is possible:

fX (Cσ 10M/2 ) = f1 (Cσ )f2 (10M/2 ). (28)

This will be the case if fX (X) is a power-law function. Then

Z ∞
W≥M = AC f2 (10M/2 )10M/2 dM (29)
M

with the proportionality coefficient

Z ∞
AC = 1, 151 f1 (Cσ )fC (Cσ )Cσ dCσ . (30)
0

Thus, under the factorizing assumption for fX , the randomness of the stress drop influences the distribu-
tion of magnitudes by modifying its proportionality factor ( 30) only. This is the case for the Gutenberg-
Richter distribution.
Indeed, let us assume a power–law PDF of a size of potential rupture surfaces in an unlimited medium:
fX (X) ≈ AX X −q (here q > 0 and AX is a proportionality constant). Note that such a PDF cannot be
exactly valid because of an integration singularity at X = 0. We assume that a power-law function is
a good approximation of a real PDF of potential rupture surfaces above a certain very small size (which
corresponds to a magnitude significantly smaller than M0 ). Thus, a PDF fX (X) is strongly decreasing
with X. Power–law size distributions are typical for natural fractal–like sets (Scholz, 1990, Shapiro and
Fayzullin, 1992, Shapiro 1992). Such a type of self–similarity seems also to be a reason of the Gutenberg–
Richter frequency–magnitude distribution of earthquakes (Shearer, 1999, Turcotte et al., 2007; Kanamori
and Brodsky, 2004). Equations ( 24) and ( 28) yield the corresponding b-value of the resulting frequency–
magnitude distribution in the unlimited medium:

b = (q − 1)/2. (31)

Indeed, a power-law PDF fX (X) with q = 2b + 1 gives

f1 (Cσ ) = AX Cσ−2b−1 , (32)

and
f2 (10M/2 ) = 10−bM 10−M/2 . (33)
Then equation ( 29) provides:
W≥M = 10a−bM , (34)
where  ∞ 
AX
Z
a = log10 fC (Cσ )Cσ−2b dCσ . (35)
2b 0

Therefore, a power–law size distribution of potential rupture surfaces leads to the Gutenberg-Richter
magnitude distribution in a rather general case of an arbitrary statistically distributed stress drop.

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 119

In order to take into account a finiteness of the stimulated volume we must include the quantity Wg (see
equation ( 7) and the consideration below this equation) as a factor under the integral in equation ( 27).
Taking also into account equation ( 28) and results ( 31)-( 35) we obtain:
Z ∞Z ∞
W≥M = 1, 151AX 10−bM Wg (Cσ 10M/2 )Cσ−2b fC (Cσ )dM dCσ . (36)
0 M

Note that the quantity Wg is usually a function of a ratio of a potential-rupture scale X and a charac-
teristic scale of the stimulated volume Y (e.g., Y = L, γ, Lmin etc.; see equations ( 17)-( 22)). Thus,
the dependence of Wg on Cσ can be eliminated by introducing a characteristic magnitude MY so that
Y = Cσ 10MY /2 . Using this the quantity Wg can be expressed (at least, approximately) as a function
Wgm (M − MY ), which is directly obtained from Wg by corresponding substitution of the argument.
Generally, the magnitude MY is an unknown quantity effectively representing the range of induced mag-
nitudes in the sense of the equivalence of equation ( 36) to the following equation:
Z ∞
ad
W≥M = 10 10−bM Wgm (M − MY )dM. (37)
M
R∞
with the proportionality coefficient 10ad = 1, 151AX 0 Cσ−2b fC (Cσ )dCσ = 2, 303b10a. If Cσ−2b fC (Cσ )
tends to a narrow, δ-function like distribution around a representative value Cσ , then in accordance with
equation ( 25) MY will be directly given by Y = Cσ 10MY /2 . In reality Cσ is restricted to a limited range
between approximately 1 and 1000. A fitting of equation ( 37) to a real frequency-magnitude distribution
of an induced seismicity yields estimates not only of the b-value but also of the characteristic magnitude
MY . Using its relation (equation ( 25)) to the scale Y one can estimate also a characteristic stress drop
from a representative value of Cσ . Note that in the case of a lower-bound probability Wgl , due to the
vanishing probability Wvol for X > Y , the characteristic magnitude MY is a limiting value for a largest
possible magnitude of an induced earthquake:

MY = log10 Y 2 + (log10 ∆σ − log10 C)/1.5 − 6.07. (38)

Equation ( 37) can be also represented in the following form:

W≥M = 10a−bM+∆(M−MY ) , (39)

where  Z ∞ 
−bm
∆(M − MY ) = log10 2, 303b 10 Wgm (m + M − MY )dm (40)
0
is a function correcting the magnitude distribution for the finiteness of the stimulated volume. This
function can be very roughly estimated in the following way. The exponential function under the integral
is a much more quicker decreasing function than Wgm (m + M − MY ). Thus, by the integration we
can very roughly assume that the last function is a constant equal to Wgm (M − MY ) and we obtain:
∆(M − MY ) ≈ log10 Wgm (M − MY ). It shows clearly that if M is significantly smaller than MY ,
this will lead to a magnitude distribution indistinguishable from the classical Gutenberg–Richter’s one
(because Wgm → 1). By M → MY the magnitude distribution will quickly drop down in the case
Wgm → 0 (corresponding to Wg = Wgl , see equation 14). Equation ( 39) can be used to further modify
equation ( 2):
log10 NM (t) = log10 Q(t) − bM + ∆(M − MY ) + Σ. (41)
Figure 2 shows theoretical cumulative frequency-magnitude curves (i.e., the quantities log10 NM as func-
tions of M ) for a given time elapsed since the injection start. In the Figure, the elapsed time has been
involved implicitly only. It defines corresponding geometrical sizes (Lmin , Lint , Lmax ) reached by
the growing cloud of the seismicity. It defines also a value of the Gutenberg-Richter quantity a(t) =
log10 Q(t) + Σ. We assigned to these quantities values of typical orders of magnitudes as well as b = 1.5.
Then we numerically computed different functions ∆(M − MY ) using equation ( 40). The functions
Wgm (m + M − MY ) were obtained using the substitution ( 25) into the three functions Wg (X) given
by equations ( 14)-( 16), respectively. To compute the lower bound of the quantity log10 NM for the case
of an ellipsoidal stimulated volume the function Wg (X) was substituted by the approximating function

Triggered versus Induced Events and Bounds of the Magnitude Probability

120

Wel (X) defined by equation ( 45). To compute the lower bound of the quantity log10 NM for the case
of a cuboidal stimulated volume the function Wg (X) was substituted by the function Wcub (X) defined
by equation ( 19). To compute the uppermost bound of log10 NM for the case of a cuboidal stimulated
volume the function Wg (X) was substituted by the 1/Wcc (X), which is reciprocal to the one given by
equation ( 22). Finally, to compute the uppermost bound of log10 NM for the case of an ellipsoidal stim-
ulated volume the function Wg (X) was substituted by the quantity 1/Wc (X), which is reciprocal to the
approximative function given by equation ( 20) and the quantity γc from equation ( 21). Two situations
are represented: a geothermal- and a hydraulic-fracturing types of stimulated volumes. The both parts
of the Figure contain five following curves (from the lowest to the uppermost ones): a lower bound for
an ellipsoidal stimulated volume, a lower bound for a cuboidal stimulated volume, a Gutenberg-Richter
straight line, an upper bound for a cuboidal stimulated volume, and finally, an upper bound for an ellip-
soidal stimulated volume.
L =150.00m L =250.00m L =400.00m L =10.00m L =50.00m L =400.00m
min int max min int max
8 8
Elllower bound Elllower bound
Ellupper bound Ellupper bound
7 7
Cublower bound Cublower bound
Cubupper bound Cubupper bound
6 Gutenberg−Richter 6 Gutenberg−Richter

5 5
log10 Nev

log10 Nev

4 4

3 3

2 2

1 1

0 0
−2 −1 0 1 2 3 4 5 −2 −1 0 1 2 3 4 5
magnitude magnitude

Figure 2: Theoretical frequency-magnitude curves: the lower bound the case of an ellipsoidal stimulated volume; the
lower bound for the case of a cuboidal stimulated volume; the Gutenberg-Richter distribution; the uppermost bound
for the case of a cuboidal stimulated volume; the uppermost bound for the case of an ellipsoidal stimulated volume.
To the left: a geothermal-type of a stimulated volume. To the right: a hydraulic-fracturing type of a stimulated
volume.

Note that the curves for ellipsoidal stimulated volumes are approximations only. In contrary, the curves
for cuboidal volumes are exact. However the equations for cuboidal stimulated volumes assume rup-
ture surfaces inclined under an angle φ to the plane of the maximum and intermediate axes only. To
demonstrate the influence of the angle we show such curves in Figure 3 for different values of φ. Finally,
Figure 4 shows an example how a sophisticated geometric form of the stimulated volume can influence
the frequency-magnitude distribution (a lower bound). Here a situation corresponding to two intersecting
ellipsoids has been numerically simulated.
A consideration of Figures 2 - 4 along with Figure 1 shows that if the seismicity statistics tends to the
lower bound then a fitting of the Gutenberg-Richter straight line will produce a systematically overesti-
mated b-value. This effect will be the stronger the smaller is the size of the stimulated volume. Especially
important is the Lmin scale. Thus, the effect will be especially strong for the hydraulic-fracturing type of
the geometry. This effect will be also strong for small time periods elapsed from an injection start. For
small injection times stimulated volumes are small. Thus, the effect will result in a decrease of b-value
estimates with injection times. This effect can be easily understood from equation ( 39). For the lower
bound the quantity ∆(M − MY ) is negative. Its absolute value is the larger the smaller MY is and there-
fore, the smaller is the size of the stimulated volume. By fitting the Gutenberg-Richter straight line this
quantity will directly contribute to the values of the parameters b and a. It will decrease a and increase b.

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 121

This effect will act in an opposite direction, if the event statistics follows the uppermost bound. It would
increase a and decrease b values.
In the following section we compare several observed frequency-magnitude distributions to theoretical
bounds. The differences between theoretical curves for ellipsoidal and cuboidal volumes are not signif-
icant. Moreover, the angle φ is not known. In spite of the fact that φ = 30◦ seems to be a reasonable
approximation, in reality the angle can be broadly distributed. Thus, we attempted to fit real data by
theoretical approximations for ellipsoidal volumes.

L =150.00m L =250.00m L =400.00m L =10.00m L =50.00m L =400.00m

min int max min int max
8 8
lower bounds lower bounds
upper bounds upper bounds
7 Φ = 0o 7 Φ = 0o
Φ = 5o Φ = 5o
o
Φ = 10 Φ = 10o
6 Φ = 15o 6 Φ = 15o
Φ = 20o Φ = 20o
Φ = 25o Φ = 25o
5 5
Φ = 30o Φ = 30o
Φ = 35o Φ = 35o
log10 Nev

Φ = 40o
log10 Nev Φ = 40o
4 4
Φ = 45o Φ = 45o
Gutenberg−Richter Gutenberg−Richter
3 3

2 2

1 1

0 0
−2 −1 0 1 2 3 4 5 −2 −1 0 1 2 3 4 5
magnitude magnitude

Figure 3: The same as Figure 2 but cuboidal stimulated volumes and different angles φ.

Figure 4: Theoretical frequency-magnitude curves (lower bounds) for a stimulated volume in a form of two inter-
secting ellipsoids.

Triggered versus Induced Events and Bounds of the Magnitude Probability

122

Discussion of case studies

We have compared equation ( 41) to frequency-magnitude distributions in several case studies. We con-
sidered two geothermal locations in crystalline rocks, Basel (Häring et al., 2008) and Soultz 1993 (Baria
et al., 1999). Further, we have included a Paradox-Valley data set obtained by an injection of a salt water
into deep carbonate rocks (Ake et al, 2005). Finally, we have also included three hydrocarbon locations: a
hydraulic-fracturing stage (A) in gas shales (Canada), a hydraulic fracturing stage (B) of a tight-gas reser-
voir at the Cotton Valley (Rutledge and Phillips, 2003), and a stage of an untypical hydraulic fracturing
in Barnett Shale (Maxwell et al., 2009).
Corresponding fitting results are shown in Figures 5- 10. In the first step we attempt to fit the real data
by a standard Gutenberg-Richter cumulative distribution. We call then the resulting straight line and
its parameter as apparent parameters of the Gutenberg-Richter distribution. We observe that the real
frequency-magnitude distributions are usually well restricted between the lower bound and the fitting
straight line corresponding to the apparent Gutenberg-Richter distribution. Note that this line is located
lower than the bound Wgu0 (see eq. 15) corresponding to the reconstructed Gutenberg-Richter distribu-
tion. Moreover, nearly all data sets show a tendency of the seismicity to be better represented just by the
lower bound. We attempt to reconstruct the parameters of the "real" Gutenberg-Richter cumulative dis-
tribution from fitting the data by corresponding theoretical curves of the lower bound. The all 6 data sets
allowed to re-estimate the Gutenberg-Richter quantities a and b. They also yield estimates of maximum
expected induced magnitudes and of the stress drops. The re-estimated b-values are systematically lower
than the parameters obtained by the apparent Gutenberg-Richter fit.

Soultz ‘93 data

lower bound
4 Gutenberg−Richter
GRLB

3.5

2.5
GR LB
log10 Nev

b = 1.35 b = 1.06
2 a = 2.47 a = 2.85

∆σ = 382.782 Pa
1.5 Mmax = 1.31

1 MC = −0.71

0.5

0
−2 −1.5 −1 −0.5 0 0.5 1 1.5
magnitude

Figure 5: Fitting the frequency-magnitude distribution of the seismicity induced by the Soultz 1993 injection.

We start with the data sets permitting rather simple interpretation. They are shown in Figures 5- 7.
We consider a fitting the frequency-magnitude distribution of the seismicity induced by the Soultz 1993
injection (Fig 5). The axes of the ellipsoid are 440m, 1400m, 1740m. The effective sphere scale γ was
approximately 700m. A conventional Gutenberg-Richter fitting yields a = 2.5, b = 1.4. The bound
Wgl yields: a = 2.9, b = 1.1, MY = 1.3, ∆σ ≈ 400P a. The complete data set is well described

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 123

by the lower-bound approximation. Thus the inducing of events seems to indeed require pore pressure
perturbation involving a nearly total rupture plain.
A similar tendency shows a data set from hydraulic fracturing of gas shales in Canada (Fig 6). The axes of
the ellipsoid are 22m, 40m, 480m. The effective sphere scale γ was approximately 33m. A conventional
Gutenberg-Richter fitting yields a = 0.9, b = 1.3. The bound Wgl yields: a = 1.9, b = 0.9, MY =
0.15, ∆σ ≈ 0.6M P a.

HRB stage A data

lower bound
Gutenberg−Richter
3 GRLB

2.5

2
GR LB
log10 Nev

b = 1.33 b = 0.89
a = 0.90 a = 1.89
1.5

∆σ = 569.88 kPa
Mmax = 0.15
1
MC = −1.51

0.5

0
−2.5 −2 −1.5 −1 −0.5 0 0.5
magnitude

Figure 6: Fitting the frequency-magnitude distribution of the seismicity induced by a hydraulic fracturing stage (A)
in a gas-shale deposit in Canada.

Fitting the frequency-magnitude distribution of the seismicity induced by hydraulic fracturing at one of
locations in Barnett Shale (Fig 7) seems to be also similar. A conventional Gutenberg-Richter fitting
yields a = −5.0, b = 2.7. The bound Wgl yields: a = −0.5, b = 1.3, MY = −1.8, ∆σ = 1P a.
In the all three above examples the lower bound seems to be well suited. We explain the fact that the
lower-bound curve somewhat underestimate the number of events in the intermediate- to high-magnitude
range by a too rough analytical approximation of the real rupture statistics. Also an influence of the
geometry, which is more complex than just an ellipsoid (see Figure 4) or a rather restricted angular
spectrum of the rupture orientations (see Figure 3) could also contribute to this effect.
Somewhat more sophisticated interpretation seems to be required by the data sets shown in Figures 8- 10.
Fitting the frequency-magnitude distribution of the seismicity induced by the Paradox-Valley injection
(Fig 8) yields the following Gutenberg-Richter parameters: a = 3.6, b = 0.83. Fitting of equation (
41) corresponding to the lower bound provides nearly unchanged Gutenberg-Richter parameters: a =
3.6, b = 0.75. In addition we estimate a characteristic stress drop, ∆σ = 0.01M P a and the maximum
magnitude defining the lower bound: MY = 3.8.
Again we observe that a dominant majority of events follows well the lower-bound approximation. How-
ever, two data points corresponding to high-magnitude events return backward to the classical Gutenberg-
Richter distribution. Thus apparently, the corresponding three large-magnitude events were triggered by
just a pore-pressure-related perturbing of nucleation spots on their rupture surfaces.
A very similar situation can be observed on the data set corresponding to a hydraulic fracturing of the
tight sand gas deposit of the Cotton Valley, Stage B (Fig 9). Fitting the Gutenberg-Richter frequency-

Triggered versus Induced Events and Bounds of the Magnitude Probability

124

Barnett Shale data

lower bound
3
Gutenberg−Richter
GRLB

2.5

GR LB
log10 Nev

b = 2.72 b = 1.31
1.5 a = −5.02 a = −0.48

∆σ = 1.427 Pa
Mmax = −1.84
1

MC = −2.73

0.5

0
−3.4 −3.2 −3 −2.8 −2.6 −2.4 −2.2 −2 −1.8
magnitude

Figure 7: Fitting the frequency-magnitude distribution of the seismicity induced by hydraulic fracturing at one of
locations in Barnett Shale.

magnitude distribution yields a = −0.9, b = 1.9 Fitting of equation ( 41) corresponding to the lower
bound provides the following update of the Gutenberg-Richter parameters: a = 1.1, b = 1.2. In addition
we estimate a characteristic stress drop, ∆σ = 0.016M P a and the maximum magnitude defining the
lower bound: MY = −0.9. It seems that the rupture surfaces of two big events were not completely
included into the stimulated volume. For their triggering an excitation of rather large nucleation domains
was sufficient (we conclude this from the fact that they are still below the Gutenberg-Richter distribu-
tion indicating that their rupture surface is possibly larger than just a nucleation spot expected by this
distribution).
One more example of a similar situation is given by the Basel data set (Fig 10). A conventional Gutenberg-
Richter fitting yields a = 4.3, b = 1.4. For the fitting of the lower-bound equation ( 41) we took the axes
of the stimulated ellipsoid being equal to 100m, 760m, 920m. The bound Wgl yields close results:
a = 4.3, b = 1.3. Additionally, ∆σ = 12.5M P a and the maximum magnitude defining the lower bound:
MY = 3.05. It seems that for the triggering of a majority of events in Basel stimulation of their nucleation
spots was sufficient.
A comparison of these two groups of case studies indicates a possibility to separate between triggered
and induced events. We define the induced events as those for which perturbing of their nearly complete
rupture surface is necessary. Then their statistic should follow the lower bound. The triggered events
should be then defined as those perturbing of nucleation spots of their rupture surface would be sufficient
for their occurrence. Their statistics should follow the reconstructed Gutenberg-Richter distribution. It
seems that some geothermal reservoirs include triggered events (e.g., the Basel case study). This is seldom
but also possible for hydraulic fracturing of hydrocarbon reservoirs. Note also that for an event triggering
a perturbation of an arbitrary element of its rupture surface is not sufficient. This would correspond to
the uppermost bound ( 16). This bound strictly contradicts with our observations. The data show that
triggering requires a perturbation of a significant part of the rupture surface necessarily including the
nucleation domain (which we model by the rupture center). It seems also that the induced events are
much more common than triggered ones.

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 125

Paradox Valley data

3.5 lower bound
Gutenberg−Richter
GRLB

2.5

GR LB
log10 Nev

2
b = 0.83 b = 0.75
a = 3.59 a = 3.60

1.5
∆σ = 10.05 kPa
Mmax = 3.82

1
MC = 0.50

0.5

0
−1 0 1 2 3 4 5
magnitude

Figure 8: Fitting the frequency-magnitude distribution of the seismicity induced by the Paradox-Valley injection.

Our approach is not restricted to fluid-induced seismicity. We hypothesize that it is applicable for any
type of seismicity induced in restricted rock volumes. Figure ( 11) shows how well the lowermost bound
describes the frequency-magnitude distribution of an aftershock series of the Antofagasta (North of Chile)
1995 earthquake (M8.0) registered during three months (Aug. 10 to Oct. 11, 1995, see for further details
and references Shapiro et al, 2003)). Note that all aftershocks here seem to be just induced events.

Conclusions
In this paper we analyze influence of the finiteness of rupture surfaces and of stimulated volumes on
the statistics of magnitudes of induced earthquakes. A power-law size distribution of potential rupture
surfaces leads to the Gutenberg-Richter magnitude distribution in a rather general case of an arbitrary
statistically distributed stress drop.
We consider both, rupture surfaces located completely within or intersecting only the stimulated vol-
ume. We propose lower and upper bounds of frequency-magnitude distributions of induced events taking
different types of geometric relations between a stimulated volume and a potential rupture surface into
account.
Observations show that by borehole fluid injections at geothermal and hydrocarbon reservoirs, the frequency-
magnitude statistics of induced events tends to be in agreement with the lower bound of the event prob-
ability. This indicates that a rupture of a fluid-induced earthquake seems to be mainly probable along a
potential rupture surface located nearly completely inside a stimulated rock volume.
Fitting the lower bound to the magnitude distributions can provide an estimate of a largest expected
induced magnitude and a characteristic stress drop, in addition to improved estimates of the Gutenberg-
Richter a- and b- parameters. Because the statistics of induced seismicity tends to the lower bound, a
direct fitting of the Gutenberg-Richter straight line to frequency-magnutude data will produce a system-
atically overestimated b-value. This effect will be the stronger the smaller is the size of the stimulated
volume. An overestimating of the b-value will be especially strong for the hydraulic-fracturing type of

Triggered versus Induced Events and Bounds of the Magnitude Probability

126

Cotton valley B data

lower bound
3
Gutenberg−Richter
GRLB

2.5

GR LB
log10 Nev

b = 1.89 b = 1.16
1.5 a = −0.86 a = 1.13

∆σ = 16.53 kPa
Mmax = −0.87
1

MC = −1.97

0.5

0
−2.5 −2 −1.5 −1 −0.5
magnitude

Figure 9: Fitting the frequency-magnitude distribution of the seismicity induced by the Cotton-Valley Stage B injec-
tion. The axes of the ellipsoid are 10m, 54m, 420m. The effective sphere scale γ was approximately 17m.

the geometry of stimulated volumes. This effect will be also strong for small time periods elapsed from
an injection start. For small injection times stimulated volumes are small. The effect results in a decrease
of b-value estimates with injection times.
The main geometric factor limiting the probability to induce a large-magnitude event is the minimum
principal axis of a fluid- stimulated rock volume. The controlling role of the minimum principal axis of
the stimulated volume seems to be supported by the observations.
Our results indicate a possibility to separate between triggered and induced events. We define the induced
events as those for which perturbing of their nearly complete rupture surface is necessary. Their statistics
follows the lower bound. The triggered events are the events, which occur due to a perturbing of nucle-
ation spots of their rupture surfaces only. Their statistics is given by the reconstructed Gutenberg-Richter
distribution. It seems that geothermal reservoirs include some triggered events. This is seldom but also
possible for hydraulic fracturing of hydrocarbon reservoirs. Note also that for an event triggering a pertur-
bation of an arbitrary element of its rupture surface is not sufficient. A triggering requires a perturbation
of a significant part of the rupture surface necessarily including the nucleation domain. It seems also that
the induced events are much more common than triggered ones.
Our approach is not restricted to fluid-induced seismicity. We hypothesize that it is applicable for any
type of seismicity induced in a restricted rock volume, e.g., aftershock series of tectonic events.

Acknowledgments

We are grateful for funding from the PHASE project and from the Federal Ministry for the Environment,
Nature Conservation and Nuclear Safety in the frame of the project MAGS. We are indebted to H. Kaieda,
U. Schanz, M. Häring, K. Mahrer, J. Rutledge, S. Maxwell, A. Gerardi, R. Baria A. Jupe, T. Urbancic, A.
Baig and A Wuestefeld who provided us a great assistance with accessing their microseismic data.

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 127

Basel data
lower bound
3.5 Gutenberg−Richter
GRLB

2.5

GR LB
log10 Nev

2 b = 1.44 b = 1.33
a = 4.28 a = 4.32

1.5
∆σ = 12.46 MPa
Mmax = 3.05

1
MC = 0.55

0.5

0
0 0.5 1 1.5 2 2.5 3 3.5
magnitude

Figure 10: Fitting the frequency-magnitude distribution of the seismicity induced by the Basel injection.

References
Avouac, J.-P. (2012). Earthquakes: Human-induced shaking. Nature Geoscience,
5(doi:10.1038/ngeo1609):763–764.
Baisch, S., Voros, R., Weidler, R., and Wyborn, D. (2009). Investigation of Fault Mechanisms during
Geothermal Reservoir Stimulation Experiments in the Cooper Basin, Australia. Bulletin of the Seismo-
logical Society of America, 99(1):148–158.

Cornet, F., Bérard, T., and Bourouis, S. (2007). How close to failure is a granite rock mass at a 5-km
depth? International Journal of Rock Mechanics and Mining Sciences, 44(1):47–66.

Dinske, C. and Shapiro, S. (2012). Seismotectonic state of reservoirs inferred from magnitude distribu-
tions of fluid-induced seismicity. J. Seismology, doi:10.1007/s10950-012-9292-9.

Dyer, B. C., Schanz, U., Ladner, F., Haring, M. O., and Spillman, T. (2008). Microseismic imaging of a
geothermal reservoir stimulation. The Leading Edge, 27(7):856–869.

Garagash, D. and Germanovich, L. (2012). Nucleation and arrest of dynamic slip on a pressurized fault.
Journal of Geophysical Research, 117(B10310):doi:10.1029/2012JB009209.

Giardini, D. (2009). Geothermal quake risks must be faced. Nature, 462(7275):848–849.

Häring, M., Schanz, U., Ladner, F., and Dyer, B. (2008). Characterisation of the Basel 1 enhanced
geothermal system. Geothermics, 37(5):469–495.
Kanamori, H. and Brodsky, E. E. (2004). The physics of earthquakes. Reports on Progress in Physics,
67(8):1429–1496.
Lay, T. and Wallace, T. (1995). Modern Global Seismology. Academic Press.

Triggered versus Induced Events and Bounds of the Magnitude Probability

128

CINCA data
3.5 lower bound
Gutenberg−Richter
GRLB

2.5

GR LB
log10 Nev

2
b = 0.49 b = 0.45
a = 4.06 a = 4.06
1.5

∆σ = 764.483 Pa
Mmax = 5.63
1

MC = 1.88

0.5

0
0 1 2 3 4 5 6
magnitude

Figure 11: The frequency-magnitude distribution of an aftershock series of the Antofagasta (North of Chile) 1995
earthquake (M8.0).

Majer, E. L., Baria, R., Stark, M., Oates, S., Bommere, J., Smith, B., and Asanuma, H. (2007). Induced
seismicity associated with enhanced geothermal systems. Geothermics, 36:185 – 222.

McGarr, A., Simpson, D., and Seeber, L. (2002). Case histories of induced and triggered seismicity.
In International Handbook of Earthquake and Engineering Seismology, Int. Geophys. Ser., vol. 81A,
edited by W. H. K. Lee et al., pages 647–665. Elsevier, New York.

Parotidis, M., Shapiro, S., and Rothert, E. (2004). Back front of seismicity induced after termination of
borehole fluid injection. Geophysical Research Letters, 31:doi:10.1029/2003GL018987.

Scholz, C. H. (1990). The Mechanics of Earthquakes and Faulting. Cambridge University Press.

Shapiro, S. (1992). Elastic waves scattering and radiation by fractal inhomogeneity of a medium. Geo-
physical Journal International, 110:591–600.

Shapiro, S. and Fayzullin, I. S. (1992). Fractal properties of fault systems by scattering of body seismic
waves. Tectonophysics, 202:177–181.

Shapiro, S., Krüger, O. S., Dinske, C., and Langenbruch, C. (2011). Magnitudes of induced earth-
quakes and geometric scales of fluid-stimulated rock volumes. Geophysics, 76:WC53–WC61,
doi:10.1190/GEO2010–0349.1.

Shapiro, S. A. and Dinske, C. (2009). Scaling of seismicity induced by nonlinear fluid-rock interaction.
J. Geophys. Res., 114(B9):B09307.

Shapiro, S. A., Dinske, C., and Kummerow, J. (2007). Probability of a given-magnitude earthquake
induced by a fluid injection. Geophys. Res. Lett., 34(22):L22314.

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 129

Shapiro, S. A., Dinske, C., Langenbruch, C., and Wenzel, F. (2010). Seismogenic index and magnitude
probability of earthquakes induced during reservoir fluid stimulations. The Leading Edge, 29(3):304–
309.
Shapiro, S. A., Patzig, R., Rothert, E., and Rindschwentner, J. (2003). Triggering of microseismicity
due to pore-pressure perturbation: Permeability related signatures of the phenomenon. PAGEOPH,
160:1051–1066.
Shearer, P. (1999). Introduction to seismology. Cambridge University Press, Cambridge.
Turcotte, D. L., Holliday, J. R., and Rundle, J. B. (2007). Bass, an alternative to etas. Geophys. Res. Lett.,
34(12):L12303.
Zoback, M. D. and Gorelick, S. M. (2012). Earthquake triggering and large-scale geologic storage of
carbon dioxide. PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED
STATES OF AMERICA, 109(DOI: 10.1073/pnas.1202473109):10164–10168.

Appendix A
Probability of an arbitrary oriented disc belonging to an ellipsoidal volume In contrast to the exact
results ( 17), ( 19), ( 20) and ( 22) here we propose an approximation of the sought-after probability
Wel (X) (here X is the diameter of the rupture). We consider an ellipsoid with the principal axes Lmin <
Lint < Lmax . If the axes are close to each other, then a good approximation of the Wel (X) will be given
by Wsp (X/γ) where γ is given by equation ( 18). Let us consider another quite a realistic geometry of
stimulated volumes: Lmin << Lint < Lmax .
We consider further a rupture in a form of a plane disc of diameter X of an arbitrary orientation with a
center at a point P inside of such an ellipsoidal stimulated volume. We will concentrate firstly on large
discs with X > Lmin . The centers of large discs completely belonging to the ellipsoid are approximately
located inside the following ellipsoidal volume:
π
Vlarge = Lmin (Lint − X)(Lmax − X). (42)
6
We consider a sphere S1 defined by normals of the length X/2 (the sphere’s radius) at the point P for all
possible orientations of the disc. If X > Lmin (i.e., large discs) then such a sphere will always intersect
the surface of the ellipsoid. We estimate approximately a part of the surface of this sphere (in relation
to the sphereâĂŹs complete surface), where a normal can have its end-point under the condition that a
corresponding disc still belongs to the stimulated volume. For this we consider such a sphere intersecting
with a "side surfaces" of the volume. The side surfaces are two ellipsoid’s surface halves spanned on the
axes Lint and Lmax . Further, we approximate these surfaces just by planes (we call them side planes). Let
us then consider a sphere S1 of radius X/2 with the center at P at a minimum distance y from a side plane.
This sphere intersects this side plane along a circle. In order to belong to the volume a disc must have
a normal located inside of a cone with the symmetry axes coinciding with the normal from p P to the side
plane. The sphere S1 and this cone define a spherical segment of the height h = (X − X 2 − 4y 2 )/2
and the surface πXh. A probability of a disc to have an orientation necessary for belonging to the
stimulated
p volume is equal to the ratio of this surface to the surface of the half of the sphere S1, i.e.,
1 − 1 − 4y 2 /X 2 . Note that the effect of a possible intersecting of the sphere S1 with the second side
plane is taken automatically into account. Indeed, the largest intersection is of importance only because
of its symmetric effect on permitted orientations of the discs The probability of large discs inside of the
stimulated volume can be estimated then by the following integral over y:
Z Lmin /2 p (Lint − X)(Lmax − X)
Wlarge (X) = 2( (1 − 1 − 4y 2 /X 2 )dy) . (43)
0 Lmin Lint Lmax
The integration yields:
r
1 L2min X X X X
Wlarge (X) = (1 − 1− − arcsin )(1 − )(1 − ) (44)
2 X2 2Lmin Lmin Lint Lmax

Triggered versus Induced Events and Bounds of the Magnitude Probability

130

From the derivation it is clear that with increasing X ≤ Lint the estimate Wlarge (X) will adequately
decrease to zero. However, for small X the function Wlarge (X) becomes inadequate. In the point
X = Lmin it must be reasonably combined with the function Wsp (X/γ). Thus, we propose the following
approximation of Wel (X):

Wel (X) ≈ Wsp (X/γ), if X ≤ Lmin ,

Wsp (Lmin /γ)
Wel (X) ≈ Wlarge (X) , if Lmin ≤ X ≤ Lint . (45)
Wlarge (Lmin )

Triggered versus Induced Events and Bounds of the Magnitude Probability

Annual Report 2012 131

Numerical Simulations of Pore Pressure Diffusion in Hydraulically

Heterogeneous Media

J. Galindo Guerreros1 , C. Dinske1 , C. Langenbruch1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
In this work we analyse the influence of log-normal hydraulically heterogeneous fractal random media on
pore pressure diffusion. For this purpose, time dependent numerical simulations of fluid injections were
computed for a reservoir rock described as a two-dimensional fractal random media using Comsol, a finite
element solver. We obtain the distribution of pore pressure changes in the medium for different fractal
dimensions from Df =2.1 to 2.9 for simulation times up to 6 days. A general finding of this study is that
an increasing complexity of the heterogeneity with increasing fractal dimension does not significantly
influence the average distribution of the pore pressure perturbation. However, if we compare the pore
pressure profiles, we observe that the pore pressure changes in the fractal models are smaller than in
a homogeneous medium characterised by an effective diffusivity equal to the mean of the log-normal
distribution. The calculation of diffusion flux magnitudes for the hydraulically heterogeneous models
show interconnected high-permeable ‘channels’. We observe that the channel widths decrease and the
diffusive flux magnitudes increase with higher fractal dimension.

Introduction
The importance of estimating hydraulic transport properties, such as diffusivity and permeability, is
widely recognised in the area of geothermal and hydrocarbon reservoir characterisation. Especially, the
ability of predicting the fracture behaviour of the medium during fluid injection experiments which is
linked to seismicity plays a crucial role for the field production optimization as well as risk estimation.
If seismicity is triggered by the change in pore pressure, i.e. a critical point is exceeded, than the gener-
ation of new fractures or the reactivation of pre-existing fractures can be related to diffusion process of
relaxation of pore pressure (see Pearson, 1981). Predominantly, homogeneous models were used in our
working group to describe pressure diffusion during fluid injection experiments (see Shapiro et al., 1999).
Here, we address the question to which extent a heterogeneous model affects the pore pressure perturba-
tion. Furthermore, the comparison between heterogeneous and homogeneous models, characterised by an
effective diffusivity D, would give an answer to the question whether it is necessary to use heterogeneous
models or not.
It has been shown that the scaling behaviour of heterogeneities in the upper crust is of fractal nature (see
e.g. Dolan and Bean, 1997; Dolan et al., 1998). Therefore, we simulate the distribution of pore pressure in
a reservoir rock described by a two-dimensional heterogeneous fractal random distribution of diffusivities
with different fractal dimensions Df .

Heterogeneous Distribution of Diffusivity

The change of pore pressure p with time t in a heterogeneous medium can be described by the following
differential equation of diffusion:  
∂p ∂ ∂
= Dij p (1)
∂t ∂xi ∂xj

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

132

with Dij as the hydraulic diffusivity tensor with xi , xj and the radius vector r describing the distance
from injection point to an observation point in the medium.
Petrophysical properties in the Earth’s crust are observed more or less skewed characterised by low mean
values and high variances (Limpert et al., 2001). They often fit the log-normal distribution as it is the
case for reservoir properties such as porosity and permeability. The probability density function of a
log-normal random variable is defined in this study by:
 
1 1
pdf (x) = √ exp − 2 (log(x) − µ)2 , (2)
x · σ 2π 2σ
with the standard deviation σ (or the variance σ 2 ), the mean µ and the random generated value x. Figure
1 shows the log-normal distributed values and their normal distribution in logarithmic space for σ = e1.4
and µ = e−2 , which were used to create the δ correlated medium.

Figure 1: Log-normal (top left) and corresponding normal distribution in logarithmic space (top right) of diffusivity
in a two-dimensional δ correlated medium (bottom) with σ = e1.4 and µ = e−2 .

Several analysis have proved (see Leary and Al-Kindy, 2002) that the physical properties of reservoirs,
such as porosity, permeability and hence diffusivity, show power-law scaling in the frequency domain,
i.e. considerable short-scale variety of properties. The presented power-spectral density plot in Figure 2
illustrates a gradual linear decay, which is typically described by a ‘band-limited’ von Karman function
or fractal distribution.
Based on this, a two-dimensional ‘unbounded’ fractal scaling is simulated using the following relation,
which is later applied to our synthetic data:
K −β = P S(K), (3)
p
where K = x2 + y 2 is the wave number with its coordinates x and y, and β = E +2H is the power-law
scaling exponent of the spectrum related to the Euclidean dimension E and the Hurst exponent H. The
Hurst exponent is related to the fractal dimension Df = (E + 1) − H. In a two-dimensional medium
the Euclidean dimension E = 2 and the fractal dimension 2 ≤ Df ≤ 3. The greater Df the higher is the
complexity of the medium.
We create a hydraulically heterogeneous fractal distribution of diffusivity D by multiplying the spectrum
of the δ correlated medium of diffusivity (eq. 2) with the square root of the fractal scaling (eq. 3) and
taking the inverse Fourier transform of the product:
h √ i
D = F −1 F (f (x)) · K −β , (4)

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

Annual Report 2012 133

Figure 2: Power-spectral density of diffusivity fluctuations computed from borehole logging at the German Conti-
nental Deep Drilling Program (KTB).

with F denoting the Fourier transform. The resulting random media for three different fractal dimensions
Df =2.1, 2.5 and 2.9 are shown in Figure 3.

Figure 3: Figures showing from left to right the effect of the fractal dimensions Df =2.1, 2.5 and 2.9 on the
log(diffusivity) distribution. The higher Df the greater the complexity of the medium.

Pore Pressure diffusion

The generated heterogeneous fractal random media of diffusivity distribution are imported into Comsol
Multiphysics, a finite element solver program. In order to consider every generated point during the cal-
culation of the pore pressure diffusion a regular finite element mesh consisting of neighbouring triangles
with spatial resolution equal to the resolution of the random media is created separately. A comparison
between regular and irregular meshes is presented in Figure 4. The circular pressure perturbation source
with a constant amplitude of 10 Pa for all simulation times is implemented into the center of the model.
The hydraulic model dimension is 400 x 400 cells incorporating 800 x 800 mesh elements. Simulation
times run up to 518400 seconds what corresponds to a total of 6 days.

Results
We simulate six days of fluid injection with constant amplitude of 10 Pa injection pressure to obtain the
pore pressure perturbations for the homogeneous model with an effective diffusivity D equal to the mean
of the log-normal distribution illustrated in Figure 5 (left). Furthermore, we present in Figures 6 - 8 the
pore pressure changes (left), difference plots (mid) and the diffusive flux magnitudes (right) for the three
heterogeneous models with fractal dimensions Df =2.1, 2.5 and 2.9 after three (top) and six (bottom) days
of injection.

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

134

Figure 4: Zoomed selection of a regular (left) and an automatically generated irregular mesh (right). The model
dimension is 400 x 400 cells.

The modelling results show that the diffusion of pore pressure strongly depends on the fractal dimension
Df . As illustrated in Figure 3, an increase of the fractal dimension leads to an increase of the range of
diffusivities. This increase therefore allows for a faster propagation of pore pressure. Thus, we observe a
broader distribution of high pore pressures at fractal dimension Df =2.9 than in the other models. If we
compare pore pressure profiles for hydraulically heterogeneous and homogeneous models for all times
then we observe that the pressure is decreased up to 20 % (see Figure 9).
The middle column in Figures 6 - 8 shows difference plots where the pore pressure perturbation of the
heterogeneous model is subtracted from that of the homogeneous model. The larger differences are
predominantly located in the left and upper part of the model due to the preferred flow direction to the
bottom right corner. Comparing the differences, we observe that the pore pressure changes in the model
with low fractal dimension (Df =2.1) is decreased up to 30 %. With increasing fractal dimension, the
difference plots indicate also regions of increased pore pressure changes in the right lower part of the
model while the absolute maximum of the differences decreases to about 10 %.

Figure 5: Pore pressure perturbation (left) and diffusive flux magnitudes (right) after three (top) and six (bottom)
days of injection with a constant amplitude of 10 Pa in a homogeneous model.

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

Annual Report 2012 135

Figure 6: Pore pressure perturbation (left), difference plots (middle) and diffusive flux magnitudes (right) after three
(top) and 6 (bottom) days of injection for a heterogeneous diffusivity model with fractal dimension Df =2.1.

Figure 7: Pore pressure perturbation (left), difference plots (middle) and diffusive flux magnitudes (right) after three
(top) and 6 (bottom) days of injection for a heterogeneous diffusivity model with fractal dimension Df =2.5.

The right column in Figures 6 - 8 illustrates the diffusion flux magnitudes given in mP2a·s . It describes
the amount of pressure transport in space during a certain time interval from regions of high to low pore
pressure perturbation. As it can be seen, the diffusion flux magnitude for the model with Df =2.9 is ap-
proximately four times higher than for the homogeneous model. In general, the diffusive flux magnitudes
increase with the fractal dimension Df and interconnected permeable ‘channels’ are created. In compar-
ison to the heterogeneous models, such channels are not generated in the homogeneous model, where we
observe radial flow only (see Figure 5 right). Furthermore, we observe in the heterogeneous models, that

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

136

Figure 8: Pore pressure perturbation (left), difference plots (middle) and diffusive flux magnitudes (right) after three
(top) and 6 (bottom) days of injection for a heterogeneous diffusivity model with fractal dimension Df =2.9.

Figure 9: Pore pressure profiles along x-direction through the center of the model for the homogeneous medium (top
left) and the three heterogeneous media with Df =2.1, 2.5 and 2.9 (top right, bottom left and bottom right) after 7
days of injection with 10 Pa. The legend depicts the considered time in seconds.

the channel widths become thinner with increasing fractal dimension. This explains that the magnitudes
of the diffusive flux are higher for models with higher fractal dimensions. With ongoing injection time

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

Annual Report 2012 137

the diffusive flux magnitudes decrease which is in accordance with Fick’s first law of diffusion.

Conclusion
The presented study shows that modelling with log-normal hydraulically heterogeneous fractal media
influences the distribution of pore pressure changes caused by fluid injections. In our case, the pore
pressure is reduced up to 20 % as it was indicated by the difference plots. Furthermore, the effect of
varying fractal dimensions is striking regarding the diffusion flux magnitudes which increase with higher
fractal dimension. We observe the creation of interconnected permeable channels whose widths decrease
with increasing fractal dimension which leads to a higher transport of the pressure in the medium.
In order to provide more realistic results this study will be extended in future to a three-dimensional
model. We plan to generate synthetic seismicity and to compare them with real data. Consequently, this
study can become more valuable for application in hydrocarbon and geothermal reservoirs.

Acknowledgments
We thank the sponsors of the PHASE consortium for supporting the research presented in this paper.

References
Dolan, S. and Bean, C. J. (1997). Some remarks on the estimation of fractal scaling parameters from
borehole wire-line logs. GLR, 24.
Dolan, S., Bean, C. J., and Riolett, B. (1998). The broad-band fractal nature of heterogeneity in the upper
crust from petrophysical logs. GJI, 132.
Leary, P. and Al-Kindy, F. (2002). Power-law scaling of spatially correlated porosity and
log(permeability) sequences from north-central North Sea Brae oilfield well core. Geophys J Int,
148:426–442.
Limpert, E., Stahel, W. A., and Abbt, M. (2001). Log-normal distributions across the sciences: Keys and
clues. BioScience, 51(5):pp. 341–352.
Pearson, C. (1981). The relationship between microseismicity and high pore pressures during hydraulic
stimulation experiments in low permeability granitic rocks. J Geophys Res, 86(B9):7855–7864.
Shapiro, S. A., Audigane, P., and Royer, J.-J. (1999). Large-scale in situ permeability tensor of rocks
from induced microseismicity. Geophys J Int, 137:207–213.

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

138

Numerical Simulations of Pore Pressure Diffusion in Hydraulically Heterogeneous Media

Annual Report 2012 139

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure

Changes Induced by Reservoir Stimulations

C. Dinske1 , C. Langenbruch1 , J. Galindo Guerreros1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary

We interpret borehole logging measurements from the German Continental Deep Drilling Program (KTB)
to construct a hydraulic diffusivity profile for the central part of the main borehole. The profile indicates
strong fluctuations of diffusivity ranging over three orders of magnitude. We find that the variations can
statistically be described by a log-normal distribution and by a power-law scaling in the corresponding
power spectrum. We generate 2D fractal random media as representations of hydraulically heteroge-
neous rock which obey the statistical properties of the derived diffusivity log. We numerically simulate
the diffusion of pore pressure perturbations caused by a fluid injection in such a heterogeneous rock.
We compare the result of the modelling with a model where a homogeneous distribution of an effective
hydraulic diffusivity (equal to the mean of the log-normal distribution of diffusivity variations) was ap-
plied. We find that the magnitude of pore pressure changes is affected, in particular in the vicinity of the
injection borehole where regions of both strong enhancement and reduction exist. Another feature is the
development of flow paths along which the injection pressure is transported. We conclude that these flow
paths reflect higher permeable fractures embedded in the low porosity rock matrix.

Introduction

The interpretation of borehole logging measurements provides information on the composition of the
rock and insights into its petrophysical properties. Nowadays, log interpretation is applied on a broad ba-
sis. Particularly, it has become an essential tool for the characterisation and development of hydrocarbon
reservoirs (e.g. Ellis, 1987). Based on empirical and theoretical relations, important material properties
are derived from borehole logs, such as the porosity and the permeability of the rock surrounding the
borehole. In general, logging interpretations show that measured as well as derived parameters strongly
fluctuate along the borehole indicating the heterogeneity of these parameters in the Earth crust (Leary and
Al-Kindy, 2002).
In this paper, we will analyse the influence of hydraulic heterogeneity on the pore pressure which is per-
turbed by fluid injections in hydrocarbon and geothermal reservoirs. It is well known that these perturba-
tions play an important role in the re-activation of fracture systems (Pearson, 1981) and we can therefore
study the significance of heterogeneously distributed hydraulic diffusivity for the triggering of seismicity.
This relates to the question whether the assumption of a homogeneous distribution of hydraulic diffusivity
in the SBRC (Seismicity Based Reservoir Characterisation) approach (e.g. Shapiro et al., 1999) is almost
perfectly adequate to describe the medium as an effective medium. For our study, we will use borehole
logs from the German Continental Deep Drilling Program (KTB) to derive a hydraulic diffusivity pro-
file. Based on the statistical characteristics of this profile we will create random media of uncorrelated
heterogeneously distributed diffusivity. It has been observed that a fractal scaling is a common feature of
properties in nature (Bak et al., 1987). For this reason we will accordingly spatially correlate the random
media. We will then numerically simulate pore pressure perturbations which would be caused by fluid
injections in such heterogenous media.

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
140

Computation of hydraulic diffusivity profile

For the analysis of pressure diffusion in heterogenous media we define in a first step the field of hydraulic
diffusivity D which will be assigned to the medium. Following the works by Peschnig et al. (1997) and
Pape et al. (1999) we interpret borehole logging measurements to determine the porosity - permeability
structure. The hydraulic diffusivity is then computed using the permeability - diffusivity relation given by
Rudnicki (1986). The data are from the KTB scientific drill site in southeast Germany where extensive
logging measurements were carried out. The borehole logs are freely available from the ICDP (Interna-
tional Continental Drilling Program) web page. We focus on the depth section between 4400 m to 6100 m
of the main hole. This has the reason that during a fluid injection in 2000 most of the seismicity was in-
duced in a depth from 5 km to 6 km (Baisch et al., 2002). We preprocessed the logging data by applying
a moving window averaging filter to eliminate influences of the logging tools on the measurements.

Figure 1: Measured (black) and computed (red) borehole logs at KTB main hole. From left to right: resistivity
laterolog deep, resistivity laterolog shallow, tube wave inverse velocity, porosity and permeability.

Porosity-permeability structure from borehole logging

With the aim to derive a permeability log we apply a two-stage process. The first stage provides a
continuous porosity profile of the drilled rock. This is achieved by applying a statistical method proposed
by Peschnig et al. (1997). They performed a multivariate regression analysis to construct porosity profiles
for the complete pilot hole (0 - 4000 m) and for the central part of the main hole (3000 m - 6000 m). For
the amphibolite section which represents most of the depth interval under consideration, the following
statistically significant log combination was found to calculate the porosity φ (Peschnig et al., 1997):

log(φ) = F0 + F1 log (LLD) + F2 log(LLS) + F3 log(DT ST ). (1)

LLD and LLS are laterolog resistivity (deep, shallow), DT ST is inverse velocity of the Stoneley tube
wave (Figure 1). The derived factors F0 to F3 are: F0 = −19.668, F1 = −0.914, F2 = 0.629, and
F3 = 7.327 (Peschnig et al., 1997). The resulting porosity-depth profile is shown in Figure 1. We see
that the porosity varies along the whole profile with maximum values in the order of 1.5 %.

From the porosity log we compute the permeability of the rock along the borehole. We use the method by
Pape et al. (1999) who derived a porosity - permeability relationship based on the assumption of a fractal

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
Annual Report 2012 141

pore-space geometry. They extended the classical Kozeny-Carman equation to a fractal model of porous
rock and established a power-law relation between permeability κ and porosity φ. For magmatic and
metamorphic rock where pore-space is mainly defined by microfissures and microcracks, the following
relation was proposed (Pape et al., 1999):

κ = Aφ + Bφe1 + C(10φ)e2 , (2)

where the exponents are given by the Archie exponent m and the fractal dimension of pore-space structure
Dφ , i.e. e1 = m and e2 = m + 2/(c1 (3 − Dφ )). Pape et al. (1999) suggested c1 = 1 and determined
m = 1.25 and Dφ = 2.24. The coefficients A, B and C were estimated empirically using calibration
measurements of porosity and permeability of core samples which yielded A = 0, B = 234 and C =
2094. Using this porosity-permeability relationship we calculate the permeability profile for the depth
interval 4400 m - 6100 m from the porosity. Figure 1 illustrates the result which reveals that permeability
varies over three orders of magnitude with a mean value of about 10−18 m2 . The result is in a good
agreement with permeability measurements taken on core samples in laboratory. Nover et al. (1995)
investigated several amphibolites (located between 4200 m and 5000 m) and determined permeabilities
ranging from 10−20 m2 to 10−16 m2 .

Figure 2: Measured (black) and computed (red) borehole logs at KTB main hole. From left to right:rock density,
P-wave velocity, S-wave velocity, Lamé parameter λ, Lamé parameter µ, undrained Lamé parameter λu , Biot coef-
ficient, and hydraulic diffusivity.

Determination of hydraulic diffusivity

According to Rudnicki (1986) the hydraulic diffusivity D can be computed from the permeability κ:

κ (λu − λ)(λ + 2µ)

D= , (3)
η (λu + 2µ)α2
where η is the viscosity of the fluid which fills the pore space. Equation 3 also includes the Lamé parame-
ters, λu , λ and µ, and the Biot parameter α, respectively. Here we assume that these (poro)elastic moduli
are heterogeneously distributed as well. Therefore we determine λu , λ, µ and α from the interpretation
of borehole logging measurements.
We compute the Lamé parameters µ and λ from the sonic velocities (vP and vS ) and rock density (ρ)
logs using the standard notations for shear wave and compressional wave velocities:

µ = vS2 ρ and λ = vP2 ρ − 2µ. (4)

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
142

mineral f Kgr [GP a]

amphibole 0.645 90
plagioclase 0.278 75
biotite 0.055 60
garnet 0.022 175

Table 1: Fractions f and grain bulk moduli Kgr of the rock-constituting minerals of KTB amphibolite used to
calculate the bulk modulus Ksolid . Data from Haimson and Chang (2002).

The borehole logs of measured density and sonic velocities as well as the two calculated elastic moduli
are shown in Figure 2. Next, we compute the poroelastic moduli α and λu according to:

K
α=1− . (5)
Ksolid
1
P fi
Ksolid is the bulk modulus of the rock-constituting material, i.e., Ksolid = i Kgri , with i mineral
fractions f and corresponding mineral bulk modulus Kgr (Table 1),
and

2
λu = Ku − µ, (6)
3
with (see e.g. Detournay and Cheng, 1993):

Ksolid (K + F ) Kf (Ksolid − K)
Ku = ,F = . (7)
Ksolid + F φ(Ksolid−Kf )

(a) (b)

Figure 3: Correlation plot of (a) zero-mean porosity (amplified by factor 150) and zero mean logarithms of perme-
ability and diffusivity and (b) zero-mean logarithms of porosity, permeability and diffusivity. Horizontal lines mark
two narrow zones of facies change (amphibolite −→ gneiss) (Peschnig et al., 1997). Data from the gneiss zones are
excluded from the following statistical analysis.

Kf is the bulk modulus of water and φ is the porosity. Figure 2 shows the resulting α and λu logs. Finally,
we compute the hydraulic diffusivity log according to Equation 3 which is presented in Figure 2. An ex-
amination of the calculated logs clearly indicates that rather than being uniform, porosity, permeability
and diffusivity are strongly fluctuating along the whole profile. We also observe that the fluctuations in

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
Annual Report 2012 143

porosity correlate with fluctuations in log(permeability) and log(diffusivity) (Figure 3). However, the
correlations become stronger if we consider also the porosity in logarithmic space. The correlation coef-
ficient between porosity and diffusivity then increases from 70 % to 88 %. Interestingly, this observation
differs from the conclusions of Leary and Al-Kindy (2002) who find a linear relation between porosity
and log(permeability). A possible explanation could be provided by the different rock formations under
consideration. Leary and Al-Kindy (2002) investigated high-porosity sandstones whereas in our study we
examine low-porosity crystalline basement rock.

Generation of random medium with fractal scaling

We use the computed hydraulic diffusivity profile to generate a random medium which obeys the sta-
tistical properties of the derived log. For this purpose, we first analyse the statistical parameters which
describe the fluctuations of fluid transport properties of the rock along the borehole. Afterwards, we build
spatially correlated media of hydraulic heterogeneity as input for the modelling of pore pressure.

(a) (b) (c)

Figure 4: Observed (black lines) and fitted (blue bars) probability density functions of (a) porosity, (b) permeabil-
ity and (c) hydraulic diffusivity fluctuations. Top figures show data in linear scale, bottom figures show data in
logarithmic scale.

Statistical analysis
We determine the probability density functions (PDF) of porosity, permeability and hydraulic diffusivity.
As illustrated in Figure 4 the variations of the three physical quantities are best described by a log-normal
distribution. Such a distribution is often observed in nature (e.g. Limpert et al., 2001) Transferring the
data to the logarithmic domain, they fit a Gaussian (normal) distribution (Figure 4). Mean value and
standard deviation of the normal distribution are then used to generate log-normally distributed random
numbers of hydraulic diffusivity. The random numbers are assigned to a 2D model consisting of 1000 x
1000 cells which results in a spatially uncorrelated random medium (Figure 6).

Estimation of fractal parameters Df and H

The generated 2D model would show a flat power spectral density because of the statistically uncorrelated
random numbers. However, it is well known that fluctuations of parameters in rock physics as well as in

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
144

Figure 5: Power spectra of the calculated borehole logs presented in Figures 1 and 2. The linear fits to the slopes
provide the power-law exponent β.

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
Annual Report 2012 145

other physical sciences obey a power-law scaling behaviour in either frequency or wavenumber domain
(see e.g. Bak et al., 1987). Such a scaling is often called 1/f (resp. 1/k) noise or pink noise and it
describes fractal structures. Figure 5 show power spectra of the computed borehole logs. The spectra
clearly indicate that all parameters fluctuate inversely with their spatial frequency. It means the power
spectral density follows a scaling according to k −β as function of the wavenumber k in the whole range,
i.e., from small to large structures. The power-law exponent β estimated by linear regression ranges from
1.03 (diffusivity) to 1.42 (permeability). β is related to the fractal dimension of the medium Df via the
Hurst exponent H and the Euclidean dimension E:
β = E + 2H and Df = (E + 1) − H. (8)
To take into account the observed spatial correlation, we apply a fractal scaling to the uncorrelated 2D
medium. For the hydraulic diffusivity log, the Hurst exponent is H = 0.016 resulting in a fractal di-
mension Df = 2.984. Since for a 2D medium the fractal dimension is bounded by 2 ≤ Df ≤ 3, the
obtained fractal dimension indicates that the heterogeneous medium has a high degree of complexity
dominated by rather small-scale structures. The spatial correlation is realised by filtering the 2D medium
in the wavenumber domain. The Fourier transform F of the simulated log-normally distributed hydraulic
diffusivity field is multiplied with the square root of 1/k power-law scaling:
√
D(k) = F(D(x, y)) k −β . (9)
The back transformation to the spatial domain then provides the spatially correlated fractal 2D medium
of heterogeneous distribution of hydraulic diffusivity (Figure 6).

(a) (b) (c)

Figure 6: (a) Probability density function of simulated hydraulic diffusivity variations. (b) 2D random medium
realisation of uncorrelated heterogeneous distribution of hydraulic diffusivity. (c) The same 2D random medium
realisation but spatially correlated with fractal dimension Df = 2.984. Top figures show data in linear scale, bottom
figures show data in logarithmic scale.

Numerical modelling
We now analyse the influence of hydraulic heterogeneity on the perturbation of pore pressure caused
by fluid injections. This is achieved by numerical modelling pore pressure diffusion with the Finite-

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
146

Element software COMSOL. The simulated random media are introduced as input parameter for solving
the diffusion equation. For a comparison, we also consider a medium characterised by a homogeneous
distribution of hydraulic diffusivity.

Model set-up

We model the pore pressure perturbation in a 2D medium with spatial extent of 1000 cells in both di-
rections. In the centre of the model a line source is implemented representing the open hole section of
a borehole. The source strength corresponds to a fluid injection pressure of 10 M P a. Simulations are
computed for a time period of 20 days with temporal increment of 2 days. To avoid spatial influences
and interpolations caused by an automatically generated irregular mesh, we implement a regular mesh
consisting of equally sized elements with the same spatial resolution as the simulated diffusivity model.
It means that the mesh consists of 1000 x 1000 grid nods. Pore pressure perturbations are modelled for
two different realisations of random media as well as for a homogeneous medium (Figure 7). In the latter,
the hydraulic diffusivity represents an effective hydraulic diffusivity of the heterogeneous media, i.e., it
is equal to the mean value of the PDF of normally distributed diffusivity in logarithmic space.

(a) (b) (c)

Figure 7: Result of numerical modelling of pore pressure perturbations in two different realisations of fractal random
media of heterogeneously distributed hydraulic diffusivity ((a) and (b)), and in a homogeneous model with diffusivity
equal to the mean value of log-normal distributed hydraulic diffusivity variations (c). Top figures show the hydraulic
diffusivity model, middle and bottom figures show simulated pore pressure changes after 10 days and 20 days of
injection, respectively. Note that colorbars in all figures are in logarithmic scale.

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
Annual Report 2012 147

Modelling results
Figure 7 illustrates the result of the modelling after 10 days and 20 days of injection. Although the typical
exponential-like decay of pressure perturbations away from the source is preserved in the random media,
we notice that the overall shape is disturbed. Comparison of pore pressure fields for heterogeneous and
homogeneous case reveals that also the magnitude of perturbations is affected, mainly in the vicinity of
the source. Depending on the random medium realisation, regions of both elevated and lowered levels of
pore pressure perturbations of the order of several Megapascals are visible (Figure 8). Another interesting
feature that we observe in the heterogeneous models is the development of flow paths along which the
injection pressure is transported (Figure 9). These distinct channels connect areas of high diffusivity in
the models which favour the diffusive flux. Since the hydraulic diffusivity is linked to the porosity, we
assume that the flow paths in the models reflect high-permeable structures, such as fissures and fractures,
embedded in lower porosity rock matrix.

(a) (b)

Figure 8: Differences in pore pressure changes between heterogeneous and homogeneous medium after 10 days (top)
and 20 days of injection (bottom). (a) and (b) correspond to the two different realisations of fractal random media
shown in Figures 7 (a) and 7 (b), respectively.

Conclusions
We derived a hydraulic diffusivity borehole profile for the central part of the KTB main hole based on
the interpretation of available logging measurements. We found that the diffusivity is characterised by
a high degree of heterogeneity as well as a high complexity of its spatial distribution. The fluctuations
of hydraulic diffusivity are log-normal distributed, with a mean value of D = 0.003 m2/s. The mean

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
148

value remarkably coincides with an estimate of diffusivity from induced seismicity at the KTB. Shapiro
et al. (2006) report a value of D = 0.004 m2/s obtained for the seismicity occurring in 5 − 6 km depth.
Furthermore, we found that log(diffusivity) well correlates with log(porosity) expressed in a high correla-
tion coefficient close to 90 %. The hydraulic diffusivity profile clearly shows a power-law scaling spatial
power spectral density with exponent close to one which reveals a high, unbounded fractal dimensionality.
We used the statistical properties of the diffusivity variations and the corresponding wavenumber power
spectrum to simulate fractal random media as representations of hydraulically heterogenous rock. We per-
formed numerical computations of pore pressure perturbations which would result from fluid injections
in such random media. The models exhibit two main features of the effect of hydraulic heterogeneity.
Around the injection borehole exist regions of strong enhancement as well as reduction in pore pressure
if compared to a hydraulically homogeneous medium. Second, the heterogeneity leads to a preferred
diffusive flux along flow paths defined by connected high-permeable areas in the random media which
can be interpreted as fracture zones.

(a) (b)

Figure 9: Flow paths indicated by the diffusive flux magnitude after 20 days of fluid injection. (a) and (b) correspond
to the two different realisations of fractal random media shown in Figures 7 (a) and 7 (b), respectively.

Outlook

The here presented analysis of hydraulic heterogeneity and its influence on the diffusion of pore pressure
perturbations was carried out in 2D random media. In the next step, we will extend our analysis to
consider pressure diffusion in 3D heterogeneous media. It allows to obtain more realistic models for
reservoir simulations. Additionally, we will use the modelled injection-induced pore pressure changes
to generate synthetic seismicity catalogs. This is achieved by application of the criticality approach
(Shapiro et al., 2005; Rothert and Shapiro, 2007). However, we understood that heterogeneity of elastic
parameters significantly influences the stress state of reservoir rock and consequently the magnitude and
distribution of critical pore pressures necessary to (re)activate fractures (Langenbruch and Shapiro, 2012).
Therefore synthetic seismicity will be generated based on the results of joined modelling in elastically
and hydraulically heterogeneous random media. Since borehole logging measurements from the KTB
site have been applied in both kind of models to create a fractal random medium, we can accordingly
use the fluid-induced seismicity observed during the 2000 injection (Baisch et al., 2002) to validate our
models.

Acknowledgments

We thank the sponsors of the PHASE consortium for supporting the research presented in this paper.

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
Annual Report 2012 149

References
Baisch, S., Bohnhoff, M., Ceranna, L., Tu, Y., and Harjes, H.-P. (2002). Probing the crust to 9 km depth:
Fluid injection experiments and induced seismicity at the KTB superdeep drilling hole, Germany. Bul
Seism Soc of America, 92(6):2369–2380.
Bak, P., Tang, C., and Wiesenfeld, K. (1987). Self-organized criticality - An explanation of 1/f noise.
Phys Rev Lett, 59:381–384.
Detournay, E. and Cheng, A. H.-D. (1993). Comprehensive Rock Engineering: Principles, Practice and
Projects, volume II: Analysis and Design Method, chapter 5: Fundamentals of Poroelasticity, pages
113–171. Pergamon Press.
Ellis, D. (1987). Well logging for Earth Scientists. Elsevier New York.
Haimson, B. C. and Chang, C. (2002). True triaxial strength of the KTB amphibolite under bore-
hole wall conditions and its use to estimate the maximium horizontal in situ stress. J Geophys Res,
107(B10):doi:10.1029/2001JB000647.
Langenbruch, C. and Shapiro, S. A. (2012). Influence of elastic heterogeneity on fracture strength distri-
bution in rocks. In SEG Technical Program Expanded Abstracts 2012, pages doi:10.1190/segam2012–
0179.1, Las Vegas.

Leary, P. and Al-Kindy, F. (2002). Power-law scaling of spatially correlated porosity and
log(permeability) sequences from north-central North Sea Brae oilfield well core. Geophys J Int,
148:426–442.
Limpert, E., Stahel, W., and Abbt, M. (2001). Log-normal distributions across the sciences: Keys and
clues. BioScience, 51(5):341–352.

Nover, G., Heikamp, S., Kontny, A., and Duba, A. (1995). The effect of pressure on the electrical
conductivity on KTB rocks. Surveys in Geophysics, 16:63–81.
Pape, H., Clauser, C., and Iffland, J. (1999). Permeability predictions based on fractal pore-space geom-
etry. Geophysics, 64(5):1447–1460.
Pearson, C. (1981). The relationship between microseismicity and high pore pressures during hydraulic
stimulation experiments in low permeability granitic rocks. J Geophys Res, 86(B9):7855–7864.

Peschnig, R., Haverkamp, S., Wohlenberg, J., Zimmermann, G., and Burkhardt, H. (1997). Integrated
log interpretation in the German Continental Deep Drilling Program: Lithology, porosity and fracture
zones. J Geophys Res, 102(B8):18.363–18.390.
Rothert, E. and Shapiro, S. A. (2007). Statistics of fracture strength and fluid-induced microseismicity. J
Geophys Res, 112:B04309, doi:10.1029/2005JB003959.
Rudnicki, J. W. (1986). Fluid mass sources and point forces in linear elastic diffusive solids. Mechanics
of Materials, 5:383–393.
Shapiro, S. A., Audigane, P., and Royer, J.-J. (1999). Large-scale in situ permeability tensor of rocks
from induced microseismicity. Geophys J Int, 137:207–213.
Shapiro, S. A., Kummerow, J., Dinske, C., Asch, G., Rothert, E., Erzinger, J., Kümpel, H.-J., and Kind,
R. (2006). Fluid induced seismicity guided by a continental fault: Injection experiment of 2004/2005
at the German Deep Drilling site (KTB). Geophys Res Lett, 33:doi:10.1029/2005GL024659.
Shapiro, S. A., Rentsch, S., and Rothert, E. (2005). Characterization of hydraulic properties of rocks
using probability of fluid-induced microearthquakes. Geophysics, 70:F27–F34.

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
150

Influence of Hydraulic Heterogeneity of Rocks on Pore Pressure Changes Induced by Reservoir Stimulations
Annual Report 2012 151

Permeability as a Function of Stiff and Compliant Components of

Pores and Cracks and its Stress Dependence

S. A. Shapiro1
email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
We compare the behaviour of permeability and elastic properties of rocks as functions of the tectonic
stress and pore pressure. We propose that different parts of the pore and fracture space differently con-
tribute to a control of these two types of physical properties of rocks. The stress dependence of elastic
moduli of drained rocks is mainly controlled by deformations of the compliant part of the pore (including
fractures) space. This leads to saturating exponential functions expressing load dependences of seismic
velocities and of related seismic parameters. On the other hand a temporal evolution of seismicity induced
by fluid stimulations (including hydraulic fracturing) of rocks sometimes implies power-law dependences
of the permeability on the pore pressure. It seems that in some rocks permeability is controlled by the stiff
part of the pore space. This leads to a power-law pressure dependence of the permeability. Sometimes
permeability can be controlled by compliant pores (e.g., cracks). Then exponential type of permeability
as function of pressure can be observed. Below we propose a permeability model taking these two parts
of the porosity into account.

Introduction
Understanding of stress dependences of physical properties of rocks is important for different applica-
tions like the overpressure prediction and the seismic monitoring of hydrocarbon reservoirs. The stress
dependence of elastic moduli of drained rocks is mainly controlled by deformations of the compliant pore
space (thin cracks and vicinities of grain contacts; see e.g., Zimmerman et al, 1986; and Shapiro, 2003).
In this paper we are interested in a permeability behaviour by fluid reservoir stimulations (e.g., hydraulic
fracturing). In such situations the main changing loading component is frequently the pore pressure Pp .
One could use a pore-pressure porosity to find a pore-pressure dependent permeability. However, usually
such relations treat the porosity in a very general way. They do not specify which porosity component
controls the permeability predominantly. . On the other hand, roles of the compliant and stiff parts of
the pore space can be very different in the pressure dependence of the permeability. Below we propose
to take these two parts of the porosity into account. For this we firstly briefly formulate the deformation
of the pore space. Then we introduce the stiff and compliant parts of the pore space and consider their
strains. We conclude then with a comparative discussion of the stress dependences of elastic moduli and
of the permeability.

Deformation of the pore and fracture space

In order to formulate a stress-dependent permeability we describe a strain of the pore and fracture space
of a rock sample. For this we summarize some necessary definitions and results of Shapiro and Kaselow
(2005). We consider a thought experiment where a uniform confining stress, σij , (also called total stress)
is applied to the external surface Σ of a rock sample. Simultaneously, a pore pressure, Pp , is applied to
the internal surface Ψ. Surface Σ is the external surface of the solid part of the rock. Simultaneously it

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its Stress Dependence
152

seals also pores. In a given point x̂ of the surface Σ the applied traction τ is then

τi = σij nj (x̂), (1)

where nj (x̂) are components of the outward normal of Σ.

0
Let us assume that the confining stress and/or pore pressure has changed from the load state (σij , Pp0 )
to the load state (σij , Pp ). As a result, points of the external surface has been displaced on ui (x̂). The
displacement is assumed to be very small in comparison with the size of the rock sample.
Following Brown and Korringa (1975) we introduce a symmetric tensor
1
Z
ηij = (ui nj + uj ni )d2 x̂. (2)
Σ 2

In the case of a continuous elastic body replacing the porous rock (i.e., a differentiable displacement is
given at all its points) the Gauss’ theorem gives
1 ∂ui ∂uj 3
Z
ηij = ( + )d x̂. (3)
V 2 ∂xj ∂xi

The integrand here is the strain tensor and V is the volume of the sample. Thus, ǫij = ηij /V is the
volume averaged strain.
We introduce also another symmetric tensor
1
Z
ζij = (ui nj + uj ni )d2 x̂, (4)
Ψ 2

where Ψ is the surface of the pore space, x̂ is a point of this surface, ui is a component of the displacement
of points x̂ of this surface slightly deformed by changing the load, and ni is a component of the outward
normal to this surface (the normal is directed into the space of pores). In points, where the surface Σ seals
the pores it coincides with the surface Ψ. However, their normals are opposite.
If we assume that the pore space is completed by some material (e.g., a fluid or a clay or a cement)
then analogously with the tensor ηij quantity −ζij will denote a volume averaged strain of this material
multiplied by the volume of this material. Clear, that quantity −ζii denotes a change of the volume of
this material.
If the load on surface Ψ is hydrostatic, then the effective stress will be defined as:
e
σij = σij + Pp δij . (5)

Let us assume that in an initial stress state the location and the form of the internal surface Ψ are known.
The geometry of the pore space can be characterized by an initial location of the surface Ψ plus its
displacement due to the load. The corresponding average strain of the pore space related to the total
volume of the sample is given by the quantity φij = −ζij /V . If the rock is statistically homogeneous
then the quantity φij is independent of the scale of the sample. Thus, it is reasonable to assume that this
quantity (along with the initial configuration of Ψ and the grain parameters) will control elastic moduli of
the drained rock. We propose also that this quantity will also control (or strongly influence) the tensor of
permeability.
e
Let us consider changes of the averaged strain φij due to changes of the load state (δσij ; δPp ). These
changes of the strain will describe average changes of the geometry of the pore space. It is easy to show
that the rule for a change of φ coincides with the rule for a change of φii . Further, one can show (see
Shapiro and Kaselow, 2005) that:
dr gr dr e gr
δφij = (Sklij − Sklij − φij Sklmm )δσkl + (φSijmm − φij C gr )δPp , (6)
dr gr dr e
δφ = (Sklmm − Sklmm − φSklmm )δσkl . (7)
dr gr
Here Sklij and Sklijare drained and grain compliances of the rock sample. C gr is the compressibility of
the grain material. We observe that load-caused changes of the porosity depend on a single combination

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its Stress Dependence
Annual Report 2012 153

e
of the confining stress and of the pore pressure. This combination is an effective stress, σkl . This is a
consequence of the assumption of statistical homogeneity of the grain material. It is also consistent with
the Gassmann’s equation (see also Brown and Korringa, 1975). In relation with this fact one frequently
states that the effective stress coefficient for the porosity (i.e., the coefficient in front of the pore pressure
in the effective stress) is equal to 1. Note, that generally this is not the case for the quantity φij . The
compliances S gr (including the compressibility C gr ) are practically independent of loads.

Stiff and compliant pore spaces

Analogously to Shapiro (2003) and to Shapiro and Kaselow (2005) we separate the total porosity φ into
two parts
φ = φc + φs0 + φs ,
 
(8)
where the first part, φc , is a compliant porosity supported by thin cracks and grain contact vicinities.
According to laboratory observations we expect that the compliant porosity will close up by the compres-
sional differential stress of few hundred MPa. This corresponds to the porosity with an aspect ratio γ (a
relationship between the minimum  and maximum dimensions of a pore) less than 0, 01 (see Zimmerman
et al., 1986). The second part, φs0 + φs , is a stiff porosity supported by more or less equidimensional


(equant) pores. The aspect ratio of such pores is typically larger than 0.1.
In turn, we separate the stiff porosity into a part φs0 , which is equal to the stiff porosity in the case of
σ e = 0, and to a part φs which is a change of the stiff porosity due to a load. We assume that the
relative changes of the stiff porosity, φs /φs0 , are small. In contrast, the relative changes of the compliant
porosity (φc − φc0 )/φc0 can be very large, i.e., of the order of 1 (φc0 denotes the compliant porosity in the
unloaded case σ e = 0). Note, however, that φc and φc0 are usually very small quantities. As a rule, (e.g.,
in porous sandstones) they are much smaller than the φs0 and even than the absolute value of φs . For
example, in porous sandstones typical orders of magnitude of these quantities are φs0 = 0.1, |φs | = 0.01
and φc = 0.001.
Analogously with ( 8) we also represent φij :

φij = φcij + φs0 s

 
ij + φij . (9)

For the following, as a state with ζij = 0 we define a state of the rock with completely closed porosity.
Then, Vp0 = 0, and φ = φii . Thus, in the following we call quantity φij generalized porosity. Quantity
φs0 s
ij denotes the stiff part of φij in the unloaded state. Quantity φij denotes changes of the stiff part of
the generalized porosity due to a load. Thus, if the load is absent, φsij = 0. Further, φcij denotes the
compliant part of the generalized porosity. It can be completely closed under a compressional stress of
the order of few hundred mega-Pascal. By φc0 ij we will denote compliant part of the generalized porosity
c c
in the unloaded state. It is clear, that φ = φii , φs0 = φs0 s s
ii , and φ = φii .
s c
Taking into account that quantities φij and φij introduced above are small (of the order of strain), it is
logic to assume the first, linear approximations of the skeleton compliances as functions of these quanti-
ties. The Taylor expansion gives:
dr
Sijkl (φs0 c s drs
mn , φmn + φmn ) = Sijkl + C
drs s
θijklmn φsmn + C drs θijklmn
c
φcmn , (10)
drs
where Sijkl is the drained compliancy of a hypothetical rock with a closed compliant porosity (i.e.,
φc = 0) and the stiff porosity equal to φs0 . Further,
dr dr
s 1 ∂Sijkl c 1 ∂Sijkl
θijklmn = , θijklmn = , (11)
C drs ∂φsmn C drs ∂φcmn

where the derivatives are taken in points φs = 0 and φc = 0, respectively and C drs is the bulk compress-
ibility of the hypothetical rock mentioned above. Quantities θ are individual for a given piece of rock
considered in a loading experiment. We assume that in such an experiment as soon as a changing load
leads to the same φsij and φcij the elastic moduli will also assume the same values. We also assume that
the same configuration of the load (independent of its history) will give the same configuration of φsij and
φcij . Thus, non-hysteretic deformations only are considered.

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its Stress Dependence
154

Approximation ( 10) implies that the quantities of the form θφ are smaller than 1. Numerous labora-
tory experiments and practical experience show that the drained compressibilities depend strongly on
changes in the compliant porosity, and depend much weaker on changes in the stiff porosity. If so, the
approximation ( 10) further can be used in the following simplified form:
dr
Sijkl (φs0 , φcij + φsij ) = Sijkl
drs c
+ θijklmn φcmn C drs , (12)

Beside this approximation we will use also following simplifying assumptions. In direct additive terms
we neglect φ in comparison with 1, i.e., we will consider rocks with moderate or small porosity of the
order of 0.1 or less. Then:
drs c
δφij = (Sklij + θklijmn φcgr C drs − Sklij
mt d
)δσkl . (13)

Equations ( 10) and ( 13) together provide a description of stress dependences of rock elastic moduli. The
c c
last equation shows that tensor quantity θklijmn controls these dependences. Tensor quantity θklijmn is
analogous to the scalar dimensionless quantity called piezosensitivity in (Shapiro, 2003). Its symmetry
properties has been discussed in (Shapiro and Kaselow, 2005). Further we will call this quantity the tensor
of stress sensitivity. Because of the symmetry of the tensor φij the following symmetry is valid for the
tensor of stress sensitivity.

Stress dependences of the stiff and compliant pore spaces

We assume that stiff porosity changes with stress are independent of the changes of the compliant porosity.
This means also, that changes of the stiff porosity are independent on the fact if the compliant porosity is
closed or not. If the compliant porosity is closed then φcij = 0 and we obtain from ( 13)
gr
δφsij = (Sklij
drs
− Sklij e
)δσkl . (14)

However, if the assumption above is valid then this relationship will be valid also for an arbitrary (how-
ever, because of other assumptions, small) φc . Therefore,

δφcij = C drs θklijmn

c
φcmn δσkl
e
. (15)

These two equations can be written in the form of systems of partial differential equations, respectively:

∂φsij drs gr
e = Sklij − Sklij .
∂σkl
(16)

∂φcij drs c
e =C θklijmn φcmn . (17)
∂σkl
Taking into account that in the absence of the load quantity φs is vanishing we immediately obtain:
gr
φsij = (Sklij
drs
− Sklij e
)σkl . (18)

A complete analysis of equation system ( 17) is a challenging problem. For an orthorhombic medium with
d d d
principal stresses τ1 = σ11 , τ2 = σ22 , τ3 = σ33 acting along the symmetry axes and an assumption of
independent principal strains of the pore space Shapiro and Kaselow (2005) found the following solution:
c drs c drs c drs
φc11 = φc0
11 e
θ1 τ1 C
, φc22 = φc0
22 e
θ2 τ2 C
, φc33 = φc0
33 e
θ3 τ3 C
. (19)

Here θ1c , θ2c and θ3c stand for θ111111

c c
, θ222222 c
and θ333333 , respectively.
Thus, the stress-dependent generalize porosity (in the case of an orthorhombic symmetry) takes the fol-
lowing form:
gr
φI (τ ) = φs0 drs c0
I + (SIK − SIK )τK + φI exp (θI τI C
c drs
), (20)
where I, and K stand for 11, 22 or 33, and there is no summation over I.

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its Stress Dependence
Annual Report 2012 155

Let us now consider an arbitrary elastic characteristics Λ (e.g., a seismic velocity, a stiffness or a com-
pliance) of a porous drained body. We assume that the characteristic Λ is such that in the point where
porosity is equal to φs0 function Λ can be expanded in the Taylor series (in a way similar to equation (
10)) relative to the porosity (this should be valid for all such characteristics like seismic velocities and
elastic moduli):
Λ(φs0 s c
ij + φij , φij ) = Λ
drs sΛ s
+ θij cΛ c
φij + θij φij , (21)
where we have kept only linear part of the Taylor expansion. Further,
sΛ ∂Λ ∂Λ
θij = , θcΛ = , (22)
∂φsij ∂φcij
and the derivatives are taken at φs = 0 and φc = 0, respectively. Substituting equations ( 18) and ( 19)
into equation ( 21) we obtain:
gr
Λ(τ ) = Λdrs + θij
sΛ drs
(SijK − SijK )τK + θIcΛ φc0 c
I exp (−θI τI C
drs
), (23)
where K and I can assume one of values 1, 2 or 3 denoting 11, 22 and 33, respectively. In the exponent
there is no summation over repeating indices.
These results have a very familiar functional dependence of a form similar to A + KP − B exp (−P D).
Such type of saturating exponential functional dependences of elastic parameters on loads have been
proposed by Zimmerman et al, (1986). Figure 1 shows that equation ( 23) satisfactory describes pressure
dependences of parameters quantifying elastic anisotropy in shale under load.

Figure 1: Anisotropy parameters ǫ, γ, δ and η (quantifying maximum direction-dependent deviations of different

anisotropic stiffnesses in transversely-isotropic rocks) as functions of confining pressure in a shale (Witsker and
Shapiro, 2009). The experimental data on elastic moduli of the Kimmeridge shale are taken from Hornby (1998).
Solid lines show approximations resulting from equation ( 23).

Effective-stress coefficients and the stress dependence of permeability

Different quantities q characterizing poroelastic media depend on the applied confining stress and on the
pore pressure in different ways. In a general case this dependence can be expressed as a function of the

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its Stress Dependence
156

following combination of the loads:

∆σq = −σij − κqij Pp . (24)
Berryman (1992) addresses different forms of load combinations ∆σq as effective-stress rules for corre-
sponding properties of rocks. The quantity κqij is called an effective stress coefficient for a given physical
quantity. It is clear that in a general case this coefficient can be a tensor.
Deformations of the pore space are also related to the stress dependence of the permeability, k. To
understand the effective-stress law for the permeability Berryman (1992) proposed the following way of
argumentation. The physical dimension of the permeability is (length)2. Thus, a uniform rescaling of the
size of an isotropic porous sample (leading also to the corresponding rescaling of sizes of pores) increases
or decreases permeability proportionally to (volume)2/3. Note that such a geometric transformation of the
pore sample does not change its porosity. In other words, this rescaling does not change specific geometric
relations between the solid and the pore space. Therefore, additionally to the sample scale permeability
must depend on a non-dimensional measure of mutual geometric relations between the solid and the pore
space (i.e., the relative positioning of grains). A power-law function of the porosity frequently seems to
be such an adequate measure. For example, it is in agreement with a Kozeny-Carman type of relations
between the porosity and permeability. Taking this into account we obtain:
k ∝ φnp V 2/3 , (25)
where np is a nearly empiric exponent close to 4 or larger. Using this relation Berryman (1992) derives
that the effective-stress coefficient for permeability is
2φ(1 − α)
κk = 1 − ≤ 1, (26)
3np (α − φ) + 2φ
Equation ( 26) shows that the permeability is not just a simple function of the differential pressure. Gen-
erally, it depends on a more complex combination of the both load components, Pp and Pc . Moreover,
heterogeneous rocks can show significant deviations of κk from this equations. For example, it is fre-
quently observed that κk is larger than 1 (Berryman 1992).
We are especially interested in situations, where the main changing load component is the pore pressure
(i.e., ∆σκ ≈ −Pp ). Equation ( 27) can be taken as a basis for a modelling of pressure-dependent per-
meability. However, this equation has been derived just from dimensionality considerations. It treats the
porosity in a very general way. It does not specify which porosity component mainly controls the perme-
ability. The total connected porosity can be substituted in this equation by any other porosity function.
For example, in the previous sections we argued that the stress dependence of drained elastic properties
is mainly controlled by the compliant porosity. However, roles of the compliant and stiff parts of the
pore space can be very different in the pressure dependence of permeability. To model such a pressure
dependence we combine equation ( 27) with equation ( 20). However, to account for possibly different
contributions of the stiff and compliant porosities in the permeability we take these porosity components
with different weighting factors (Φs and Φc , respectively) introduced ad hoc:
k ∝ (Φs (φs0 + Pp (C drs − C gr )) + Φc φc0 exp (θc Pp C drs ))np V (Pp )2/3 . (27)
Here C gr is the compressibility of the grain material. From this relation we can derive several simple
models. By moderate loads the volume V will not change a lot. Thus, we can include it approximately
into a proportionality factor. Let us further assume a rock where the permeability is controlled by a stiff
porosity only (i.e., Φc = 0). In this case we obtain
C drs − C gr
k ∝ (1 + Pp )np . (28)
φs0
Let us further assume that the initial (connected!) stiff porosity is vanishing small. Then we obtain a
power-law model of the permeability:
k ∝ Ppn , (29)
Let us in contrary assume that the contribution of the stiff porosity in the connected porosity is vanishing
small. This will lead to Φs = 0. Then the permeability is controlled by the compliant porosity. This
yields:
k ∝ exp(−bp Pp ), (30)

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its Stress Dependence
Annual Report 2012 157

where bp is a fitting constant.

Figure 2: Distance versus time plot of a hydraulic fracturing induced microseismicity (points correspond to hypocen-
ters of microearthquakes) in Barnett Shale. The solid line shows two types of the temporal growth of the microseismic
cloud. On the left-hand part a usual diffusion type approximation of the triggering (t1/2 ) has been shown. On the
right-hand part a cubic root parabola (t1/3 ) has been shown. It is better matching the data.

Note that the both types of the stress dependence ( 29) and ( 30) can be approximately valid without any
relation to the compliant and stiff porosity. They can be considered as simple modelling alternatives. An-
alyzing dynamics of microseismicity induced by stimulating of gas shales and other hard rocks by fluid
injections in boreholes one can show that all these models are applicable for explaining real data. How-
ever, sometimes it seems that for shale the permeability enhancement is controlled by the stiff component
of the pore space. Then, the power-law model of the permeability seems to be especially adequate. Figure
2 shows a well-known example of hydraulic-fracturing induced seismicity in a Barnett Shale gas reservoir
(Shapiro and Dinske, 2009). The right-hand part of this Figure shows that the cloud of induced seismicity
growths as t1/3 . Such type of behaviour indicates a power-law pressure dependence of the permeability.
In contrast in the case of a permeability being an exponential function of pressure a standard diffusional
growth of the type t1/2 should be expected. Such a growth of microseismic clouds is frequently observed
by geothermal stimulations of granite.
We further propose that the following generalization of equation ( 27) can be applied in a more general
case of an orthotropic loading of orthorhombic rocks:
gr
kI ∝ (Φs (φs0 drs c0 c
I + (SIK − SIK )τK ) + Φc φI exp (θI τI C )) V (Pp )2/3 .
drs np
(31)

Here there is no summation over I.

Conclusions
The stress dependence of elastic moduli of drained rocks is mainly controlled by deformations of the
compliant part of the pore (and fracture) space. This leads to saturating exponential functions expressing
load dependences of seismic velocities and of parameters describing elastic anisotropy in shale and other

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its Stress Dependence
158

rocks. On the other hand a temporal evolution of seismicity induced by fluid stimulations (including
hydraulic fracturing) of rocks sometimes implies power-law dependences of the permeability on the pore
pressure. We propose the following explanation. In some rocks permeability seems to be controlled by the
stiff part of the pore space. This leads to power-law pressure dependence of the permeability. Sometimes
permeability can be controlled by compliant pores (e.g., cracks). Then exponential type of permeability
as function of pressure can be observed. The compliant and stiff parts of the pore space seem to make
very different contributions into a load impact on elastic properties and on hydraulic transport properties
of different rocks.

Acknowledgement
This contribution has been supported by the PHASE research consortium project of the Freie Universitaet
Berlin.

References
Berryman, J. G. (1992). Effective stress for transport properties of inhomogeneous porous rock. Journal
of Geophysical Research, 97:17409–17424.
Brown, R. J. S. and Korringa, J. (1975). On the dependence of the elastic properties of a porous rock on
the compressibility of the pore fluid. Geophysics, 40:608–616.

Hornby, B. E. (1998). Experimental laboratory determination of the dynamic elastic properties of wet,
drained shales. Journal of Geophysical Research, 103:29945–29964.
Shapiro, S. A. (2003). Elastic piezosensitivity of porous and fractured rocks. Geophysics, 68:482–486.
Shapiro, S. A. and Kaselow, A. (2005). Porosity and elastic anisotropy of rocks under tectonic stress and
pore-pressure changes. Geophysics, 70:N27–N38.
Witsker, S. and Shapiro, S. A. (2009). Stress Induced Elastic Anisotropy of Shales. PHASE Research
Project, 5th Annual Report, pages 183–196.
Zimmerman, R., Somerton, W., and King, M. (1986). Compressibility of porous rocks. Journal of
Geophysical Research, 91:12765–12777.

Permeability as a Function of Stiff and Compliant Components of Pores and Cracks and its Stress Dependence
Annual Report 2012 159

Influence of Nonlinear Fluid-rock Interaction on the Back Front of

Seismicity

N. Hummel1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
Borehole fluid injections are a common pre-requisite for the development of geothermal and hydrocarbon
reservoirs. Usually such hydraulic stimulation treatments are accompanied by microseismic activity. The
occurrence of corresponding induced seismicity is not only limited to the injection phase but can also be
observed after completion of the fluid injection. For the post-injection phase the so-called back front of
seismicity is an important phenomenon. It supports the idea that fluid induced seismicity is governed by
a diffusional pore-fluid pressure relaxation. The back front describes the distance from the injection point
at which seismicity is terminated at a given time after the end of the fluid injection. Assuming a constant
with time and pressure hydraulic transport the back front allows to estimate hydraulic diffusivity of the
rock from the spatio-temporal evolution of post-injection induced seismicity. However, fluid injections
can also significantly change hydraulic transport properties. For instance, hydraulic fracturing treatments
can considerably enhance the permeability. In this case the strongly increasing permeability becomes a
function of the pressure perturbation. For such situations we explore microseismic signatures of the back
front. We numerically consider a power-law as well as an exponential dependence of the diffusivity on the
pressure perturbation. This leads to a hydraulic transport which is described by nonlinear diffusion. We
numerically compute solutions of corresponding nonlinear diffusion equations. We introduce a medium
realization which takes into account a post-injection enhanced permeability. For different locations this
model changes the nonlinear hydraulic transport into a constant one. This hydraulic modification is done
as soon as the local pressure perturbation has reached its maximum. For such a medium we compute
pressure profiles which are used to generate synthetic microseismicity. We analyze the spatio-temporal
characteristics of post-injection induced seismicity to identify the character of the back front. Our results
show that nonlinear fluid-rock interaction still leads to a distinct and therefore detectable back front.
However, the character of the back front can significantly differ from a square root of time dependence.

Introduction
Last year we started to explore microseismic back front signatures for situations where hydraulic transport
properties become functions of the pressure (Hummel and Shapiro, 2011). For this nonlinear fluid-rock
interaction we considered two different models of a pressure-dependent diffusivity. To approximate the
complicated permeability evolution after termination of the fluid injection we assumed that the hydraulic
transport remains nonlinear. Such a medium realization corresponds to a reservoir rock where fractures
close from the injection point into the direction of fracture propagation. Hence, the application of this
model is limited because it does not account for a post-injection enhanced permeability.
In what follows we introduce a different type of medium realization. In particular, it includes the possibil-
ity of a post-injection enhanced hydraulic transport. Our model consists of a nonlinear medium diffusivity
which we freeze to a constant enhanced value. For each cell of our medium this hydraulic modification
is done as soon as the local pressure perturbation has reached its maximum. For such type of medium we
compute pressure profiles by solving corresponding nonlinear diffusion equations. In analogy to Hummel
and Shapiro (2011) we numerically generate synthetic clouds of seismic events and analyze correspond-
ing spatio-temporal features of post-injection induced seismicity. With this numerical study we expect to

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

160

gain further knowledge about the post-injection hydraulic transport. In particular, we want to address the
following questions: How does the back front look like in such media? What are the back front signatures
of seismicity induced by nonlinear fluid-rock interaction? How do microseismic back front signatures of
the frozen medium diffusivity model differ from those of the nonlinear medium diffusivity model?

Theory and modeling

Below we briefly recall the state of affairs. To study back front signatures of seismicity based on nonlinear
fluid-rock interaction we consider two different models of a pressure-dependent diffusivity. One is the
following power-law model
D(p(r; t)) = (n + 1) D0 pn , (1)
which formerly was introduced by Shapiro and Dinske (2009). This power-law model approximates the
evolution of hydraulic transport properties during fracturing for situations where the initial formation per-
meability can be nearly neglected (e.g. shales). The quantity n controls the grade of nonlinearity on the
diffusion. We call it the index of nonlinearity. Physically, n is an effective property of the reservoir. It
must depend on the lithology, the strength and stress state of rocks, elastic properties and the pore space
geometry. If n is large, the diffusivity will depend strongly on the pressure. In the limit of n = 0 the
diffusivity becomes independent of the pressure which results in a linear pressure diffusion. The corre-
sponding hydraulic transport is then characterized by the constant D0 . Therefore, D0 is a normalizing
constant which should not be interpreted independently.
In addition to the power-law model we also consider an exponential dependence of diffusivity on the
pressure perturbation p
D(p(r; t)) = D0 eκ p(r;t) . (2)
Formerly, this model was introduced by Hummel and Müller (2009). It is appropriate for rocks having
an in-situ non-negligible initial diffusivity (e.g. granitic rocks, geothermal reservoirs). The pressure
influence is given by the so-called permeability compliance κ. If κ is large the hydraulic diffusivity will
strongly depend on pressure. In the limit of κ = 0 Pa−1 the diffusivity becomes independent of the
pressure which results in a linear pore pressure diffusion. The corresponding hydraulic transport is then
characterized by the constant D0 (as in the case of the power-law model).
Taking into account both pressure-dependent diffusivity models ( 1) and ( 2) the hydraulic transport can
be described by the following nonlinear diffusion equation (introduced by Shapiro and Dinske, 2009)

∂ rd−1 p(r; t)
 
∂ d−1 ∂p(r; t)
= D(p(r; t)) r . (3)
∂t ∂r ∂r

This equation describes the spatio-temporal evolution of the pressure perturbation p in a d- dimensional
space. Here, we approximate a borehole fluid injection experiment by a point source of pressure pertur-
bation into an initially hydraulically homogeneous, spherically symmetric and isotropic fluid-saturated
medium. For the index of nonlinearity we consider values of n = 1, 2 and 3. For the injection pressure
rescaled permeability compliance we use values of κ = 0.2 Pa−1 , 0.4 Pa−1 and 0.5 Pa−1 (Hummel and
Müller, 2009). Depending on the dimensionality and the magnitudes of n and κ solutions of equation (
3) correspond to pressure profiles in rocks with a specific pressure-dependent hydraulic diffusivity. For
example solutions for d = 3 and κ = 0.5 Pa−1 correspond to pressure profiles in a highly fractured three
dimensional rock. For D0 we uses a value of 1 m2 s−1 . The injection source is realized as a Dirichlet-type
boundary condition with a time-dependent injection pressure magnitude of p0 = 10 Pa. In our simula-
tions this injection source is switched off after t = t0 = 40 s. After termination of the fluid injection
the pressure perturbation decreases. In turn, the effective normal stress increases and the rock becomes
more stable. Consequently, induced cracks and fractures close. However, fracture closing can be pre-
vented by the use of proppant. This is normally the case in hydraulic fracturing treatments which leads
to a post-injection enhanced hydraulic permeability (Economides and Nolte, 2000). To approximate the
complicated hydraulic behavior of the medium and to account for a post-injection enhanced permeability
we introduce the following model of a frozen medium diffusivity (Figure 1):
As a consequence of terminating the fluid injection a diffusional-like pressure wave will penetrate the

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

Annual Report 2012 161

Figure 1: Top: Time-dependent injection pressure source function of duration t0 = 40 s. The horizontal axis
represents the time t. Middle: Pore pressure versus time for three different distances r (r1 < r2 < r3 ). After
shut-in (dashed line) the pressure relaxation leads to a diffusional-like pressure wave with maxima pmax (r; t) (black
circles). Bottom: Medium diffusivity versus time for corresponding distances r. As soon as the medium diffusivity
has reached its maximum Dmax (pmax (r; t)) at a certain distance the diffusivity will thereafter remain constant with
Dmax (pmax (r)).

medium. At a particular location (i.e. distance from the injection point r) and time the pressure per-
turbation will increase up to a specific maximum pmax (r; t). As long as the pressure increases induced
cracks and fractures continue to grow. However, as soon as the pressure has reached pmax (r; t) it starts
to decrease. This pressure reduction leads to closing of induced cracks and fractures.
For the hydraulic medium diffusivity D(p(r; t)) we observe a similar behavior. Due to the pressure de-
pendence the medium diffusivity will also increase up to a maximum value Dmax (pmax (r; t)). However,
to account for a post-injection enhanced hydraulic diffusivity we prevent D(p(r; t)) to decrease there-
after. As soon as the medium diffusivity has reached its maximum at a particular cell of our medium
we will keep it constant with Dmax (pmax (r; t)) for the rest of the experiment. We note that this result-
ing post-injection enhanced diffusivity is a function of the distance. With increasing distance from the
wellbore this modified diffusivity decreases. Such a behavior corresponds to cracks and fractures with a
post-injection distance-dependent aperture. Close to the wellbore a fracture remains almost open. Thus,
it has a high post-injection enhanced diffusivity. As soon as the distance from the borehole increases the
fracture aperture becomes smaller. Therefore, the diffusivity of a fracture decreases. To our knowledge
this is a new type of model to approximate the complicated behavior of a post-injection enhanced dif-
fusivity. It also constitutes a significant conceptual update to the nonlinear medium diffusivity model of
Hummel and Shapiro (2011).
Finally, we compute synthetic seismicity clouds in the r − t domain by comparing the pressure evolutions
with a criticality field C(r) (further details are given in Rothert and Shapiro, 2003; Hummel and Müller,
2009). For each cell of the medium this field represents randomly distributed critical pressure values
which are necessary to trigger an event. The values of C are broadly distributed between 0 Pa and p0 .
Large C values correspond to stable cracks while small C values represent weak cracks. We note that
all parameters presented here are the same as reported in Hummel and Shapiro (2011). Thus we are able
to compare the results of the nonlinear medium diffusivity model (Hummel and Shapiro, 2011) with the
ones based on the frozen medium diffusivity model (presented in this paper).

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

162

Determination of back front signatures

To study microseismic back front signatures for nonlinear fluid-rock interaction we analyze spatio-temporal
characteristics of post-injection induced seismicity. This evaluation is done in analogy to Hummel and
Shapiro (2011).
First we apply the back front concept to our synthetic event clouds. This method has been introduced by
Parotidis et al. (2004) assuming that microseismic events can only be triggered for increasing pore pres-
sures. It supports the idea that fluid induced seismicity is governed by a diffusional pressure relaxation.
The back front describes the distance from the injection point rbf at which seismicity is terminated at a
given time t after the end of the fluid injection
s    
t t
rbf = 2 d Dbf t − 1 ln . (4)
t0 t − t0

Assuming a constant with time and pressure hydraulic transport the back front allows to estimate a scalar
hydraulic diffusivity Dbf of the rock from the spatio-temporal evolution of post-injection induced seis-
micity. Hence, we fit equation ( 4) as envelope of the domain of seismic quiescence in the r − t plot. As
a result we obtain a specific diffusivity estimate Dbf .
For situations where the t1/2 - dependent back front parabola (equation 4) does not properly describe the
envelope of seismic quiescence we numerically determine the position of the back front. We do this by
computing the spatio-temporal pressure maxima pmax (r; t) for times t > t0 . We obtain the back front
location by tracking these rpmax - tpmax values in the r − t plot. We then consider the following fitting
function
rbf = a (t − t0 )b , (5)
which has two fitting parameters a and b. The best-fit function for the numerically determined location
of the back front allows us to determine its character.
To better understand what the back front provides in the case of nonlinear fluid-rock interaction we
compare the obtained diffusivity estimates Dbf with two other diffusivity values. One is the effective
diffusivity Def f which we compute for the shut-in time according to (Hummel and Shapiro, 2011)
−1
Zrf e

1 1
Def f (rf e ; tf e ) =  dr . (6)
rf e D(p(r; t0 ))
0

This single value represents the effective hydraulic state of the pressure-stimulated medium immediately
before termination of the fluid injection. Here, rf e corresponds to the radial distance of the farthest
microseismic event at the shut-in time t0 .
Finally, the back front diffusivity estimate Dbf is compared with a heuristic diffusivity estimate Dh .
This value is obtained from the triggering front concept. This method has been introduced by Shapiro
et al. (1997, 1999, 2002) assuming a time- and pressure-independent hydraulic transport. The triggering
front approximately describes the envelope of injection phase seismicity in the r − t domain. For a
homogeneous and isotropic medium, the triggering front is approximately given by
p
r = 4 π Dh t . (7)

It allows to estimate a scalar hydraulic diffusivity Dh from the spatio-temporal distribution of injection
phase induced seismicity. For the linear diffusion limit, triggering front and back front provide the same
diffusivity estimate, Dh = Dbf = D0 . However, for a pressure-dependent hydraulic transport the
medium is pressurized by the nonlinear diffusion. For such situations Hummel and Shapiro (2012) show
that the heuristic diffusivity estimates obtained from the triggering front are estimates of the effective
diffusivity.

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

Annual Report 2012 163

Results
Pore-fluid pressure profiles
In analogy to Hummel and Shapiro (2011) we first compute pressure profiles as solutions of the nonlinear
diffusion equation. We note that these pressure profiles are based on the frozen medium diffusivity
model. For power-law and exponential diffusion we observe the following characteristics (Figures 2, 3,
4, 5, 6, 7): Most noticeable is the effect of geometrical spreading. In 2- and 3-D, the pressure perturbation
injected with a constant magnitude is able to propagate into additional directions. As a result, the pressure
profiles are characterized by a strong spatial pressure decay in the vicinity of the injection point.
For increasing nonlinear fluid-rock interaction (i.e. increasing indices of nonlinearity n or permeability
compliances κ) the pressure profiles penetrate deeper into the formation. Additionally, for the power-law
diffusion model, the effect of the medium pressurization is stronger than for the exponential diffusion
model. This feature can be seen most clearly in 1-D and 2-D. For the same observation time the power-
law pressure diffusion is faster than the exponential pressure diffusion. Correspondingly, the penetrated
distance is larger as well. Furthermore, the tips of the pressure profiles differ in their shape. Their
geometry depends on the type of nonlinear fluid-rock interaction. For the power-law diffusion model the
tips of the profiles are characterized by a distinct pressure step. Ahead, the pressure perturbation is still
zero and the medium has not yet been pressurized. In contrast, the profile tips of the exponential diffusion
model show a smooth transition. For small permeability compliances, this transition behavior is quite
similar as for linear diffusion-based profiles (Hummel and Shapiro, 2011). With increasing nonlinearity,
the pressure profiles also change their character from concave to convex. However, this behavior is limited
to the injection phase.
The features described so far are consistent with the ones based on the nonlinear medium diffusivity model
(Hummel and Shapiro, 2011). As soon as the injection source is switched off the pressure decreases. With
ongoing observation time a damped diffusional-like wave penetrates further into the medium. For the
frozen medium diffusivity model the pressure profiles are characterized by a slow increase which results
in a specific spatio-temporal pressure maximum pmax (r; t). With ongoing observation time, the pressure
relaxation reduces the amplitude of these pressure maxima. As soon as the pressure has reached its
maximum at a particular distance it drops to zero thereafter. Additionally, one can see that with increasing
nonlinear fluid-rock interaction the pressure maxima move closer to the tips of the profiles. Consequently,
at the front, an increase in nonlinearity results in a pressure compression with strong gradients. However,
with increasing dimension, this effect becomes less pronounced.

Clouds of synthetic seismicity

Clouds of synthetic seismicity are generated by comparing the computed pressure fields with the critical-
ity field (Figures 2, 3, 4, 5, 6, 7). In analogy to Hummel and Shapiro (2011) we observe that clouds of
seismicity based on the power-law diffusion model show a sharp upper boundary. Ahead, there are no
seismic events and the rock has not yet been pressurized. As explained in Hummel and Shapiro (2012)
this distinct upper boundary of the cloud results from the characteristic shape of the pressure profiles.
Since these profiles are characterized by tips where the pressure quickly drops to zero there is a clear sep-
aration between the already stimulated and virgin rock. This distinct demarcation is responsible for the
sharp upper contour of the seismicity cloud. In comparison, seismicity clouds based on the exponential
diffusion model have a more diffuse upper boundary. This effect is due to the smooth transition occurring
at the tips of corresponding pressure profiles.
With termination of the fluid injection a growing domain of seismic quiescence evolves in the r − t plot.
This domain becomes more pronounced as the nonlinear fluid-rock interaction increases. However, in
most cases, fitting the t1/2 − dependent back front (equation 4) as envelope of this seismically quiet area
provides unsatisfying results. The obtained diffusivity values seem to be always overestimated. Only for
a weak nonlinearity and 3-D exponential diffusion the back front fits to the domain of seismic quiescence.
As in most cases, the t1/2 − dependent back front does not describe the spatio-temporal distribution of
post injection induced seismicity we determine the back front character by numerically evaluating its
location.

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

164

1D powerlaw diffusion, n = 1 1D powerlaw diffusion, n = 1

10 600
232 events
9 t = 40 s 2
Dh = 3.9 m /s
0

500 Pmaxbf
8
2
Dbf = 28.6 m /s
Pore−fluid pressure [Pa]

7 21⋅ (t − t0)
0.25
400
2
6 Deff(t0) = 6.7 m /s

Distance [m]
5 t = 60 s 300

4 t = 80 s
200
3

2
t = 100 s 100
1

0 0
0 100 200 300 400 500 0 20 40 60 80 100
Distance [m] Time [s]

1D powerlaw diffusion, n = 2 1D powerlaw diffusion, n = 2

10 600
743 events
9 t = 40 s 2
Dh = 27.7 m /s
0

500 Pmaxbf
8
2
Dbf = 289.1 m /s
Pore−fluid pressure [Pa]

7 71⋅ (t − t0)
0.2
400
2
6 Deff(t0) = 66.1 m /s
Distance [m]

5 300
t = 60 s t = 80 s
4
200
3

2 t = 100 s
100
1

0 0
0 100 200 300 400 500 0 20 40 60 80 100
Distance [m] Time [s]

1D powerlaw diffusion, n = 3 1D powerlaw diffusion, n = 3

10 600

9 t = 40 s
0

8 500
Pore−fluid pressure [Pa]

7
400
6
Distance [m]

5 300

4 2361 events
2
Dh = 247.2 m /s
t = 100 s 200
3 Pmaxbf
t = 60 s t = 80 s D = 3245.1 m2/s
2 bf
100 0.22
213⋅ (t − t0)
1 2
Deff(t0) = 724.9 m /s
0 0
0 100 200 300 400 500 0 20 40 60 80 100
Distance [m] Time [s]

Figure 2: Snapshots of pore-fluid pressure profiles (left) and corresponding r − t plot of synthetic microseismicity
(right) based on 1-D power-law diffusion. After shut-in the pressure at the source drops to zero. Simultaneously, a
diffusional-like wave penetrates further into the medium. With increasing distance from the wellbore, the pressure
profiles increase to a specific maximum. Thereafter, the pressure finally drops to zero. For increasing indices of
nonlinearity n these pressure maxima are shifted closer to the tip of the profiles. At the front, this results in a
pressure compression including strong gradients. The distinct demarcation between the virgin and already pressurized
rock leads to the sharp upper boundary of the seismicity clouds. For increasing pressure influence (i.e. increasing
values of n) the domain of seismic quiescence increases. However, the t1/2 − dependent back front (black dashed
parabola) does not properly describe the spatio-temporal evolution of post injection seismicity. This is confirmed
by the exponents b of the best fit function (red) describing the numerically determined location of the back front
(black curve). In all three cases the magnitudes b are significantly smaller than 0.5. Compared to the heuristic
diffusivity estimate Dh obtained from the triggering front (blue) and the computed effective diffusivity value Def f
the diffusivity estimates of the back front are always the largest.

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

Annual Report 2012 165

−1
1D exponential diffusion, κ = 0.2 Pa 1D exponential diffusion, κ = 0.2 Pa−1
10 200
135 events
2
9 180 Dh = 1.7 m /s
t0 = 40 s
Pmaxbf
8 160 2
Dbf = 8.9 m /s
Pore−fluid pressure [Pa]

7 140 11⋅ (t − t0)

0.26

2
6 120 Deff(t0) = 2.3 m /s

Distance [m]
5 100
t = 60 s
4 80
t = 80 s
3 60
t = 100 s
2 40

1 20

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Distance [m] Time [s]

−1
1D exponential diffusion, κ = 0.4 Pa 1D exponential diffusion, κ = 0.4 Pa−1
10 200
273 events
2
9 t0 = 40 s 180 Dh = 4.1 m /s
Pmaxbf
8 160 2
Dbf = 34.7 m /s
Pore−fluid pressure [Pa]

7 140 27⋅ (t − t0)

0.19

2
6 120 Deff(t0) = 7.3 m /s
Distance [m]

5 100

4 80
t = 80 s
3 t = 60 s 60
t = 100 s
2 40

1 20

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Distance [m] Time [s]

−1
1D exponential diffusion, κ = 0.5 Pa 1D exponential diffusion, κ = 0.5 Pa−1
10 200
t = 40 s 413 events
2
9 0
180 Dh = 8.5 m /s
Pmaxbf
8 160 2
Dbf = 75.8 m /s
Pore−fluid pressure [Pa]

7 140 41⋅ (t − t0)

0.17

2
6 120 Deff(t0) = 9.4 m /s
Distance [m]

5 100

4 80

3 60
t = 80 s
t = 60 s t = 100 s
2 40

1 20

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Distance [m] Time [s]

Figure 3: Snapshots of pore-fluid pressure profiles (left) and corresponding r − t plot of synthetic microseismicity
(right) based on 1-D exponential diffusion. The tips of the pressure profiles are characterized by a smooth transition.
This smooth transition leads to the diffuse distribution of events at the upper boundary of corresponding seismicity
clouds. With increasing pressure influence (i.e. increasing values of κ) the domain of seismic quiescence increases.
The t1/2 − dependent back front (black dashed parabola) does not properly describe the spatio-temporal evolution of
post injection seismicity. This is confirmed by the exponents b of the best fit function (red) describing the numerically
determined location of the back front (black curve). In all three cases the magnitudes b are significantly smaller than
0.5. Compared to the heuristic diffusivity estimate Dh obtained from the triggering front (blue) and the computed
effective diffusivity value Def f the diffusivity estimates of the back front are always the largest.

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

166

Radial 2D powerlaw diffusion, n = 1 Radial 2D powerlaw diffusion, n = 1

10 200
2976 events
2
9 180 Dh = 1.4 m /s
Pmaxbf
8 160 2
Dbf = 4.3 m /s
Pore−fluid pressure [Pa]

7 140 12⋅ (t − t0)

0.23

t = 40 s 2
0 Deff(t0) = 1.9 m /s
6 120

Distance [m]
5 100

4 80
t = 60 s
3 60
t = 80 s
2 40
t = 100 s
1 20

0 0
0 50 100 150 200 0 20 40 60 80 100
Distance [m] Time [s]

Radial 2D powerlaw diffusion, n = 2 Radial 2D powerlaw diffusion, n = 2

10 200
10080 events
2
9 180 Dh = 6.6 m /s
Pmaxbf
8 160 2
Dbf = 29.1 m /s
Pore−fluid pressure [Pa]

7 140 35⋅ (t − t0)

0.18
t = 40 s
0 2
6 120 Deff(t0) = 14.9 m /s
Distance [m]

5 100

4 80
t = 60 s t = 80 s
3 60

2 t = 100 s 40

1 20

0 0
0 50 100 150 200 0 20 40 60 80 100
Distance [m] Time [s]

Radial 2D powerlaw diffusion, n = 3 Radial 2D powerlaw diffusion, n = 3

10 200

9 180

8 160
t = 40 s
0
Pore−fluid pressure [Pa]

7 140

6 120
Distance [m]

5 100

4 80 31614 events
2
Dh = 44.2 m /s
t = 100 s
t = 60 s t = 80 s
3 60 Pmaxbf
D = 240.6 m2/s
2 40 bf
0.16
101⋅ (t − t0)
1 20 2
Deff(t0) = 107.3 m /s
0 0
0 50 100 150 200 0 20 40 60 80 100
Distance [m] Time [s]

Figure 4: Snapshots of pore-fluid pressure profiles (left) and corresponding r − t plot of synthetic microseismicity
(right) based on 2-D power-law diffusion. The effect of geometrical spreading leads to the pressure drop close to the
injection point. However, with increasing distance from the wellbore the shape of the profiles changes from concave
to convex. As in the 1-D case, the seismicity clouds are characterized by a sharp upper boundary. For increasing
pressure influence (i.e. increasing values of n) the domain of seismic quiescence increases. The t1/2 − dependent
back front (black dashed parabola) does not properly describe the spatio-temporal evolution of post √ injection seismic-
ity. The numerically determined location of the back front (black solid line) deviates from the t-behavior. This is
confirmed by the best-fit function (red). In all cases their exponents b are significantly smaller than 0.5. Compared to
the heuristic diffusivity estimate Dh obtained from the triggering front (blue) and the computed effective diffusivity
value Def f the diffusivity estimates of the back front are always the largest.

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

Annual Report 2012 167

−1
Radial 2D exponential diffusion, κ = 0.2 Pa Radial 2D exponential diffusion, κ = 0.2 Pa−1
10 100
1649 events
2
9 90 Dh = 1.4 m /s
Pmaxbf
8 80 2
Dbf = 1.7 m /s
Pore−fluid pressure [Pa]

7 70 6⋅ (t − t0)
0.3

t0 = 40 s 2
6 60 Deff(t0) = 1.1 m /s

Distance [m]
5 50

4 40
t = 60 s
3 30
t = 80 s
2 20
t = 100 s
1 10

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Distance [m] Time [s]

−1
Radial 2D exponential diffusion, κ = 0.4 Pa Radial 2D exponential diffusion, κ = 0.4 Pa−1
10 100
3953 events
2
9 90 Dh = 2.5 m /s
Pmaxbf
8 80 2
Dbf = 4.1 m /s
t = 40 s
Pore−fluid pressure [Pa]

7 0 70 12⋅ (t − t0)
0.23

2
6 60 Deff(t0) = 1.9 m /s
Distance [m]

5 50

4 t = 60 s 40

3 t = 80 s 30

2 20
t = 100 s
1 10

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Distance [m] Time [s]

−1
Radial 2D exponential diffusion, κ = 0.5 Pa Radial 2D exponential diffusion, κ = 0.5 Pa−1
10 100
6146 events
2
9 90 Dh = 3.3 m /s
Pmaxbf
8 t0 = 40 s 80 2
Dbf = 7.4 m /s
Pore−fluid pressure [Pa]

7 70 18⋅ (t − t0)
0.19

2
6 60 Deff(t0) = 3 m /s
Distance [m]

5 50

4 40
t = 60 s
3 t = 80 s 30

2 20
t = 100 s
1 10

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Distance [m] Time [s]

Figure 5: Snapshots of pore-fluid pressure profiles (left) and corresponding r − t plot of synthetic microseismicity
(right) based on 2-D exponential diffusion. The effect of geometrical spreading leads to the pressure drop close to
the injection point. However, with increasing distance from the wellbore the shape of the profiles changes from
concave to convex. As in the 1-D case, the tips of the pressure profiles are characterized by a smooth transition. This
smooth transition domain leads to the diffuse distribution of events at the upper boundary of corresponding seismicity
clouds. With increasing pressure influence (i.e. increasing values of κ) the domain of seismic quiescence increases.
The t1/2 − dependent back front (black dashed parabola) does not properly describe the spatio-temporal evolutions
of post
√ injection seismicity. The numerically determined location of the back front (black solid line) deviates from
the t-behavior. This is confirmed by the best-fit function (red). In all cases their exponents b are significantly
smaller than 0.5. Compared to the heuristic diffusivity estimate Dh obtained from the triggering front (blue) and the
computed effective diffusivity value Def f the diffusivity estimates of the back front are always the largest.

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

168

Radial 3D power−law diffusion, n = 1

50
6611 events
0.4
45 2.2 * (t)
Pmax
bf
40 D = 0.4 m2/s
bf
0.18
35 5.3 * (t − t )
0
2
Deff(t0) = 0.3 m /s
30

Distance [m]
25

20

15

10

0
0 20 40 60 80 100
Time [s]

Radial 3D powerlaw diffusion, n = 2 Radial 3D power−law diffusion, n = 2

10 50
17786 events
9 0.38
45 3.3 * (t)
Pmax
bf
8 40 D = 1 m2/s
bf
0.13
8.8 * (t − t )
Pore−fluid pressure [Pa]

7 35 0
2
Deff(t0) = 0.6 m /s
6 30
Distance [m]

5 25

4 t0 = 40 s 20

3 15
t = 80 s
2 t = 60 s 10
t = 100 s

1 5

0 0
0 5 10 15 20 25 30 0 20 40 60 80 100
Distance [m] Time [s]

Radial 3D powerlaw diffusion, n = 3 Radial 3D power−law diffusion, n = 3

10 50
45797 events
9 0.37
45 5.9 * (t)
Pmax
bf
8 40 D = 3.5 m2/s
bf
0.1
17.6 * (t − t )
Pore−fluid pressure [Pa]

7 35 0
2
Deff(t0) = 2.4 m /s
6 30
Distance [m]

5 25
t0 = 40 s
4 20

3 15
t = 100 s
t = 60 s t = 80 s
2 10

1 5

0 0
0 5 10 15 20 25 30 0 20 40 60 80 100
Distance [m] Time [s]

Figure 6: Snapshots of pore-fluid pressure profiles (left) and corresponding r − t plot of synthetic microseismicity
(right) based on 3-D power-law diffusion. The strong pressure drop close to the injection point results from the
geometrical spreading. As in the 1-D and 2-D case, the pressure at the tips of the profiles drops to zero. This pressure
drop separates the virgin from the already pressure-stimulated rock. It is responsible for the sharp upper boundary of
the seismicity clouds.
√ For increasing pressure influence (i.e. increasing values of n) the domain of seismic quiescence
increases. The t− dependent back front (black dashed parabola) does not properly describe the spatio-temporal
evolution of post injection seismicity. However, their diffusivity estimates almost agree with the computed effective
diffusivity values. The magnitudes of√ the best fit function (red) confirms that the numerically determined back front
(black solid line) deviates from the t-behavior. In all cases their exponents b are significantly smaller than 0.5. In
comparison to 1- and√2-D, the character of the seismicity cloud has changed. The triggering front (blue) is not any
more described by a t− dependence (further details are given in Hummel and Shapiro, 2012).

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

Annual Report 2012 169

−1
Radial 3D exponential diffusion, κ = 0.2 Pa Radial 3D exponential diffusion, κ = 0.2 Pa−1
0.2 100
2944 events
2
0.18 90 Dh = 1.2 m /s
Pmaxbf
0.16 80 2
Dbf = 1.5 m /s
t0 = 40 s
0.36
Pore−fluid pressure [Pa]

0.14 70 5⋅ (t − t0)
2
0.12 60 Deff(t0) = 1 m /s

Distance [m]
0.1 50

0.08 40
t = 60 s
0.06 30

0.04 t = 80 s 20
t = 100 s
0.02 10

0 0
0 10 20 30 40 50 0 20 40 60 80 100
Distance [m] Time [s]

−1
Radial 3D exponential diffusion, κ = 0.4 Pa Radial 3D exponential diffusion, κ = 0.4 Pa−1
0.2 100
5288 events
2
0.18 90 Dh = 1.5 m /s
Pmaxbf
0.16 80 2
t0 = 40 s Dbf = 1.6 m /s
0.35
Pore−fluid pressure [Pa]

0.14 70 5⋅ (t − t0)
2
0.12 60 Deff(t0) = 1 m /s
Distance [m]

0.1 50

0.08 t = 60 s 40

0.06 30
t = 80 s
0.04 20
t = 100 s
0.02 10

0 0
0 10 20 30 40 50 0 20 40 60 80 100
Distance [m] Time [s]

−1
Radial 3D exponential diffusion, κ = 0.5 Pa Radial 3D exponential diffusion, κ = 0.5 Pa−1
0.2 100
8018 events
2
0.18 t0 = 40 s 90 Dh = 1.8 m /s
Pmaxbf
0.16 80 2
Dbf = 1.6 m /s
0.33
Pore−fluid pressure [Pa]

0.14 70 5⋅ (t − t0)
2
0.12 60 Deff(t0) = 1.1 m /s
Distance [m]

t = 60 s
0.1 50

0.08 40
t = 80 s
0.06 30

0.04 20
t = 100 s
0.02 10

0 0
0 10 20 30 40 50 0 20 40 60 80 100
Distance [m] Time [s]

Figure 7: Snapshots of pore-fluid pressure profiles (left) and corresponding r − t plot of synthetic microseismicity
(right) based on 3-D exponential diffusion. The strong pressure drop close to the injection point results from the
geometrical spreading. The smooth pressure transition at the tips of the profiles leads to the diffuse upper boundary of
corresponding seismicity clouds. For increasing pressure influence (i.e. increasing values of κ) the domain of seismic
quiescence increases. Interestingly we observe that the t1/2 − dependent back front (black dashed parabola) now
describes the spatio-temporal evolution of post injection seismicity. It also agrees with the numerically determined
location of the back front (black solid line). Their character is quite close to a cubic-root of time dependence (shown
by the red curve). In general, all three diffusivity values (Dbf , Dh , and Def f ) show quite good agreement. This is
due to the fact that the corresponding hydraulic transport is quite similar compared to linear diffusion. In other words
the effect of nonlinear diffusion is less pronounced as the effect of geometrical spreading. For κ ≫ 1 nonlinear
effects should become apparent. However, to our knowledge, such large magnitudes are not reported in the literature.

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

170

For this we first track the spatio-temporal pressure maxima in the r − t domain. We then fit this resulting
curve using equation 5. In this way we obtain a best fit function with a characteristic exponent b. Our
results show that in most cases the best fit function has exponents b which are significantly smaller than
0.5 (representing a t1/2 signature). In particular, the magnitudes of b range from 0.1 to 0.3. Additionally,
we still observe that for the 1-D and 2-D power-law diffusion model and an index of nonlinearity n = 3,
the best fit function slightly deviates from the numerically determined back front location.
In order to obtain reference values for Dbf we determine a heuristic diffusivity and an effective diffusivity
value. For the heuristic diffusivity value we fit the triggering front (equation 7) to the upper bound of
injection-phase induced seismicity in the r − t plot. In contrast, the effective diffusivity value is computed
according to equation ( 6). Except of the seismicity clouds based on 3-D exponential diffusion, the back
front diffusivity estimates are larger compared to the corresponding effective and heuristic diffusivity
values. However, for 3-D exponential diffusion the diffusivity estimates obtained from equation ( 4)
agree with both the heuristic diffusivity estimate Dh and the computed effective diffusivity Def f . This is
due to the fact that the corresponding hydraulic transport is still quite similar compared to linear diffusion
(i.e. time- and pressure-independent). In this case, the effect of geometrical spreading is larger compared
to the effect of nonlinear diffusion. However, for permeability compliances κ ≫ 1 nonlinear effects
should become apparent. Admittedly, to our knowledge, such large magnitudes are not reported. As
a consequence, for three dimensional exponential diffusion, a t1/2 -dependent back front approximately
describes the spatio-temporal evolution of corresponding post-injection seismicity.

Frozen versus nonlinear medium diffusivity

In what follows we compare the results obtained from the frozen medium diffusivity model with the ones
obtained from the nonlinear medium diffusivity model. Thus, we suggest to have the last year’s Phase
Report on hand. According to our modeling, the medium diffusivity can only be modified after termina-
tion of the fluid-injection. Hence, the injection phase remains unaffected of these hydraulic modifications.
Consequently, for the injection phase, the spatio-temporal features of corresponding pressure fields and
synthetic event clouds are the same. However, once the fluid injection is terminated, differences in the
results of both models become apparent. First of all, a general difference is given by the two models itself.
The nonlinear medium diffusivity model is not able to account for a post-injection enhanced hydraulic
transport. In contrast, the frozen medium diffusivity model is capable to compensate this drawback. With
this model we simulate a rock containing induced fractures with a post-injection distance-dependent aper-
ture. This major difference in both models leads to different post-injection characteristics in the computed
pressure evolutions. For instance, the shape of the pressure profiles based on the nonlinear medium diffu-
sivity is not the same as the ones based on the frozen medium diffusivity. The latter have a lower pressure
increase as well as smaller maximum pressure magnitudes. Additionally, the spatio-temporal pressure
maxima are shifted closer to the tips of the profiles. This leads to a pressure compression with denser gra-
dients. In contrast, pressure profiles based on the nonlinear medium diffusivity model are characterized
by a bulgy-like shape. Although their maximum pressure magnitudes are higher, the pressure gradients
at the tips of the profiles are less dense.
Despite of these differences in the pressure evolutions we observe only minor differences in the spatio-
temporal characteristics of synthetic event clouds. Since the pressure magnitudes are generally higher for
the nonlinear medium diffusivity model corresponding seismicity clouds comprise more events. How-
ever, their distribution does not provide significantly different spatio-temporal characteristics compared
to the ones from the frozen medium diffusivity model.

Conclusions
We continued to study back front signatures of seismicity induced by nonlinear fluid-rock interaction.
For this we considered two models describing a power-law and an exponential pressure dependence of
hydraulic diffusivity. To account for a post-injection enhanced hydraulic transport we introduced a model
of a frozen medium diffusivity. In this model, the nonlinear medium diffusivity is modified to a constant
enhanced diffusivity as soon as the pressure perturbation has reached its spatio-temporal maximum. Both

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

Annual Report 2012 171

pressure-dependent diffusivity models lead to nonlinear diffusion equations which we solved for a time-
dependent source function. After termination of the fluid injection the pressure profiles are characterized
by a pressure increase which results in pressure maxima. With increasing nonlinearity these spatio-
temporal pressure maxima are shifted closer to the tips of the profiles. At the front this leads to an
enhanced pressure compression including denser pressure gradients. On the basis of computed spatio-
temporal pressure evolutions we generated synthetic clouds of microseismicity. We analyzed spatio-
temporal characteristics of synthetic microseismicity and observed that nonlinear diffusion still leads to
a distinct and thus detectable back front. However, our results show that in most cases the character of
the back front significantly deviates from a square-root of time-dependent behavior. Only in the case
of 3-D synthetic seismicity based on exponential diffusion a t1/2 − dependent back front approximately
describes the evolution of post-injection induced events quite well. The corresponding estimates obtained
from the back front are very similar to the computed effective diffusivity values and heuristic estimates
determined by the triggering front. The comparison of the results obtained from the frozen and nonlinear
medium diffusivity model shows that the synthetic seismicity distributions and their corresponding spatio-
temporal characteristics are almost identical.

Acknowledgments
We thank the Federal Ministry for the Environment, Nature Conservation and Nuclear Safety (BMU) as
sponsor of the MAGS project and the sponsors of the PHASE consortium for supporting the presented
research.

References
Economides, M. J. and Nolte, K. G. (2000). Reservoir stimulation. John Wiley & Sons, Inc, New York.
Hummel, N. and Müller, T. M. (2009). Microseismic signatures of non-linear pore-fluid pressure diffu-
sion. Geophysical Journal International, 179(3):1558 – 1565.
Hummel, N. and Shapiro, S. A. (2011). Microseismic back front signatures for nonlinear fluid-rock
interaction. 7th Annual PHASE Report, pages 135 – 151.
Hummel, N. and Shapiro, S. A. (2012). Microseismic estimates of hydraulic diffusivity in case of non-
linear fluid-rock interaction. Geophysical Journal International, 188:1441 – 1453.
Parotidis, M., Shapiro, S. A., and Rothert, E. (2004). Back front of seismicity induced after termination
of borehole fluid injection. Geophysical Research Letters, 31(L02612).
Rothert, E. and Shapiro, S. A. (2003). Microseismic monitoring of borehole fluid injections: data model-
ing and inversion for hydraulic properties of rocks. Geophysics, 68(2):685 – 689.
Shapiro, S. A., Audigane, P., and Royer, J.-J. (1999). Large-scale in situ permeability tensor of rocks
from induced microseismicity. Geophysical Journal International, 137:207 – 213.
Shapiro, S. A. and Dinske, C. (2009). Scaling of seismicity induced by nonlinear fluid-rock interaction.
Journal of Geophysical Research, 114(B09307).
Shapiro, S. A., Huenges, E., and Borm, G. (1997). Estimating the crust permeability from fluid-injection-
induced seismic emission at the KTB site. Geophysical Journal International, 131(F15 – F18).
Shapiro, S. A., Rothert, E., Rath, V., and Rindschwentner, J. (2002). Characterization of fluid transport
properties of reservoirs using induced microseismicity. Geophysics, 67(1):212 – 220.

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

172

Influence of Nonlinear Fluid-rock Interaction on the Back Front of Seismicity

Annual Report 2012 173

Permeability Tensor Characterization from the Back Front of

Seismicity

N. Hummel1 and S. A. Shapiro1

email: [email protected]
1
Freie Universität Berlin, FR Geophysik

Summary
After termination of borehole fluid injections seismic activity can still be observed. Previously, this phe-
nomenon has been analyzed assuming that the main triggering mechanism is governed by a linear pore
pressure diffusion. The so-called back front of seismicity has been introduced to characterize the seis-
micity evolution of the post injection phase. It allows to determine a scalar hydraulic diffusivity estimate
from the spatio-temporal evolution of induced events. The back front describes the distance from the
injection point at which seismicity is terminated at a given time after the end of the fluid injection. How-
ever, fluid injections can considerably enhance the permeability which then becomes a function of the
pore pressure. Additionally, most rocks are elastically anisotropic which implies an anisotropic hydraulic
transport. In respect of the corresponding permeability tensor we explore what kind of diffusivity esti-
mates are provided by the back front. For this we consider seismicity recorded during the borehole fluid
injection experiment at Ogachi (Japan) and Fenton Hill (New Mexico). We apply a scaling approach
which is based on a model of a factorized anisotropic pressure dependence of permeability. With this
scaling approach we transform clouds of hypocenters of events obtained in a hydraulically anisotropic
medium into a cloud which would be obtained in an equivalent isotropic but still nonlinear medium. We
then apply the back front concept which provides a certain diffusivity estimate. This diffusivity estimate
is used together with scaling factors from the transformation approach to reconstruct the corresponding
diffusivity tensor. Our results show that the back front provides a diffusivity value which corresponds to
the smallest diffusivity tensor element.

Introduction
Borehole fluid injections are a common pre-requisite for the development of geothermal (see for example
Majer et al., 2007) and hydrocarbon reservoirs (Economides and Nolte, 2000). Usually such hydraulic
stimulation treatments are accompanied by microseismic activity. The occurrence of induced seismicity
is not only limited to the injection phase. Even after termination of the fluid injection seismic activity can
be monitored, as for instance at Barnett Shale (Shapiro and Dinske, 2007), Basel (Häring et al., 2008;
Dinske et al., 2008), the Canyonsands (Fischer et al., 2008), Cooper Basin (Baisch et al., 2006; Haeckel
et al., 2011), Cotton Valley (Urbancic et al., 1999; Shapiro et al., 2006), Fenton Hill (Fehler et al., 1998),
Soultz-sous-Forêts (Dyer et al., 1994). Assuming that pore-fluid pressure diffusion is the main triggering
mechanism of fluid injection induced seismicity, it has been shown that hydraulic transport properties can
be determined from the rate of growth of the microseismic cloud. For the injection phase this hydraulic
characterization can be done with the concept of the triggering front (Shapiro et al., 1997, 1999, 2002).
The triggering front approximately describes the outermost envelope of induced seismicity in the r − t
domain. Here, r describes the radial distance of events from the injection point and t corresponds to the
occurrence time of events. For a homogeneous and isotropic medium the triggering front is given by
p
r = 4 π D0 t , (1)
where the constant with time and pressure hydraulic transport is defined by the scalar hydraulic diffusivity
D0 . Following this philosophy a similar method has been developed for the post injection phase. This

Permeability Tensor Characterization from the Back Front of Seismicity

174

method uses the so-called back front of seismicity. After termination of the fluid injection an area of
seismic quiescence can evolve in the r − t domain. This effect can be explained assuming that seismic
events only can be triggered for increasing pore pressures (Parotidis et al., 2004). Despite shut-in the pore
pressure is still able to increase as a function of time and distance from the injection point. This increase
in pressure continues as long as the pressure reaches its maximum. Thereafter, the pressure starts to
decrease. In turn this pressure decrease results in an increasing effective normal stress which leads to a
strengthening of induced cracks and fractures. Therefore, seismic events can not be triggered at locations
where the pressure has already reached its maximum. The back front describes this distances from the
injection point r at which seismicity is terminated at a given time t after the end of the fluid injection
(Parotidis et al., 2004): s    
t t
r = 2 d D0 t − 1 ln . (2)
t0 t − t0
Here, d describes dimensionality. The back front allows to determine a scalar hydraulic diffusivity es-
timate D0 from the spatio-temporal evolution of induced events. For a time- and pressure-independent
hydraulic transport in an isotropic medium triggering front and back front provide the same diffusivity
estimate. Additionally, for the case of a classical one dimensional hydraulic fracture the back front al-
lows to characterize the hydraulic transport of the fractured domain (Shapiro et al., 2006). Taking into
account the dynamic viscosity of the reservoir fluid µ and the poroelastic compliance S (uniaxial storage
coefficient) the diffusivity estimates can be converted into permeability estimates κ (Jaeger et al., 2007):
κ −1
D= S . (3)
µ
We note that the poroelastic compliance is related to porosity and different bulk moduli for the pore-fluid,
the rock skeleton, the grain material as well as the Biot coefficient (see for example Shapiro et al., 2002).
However, fluid injections like for example hydraulic fracturing treatments can considerably enhance the
permeability (Economides and Nolte, 2000). In this case, permeability becomes a function of the pressure
which results in a nonlinear pressure diffusion (Shapiro and Dinske, 2009; Hummel and Müller, 2009).
For such situations Hummel and Shapiro (2012) show that the triggering front provides diffusivity esti-
mates of an effective diffusivity which characterizes the medium after stimulation including a possible
hydraulic fracturing of the rock. In addition to a nonlinear hydraulic transport most rocks are elasti-
cally anisotropic. Such a nature controls the distribution of microseismicity which is not only caused by
an anisotropic permeability. For instance, an anisotropic stress distribution is strongly influenced by an
anisotropic elasticity and strength of the rock (see e.g. Langenbruch and Shapiro, 2012). However, usu-
ally, the anisotropy of hydraulic permeability is much stronger than the anisotropy of elastic properties.

Here, we explore what kind of diffusivity estimates are provided by the back front in respect of the corre-
sponding permeability tensor. Since the back front has been derived for a hydraulically isotropic medium
we apply a microseismic data transformation. This transformation approach is based on a model of a
factorized anisotropic pressure dependence of permeability. Hence, it accounts for a hydraulic anisotropy
and a pressure-dependent permeability. With this scaling approach we transform clouds of hypocenters
of events obtained in a hydraulically anisotropic nonlinear medium into a cloud which would be obtained
in an equivalent isotropic nonlinear medium. We apply this scaling approach to seismicity recorded dur-
ing the borehole fluid injection experiment at Ogachi (Japan) and Fenton Hill (New Mexico). We apply
both the triggering front and the back front to determine a hydraulic diffusivity. This diffusivity estimate
is used together with scaling factors from the transformation approach to reconstruct the corresponding
diffusivity tensor. Our results show that the back front provides a diffusivity value which corresponds to
the smallest diffusivity tensor element.

Transformation of the microseismic event cloud

Most rocks are elastically anisotropic which implies an anisotropic hydraulic transport. Since the back
front has been introduced for hydraulically isotropic media we have to account this anisotropy.
One way has been proposed by Shapiro et al. (1999) for the case of linear diffusion. Their idea is to replace

Permeability Tensor Characterization from the Back Front of Seismicity

Annual Report 2012 175

the entire heterogeneous seismically active rock volume by an effective homogeneous medium with an
upscaled anisotropic permeability tensor. For the case of such an effective homogeneous anisotropic
medium the diffusional relaxation of the pore pressure perturbation in the principal coordinate system of
the diffusivity tensor D is described by:

∂p ~ D∇
~ p.
=∇ (4)
∂t
Here, D is a constant diagonal matrix with corresponding diffusivity tensor components D011 , D022 and
D033 . Further they introduce a new system of coordinates [x′ , y ′ ,z ′ ]:
x y z
x′ = p , y′ = p , z′ = p , (5)
3 D011 /tr(D) 3 D022 /tr(D) 3 D033 /tr(D)

with tr(D) = D011 + D022 + D033 . Substituting these new coordinates into equation 4 results in an
isotropic diffusion equation. The corresponding hydraulic transport is controlled by the mean value of
the diagonal diffusivity tensor elements tr(D)/3. In this way their scaling approach transforms a cloud of
seismic events from one corresponding effective hydraulically homogeneous anisotropic medium into a
microseismic cloud corresponding to an equivalent hydraulically homogeneous isotropic medium. How-
ever, their main assumption is a time- and pressure independent hydraulic diffusivity.
In the following we extend this approach for the more general case where hydraulic transport proper-
ties can be functions of the pressure. For such situations the hydraulic transport results in a nonlin-
ear diffusion. Consequently, our equations differ from the one of Shapiro et al. (1999). For the event
cloud transformation we consider a three dimensional nonlinear hydraulically anisotropic medium. The
spatio-temporal evolution of the pressure perturbation in the principal coordinate system of the hydraulic
diffusivity tensor can be described by
∂p ~ D(p) ∇ ~ p.
=∇ (6)
∂t
The pressure-dependent diffusivity tensor D(p) is a diagonal matrix with corresponding tensor compo-
nents  
D11 (p) 0 0
D(p) =  0 D22 (p) 0 . (7)
0 0 D11 (p)
Actually, there are two different pressure-dependent diffusivity models under consideration. A power-
law dependence of diffusivity on the pressure is appropriate for rocks having an almost negligible initial
hydraulic diffusivity such as shales (Shapiro and Dinske, 2009). In contrast, an exponential dependence
of diffusivity on the pressure is applicable for rocks which have an in-situ and non-negligible initial
diffusivity such as granitic rocks (Hummel and Müller, 2009; Hummel and Shapiro, 2012). Both models
effectively address the impact of a massive stimulation or a growing hydraulically fractured domain on the
pressure diffusion and associated microseismicity. Pre-existing joints or planes of weakness can be almost
or completely closed at a zero injection pressure. Consequently, their contribution to the diffusivity is very
low or zero. In contrast, an increase in pressure can cause these joints to open which then contribute to an
enhanced hydraulic transport. This type of behavior is described by the pressure-dependent diffusivity.
Both these models include the normalizing constant D0 which describes the hydraulic transport in the
limit of a zero pressure (i.e. time- and pressure-independent hydraulic transport). In the following, we
suppose that for all three components of the principal coordinate system the dependence of hydraulic
diffusivity on the pressure perturbation is the same. We furthermore assume that anisotropy only affects
the parameter D0 . Hence, the corresponding diffusivity tensor is given as
 
D011 f (p) 0 0
D(p) =  0 D022 f (p) 0 . (8)
0 0 D033 f (p)

For the case of a linear diffusion in a hydraulically anisotropic medium the hydraulic transport is con-
trolled by the diffusivity tensor elements D011 , D022 and D033 (according to equation 4). Consequently,

Permeability Tensor Characterization from the Back Front of Seismicity

176

the case of a nonlinear diffusion in a hydraulically anisotropic medium can be described by decomposing
the diffusivity tensor (equation 8) into a pressure-dependent part and an anisotropic part:
 
D011 0 0
D(p) =  0 D022 0  f (p) , (9)
0 0 D033
where generally, f (p) can be an arbitrary functional dependence on pressure (for instance power-law or
exponential dependence). The anisotropic part controls the directivity of the stimulation and of the hy-
draulic transport while the pressure-dependent part describes the impact of a hydraulic stimulation on the
permeability. However, for field microseismic data, it is very difficult to determine the functional depen-
dence f (p). Therefore, we following Hummel and Shapiro (2012) and consider a heuristic diffusivity.
This heuristic diffusivity effectively describes the corresponding enhanced pressure-dependent hydraulic
transport in a replacing homogeneous medium:
 
Dh11 0 0
D≈ 0 Dh22 0 . (10)
0 0 Dh33
Similar to the linear diffusion transformation approach (Shapiro et al., 1999) we factorize the anisotropic
diffusivity by introducing new coordinates [x′ , y ′ , z ′ ]
s s s
D D D
x′ = x , y′ = y , z′ = z , (11)
Dh11 Dh22 Dh33

where p
D= 3
Dh11 Dh22 Dh33 . (12)
Note that there is no principal difference between 5 and 11. The only difference is the choice in the
constant of normalization D. Substituting these new coordinates into equation 6 results in an isotropic
nonlinear diffusion equation (further details are given in Hummel and Shapiro, 2011). In other words, we
approximate the real medium by a model with a nonlinearity and a factorized anisotropy. Although the
anisotropic permeability depends on the geological structure and the strength of the rock, we assume that
the type of pressure dependence of permeability is independent of a specific direction.
To apply this model of a factorized anisotropy (equation 11) to microseismic data we first have to iden-
tify the directivity of the hydraulic transport. To reveal this directivity and the influence of hydraulic
anisotropy on the distribution of induced events we consider the particular shape of the seismicity cloud.
We determine the characteristic lengths L from the geometry which the seismicity cloud has at the end of
the injection phase. For this we carry out a principal component analysis and rotate the event cloud into
the principal coordinate system (PCS). For the corresponding principal X- Y - and Z direction the charac-
teristic lengths are Lmax , Lint and Lmin . Since very large outliers are not further considered we note that
these characteristic lengths still represent upper limits of the seismically stimulated rock volume. Conse-
quently, events which may have been triggered by static stress changes might be also included. Note also
that the characteristic lengths are site specific. Their decreasing order must be linked not necessarily to
the respective axes X, Y and Z. For instance, situations in which the Lmax corresponds to the Y or Z
axis can not be excluded. In any case, these characteristic lengths are used to determine scaling factors
FXscal , FYscal , FZscal according to
Lmax Lint Lmin
FXscal = , FYscal = , FZscal = , (13)
Lscal Lscal Lscal
√
where Lscal = 3 Lmax · Lint · Lmin . Basically, these scaling factors correspond to the inverse of the
square root factors given in equation 11. Although the factors of equation 13 depend on the final shape
of the cloud we do not only transform seismicity which is induced at the end of the injection phase. In
contrast, we scale the coordinates of all events j which have been induced during the complete fracturing
treatment according to (compare with equations 11)
Xj Yj Zj
Xtransf. j = , Ytransf. j = , Ztransf. j = . (14)
FXscal FYscal FZscal

Permeability Tensor Characterization from the Back Front of Seismicity

Annual Report 2012 177

With these new locations Xtransf , Ytransf , Ztransf the event cloud is transformed into a cloud which
would result in a hydraulically nonlinear equivalent isotropic medium. In the end, information about the
directionality of the hydraulic transport can be obtained from the scaling factors. The combination of
equations 11 and 14 allows to reconstruct the factorized diffusivity tensor according to
Dh11 = Fx2 · D
Dh22 = Fy2 · D (15)
Dh33 = Fz2 · D.

Application to real data

Ogachi, Japan
The purpose of the borehole fluid injection experiment at the Ogachi site, Japan, was to artificially create
a shallow geothermal reservoir as a Hot Dry Rock (HDR) system (Kitano et al., 2000; Ito, 2003; Kaieda
et al., 2000, 2005, 2010). For this, a 1000 m deep injection well (OGC-1) was drilled almost vertically
into the naturally fractured granitic basement. The last 10 m of the borehole where left uncased in order
to have an open hole section from 990 m to 1000 m (Figure 1). In 1991, pressurized water was injected
into the rock for almost 11 days. Flow rates were increases step-wise up to 13 l/s and caused wellhead
pressures of at most 20 MPa. Seismicity was recorded with a near-surface network deployed around the
injection well. The monitoring system consisted of a 10 station network of three component geophones
which where installed in shallow boreholes at 30 m to 50 m. Additionally, a single three component
geophone was installed in the observation well at a depth of 380 m. As a result of the fluid injection 1550
events have been located in a depth of about 700 m to 1300 m. At first, induced events are distributed
around the injection well with a most southern extend of almost 250 m. With ongoing injection the
hypocenter distribution propagates NNE.

Figure 1: Distribution of 1550 induced events recorded during the 1991 stimulation experiment at the Ogachi site,
Japan. Data are courtesy of H. Kaieda (Central Research Institute of Electric Power Industry, CRIEPI)

Permeability Tensor Characterization from the Back Front of Seismicity

178

Figure 2: Temporal evolution of the seismicity cloud in the principal coordinate system. Vertical and horizontal
scales are equal. The double arrows indicate the characteristic lengths of the seismicity cloud. These dimensions are
used to reveal the directivity of hydraulic anisotropy and to transform the event cloud into one which would have
occurred in an equivalent isotropic but still nonlinear medium.

Figure 3: Temporal evolution of the transformed event cloud which would be obtained in a hydraulically equivalent
isotropic medium. Vertical and horizontal scales are equal and the same as in Figure 2. With increasing time T the
seismicity cloud increases in size and number of induced events while its geometry remains almost the same. Note
that for all three snapshots the scaling was taken the same.

Permeability Tensor Characterization from the Back Front of Seismicity

Annual Report 2012 179

We next apply our scaling approach to transform the cloud of hypocenters of events obtained in a hy-
draulically anisotropic nonlinear medium into a cloud which would be obtained in an equivalent isotropic
but still nonlinear medium. For this we consider the seismicity cloud in the principal coordinate system
(PCS). To reveal the influence of hydraulic anisotropy we evaluate the characteristic lengths L of the
seismicity cloud. In the PCS of the cloud we obtain Lmax = 1142 m for the X− direction, Lint = 640 m
for the Y − direction and Lmin = 490 m for the Z− direction (Figure 2). These dimensions are used
to determine scaling factors according to equation 13. With these scaling factors we finally transform
the seismicity cloud into a cloud which would result in an equivalent hydraulically isotropic nonlinear
medium. Figure ( 3) shows how well the rescaling obtained from the final shape of the cloud works
during the total process time. As the transformed event cloud increases in size and number of induced
events, their geometry remains almost the same. This supports our model of a nonlinear permeability
with a factorized anisotropy.

In the following we analyze spatio-temporal characteristics of both the original and the transformed event
clouds (Figure 4). For this we compute the corresponding r − t plot and apply the triggering front and the
backfront. Firstly, we consider the original cloud and fit the triggering front (equation 1) as envelope of
injection phase seismicity. As a result we obtain a diffusivity estimate of 0.08 m2 /s. With the termination
of the fluid injection an area of seismic quiescence evolves in the r − t plot. For this post injection phase

16 20

Pressure [MPa]
Flow rate [l/s]

8 10

Pressure Injection rate

0 0
0 2 4 6 8 10 12
r−t plot
1500
Dh = 0.08 m2/s Dbf = 0.02 m2/s
Distance [m]

1000

500

0
0 2 4 6 8 10 12
r−t plot of transformed seismicity
1000 2 2
Dh = 0.04 m /s Dbf = 0.03 m /s
Distance [m]

500

0
0 2 4 6 8 10 12
Time [d]

Figure 4: Top: Engineering data showing wellhead pressure and injection rate. Middle: r − t plot of induced
seismicity being recorded during and after the fluid-injection experiment. Triggering front and back front provide
different hydraulic diffusivity estimates. Bottom: r − t plot of seismicity which would be obtained in a hydraulically
equivalent isotropic medium. For such a medium, triggering front and back front provide almost the same diffusivity
estimate.

Permeability Tensor Characterization from the Back Front of Seismicity

180

we fit the back front (equation 2) as envelope of the domain where no seismic events occur. This provides
a smaller diffusivity estimate of 0.02 m2 /s. Evidently, both diffusivity estimates are different although
they have the same order of magnitude.
Secondly, we consider the transformed hypocenter cloud which would result in an equivalent hydrauli-
cally isotropic but still nonlinear medium. For such a medium the heuristic diffusivity estimates provided
by the triggering front and back front characterize the effective hydraulic transport. As in the case of a
time- and pressure-independent diffusion we expect both values to be the same. Our results show that
this is almost confirmed. The triggering front provides a diffusivity estimate of 0.04 m2 /s. In contrast, the
back front yields a value of 0.03 m2 /s. Evidently, both estimates are almost identical.
Finally, information about directionality can be obtained from the scaling factors. They allow us to re-
build the factorized diffusivity. Following equation 15 the effective hydraulic diffusivity along the princi-
pal axes of the diffusivity tensor can be reconstructed. For the back front diffusivity estimate of 0.03m2/s
we obtain
D = diag(Dh11 ; Dh22 ; Dh33 ) = diag(0.08; 0.03; 0.01) m2/s . (16)
Within the hydraulically anisotropic nonlinear medium the fastest hydraulic transport takes place in the
11 direction. It is characterized by an effective diffusivity of 0.08 m2 /s. Exactly the same diffusivity is
obtained from fitting the triggering front as envelope of seismicity in the r − t plot. Ahead of this front
there are only a few events and the medium has practically not yet been pressurized by the diffusion.
Hence, the first events which occur at a particular time must have been triggered by the pressure diffusion
along the hydraulically fastest direction. Consequently, the triggering front provides a diffusivity estimate
which corresponds to the largest heuristically effective hydraulic diffusivity tensor component.
In contrast, the slowest hydraulic transport occurs in the 33 direction. It is characterized by a diffusivity
magnitude of 0.01 m2 /s. Since seismic events can not be triggered at locations where the pressure has
already reached its maximum, the latest possible events should be triggered by the slowest penetrating
pressure maximum. This hypothesis is confirmed by our results. In Figure 4 one can see that below
the back front parabola there is not one single seismic event. With 0.02 m2 /s this diffusivity estimate
provided by the back front almost coincides with the least principal effective diffusivity of 0.01 m2 /s.
Consequently, the back front provides a diffusivity estimate which corresponds to the least hydraulic dif-
fusivity tensor component.

Finally, we convert the diffusivity tensor components into corresponding permeability values. For this we
assume a uniaxial storage coefficient of S −1 = 4.36 · 1010 Pa and a dynamic fluid viscosity (at 240◦ C)
of µ = 1.2 10−4 Pa · s (see Audigane et al., 2002). Following equation 3 we obtain the reconstructed
permeability tensor
κ = diag(22; 8.3; 2.8) · 10−17 m2 . (17)
In comparison to the permeability tensor estimates given in Audigane et al. (2002)

κAudigane et al. = diag(23.7; 6.9; 5.0) · 10−17 m2 (18)

our results are in good agreement.

Fenton Hill, New Mexico

In the following we apply the developed theory to microseismic data recorded during the massive hy-
draulic fracturing (MHF) treatment carried out 1983 at the Fenton Hill hot dry rock site. The purpose of
the Phase II program was to create an artificial geothermal reservoir within a low permeability granitic
rock (House, 1987; Fehler, 1989; Fehler et al., 1998). The idea of the fluid injection experiment was to
create several vertical fractures which should then establish a hydraulic connection between an injection
well and a production well. Following this idea, two wells where drilled into the highly jointed granitic
basement. They both had a 35◦ inclined open hole section which were only 380 m apart vertically.
In December 1983 pressurized water was injected for almost 61 h into the rock through an isolated section
of the wellbore. This injection interval extended from the casing shoe at a depth of 3531 m (11585 ft) to
a sand-barite plug top at a depth of 3550 m (11648 ft). During this stimulation experiment almost 21600
m3 of water was injected with wellhead pressures of about 48 MPa (Dreesen and Nicholson, 1985). As

Permeability Tensor Characterization from the Back Front of Seismicity

Annual Report 2012 181

Figure 5: Distribution of 11362 induced events recorded during the December 1983 massive hydraulic fracturing
treatment at Fenton Hill, New Mexico. Data are courtesy of M. Fehler (Los Alamos National Laboratory).

a result of the fluid injection more than 11300 microseismic events have been recorded and located with
accuracies of 20 m to 30 m (House, 1987). The monitoring system consisted of two different types of
geophones. Groups of three component geophones have been installed in three different boreholes in
depths of 2400 m, 2850 m and 3300 m. Vertical component geophones have been deployed in two shal-
low boreholes at depths of 570 m and 820 m (House, 1987). However, during the entire experiment there
where disruptions in the seismic network. Although the monitoring system was at least partially in oper-
ation there are three time periods where events have not been located. These time periods include the first
five hours, the time between 50 h and 55 h as well as the time between 30 h and about 40 h. In the latter
case no events have been located because there were only three stations in operations (Fehler et al., 1998).

Figure 5 shows the distribution of induced events together with the injection interval and a schematic
borehole (since we have no exact information about the borehole geometry and orientation). The ma-
jority of events has been triggered within a quite limited rock volume around the injection point. To
the outside, the number of diffusively distributed events decreases. The vertical extend of the seismicity
cloud includes a depth range from almost 3000 m to 4000 m. The map view projection of the hypocenter
distribution shows that the orientation of the cloud strikes NNW.
We next apply our scaling approach to transform the cloud of hypocenters of events obtained in a hy-
draulically anisotropic nonlinear medium into a cloud which would be obtained in an equivalent isotropic
but still nonlinear medium. For this we consider the seismicity cloud in the principal coordinate system
(PCS). To reveal the influence of hydraulic anisotropy we evaluate the characteristic lengths L of the seis-
micity cloud. We obtain Lmax = 1033 m, Lint = 791 m and Lmin = 504 m (Figure 6). We note that these
dimensions correspond to the volume in which most events have been induced. Very large outliers are
not further considered. These dimensions are used to determine scaling factors according to equation 13.
With these scaling factors we finally transform the seismicity cloud into a cloud which would result in an
equivalent hydraulically isotropic nonlinear medium. Figure ( 7) shows how well the rescaling obtained
from the final shape of the cloud works during the total process time. As the transformed event cloud
increases in size and number of induced events, their geometry remains almost the same.

Permeability Tensor Characterization from the Back Front of Seismicity

182

Figure 6: Temporal evolution of the seismicity cloud in the principal coordinate system. Vertical and horizontal
scales are equal. The double arrows indicate the characteristic lengths of the seismicity cloud. These dimensions are
used to reveal the directivity of hydraulic anisotropy and to transform the seismicity cloud into a cloud which would
have occurred in an effective isotropic medium.

Figure 7: Temporal evolution of the transformed event cloud which would be obtained in a hydraulically equivalent
isotropic medium. Vertical and horizontal scales are equal and the same as in Figure 6. With increasing time T the
seismicity cloud increases in size and number of induced events while its geometry remains almost the same. Note
that for all three snapshots the scaling was taken the same.

Permeability Tensor Characterization from the Back Front of Seismicity

Annual Report 2012 183

In the following we analyze spatio-temporal characteristics of both the original and the transformed event
clouds (Figure 8). For this we compute the corresponding r − t plot and apply the triggering front and the
backfront. Firstly, we consider the original cloud and fit the triggering front (equation 1) as envelope of
injection phase seismicity. As a result we obtain a diffusivity estimate of 0.16 m2 /s. With the termination
of the fluid injection an area of seismic quiescence evolves in the r − t plot. For this post injection phase
we fit the back front (equation 2) as envelope of the domain where no seismic events occur. This provides
a smaller diffusivity estimate of 0.07 m2 /s. Evidently, both diffusivity estimates are different although
they have the same order of magnitude.
Secondly, we consider the transformed hypocenter cloud which would result in an equivalent hydrauli-
cally isotropic but still nonlinear medium. For such a medium the heuristic diffusivity estimates provided
by the triggering front and back front characterize the effective hydraulic transport. As in the case of a
time- and pressure-independent diffusion we expect both values to be the same. Our results show that
this is almost confirmed. The triggering front provides a diffusivity estimate of 0.11 m2 /s. In contrast, the
back front yields a value of 0.07 m2 /s.
Finally, information about directionality can be obtained from the scaling factors. They allow us to re-
build the factorized diffusivity. Following equation 15 the effective hydraulic diffusivity along the princi-
pal axes of the diffusivity tensor can be reconstructed. For the back front diffusivity estimate of 0.07m2/s
we obtain
D = diag(Dh11 ; Dh22 ; Dh33 ) = diag(0.14; 0.08; 0.03) m2 /s (19)
Within the hydraulically anisotropic nonlinear medium the fastest hydraulic transport takes place in the
11 direction. It is characterized by an effective diffusivity of 0.14 m2 /s. Fitting the triggering front as
envelope of seismicity in the r − t plot provides almost the same diffusivity estimate of 0.16 m2 /s. Ahead
of this front there are only a few events and the medium has practically not yet been pressurized by

150 60 Pressure [MPa]

Flow rate [l/s]

100
30
50
Treating pressure Flow rate
0 0
0 20 40 60 80
r−t plot
1000 2 2
D = 0.16 m /s D = 0.07 m /s
Distance [m]

h bf

500

0
0 20 40 60 80
r−t plot of transformed seismicity
1000 2 2
D = 0.11 m /s D = 0.07 m /s
Distance [m]

h bf

500

0
0 20 40 60 80
Time [h]

Figure 8: Top: Engineering data showing wellhead pressure and injection rate. Middle: r − t plot of induced
seismicity being recorded during and after the fluid-injection experiment. Triggering front and back front provide
different hydraulic diffusivity estimates. Bottom: r − t plot of seismicity which would be obtained in a hydraulically
equivalent isotropic medium. For such a medium, triggering front and back front should provide the same diffusivity
estimate. However, we observe a small difference in the corresponding diffusivity value.

Permeability Tensor Characterization from the Back Front of Seismicity

184

the diffusion. Hence, the first events which occur at a particular time must have been triggered by the
pressure diffusion along the hydraulically fastest direction. Consequently, the triggering front provides
a diffusivity estimate which corresponds to the largest heuristically effective hydraulic diffusivity tensor
component.
In contrast, the slowest hydraulic transport occurs in the 33 direction. It is characterized by a diffusivity
magnitude of 0.03 m2 /s. However, for the event cloud in the PCS, the back front provides a diffusivity
estimate of 0.07 m2 /s. This estimate is closer to the reconstructed intermediate diffusivity tensor compo-
nent of 0.08 m2 /s, but still smaller. Additionally, Figure 8 shows that in both r − t plots there are some
events below the back front parabola. However, since seismic events can not be triggered at locations
where the pressure has already reached its maximum, the latest possible events should be triggered by the
slowest penetrating pressure maximum. Thus, we conclude that the back front still provides a diffusivity
estimate which corresponds to the least hydraulic diffusivity tensor component.
Finally, we convert the diffusivity tensor components into corresponding permeability values. Following
equation 3 we compute corresponding permeability values. Assuming a uniaxial storage coefficient of
S −1 = 1.68 · 1011 Pa and a dynamic viscosity of the reservoir fluid (salt water at 150◦ C) of µ = 1.9 10−4
Pa · s (Shapiro et al., 2002) we compute the reconstructed permeability tensor to

κ = diag(15.8; 9.0; 3.4) · 10−17 m2 . (20)

The permeability magnitudes are in agreement with the ones computed from Shapiro et al. (2002).

Conclusion
Understanding the spatio-temporal characteristics of induced seismicity contributes to reservoir charac-
terization. Spatio-temporal features can provide important information about the interaction between
fluid and reservoir rock. Here we have analyzed spatio-temporal characteristics of induced seismicity
recorded during stimulation experiments at Ogachi, Japan and Fenton Hill, New Mexico. To approximate
the real medium we have considered a model with a nonlinear permeability and a factorized anisotropic
permeability where the pressure dependence itself is independent of a specific direction. For Ogachi and
Fenton Hill this factorized model seems to be confirmed by microseismic data. We have proposed a scal-
ing approach to transform clouds of microseismic events obtained in a hydraulically anisotropic nonlinear
medium into those which would be obtained in an equivalent isotropic but still nonlinear medium. For the
original and the transformed event cloud we have applied both the triggering front and the back front. For
Ogachi as well as Fenton Hill we have reconstructed the diffusivity tensor for a hydraulic transport in an
anisotropic nonlinear medium. Our results show that the triggering front provides estimates of the diffu-
sivity which correspond to the largest diffusivity tensor component. In contrast, the diffusivity estimates
obtained from the back front correspond to the smallest diffusivity tensor component.

Acknowledgments
We thank the sponsors of the PHASE consortium for supporting the research presented in this paper.

References
Audigane, P., Royer, J.-J., and Kaieda, H. (2002). Permeability characterization of the Soultz and Ogachi
large-scale reservoir using induced microseismicity. Geophysics, 67(1):204–211.

Baisch, S., Weidler, R., Vörös, R., Wyborn, D., and de Graaf, L. (2006). Induced Seismicity during the
Stimulation of a Geothermal HFR Reservoir in Cooper Basin, Australia. Bulletin of the Seismological
Society of America, 96(6):2242 – 2256.

Dinske, C., Taranczewski, T., Shapiro, S. A., and Ladner, F. (2008). Analysis of Induced Microseismicity
of a Geothermal Reservoir Stimulation in Basel, Switzerland. 4th Annual PHASE Report.

Permeability Tensor Characterization from the Back Front of Seismicity

Annual Report 2012 185

Dreesen, D. S. and Nicholson, R. W. (1985). Well completion and operations for MHF of Fenton Hill
HDR Well EE-2. Trans. Geotherm. Res. Coun., 9(II):89 – 94.

Dyer, B., Juppe, A., and Jones, R. (1994). Microseismic results from the European HDR Geothermal
Project at Soultz-sous-Forets, Alsace, France. IR03/24, CSM associated Ltd.

Economides, M. J. and Nolte, K. G. (2000). Reservoir stimulation. John Wiley & Sons, Inc, New York.

Fehler, M., House, L., Phillips, W. S., and Potter, R. (1998). A method to allow temporal variation
of velocity in travel-time tomography using microearthquakes induced during hydraulic fracturing.
Tectonophysics, 289:189 – 201.

Fehler, M. C. (1989). Stress Control of Seismicity Patterns Observed During Hydraulic Fracturing Exper-
iments at the Fenton Hill Hot Dry Rock Geothermal Energy Site, New Mexico. International Journal
of Rock Mechanics and Mining Sciences & Geomechanics Abstracts, 26(34):211 – 219.

Fischer, T., Hainzl, S., Eisner, L., Shapiro, S. A., and Le Calvez, J. (2008). Microseismic signatures of
hydraulic fracture growth in sediment formations: Observations and modeling. Journal of Geophysical
Research, 113(B02307).

Haeckel, A., Dinske, C., and Shapiro, S. A. (2011). SBRC of a Geothermal Reservoir in Cooper Basin,
Australia. 7th Annual PHASE Report, pages 189–201.

Häring, M. O., Schanz, U., Ladner, F., and Dyer, B. C. (2008). Characterisation of the Basel 1 enhanced
geothermal system. Geothermics, 37:469 – 495.

House, L. (1987). Locating microearthquakes induced by hydraulic fracturing in crystalline rock. Geo-
physical Research Letters, 14(9):919 – 921.

Hummel, N. and Müller, T. M. (2009). Microseismic signatures of non-linear pore-fluid pressure diffu-
sion. Geophysical Journal International, 179(3):1558 – 1565.

Hummel, N. and Shapiro, S. (2012). Microseismic estimates of hydraulic diffusivity in case of non-linear
fluid-rock interaction. Geophysical Journal International, 188:1441–1453.

Hummel, N. and Shapiro, S. A. (2011). Microseismic back front signatures for nonlinear fluid-rock
interaction. 7th Annual PHASE Report, (7):135 – 151.

Ito, H. (2003). Inferred role of natural fractures, veins, and breccias in development of the artificial
geothermal reservoir at the Ogachi Hot Dry Rock site, Japan. Journal of Geophysical Research,
108(B9).

Jaeger, J., Cook, N., and Zimmerman, R. (2007). Fundamentals of Rock Mechanics. Wiley-Blackwell, 4
edition.

Kaieda, H., Ito, H., Kiho, K., Suzuki, K., Suenga, H., and Shin, K. (2005). Review of the Ogachi HDR
Project in Japan. Proceedings of World Geothermal Congress, Antalya, Turkey, 24–29 April, 2005.

Kaieda, H., Jones, R., Moriya, H., Sasaki, S., and Ushijima, K. (2000). Ogachi HDR Reservoir Evalua-
tion by AE and Geophysical Methods. Proceedings of World Geothermal Congress, Kyushu–Tohoku,
Japan, May 28–June 10, 2000.

Kaieda, H., Sasaki, S., and Wyborn, D. (2010). Comparison of Characteristics of Micro-Earthquakes
Observed During Hydraulic Stimulation Operations in Ogachi, Hijiori and Cooper Basin HDR Projects.
Proceedings of World Geothermal Congress, Bali, Indonesia, 25–29 April, 2010.

Kitano, K., Hori, Y., and Kaieda, H. (2000). Outline of the Ogachi HDR Project and Character of the
Reservoir. Proceedings of World Geothermal Congress 2000, Kyushu–Tohoku, Japan, May 28–June
10, 2000.

Permeability Tensor Characterization from the Back Front of Seismicity

186

Langenbruch, C. and Shapiro, S. A. (2012). Modelling of Fracture Strength Distribution in Elastically

Heterogeneous Rocks. Proceedings, European Association of Geoscientists & Engineers Annual Con-
ference, Paper C043, (C043).
Majer, E. L., Baria, R., Stark, M., Oates, S., Bommer, J., Smith, B., and Asanuma, H. (2007). Induced
seismicity associated with Enhanced Geothermal Systems. Geothermics, 36:185–222.
Parotidis, M., Shapiro, S., and Rothert, E. (2004). Back front of seismicity induced after termination of
borehole fluid injection. Geophysical Research Letters, 31(L02612).
Shapiro, S., Huenges, E., and Borm, G. (1997). Estimating the crust permeability from fluid-injection-
induced seismic emission at the KTB site. Geophysical Journal International, 131(F15 - F18).
Shapiro, S. A., Audigane, P., and Royer, J.-J. (1999). Large-scale in situ permeability tensor of rocks
from induced microseismicity. Geophysical Journal International, 137:207–213.
Shapiro, S. A. and Dinske, C. (2007). Induced seismicity in Barnett Shale: a volumetric hydraulic
fracturing. 3rd Annual PHASE Report, pages 13 – 21.
Shapiro, S. A. and Dinske, C. (2009). Scaling of seismicity induced by nonlinear fluid-rock interaction.
Journal of Geophysical Research, 114(doi:10.1029/2008JB006145).
Shapiro, S. A., Dinske, C., and Rothert, E. (2006). Hydraulic-fracturing controlled dynamics of micro-
seismic clouds. Geophysical Research Letters, 33(L24321).
Shapiro, S. A., Rothert, E., Rath, V., and Rindschwentner, J. (2002). Characterization of fluid transport
properties of reservoirs using induced microseismicity. Geophysics, 67:212 – 220.
Urbancic, T., Shumila, V., and Rudledge, J. (1999). Determining hydraulic fracture behavior using mi-
croseismicity. Vail Rock ’99, 37th U.S. Rock Mechanics Symposium, Vail, Colorado.

Permeability Tensor Characterization from the Back Front of Seismicity

Annual Report 2012 187

Members and Sponsors

188
Annual Report 2012 189

PHASE Research Team

Professor

Serge A. Shapiro
received his Diploma (1982) from the Moscow State Uni- Freie Universität Berlin
versity and Ph.D. (1987) from the Moscow Research In- Malteserstr. 74-100, Build. D, Room 146
stitute VNIIGeosystem, both in Geophysics. From 1982 D-12249 Berlin
to 1992, he worked for the VNIIGeosystem as a research Tel.: ++49 (0)30 838 70839
scientist. In 1991-1997 he was a research scientist at the Fax.: ++49 (0)30 838 70729
Geophysical Institute of Karlsruhe University, Germany. email: [email protected]
During this time he received an Alexander von Humboldt
scholarship and a Heisenberg professorship. From 1997
till 1999, he was a professor of Applied Geophysics at the
School of Geology of the Polytechnic Institute in Nancy,
France, where he cooperated with the GOCAD consor-
tium. Since 1997 till 2006 he was one of Principal Inves-
tigators of the WIT university consortium. Since Febru-
ary 1999, he has been a full professor of Geophysics at
the Freie Universität Berlin. Since 2004 he has been the
Research Director of the PHASE university consortium.
His interests include the physics of seismicity, exploration
seismology, rock physics, and forward and inverse scatter-
ing problems. In 2002 he received the Best Paper in Geo-
physics Award of the SEG. In 2004 he was elected a Fel-
low of The Institute of Physics (UK). Memberships: SEG,
EAGE, AGU, and German Geophysical Society (DGG).

PHASE Research Team

190

Scientific Staff & Postdocs

Beatrice Cailleau
Has studied Mechanics in 1992-1996 and received a Freie Universität Berlin
diploma in Geophysics from the University of Grenoble, Malteserstr. 74-100, Build. D, Room 136
France, in 1997. She worked one year in computer pro- D-12249 Berlin
gramming in the company Cap Gemini Paris, and then Tel.: ++49 (0)30 838 70905
joined the Graduate School of Geomar - Kiel University Fax.: ++49 (0)30 838 70729
to specialise in finite and boundary element modelling ap- email: [email protected]
plied to Geosciences. In 2003, she obtained a PhD in Geo-
physics focusing on the numerical modelling of volcano-
tectonics interaction. As a research scientist, she spent 2
years at the Geodesy Lab of Miami, USA, and two years
at the Geodynamics Group of GFZ Potsdam to work on
plate margin deformation of Central and South America.
Since 2008, she is at Free University of Berlin investi-
gating the role of fluid and stress coupling in earthquake
triggering by means of finite element modelling.

Carsten Dinske
received his diploma in Geophysics from the University of Freie Universität Berlin
Potsdam in 2004. Before he became a research associate Malteserstr. 74-100, Build. D, Room 141
at Freie Universität in January 2005 he was a research as- D-12249 Berlin
sistance at GFZ Potsdam supporting the KTB seismolog- Tel.: ++49 (0)30 838 70175
ical monitoring group during a long-term fluid injection Fax.: ++49 (0)30 838 70729
experiment. In 2011, he finished his PhD studies and re- email: [email protected]
ceived a doctoral degree. His work mainly focuses on im-
proved understanding of injection-induced seismicity re-
lated phenomena. He is a member of AGU, EAGE and
SEG.

Oliver Krüger
Oliver joined the group in 2000 and got his degree in geo- Freie Universität Berlin
physics 2002. From 2000 to 2009 finite difference meth- Malteserstr. 74-100, Build. D, Room 130
ods were his scientific tools. With them he examined, D-12249 Berlin
among other things, the seismic wave propagation in gas Tel.: ++49 (0)30 838 70908
hydrates, reflection coefficients of fractured media and the Fax.: ++49 (0)30 838 70729
applicability of effective media theories. The estimation email: [email protected]
of the diffusivity of the underground rock on the basis of
spatial - temporal evolution of micro-earthquakes was his
research subject from 2009 to 2010. Since 2010, he inves-
tigates the magnitude frequency distribution of induced
micro-earthquakes.

PHASE Research Team

Annual Report 2012 191

Jörn Kummerow
received his diploma in Geophysics from Free Univer- Freie Universität Berlin
sity, Berlin, in 1998. In his diploma thesis, he worked Malteserstr. 74-100, Build. D, Room 112
on travel time tomography of two seismic profiles near D-12249 Berlin
Calama, Northern Chile. Between 1998 and 2003, he was Tel.: ++49 (0)30 838 70589
member of the TRANSALP Group at the GFZ Potsdam Fax.: ++49 (0)30 838 70729
and studied the crustal and upper mantle seismic structure email: [email protected]
of the Eastern Alps. He finished his PhD in 2003. Since
2003, he has been working in the KTB seismic monitor-
ing team focusing on fluid-induced microseismicity. In
August 2005 he joined the microseismicity group at the
Free University, Berlin and continues his research in seis-
mic source location, multiplet analysis as well as source
parameter determination. He is member of EAGE, EGU
and SSA.

Sibylle I. Mayr
studied Physics at the University of Stuttgart from 1988 - Freie Universität Berlin
1993 and worked 1992/93 on her diploma thesis at the In- Malteserstr. 74-100, Build. D, Room 105
stitute of Geophysics (Stuttgart). The subject of the theses D-12249 Berlin
was the ’Investigation of the elastic behaviour of porous Tel.: ++49 (0)30 838 70587
sediments at static stress’. In 1994/95 she worked at the Fax.: ++49 (0)30 838 70729
Institute of Geotechnical Engineering, Stuttgart. From email: [email protected]
1995 to 2008 she was Member of the Department of Ap-
plied Geophysics, Technical University Berlin (TUB). Till
2000 she was Assistant Lecturer and working as Scientific
Employee on the ultrasonic behaviour of sandstones un-
der hydrostatic pressure. Afterwards she was employed
on an ’Elektro Kinetic Sounding’ Project till 2002. In
2002 she received her doctorate in Natural Sciences by
the TUB. From 2002 to 2008 she worked on two Petro-
physics Projects (Chicxulub and Chesapeake) at the TUB.
In August 2008 she joined the microseismicity group at
the Free University Berlin again working in the field of
rock physics and concentrating on pore pressure induced
microseismicity. From May 2011 to March 2012 she was
Gast Dozentin for Experimental Rock Physics at the Insti-
tut für GeologischeWissenschaften/FU Berlin Since 2012
she works on a DGMK- Project now dealing with stress
induced anisotropy. Membership: SEG, German Geo-
physical Society (DGG).

PHASE Research Team

192

Erik H. Saenger
received his diploma in Physics in March 1998 and his ETH Zurich
Ph.D. in November 2000 from the University of Karl- Geological Institute
sruhe. Since January 2001 he has been a research asso- LEB D 3
ciate at the Freie Universität Berlin. During this time, Leonhardstrasse 19
he focused on wave propagation in fractured and multi 8092 Zurich, Switzerland
phase materials. Erik has been the head of the Numeri- email: [email protected]
cal Rock Physics Group. Currently, he is working at the
ETH Zurich. He is member of the DGG, DPG, SEG, and
EAGE.

Peter Wigger
received his Diploma in Geophysics 1977 from the Freie Freie Universität Berlin
Universität Berlin (FUB). Then he worked in the explo- Malteserstr. 74-100, Build. D, Room 107
ration company PRAKLA-SEISMOS. He returned to the D-12249 Berlin
FUB and was engaged in a number of projects explor- Tel.: ++49 (0)30 838 70586
ing the continental lithosphere by controlled source seis- Fax.: ++49 (0)30 838 70728
mology. 1983 he finished his Ph.D. at the FUB (thesis email: [email protected]
about the Apennin mountain chain and the geothermal
anomaly in Tuscany, Italy) and since that time his research
was focussed to the South American Andes, using wide
and steep angle seismics as well as seismological meth-
ods. Since 1985 he also acted as the research manager for
two large interdisciplinary geoscientific programs based at
the FUB: Research Group "Mobility of active continental
margins" (until 1990) and Collaborative Research Center
(SFB 267) "Deformation Processes in the Andes" (until
2005).

PHASE Research Team

Annual Report 2012 193

PhD Students

Wasja Bloch
gained his Bachelor’s degree in Geosciences at the Uni- Freie Universität Berlin
versity of Göttingen in 2009, and started working as a Malteserstr. 74-100, Build. D, Room 107
student research assistent for the FU Geophysics Group D-12249 Berlin
in early 2011. In mid 2012 he graduated at the FU Tel.: ++49 (0)30 838 70586
Berlin with a thesis about local seismicity in the north- Fax.: ++49 (0)30 838 70729
ern Chilean forearc region. He then continued his work email: [email protected]
about the northern Chilean subduction zone as a Ph.D.
student. As of early 2013 the focuses on the diverse seis-
mogenic processes within the underthrusted plate in sub-
duction regimes.

Stine Gutjahr
started as a student assistant in our imaging group and re- Freie Universität Berlin
ceived her diploma in 2009. For her diploma theses she Malteserstr. 74-100, Build. D, Room 127
analysed active seismic data from the San Andreas fault D-12249 Berlin
at Parkfield related to the San Andreas Fault Observatory Tel.: ++49 (0)30 838 70597
at Depth. As a PhD student she reprocesses old industry Fax.: ++49 (0)30 838 70729
seismic reflection data in order to image deep structures email: [email protected]
of the San Andreas fault in the Cholame region.

Nicolas Hummel
studied Geophysics at the University of Karlsruhe. In Freie Universität Berlin
2008 he received his diploma from the Faculty of Physics. Malteserstr. 74-100, Build. D, Room 142
From 2004 to 2006 he was research assistant at the Black D-12249 Berlin
Forest Observatory testing different data acquisition sys- Tel.: ++49 (0)30 838 70443
tems and actively participated in several seismological ex- Fax.: ++49 (0)30 838 70729
periments. Between 2006 and 2008 he joined the seismic email: [email protected]
imaging group followed by the rock physics group where
he finally wrote his thesis about microseismicity and pore
pressure diffusion in porous media. Since October 2008
he is research associate at Freie Universität Berlin and a
member of the Phase university consortium. His Ph.D.
topic is about physics of induced microseismicity espe-
cially for hydraulic fracturing. His focus is on transport
processes responsible for seismicity triggering and the
determination of hydraulic transport properties. He is a
member of EAGE and SEG.

PHASE Research Team

194

Cornelius Langenbruch
joined our research group in summer 2005 as a teaching Freie Universität Berlin
assistant for the department of Geophysics. In 2006 he Malteserstr. 74-100, Build. D, Room 132
also started to support the work of the microseismicity D-12249 Berlin
group analyzing the temporal evolution of fluid-induced Tel.: ++49 (0)30 838 70867
seismicity. In July 2008 he received his Diploma in Geo- Fax.: ++49 (0)30 838 70729
physics from the Freie Universität Berlin. Since August email: [email protected]
2008 he is a Ph.D. student at the Freie Universität Berlin.
His current research focuses on the physics, numerical and
geo-mechanical modeling and the statistics of fluid injec-
tion induced seismicity. In 2010 he received the award
for young scientists of the GtV-BV Geothermie. He is a
member of EAGE, EGU, AGU and SEG.

Antonia Oelke
studied physics before she joined our group in 2009. For Freie Universität Berlin
her diploma thesis, she closely worked with Stefan Buske Malteserstr. 74-100, Build. D, Room 145
on the reprocessing of seismic profiles in New Zealand. D-12249 Berlin
Since 2010, she is working as a Ph.D. student with the Tel.: ++49 (0)30 838 70433
focus on the modelling of the reflection coefficient of Fax.: ++49 (0)30 838 70729
a hydraulic fracture, and she became a student of the email: [email protected]
GEO.SIM graduate school in 2011. She is a member of
EAGE.

Anton Reshetnikov
studied physics and computer sciences at Department of Freie Universität Berlin
Computational Physics, Faculty of Physics, Saint Pe- Malteserstr. 74-100, Build. D, Room 129
tersburg State University, Russia. He received his MSc D-12249 Berlin
diploma in 2006. From 2001 to 2007 he worked for the Tel.: ++49 (0)30 838 70598
GEOVERS ltd., Moscow, Russia as a software engineer, Fax.: ++49 (0)30 838 70729
senior geophysicist and project manager, where he has ob- email: [email protected]
tained the experience of software designing and develop-
ing for seismic data processing, solutions of forward and
inverse problems of geophysics, large scale calculations.
Then he joined the microseismicity group at Free Univer-
sity, Berlin in September 2007. Now he is working on
microseismicity location and imaging problems. His re-
search interests include seismic data processing, numer-
ical methods in imaging and modeling of seismic wave
propagation for heterogeneous media.

PHASE Research Team

Annual Report 2012 195

Karsten Stürmer
joined our group as a student assistant in numerical rock Freie Universität Berlin
physics. In September 2009 he finished his master in this Malteserstr. 74-100, Build. D, Room 137
topic and became a full member of our research staff. D-12249 Berlin
Now he studies waveform similarities in a close collabo- Tel.: ++49 (0)30 838 70461
ration with Joern Kummerow. Furthermore he is an active Fax.: ++49 (0)30 838 70729
member of the SEG student section Student Geophysical email: [email protected]
Society of Berlin.

Changpeng Yu
Changpeng Yu studied geophysics at China University Freie Universität Berlin
of Geosciences (Wuhan) since 2004 and received his Malteserstr. 74-100, Build. D, Room 127
diploma in 2008. In 2007, he joined the Center for Geo- D-12249 Berlin
physics Prospecting in CUG as a research assistant and Tel.: ++49 (0)30 838 70597
then became a formal master student the next year. In Fax.: ++49 (0)30 838 70729
his former group, his work covered reflection seismic data email: [email protected]
processing, velocity model building, attenuation analysis
and full-waveform inversion. After receiving his MSc
diploma at CUG, he joined our group in Sep 2011 with
support from China Scholarship Council. Now his work
mainly focuses on anisotropic phenomena revealed by mi-
croseismic observations.

PHASE Research Team

196

MSc & Bachelor Students

Julio Galindo Guerreros

wrote his bachelor thesis in Dr. Brasse’s Magnetotelluric Freie Universität Berlin
group with focus on 2D forward modelling of anisotropic Malteserstr. 74-100, Build. D, Room 141
data. Afterwards, he became member of Prof. Kauf- D-12249 Berlin
mann’s Geodynamics working group, where he conducted Tel.: ++49 (0)30 838 70175
several near surface measurements both for teaching and Fax.: ++49 (0)30 838 70729
scientific purposes in the framework of the Karst Project. email: [email protected]
In 2012 he joined our group as student assistant and teach-
ing fellow. Currently, he is working with Carsten Dinske
on numerical simulations of fluid induced seismicity in
hydraulically heterogeneous media. Moreover, he is SEG
Student Chapter President of the Student Geoscientific
Society Berlin & Potsdam and member of the SEG and
EAGE.

Jonas Folesky
received his BSc of Science in Physics from TU-Berlin in Freie Universität Berlin
2011. Since then he is a student Member of our Group. Malteserstr. 74-100, Build. D, Room 145
Currently he is working with Jörn Kummerow on back D-12249 Berlin
projection imaging and modeling using a finite difference Tel.: ++49 (0)30 838 70433
program by Erik Saenger. Fax.: ++49 (0)30 838 70729
email: [email protected]
berlin.de

Steven Franke
is a student member in the seismic working group since Freie Universität Berlin
2012. He is currently working on an overview of a by Malteserstr. 74-100, Build. D, Room 136
hydraulic fracturing induced microseismic data set, which D-12249 Berlin
is also going to be the topic of his bachelor thesis. Tel.: ++49 (0)30 838 70184
Fax.: ++49 (0)30 838 70729
email: [email protected]

Friederike-Josephine Fußek
joined our group in 2012. She works in the rock physics Freie Universität Berlin
group of Sibylle Mayr, where she will write her bachelor Malteserstr. 74-100, Build. D, Room 105
thesis. Furthermore, she is President of the EAGE Student D-12249 Berlin
Chapter of the Student Geoscientific Society e.V.. Tel.: ++49 (0)30 838 70587
Fax.: ++49 (0)30 838 70729
email: friederike-josephine.fussek@t-
online.de

PHASE Research Team

Annual Report 2012 197

Alexander Häckel
is a student member and teaching assistant. In his Bache- Freie Universität Berlin
lor Thesis he applied the seismicity based reservoir char- Malteserstr. 74-100, Build. D, Room 136
acterization on the Cooper Basin microseismic data set. D-12249 Berlin
Tel.: ++49 (0)30 838 70461
Fax.: ++49 (0)30 838 70729
email: [email protected]

Aurelian Röser
joined our group as a student assistant in 2009. In Freie Universität Berlin
2011, he received his bachelor degree, for which he fo- Malteserstr. 74-100, Build. D, Room 145
cused his research on common patterns in frequency- D-12249 Berlin
magnitude distributions of natural seismicity along con- Tel.: ++49 (0)30 838 70433
vergent plate boundaries. For his master thesis, Aurelian Fax.: ++49 (0)30 838 70729
investigates stress drop behavior by means of the analysis email: [email protected]
of frequency-magnitude distributions during aftershock
series. Moreover, he works as a teaching fellow and the
student advisor for students majoring in geophysics. Au-
relian is the Chairman of the Student Geoscientific Soci-
ety and a member of the SEG Committee on University
and Student Programs as well as of SEG, EAGE, AAPG,
and SPE.

Andrei Rusu
is a student member in the seismology working group Freie Universität Berlin
since early 2012. In his Bachelor thesis he analyzed the Malteserstr. 74-100, Build. D, Room 137
microseismic dataset from the Basel geothermal reservoir, D-12249 Berlin
for which he calculated the fault plane solutions. Cur- Tel.: ++49 (0)30 838 70461
rently he is working on determining fault mechanisms for Fax.: ++49 (0)30 838 70729
multiplet events. email: [email protected]

Tetsuro Taranczewski
is a student member in the seismic working group since Freie Universität Berlin
2008. He is currently working on Bohemia swarm earth- Malteserstr. 74-100, Build. D, Room 136
quakes. He is also a teaching assistant. D-12249 Berlin
Tel.: ++49 (0)30 838 70905
Fax.: ++49 (0)30 838 70729
email: [email protected]

PHASE Research Team

198
Annual Report 2012 199

List of Sponsors

BG Group
Thames Valley Park
Reading
Berkshire RG6 1PT United Kingdom
Phone: +44 (0) 118 935 3222
Fax: +44 (0) 118 935 3484
www.bg-group.com
BGP INC.
China National Petroleum Corporation
No. 2 Hengtong North street
Zhuozhou, 072750, Hebei Prov.
P. R. China
Phone: +86 -(0)312-3737057
Fax: +86 (0)312-3738015
www.bgp.com.cn

BP Exploration Operating Company Limited

Chertsey Road
Sunbury on Thames
Middlessex TW 16 7LN Great Britain
Phone: +44 (0) 1932 762000
Fax: +44 (0) 1932 738411
www.bp.com

Cairn Energy India Pty Ltd

Vipul Plaza, Suncity, Sector-54
Gurgaon-122002
Haryana
India
Phone: +91 124 459 3207
Fax: +91 124 28809320
www.cairnindia.com

Chevron Energy Technology Company

6001 Bollinger Canyon Rd
San Ramon, California, 94583
USA
Phone: +1 925 842 0248
Fax: +1 925 842 2111
www.chevron.com

List of sponsors
200

ExxonMobil Upstream Research Company

P.O. Box 2189
Houston, Texas, 77525-2189
USA
Phone: +1 713 431 4832
Fax: +1 713 431 6333
www.exxonmobile.com

Gaz de France
Produktion Exploration Deutschland GmbH
Waldstraße 39
49808 Lingen (Ems)
Germany
Phone: +49 (0) 591 612-0
Fax: +49 (0) 591 6127-000
www.gazdefrance.com

RWE Dea AG
Überseering 40
22297 Hamburg
Germany
Phone: +49 (0) 40 6375-0
Fax: +49 (0) 40 6375-3175
www.rwe.com/web/cms/en/53846/rwe-dea/

Total Office Pau

Avenue Larribau
64018 Pau
France
Phone: +33 (0) 559 83 40 00
Fax: +33 (0) 559 83 44 81
www.total.com

List of sponsors
Annual Report 2012 201

altcom
13 North Parade, Penzance
Cornwall, TR18 4SL
United Kingdom
Phone: +44 (0)1736 368254
Fax: +44 (0)1736 368260
www.altcom.co.uk

ESG
20 Hyperion Cour
Kingston, ON
Canada
Phone: +1 613 548 8287
Fax: +1 613 548 8287
www.esgsolutions.com

Geo Explorers Ltd.

Wasserturmplatz 1
CH-4410 Liestal
Switzerland
Phone: +41 61 821 60 40
Fax +41 61 821 60 44
www.geothermal.ch

Norsar
Instituttveien 25
N-2007 Kjeller
Norway
Phone: +47 63 80 59 00
Fax: +47 63 81 89 19
www.norsar.no

List of sponsors