You are on page 1of 5

Uncertainty of micro-seismic events location during fluid injection

Valeri Korneev1 and Thomas Daley1


1

Lawrence Berkeley National Laboratory, USA

submitted to:
1st Joint International Workshop for the Earths Surface and Subsurface 4D Monitoring in 2012
Venue: KACST Conference Hall, Riyadh, Saudi Arabia
Date: January 8 (sun.) - 11(wed.), 2012

ABSTRACT
The rate of fluid injection into underground reservoirs varies with injection pressure.

Use of too high pressure

can cause fracturing resulting in increased level of micro-seismicity (MS). Among potential applications, the
storage of CO2 underground (carbon sequestration) is one where cap rock integrity is paramount. Accurate
locations of the MS events can be critically important for monitoring fluid injection, evaluating extent of fracture
propagation within or outside the reservoir, and timely decision making about injection regimes.
Accuracy of the MS location depends on available data, signal to noise ratios of the seismic records, complexity
of the local geology and the location method, namely accuracy of the forward modeling approach.

Commonly a

ray method is utilized for computing the theoretical traveltimes, because it is relatively fast and can be used for
iterative location schemes.

To ensure stability of the ray method it uses smoothed simplified models, which

potentially can severely compromised the result.

For example, if a receiver and/or the source are located close to

an interface, the recorded traces have several waves contributing in the direct P- and S- wave arrivals with the
resulting polarization being very different

from those obtained by the ray method.

For horizontally layered

models, we propose and test a new approach based on the use of finite-difference (FD) computations.

FD

method allows us to model the fields with all complexities related to the model, and gives accurate estimates for
the traveltimes and effective polarization of waves used in the MS locations.

For a model based on the In Salah

sequestration site, we compute the field characteristics of the modeled field just once for all potential source
points using the reciprocity principle. The resulting inversion is straightforward and allows estimation of the
spatial uncertainty as a function of the noise in the data.

At far offsets the model complexity causes an inherent

nonuniquness, suggesting use of modified observation schemes for more robust MS locations.

Introduction
Uncertainty of a micro-seismic event (MSE) location is a complex problem because of the variety of
sources affecting the result (Swanson et al., 1992; Zimmer et al., 2009). In practice, all of the MSE
parameters including hypocenter position, an origin time, orientation and a size of a rupture (moment
tensor components) are unknown and are to be determined using the seismic records from the local
networks. The key data elements of the seismic records include travel times of the first breaks for
direct P- and S- waves and polarization of the P- wave. Thus, for a homogeneous (and known)
velocity models the difference in S- and P- traveltimes is proportional to a source-receiver distance
while the polarization points at the source location. Single point measurement has an inherent
ambiguity of two possible (opposite) directions towards the source because in general the first break
can be both positive or negative depending on rupture orientation. This ambiguity can be resolved
by the use of sensors placed at different levels in an observation borehole.
The following factors contribute into the uncertainty in MSE locations:
Ambient noise translates into errors in data picking for traveltimes and polarization angles. Small
magnitude events therefore in general can be located less accurately compare to stronger events.
Network density. Denser networks generally provide more accurate locations of MSE .
However, use of multiple wells for sensors can be excessively expensive. For the given velocity
models the gains due to extra sensors can be evaluated by numerical modeling.
Velocity model. Proper P- and S- velocity (as well as density) models are required to compute the
expected traveltimes and incident angles for the inversion. Obtaining a good velocity model can be
done using independent estimations from surface seismic data, well logs, vertical seismic profiling,
cross-well observations and local geological maps.
Forward modeling method. Even if the velocity model is exactly known the computation of the first
arrival times can be not a trivial task. Most of MSE inversion methods are iterative and need
multiple computation of traveltimes to reach the minimal misfits between the modeled and recorded
data. The correspondent modeling algorithms need to work fast which is usually achieved by using a
ray-tracing method, which has several principal problems. First, the ray-tracing method is a high
frequency asymptote for the exact solution and this assumption can conflict with the real situation.
Second, the ray-tracing method is hard to apply at high offsets in the layered media because of
numerous possible multiple rays with similar wave paths. Third, usually, at large offsets, the first
arrivals are comprised by the head waves due to the presence of the layers with the fast velocities.
Part of the propagation path for these waves goes along the interfaces of high velocity layers which
present the zones of numerical instabilities for the ray-path method. To prevent such instabilities a
spatial velocity model smoothing is usually applied. Therefore the true velocity model gets distorted
contributing into increase of location uncertainty. Use of eikonal equation solver is compromised by
its relatively long computation time. Moreover, this solver cannot provide a needed polarization
evaluation for medium and large offsets when either source or the receiver are located in a close
proximity of a model interface. We roughly estimate close proximity as a distance up to eighth of a
dominant wavelength. Indeed, on both sides of the interface we have a close interference of either
reflected or refracted P- and S- waves, with resulting polarization being different from either incident
or purely refracted P- waves as it usually assumed in MSE location techniques.
In this study we demonstrate, that in the case of horizontally-layered medium it is possible to compute
the fields of traveltimes for P- and S- waves, as well as incident P- wave angle using two runs of 2D
finite-difference (FD) elastic code and using reciprocity principle. Reciprocity principle states, that
traveltime from a source in A to a receiver in B is the same as a traveltime from a source in B to a
receiver in A. Similar statement is valid for the polarization. The obtained fields carry all
mentioned complexity of the waves recorded and/or excited in the vicinity of the interfaces and can be
readily used for mapping of MSE locations and their uncertainties without applications of any

iterative schemes (Gibowicz and Kijko, 1993). The examples use real case models from InSalah CO2
injection project.
InSalah CO2 injection project
Recognizing the importance of the role that In Salah could play in advancing secure and economic
storage of CO2, the gas operators established an international Joint Industry Project (JIP). The JIP runs
an extensive CO2 monitoring and verification program; the findings are shared widely with the
scientific community, regulators, policy makers and NGOs to advance CCS as an important climate
change mitigation technology. The JIP has three objectives:

To provide assurance that secure geological storage of CO2 can be verified cost-effectively
and that long-term assurance can be provided by short-term monitoring.
To demonstrate to stakeholders that industrial-scale geological storage of CO2 is a viable
greenhouse gas mitigation option.
To set precedents for the regulation and verification of the geological storage of CO2,
allowing eligibility for greenhouse credits.

In recognition of the projects importance the JIP receives assistance from the US Department of
Energy, the EU Directorate of Research and collaborates with leading technology providers from
around the world.
The separated CO is transported from the Krechba central processing facility to three injection wells,
pumping the CO in an isolated zone in the Krechba reservoir for permanent storage. The target zone is
20 meter thick and lies two kilometers below the surface. The CO is a supercritical fluid at these
conditions of pressure and temperature. Among the other monitoring methods being employed at
Krechba is microseismic monitoring (find more information at http://www.insalahco2.com).
Approach
For data inversion we need to find a global minimum for misfits of measured and modelled data. If
the model is laterally-homogeneous then a single 2D modeling result provides traveltime and incident
angle solutions for all the vertical planes containing any source-receiver pair. The azimuthal
orientation of this plane can be determined from the polarization of the first arrival. Placing a source
in an actual sensor position and using a dense receiver coverage we can obtain the forward modelling
solutions for the whole volume of study after a single code execution. First breaks of P- waves and
incident angles can be easily determined automatically for the whole field using an available picking
algorithm. The S- wave traveltimes can be determined in the same manner using the actual Svelocity model as a P- velocity in the FD code input. Thus just two runs are needed to determine all
the fields necessary for the inversion of data recorded at a particular receiver level (depth).
Similarly, the forward modeled data can be computed and stored for all available levels. This part of
the inversions should be done just once unless the velocity model need to be modified.
When actual MSE data are recorded and picked, they can be subtracted from the modelled fields and
combined using any desired norm. We have found that L1 norm is the most effective for the MSE
location purposes because it directly gives the location uncertainty maps. The MSE location can be
simply found after the global search of the maps, which also allow a direct visual analysis. The
resulting MSE location procedure is simple, fast and accurately takes into account all the field
complexity resulting from the layering of the model.
A similar approach can generally be applied to an arbitrarily heterogeneous 3D model, which is an
often case in mining seismology (Baig, et al., 2009). Such problem however, requires significant
computational resources, because of the need to compute and store a 3D set of traces. Special
optimization algorithms (based on the need to compute just the first arrivals) can be applied here.

Example
We used the velocity model shown on Figure 1 (the model is developed by the NORSAR InSalah
group; Volker Oye, personal communication). Shown is the reservoir at a depth of 2000 m.

Figure 1 Velocity model for InSalah injection project.

Figure 2 Direct P- traveltime map (left panel) and incident angle map(right panel) for the model
shown on Figure 1. Computed using 2D FD code.
The P- traveltime and incident angle maps for a sensor at 80 m depth are shown on Figure 2. We used
550 x 370 receiver grid with 10 m spacing to provide the coverage. The CPU running time on an 8
core PC is 82 minutes. The incident angle map has a complex character at medium and large offsets
due to complex layered model structure.

Figure 3. Relative misfit maps for an MSE at (4500 m, 1000 m): direct S- direct P traveltime (left
panel) and incident angle (right panel).

Figure 4. Combined misfit map for


an MSE at (4500 m, 1000 m

Figure 5. Combined misfit map for


an MSE at (1000 m, 2000 m

Relative misfits for an MSE event at (4500 m, 1000 m) are shown on Figure 3. Non uniqueness of the
incident angle misfit transfers to a non uniqueness of the combined misfit (Figure 4). The maps on
Figures 3 and 4 show the area of uncertainty for a given noise/signal level given on a color scale.
Thus, a 25% error in raw data transfers to about 2000 m by 1200 m uncertainty area. Adding more
sensors in the same well does not bring much of an improvement, although allowing resolving a
possible ambiguity of up- down- ward propagation of MSE waves. Note, that application of extra
wells in the case shown on Figure 4 will unlikely bring much more of the resolution if the
source-receiver offsets stay about the same. This is a result of several (three) local minimum zones
having close horizontal location.
For a smaller offsets, the MSE location is much improved (Figure 5), because for near vertical rays
ambiguity of incident angles vanishes. Small offsets provide the best spatial resolution.
Conclusions
We have developed an approach of non-iterative fast inversion of MSE data. Application of FD
modelling allows accurately account for model complexity and rays multipathing. The forward
modelling maps can be computed at the first stage of the inversion and later used for instantaneous
computation of location maps and finding their global minima. These maps can be analysed visually
giving information about the location uncertainty as a function of data signal/noise ratios.
Acknowledgements
This work was supported by the GEO-SEQ project for the Assistant Secretary for Fossil Energy,
Office of Coal and Power Systems through the National Energy Technology Laboratory, of the U.S.
Department of Energy, under contract No. DE-AC02-05CH11231.
References
Baig, A., Urbancic, M. Seibel, and V. Shumila, 2011, Locating Microseismicity in
three-dimensionally heterogeneous reservoirs, CSPG, CSEG, CWLS Convention.
Gibowicz, S. J. and A. Kijko, 1993, An introduction to mining seismology, Academic Press.
Swanson, P.L. , Boler, F.M., Estey, L.H., and S. Billington, 1992, Accuracy and precision of
microseismic event locations in rock burst research studies, report, Washington, D.C., U.S. Dept. of
the Interior.
Zimmer, U., Bland, H., Du, J., and N. Warpinski, 2009, Accuracy of microseismic event locations
recorded with single and distributed downhole sensor arrays, Expanded abstracts, SEG Annual
Meeting.

You might also like