You are on page 1of 5

Downloaded 10/21/14 to 167.205.22.105. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.

org/

Role of 1D MT inversion in a 3D geothermal field


Dhananjay Kumar*, and G. Michael Hoversten, Chevron Energy Technology Company, Gregg Nordquist,
Chevron Geothermal and Power Operations, William Cumming, Cumming Geoscience
Summary
An extensive magnetotelluric (MT) survey comprised of 85
sites has been acquired over the Darajat geothermal field in
Indonesia to map the geothermal reservoir and the
overlying clay cap. The rouged topography and the
geometry of the margin of the clay cap makes the resistivity
structure 3D at reservoir depth. Although 3D MT inversion
is now commonly used in geothermal interpretations 1D
and 2D MT inversions are still effective tools for a variety
of tasks such as quality assurance. Lower dimensional
inversion can also play two critical roles in determining and
assessing the resistivity model derived by 3D inversion: 1)
by providing a good starting model to reduce the
computational cost of the 3D inversion, and 2) by providing
a computationally feasible path to stochastic inversion of
the data that provides realistic parameters standard
deviations for use in assessing reliability of the resistivity
model. Using a spatially constrained 1D stochastic
inversion of the MT data, we investigate the common claim
that 1D inversion can provide a pseudo 3D model which
closely matches the 3D inversion for the overburden and
clay cap layers. The discrepancy between the pseudo 3D
and true 3D inverse models increases with depth, however
the presence of the core resistive feature of the field is still
indicated at approximately the same depth as found in the
true 3D model. Analysis of the 1D model parameter
probability density functions shows that layer thicknesses
are better determined than layer resistivities.
Introduction
The magnetotelluric (MT) method has been widely used to
resolve resistivity patterns in geothermal reservoirs,
because MT method can resolve the low resistivity, low
permeability hydrothermal smectite clay cap that acts as the
trap over high temperature geothermal reservoirs in both
sedimentary and igneous environments (Ussher et al.,
2000). The temperature related transition from smectite to
more resistive and more brittle illite and chloride is closely
correlated with the top of permeable geothermal reservoirs.
Geothermal MT success case histories have sometimes
focused on mapping the base of the low resistivity smectite
clay cap using 1D layered inversion (Anderson et al.,
2000). Although 3D MT inversion has been increasingly
emphasized in geothermal MT exploration (Cumming, et
al., 2000; Uchida and Sasaki, 2006; Arnason et al., 2010)
and is now in-house application at Chevron, it is still
relatively expensive and 1D inversion remains the most
common imaging methods used for quality assurance and
uncertainty testing (Cumming and Mackie, 2010).

2010 SEG
SEG Denver 2010 Annual Meeting

Figure 1: Darajat geothermal field is located in the West


Java province of Indonesia, about 150km southeast of
Jakarta (figure modified from Hadi et al., 2005).
A total of 85 MT sites, with usable 5-component MT data
from 0.001 to 120 Hz, were acquired during surveys in
1997 and 2004 over the Darajat geothermal field in
Indonesia (Figure 1) to derive an electrical resistivity model
of the geothermal reservoir. The low resistivity pattern
imaged by earlier MT inversions matched the smectite-illite
clay transition determined from wells and conformed to the
reservoir isotherms reasonably well. However, ambiguity
was particularly evident on the margins of the reservoir for
targeting makeup wells. Because of the geothermal power
facilities, data quality is compromised in some areas, but it
is unclear to what extent this affects the reliability of the
MT inversions.
It is well known and widely appreciated that the inversion
of MT data is non-unique, like most geophysical imaging
methods. The reliability of an MT inversions resolution of
earth properties depends not only on the noise level in the
MT data and the accuracy of the MT forward response
calculation for a particular earth model, but also on the
criteria used to choose a single final model from the many
non-unique possibilities. Currently the computationally
practical 3D inversion algorithms fall into the category of
gradient based algorithms subject to regularization
constraints designed to weight the inversion towards more
realistic models. As is the case with the 3D inversion used
in this paper (Commer and Newman, 2008), these
techniques find a single model which fits the observed data.
If estimates of model parameter resolution are needed (e.g.
how well is the thickness of the clay cap resolved) they
cannot be accurately derived from deterministic inversion

1107

Downloaded 10/21/14 to 167.205.22.105. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

1D MT inversion in a 3D geothermal field


using model covariance matrix for two reasons: 1) a high
degree of parameter correlation is induced by regularization
of pixilated models, and 2) no formal theory exists to
account for the off diagonal cross-correlation coefficients
of the matrix. Even when the number of parameters can be
reduced thereby reducing the parameter correlation
standard deviations derived from the diagonal of the
correlation matrix significantly underestimate the true
parameter variance as measured by the model parameter
probability density function (pdf) (Trainer and Hoversten,
2009).
Of the methods commonly used to characterize model
uncertainties, stochastic inversion is arguably the most
general, but it is currently not feasible in 3D. As a costeffective alternative, we demonstrate that 1D stochastic
inversion can provide significant information about model
uncertainty and can provide a good starting point for 3D
inversion. We show that while statistical moments such as
the median and mode of the parameter pdfs can be used to
generate a single model, they are not guaranteed to provide
a model that best fits the data. This can be accomplished
by a gradient based inverse starting from the model
constructed from mode values of the parameter pdfs.
MT inversion
Because the on-diagonal elements of the MT impedance
were less reliable only the off-diagonal elements of the
impedance (Z xy and Z yx ) are used in the 3D inversion. The
inversion model included the detailed topography model;
and therefore no static corrections were applied. Of the
many MT parameters that can be used for 1D inversion, we
chose the determinant of impedance (Berdichevskiy and
Dmitriev, 1976) because many investigators have found
that it works well in geothermal setting and requires no
interpreter input (Anderson et al, 2000; Arnason et al.,
2010).
We have developed a stochastic Markov Chain Monte
Carlo (MCMC) inverse algorithm (Gilks et al., 1996) and a
deterministic least squares 1D MT inverse algorithm which
invert all available MT soundings simultaneously subject to
lateral and vertical smoothing of the 1D model parameters.
The stochastic MCMC algorithm without lateral smoothing
is described in Trainor and Hoversten (2009). The lateral
smoothing approach was demonstrated by Viezzoli et al.,
2008 for airborne EM data, where the set of data locations
are triangulated to generate a list of nearest neighbors for
each site. The lateral smoothing between sites is scaled by
the inverse distance to the neighbors. Vertical smoothing of
layer resistivity and thickness can be done if a large number
of layers are specified. In this work we test the properties
of the algorithm using only 4 layers to represent the major
elements of the model.

2010 SEG
SEG Denver 2010 Annual Meeting

The lateral smoothing using nearest neighbor sites as


constrain in the 1D stochastic MCMC algorithms is added
as prior which modifies the likelihood function as:

1
f ( E | x C ) = B exp err
2

(1)

with

( )

c
N 2 nfreq e
N NN ( l ) NP

ijk M ijk x
+ 1 p ln p mn
err =

p ln
ijk
i =1 j =1 k =1
l =1 m =1 n =1 d lm

The likelihood function describes the posterior probability


of candidate model xC given the observed data. Embedded
in this exponential function (equation 1) is the misfit
between the observed data e and the calculated data M,
each at frequency k scaled by the data errors . Thus, the
total number of data is the product of the number of
frequencies (nfreq) and number of sites (N) multiplied by
two for the real and imaginary parts. NN(l) is the number
of nearest neighbors to the lth site and NP is the number of
parameters, p, in the 1D models, and d is the distance
between two sites. is the Lagrange multiplier controlling
the tradeoff between data fit and model smoothness.
The model update, mi+1, at the (i+1)th iteration for the
regularized least square inversion (LSQ) can be expressed
as (Newman and Alumbaugh, 1997):

(2)
mi +1 = G T G + W T W G T Dd ,
with G = DA , and d = (d d i + Ami ) , where m is the
model parameter vector, D is the data-weighting matrix, A
is the model sensitivity matrix (the Jacobian), is the
model smoothing trade-off parameter, W is the
regularization matrix, and the superscript T denotes
transpose of a matrix. The data-weighting matrix is the
inverse of the data standard deviations. The spatial model
smoothing is applied as in the stochastic inversion via W.
Results
Figure 2 shows a depth slice through a 3D conductivity
model at 0 m elevation (mASL) (~ 1800 m below the
surface) from 3D inversion using the code of Commer and
Newman (2008). The 3D inversion was started from a 0.1
S/m half space below topography. The figure shows
contour of the top of the Andesite intrusion as mapped from
boreholes. The resistivity structure and the top Andesite are
well correlated, and are also consistent with gravity data
(Rajeki et al., 2010). The 3D resistivity structure implies
that the Andesite may still be present at 0 mASL to the
north where contours from wells do not match the 3D
model. The ambiguity may be related to variation in rock
properties or to uncertainty in the resistivity model.
Because a thorough assessment of model uncertainty using

1108

S/m
9203000

9202000

9201000

Site: DJ-35
(a)
100

10

Figure 2. Depth slice of conductivity from 3D MT inversion


(blue color shows resistive body) at 0 (mASL) elevation
through Darajat Geothermal field. East-west black line is
location of cross section shown in Figure 5. Major faults
and the top of Andesite elevation contours are also posted.
The unit of conductivity is Siemens per meter (S/m).
The stochastic MCMC inversion provides complete pdfs
of layer parameters from which we must choose one of the
moments of the distribution (mean, mode or median) if a
single value is desired. The mode of the distribution
represents the most likely value so we use that to construct
model resistivities and thicknesses. These models are not
however guaranteed to provide the best fit to the observed
data in a least squares sense. For that we use the modes
from the MCMC to create starting models for the LSQ
inversion. Figure 3 presents examples of a bad and a good
data fits for 1D constrained inversion, and Figure 4 shows
example of pdfs at site 35. We have observed that the LSQ
inversion performs much better using starting models
generated from the MCMC.
By comparing the 1D inversion with the 3D inversion
(Figure 5) we find a reasonably good agreement for
mapping of the overburden and conductive clay cap (layers
1 and 2: b, c and d). Below the conductive layer the match
with the 3D model is generally poor. Generally the best
determined parameter from the 1D MCMC as indicated by
the standard deviation from the pdfs (Figure 4) is the
resistivity of the clay cap. Clay cap from 1D inversion is
thicker than 3D inverted model that indicate a need to use
more layers in 1D inversion. Based on the uncertainty
analysis of 1D stochastic MCMC inverted models the layer
thicknesses seem to be better determined than layer
resistivities, however this is different from conventional
wisdom (Simpson and Bahr, 2005).

2010 SEG
SEG Denver 2010 Annual Meeting

frequency

804000

frequency

803000

frequency

802000

0.1
10
Frequency (Hz)

Site: DJ-25

(d)

10
0.001

0.1

10

Frequency (Hz)

Figure 3: Data mismatch. Comparing 1D inverted model


synthetics with field data from two MT sites: site DJ-35
where 1D MCMC provides good data fitting (a and b) and
site DJ-25 where 1D MCMC provides poor data fitting (c
and d). Synthetic based on 1D LSQ is shown by red circle
points and synthetics based on 1D MCMC (using mode of
model pdf) is shown in blue + mark. 1D field data are
shown by the black X points and errors in data are shown
by black dashed lines.

frequency

801000
Easting (m)

Site: DJ-25

(c)
1
0.001

2
dev

1 h1

frequency

800000

Site: DJ-35
(b)

90

h1

frequency

799000

10

h2

frequency

798000

Phase (degrees)

10

9200000

9199000

90

100

Phase (degrees)

9204000

App. Resistivity (ohm-m)

9205000

App. Resistivity (ohm-m)

a 3D MT inversion algorithm is a computationally


impractical task, it was hoped that the resolution of this
feature could be tested using a 1D stochastic approach.

Northing (m)

Downloaded 10/21/14 to 167.205.22.105. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

1D MT inversion in a 3D geothermal field

h3

2 h2

4-layer
model

3 h3
4

Figure 4: Parameter pdfs after 1D MCMC constrained


inversion for a 4 layer 1D model at Site 35. The upper 4
are layer resistivities and the lower 3 are layer thicknesses.
The statistical measures are also marked.

1109

1D MT inversion in a 3D geothermal field

35

34

33

Discussions and Conclusions

420b

(a)

Z (m)

1000

Red: from (b)


White: from (c)
Black: from (d)

1000

3D Inversion
(b)

Z (m)

1000

1000

1D MCMC unconstrained
(c)

Z (m)

1000

1000

1D MCMC constrained
(d)

Z (m)

Downloaded 10/21/14 to 167.205.22.105. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

site

1D LSQ constrained
798500

X (m)

803500

Figure 5: West to East cross-section at easting 9202500


through 3D inversion (a), 1D MCMC unconstrained
inversion (b), 1D MCMC constrained inversion (c), and 1D
LSQ constrained inversion (d). The layer interfaces from 4layer 1D inversions are shown on the 3D inverted model
(a). Note that 1D MCMC unconstrained inverted model is
also shown for comparison with constrained inversions.
Model parameters from 1D MCMC shown here is mode of
pdfs, and the mode of pdfs from (c) was used as starting
model for LSQ constrained inversion (d). Top and base of
reservoir is marked on the 1D inverted models (b, c and d).

2010 SEG
SEG Denver 2010 Annual Meeting

In geothermal exploration a high quality MT data set can be


used to resolve the depths to and the geometry of the claycap overlying the geothermal reservoir. Beneath the
conductive clay cap the resolution worsens, and details of
the resistivity structure at reservoir depths are uncertain.
When integrated with other geologic and geochemical data
the geometry of the clay cap as defined with a detailed
inversion of the MT data provides important constraints on
the location and size of the reservoir. By implementing a
1-D stochastic inversion we aim to better understand and
quantify the uncertainties of the resistivity structure of the
shallow high resistivity overburden and the clay cap. This
should help to provide meaningful constraints for the
shallow structure in the 3-D modeling and may in turn
improve confidence in deeper features at reservoir depths.
Testing of the MCMC inversion shows that the mode of
model parameter pdfs from the 1D stochastic MCMC
inversion provides a pseudo 3D resistivity model that is in
general matches with the 3D inverse model. The accuracy
of match between 1D and 3D derived models is best for the
layer above the clay cap and the clay cap and then
decreases with depth. The advantages of 1D MCMC
inversion are, 1) speed, and 2) uncertainties in model
parameters. An important next step will be to develop
work flows that can use the information obtained from
these uncertainties to provide starting model for 3D
inversion. The mode of the model parameter pdfs
generally provides a good match to data, but if a single
best-fitting model is desired LSQ provides this when the
starting model is constructed from the mode of the pdfs
after 1D MCMC inversion.
The uncertainty analysis based on these tests of 1D
stochastic MCMC inversion of Darajat data indicates that
the resistivity of the clay cap is best resolved and in general
layer thicknesses are better resolved than layer resistivities;
these conclusions require further verification. In future one
can 1) work on better understanding of the effects of noise
in data and the effects on inverted models, 2) use controlled
source electromagnetic data (e.g., time domain EM) to
constrain relatively resistive geothermal reservoir, 3) make
use of seismic data (e.g., micro earthquake data) to
correlate seismic velocity and resistivity models to
constrain subsurface model, 4) make weighting () for
spatial constrain in 1D MCMC as variable, and 5) try 2D
MCMC.
Acknowledgements
We thank Chevron ETC and Chevron Geothermal and
Power for permission to publish.

1110

Downloaded 10/21/14 to 167.205.22.105. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

EDITED REFERENCES
Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2010
SEG Technical Program Expanded Abstracts have been copy edited so that references provided with the online metadata for
each paper will achieve a high degree of linking to cited sources that appear on the Web.
REFERENCES

Anderson, E., D. Crosby, and G. Ussher, 2000, Bulls-eye!- simple resistivity imaging to reliably locate
the geothermal reservoir: Proc. WGC, 909-914.
rnason, K., H. Eysteinsson, and G. P. Hersir, 2010, Joint 1D inversion of TEM and MT data and 3D
inversion of MT data in the Hengill area, SW Iceland: Geothermics, 39, no. 1, 1334,
doi:10.1016/j.geothermics.2010.01.002.
Berdichevskiy, M. N., and V. I. Dmitriev, 1976, Basic principles of interpretation of Magnetotelluric
sounding curves, in A. Adam, ed., Geoelectric and geothermal studies: Budapest, Akademai Kiado,
165-221.
Commer, M., and G. A. Newman, 2008, New advances in three-dimensional controlled-source
electromagnetic inversion: Geophysical Journal International, 172, no. 2, 513535,
doi:10.1111/j.1365-246X.2007.03663.x.
Cumming, W., and R. Mackie , 2010, Resistivity imaging of geothermal resources using 1D, 2D and 3D
MT inversion and TDEM static shift correction illustrated by a Glass Mountain Case History: Proc.
WGC.
Cumming, W., G. Nordquist, and D. Astra, 2000, Geophysical exploration for geothermal resources: an
application for combined MT-TDEM: SEG expanded Abstract.
Gilks, W., S. Richardson, and D. Spiegelhalter, 1996, Markov chain Monte Carlo in Practice: Chapman &
Hall/CRC Press.
Hadi, J., C. Harrison, J. Keller, and S. Rejeki, 2005, Overview of Darajat reservoir characterization: a
volcanic hosted reservoir: Proc. WGC, Turkey, p. 11.
Newman, G. A., and D. L. Alumbaugh, 1997, Three-dimensional massively parallel electromagnetic
inversion-I. Theory: Geophys. J. Int. 128, no. 2, 345354, doi:10.1111/j.1365-246X.1997.tb01559.x.
Rejeki, S., R. Dave, G. Nordquist, and A. Fitriyanto, 2010, Geologic conceptual model update of the
Darajat geothermal Field, Indonesia: Proc. WGC, Bali, Indonesia.
Simpson, F., and K. Bahr, 2005, Practical magnetotellurics: Cambridge University Press.
Trainor , W., and G. M. Hoversten, 2009, Practical challenges of stochastic inversion implementation for
geophysical problems: SEG Expanded Abstracts, 28, no. 1, 734738.
Uchida, T., and Y. Sasaki, 2006, Stable 3D inversion of MT data and its application to geothermal
exploration: Exploration Geophysics, 37, no. 3, 223230, doi:10.1071/EG06223.
Ussher, G., C. Harvey, R. Johnstone, and E. Anderson, 2000, Understanding the resistivities observed in
geothermal systems: Proc. WGC, 1915-1920.
Viezzoli, A., A. V. Christiansen, E. Auken, and K. Srensen, 2008, Quasi-3D modeling of airborne TEM
data by spatially constrained inversion: Geophysics, 73, no. 3, F105F113. doi:10.1190/1.2895521

2010 SEG
SEG Denver 2010 Annual Meeting

1111

You might also like