Sepulveda Et Al, 2012, Kriging Predictions of Drill-Hole Stratigraphy and Temperature Data From The Wairakei Geothermal Field, New Zealand - Implications For Conceptual Modeling

ARTICLE IN PRESS
G Model
GEOT-887; No. of Pages 19
Geothermics xxx (2012) xxxxxx
Contents lists available at SciVerse ScienceDirect
Geothermics
journal homepage: www.elsevier.com/locate/geothermics
Kriging predictions of drill-hole stratigraphy and temperature data from the

Wairakei geothermal eld, New Zealand: Implications for conceptual modeling
F. Seplveda a, , M.D. Rosenberg b , J.V. Rowland c , S.F. Simmons d
a
Contact Energy Limited, Private Bag 2001, Taupo, New Zealand

GNS-Science, Wairakei Research Centre, Private Bag 2000, Taupo, New Zealand
IESE, University of Auckland, Private Bag 92019, Auckland, New Zealand
d
Hot Solutions Ltd, PO Box 32-125, Devonport 0744, Auckland, New Zealand
b
c
a r t i c l e
i n f o
Article history:
Received 18 July 2010
Received in revised form 9 December 2011
Accepted 3 January 2012
Available online xxx
Keywords:
Wairakei
Geothermal
Temperature
Stratigraphy
Prediction
Kriging
Indicator Kriging
Universal Kriging
a b s t r a c t
Drill-hole temperature and stratigraphic datasets from the Wairakei geothermal eld were used for
geostatistical predictions using Kriging. In order to adequately constrain Kriging models, anisotropy and
trends associated with temperature and stratigraphy were studied using standard variogram analysis, in
combination with new regional and local structural data, revised gravity, and available geoscientic and
reservoir data. This combined analysis lead to the incorporation of horizontal anisotropy (horizontal to
vertical correlation ranging from 8:1 for regional stratigraphic units to 4:1 for local rhyolite bodies) in
the case of stratigraphic models and variable anisotropy in the case of temperature models. In the latter,
the variable anisotropy was represented by two end members: an isotropic model (horizontal to vertical
correlation of 1:1) representative of depths >2000 mGL, and an anisotropic model (horizontal to vertical
correlation of 3:1) representative of depths <1000 mGL. Kriging models of temperature also incorporated
a vertical trend which is a combination of two end members at Wairakei: Boiling-Depth-Point Curve
(convective) and linear (conductive). The Kriging models succeeded in identifying the primary geological
controls on temperature distribution: major upows largely controlled by structures at depth (>1000 m
depth) and shallow (<1000 m depth) outows stratigraphically channelled through formation contacts
and rhyolite edges. A combination of stratigraphy and faults explain local cold downows in shallow
(7501000 m depth) parts of the eld.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
The Wairakei geothermal eld, New Zealand, was the rst
liquid-dominated reservoir in the world to be developed for electricity generation, with production starting in 1958. At present,
Wairakei continues to generate electricity with an installed capacity of approximately 170 MWe, and it is projected to exceed the
300 MWe mark by 2013. The expansion of a geothermal eld invariably poses challenges in terms of the denition of production
and injection drilling targets, both of which are equally important
in the current scheme of sustainable development of geothermal
resources. Conceptual models of geothermal reservoirs play a central role in the denition of drilling strategies, and also dynamically
evolve as more drill-hole data becomes available. As a result of
long-term production at Wairakei, drill-hole datasets have become
increasingly available to assist the elaboration of geological conceptual models and numerical simulations of exploitation effects.
Analysis and interpretation of large geothermal drill-hole datasets
can be challenging, but multidisciplinary analysis can be optimised

by use of geostatistical interpolation techniques (e.g., Fabbri, 2001;
Teng and Koike, 2007). In this study, we applied Kriging to drill
hole datasets of temperature and stratigraphy for geostatistical
modelling to illustrate the utility and limitations of Kriging for prediction of drill-hole parameters at Wairakei, to exemplify the value
of existing geoscientic and reservoir knowledge in providing constraints to geostatistical models, and to characterize and discuss
correlations between the subsurface temperature and stratigraphy
with emphasis in the deep architecture of the eld. Geostatistical models are used to predict the value of an attribute in an
unsampled location using attribute values known at sampled locations. Statistical condence of the predictions naturally decreases
towards peripheral or deep areas of a reservoir which tend to be
less explored. In this context, Wairakei offers a unique opportunity to test geostatistical predictions based on cumulative drilling,
geoscientic, and reservoir data.
2. Geology of the Wairakei geothermal eld
Corresponding author. Tel.: +64 7 3761959; fax: +64 7 3748472.

E-mail address: fabian.sepulveda@contactenergy.co.nz (F. Seplveda).
The Wairakei geothermal eld is located in the Taupo Volcanic

Zone (TVZ), an extensional volcanic arc that has been active during
0375-6505/$ see front matter 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.geothermics.2012.01.002
Please cite this article in press as: Seplveda, F., et al., Kriging predictions of drill-hole stratigraphy and temperature data from the Wairakei
geothermal eld, New Zealand: Implications for conceptual modeling. Geothermics (2012), doi:10.1016/j.geothermics.2012.01.002
G Model
2
ARTICLE IN PRESS
F. Seplveda et al. / Geothermics xxx (2012) xxxxxx
Fig. 1. Map of Central Taupo Volcanic Zone, showing geothermal areas as dened by resistivity data (Schlumberger surveys; 30 Ohm-m boundary from Bibby et al., 1995) with
respect to residual gravity anomalies (Bibby et al., 1995), earthquake data (period 19872011; 05 km depth; source GEONET; data ltered as in Bryan et al., 1999), caldera
margins (after Wilson et al., 1995 and Gravley et al., 2007) and the TVZ rift architecture (after Rowland and Sibson, 2004). Base layer is shaded relief map (25 m Digital Terrain
Model). Map projection: New Zealand Map Grid (m). Abbreviations for geothermal areas as follows (from north to south): TT = Taheke-Tikitere; RT = Rotoma; KA = Kawerau;
RO = Rotorua; WW = Waimangu-Waiotapu; RE = Reporoa; TK = Te Kopia; OK = Orakei-korako; NG = Ngatamariki; BO = Broadlands-Ohaaki; MK = Mokai; RW = Rotokawa;
WT = Wairakei-Tauhara.
the last 1.6 Ma (Fig. 1; Wilson et al., 1995; Houghton et al., 1995)
in response to oblique subduction of the Pacic plate beneath the
Australian Plate. The central segment of the TVZ, extending from
Kawerau geothermal eld in the north to Lake Taupo in the south
(Fig. 1), represents the most active silicic volcanic province on
earth (780 km3 /61 kyr), with a number of associated ignimbrite and
caldera-forming eruptions, which represent more than 90% of the
total erupted magma of the TVZ (Wilson et al., 1995).
The central TVZ marks the concentration of the majority of hightemperature geothermal systems of New Zealand (Fig. 1), with
magmatism as the primary heat source. The depth of such heat
source remains unconstrained at the scale of individual geothermal systems, but regional seismic and MT studies in the central
TVZ identify low resistivity or seismically anomalous regions at
depths of 5 km to >10 km, as an indication of partially molten rock
(Sherburn et al., 2003; Heise et al., 2007). It is also worth noting in
the explored vertical range of the TVZ (<3 km depth), drilling evidence of magma bodies is lacking and evidence of plutonic rocks is
relatively rare (Browne et al., 1992; Milicich et al., 2011).
The central TVZ undergoes NW-SE extension at rates on the
order of 78 mm/yr, which is mostly accommodated by faulting
and tectonic subsidence (Villamor and Berryman, 2001; Nicol

et al., 2006). Active structures mainly consist of NE-SW trending,
high-angle normal faults, and subvertical tension cracks, which
are collectively referred to as rift structures. The spatial relationships between geothermal activity (as delineated by shallow low
resistvity anomalies; Bibby et al., 1995, 1998), rift and caldera structures, and modern seismicity (<5 km depth) are shown in Fig. 1. A
TVZ rift boundary is shown in Fig. 1 to indicate the extent of hightemperature geothermal activity, caldera structures (ca. <330 ka
old; Wilson et al., 1995), and active rift structures and tectonic
subsidence, mainly interpreted from morpho-tectonic analysis (i.e.,
surface fault scarps and tracers, and graben structures; Rowland
and Sibson, 2004). Earthquake locations (Bryan et al., 1999; Fig. 1)
concentrate within the TVZ boundary conrming the potentially
active character of most mapped faults. However, earthquake data
reveals seismic gaps and NS-trending seismicity clusters (northern part of TVZ; Fig. 1) not coincident with active faults. It is also
noted that geothermal locations vary from relatively seismic (e.g.,
Kawerau) to relatively aseismic (e.g., Ohaaki), and from having
strong fault correlation (e.g., Kawerau, Te Kopia; Orakei-korako) to
unclear fault correlation (all others). The general observation is that
G Model
ARTICLE IN PRESS
Effusive units and some
sub-units interpolated separately
Poihipi Rhyolites
Tahorakuri Group
Ohakuri Group
Pre-volcanic
basement
Whakamaru
Ignimbrites
Wairakei Ignimbrite
Rautehuia Breccia
Stockyard Ignimbrite
Waikora Unit
Karapiti 2a
Karapiti 2b and 3 Rhyolite
Waiora Valley
Andesite
Waira Formation (WF)
Wa5
Wa4
Wa3
Wa2
Wa1
Waiora Formation
Andesites, dacites
and rhyolites
340320 ka Whakamaru ignimbrites (Wilson et al., 1986;

Brown et al., 1988). Estimated volume >1000 km3 .
Discontinuous distribution across Wairakei.
Pre-Whakamaru pyroclastics and lavas with intercalations
of tuffaceous sandstone and local greywacke-bearing
conglomerate (Waikora Unit). For further details, refer to
Rosenberg et al. (2009) and Bignall et al. (2010)
Mesozoic greywacke and argillite. Not recognized in the
Wairakei area. Known to be deeper than 2.9 km in the Te
Mihi area. Recently found in east Tauhara.
No distinction of WF subunits for

interpolation. Effusive units
interpolated separately
Te Mihi Rhyolites
Post-Whakamaru, pre-Huka Falls Fm volcanic (pyroclastic)

rocks with interbedded lake sediments. Several effusive
units interbedded. Top locally gradational with Huka Falls
Formation. Basal ignimbrites (Wa1)
No distinction of HFF
subunits for interpolation
Te Mihi Rhyolites
HFF Upper
HFF Middle
HFF Lower
Hu4
Hu3
Hu2
Hu1
Huka Falls
Formation (HFF)
Pumice cover
Oruanui Formation
Subunit
Pumice cover
Wairakei Breccia
Huka Falls Formation (HFF)
Comment
Main effusive units
Formation
Formation
Subunit
This study (after Rosenberg et al., 2009)

Grindley (1965)
Table 1
Summary of stratigraphic denitions used in this study.
A range of stratigraphic models have been proposed for

Wairakei (Grindley, 1965; Steiner, 1977; Healy, 1984; Wood, 1994;
Wood and Browne, 2000; Rosenberg et al., 2009; Bignall et al.,
2010), of which the Grindley (1965) illustration was inuential
for several decades. In this study, a number of re-interpretations,
re-denitions, and changes in nomenclature are introduced after
Rosenberg et al. (2009), as a result of ongoing drilling and detailed
lithological and petrographic re-examinations of core and cuttings.
This study focuses on stratigraphic units underlying Huka Falls
Formation, and accordingly, a detailed discussion of the shallow
stratigraphy of the Wairakei eld is beyond the scope of this
paper. For the purposes of interpolation, no distinctions were made
between the members of Huka Falls Formation and Waiora Formation as described in Table 1 (Grindley, 1965; Rosenberg et al., 2009).
Stratigraphy units of interest here include, from top to bottom: Waiora Formation (Grindley, 1965), Whakamaru Ignimbrites
(known locally as Wairakei Ignimbrite) (Wilson et al., 1986), and
Tahorakuri Formation (Gravley et al., 2006; Table 1). A range of predominantly rhyolitic effusive units occur within these stratigraphic
layers, and these are modeled as separate entities, with the aim
to highlight the heterogeneous character of the stratigraphic units.
The denition of Stockyard Ignimbrite and Poihipi Rhyolites as subunits of Tahorakuri Formation (i.e., pre-Whakamaru Ignimbrites;
Table 1) has been recently established, following revision of stratigraphy and identication of Whakamaru Ignimbrite in wells WK248,
WK253 and WK259 (Bignall et al., 2010).
The Whakamaru Group Ignimbrites collectively represent the
most voluminous ignimbrite complex of the TVZ (>1500 km3
of erupted magma) and an important stratigraphic marker for
the interpretation of deep stratigraphy and fault geometry at
Wairakei, due to both regional extent and relatively easy recognition (crystal type and abundance distinct from other widespread
TVZ ignimbrites; Brown et al., 1998; Saunders et al., 2010). The
caldera boundary from which these ignimbrites derive encircle the
Wairakei, Rotokawa, and Mokai geothermal systems (Wilson et al.,
1986; Houghton et al., 1995; Brown et al., 1998; Fig. 1). In spite
of its regional extent, this unit is highly discontinuous across the
Wairakei eld (Rosenberg et al., 2009). Fewer than ten wells have
penetrated the entire Whakamaru Ignimbrite, revealing a variable
thickness ranging from <100 m (WK248) to >1000 m (WK121). A
number of relatively deep wells such as WK301 (eastern Wairakei,
2000 m deep; Fig. 2) and WK247 (Te Mihi area; 2750 m deep) did
not encounter Whakamaru Ignimbrites; a fact used to aid interpret
locations of the deep structures.
Where Whakamaru Ignimbrites are absent, the denition of the
base of Waiora Formation is generally more uncertain. Based on its
current understanding, the base of the Waiora Formation occurs
Remarks
2.1. Local stratigraphy and structure
1.8 ka Taupo eruption and associated reworked deposits

26.5 ka Oruanui eruption (Wilson, 2001)
Silt/sand grade lake sediments (equivalent Hu3 and H4)
Pumice breccia
Silt/sand grade lake sediments (equivalent Hu1)
Magmatic-hydrothermal breccia
systematic spatial correlations of geothermal activity with faults,

caldera boundaries and seismicity are equivocal based on available
data (Fig. 1).
Both rift and caldera structures are major components of the
architecture of the central TVZ, the distinction of which is sometimes subtle in the TVZ, due to the close control exerted by rift
structures on the inception, development, and evolution of calderas
(Spinks et al., 2005; Gravley et al., 2007; Seebeck et al., 2010;
Rowland et al., 2010). Oblique structures, dened here as structures
with orientations other than NE-trending, have also been documented within the central TVZ (Tienfeng and Hedenquist, 1981;
Acocella et al., 2003; Rowland and Sibson, 2001, 2004) and along
the northern, off-shore prolongation of the TVZ (Wright, 1992;
Lamarche et al., 2006; Mouslopoulou et al., 2008). As discussed
above, oblique structures can also be inferred from regional seismicity (Fig. 1).
Consolidated into a single Supercial

Deposits unit for interpolation
G Model
4
ARTICLE IN PRESS
Fig. 2. Location map of the Wairakei Geothermal Field, as dened by the resistivity boundary (5 Ohm-m inner boundary; Risk, 1984), showing geothermal wells, surface faults
(Grindley, 1961; GNS Active Fault Database http://data.gns.cri.nz/af/) and the boundaries of the temperature and stratigraphy models presented in this study. Abandoned
wells (from which temperature and stratigraphy data are also available) not shown.
on average at ca. 600 m depth beneath the Western Bore Field, and
sharply deepens to the west and east, being logged at ca. 1450 m
depth in well WK301 (reinjection area; Fig. 2) directly overlying
Tahorakuri Formation. To the east of WK301, the base of Waiora
Formation becomes shallower (ca. 850 m depth in wells WK305
and WK307, and ca. 600 m depth in wells WK314 and WK315;
Fig. 2), but is much deeper again, south-eastwards into the Tauhara
geothermal eld.
The pre-volcanic basement, which is mostly made up of Mesozoic greywacke in the central TVZ (Wood et al., 2001; Mortimer,
2004), has not been intersected by drill-holes in Wairakei yet,
and the deepest wells drilled in the Te Mihi and reinjection areas
(WK247; WK317; Fig. 2) indicate the basement there is at least
2.9 km below surface. In the eastern side of Tauhara, well TH17
intersected greywacke basement at a depth of ca. 2000 m. This is
consistent with similar drilling ndings from other eastern elds
(e.g., Rotokawa, Ohaaki, Kawerau), and validates the general correlation of high residual gravity anomaly values and relatively
shallow greywacke basement, to the east of the TVZ (Fig. 1).
The predominance of NE-trending (and subordinate WNWtrending) surface fault expressions at the Wairakei Field matches
the broader tectonic grain of the Taupo Volcanic Zone and is consistent with NW-SE directed widening of the central TVZ (Grindley,
1961; Rowland and Sibson, 2001; Fig. 2). Interpreted drill-hole data
(e.g. stratigraphic offsets) and structural imaging via acoustic logs
(e.g. McLean and McNamara, 2011) show active faults are dominantly NE-striking, normal and dip steeply (6080 ). Earthquake
hypocentres in the central TVZ (including the Wairakei area) plot
as deep as 69 km where the brittle-ductile transition is inferred
to occur (Bryan et al., 1999; Sherburn et al., 2003; Bannister et al.,
2004; Harrison and White, 2004). Faulting is therefore likely to be
active through the entire depth extent of the Wairakei-Geothermal
system.
3. Data and methods
In this study, geostatistical models of temperature and stratigraphy were computed using the Kriging technique (e.g., Olea,
1999). Temperature predictions were obtained by use of ordinary Kriging (referred to as Kriging) and universal Kriging,
and stratigraphy predictions were computed by use of indicator Kriging. In simple terms, Kriging and indicator Kriging are
used for spatial prediction of continuous (numeric data) and
categorical variables, respectively. All Kriging algorithms were
implemented in mathematical modeling software package MATLAB (http://www.mathworks.com). A detailed description of the
G Model
ARTICLE IN PRESS
Fig. 3. Temperature data, including drill-hole temperature and thermal manifestations (1950present), and boundary conditions used for this study. Map projection: New
Zealand Map Grid.
Kriging algorithms is beyond the scope of this paper, but general

overview is provided in Section 3.2. For more details on the Kriging
algorithms, as implemented here, readers can refer to Olea (1999)
and Deutsch and Journel (1992).
The following steps were undertaken as part of geostatistical modeling: drill-hole data validation, exploratory data analysis
(anisotropy and trends), analysis of geoscience/reservoir information to assist interpretation of anisotropy and trends, and spatial
prediction.
3.1. Drill-hole data validation
Temperature and stratigraphy drill-hole data used in this study
include more than 200 geothermal wells covering an area of
9 km 7 km and a 3 km depth range (Fig. 3). For the purpose of
geostatistical modeling, vertical resolution of drill-hole temperature data was standardized to 10 m (this means, datasets collected
at higher resolution were reduced in size). Horizontal resolution
is comparatively poorer and constrained by well spacing (from
50 m in densely drilled areas to several km in peripheral or deep
areas). Measured temperatures were used as a proxy for formation
temperatures. Measured temperatures can be affected by a series
of artifacts which were minimized or eliminated when possible.
Examples include lack of thermal equilibration, effects induced by
production, and internal ow or development of shallow steam
zones.
In order to minimise artifacts due to lack of thermal equilibrium, we selected downhole temperature logs taken 28 or more
days after well completion. Worth noting is temperature logs

may potentially equilibrate over longer periods (on the order of
months; Horner Method; Dowdle and Cobb, 1975; Verma et al.,
2006).
Measured temperatures can be affected by extraction (e.g., as
an indirect effect of pressure drawdown, or cooling due to ineld
injection). Bixley et al. (2009) documented a temperature change
up to 15 C in the production areas of Wairakei as a result of
uid extraction, the bulk of which took place during the period
19601970. Accordingly, post-1970, temperature logs were preferentially selected to minimize artifacts due to temperature changes
with time.
The effects of internal ow or development of shallow steam
zones were assessed on a well-by-well basis and removed where
identied. Discussion on how to recognize these artifacts is beyond
the scope of this paper, but readers can refer to Grant and Bixley
(2011) for guidelines.
The following boundary conditions (Fig. 3) were used to constrain Kriging models:
Surface temperature from thermal manifestations (GNS
database). These data were merged into input drill-hole dataset
for Kriging predictions
Minimum temperature of 10 C, assigned at selected topographic
locations (except for areas within 250 m of thermal manifestations or drill-hole data) using a coarse 500 500 m regular grid
(Fig. 3). These data were merged into input drill-hole dataset
for Kriging predictions. This minimum temperature was also
ARTICLE IN PRESS
G Model
6
implemented as post-prediction correction (i.e., all values below

10 C were converted to 10 C)
Maximum temperature of 270 C (Bixley et al., 2009). This was
implemented as a post-prediction correction (i.e., all values above
270 were converted to 270 C).
It is noted that while interpreted temperatures (input for Kriging models) are eventually corrected for the artifacts above, there
always remains a degree of subjectivity in the interpretations.
Stratigraphy can similarly be subject to uncertainty, mainly because
of its subjective nature (interpretation) and also due to the difculties and limitations inherent to the recovery of geological
samples during drilling. The majority of the drill-hole material used
in Wairakei for stratigraphic descriptions is cuttings (95% versus
5% remaining core), which is commonly intensely hydrothermally
altered, generally undated, and locally discontinuous due to blind
drilling. Accordingly, many stratigraphic denitions presented here
are still regarded as provisional, and Kriging models of stratigraphy
presented here may be potentially subject to revision.
where (h) is the variogram, T is temperature, h is the lag distance

vector (separation between data pairs), n(h) is the number of data
pairs separated by h, and Xi = (xi , yi , zi ) is the location of the sampling
point i. A graphical representation of Eq. (4) is the (h) versus h
plot, referred to as the empirical variogram (examples discussed in
Section 4.1).
Temperature increases with depth in response to the prevailing thermal regime. In geostatistical terms, temperature is said
to be a non-stationary variable characterised by a vertical drift.
In a conductive thermal regime, the vertical drift can be approximated to a linear trend, whereas convective regimes (typically
high-temperature geothermal systems like Wairakei), the vertical
drift will be governed to a large extent by the boiling-depth-point
(BDP) curve. The mathematical treatment of a drift in Kriging consists of adding a further constraint to the Lagrange optimisation
problem discussed in Eqs. (1)(3), as follows:
2
L(i , ) = (Xo ) + 2o

n
3.2. Kriging applied to temperature
+2
Kriging is an exact interpolator, meaning data are honored

at sample locations. Kriging also has the ability to identify and
minimize screening and clustering effects (e.g., wells in a line or
clustered locally). All these properties are relevant to the study of
drill-hole datasets, which tend to be highly scattered in 3D space.
Kriging also can handle anisotropy and spatial trends inherent
to data. Variogram analysis, sometimes referred to as exploratory
data analysis (EDA), is an important step in geostatistical analysis
where principal directions of anisotropy and the presence of a trend
can be analyzed (EDA is discussed in more detail in Section 3.3).
Kriging is based on a generalized form of linear regression, which
in the case of temperature takes the form (Olea, 1999):
T (Xo ) =
n
Ti i
(1)
i=1
where Ti = measured temperatures at the sample points i = 1. . .n,

i = weight coefcients, and T(Xo ) = prediction of temperature at the
location Xo = (xo , yo , zo ), with i coefcients meeting the condition:
n
i = 1
(2)
i=1
Kriging minimizes the variance of the estimation error 2 (Xo )

which can be expressed as a function of i coefcients. In practical terms, variance is a measure of spatial variability. From a
mathematical point of view, the Kriging problem is a constrained
optimisation, this is, minimising the variance subject to a constraint
[condition (2)] which is solved through the Lagrange method of
multipliers (Olea, 1999, and references therein):
L(i , ) = 2 (Xo ) + 2
n
i = 1
(3)
i=1
where L(i ,) is the Lagrangian function and is a Lagrange multiplier.

The variance is minimised using a minimum square error
approach, for which the following empirical estimator of the variance 2 (Xo ) is adopted (Olea, 1999),
1
2
[T (Xi + h) T (Xi )]
2n(h)
n(h)
(h) =
i1
(4)
j1
j
n
n
i 1
i=1
i fj (Xi ) fi (Xo )
(5)
i1
where fj is a polynomial function representing the drift (Olea, 1999).

The problem in Eq. (5) is usually referred to as a universal Kriging
problem. Conceptually, Kriging models are composed by a residual random function (the prediction) plus a deterministic function
fj (the drift). In conventional Kriging (Eq. (3)), the drift is constant
and unknown, and in universal Kriging (Eq. (5)), the drift is variable
and modeled. Universal Kriging shares all the properties of conventional Kriging: it is a minimum square error, exact interpolator
that automatically corrects for clustering in the sampling, and the
observations take weights under a screen effect (Olea, 1999). In this
study, both Kriging and universal Kriging predictions are shown for
comparative purposes, although universal Kriging predictions are
ultimately adopted for prediction of temperature at Wairakei (see
Section 4.3 for details).
3.3. Kriging applied to stratigraphy
In indicator Kriging, the m available stratigraphic units of a drillhole dataset are categorized with an arbitrary integer kj (j = 1,. . .,m).
By denition, a categorical variable is assigned discrete values as
opposed to continuous numeric variables (e.g. temperature). With
stratigraphic data being reduced to categories, the following binary
transformation is applied to stratigraphic data:

Ij (Xi ) =
1 if K(Xi ) = Kj
0
(6)
otherwise
where Ij (Xi ) is the indicator transform of the stratigraphic unit j at

the sampling point Xi = (xi , yi , zi ) and k(Xi ) is the stratigraphic class
observed at Xi , with i = 1,. . .,n (n is the number of sampling points).
Basically, Ij (Xi ) = 1 where kj is known to be present, and Ij (Xi ) = 0
where kj is known to be absent. Note that unlike temperature,
where data is available nominally at 10 m vertical resolution, stratigraphic data is of variable vertical resolution, which is determined
by the location of all stratigraphic contacts of the stratigraphy
class kj .
In the case of categorical variables (i.e., stratigraphy), a variogram model IMj may be calculated (subject to data availability)
and tted to each empirical indicator variogram of the categories
kj , j = 1,. . .,m (Eq. (1)). Such variogram models are used as a proxy
for the local probability distribution of the stratigraphic class kj .
At every interpolation location Xo , the stratigraphic class with the
G Model
ARTICLE IN PRESS
highest probability is assigned, this is:
K(Xo ) =
k1 if IM1 (Xo ) = max{IM1 (Xo ), IM2 (Xo ), . . . , IMm (Xo )}
k2 if IM2 (Xo ) = max{IM1 (Xo ), IM2 (Xo ), . . . , IMm (Xo )}
..
.
km
if IMm (Xo ) = max{IM1 (Xo ), IM2 (Xo ), . . . , IMm (Xo )}

(7)
where K(Xo ) is the predicted stratigraphic class at Xo . The fact that

Indicator Kriging is an exact interpolator implies that a value of
Ij (Xo ) = 1 equivalent to 100% outcome probability is assigned at
Xo = Xi , provided Ij (Xi ) = 1.
3.4. Exploratory data analysis
Empirical variograms are used to quantify the average variation
of an attribute (such as temperature) as a function of lag distance
and direction. Changes in direction producing signicant changes
in the empirical variograms are generally interpreted in terms of
anisotropy. The mathematical treatment of anisotropy in Kriging
is beyond the scope of this paper, but readers can refer to Deutsch
and Journel (1992) and Leuangthong et al. (2008) for details.
Although variogram analysis is a 3D problem and all directions
should be eventually studied, in practice, variogram directions and
lag intervals can be chosen selectively subject to the condition that
all relevant directions be identied and reasonably characterised,
Lag intervals/increments be large enough to contain a statistically
representative number of pairs (the minimum statistical condence is set at 100 pairs here at least 30 pairs per lag interval is
recommended by Olea, 1999), and lag intervals be small enough so
that empirical variograms capture relevant variations for variogram
tting (discussed below).
In areas/directions with poor data coverage, there is usually a
trade-off between statistical condence (i.e., number of data pairs
at each lag interval) and level of detail (i.e., number of lag intervals)
of variograms.
The generalized use of variograms in geostatistics is supported
by the following empirical observations:
1) Two adjacent points tend to show similar attribute values
whereas two distant points tend to show greater variation in
these values (also known as Toblers Law; Tobler, 1970). From a
geostatistical point of view, the correlation of two data points is
proportional to the separation distance between them.
2) Under stationary conditions (this is, trend or drift not present;
see Section 4.1 for discussion on drift), the variation in the
attribute values increases up to a certain critical lag distance, referred to as range, beyond which variance reaches a
plateau referred to as sill. From a statistical point of view,
sampling points separated by a distance greater than the range
are uncorrelated.
Variance is estimated empirically using available data and then
it is mathematically modeled by tting a variogram model. The
latter is a mathematical function of lag distance, sill, range and
direction. Commonly used variogram models include spherical,
exponential and Gaussian (Deutsch and Journel, 1992; Olea, 1999).
Trial and error is a common practice for variogram tting (Olea,
1999). Because of point 2) above, a good variogram t at small lag
distances, and a poor t at big lag distances is better than a moderate
t at all lag distances.
The utility of variogram analysis in Kriging as to characterizing
anisotropy and spatial trends is limited by available data. Variograms may not be statistically representative in peripheral or deep
areas of a study area due to relative lack of data. In the absence

of representative variograms, available geoscientic and reservoir
data can provide bounds to geostatistical models. In this context,
surface fault lineaments (both regional and local) and gravity data
were analyzed in this study to better characterize the structural
setting of the Wairakei eld, with emphasis on the implications for
anisotropy analysis (see Section 4.1 for details).
Regional structural lineaments were obtained through the analysis of a digital terrain model of the TVZ (25 model spatial
resolution). Local structures were mapped in the vicinity of the
Wairakei Geothermal Field (Fig. 6). Fractures, faults, and veins were
recognized in exposures of Huka Falls Formation and supercial
deposits (Oruanui Formation and Taupo Formation). A description
of these formations can be found in Rosenberg et al. (2009) and
Wilson (2001).
In this study, residual gravity anomalies of Wairakei (Hunt,
1991) were used for identication of subsurface structural trends
and further examined via gradient analysis, with sharp gravimetric
variations being used as a proxy for the delineation of subsurface
structures.
4. Results and discussion
4.1. Anisotropy and drift analysis of temperature
Vertical and horizontal directions are regarded as the most
useful for variogram analysis because geological controls on temperature are mainly represented by stratigraphy and structures,
meaning principal directions of correlation are likely to be subhorizontal and subvertical, respectively. This holds valid for a young
extensional regime like the TVZ, but it may not hold true for
other geological settings. Also, temperature is affected by a vertical increase with depth, or vertical drift, meaning the vertical and
horizontal directions are best suited to characterise main drift and
drift-free directions, respectively. This holds valid for all geothermal settings.
Fig. 4 shows horizontal variograms of temperature in the directions EW, NE, NS, NW, and omnidirectional. It follows from this
gure that differences in direction do not produce signicant
changes among the horizontal variograms. This poses an apparent discrepancy between variogram analysis and surface fault data,
the latter supporting a dominant NE-trending direction of correlation. In this context, geological and reservoir engineering evidence,
including existing and new data, are briey described below, both
on regional and local scale, which in view of the authors supports complex anisotropy patterns (i.e., other than NE-trending)
at Wairakei.
4.1.1. New and existing structural evidence
Rose diagrams from regional oblique lineaments show systematic WNW- and NNW-trending populations (this study; Fig. 6).
Other regional structural studies (Rowland and Sibson, 2001, 2004;
Acocella et al., 2003; Rowland et al., 2010) describe the TVZ as a
segmented rift, with a series of transitional zones or accommodation zones, which may be manifested through local discontinuities
in NE-trending fault activity, and/or seismic gaps (Fig. 1), and/or
occurrence of structures (faults and lineaments) running roughly
perpendicular to the NE-trending rift axis (Fig. 7). These studies
show the Wairakei geothermal eld may be spatially associated
with stretched and magma-intruded crust in one such accommodation zone (Fig. 1). At the scale of the Wairakei area, other
indicators favour the existence of oblique structures. These include
rose diagrams of orientations of local fractures, faults, and veins
around the Wairakei geothermal eld (from this study) as shown
in Fig. 6. Veins, which are fossil remnants of structural pathways
G Model
8
ARTICLE IN PRESS
Fig. 4. Directional variograms obtained along selected horizontal directions using temperature data from Wairakei.
for hydrothermal uid ow, follow not only a NE-trending, but

also a preferential EW-trending direction. Also, some documented
faults in the southern part of Wairakei trend WNW (Grindley, 1961;
Fig. 2), and, thermal manifestations of Lake Taupo (10 km south of
Wairakei) are strongly aligned in a NW direction (de Ronde et al.,
2002).
4.1.2. Existing and revised gravity data
In general, sharp variations in residual gravity anomalies tend
to highlight structural discontinuities associated with density
changes, generally, changes in basement geometry (basement

being denser than volcaniclastic cover), or borders of basin structures (e.g., tectonic grabens, calderas) within which relatively
low-density volcanoclastic deposits accumulate. In the TVZ, negative residual anomalies are commonly interpreted as indicative of
buried caldera and many geothermal systems sit within or near the
edge of such calderas, which tend to be of irregular strike (Bibby
et al., 1995; Fig. 1). Detailed gravity mapping at Wairakei (Hunt,
1991; Hunt et al., 2009; Fig. 7) suggests both NE and NW-trending
discontinuities are present at subsurface (this study; Fig. 7B).
G Model
ARTICLE IN PRESS
Fig. 5. (A) Ominidirectional horizontal variograms for <1000 mGL and >1000 mGL depth intervals; (B) Vertical variogram of temperature, along with synthetic BDP and linear
vertical variograms.
Fig. 6. Map of structural lineaments of the TVZ. In the upper-left corner, rose diagrams show dominant directional trends of regional lineaments, with lineaments weighted
by length. In the lower-right corner, rose diagrams show dominant directional trend of faults, fractures, and veins (weighted by frequency) exposed in the vicinity of the
Wairakei area (eld data locations shown). Abbreviations of geothermal areas (red labels) as in Fig. 1.
ARTICLE IN PRESS
G Model
10
Fig. 7. (A) Residual gravity map of Wairakei (Hunt, 1991); (B) map of rst-derivative of residual anomalies, showing areas of steep gravimetric gradient (>10 mGal/km; areas
in yellow, orange and red) as proxies for subsurface structures (this study). Map projection: New Zealand Map Grid.
4.1.3. Reservoir studies

The hydrological connection between the Wairakei and Tauhara
geothermal elds, and overall NW-SE disposition of the greater
Wairakei-Tauhara system, supports a NW-SE connectivity of the
reservoir (Bixley et al., 2009). Reservoir modeling and simulation
results point to dominantly horizontal (i.e., stratigraphic) permeability at shallow depths in Wairakei, which can potentially mask
shallow structural patterns (Mannington et al., 2004).
In the absence of conclusive evidence for anisotropy, horizontal
anisotropy is not adopted as it represents a potential bias to Kriging
models. Accordingly, thermal variance is modelled using a cubic
model of variance of the form:
(h) =

2
Co + C
Co + C
h
a
35
4
3
h
a
7
2
5
h
a
3
4
and McNamara (2011) noted a close correlation between feed

zones interpreted from completion tests [depths greater than 2000
meters below ground level (mGL)] and sub-vertical, open fractures
imaged through acoustic logs in wells WK404 and WK407. No clear
correlation with stratigraphy could be established.
7
h
a
h<a
(8)
ha
where h is the lag distance (m), a is the range (m), C is the sill (dimensionless) and Co is the nugget effect (microvariance at h 0), with
a = 6300, C = 4500 and Co = 0. The graphical representation of these
parameters is shown in Fig. 4E (omnidirectional variogram). It is
noted that the NS direction (Fig. 4A) stands out with a variance well
above the model variogram for lag distances greater than 4000 m.
As discussed in Section 3.2, however, the poor tting at large lag
distances is not a concern provided good t is achieved at small lag
distances (in this instance, <4000 m).
Fig. 4 provides an estimate of the average horizontal range (correlation) for temperature data. Horizontal correlation is dictated
by lateral connectivity in the reservoir, which in turn is a function of stratigraphic permeability. Reduction of primary porosity
with increasing depth has been documented in a range of geothermal elds, including Wairakei (e.g., Stern, 1982; Stimac et al.,
2004; Mannington et al., 2004), with the potential implication
of progressive reduction in stratigraphic permeability relative to
structural permeability. In agreement with this hypothesis, McLean
Fig. 8. Temperature proles from selected outeld wells, exemplifying the presence
of a dominantly linear drift from 0 to 600 mRL.
G Model
ARTICLE IN PRESS
11
Fig. 9. Examples of omnidirectional indicator variograms calculated along horizontal directions using binary-reduced stratigraphy data from Wairakei.
In order to test changes in the horizontal correlation of

temperature with depth at Wairakei, omnidirectional horizontal variograms were computed separately for depth intervals of
01000 mGL and 10003000 mGL (Fig. 5A). Note the number of
lag intervals of the 10003000 mGL variogram is less than that of
01000 mGL variogram. The reduction of lag intervals at greater
depths was adopted to increase statistical condence per lag
interval, although at the expense of loss of detail (Fig. 5A; see
Table 2 for details). The emerging 10003000 mGL variogram
remains highly noisy and only an indicative horizontal range of
ca. 2000 m (about a third of the horizontal range used to model
data in the 01000 mGL depth interval) is shown in Fig. 5A, to
account for the apparent lack of correlation beyond a lag distance
of 2000 m. Although the magnitude of the horizontal range that
best models data in the 10003000 mGL range is relatively difcult to constrain with accuracy comparable to the 01000 mGL
depth interval, the comparison of the two variograms clearly
points to an overall reduction of horizontal correlation with depth
(Fig. 5A).
The variogram in Eq. (8) cannot successfully model the vertical variogram of temperature (Fig. 5B). The continuous increase
of variance (lack of plateau or sill) that characterises this vertical
variogram is diagnostic of non-stationary variables affected by a
drift. At Wairakei, the vertical drift of temperature is a mixture
of two end members: linear-conductive (dominant outeld, e.g.,
Fig. 8) and boiling-depth-point (BDP) curve (dominant ineld;
Bixley et al., 2009). In order to incorporate the drift in Universal
Kriging, a polynomial order for the function f(Xi ) (as in Eq. (5))
must be specied. In this study, a 3rd order polynomial of the form
f(Xi ) = a0 + a1 z + a2 z2 + a3 z3 was found to adequately model the BDP
curve. The drift is handed automatically in the universal Kriging
systems of equations without the need of resorting to the estimation of the polynomial coefcients ai above. This means, Kriging
has the exibility to compute a 3rd degree polynomial in the presence of a drift locally dominated by the BDP end member, and a
1st degree (linear) polynomial where a conductive regime is dominant by selectively setting polynomial coefcients to zero (e.g.,
coefcients a2 and a3 above). The general approach to successfully
Fig. 10. Temperature distribution at z = 0 mRL from Kriging model, assuming isotropic variance. Map projection: New Zealand Map Grid.
G Model
ARTICLE IN PRESS
12
Table 2
Detail of omnidirectional, horizontal variogram for data in the 01000 mGL depth interval (plotted in Fig. 5A). Number of lag distances in class is used as a measure of statistical
condence in the lag interval. In general, the statistical condence for each lag interval (which is proportional to the number of classes) in the 10003000 mGL variogram
is remarkably lower (although above the 100 pair mark suggested by Olea, 1999) relative to data in the 01000 mGL depth interval (note: 10003000 mGL variogram with
reduced number of lag intervals plotted in Fig. 5A not shown here).
Lag class No.
Average lag distance h (m)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
125
375
625
875
1125
1375
1625
1875
2125
2375
2625
2875
3125
3375
3625
3875
4125
4375
4625
4875
5125
5375
5625
5875
6125
6375
6625
6875
7125
7375
Omnidirectional, horizontal variogram (01000 mGL)
Omnidirectional, horizontal variogram (10003000 mGL)
No. of distances in class
Semivariogram Value
No. of distances in class
Semivariogram Value
2178
9460
17,776
22,157
19,984
27,976
46,790
67,660
68,068
113,196
84,926
51,425
51,873
42,420
50,944
73,533
103,443
96,850
96,774
69,304
37,566
16,475
9618
16,811
31,904
30,587
30,408
10,059
6170
442
100
274
295
471
689
771
1467
1235
2012
2295
2685
2813
3320
3792
4087
4376
3941
4861
4482
4345
4564
5790
8297
3746
3857
3546
3908
4225
5879
4945
88
568
1083
979
1470
348
420
350
293
715
738
499
975
427
696
2948
2889
2808
983
4173
1394
672
759
775
1746
478
262
10
31
8
485
210
297
512
71
511
2419
956
1352
3811
5470
10,560
6876
994
1377
806
1627
2998
3181
4470
2566
6921
5998
5436
1386
4201
4195
18,403
2166
1088
model a variable drift is to set the polynomial order to t the most

complex known form of the drift.
The most challenging part of the universal Kriging problem is
the need to specify the variogram of the residuals (i.e., temperature
without drift), not the variogram of the regionalized variable (temperature + drift). A common way to estimate the variogram of the
residuals is by modelling the drift-free direction (horizontal direction in the case of temperature). In the case of temperature data
from Wairakei, it has been shown that the sill of the horizontal variogram is relatively constant over the entire vertical depth interval
of study (sill = 4500, Fig. 5A). A relatively safe assumption is therefore that the range of the vertical variogram of the residuals is the
same as that of the horizontal variogram. An independent, semiquantitative verication of this assumption is made by comparing
the vertical variogram of temperature against synthetic variograms
of linear trend (45 C/km) and BDP curve (Fig. 5B). This comparison shows the vertical variogram of temperature is in the expected
range of variance for a variogram with a superimposed vertical drift
represented by a combination of the two end members hypothesised above.
The range of the horizontal variogram decreases with depth
(Fig. 5A) posing some questions around the magnitude and
variability (i.e., constant versus variable) of the range of the vertical variogram of residuals. The following geoscientic analysis
is used to guide the choice of the vertical variogram of the
residuals:
Faults can be recognized from surface (surface fault traces)
through depths of ca. 20002500 m (acoustic logs; McLean and
McNamara, 2011) down to 69 km depth at the TVZ (distribution
of seismicity; Bryan et al., 1999). Based on this ubiquitous presence of faults in the vertical depth range under study (03 km),
subvertical connectivity of the reservoir (vertical range of variogram of residuals along z direction) can be assumed constant.
Stratigraphic permeability is variable in the vertical depth range
under study, being greater than vertical permeability at shallow depths (<1000 mGL; Mannington et al., 2004). This implies,
the average vertical range of the variogram of the residuals in
the vertical direction must be lower than 6300 m, which is the
prevailing average horizontal correlation at depths <1000 mGL
(Figs. 4E and 5A).
Horizontal and vertical correlation of temperature is similar at
depths >2000 m. This hypothesis relies on indirect evidence from
reservoir models which produce a reasonable match of measured
temperatures by adopting comparable horizontal and vertical
permeability towards the base of the models (Mannington et al.,
2004).
The above three conditions are satised by adopting a range of
2000 m for the vertical variogram of the residuals in the z direction.
Let d(Xo ) be the depth in meters of the prediction location Xo . An
anisotropy factor p can be dened as:
p(Xo ) =
1
2000 d(Xo )
2000
0
d(Xo ) < 1000 m

d(Xo ) 1000 m and d (Xo ) 2000 m
d(Xo ) > 2000 m
(9)
Kriging temperatures are ultimately calculated under variable

anisotropy as:
T (Xo ) = Tanisotropic (Xo )p(Xo ) + Tisotropic (Xo )(1 p(Xo ))
(10)
G Model
ARTICLE IN PRESS
13
Fig. 11. Kriging models of temperature for NE-trending cross section extending from points A to B, as shown in Fig. 10. (A) Isotropic thermal structure, no drift; (B) Isotropic
thermal structure, drift (3rd order polynomial); (C) Anisotropic thermal structure (horizontal to vertical correlation of 3:1), no drift; (D) Anisotropic thermal structure
(horizontal to vertical correlation of 3:1), drift (3rd order polynomial); (E) Temperature model using anisotropy as in (D) from 0 to 1000 mGL, isotropy as in (B) from 1000
to 3000 mGL, and variable anisotropy between 1000 and 2000 mGL, computed as mixing line between (B) and (D) using anisotropy coefcient as a function of depth (1 at
1000 mGL, 0 at 2000 mGL).
where Tanisotropic is the Kriging prediction of temperature for an

anisotropic model (vertical to horizontal correlation of 1:3) with
vertical drift, and, Tisotropic is the Kriging prediction of temperature
for an isotropic model (vertical to horizontal correlation of 1:1) with
vertical drift.
4.2. Stratigraphical anisotropy

Both vertical indicator variograms and horizontal, omnidirectional indicator varigorams were computed for all stratigraphic
units. Based on preliminary analysis (not shown), a signicant proportion of indicator variograms were regarded as lacking statistical
validity (applying criteria as in Section 3.2). Detailed, horizontal
directional analysis was not performed, partly due to the anticipated limitations in statistical validity, but also due to lacking
geological justication. Regarding the latter, a well-established fact
at the TVZ and Wairakei is volcanic and sedimentary stratigraphic
units are by their origin and nature of emplacement, sheet-like in
morphology. The predominantly rhyolitic lava bodies at Wairakei
have higher aspect ratios than pyroclastic or volcaniclastic strata,
but are still more extensive laterally than vertically (Rosenberg
et al., 2009).
Following this observation, statistically sound indicator variograms were identied (examples in Fig. 9) and used to derive
representative horizontal ranges, which were extrapolated to other
stratigraphic units. The ratio of horizontal to vertical correlation
(this is, the ratio between horizontal range axy and vertical range
az ) was approximated using general thickness relationships. In general, horizontal to vertical correlation ratios used in this study
varied from az /axy ratio of 1:4 (typically used for rhyolites such as
Karapiti 2a; Fig. 6B) to az /axy ratio of 1:8 (typically used for regional
stratigraphic units such as Huka Falls Formation; Fig. 9A).
4.3. Spatial prediction

The interpretation of anisotropy can be supported by combined variogram analysis and geoscience. In practice, prediction
models can also be used to assist interpretations of variograms
and anisotropy. In other words, model variograms are used to
constrain Kriging models and Kriging models are used to constrain model variograms. Although this may sound like a circular
argument, a preliminary prediction model (e.g., horizontal map
of temperature) computed under the deliberate assumption of
isotropy is a rst, necessary and practical step to pre-assess the
G Model
14
ARTICLE IN PRESS
Fig. 12. Kriging models of temperature for NW-trending cross section extending from points C to D, as shown in Fig. 10. Explanations as in Fig. 11.
presence of anisotropy (specically, directional anisotropy in the

horizontal plane).
Geoscientic evidence was already revealing a complex pattern
of anisotropy at Wairakei (Section 4.1). Kriging temperatures at z = 0
meters below sea level (approximately 450 mGL; Fig. 10), obtained
under the assumption of isotropic horizontal variance (as in Eq.
(7)), provide further insight into what anisotropy patterns look like.
High temperature regions of Wairakei tend to concentrate along
NE-trending (e.g., Te Mihi area) and NW-trending directions (Eastern Bore Field). Strictly speaking, temperature data at Wairakei are
characterized by two directions of maximum correlation which are
mutually perpendicular (at the depth of ca. 450 m). Whether it is
an isotropic thermal structure or a thermal structure with more
than one direction of correlation, these scenarios are potentially
undistinguishable in variogram analysis (Fig. 4). Following these
observations, the isotropic model of variance used for Kriging models of temperature is a practical approach, although it does not
necessarily imply the existence of an underlying isotropic thermal
structure.
A sharp, NW-trending thermal boundary is evident to the northeast of Wairakei, closely coinciding with the resistivity boundary
(Fig. 10). The close spatial correlation between thermal features from Geyser Valley (location as in Figs. 3 and 10) and the
NW-trending boundary indicates the presence of a subvertical,

peripheral leakage from the geothermal system to surface, but at
the same time a NW-trending barrier for lateral ow. The nature
of the NW-trending boundary is unknown, but it is likely to put in
contact permeable units (ineld) and impermeable units (outeld).
Well WK401 shows a peak of 150 C at ca. 400 mGL followed by
a temperature inversion. This is an indication of a thermal outow
sourced from Wairakei and owing to the south which leaks to the
surface at Karapiti thermal area (about 1 km north of WK401). It is
noted this thermal outow is not observed in adjacent wells WK404
and WK407 (immediately south-east of WK401; Figs. 3 and 10).
Figs. 11 and 12 show temperature distribution along NE- and
NW-trending vertical cross sections (orientations in Fig. 10), based
on Kriging models. For comparative purposes, both Kriging (no
drift) and universal Kriging (drift) predictions are shown, for both
isotropic models, anisotropic models, and variable anisotropy (Eq.
(10)) models.
The lack of a vertical drift (Figs. 11A, C, 12A and C) produces
unrealistic temperature inversions at depths >2500 mGL, providing a validation for the use of a vertical drift and illustrating how
inadequate geostatistical constraints can produce misleading predictions. In similar lines, the most accepted conceptual model for
Wairakei is that Eastern Bore Field is an outow structure sourced
G Model
ARTICLE IN PRESS
15
Fig. 13. (A) Kriging models of stratigraphy for NE-trending cross section extending from points A to B, as shown in Fig. 10; (B) Interpreted stratigraphy and structure based
on (A).
from Te Mihi upow (e.g., Bixley et al., 2009; Glover and Mroczek,
2009). Note the isotropic model in Fig. 12B fails to capture the
connectivity between Eastern Bore Field and Te Mihi.
Figs. 13A and 14A show Kriging models of stratigraphy for
the cross sections above (same directions as in Figs. 11 and 12).
Each Kriging model is used to interpret stratigraphy and structure
(Figs. 13B and 14B, respectively). In this context, Kriging models of
stratigraphy are not a nished product, but a guide for interpretation. Interpretations are primarily based on Kriging predictions,
but some assumptions are also made on the basis of gravity (Fig. 7).
In particular, the large scale variations of gravity are interpreted
in terms of basement depth, and these variations indicate deepening of the basement towards the northwest (i.e., decrease of
residual anomalies to the northwest; Fig. 7A). Even though basement is not depicted in the geological cross sections, relatively deep
stratigraphic units like Waiora Formation are likely to mimic to
some extent the geometry of the basement. As a result, the base
of Waiora Formation is interpreted to deepen to the northwest
following the increase of negative residual anomalies (left hand

side of NW-trending cross section; Fig. 14B).
Figs. 13B and 14B are used to support interpretations in terms of
geological controls on thermal discontinuities across Wairakei. For
this purpose, Kriging models of temperature are superimposed on
interpreted geology (Figs. 15 and 16). Fig. 15 (NE-trending cross section) shows the Eastern Bore Field outow is primarily controlled
by stratigraphy, particularly, the interface between Wairakei Ignimbrite and Waiora Formation. In this area, Waiora Valley Andesite
is also a relevant stratigraphic unit and likely to contribute to
stratigraphic permeability (with permeability concentrated along
brecciated margins). Other outow structures are evident which
tend to follow Karapiti 2b Rhyolite and Karapiti 3 (Figs. 15 and 16).
Karapiti 2b Rhyolite is a drilling target for mid-depth production
and it is interpreted that most lateral ows concentrate in the brecciated margins of the rhyolite. To date, it has not been conclusively
resolved whether Karapiti 2a and 3 are different or equivalent ow
units.
G Model
16
ARTICLE IN PRESS
Fig. 14. (A) Kriging models of stratigraphy for NW-trending cross section extending from points C to D, as shown in Fig. 10; (B). Interpreted stratigraphy and structure based
on (A).
The variations in thickness of Wairakei Ignimbrite across

Wairakei, particularly at the transition from Te Mihi-Western Boreeld (from wells WK219 to WK212), can be explained in terms of a
piecemeal caldera, as dened by Cole et al. (2005). According to
these authors, piecemeal calderas are characterized by multiple
localized collapses, leading to intra-caldera deposits of highly variable thickness and irregular caldera oors. In areas where Wairakei
Ignimbrite is completely absent (e.g., WK317 area), it is not clear
whether caldera geometry or post-Whakamaru caldera events
obliterating Wairakei Ignimbrite are the cause of such a stratigraphic discontinuity. Whatever the case may be, Fig. 16 shows
how puzzling deep architecture remains to be at Wairakei.
Signicant stratigraphic offsets occur in the vicinity of the Te
Mihi upow, which support the existence of fault zones as likely
conduits for vertical permeability (Figs. 15 and 16). Fig. 16 also
portrays high temperature regions at depth in the WK317 area.

Kriging models suggest a deep outow structure which connects
with the shallower Eastern Bore Field outow. This connection is
likely to be an artifact of the Kriging predictions.
Relatively shallow (7501000 mGL) temperatures inversions
have been found in Te Mihi, both in the WK243 area (four wells
drilled off this pad, namely, WK243WK246) and in WK257. Based
on Kriging predictions, temperature inversions are interpreted in
terms of cold inows from steam-heated aquifers hosted at the base
of Waiora Formation, and top of rhyolites Karapiti 2b and/or Karapiti 3, which permeate into deeper stratigraphic levels through a
combination of intra-formational permeability (dominant in the
WK243 area) and faults (dominant at WK257; Fig. 16). All cold
inows are interpreted to be driven by pressure drawdown associated with long-term production at Wairakei.
G Model
ARTICLE IN PRESS
17
Fig. 15. NE-trending cross section with superposition of Kriging model of temperature (Fig. 11E) and interpreted geology (Fig. 13B).
Fig. 16. NW-trending cross section with superposition of Kriging model of temperature (Fig. 12E) and interpreted geology (Fig. 14B).
5. Conclusions
In this paper multidisciplinary drill-hole datasets and cumulative geoscientic and reservoir knowledge from Wairaikei have
been used to test the applicability of geostatistical techniques
for characterizing the permeability and temperature in Wairaikei
geothermal reservoir. The geoscience and reservoir data provide
constraints to geostatistical models, particularly in relation to interpretation of variograms, anisotropy and drift. The analysis has
illustrated the geothermal reservoir complexities at Wairakei in
terms of temperature anisotropy, and provided a methodology to
handle multiple anisotropy. Results show that kriging predictions
of temperature can realistically reect anisotropy variations with
depth from dominantly anisotropic at shallow depths (<1000 m),
to dominantly isotropic at greater depths (>3000 m). Based on geostatistical models, two cross sections have been presented in this
study, which provide a new eld wide understanding of the correlation of stratigraphy, faults, temperature, and uid ow paths.
A potential limitation of the geostatistical methods and models

presented here is model condence that is limited by data distribution. This implies interpretations carry generally low condence
in peripheral and/or deep areas, In addition, uncertainty in source
data (e.g., temperature artifacts; subjectivity of stratigraphic interpretations) is not quantiable by Kriging.
Acknowledgements
This work was funded by VCUDF Grant No. 23278 (University
of Auckland, New Zealand), Foundation for Research, Science and
Technology (projects PROJ-20199-GEO-GNS and UOAX0713
University of Auckland) and Contact Energy. Thanks are extended
to Michael OSullivan, Angus Yeh, and Juliet Newson (Dept. of
Engineering Science, University of Auckland) for their guidance
in the compilation of temperature data. We also acknowledge the
New Zealand GEONET project, and its sponsors EQC and Ministry of
Science and Innovation (formerly FRST) for providing earthquake
G Model
18
ARTICLE IN PRESS
data and GNS Science for providing fault (New Zealand Active
Faults Database) and thermal manifestations data. Special thanks
also to anonymous reviewers for their useful comments.
References
Acocella, V., Spinks, K., Cole, J., Nicol, A., 2003. Oblique back arc rifting of Taupo
Volcanic zone, New Zealand. Tectonics 22 (4), 1921.
Bannister, S., Bryan, C.J., Bibby, H.M., 2004. Shear wave velocity variations across the
Taupo Volcanic Zone, New Zealand, from receiver functions. Geophysica Journal
International 159, 291310.
Bibby, H.M., Caldwell, T.G., Davey, F.J., Webb, T.H., 1995. Geophysical evidence on the
structure of the Taupo Volcanic Zone and its hydrothermal circulation. Journal
of Volcanology and Geothermal Research 68, 2958.
Bibby, H.M., Caldwell, T.G., Risk, G.F., 1998. Electrical resistivity image of the upper
crust within the Taupo Volcanic Zone, New Zealand. Journal of Geophysical
Research 103 (B5), 96659680.
Bignall, G., Milicich, S.D., Ramirez, L.E., Rosenberg, M.D., Kilgour, G.N., Rae, A.J., 2010.
Geology of the Wairakei-Tauhara geothermal system, New Zealand. In: Proceedings of the 2010 World Geothermal Congress, April 2530, Bali, Indonesia, p.
8.
Bixley, P.B., Clothworthy, A.W., Mannington, W., 2009. Evolution of the Wairakei
Geothermal Reservoir during 50 yr of production. Geothermics 38, 145154.
Brown, S.J.A., Wilson, C.J.N., Cole, J.W., Wooden, J., 1998. The Whakamaru group ignimbrites, Taupo Volcanic Zone, New Zealand: evidence for reverse tapping of a
zoned silicic magmatic system. Journal of Volcanology and Geothermal Research
84, 137.
Browne, P.R.L., Graham, I.J., Wood, C.P., 1992. Subsurface andesite lavas and plutonic rocks in the Rotokawa and Ngatamariki geothermal systems, Taupo
Volcanic Zone, New Zealand. Journal of Volcanology and Geothermal Research
51, 199215.
Bryan, C.J., Sherburn, S., Bibby, H.M., Bannister, S.C., Hurst, A.W., 1999. Shallow seismicity of the central Taupo Volcanic Zone, New Zealand: its distribution and
nature. New Zealand Journal of Geology and Geophysics 42, 533542.
Cole, J.W., Milner, D.M., Spinks, K.D., 2005. Calderas and caldera structures: a review.
Earth-Science Reviews 69, 196.
de Ronde, C.E.J., Stoffers, P., Garbe-Schnberg, D., Christenson, B.W., Jones, B.,
Manconi, R., Browne, P.R.L., Hissman, K., Botz, R., Davy, B.W., Schmitt, M., Battershill, C.N., 2002. Discovery of active hydrothermal venting in Lake Taupo, New
Zealand. Journal of Volcanology and Geothermal Research 115, 257275.
Deutsch, C., Journel, A., 1992. GSLIB: Geoestatistical Software Library and Users
Guide. Oxford University Press, New York, p. 369.
Dowdle, W.L., Cobb, W.M., 1975. Static formation temperature from well logs an
empirical method. Journal of Petroleum Technology 11, 13261330.
Fabbri, P., 2001. Probabilistic Assessment of Temperature in the Euganean Geothermal Area (Veneto Region, NE Italy). Mathematical Geology 33/6, 745760.
Glover, R.B., Mroczek, E.K., 2009. Chemical changes in natural features and well
discharges in response to production at Wairakei, New Zealand. Geothermics
38 (1), 117133.
Grant, M.A., Bixley, P.B., 2011. Geothermal Reservoir Engineering, 2nd edition. Elsevier, p. 359.
Gravley, D.M., Wilson, C.J.N., Rosenberg, M.D., Leonard, G.S., 2006. The nature and age
of the Ohakuri Formation and Ohakuri Group in surface exposures and geothermal drill hole sequences in the central Taupo Volcanic Zone, New Zealand. New
Zealand Journal of Geology and Geophysics 49, 305308.
Gravley, D.M., Wilson, C.J.N., Leonard, G.S., Cole, J.W., 2007. Double trouble: paired
ignimbrite eruptions and collateral subsidence in the Taupo Volcanic Zone, New
Zealand. Geological Society of America Bulletin 119, doi:10.1130/B25924.1.
Grindley, G.W., 1961. Geological Map of New Zealand 1:63360, Sheet N94, Taupo.
Department of Scientic and Industrial Research, Wellington, New Zealand.
Grindley, G.W., 1965. The geology, structure and exploitation of the Wairakei
Geothermal Field, New Zealand. In: New Zealand Geological Survey Bulletin. s.
75. Department of Scientic and Industrial Research, Wellington, New Zealand.
Harrison, A.J., White, R.S., 2004. Crustal structure of the Taupo Volcanic Zone, New
Zealand: stretching and igneous intrusion. Geophysical Research Letters 31,
doi:10.1029/2004GL019885.
Healy, J., 1984. Wairakei Geothermal Field in review. In: Geothermal Circular
JH/10, New Zealand Geological Survey. Department of Scientic and Industrial
Research, p. 34.
Heise, W., Bibby, H.M., Caldwell, T.G., Bannister, S., Ogawa, Y., Takakura,
S., Uchida, T., 2007. Melt distribution beneath a young continental rift:
the Taupo Volcanic Zone, New Zealand. Geophysical Research Letters 34,
doi:10.1029/2007GL029629.
Houghton, B.F., Wilson, C.J.N., McWilliams, M.O., Lanphere, M.A., Weaver, S.D.,
Briggs, R.M., Pringle, M.S., 1995. Chronology and dynamics of a large silicic magmatic system: Central Taupo Volcanic Zone, New Zealand. Geology 23, 1316.
Hunt, T.M., 1991. Gravity Anomaly Map of Wairakei Geothermal Field. Department
of Scientic and Industrial Research, Geology & Geophysics Contract Report
1991/1, p. 23.
Hunt, T.M., Bromley, C.J., Risk, G.F., Sherburn, S., Soengkono, S., 2009. Geophysical
investigations of the Wairakei Field. Geothermics 38 (1), 8597.
Lamarche, G., Barnes, P.M., Bull, J.M., 2006. Faulting and extension rate over the last
20,000 years in the offshore Whakatane Graben, New Zealand continental shelf.
Tectonics 25 (4), TC4005 (24 pp).
Leuangthong, O., Khan, K.D., Deutsch, C.V., 2008. Solved Problems in Geostatistics.
John Wiley and Sons, p. 205.
Mannington, W., OSullivan, M., Bullivant, D., 2004. Computer modelling of the
Wairakei-Tauhara geothermal system, New Zealand. Geothermics 33, 401
419.
McLean, K., McNamara, D., 2011. Fractures interpreted from acoustic formation
imaging technology: correlation to permeability. In: Proceedings 36th Workshop on Geothermal Reservoir Engineering, Stanford University, Stanford,
California, USA, SGP-TR-191, pp. 103112.
Milicich, S.D., Wilson, C.J.N., Bignall, G., Pezaro, B., Charlier, B.L.A., Wooden, J.L.,
Ireland, T.R., 2011. Buried rhyolite in the Kawerau Geothermal Field, Taupo
Volcanic Zone, New Zealand: sources of a rejuvenated geothermal system.
In: Proceedings New Zealand Geothermal Workshop, Auckland, New Zealand,
November, p. 5.
Mortimer, N., 2004. New Zealands geological foundations. Gondwana Research 7
(1), 261272.
Mouslopoulou, V., Nicol, A., Walsh, J.J., Beetham, D., Stagpoole, V., 2008. Quaternary
temporal stability of a regional strike-slip and rift fault intersection. Journal of
Structural Geology 30 (4), 451463.
Nicol, A., Walsh, J., Berryman, K., Villamor, P., 2006. Interdependence of fault displacement rates and paleoearthquakes in an active rift. Geology 34, 865868.
Olea, R., 1999. Geostatistics for Engineers and Earth Scientists. Kluwer Academic
Publishers, p. 301.
Risk, G.F., 1984. Electrical resistivity survey of the Wairakei geothermal eld. In:
Proceedings of 6th New Zealand Geothermal Workshop, Auckland, New Zealand,
vol. 6, pp. 123128.
Rosenberg, M.D., Bignall, G., Rae, A.J., 2009. The Geological framework of the
Wairakei-Tauhara Geothermal System, New Zealand. Geothermics 38 (1),
7284.
Rowland, J.V., Sibson, R.H., 2001. Extensional fault kinematics within the Taupo Volcanic Zone, New Zealand: soft-linked segmentation of a continental rift system.
New Zealand Journal of Geology and Geophysics 44, 271283.
Rowland, J.V., Sibson, R.H., 2004. Structural controls on hydrothermal ow in a
segmented rift of the Taupo Volcanic Zone, New Zealand. Geouids 4, 259283.
Rowland, J.V., Wilson, C.J.N., Gravley, D.M., 2010. Spatial and temporal variations in
magma-assisted rifting, Taupo Volcanic Zone, New Zealand. Journal of Volcanology and Geothermal Research 190 (12), 89108.
Saunders, K.E., Baker, J.A., et al., 2010. Microanalysis of large volume silicic magma
in continental and oceanic arcs: Melt inclusions in Taupo Volcanic Zone and
Kermadec Arc rocks, South West Pacic. Journal of Petrology 190 (12), 203218.
Seebeck, H., Nicol, A., Stern, T.A., Bibby, H.M., Stagpoole, V.M., 2010. Fault controls on
the geometry and location of the Okataina Caldera, Taupo Volcanic Zone, New
Zealand. Journal of Volcanology and Geothermal Research 190 (1/2), 136151,
doi:10.1016/j.jvolgeores.2009.04.011.
Sherburn, S., Bannister, S., Bibby, H., 2003. Seismic velocity structure of the central
Taupo Volcanic Zone, New Zealand, from local tomography. Journal of Volcanology and Geothermal Research 122, 6988.
Spinks, K.D., Acocella, V., Cole, J.W., Basset, K.N., 2005. Structural control of volcanism and caldera development in the transtensional Taupo Volcanic Zone, New
Zealand. Journal of Volcanology and Geothermal Research 144, 722.
Steiner, A., 1977. The Wairakei Geothermal Area, North Island, New Zealand: its
subsurface geology and hydrothermal rock alteration. New Zealand Geological
Survey Bulletin 90, 1134.
Stern, T.A., 1982. Seismic and gravity investigations of the Central Volcanic Region,
North Island, New Zealand. PhD Thesis, Institute of Geophysics, Victoria University of Wellington, New Zealand, p. 318.
Stimac, J.A., Powell, T.S., Golla, G.U., 2004. Porosity and permeability of the Tiwi
geothermal eld, Philippines, based on continuous and spot core measurements.
Geothermics 33, 87107.
Teng, Y., Koike, K., 2007. Three-dimensional imaging of a geothermal system using
temperature and geological models derived from a well-log dataset. Geothermics 36, 518538.
Tienfeng, W., Hedenquist, J.W., 1981. A reassessment of the structural control of
the Broadlands Geothermal Field, New Zealand. In: Proceedings of the 3rd New
Zealand Geothermal Workshop, Auckland, New Zealand, pp. 195202.
Tobler, W.R., 1970. A computer movie simulating urban growth in the Detroit region.
Economic Geography 46, 234240.
Verma, S.P., Andaverde, J., Santoyo, E., 2006. Statistical evaluation of methods for
the calculation of formation temperatures in geothermal and oil wells using an
extension of the error propagation theory. Journal of Geochemical Exploration
89, 398404.
Villamor, P., Berryman, K., 2001. A late Quaternary extension rate in the Taupo Volcanic Zone, New Zealand, derived from fault slip data. New Zealand Journal of
Geology and Geophysics 44, 243269.
Wilson, C.J.N., 2001. The 26.5 ka Oruanui eruption, New Zealand: an introduction
and overview. Journal of Volcanology and Geothermal Research 12, 133174.
Wilson, C.J.N., Houghton, B.F., Lloyd, E.F., 1986. Volcanic history and evolution of the
Maroa-Taupo area, central North Island. In: Smith, I.E.M. (Ed.), Late Cenozoic
Volcanism in New Zealand. Royal Society of New Zealand Bulletin 23, 194223.
Wilson, C.J.N., Houghton, B.F., McWilliams, M.O., Lanphere, M.A., Weaver, S.D.,
Briggs, R.M., 1995. Volcanic and structural evolution of the Taupo Volcanic Zone,
New Zealand: a review. Journal of Volcanology and Geothermal Research 68,
128.
Wood, C.P., 1994. The Waiora Formation Geothermal Aquifer, Taupo Volcanic Zone,
New Zealand. In: Proceedings 16th New Zealand Geothermal Workshop, pp.
121126.
G Model
ARTICLE IN PRESS
Wood, C.P., Brathwaite, R.L., Rosenberg, M.D., 2001. Basement structure, lithology and permeability at Kawerau and Ohaaki geothermal elds, New Zealand.
Geothermics 30 (4), 461481.
Wood, C.P., Browne, P.R.L., 2000. Wairakei Geothermal Power Project 40th Anniversary: Geology and Hydrothermal Alteration. In: Proceedings 22nd New Zealand
19
Geothermal Workshop, Auckland University, Auckland, New Zealand, pp.

285290.
Wright, I.C., 1992. Shallow structure and active tectonism of an offshore continental back-arc spreading system: the Taupo Volcanic Zone. New Zealand. Marine
Geology 103 (13), 287309.

Sepulveda Et Al, 2012, Kriging Predictions of Drill-Hole Stratigraphy and Temperature Data From The Wairakei Geothermal Field, New Zealand - Implications For Conceptual Modeling

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Sepulveda Et Al, 2012, Kriging Predictions of Drill-Hole Stratigraphy and Temperature Data From The Wairakei Geothermal Field, New Zealand - Implications For Conceptual Modeling

Diunggah oleh

Hak Cipta:

Format Tersedia

ARTICLE IN PRESS

Geothermics xxx (2012) xxxxxx

Contents lists available at SciVerse ScienceDirect

Kriging predictions of drill-hole stratigraphy and temperature data from the

Contact Energy Limited, Private Bag 2001, Taupo, New Zealand

can be challenging, but multidisciplinary analysis can be optimised

Corresponding author. Tel.: +64 7 3761959; fax: +64 7 3748472.

The Wairakei geothermal eld is located in the Taupo Volcanic

and tectonic subsidence (Villamor and Berryman, 2001; Nicol

340320 ka Whakamaru ignimbrites (Wilson et al., 1986;

No distinction of WF subunits for

Post-Whakamaru, pre-Huka Falls Fm volcanic (pyroclastic)

Huka Falls Formation (HFF)

This study (after Rosenberg et al., 2009)

A range of stratigraphic models have been proposed for

2.1. Local stratigraphy and structure

1.8 ka Taupo eruption and associated reworked deposits

systematic spatial correlations of geothermal activity with faults,

Consolidated into a single Supercial

F. Seplveda et al. / Geothermics xxx (2012) xxxxxx

Kriging algorithms is beyond the scope of this paper, but general

days after well completion. Worth noting is temperature logs

F. Seplveda et al. / Geothermics xxx (2012) xxxxxx

implemented as post-prediction correction (i.e., all values below

where (h) is the variogram, T is temperature, h is the lag distance

L(i , ) = (Xo ) + 2o

3.2. Kriging applied to temperature

Kriging is an exact interpolator, meaning data are honored

where Ti = measured temperatures at the sample points i = 1. . .n,

Kriging minimizes the variance of the estimation error 2 (Xo )

where L(i ,) is the Lagrangian function and is a Lagrange multiplier.

where fj is a polynomial function representing the drift (Olea, 1999).

where Ij (Xi ) is the indicator transform of the stratigraphic unit j at

highest probability is assigned, this is:

k1 if IM1 (Xo ) = max{IM1 (Xo ), IM2 (Xo ), . . . , IMm (Xo )}

k2 if IM2 (Xo ) = max{IM1 (Xo ), IM2 (Xo ), . . . , IMm (Xo )}

if IMm (Xo ) = max{IM1 (Xo ), IM2 (Xo ), . . . , IMm (Xo )}

where K(Xo ) is the predicted stratigraphic class at Xo . The fact that

areas of a study area due to relative lack of data. In the absence

for hydrothermal uid ow, follow not only a NE-trending, but

changes, generally, changes in basement geometry (basement

F. Seplveda et al. / Geothermics xxx (2012) xxxxxx

4.1.3. Reservoir studies

and McNamara (2011) noted a close correlation between feed

In order to test changes in the horizontal correlation of

F. Seplveda et al. / Geothermics xxx (2012) xxxxxx

Average lag distance h (m)

Omnidirectional, horizontal variogram (01000 mGL)

Omnidirectional, horizontal variogram (10003000 mGL)

No. of distances in class

No. of distances in class

model a variable drift is to set the polynomial order to t the most

d(Xo ) < 1000 m

Kriging temperatures are ultimately calculated under variable

where Tanisotropic is the Kriging prediction of temperature for an

4.2. Stratigraphical anisotropy

4.3. Spatial prediction

presence of anisotropy (specically, directional anisotropy in the

NW-trending boundary indicates the presence of a subvertical,

following the increase of negative residual anomalies (left hand

The variations in thickness of Wairakei Ignimbrite across

portrays high temperature regions at depth in the WK317 area.

A potential limitation of the geostatistical methods and models

where (h) is the variogram, T is temperature, h is the lag distance