Geothermics 76 (2018) 26–37

Contents lists available at ScienceDirect

Geothermics
journal homepage: www.elsevier.com/locate/geothermics

Transient pressure analysis of geothermal wells fractured during well testing T

a,b a,⁎
Hafeza A. Bakar , Sadiq J. Zarrouk
a
Department of Engineering and Science, The University of Auckland, Private Bag 92019, Auckland, New Zealand
b
Faculty of Engineering, Universiti Malaysia Sabah, 88999 Kota Kinabalu, Sabah, Malaysia

A R T I C LE I N FO A B S T R A C T

Keywords: Fracturing during injectivity testing can take place in geothermal wells when the reservoir has low permeability
Transient pressure analysis or when the well has significant skin damage. The transient behavior (pressure falloff) of these wells cannot be
PyTOUGH matched using existing well test analysis methods. At the same time, modelling fracturing in geothermal re-
Injection test servoirs using rock mechanics and commercial finite element software is complicated due to several field un-
Geothermal wells
certainties (e.g. formation height, reservoir permeability and porosity). In addition, rock mechanics data (rock
Fractured wells
stress, strain and Young’s modulus) are normally unknown in geothermal fields. This makes it difficult to develop
Fracture clapping
an appropriate fracture model that matches the field test data.
This study attempts to develop a fracture model without integrating rock mechanics. The model is setup with
a simple grid using the TOUGH2 geothermal reservoir simulator and validated using the advanced pressure
derivative transient analysis. Multiple subsets of fracture geometries were developed to represent the different
stages of fracture closure during pressure falloff. The PyTOUGH code was used to simplify the running of the
different fracture stages.
The results are very promising and provide a clear justification and explanation for the commonly en-
countered fractured well behavior. This model should be of use in matching data from geothermal wells with
similar pressure response.

1. Introduction example, a reservoir deeper than 1500 m, a vertical fracture is likely to

form as the least principal stress is likely in a horizontal direction
Injection/falloff tests are commonly carried out upon completion of (Hubbert and Willis, 1972; Ahmed, 2006). In some shallow formations
drilling geothermal wells. This is to obtain the wells injectivity, esti- on the other hand, least principal stress is usually the overburden
mate the formation permeability, Skin and reservoir pressure. (vertical) stress, thus fracture will propagate in a horizontal direction
During the water injection part of the test, it is possible that frac- (Hubbert and Willis, 1972; Ahmed, 2006).
turing of reservoir rock occurs when the reservoir has low permeability Fracture propagation criteria (vertical or horizontal) is different in
or where there is skin damage near the well (Clark, 1968; Nikravesh geothermal wells as will be demonstrated later.
et al., 1996; Abou-Sayed et al., 1996). When this takes place, all existing Fig. 1 presents the typical pressure behaviour of a fracture forma-
(conventional) well test analyses techniques will fail to match the re- tion during an injectivity test. The breakdown pressure is the maximum
servoir pressure response. This is because fracture (rock) mechanics pressure where the formation breaks down and initiates a fracture
should be considered in the solution, while conventional well test generation. The fracture propagates as injection continues and at the
analyses methods, do not account for them. same time fluid leaks off from the fracture to the surrounding rock.
Well fracturing can take place when high-pressure fluid is in- When the injection is ceased, a rapid pressure drop is observed due to
troduced into the reservoir (Hubbert and Willis, 1972; Yew and Weng, fluid leak off through the fracture into the surrounding matrix. The
2014; Fjaer et al., 2008). This behavior is known to drilling engineers as pressure at this point is called instantaneous shut-in pressure (ISIP). The
it can result in problems during cementing of deep well casings. This is ISIP signifies the start of fracture closure. Fluid continues to leak off
because the fluid pressure during injection can exceed the minimum from the still open fracture into the formation and fracture opening
principal stress of the rock and causes splitting/fracture of the rock decreases until the fracture closes. Then, the fluid pressure transient
resulting in loss of cement (poor cementing job). Fractures form per- gradually transitions into pseudo-linear or radial flow.
pendicular to the least principal stresses. In petroleum reservoirs for Rock mechanics are commonly integrated with fluid mechanics for

⁎
Corresponding author.
E-mail address: s.zarrouk@auckland.ac.nz (S.J. Zarrouk).

https://doi.org/10.1016/j.geothermics.2018.06.010
Received 12 June 2017; Received in revised form 1 June 2018; Accepted 23 June 2018
0375-6505/ © 2018 Elsevier Ltd. All rights reserved.
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Fig. 1. Pressure response of a fractured well during injection test (after Hubbert
and Willis, 1957). Fig. 2. Fall-off pressure response from an injectivity test of Well-25A-33 (data
provided by Mr. Ken Tyler Nevada University).

fracture modelling and it has been broadly implemented in commercial

finite element software (Ingraffea and Heuze, 1980; Noorishad et al., et al., 2011). This field has been producing electricity since 1989 and
1982; Rutqvist et al., 2002). Many studies from the petroleum industry currently operating at 16 MWe (McLachlan et al., 2011; Ohren et al.,
have reported good results from the use of this kind of facture models 2011).
(Bai et al., 1999; Jing et al., 2001; Detournay, 2004). Nevertheless, this
is not the case for geothermal fields, because unlike oil and gas fields, 2.2. Well-25A-33
geothermal fields have more uncertainties and limited information on
rock mechanics. These uncertainties and unknowns can lead to poor Well 25A-33 was drilled in 2009 to a total depth of 2741 m (Ohren
matching of the field data. et al., 2011). The casing depth of the well is 2122 m and a perforated
In this work we will use numerical reservoir simulation to model the liner was installed from 2122 m to 2741 m. The main feed zone is found
facture behavior of such wells during pressure fall off using the at 2183 m to 2186 m. An injection falloff test was carried out on this
TOUGH2 geothermal modeling software (Pruess et al., 1999). A python well on January 2010. The pressure response at 2184 m and pressure
script was written to simplify the running of multiple models. derivative on log-log plot of the test are shown in Figs. 2 and 3 re-
Prior to this study, parallel coupling between TOUGH2 and spectively. There is no record of stable injection pressure prior to fall-
ABAQUS™ solid mechanics software was attempted to match the well off, therefore it was assumed based on the best estimates prior stopping
test data from several geothermal wells. Unfortunately, the results were water reinjection (Table 1). Deflagration (permeability enhancement
not satisfactory with poor data matches between simulated and real with down-hole explosives) process was carried out during September
data. This is likely due to the unknown rock properties of these fields and December 2010 at the upper zone and shallow part of the well to
and limited control on some input and solution parameters in improve the well performance. The well has an average production rate
ABAQUS™ software. of 43 kg/s at 169 °C at the wellhead.
The objective of this study is to develop a fracture model without The unusual shape (sharp dip) of the differential pressure (dP) and
explicit fracture mechanics modeling. The model is solely setup using a the pressure derivative (dP`) (Horne, 1995) on the log-log plot (Fig. 3)
geothermal reservoir simulator to explore the type, extent (length) and shows a clear change in profile at the ISIP point. It is impossible to
permeability of fracture. The result is then analysed and matched using model/match this behavior with all existing well test analysis models.
advanced transient pressure analysis method for further data validation At the same time not knowing the exact pressure prior to stopping in-
using well test data from the Soda Lake geothermal field (Nevada, USA) jection result in a wellbore storage stage of the early dP curve that is not
and the Wairakei geothermal field (New Zealand). Sensitivity analysis is unity slope, which is an indication that the estimated initial pressure
carried out to identify the factors, which may affect the model and data was not exact.
matching result. It is important to note that the distinct change in the pressure falloff
(Fig. 2) and the corresponding sharp drop in the pressure derivative
2. Case study: Soda Lake geothermal field, Nevada, USA (Fig. 3) is different from down flow (internal flow) pressure response.
McLean and Zarrouk (2015) showed the pressure falloff response due to
2.1. Overview down flow. While the shape of the curve is relatively similar, the
sudden drop (dip) in pressure derivative is much steeper and more
Soda Lake geothermal field is located in the Carson Dessert, west pronounced in the case of fracture closure (fracture clapping). At the
central Nevada. The first geothermal waters were discovered in 1903 same time, other well test data and the set casing depth information can
when a well drilled at the site of a hot spring northeast of Soda Lake also help identify the difference. We also argue that other artificial well
encountered hot water at a depth of 18 m (Sibbett, 1979). test effects such as wellbore heating after cold-water injection and tool
Geothermal drilling activity began in the mid 1970s and are still movement due to wireline thermal expansion should be; limited, much
ongoing. Up to 2011 (McLachlan et al., 2011), there were more than 28 slower and less significant than the ISIP.
deep production, injection, observation and temperature gradient in-
vestigation wells onsite. In addition, there are 70 other shallow ground 3. General model setup
water wells and survey drill-holes.
Soda Lake field has proved reserves of 20 MWe and indicated re- The pressure falloff behavior of Well 25A-33 (Fig. 2) is similar to
sources of 41 MWe (McLachlan et al., 2011). There are two power- that of Fig. 1 indicating that fracturing took place during the injection
producing facilities in the field with a total capacity of 23MWe (Ohren test. Nonetheless, the type of fracture (vertical or horizontal) is

27
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Fig. 3. Differential pressure (dP) and pressure derivative (dP’) plots of Well 25A-33 during fall-off.

Table 1
Input parameters for the models for well 25A-33.
Parameter Value Unit (SI)

(a)
Fracture half-length (Vertical) 100 m
Fracture radius (Horizontal) 100(a) m
Reservoir height 600(m) m
Vertical fracture width 12.5(a) m
Horizontal fracture height 0.05(a) m
Initial reservoir pressure 210(a) bar
Initial reservoir temperature 175(m) °C
Reservoir permeability 0.35(c) mD
Reservoir (matrix) porosity 0.1(a) –
Fracture porosity 0.1(a) –
Injection rate 7.57(m) kg/s

Note: (a) Assumed data, (m) measured data, (c) calculated data.

unknown. Therefore, two numerical models of vertical and horizontal

fractures were setup using the TOUGH2 simulator (Pruess et al., 1999)
based on an injection test into Well 25A-33 conditions. A PyTOUGH
script (Croucher, 2013) was written for automating the models running
and data matching. The computation grid structures for both models
are as shown in Figs. 4 and 5 respectively. The reservoir model input
parameters for both models are given in Table 1. These input para-
meters were based on measured field data when possible, while the
Fig. 4. The grid structure for vertical fracture model using one layer and as-
remaining parameters were either calculated or assumed based on ex-
signing different permeability to the fracture blocks.
perience. However, we have carried a sensitivity study to investigate
the significance of all the assumed parameters as will be discussed in
Section 5. work.
The vertical fracture model (Fig. 4) concept adapted a simple ver- The width of a fracture for vertical model is represented by the
tical fracture models (Perkins and Kern, 1961) such as Khristianovich- fracture block size as shown in Fig. 4 with red blocks. The width of the
Geertsma-de Klerk (KGD) (Geertsma and De Klerk, 1969) and Perkins- fracture is approximately 12.5 m. The width of fracture for horizontal
Kern-Nordgren (PKN) (Nordgren, 1972). The minimum in-situ stress is model on the other hand is represented by horizontal fracture height
in the horizontal plane and the vertical fracture plane is perpendicular with 0.05 m (Table 1). The red blocks as shown in Fig. 5 represents
to the minimum in-situ stress direction. horizontal fracture disc. The fracture width remains constant
Meanwhile for horizontal fracture model (Fig. 5), the concept was throughout the simulation run, while the fracture half-length and radius
referred to penny-shaped (also known as pancake) fracture model were changed to match the field data. Both numerical models (Figs. 4
(Perkins and Kern, 1961; Savitski and Detournay, 2002; Abe et al., and 5) have constant pressure (Dirichlet) side boundary to represent
1976). The fracture becomes radial shaped when the vertical distribu- infinite acting reservoirs, while having closed upper and lower
tion of minimum in-situ stress is uniform. However, if the horizontal boundaries.
stresses (x and y directions) are not equal, the fracture will then have an Note that, the number of vertical (Fig. 4) and horizontal (Fig. 5)
ellipse shape. The ellipse shaped fracture was not considered in this grids in the model layer is the same. The vertical fracture model is made

28
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Fig. 6. Fall-off pressure response of vertical fracture model and field data (Well-
25A-33).

Fig. 5. The grid structure for horizontal fracture model, made from 0.05 m
height fracture layer sandwiched between two reservoir layers each ½ reservoir
height.

from one layer with a vertical fracture chosen using some of the existing
model blocks (Fig. 4), which in the base case equals to 12.5 m (Table 1).
This do not suggest that the actual fracture is 12.5 m wide in real re-
servoirs, but we did not refine the model grid to the real reservoir di-
mensions. This will be further investigated with a mesh refinement
study in Section 5. Fig. 7. Fall-off pressure response of horizontal fracture model and field data
While, in the horizontal model we have added an additional layer (Well-25A-33).
0.05 m wide to represent the fracture layer, which is sandwiched be-
tween two reservoirs layers each ½ reservoir height (Table 1).
The following assumptions were made in this study:

• Symmetrical fracture growth from an injection point during pres-

sure falloff.
• Only one fracture is used for each model
• The vertical fracture is a straight line.
• Homogeneous reservoir rock.
• Uniform rock matrix porosity and permeability.
• Fluid rheology is following Darcy law
• The injection rate is constant during injection period.
4. Modelling results

Both vertical and horizontal fracture models were run for full in- Fig. 8. Differential pressure (dP) and pressure derivative (dP') plots of simu-
jection and pressure fall-off period as shown in Figs. 6 and 7 respec- lated data for vertical fracture model and field data.
tively. There are about twenty PyTOUGH runs of multiple models for
each fracture model with changing fracture size and fracture perme- grid used for each model, this will be further investigated in Section 5.
ability to match the measured field data. This pressure response in Figs. 6 and 7 were further analysed using
Overall, the optimized vertical fracture model gives a very good fit the pressure difference and its derivative on the log-log plot of both
to the pressure data as shown in Fig. 6. The optimized horizontal fracture models as shown in Figs. 8 and 9. The pressure derivative result
fracture model meanwhile gave a satisfactory early behavior but poor from the vertical fracture model (Fig. 8) shows a satisfactory match
match in the later time to the field data (Fig. 7). Reasonable matching with the field data in the early period (0.0001–0.1 h) and a good match
of the field data is observed at times between 1.94 h and 2.50 h, while in the later period (0.12–1.50 h). On the other hand, the pressure de-
poor data matched is found from time 2.50 h till 4.00 h. This indicates rivative plots for horizontal fracture model (Fig. 9), gave a relatively
that vertical fracture propagation is more suitable compared to hor- poor match to the field data.
izontal fracture design for this well. The difference in the results be- It can be identified from Figs. 8 and 9 that the field pressure
tween vertical and horizontal fracture models may also be cause by the

29
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Fig. 11. Fracture radius and fracture permeability over time for horizontal
Fig. 9. Differential pressure (dP) and pressure derivatives (dP') plots of simu-
fracture model.
lated data for horizontal fracture model and field data.

derivative displays two humps and a sharp dip between those humps. Fig. 11 on the other hand gives the fracture radius length and
The first hump represents the quick pressure drop while the fracture is fracture permeability over time for horizontal fracture model. Unlike
still open after injection ceases. The sharp dip in the pressure derivative the vertical fracture model, the fracture radius/length and fracture
is the fracture closing. The later large hump depicts normal pressure permeability decreases over time but the fracture stays open at 2.08 h.
declining before reaching steady (infinite acting) reservoir pressure (i.e. Then, fracture radius grows back to 25 m and remains at that radius
zero slope for the derivative pressure). The vertical fracture model until the end of time, while fracture permeability exhibits an increment
pressure derivative (Fig. 8) has shown close fit to the two humps. Yet at time 2.3 h and stay constant at about 0.95 Darcy.
the sharp dip has satisfactory match. On the other hand, on the hor- It is clear that the model with the vertical fracture is more re-
izontal fracture model pressure response (Fig. 9), no clear hump is presentative of the pressure response of this well rather than a hor-
shown in the pressure derivative plots and the dip is very small. The izontal fracture. However, despite the poor match of the horizontal
field and model data match are also relatively poor. model to the history data (Fig. 7) and the pressure derivative (Fig. 9), it
Fig. 10 shows the fracture half-length and fracture permeability is also indicating there is a fracture clapping effect taking place.
responses over time during pressure fall-off for vertical fracture model.
It can be found that the fracture half-length and permeability decreases 5. Sensitivity analysis
immediately after injection shut down until ISIP at 2.08 h (from the
start of the test). After that, the fracture half-length as well as fracture For the sensitivity analysis, a reference model was set up using the
permeability gradually increase until certain time and return to fracture vertical fracture model described in Section 3. The vertical fracture
closure mode. Fracture half-length and fracture permeability increase model is used because the design is more suitable as it gave a good
after 2.08 h may be due to; fracture closure near the wellbore resulted match for well 25A-33 pressure data compared to the horizontal frac-
in a buildup of pressure in the remaining part of the fracture in contact ture design. The range of parameters used in the sensitivity analysis are
with low permeability rock matrix. This results in a partial re-opening slightly below and above the reference values (Table 1). The sensitivity
of the fracture to allow some flow from the fracture into the borehole of these parameters are summarised in Table 2 and discussed in the
driven by pressure difference. This effectively mean that the fracture following section.
does not fully open or close at the same time, there will be parts of the
fracture that closes, while other parts remain open for longer time. Van 5.1. Fracture half-length
der Baan et al. (2016) reported similar effect while studying hydraulic
fracture propagation and gave it the name “fracture clapping”. The starting open fracture half-length used in the reference model
was 100 m and the range used for the sensitivity study varies from 50 m
to 250 m. The results show that this parameter has no significant effect
on the shape of derivative plots compared with the reference model as
demonstrated in Fig. 12. This indicates that the starting open fracture
half-length has low impact on the modeling results.
The fracture half-length (Fig. 13) and fracture permeability (Fig. 14)

Table 2
Reference model, range tested value and sensitivity result for well 25A-33.
Parameter Reference model Range Degree of
value tested Sensitivity

Fracture half length (m) 100 50, 150, Low

250
Vertical fracture height 600 100, 200, High
(m) 400
Fracture porosity 0.1 0.15, 0.2, Low
0.25
Initial reservoir 210 200, 215, High
Fig. 10. Fracture half-length and fracture permeability over time for vertical
pressure (bar) 220
fracture model.

30
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Fig. 12. Derivative plots of various fracture half-length used in the model of well 25A-33.

25A-33. The actual vertical reservoir height is normally subjective and

in most cases unknown (McLean and Zarrouk, 2017). A range of re-
servoir height from 100 m to 400 m were tested, with vertical fracture
height extending to the full reservoir height. A derivative plot com-
paring the results shows that the model is sensitive to different vertical
reservoir/fracture height (Fig. 15). Fig. 15 shows that data match
around the pressure dip section has improved with reducing height; this
is likely because actual fractures in reservoir has limited vertical/sub-
vertical extent. The vertical fracture height of 400 m demonstrates the
best overall match.
Figs. 16 and 17 on the other hand show the fracture half-length and
fracture permeability responses over time during fall-off. The fracture
half-length is reduced as the injection ceases indicating fracture is
closing gradually. Similarly, fracture permeability decreases as fracture
is closing.
Fig. 13. Fracture half-length responses for the effect of changing fracture half For vertical fracture height of 400 m (Fig. 17), similar to the re-
–length setup in the model of well 25A-33. ference model, it exhibits a gradual increase in fracture half-length and
fracture permeability for some period of time after fracture closure and
return to reservoir condition at the end of the test. However, as for
100 m and 200 m heights, there is no indication of fracture closure and
the fracture remains open until the end of the test, which is unrealistic.
Gradual fracture opening after closure or fracture remains open after
injection ceases may due to skin, well damage (due to the slow pressure
leak-off) or minor seismic activity effects (McLean and Zarrouk, 2017;
Davies et al., 2012).
Investigating fracture permeability on the other hand found that
vertical fracture height of 100 m shows a significant increase of per-
meability compared to other heights (200 m, 400 m and 600 m (ref)).
The fracture permeability increases up to about 9.0 Darcy after 2.79 h
and remains constant until the end of the test. Other fracture height
meanwhile shows similar fracture permeability trends with the re-
ference model.
Fig. 14. Fracture permeability responses while changing fracture half-length It is clear from Figs. 16 and 17 that while the actual reservoir/
setup in the model of well 25A-33. fracture height remains unknown, the realistic estimate of the injection
induced fracture height in this case is more than 100 m.
responses after injection ceased also demonstrate no significant differ-
ence from that of the reference model. 5.3. Fracture porosity

The reference model has a fracture porosity of 0.1 (Table 1) and the
5.2. Vertical reservoir/fracture height (thickness) range tested is from 0.15 to 0.3 (Table 2). Results show that the model
is insensitive to this parameter as shown in the derivative plots of
The reference model has a vertical reservoir height of 600 m Fig. 18. This indicates that fracture porosity has no significant effect to
(Table 1) based on the estimated height of the open-hole section of well the data matching. Any practically realistic fracture porosity can be

31
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Fig. 15. Differential pressure (dP) and derivative pressure (dP’) plots of various vertical fracture heights.

5.4. Initial reservoir pressure

The reference model used a reservoir pressure of 210 bar (Table 1),
which was not available from well testing data. A reservoir pressure
range from 200 bar to 220 bar was investigated. The results show that
data matching at the dip and the second hump sections have been af-
fected significantly (Fig. 21). Data fit at the dip area are better as the
reservoir pressure increases from 200 bar to 220 bar, yet slightly off the
field data at 220 bar. Similarly, data matching at the second hump
demonstrate improvement from 200 bar to 215 bar and off the field
data for reservoir pressure of 220 bar. This indicates that this parameter
is important and sensitive to the model.
Figs. 22 and 23 show fracture half-length and fracture permeability
responses for the reservoir pressure range tested respectively. While, for
the fracture half-length, most reservoir pressures tested shows that the
Fig. 16. Fracture half-length responses for different vertical fracture/reservoir length reduces since injection ceased until full fracture closure except
heights (100 m, 200 m, 400 m and 600 m (ref)).
for 220 bar, where the fracture remains open. This is a clear indication
that the reservoir pressure is less than 220 bar.
The fracture for reservoir pressure of 200 bar gradually opens for
some time and closes again at the end of the test, similar to the re-
ference model. However, for reservoir pressures of 215 bar and 220 bar,
the fracture slowly open up after the closure point and remain open
until the end of the test.
Fracture permeability (Fig. 23) for the range of reservoir pressure
tested also shows reduction at the early period of fall off, then a slight
increase after closure and finally return to reservoir permeability at the
end of the test. Nevertheless, reservoir pressure for 215 bar and 220 bar
exhibit quite different behaviors at the end of the test, where the
fracture permeability did not return to the initial reservoir perme-
ability. This is because the model show that the fracture remains open
(Fig. 22) even when the test has ended, which is un-realistic. Therefore,
it is important to make sure the initial reservoir pressure used in the
simulation is as accurate as possible.
Fig. 17. Fracture permeability for various vertical fracture heights (100 m,
200 m, 400 m and 600 m (ref)).
5.5. Summary of sensitivity analysis

used in the modeling study.

The results from the sensitivity analysis are summarised in Table 2.
This result is further confirmed in fracture half-length and fracture
It is identified that some parameters are demonstrated to have low
permeability responses in Figs. 19 and 20 respectively. The responses
degree of sensitivity in the model, which are fracture half-length and
for fracture half-length and permeability of the range tested are same as
fracture porosity. On the other hand, vertical fracture height and initial
the reference model response. The porosity is more related to the pro-
reservoir pressure have high degree of sensitivity to the modeling re-
portion of fluid stored in the fracture.
sults.
No attempt was made to investigate the effect of changing reservoir
For parameters with low sensitivity, it indicates that the require-
matrix porosity at this stage as we feel it will have a lesser impact than
ment to use the exact or correct value of these paramaters in the model
the facture porosity.
is less important. However, the lack of sensitivity is only determined

32
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Fig. 18. Differential pressure (dP) and derivative pressure (dP’) plots for different fracture porosities.

6. Fracture grid refinement study

The fracture (model) block size used in vertical fracture model

discussed in Section 3 is used as the reference fracture block size in this
study, which is 12.5 m (coarse grid) in the x and y directions. The
fracture block size is further refined to 6.25 m (moderate) and 3.13 m
(fine) to see the significance of fracture block size on the data matching
results.
The derivative plots comparing these fracture sizes show that frac-
ture block size gives significant effect to the data matching result. It is
also found that data matching at the dip area is good when the fracture
block size is refined to 6.25 m as shown in Fig. 24. As the fracture block
is further refined however, the matching at the dip area go slightly off
below the field data as shown in Fig. 24.
Fig. 25 gives the fracture permeability change over time during fall-
Fig. 19. Fracture half-length responses during fall-off for various fracture off for the different fracture block sizes. Fig. 25 shows that fracture
porosity. permeability after injection ceases is significantly affected by the re-
finement of fracture block size. The fracture permeability response from
the finest fracture size model show quite significant differences com-
pared to the other two fracture block sizes models. The fracture per-
meability in the fine mesh model increased quite significantly after the
initial fracture closure point compared to the other models (Fig. 25).
This is because the fracture remains open for longer (Fig. 26) to allow a
better match to the pressure derivative (Fig. 24).
Fig. 26 shows that the fracture half-length is almost similar for all
fracture block sizes after injection ceased. Fracture half-length gradu-
ally decreased as the injection stops and comes to closure at one point.
Then the fracture opened up for some period and closed up at the end of
the test.
The injection-induced fracture normally has a width of only few
milliliters in practice, therefore further model grid/mesh refinement
study is required. However, we feel that the results thus far are a good
proof of concept. It is also possible to use a simple radial (one dimen-
Fig. 20. Fracture permeability for various fracture porosities. sional) grid modified with a vertical fraction and Fractional Direction
model in the near well blocks (Zarrouk et al., 2007) to match this be-
within a range near to the referenced value. havior. However, this model will require different interpretations to the
Parameters with high degree of sensitivity on the other hand re- output result as it will include a new model dimension which is nor-
quires a very careful and reliable value to be used in the model. Results mally somewhere between 1–3 dimensions (Zarrouk et al., 2007).
demonstrated that the use of a slightly different value of the parameters
showed a very significant effect to the data matching process.
7. Other geothermal wells with similar behavior

There are several other reported geothermal wells that show similar
pressure response after injection tests as shown above in Well-25A-33.

33
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Fig. 21. Differential pressure (dP) and derivative pressure (dP’) plots of various initial reservoir pressures.

Fig. 22. Fracture half-length responses for various initial reservoir pressures. Fig. 24. Differential pressure (dP) and pressure derivative (dP’) plots for dif-
ferent fracture block sizes of vertical fracture model.

Fig. 23. Fracture permeability responses for various initial reservoir pressures
during fall-off.
Fig. 25. Fracture permeability responses of different fracture block sizes used in
vertical fracture model.
Well-Y from Soda Lake, USA (Fig. 27) and Well-Z from Wairakei, New
Zealand (Fig. 28) are two examples. The log-log plots shown in these
figures were based on measured pressure data recorded during field
injection fall of testing (completion) testing.

34
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

Table 3
Input parameters for fracture models of Well Z from Wairakei, New Zealand.
Parameter Value Unit (SI)

(a)
Fracture half-length 100 m
Fracture radius 100(a) m
Layer height 340(m) m
Vertical fracture height 340(a) m
Horizontal fracture height 0.05(a) m
Initial reservoir pressure 71.2(a) bar
Initial reservoir temperature 30(m) °C
Reservoir permeability 15(c) mD
Reservoir porosity 0.1(a)
Injection rate 12.53(m) (0 s – 3523 s) kg/s
25.33(m)(3523 s – 6283 s)
37.86(m) (6283 s – 8774 s)
25.33(m) (8774 s – 10,590 s)
37.86(m) (10,590 s – 11956s)
Fig. 26. Fracture half-length responses for various fracture block sizes used in
the model. Note: (a) Assumed data, (m) measured data, (c) calculated data.

7.1. Model setup

Vertical and horizontal fracture models were setup using the same
approach as carried out in Well-25A-33. This is due to no definite in-
dication of a fracture orientation based on the available data. A
PyTough script was written for automating the models and data
matching. The fracture grid diagrams for both models are comparable
to diagrams as shown in Figs. 4 and 5, while the input data for these
models are listed in Table 3.

7.2. Modelling results

After running multiple TOUGH2 models, it is determined that the

optimized vertical fracture model has given a better overall data
Fig. 27. The log-log plot of differential pressure and pressure derivative for matching compared to optimized horizontal fracture model as shown in
Well-Y from Soda Lake, USA (data provided by Mr. Ken Tyler Nevada Fig. 29. The result from horizontal fracture model starts to show poor
University). data matching after 3.4 h.
Pressure derivative plots (Fig. 30) also confirmed that simulated
data from vertical fracture model gives reasonable data matching
overall except on the later dip section. This second dip can be related to
other wellbore or reservoir effects, but it is not the focus of the current
modeling study. It is also noted that there is a slight off data match at
the first dip. This is due to the estimated vertical fracture height, which
is unknown and solely based on the depth of the open-hole section of
the well. Further refinement of the reservoir/fracture height can result
in better match as demonstrated in well-25A-33 case study.
Fig. 31 on the other hand shows the fracture half-length and fracture
permeability response during the fall-off for vertical fracture models.

Fig. 28. The log-log plot of differential pressure and pressure derivative for
Well-Z from Wairakei, New Zealand (Contact Energy Ltd. with kind permis-
sion).

There are two humps and a sharp dip in between found in the
pressure derivative plots. These behaviors are similar to Well-25A-33,
which indicate a comparable ISIP and fracture closure is taking place in
these wells. Similar to Well-25A-33 the exact fracture orientation
(vertical or horizontal) cannot be determined without running multiple
TOUGH2 models to this field data. Well-Z pressure data is modelled and
discussed in this work using the same approach applied to Well-25A-33.
Fig. 29. Pressure response of Well-Z for vertical fracture and horizontal fracture
models against field data.

35
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

other wells for example deep (> 4000 m) enhanced geothermal

systems (EGS) wells (Llanos et al., 2015). However, this system is
not a typical EGS site, it is an exception from the regular strike-slip
or normal faulting in such depth.
4 The method presented in this work simplifies well test analysis
without including rock mechanics. It can be used to match field data
of other fractured wells with similar pressure response (e.g. Wells Y
and others). This will allow the estimation of the reservoir perme-
ability, fracture half-length, when conventional well test analysis
methods are not be successful.
5 Fracture permeability plays an important role in matching of si-
mulated model with a field data.
6 Fracture half-length and fracture permeability may increase (frac-
ture clapping) after fracture closure, depending on the flow regime
associated with vertical fracture.
7 Vertical fracture height and initial reservoir pressure have sig-
nificant effect on the models result. Sensitivity analysis shows that
Fig. 30. The differential pressure (dP) and pressure derivative (dP’) plots of the data matching can be improved by adjusting these parameters.
vertical fracture model, horizontal fracture model and field data of Well-Z.
However, careful estimation of these parameters should be based on
field observation and experience since estimation of these values
can be subjective.
8 Fracture block size also demonstrates significant impact to the data
matching result. Finer block size shows better data fit compared to
coarse block size, this indicates that these models are mesh size
dependent and should be considered with care. To overcome this
problem, it is theoretically possible to replace the two or three di-
mensional Cartesian model grid and the associated explicit (vertical
or horizontal) fractures with a one-dimensional radial model. The
near wellbore blocks can be modified using the fractional dimension
model (Zarrouk et al., 2007) to provide a good match with field
data.
9 Fracture porosity and fracture length/radius have no significant
effect on matching field data.
10 The later time dip in the pressure response seen in Well Z (Fig. 28) is
Fig. 31. Fracture half-length and fracture permeability responses from vertical a result of wellbore/reservoir effects, which is not taken into con-
fracture model for Well-Z. sideration in data matching. No attempt was made at this stage to
match the later part of the pressure response.
Both parameters decrease initially as injection ceases until fracture
closure. Then, these parameters started to increase for a certain period Acknowledgments
and later return to reservoir conditions. The increment period for ver-
tical fracture model are determined between 3.32 h–3.62 h. The in- The authors would like to thank Mr. Ken Tyler from the University
crease in fracture length and fracture permeability is caused by the Nevada, Reno for proving the Soda Lake well 25A-33 and Well-Y test
reopening of the fracture due to fracture pressure buildup from the slow data and Mrs Katie Maclean of Contact Energy Ltd. for the Wairakei
fluid leak off into the surrounding rock matrix. well test data. The authors also gratefully acknowledged Ministry of
Higher Education Malaysia (MOHE) for the financial support.
8. Conclusions
References
The following conclusions can be drawn from this modelling study:
Abe, H., Keer, L., Mura, T., 1976. Growth rate of a penny‐shaped crack in hydraulic
1 The distinct very sharp drop in the pressure derivative during early fracturing of rocks, 2. J. Geophys. Res. 81 (35), 6292–6298. http://dx.doi.org/10.
1029/JB081i035p06292.
pressure falloff is a clear indication of ISIP and fracture closure. Abou-Sayed, I., Schueler, S., Ehrl, E., Hendricks, W., 1996. Multiple hydraulic fracture
However, careful examination of the data is needed to identify other stimulation in a deep horizontal tight gas well. J. Pet. Technol. 48 (02), 163–168.
effects such as flow disruptions, down flow, heating during pressure http://dx.doi.org/10.2118/30532-JPT.
Ahmed, T., 2006. Reservoir Engineering Handbook (Trans. Ed.^Eds. ed. Vol.). Gulf
falloff and tool movement due to thermal expansion of the wireline. Professional Publishing 0080480683.
2 Unfortunately the early pressure response during the reinjection Bai, M., Meng, F., Elsworth, D., Abousleiman, Y., Roegiers, J.C., 1999. Numerical mod-
part of the 25A-33 well test was not available to us. Having this data elling of coupled flow and deformation in fractured rock specimens. Int. J. Numer.
Anal. Methods Geomech. 23 (2), 141–160. http://dx.doi.org/10.1002/(SICI)1096-
would have helped relate this analysis to the well injectivity and 9853(199902)23:2<141::AID-NAG962>3.0.CO;2-G.
identifying the fracturing pressure, which could be related to the Clark, K.K., 1968. Transient pressure testing of fractured water injection wells. Soc. Pet.
ISIP and closure pressures. Eng. 20 (6), 639–645. http://dx.doi.org/10.2118/1821-PA. SPE-1821-PA, June
1968.
3 Vertical fracture model gives reasonably good match to the field Croucher, A., 2013. PyTOUGH User’s Guide. Department of Engineering Science,
data presented in this study while the horizontal fracture models University of Auckland, Auckland, New Zealand. https://github.com/acroucher/
was not able to match well 25A-33 and Well- Z data. This indicates PyTOUGH.
Davies, R.J., Mathias, S.A., Moss, J., Hustoft, S., Newport, L., 2012. Hydraulic fractures:
vertical fracture propagation during the testing of these conven-
how far can they go? Mar. Pet. Geol. 37 (1), 1–6. http://dx.doi.org/10.1016/j.
tional geothermal wells in two-phase geothermal systems. marpetgeo.2012.04.001.
Horizontal fractures (e.g. thrust faults) should not be ruled out in Detournay, E., 2004. Propagation regimes of fluid-driven fractures in impermeable rocks.
Int. J. Geomech. 4 (1), 35–45. http://dx.doi.org/10.1061/(ASCE)1532-3641(2004)

36
H.A. Bakar, S.J. Zarrouk Geothermics 76 (2018) 26–37

4:1(35)#sthash.GuvZyCfi.dpuf. dx.doi.org/10.2118/31103-MS.
Fjaer, E., Holt, R., Horsrud, P., Raaen, A., Risnes, R., 2008. Mechanics of hydraulic Noorishad, J., Ayatollahi, M., Witherspoon, P., 1982. A finite-element method for coupled
fracturing. Dev. Pet. Sci. 53, 369–390. http://dx.doi.org/10.1016/S0376-7361(07) stress and fluid flow analysis in fractured rock masses. Paper Presented at the
53011-6. International Journal of Rock Mechanics and Mining Sciences & Geomechanics
Geertsma, J., De Klerk, F., 1969. A rapid method of predicting width and extent of hy- Abstracts. http://dx.doi.org/10.1016/0148-9062(82)90888-9.
draulically induced fractures. J. Pet. Technol. 21 (12). http://dx.doi.org/10.2118/ Nordgren, R., 1972. Propagation of a vertical hydraulic fracture. Soc. Pet. Eng. J. 12 (04),
2458-PA. 1,571-571,581. 306–314. http://dx.doi.org/10.2118/3009-PA.
Horne, R.N., 1995. Modern Well Test Analysis: A Computer Aided Approach, 2nd addi- Ohren, M., Benoit, D., Kumataka, M., Morrison, M., 2011. Permeability recovery and
tion. Petroway, Inc. ISBN 0-9626992-1-7. enhancements in the Soda Lake geothermal field, fallon, Nevada. Geotherm. Resour.
Hubbert, M.K., Willis, D.G., 1957. Mechanics of hydraulic fracturing. Pet. Trans. AIME Counc. Trans. 35, 493–497.
210, 153–168. Perkins, T., Kern, L., 1961. Widths of hydraulic fractures. J. Pet. Technol. 13 (09),
Hubbert, M.K., Willis, D.G., 1972. Mechanics of hydraulic fracturing. Mem.-Am. Assoc. 937–949. http://dx.doi.org/10.2118/89-PA.
Pet. Geol. (United States) 18, 239–257. Pruess, K., Oldenburg, C., Moridis, G., 1999. TOUGH2 User’s Guide Version 2. Lawrence
Ingraffea, A.R., Heuze, F.E., 1980. Finite element models for rock fracture mechanics. Int. Berkeley National Laboratory.
J. Numer. Anal. Methods Geomech. 4 (1), 25–43. http://dx.doi.org/10.1002/nag. Rutqvist, J., Wu, Y.S., Tsang, C.F., Bodvarsson, G., 2002. A modeling approach for ana-
1610040103. lysis of coupled multiphase fluid flow, heat transfer, and deformation in fractured
Jing, L., Ma, Y., Fang, Z., 2001. Modeling of fluid flow and solid deformation for fractured porous rock. Int. J. Rock Mech. Min. Sci. 39 (4), 429–442. http://dx.doi.org/10.
rocks with discontinuous deformation analysis (DDA) method. Int. J. Rock Mech. 1016/S1365-1609(02)00022-9.
Min. Sci. 38 (3), 343–355. http://dx.doi.org/10.1016/S1365-1609(01)00005-3. Savitski, A., Detournay, E., 2002. Propagation of a penny-shaped fluid-driven fracture in
Llanos, E.M., Zarrouk, S.J., Hogarth, R.A., 2015. Numerical model of the Habanero an impermeable rock: asymptotic solutions. Int. J. Solids Struct. 39 (26), 6311–6337.
geothermal reservoir, Australia. Geothermics 53, 308–319. http://dx.doi.org/10. http://dx.doi.org/10.1016/S0020-7683(02)00492-4.
1016/j.geothermics.2014.07.008. Sibbett, B.S., 1979. Geology of the Soda Lake geothermal area, United States. http://dx.
McLachlan, H.S., Benoit, W.R., Faulds, J.E., 2011. Structural framework of the Soda Lake doi.org/10.2172/5716536. DOE/ET/28392-34., pp 32. http://data.nbmg.unr.edu/
geothermal area, Churchill County, Nevada. Geotherm. Resour. Counc. Trans. 35, public/Geothermal/GreyLiterature/Sibbett_Geol_SodaLake_197.
925–930. Van der Baan, M., Eaton, D.W., Rreisig, G., 2016. Stick-split mechanism for anthropogenic
McLean, K., Zarrouk, S.J., 2015. Geothermal well test analysis using the pressure deri- fluid-induced tensile rock failure. Geol. Soc. Am. 4 (7) 603-506. https://sites.
vative: some common issues and solutions. Geothermics 55, 108–125. ualberta.ca/∼vanderba/papers/VEP16.pdf.
McLean, K., Zarrouk, S.J., 2017. Pressure transient analysis of geothermal wells: a fra- Yew, C.H., Weng, X., 2014. Mechanics of Hydraulic Fracturing, Secoend edition. Gulf
mework for numerical modelling. Renew. Energy 101, 737–746. http://dx.doi.org/ Professional Publishing eBook ISBN: 9780124200111. Hardcover ISBN:
10.1016/j.renene.2016.09.031. 9780124200036, pp 244.
Nikravesh, M., Kovscek, A., Johnston, R., Patzek, T., 1996. Prediction of formation da- Zarrouk, S.J., O’Sullivan, M., Croucher, A., Mannington, W., 2007. Numerical modelling
mage during fluid injection into fractured, low permeability reservoirs via neural of production from the Poihipi dry steam zone: Wairakei geothermal system, New
networks. Paper Presented at the SPE Formation Damage Control Symposium. http:// Zealand. Geothermics 36, 289–303.

37