You are on page 1of 12

SPE

SPE 21272
Estimating Gas Desorption Parameters From Devonian Shale
Well Test Data
H.S. Lane, Texas A&M U.; D.E, Lancaster, S.A, Holditch &Assocs. inc.; and A.T. Watson,
Texas A&M U.
SPE Members

~pflfght 1990, SOCiefYof Petroleum Engineers, Inc.


hris paper

was

prepsred for presentation at the SPE Eastern Regional Meeting held Irr Columbus, Ohio, CICtObW31-November 2, 19S0.

kIis paper wee selected for preeenlafion by an SPE Program Committee following review of information contained in an sbetrac: aubmifted by the author(s), Contente of the pqMr,
IS presented, have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(e). The material, as presented, doee not necessarily reflect
my positionof the society of Petroleum Engineere, ite offlcare, or mambers. Papera presented at SPE meetings are subject to publication review by Editorial Committees of the Society
)f PetroleumEngineers. Perm@ion to copy ie restrictedto an sbatractof not more than 200 wor@. Iliuefrafionsmay notbe mpied. The abstractshouldcontainconspicuousacknowtedgmenf
~fwhere and by whom the paper is presented. Write Publication Manager, SPE, P.O. Sox S3333S, Richardson, TX 75083-3S3S U.S.A. Telex, 730989 SPEDAL.

ABSTRACT

has been reported that the total gas production for a well
producing from a formation containing adsorbed gas can be
an order of magnitude higher than that for a well producing
from a conventional formation of similar properties.12 It is
thus very desirable to properly account for adsorbed gas
when modeling Devonian shale reservoirs.

The feasibility of detecting and accurately estimating


gas desorption parameters from a history match of Dev~
nian shale welI-test pressure data is examined. Both drawdown and buildup tests are analyzed, and baaed on the
results of these analyses a desorption-specific well-test design is proposed. The results from a simulated desorptionspecific test suggest that it is possible to characterize gas
desorption from a well test with reasonable accuracy, even
when the effects of desorption are partially masked by wellbore storage and skin effects.

Unfortunately, while thousands of wells have been


completed in the Devonian shales, little work haa been done
to chsxacterize the desorption of natural ~ss &or, shale surfaces. Although it is possible to deterrmne gas desorption
isotherms from laboratory measurements on shale core samples, such data are very scarce and are not currently being
measured on a widespread basis. Alternate sources of information on gas desorption in the Devonian shales are thus
being investigated. Two such sources of information are
measured production and well-test pressure data.

INTRODUCTION
The Devonian shales of the eastern United States are
considered a major potential source of do
tic natural gas,
contsinin an estimated 1860 to 2580 scf of gaa.l *2 It is
believed t% at natural gas is stored iJ
e Devonirm shales
as both conventional {free gas and a- adsorbed gas, or
as that is physically attached to the surface of the shales
t y Van der Waals-type forces. The existence of adsorbed
gss in the Devonian shales is usually sasociated with the
presence of kerogen,3 a coal-like orgmic material, as well
as with certain clay minerals such as illite4 which may be
major constituents of the shale,

In a previous study, the authors investigated the feasibility of characterizing gas desorption in the Devonian
shales from an analysis of production data.lO~l113 While
it was determined that quantitative estimates for gas desorption parameters could be obtained from a production
data history match (and that the presence of gaa desorption
could be detected statlsticrdly under certain condhions),
the uncertainty associated with these parameter estimates
was judged to be unsatisfactory. In this paper, the feasibllity of detecting and estimating gas desorption parameters
from an analysis of Devonian shale well-test data is examined. Objective, statistically-based techniques are used for
estimating gas desorption parameters, evaluating the accuracy of the resulting parameter estimates, and determining whether the presence of gas desorption can be detected
from measured well-test pressure data. Both drawdown
and buildup tests are investigated, and a deso tion specific well-test design is proposed. The results in3 cate that
it is possible to characterize gas desorption from a well test
with r~onable
accuracy.

Previous investigators have suggested that 50% or


more of the gas stored in the Devonian shales may exist as
adsorbed gas. These estimates have been derived from laboratory measurements of gas-shale adsorption isothermss6
aa well as from engineering studies based on shale gsscontent data.-g Despite this evidence that adsorbed gas
may account for a significant portion of the total gas. content of the Devonian shales, however, adsorption is still
considered to be an unconventional gaa storage mechanism
and is typically ignored in reservoir engineering analyses.
It haa been observed that when the desorption of this
adsorbed gas phase is ignored, siggifica.nt errors may arise
in production forecasts made for Devonian shale wells, particulaxly if the wells have been stimulated,lOJ1 In fact, it

BACKGROUND
It has been noted that significant quantities of Devonian shale gas may exist as adsorbed gaa.6-g The amount
of gas stored in the adsorbed state is a function of the free

References and illustrations at end of paper


113

,.

ESTIMATING GAS DESORPTION PARAMETERS


manw nvvnMr,iw
CUATw wulT_m7cfr
n~~fi

ESTIMATING

gas equilibrium pressure. Under theassumption ofisothermal conditions, this functional relationship can usually be
modeled by the Langmuir isotherm:
blp
Varf, =
1 + bzp

en=

**97*

DESORPTION

Here, we investigate the qua.ntitativc estimation of gas


desorption parameters from an analysis of hypothetical Devonian shale well-test data. The accuracy of the resulting
desorption parameter estimates is evaluated using standard
statistical techniques, and a reservoir-model selection procedure is employed to determine whether gas desor tion
can be detected statistically from an analysis of wel1-test
data. The methodologies used for the parameter cstinlation, error analysis, and model selection are summarized
and discussed next.

(1)

where V=d$represents the amount of g% adsorbed Per unit


volume (or mass) of reservoir rock, p IS the free gas equilibrium pressure, and bl and bz are ind~pendent desorption
parameters that depend on the formatlo,n and its temperature. Although alternate paramete~uzat?onsfor the Langmuir isotherm exist, the parameter]zatlon shown here is
particularly useful since a linear desorption isotherm can
be obtained from Eq. 1 by setting b2=0.

Pmuncter

Estimation

A nonlinear parameter estimation (automatic history


matching) algorithm is used to estimate values for unknown
reservoir rock propcrt.ies, including the desorption pamrneters, from a set of measured well-test pressure data. The
estimates for the unknown parameters are taken to be those
values that minimize a weighted-least-squares performance
index, defined by:

Fig. 1 shows adsorbed and total gas contents for two


shale samples as measured by Thomas and Frost.5 Also
shown are least-square fits of the adsorption data using
Eq. 1. Note that the Langmuir model does an excellent job
of describing the measured data. It thus appears reasonable to incorporate the Langmuir isotherm into a reservoir
model to account for the presence of adsorbed gas, The additional term describing gas adsorption will appear in the
accumulation term of the gas diffusivity equation:

(4)
where the vector IZn,contains a set. of measured well-test
data and ~, contains corresponding values from a numerical reservoir simulator. The weighting matrix, W, is specified so that maximum likelihood estimates are obtained for
the unknown parameters, P. For well-test pressure data,
measurement errors are assumed to be independent and
identically distributed so that W can be set equal to the
identity matrix. Eq. 4 is minimized subject to a set of constraints (upper and lower bounds) that are impc.sed on the
estimated parameter. Details of the constrained mininlizattifi procedure are reported elsewhere. *416

(2)

The subscript sc in Eq. 2 indicates that the adsorbed gas


density is evaluated at standard conditions. Since V.da is
a function of pressure only, the chain rule can be used (in
addition to the Real Gas Law) to rewrite Eq. 2 in a more
classical form:
(3)

It is well known that multiple local minima are often encountered when solving nonlinear least-squares minimization problems. This means that, depending upon the
.
initial estimates for the parameters ~ supplied by the user,
the minimization of Eq. 4 may terminate at a local minimum rather than at the desired global minimum, giving rise
to potentially large errors in the resulting parameter estimates. To overcome thk problem, a multistart approach is
employed whereby the minimization algorithm is initiated
from several carefully selected starting points (sets of initial
parameter estimates). Details of this multistart procedure
are given in Ref. 15.

where dvad,/dp,
the slope of the isotherm, can be readily
determined from Eq. 1. Note that Eq. 3 assumes radial gas
flow and homogeneous reservoir properties.
Bumb and McI<ee12 derived an approximate solution
to a form of Eq. 3 in order to examine gas-well testing in the
presence of desorption, They evaluated the pressure dependent pro erties (including dV=~,/dp) at an average pressure
so that t! e equation could be linearized and solved analytically. Their results indicated that in the presence of gas
desorption, semilog plots of pseudopressure vs. pseudotime
were similar in shape to those for systems without adsorbed
gas, While they did not attempt to determine (estimate)
the desorption isotherm from well-test data, they concluded
that desorption could not be detected from a single-well
production test. They further concluded that the effect of
desorption would show up as an increased effective compressibility, which is equivalent to a negative skin factor for
a production well. Bumb and McKees study considered
only pressure drawdown tests for infinite-acting systems.

Parameter Accuracies
R(servoir parameter estimates obtained from a history
match of well-test data will contain various levels of estimation error or uncertainty. These parameter errors will
depend on the level of measurement noise present in the
data, the sensitivity of the numerical reservoir model to
small changes in the estimated parameter values, the degree
of correlation that exists among the estimated parameters,
and the number and range of the available data. Two methods are used to analyze these parameter errors: a linearized
covariance analysis and a Monte Carlo analysis.

In this study, the method of nonlinear parameter estimation (or automatic history matching) is used to analyze
well-test data in order to detect and estimate gas desorption. A numerical reservoir simulator is employed so that
nonlinear, pressure-dependent gas properties can be properly accounted for. This approach allows for a much more
rigorous (and accurate) analysis of well-test data, since the
parameter estimation process does not rely on a subjective
(i.e., visual) analysis of the data but is instead based on
sound statistical principles. To our knowledge, the estimation of gas desorption parameters from well-test data has
not previously been addressed.

The covariance analysis is based on the assumption


that the reservoir model is suitably approximated as a linearized function of the parameters ~ in the vicinity of the
minimum of Eq. 4. Under this assumption, confidence intervals can be calculated for each parameter from sn estimate of the covariance matrix of parameter errors. 1718The
size of a parameter confidence interval rellects the uncertainty in the final estimate for the parameter: the larger the
114

rcl

n.a.

,ZLLIL

LANL,

V.L.

LANti ASLJ$K, A.l .

WIiLSUN

parameters, we can say that gas desorption can be detected


from the history match since simpler models (that do not
include desorption) were determined to be inadequate.

confidence interval the more uncertain (or less accurate) the


estimate.
It has been observed that the assumptions inherent in
a covariance analysis may ofttm be violated for nonlinear
reservoir history matching problems (particularly for those
problems requiring several parameters to be estimated simultaneously), resulting in very conservative (i.e., overly
large) estimates for parameter uncertainties.ll However,
parameter confidence limits can also be determined without msMng such assumptions about the behavior of the
reservoir model by using a Monte Carlo approach. This approach uses the results of a large number of history matches,
each involving the same set of underlying (or true) welltest data but a different set of measurement noise, to define
uncertainty limits for each estimated parameter. Unlike a
covariance analysis, a Monte Carlo error analysis can also
take the effects of imposed parameter constraints into account when evaluating parameter uncertainties. Details of
the Monte Carlo analysis are provided in Refs. 11 and 13.

Depending on the structure of the hierarchy of candidate modelst however, it may not always bc possible to
determine a single most appropriate model for describing
a set of well-test data. This situation may arise if one or
more pairs of candidate models are non-nested. For example, in Fig. 2 the 1P,D2 and lP,S models are both nested
with the most complete reservoir model ( 1P,D2,S , but they
are not nested with each other. That is, neit Ler model
contains the other since the each have unknown parameters that the other does not { the lP,S model includes skin
while the 1P,D2 model includes the desorption parameters).
Thus, the F-test cannot be used to directly compare these
two non-nested models. It is therefore possible that all the
other candidate reservoir models in the hierarchy could be
discarded on the basis of repeated F-tests? leaving these
two non-nested models as remaining candidates. If this
case were to arise, then additional independent information would be required to select the more appropriate of
the two models, so we would say that the effects of gas desorption could not be differentiated from skin effects on the
basis of the history match.

Model Selection
When analyzing a set of well-test data, an engineer will
typically have several different candidate reservoir models
at his disposal, all of which may be consistent with the
information available about the reservoir. While each candidate model will have a different set of unknown reservoir
parameters (which must be estimated from the data), the
most appropriate model for describing the data may not
be readily apparent. Choosing an appropriate model, however, is a very important task since an inappropriate selection may lead to parameter estimates that provide a poor
representation of the actual reservoir system.

RESULTS
Hypothetical pressure data were analyzed to determine
if gas desorption could be detected and/or estimated accurately from a well-test. Both drawdown and buildup tests
were analyzed, and based on the results of those analyses a
desorption-specific well-test design was proposed. The results from a simulated desorption-specific test suggest that
it is possible to characterize gas desorption from a well test
with reasonable accuracy, even in the presence of wellbore
storage and skin effects.

In classical well-test analyses (type-curve and semilog),


the shapes of curves constructed from the pressure data
(or derivatives of the data) are often used to help select a
reservoir model. This may be suitable for cases that can
be represented by relatively few unknown parameters, or
when parameters are not highly correlated. In other situations, however, a more rigorous method of choosing a
reservoir model is desired. In this study, for example, we
are interested in estimating the desorption parameters in
addition to the other reservoir parameters that one might
want to estimate from a well test. We would also like to
know whether the presence of desorption can actually be
detected from the measured data. Our purposes are best
served by using a statistically-baaed model selection procedure deacrib~d by Watson et al.~g

For each well-test scenario considered, a numerical gasreservoir simulato~ was used to simulate pressure data
that included the effects of gas desorption (as well as other
effects such as wellbore storage and skin). Random, normal deviates were then added to the simulated data to obtain measured well-test data. These measured data were
then history matched with the various candidate reservoir
models shown in Fig. 2. Parameter accuracies were then
evaluated and F-tests were erformed to determine if gas
desorption could be detecte J and/or estimated accurately
from the well-test analysis,
Note that the hierarchy in Fig. 2 does not contain parameters representing wellbore storage or boundary effects.
While these additional parameters can be easily included
in the hierarch , their effects are usually obvious from a visual analysis o ? well-test data and so statistically testing for
their presence is usually not required. On the other hand,
the effects of desorption and skin ma not be obvious from
the data, so those parameters are inc 1uded in the hierarchy
in order to determine if they can be detected statistically
from the history matches.

To ~se~his model s~lection procedure, a hierarchy of


n~sted c~&date reservoir models, such as that shown in
The candidate models are nested
Fig. 2, IS developed.
in that each of the simpler reservoir models (those having fewer unknown parameters) is contained in the most
complete model (that h~ving the largest number of parameters).
All of the candidate models in the hierarchy are
used to history match the measured well-test data. The resulting residual performance index values (final values from
the minimization of Eq. 4) are then used to compare a pair
of nested reservoir model?, via a statistical F-test, to determine the more appropriate mode$ the more appropriate
model is retained in the hierarchy while the other model is
discarded. Repeated use of the F-test with various nested
model pairs will determine the best model in the hierarchy for describln the data. Since the fh.1 model choice
has been made on tt e basis of a statistical comparison of all
candidate models, those pammeters that are included in the
chosen reservoir model can be said to have been detected
statistically (at a given level of confidence) from the data.
Thus, if the chosen model includes one or both desorption

The base parameter values used in simulating the welltest data are iisted in Table 1. For each scenario considered
(e.g., drawdown or buildup, finite- or infinite-acting, with
or without skin and wellbore storage), four different sets
of desorption parameters were used and four sets of welltest data were simulated so that the effects of varying the
desorption isotherm could be examined. The desorption
isotherms corresponding to these four parameter sets are
shown in Fig. 3. Note that the adsorbed gas content at the
initial reservoir pressure (accountin for a proximately 70%
of the total gas in place) is identlc
~for t~efour isotherms;

11s

.
ESTIMATING

GAS DESORPTION

.WQf)M
... .. lM?VIWJTAN
--.
v..

. .

PARAMETERS

!X4A1.~ WU3.1._TU.CT llATA


----

-----

---

..

-..

12nm
OEu

Qlf)?n
LLL,

within about 570 of the true forecsst (these results are not
shown here).

only the shape of the isotherm is varied. While all four sets
of resulting well-test data were analyzed for each scenario
considered, we present only detailed results for Case 3, the
intermediate desorption isotherm.

Finally, we examine the history match results for the


lP,S model, which was ~so chosen by the model selection
procedure. Table 2 inchcates that the mewmred ressure
data can be adequately explained by usin a skin f actor of
s=-O.2640, Since the data were simulatJ with the 1P,D2
reservoir model (which had a skin factor of zero), however,
these results suggest that the effects of a significant amount
of adsorbed gas can be explained by assuming that the
formation has been slightly stimulated. WMle it seems
plausible that this assure tion would affect the forecasted
production for the well, tEe lP,S model was also found to
provide a 35-year forecast that was accurate to withiri a few
percent (these results not shown).

Drawdown Tests
We first simulated an ideal, constant-rate (q=60 Mcfd ,
finite-acting drawdown test which contained neither weli bore storage nor skin effects. Fig. 4 shows that several
thousand hours of data were required for the effects of the
closed outer boundary to be apparent for all four of the
isotherms studied. Random errors were added to these
simulated data and the resulting measured data were hlstory matched using the candidate reservoir models shown
in Fi .2 (with an additional unknown parameter representing tf e drainage radks, r~). Selected history matches for
Case 3 are shown in Fig. 5 and the corresponding parameter estimates are listed in Table 2. For convenience, we will
use the notation fror ~ Fig. 2 when referring to the various
candidate reservoir n Iodels.

One possible explanation for this accurate production


forecast (despite using the wren reservoir model) is that
boundary effects were represent t in the data, meanmg that
some information about the reservoir gas content could also
be extracted from the history match. Note from Table 2
that the estimate for r, in the lP,S model was 31.5% larger
than the true value. This would correspond to an estimated pore volume that is 7370 larger than the true pore
volume, and thus to an estimate for the f&e-gas content
that is 73% larger than the true value. This su eats that
the lP,S model is attributing some of the adsorbe7 gas that
is present in the reservoir to excess free gas. WhUe thh
excess free gas does not account for all of the adsorbed-gas
content (which was roughly 2.5 times the free-gas content),
it is apparently sufilcient to produce a reasonably accurate
long-term production forecast. If the production characteristics of the well should change, however (e.g., if the well
were hydraulically fractured), it hss been shown that significant errors can arise in a forecast if the true reservoir
model is not used for the hktory match.ll

A visual snalysis of Fig. 5 suggests that only the simplest reservoir model in the hierarchy (1P) was inadequate
for describing the measured data. The remaining candidate
models all provided essentially identical fits of the data, as is
indicated by their similar residual performance index values
(see column ~ -~Table 2). Using these residual values, the
model selection procedure suggests that either the lP,D1
or the lP,S model are appropriate for describing the datty
as discussed previously, however, the F-test cannot be used
to compare these two models directly since they are nonnested. This indicates that the effects of gas desorption
cannot be dkitinguished from skin effects when analyzing
these drawdown data, even though a skin factor was not
used in simulating the data.
Of particular interest in Table 2 are the desorption
psrameter estimates obtained for the 1P,D2 model (i.e., the
model used to simulate the data as well as for the model
chosen by the F-test (lP,D1). J ote that when the 1P,D2
model is used to match the data, significant errors arise
in both desorption parameter estimates (bl and b2). The
desorption isotherm defined by these parameter values is
plotted in Fig. 6, along with the linear isotherm estimated
from the lP,D1 model. Wile neither estimated isotherm
appears to provide a good approximation to the true but
presumed unknown) isotherm, we note that the shape o 1 the
estimated isotherms are similar to the true shape at high
pressures. This is shown more clearly in Fig. 7, in which
the estimated isotherm slopes are plotted as functions of
pressure.

Ideal infinite-acting drawdown tests were also examined by analyzing only the first 500 hours of data from the
finite-acting case. .Par~eter uncertah+ties for the infiniteacting case, including estimates for the Isotherm slope, were
only slightly larger than those for the finiteactin~ case,
while the model selection procedure produced identical resuits (i.e., the effects of skin and desorption could not be
differentiated). Production forecasts based on the hhtory
matchln results, however, indicate that while the 1P,D2
and 1P,8 1 models again do a reasonable job of accurately
forecasting 35-year production behavior (for twsumed values of pW~ and re ), the lP,S model provides a forecast that
underpredicts the true production by about 30% (see Fi .$)
despite an estimated skin factor that is within 1% o f the
estimate for the finite-acting lP,S case. These results confirm that it was the presence of boundary effects in the
data that enabled the lP,S model to provide an accurate
forecast for the finite-acting case, llmthermore, the forecasts in Fig. 8 suggest that for the infinite-acting case, it
may be very important to acquire additional independent
information in order to help differentiate between skin and
desorption effects.

Fig. 7 indicates that both desorpt~on models sre able


to estimate the true isotherm slope near the initial reservoir
prewure. Thu?, while the individual desorption parameters
cannot be estimated accurately from the drawdown test,
it appears that information about the slope of the desorption isotherm can be extracted from the measured pressure
data. (Recall from Eq. 3 that the isotherm slope, which
controls the rate of gas desorption, appears explicitly in the
diffusivity equation.) Table 3, which contains the results
of the error analyses for the true (1P,D2) model, su ests
that 9 times out of 10, the isotherm slope (dVad8/dp Y can
be estimated:0 within ab?ut +5% at high pressures (near
p=pi=665
psla) but to wdhin only k60% at lower pressures (near p=pWt=217 psia). Fig. 7 indicates that with
the lP,D1 model, for which the isotherm slope is a constant, the accuracy of the slope estimate is even poorer at
low pressures. However, both models estimate the isotherm
slope suiilciently well to rovide 35-year production forecasts (for an assumed vJ ue of pWf) that are accurate to

Both finite- and infinite-acting drawdown tests were


ako analyzed with wellbore storage tiects present in the
messured data VWij=30bbl). The results (not shown here)
indicated that ti e accuracy of the parameter esdrnates decressed only slightly horn the ideal drawdown case. This
observation is not particularly surprising, since wellbore
storage effects dominate early-time data while Fig. 4 suggests that desorption effects become more important at
much later times (i.e., when the curves diverge). The error
snidysis results also suggest that Vwb CUI be estimated to
with]n about &lYo from messured drawdown data.

116

SPE

M.K
-----

21272

I.Attll?.
-.-,->

1).l?.
----

i.AtJ17ASTER.
A-T------

Fhmlly, drawdown tests that included both wellbore


storage and skin effects were a,nalyzed. Anegativeskinfactor was usedin simulating these data see Table 1) since
many Devonian shale wells receive smal\ breakdown treatments prior to undergoing testing or production, The results of the error analysis for the finite-acting case are presented in Table 4, Note tloxn Table 4 that when both desorption and skin effects are represented in the measured
data (i.e., when skin is non-zero and must also be estimated), the slope of the desorption isotherm cannot be estimated accurately from a hisiory match when the true reservoir model (1 P,D2,S) is used. In fact, the covariance analysis for th]s case indicates that the correlation coefficients 1
amen the estimates for the parameters &l, &, S, and ~e
are J nearly unity, suggesting that a change (or error) m
one parameter estimate can be almost perfectly compensated for by changes in the other parameters, Thus, it is
not surprising that the model selection procedure chooses
the lP,S model as the single most appro riate model for
describing the data, indicating that the ez ects of gas desorption cannot be detected from a drawdown test when skin
is non-zero.

MATSON

distin shed from skin effects for the ideal buildup case,
provi r ed skin is zero.
Table 5 indicates that the desorption parameter estimates for the chosen model (1P,D2) are much closer to the
true values than they were for the ideal drawdown case (see
Table 2). The slope of the desorption isotherm defined by
these parameter estimates is plotted in Fig, 12. Note that
the estimated 1P,D2 isotherm does a very good job of approximating the true isotherm slope over a wide range of
pressures. Table 6, which lists the results of the parameter
error analyses for the ideal buildup case, indicates that (9
times out of 10 the estimated isotherm slope will be within
about +570 oft L e true slope, while the estimated desorption
parameters will be within +30 to 40% of the true vrdues.
Both of these uncertainties are significantly smaller than
those arising from the ideal drawdown analysis.
When wellbore storage effects were added to the
buildup data, the model selection procedure chose the
lP,D1 rmd lP,S models as the most appropriate for describin~ the data, indicating
thattheeffects
of desorption could
no longer be dktinguished from skin effects (even though
the skin factor was zero). The accuracy of the desorption
parameter estimates (as well as the corresponding isotherm
slope) also decreased significantly for the true model (see
Table 7). This suggests that significant information about
gas desorption may be represented in the early-time buildup
data - information that is masked by the presence of wellbore storage effects. Thus, it appears that it may be important to take measures towards minimizing wellbore st~rage,
such as by using a bottomhole shutin tool, when des~gning
buildup tests for characterizing gas desorption.

While the strong correlation between the parameters


for the case with wellbore storage and skin leads to large
uncertainties in the parameter estimates, production forecssts based on the finite-acting history matches (not shown)
were found to be accurate to within a few percent for an
assumed constant vslue of Pwf. This can again i e attributed to the presence of boun dary effects in the measured
data, since it was found that all production forecasts for the
infinite-acting case contained significant errors. Fig. 9 indicates that for the infinite-acting case, the lP,S model underpredlcts the wells production potential while the 1P,D2,S
(true) and lP,D1,S models overestimate production. Not!,
however, that while the lP,S model wilI always underestimate production (since desorption is igmored), the models
that include desorption are capable of both underestimating and overestimating production, depending on whether
the estimate for skin is higher or lower than the true value.

Finally, both wellbore storage and skin effects were


added to the simulated buildup data. The results for this
case (not shown) were similar to those for the corresponding drawdown case: the model selection procedure chose
the lP,S model as the most appropriate for describing the
measured pressure data, indicating that the effects of gas
desorption could not be detected when skin was non-zero,
and the error analyses indicated that parameter uncertainties were of the same order of magnitude (although slightly
smaller) as those listed in Table 4. Therefore, while it appears that ideal buildup tests may offer significant advantages over ideal drawdown tests when attempting to characterize gas desorption, these advantages disappear when
wellbore storage and skin effects are present in the data.

Buildup Tests
Buildup data were simulated and analyzed for the scenarios presented in the previous section. In each case, the
well was shut in and tested for 500 hours after having been
produced at.a constant rate of q=60 Mcfd for 40,000 hours.
Note that since a numerical simulator was used to history
match the buildup data, it wss also necessary to simulate the (finite-acting) drawdown period preceding the test.
Thus, unlike for a classical (type-curve or semilog) analysis,
an estimate for the drainage radius could be obtained from
a history match of the measured. buildup data.

Desorption-Specific Well Tests


The results from the previous sections suggest that it
may be difficult (or impossible) to detect the presence of
gas desorption or to estimate the gas desorption parameters accurately when wellbore storage and skin effects are
present in the well-test data. The chief difficulty arises due
to the stron correlation that exists between the estimates
for the skin f actor and the desorption parameters. That is,
the effects of gas desorption can be adequately explained
by using a skin factor that is only slightly more negative
than the true skin factor. Note from Tables 2 and 5, however, that when the lP,S model was used to history match
the well-test data for the ideal cases (for which the true skin
factor was zero , the estimated skin factor was different for
the drawdown { s=-. 2640) than for the buildup (s=-.5214).
This suggests that a single value of s (or the lP,S model
itself ) might be inadequate for describhg a set of measured
cJ~d~fg~h drawdown and shutin data were available from
-.

Fig. 10 shows the ideal buildup tests (with no wellbore


storage or skin effects) simulated with the four isotherms
Selected history matches for
considered in the study.
Case 3, the intermediate isotherm, are plotted in Fig. 11
and the corresponding parameter estimates are listed in
Table 5. While a visual analysis of Fig. 11 suggests that
there is little difference am?ng.the data fits for the various
mndldate models, Table 5 mdlcates that the residud. perfomm%
index value for the 1P,D2 (true) model IS S1 ificantly smaller than those for the lP,D1 and lP,S mo Tels.
In fact, the model selection procedure indicates that the
1P,D2 reservoir mode! is the single most appropriate candidate model in. the luerarchy for describing the measured
data. Thus, unhke for the Jdeal drawdown ~ase, it appears
that the presence of a nonlinear desorption isotherm can be
detected from the ideal buildup data, although this is not
at all obvious from a visual analysis of the history matches.
It also appears that the effects of gas desorption can be

A desorption-specific well test was thus designed and


investigated. The simulated test consi~ted of a 100 hour
constant-rate drawdown followed by a 40 hour buildup,
11/

8
/

ESTIMATING GAS DESORPTION


PARAMETERS
mnnai n.sx,nwr ... em,, = ,.ru, , lrm.
eml n,m,

with pressure data being continuously measured throughout the entire test. The well was produced at a very high
flowrate (9=120 Mcfd) so that a significant pressure drawdown could be achieved over a short period of time. An
infinite-acting drawdown was considered to be adequate
since the previous drawdown analyses indicated that the
presence of boundary effects in the data did not significantly
improve the accuracy of the resulting parameter estimates.

O--

.* *.*

The buildup analyses indicated that the presence of a


nonlinear gas desor tion isotherm could be detected from
a history match an (? the isotherm slope could be estimated
with reasonable accuracy over a wide pressure range provided skin and wellbore storage effects were not present in
the data. The addition of wellbore storage effects significantly decreased the accuracies of the history match parameter estimates and destroyed the ability of the model selection procedure to distinguish between skin and desorption
effects, indicating that steps should be taken to minimize
the wellbore volume. The addition of skin effects produced
even larger parameter errors and resulted in the failure of
the model selection procedure to detect desorption from a
history match.

We first simulated an ideal well test having no wellbore


storage or skin effects. Table 8 lists the results of the history
matches for the various candidate reservoir models. Note
that the residual performance index value for the 1P,D2
[)true model is two orders of magnitude smaller than those
or the lP,D1 and lP,S models, enabling the model selection
~)F-test procedure to detect the presence of the nonlinear
esor tion isotherm. The parameter error analysis for this
case rnot shown) suggests that, 9 times out of 10, the errors
in the desorption parameter estimates will be less than 10%
while the estimated isotherm slope well be within l~o of the
true value over a large pressure range. Thus, the proposed
well-test design appears to be a significant improvement
over either a conventional drawdown or buildup test for
the ideal case.

Based on the results of the drawdown and buildup


analyses, a desorption-specific well test was designed that
allows a nonlinear desorption isotherm to be detected even
in the presence of wellbore storage and skin effects. The
test, which requires the use of a bottomhole shutin tool
to minimize the wellbore volume, consists of a short-term,
high-flowrate, constant-rate drawdown followed by a short
buildup, with pressure data being continuously measured
throughout. Simulated results showed that the slope of the
desor tion isotherm could be determined to within about
+10 ?o over a large pressure range, even though the effects
of gas desorption were partially masked by wellbore storage
and skin effects.

When simulating pressure data for the case with wellbore storage and skin, it was assumed that a downhole
shutin !.001would be used to limit the wellbore volume so
that wehbore storage effects would not seriously affect the
buildup portion of the test. A wellbore volume of VWb=5.O
bbl was thus used (along with the other parameter values
,listed in Table 1 in simulating the data. The results of
the history mate i es for this case (not shown) were similar to the ideal case in that the model selection procedure
again detected the presence of the nonlinear desorption
isotherm. (In fact, the res:dual performance index value
for the 1P,D2,S [true] model was two orders of magnitude
smaller than those for the lP,D1 ,S and lP,S models. ) The
results of the error analysis, shown in Table 9, indicate
that in the presence of wellbore storage and skin, the uncertainty in the desorption parameter estimates was +20
to 30% while the uncertainty in the isotherm slope was
in the + 107o range. These results are again a significant
improvement (of an order of magnitude or more) over the
previous results presented for the drawdown and buldup
cases. Thus, it appears that with the suggested well-test
design, the presence of gas desorption can be detected and
the gas desorption parameters (or, more specifically, the resulting slope of the desorption isotherm) can be estimated
with reasonable accuracy from a history match, even when
the effects of desor tion are partially masked by wellbore
storage and skin e$.ects.

NOMENCLATURE
Langmuir desorption parameter, scf/ft3/psia
Langmuir desorption parameter, psia-*
gas formation volume factor, ft3/scf
gas compressibility, psia- *
least-squares objective function, psiaz
permeability, md
P = pressure, psia
re = drainage radius, ft
=
wellbore radius, ft
~w
s = skin factor
t = time, hours
amount of gas adsorbed, scf/ft3
l~s
VWb = wellbore volume bbl
W = weighting matri;
Y. = measured datum
Y= = simulated datum
6 = reservoir parameter estimated from data
79 = gas gravity (air=l.0)
P = density, lb/ft3
P = viscosity, cp
~ = porosity, fraction
Subscripts and Superscripts

SUMMARY
Drawdown and buildup tests were analyzed in order
to determine whether gas desorption could be detected
and/or estimated accurately from measured well-test pressure data. The drawdown analyses suggest that the desorption parameters could not be estimated accurately from a
history match and that the effects of gas desorption could
not be distinguished from skin effects. In the absence of
any skin effects, the slope of the desorption isotherm (which
controls the rate of gas desorption) could bc estimated to
within a few percent near the imtial reservoir pressure.
With anon-zero skin factor, however, estimates for the desorption arameters and the isotherm slope could be in error
by 100#0 or more. This may !ead to significant errors in production forecasts when boundary effects are not present in
the history matched data.

bl
b2
B~
(+
J
k

=
=
=
=
=
=

i
Sc
T
Wf
Ws
+

=
=
=
=
=
=

initial
standard conditions
transpose operator
wellbore flowing
wellbore shutin
vector

ACKNOWLEDGEMENT
This reserwh was carried out as part of a project sponsored by the Gas Research Institute under Contract No.
5086213-1446, Reservoir Engineering and Ikatment Design Technology,

I
..-

11s

,,

..----.-

U.GY.

-Xla,

.!2.

I.ru,

u,

Lun,

Wnbavm

14. Watson. A.T., and Lee, J. W.; A New Algorithm for


Automatic History Matching Production Data, paper
SPE 15228 presented at the 1986 SPE Unconvention~
Gas Technology Symposium, Louisville, May 18-21.

REFERENCES
1. Unconventional Gas $ourcew Volume IZI, Devonian
Shale, National Petroleum Council , Washington, DC
(1980).

15. Murtha, J.A., Gatens, J.M. III, Lancaster, D.E., Lane,


H, S.,Lee, W. J., Olarewaju, J. S., and Watson, A.T.:
tpractic~ Methods for Analyzing Well Test and Production Data from Devonian Shale Reservoirs, S.A.
Holditch & Associates, Inc. Topical Report to the Gas
Research Institute, GRI Contract No. 5084-213-0980
(Oct. 1987).

2. Zlelinski, R,E. and McIver, R. D.: Resource and Exploration Assessment of the Oil and Gas Potential in
the Devonian Shales of the Appalachian Basim Mound
Facility Report to the US. Department of Energy,
DOE Contract No. DP-0053-1125 (June 1982).
3. Seeder, D.J.: Porosity and Permeability of Eastern
Devonian Gas Shale: SPE Form. Eval. (Mar. 1988)
116.24.

16. Murtha, J.A., Gatens, J.M. III, Lancaster, D, E., Lane,


H.S., k:
W. J., Ohuewaju, J. S., and Watson, A.T,:
Reservoir Engineering and Treatment Design Technology, S.A. Holditch & Associates, Inc. Annual wport to the Gas Research Institute, GRI Contract No.
5086-213-1446 (Dec. 1987).

4. Schettler, P.D., Parmely, C. R., and Woomer, M.T.:


Physical Properties of Clay Minerals and their ReIationshlp to Devonian Shale, Juniata College Topic~
Report to the Gas Research Institute, GRI Contract
No. 5085-213-1143 (Sept. 1989).
Adsorpand Frost, R. R.:
5. Thomas, J. Jr.
tion Desorption Studies of Gases Through Shales,
Studies of the ;lew AlGeo{ogical and Geochemical
bany Shale Group (Devonian/Mississippian)
in Illinois, R.E. Bergstrom and N.F. Shimp (eds. ), Illinois State Geological Survey Fhml Report to the U.S,
De artment of Energy,DOE Contract No. DE-AC2176i T12142 (June 1980) 143-60.

17. Bard, Y.: Nonlinear Parameter Edimation,


Press, Inc., Washington, DC (1974).

Academic

18. Beck, J.V. and Arnold, K. J.: Parameter Estimation


in Engineenng and Science, John Wiley & Sons, New
York City (1977).
19. Watson, A. T., Gatens, J.M. III, and Lane, H.S.:
Model Selection for Well Test and Production Data
Analysis? SPE Form. EvaL (Mar. 1988) 215-21.
Jbfan20. Sawyer, W. K.: U~ers Guide and Documentation
Mathematical and Computer Serual for SUGAIUII,
vices, Inc., Morgantown, WV (1987).

6. Schettler, P.D. and Parmely, C. R.: Physicocllemicd


Properties of Methane Storage and Transport in Devonian Shales, Juniata College Annual Report to the
Gas Research Institute, GRI-90/0013 (May. 1989).
7.

m.z.

Kuuskraa,
V. A., Wicks, D. E., Sawyer, W. K., ad Esposito, P.R.: Technically Recoverable Devonian Shale
Gas in Ohio? Lewin and Associates, Inc. Report to
the U.S. Department of Energy, DOE/MC/19239-1525
(July 1983).

S. Kuuskraa, V. A,, Wicks, D. E., Sawyer, W. K., and


Esposito, P.R.: Technicall~ Recoverable Devonian
Shale Gas in West Virginia, Lewin md Assoclat~,
Report to the U.S. Department of Energy,
Inc.
DOE/MC/1923Q+750 (Dec. 1984).
9. Kuuskraa, V. A., Sedwick, K. B., Thompsm, K. B.,
Wicks, D.E., Sawyer, W.K., and Esposi o, i %: Technically Recoverable Devonian Shale Gt +- .entucky:
Lewin and Associates, Inc. Report to the
S. Department of Energy, DOE/MC/19239-1834 (May 19S5).
10. Lane, H.S., Lancaster, D. E., smdWatson, A.T.: Characterizing the Role of Desorption in G,% Production
from Devonian Shales? Energy Source?, m press.
11. Lane, H, S.,Watson, A.T., and Lancaster, D.E.: Identifying and Estimating Desor tion Fkom Devonian
Shale Gas Production Data: i .A. Holditch & Associates, Inc. Topical Report to the Gas Research Institute, GFtI Contract No. 5086-213-0980 (Aug. 1988).
12. Bumb, A.C. and McKee, C.R.: Gas-Well Testing in
the Presence of Desorption for Coalbed Methane and
Devonian Shale, SPE Form. EVUL (Mar. 1988) 17985.
13. Lane, H.S., Watson, A.T., and Lancaster, D.E.: Identifying and Estimatin Desorption nom Devonian
Shale Gas Production 6 ata, paper SPE 19794 presented at the 1989 SPE Annual Technical Conference
and Exhibhion, San Antonio, Ott. 8-11.
.

llW

SPE 21272

Table 2- History Matching Results For Irlcal, Finite-Acting Drawdown Test (Case 3)

Parsmcter Estimates
s

Model

trnr

Table 1- Parameters Used For Simulating Well-Test Dato

Freed (Known) Parameters:


Initial Reservoir Pressure, pa (psia)

665.0

Dry Gas Gravity, -yg


Wellbnre Radius, rw (ft)

0.700
o,~130

Net Pay Tbichkness, h (ft)

120.0
0.030

Formation Pornsity, #

(scf/f;~/psia)

(Ps::- )

1000.

2.000E -2

3.870E 3

109s.
1001.
1007.

9.W6E-2

1.428E 2

1.161E-2

z.590E.3

1.6XE 3

(id)

(2)

.1000

(ps;a2)
.

1P,D2,S
1P,D2
1P,D1 ,S

15X)
16.27

.1001
.1OO1
.1000

lP,D1
lP,S
1P

16.27

.Iooo

1007.

16.51

.1000
.1046

1315.

15.60

838.0

,09q53
0.0

1.655E -3

-.2640

1269.

models chrrscn by F-t w+t

Hktory Matched (Unknown) Parameters:


Formation Permeability, k (red)
Drainaga Radius, r, (ft)
Dssorptinn Parameter-.

0.100
1000.
Table 3 Error Analysis Rrwdts Frrr ldmd,
Finite-Acting Drawdown Teat (Caw. 3)

bl @f/ft3/psia)

case1
case2
c-3

S.6E-3

Caad

8. OE-2

8.OE-3
2.OE..!?

Parnmctcr
& (psia-}
case 1

c= 2
(-he 3
cm

Weffbore stnrage Vohrme, vu$ (bbl)


Skin Factor,s

nOtuaed f0rrdlc2w2s

Parnnmtcr Errnrs:
9070 Confidcncc Lknits
Monte Carlo
Covariance
Arm&siY
A;$~;

Tab]c 4 Error Analysis Results For Fhtitc- Acting


Drawdown Test lVith W,41hore Storage And Skin (Ca.sr 3)

Parametm Errvrx
90% Confidmcc Limits

0.0

Mnntc Carlo
Analysis

Pnramctm
.

%)

Cowrriallcr
Analysis
(+%)
_

6.5E4

[.1300, .Iooo]

.1393

[- .3500, .3000]

.7000

3.&E-3
2.OE-2

r,

[ .6600, .8000]
1-84.40, 400.0]
[--84.30,178.1]

.!)621
421.7
296.4

r,

[37.06, 31.30]
[ 99.50, 268.4]
[ 33.28, 2848.]
I 5.810, 2.000]

2165.
22300.
23030.

30.00
-2.00

b,
b,

at:
p=217 psia
p=300 psia

dV.~./dp

p=400 psia
p=500 psia

b,

b
v.~
s

1-58.75, 57.52]
1.47.85, 35.67]
1-34.40, 20.13]
1-19.60,10.34]

dv.a.}dp

122.56,

13.60]

4..584

1084.

al;

P=362 psia

j 100.0, 341.7]

p=4Ul psia

J-100.0, 345,0]
I -100.0, 351.5)

p=600 psia

[-6.330,3.621]

p=500 psia

p=665 psia

[-4.037,5s85]

P=GOOPhl

[- 100.0, 356.7]

p=665 psia

[ 100.0, 361.9]

impossd constraint
impnscd constraint

spE 21272

hble 6- Error Analysis Results For ideal Buildu~ Trst (Case 3)

Table 5- History Matchkzg Results For ideal Buildup Test (Caae 3)

Parameter Errors
90% Confidence Limits
Monte Cz.rlo
yn~$??
Arzal~is
[i%)

Parameter E6timate2
Model

(Pia)

true
IP,D2,S
1P,D2

12.85
12.92
24.31
24.31
26.13
797.9

lP,D1,S
lP,D1
lP,S
1P

(;:)

(scf/f~J/psia)

.1000

1000.

2.000E-2

3.870&3

.09848
.09951

1153.
9S2.1
822.1

1.338E-2
1.617E-2

5.186B3
3.050E-3

(~d)

.09562
.09562
.09507
.09969

3.945E-3
3.945&3

822.1
1347.
1264.

(J:-

Parameter

-.2350

h,

0.0

at:
p=270 psia

di<d,/dp

model chosen by F-test

Tnble S

.6095
1.718
47.06
39.76

[- 30.21, 48.60]
[-25.31, 38.00]

h
-.5214

Table 7- Error Analysis Rcsrdts Frw Buildup


Test With Wellbore Storage (C= 3)

[-.4900, .5000]
[-1,570, 1.400]

r<

p=300 psia
p=400 psia

[-5.639, 4.508]
[+.757, 3.20S]
[4.423, 4.653]

P 500 psia
p =55Spsia

[-6.767, 6.135]
[-7.906, 7.124]

History Matching Results For Idenl Dcsorptioa-Specific Well Test (Case 3)

Parameter E2timatcs

Pruameter

[-.9200, .6000]
[4.720, 3.700]
[-53.41, 400.0]

r,
b,
b,
Vmb
dvad./dp

Parameter Errors
!)0% Confidence Ltmits
~owui ante
Monte Carlo
.hmlysis
Ancdysis
(+%)
(%)

[-54.17, 204.2]

3G3.6

[-.8300, 1.000]

2.S29

at

p=217 psia
p=300 psia
p=400 psia
p=500 psia
p=558 psia

2.s96
1?.s3
424.5

Model
_

(ps;az)

true
1P,D2.S
1P,D2
IP,D1,S
lP,D1
lP,S

8.425
8.50S
434.3
434.3

bi
(scf/ft /psirr)

.1000

2.000 E-2

3.8iOE-3

.1000
.1000
.09665

2,250E_~

4.341L3
4.031)2-3

.09665
.09627

549.4

(n!d)

2.129E-2

(p.!:+

-.010s9

2.S63E-3
2.S63E-3

0.0
-.4061

model chosen by F-test


[-19.36, 28.64]
[-8.215,
[4.691,
[-11.08,
[-14.16,

11.51]
8.830]
16.90]
21.31]

Table 9- Error Analysis Results For Desorptimr-Specific


Well Test With Wellbore Storage And Skin (Case 3)

imposad constraint

Parameter

Parameter Errom
90% Confidence Limits
Monte Carlo
Covariance
Am&~is
A;&n&is

. .
k

[-.2500, .2000]

b,

v=~

[-13.24, 31,80]
[-15.65, 33.20]
[-2.520, 2.500]

[-1.550, 1.600]

&ad./dp

at,

P=253 psia

[-7.465, 7.782]

p=300 psia

[-7.536,
[-10.34,
[-13.10,
[-15.46,
[-16.50,

p :400 psia
p /500 psia
p .600 psia
P=657 psia
121

8.925]
11.27]
13.05]
14,44]
15.11]

.,
.2s01
28.05
41.94
3.475
3.670

sPE 21272

12

o
0

SAMPLE I
SAMPLE 2
LANGMUIR ISOTHERM

II
ToTAL
NUMBER OF
INDEPENDENT
PARAMETERS

0 /

0
1P, s

1P, D2

ADSORBED

lP

800

400

1200

1600
::s

PRESSURE, psia

.
.

SINGLE
POROSITY
LINEAR DESORPTION
ISOTHERM
LANGMUIR
DESORPTION
ISOTHERM
SKIN

Fig.

4.0 m
~
\
:.
0
w
z 3.0

600

2 - 111.,..,1>?

of candidate

,...,.1,

models,

HIGHLY NONLINEAR

0
a
u
5
:

0
.5
a

INTERMEDIATE

2.0

400 -

z
NEARLY LINEAR
~

200 -

1.0

Case
Case
------Case
-.Case4

I
2
3

.i

200

400
PRESSURE,

600

Fig. 3 - Lanwuir deeorptim isotherms used


well-test

800

TIME, haurs

psio

tosimulate

Fig.

data.

122

4 - Simulated data for ideal,


wellbore
storase
or skin

finite-ac
effects).

clng

drawdown case

(with

n.

=21272

4.0

TRUE
IP,
D2
-----IP,
DI

3.0

600

500

400

a\

2.0

0
.G
a
~

Lo

Measured

o
300

-------

~-

1P
IP,02 - IP,DI - IP,S

..

~-

--

~. #-

*H

-e

\
I

200

,Q-1

,0-2

100

10

102

,.3

,04

\]

0I
105

200

TIME, hours

400

PRESSURE,

600

psia

0.020

TRUE

1P, D2
1P, DI

-----

.TRUE
---JP,D2
. IP, S

\
\\

400

200

- IP, DI

---------------------0

O*

200
PRESSURE,

Pip,.

400

600

10

15

20

25

30

35

TIME, years

pskf

1 - True and estimaccd


desorption
isotherm slopes
for ideal,
f init.-acting
drawdown (Case 3).

I
5

Fig.

8 - Production
forecas:s
bescd on parameter estimates
ideal,
infinite-act
inn drawdown (C8$e 3).

deter.xned

from

spE 21272

1200

TRUE
IP, D2, S
----IP,
DI, S
.-1PS

/,
,0-

800
/

,/

Case
Case
-------Case
.Case

550

2
3
4

/
45C

/
350
400

.e -- 0

250
,/
0
150
---

0
0

15

10

20

30

25

35

102

IO-3

Fig.

[OQ

10-

TIME, years

lot

lo~

TIME, hours

9 - Production
forecasts
based on parameter estimates
determined
froa
in finite-acttns
drawdobm with uellbore
storage and skin effects
(Case 3).

Fig.

10 - Simulated data for


skin effects)
.

ideal

buildup

case

(with

n. wellbore

stcwa~e

TRUE
IP,
D2
------1P,
DI

550

=
z
~x

o Measured
---1P,
D2
.1P, DI
. .. .... . Ip,s

/flB

\
\
\

g- O.o1o
~
W

*S

/
350

T
:
#.R

450

0!015

F
o
Q 0.005

.@gq
,,2P

250

/
150 ,o~

o~
o

200

TIME, hours

Fis.

11 - Selecttd

history

matches

for

ideal

400

600

PRESSURE, psla

buildup

(Case

3).

Fig.

12 - T...
far

and
ideal

stimated

butldup

desorption
3).

(tiSO

isotherm

81OW

or

You might also like