SPE 84478
Planning and Interpreting NMR Fluid-Characterization Logs
Chanh Cao Minh, Nick Heaton, Raghu Ramamoorthy, Eric Decoster, Jim White, Elmar Junk, Schlumberger;
Ramsin Eyvazzadeh, Osama Al-Yousef, Saudi Aramco; Rosaura Fiorini, ENI Dacion B.V. and Darren McLendon,
ExxonMobil
Copyright 2003, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the SPE Annual Technical Conference and
Exhibition held in Denver, Colorado, U.S.A., 5 – 8 October 2003.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Abstract
This paper presents a logical approach to ensure successful
nuclear magnetic resonance (NMR) logging for fluid-
characterization purposes. Numerous examples are shown in a
three-stage process consisting of a) planning, b) processing
and interpretation, and c) validation and integration.
First, modeling is performed to assess the NMR contrast,
using any available information concerning anticipated fluid
types. Knowledge of downhole conditions also allows the
estimation of the NMR signal-to-noise ratio (SNR) from
which the sensitivity and statistical error can be evaluated. The
outcome of the planning process is a suite of NMR pulse
sequences that optimizes resolution of the different fluids and
minimizes error with the shortest possible acquisition time.
Interpreting fluid-characterization NMR logs is relatively
straightforward when the fluids have large NMR contrast and
obey the standard correlations. However, dissolved gas,
wettability variations, internal field gradients, and restricted
diffusion can cause deviations from model behavior and have
significant effects on NMR relaxation data. These effects need
to be recognized and accounted for.
To gain insights into the NMR-based fluid evaluation process,
we have developed a model-independent technique that gives
various sets of D-T1-T2 maps. The maps are analogous to the
crossplots commonly used in log interpretation practices and
are indispensable in the interpretation of the data. We also
explain quantitative interpretation techniques from D-T2
maps.
Integrating NMR fluid-characterization results with other
information and openhole logs constitutes the final step. As
with any technique, results need to be crosschecked with other
log data for consistency. The answer gives a water saturation
that can be verified with other tools and techniques such as the
dielectric log, resistivity logs, and the density magnetic
resonance technique method.
Introduction
Current methods to analyze fluids using suites of NMR
measurements employ model-based inversions. Two examples
of the forward-modeling approach are the MACNMR1 and
Magnetic Resonance Fluid (MRF) characterization methods.2
The MRF technique is based on physical laws that are
calibrated empirically to account for the downhole fluid NMR
responses. By using realistic fluid models, MRF aims to
minimize the number of adjustable parameters to be
compatible with the information content of typical NMR log
data. Since the model parameters are by design related to the
individual fluid volumes and properties, determination of the
parameter values (i.e., data-fitting) leads directly to estimates
of the fluids petrophysical quantities.
Any forward-model approach relies on the validity of the fluid
models employed. In “non-ideal” situations where the fluid
NMR response deviates from the model behavior (such as
internal gradient, oil-wet rocks, restricted diffusion etc.), these
techniques may lead to erroneous answers. In some
circumstances, “non-ideal” responses may be identified by a
poor fit-quality of the echo data, in which case the fluid
models can be adjusted by modifying the appropriate model
parameters. However, it may not be obvious which element of
the fluid model should be modified and what modification is
needed to get the desired answer.
The maximum entropy principle (MEP) method is a model-
independent inversion that provides a simple graphical
representation of NMR data for fluid analysis in all
environments. The graphical representations (i.e., multi-
dimensional distributions) can themselves be used directly for
interpretation or, alternatively, they may be used to guide the
selection of parameters for model-based processing such as
MRF. It is important to recognize that the MEP technique as
well as the methods to interpret D-T2 maps are applicable to
both CPMG (Carr, Purcell, Meiboom, and Gill) and DE
(diffusion editing) 3 measurements.
Planning
The purpose of the planning process is to derive an optimal
NMR suite of pulse sequences that provides the greatest
2 SPE 84478
resolution of the different fluids, the lowest possible result
error and the fastest possible acquisition time. Because crude
oils have a wide range of relaxation and diffusivity, it is useful
to divide the oil into classes of light, medium, and heavy based
on viscosity. Diffusion, T1 and T2 relaxations of water, oil and
gas are plotted versus temperature in Fig. 1. In the diffusion
plot, all three fluids are well resolved, except for water and
light oil. In the T1 plot, at low temperature water overlaps with
gas and light oil. At high temperature, gas overlaps with light
oil. In the T2 plot, the overlap straddles several fluid types,
thus explaining why it can be difficult to differentiate fluids
with T2 information only. The planning strategy is as follows:
1) For medium-light and medium-heavy oil (viscosity
range 2 cp to 100 cp), the recommendation is to use
several diffusion sequences, for example, four
sequences with similar wait-time but different echo
spacings (such as 0.2 ms, 2 ms, 4 ms, 6 ms).
2) For gas, water in large pores, and light-oil (viscosity
range 0.2 cp to 2 cp), the recommendation is to use
multi-wait sequences in addition to diffusion
sequences, for example, six sequences with three
different wait-times (such as 12 s, 3 s and 1 s) and
four different echo spacings (such as 0.2 ms, 2 ms, 4
ms, 6 ms).
3) For heavy oil (viscosity greater than 50 cp), the
recommendation is to maximize the signal-to noise
ratio (SNR) of the longest diffusion sequences, for
example, by using five sequences with different echo
spacings with several repeat stacks of the long echo
spacings sequences (such as 0.2 ms, 4 ms, 6 ms, 8 ms,
10 ms with 8 repeats). Another option is to use the
new DE sequences.
The water and oil volume errors attributed to a suite of NMR
pulse sequences and its SNR are described elsewhere.4 Note
that these errors do not include systematic errors associated
with the fluid models. If the SNR of the NMR logging tool is
characterized for environmental effects such as temperature
and salinity, one can predict the acquisition time for a desired
SNR and examine the corresponding errors. For example, for
the CMR-Plus * Combinable Magnetic Resonance logging tool
with high-logging speed capability, the MRF station time is
2.67 minutes for a typical suite of five DE sequences but 5
minutes for the equivalent suite of five CPMG sequences. Yet,
the DE suite gives lower error and wider T2 range than the
CPMG suite. As a general rule, increasing SNR will lower the
error curve, maximizing T1 information with multi-wait
sequences will extend the long T2 range, and increasing the
echo spacing with several repeats will extend the short T2
range.
Processing and Interpretation
A general model-independent method, based on the maximum
entropy principle (MEP), has been developed to analyse multi-
measurement NMR data governed by distributions of more
* volumes must be corrected for hydrogen index to give correct
Mark of Schlumberger
than one property (eg. T2, T1, and diffusion). The objective of
the method is to present complex data in an understandable
format, such as the D-T2 map, without imposing any bias
while minimizing noise-related artifacts. A more detailed
discussion of the model-independent processing method is
provided in the Appendix.
D-T2 maps generated by model-independent inversion can be
used to:
1) Observe model deviations and assist model-based
inversion codes such as MRF,
2) Identify the different fluids and derive direct
petrophysical answers from the map itself.
Fig. 2 shows an example of a D-T2 map.
Assisting Model-Based MRF Inversion and Analysis
Just as T2 distributions provide interpretable
representations of single-measurement NMR data, D-T2 maps
provide unbiased representations of multi-measurement data
suites. An overlay of the theoretical responses of the three
most encountered fluids (i.e., water, oil, and gas) shows
whether the fluid responses need any adjustments for use in
model-based inversion.
For example, the MRF model (for oil, gas, and water) states
that the water and gas diffusion constants are independent of
T2, and depend on temperature T and pressure P (for gas):
Dw ( T 2 ) = Dw (T ) . (1)
Dg (T 2 ) = Dg ( T ,P ) . (2)
For oil, the diffusion constant is linearly proportional to T2,
DO (T 2 ) = λ × T 2 . (3)
It follows from Eqs. 1 to 3 that we can overlay two horizontal
lines (i.e., at constant D values) representing the theoretical
responses of water and gas, and a diagonal line representing
the theoretical response of oil on a D-T2 map. Deviations from
the ideal fluid responses will be evident in the maps as the
data will be located away from the overlay lines. Some causes
of the deviations from ideal responses are shown in Fig. 3.
Direct Petrophysical Analysis from D-T2 Maps
Because D-T2 maps contain essentially the same
information that is in the original echo data, it is possible to
use the maps themselves as starting points for fluid
characterization. We have developed two different
approaches: the visual point-and-click approach and the log-
mean diffusion approach. The merit, applicability, and
demonstration of each method are discussed below.
Visual (Point-and-Click) Approach
In some cases, fluid distributions are well separated in the
D-T2 maps and their respective volumes can be directly
obtained by integrating the amplitudes in the D and T2
windows defined by each type of fluid. The corresponding
saturations can then be derived by dividing the respective fluid
volumes by the total fluid volume. Note that the individual
SPE 84478 3

answers. In this approach, the task of identifying the fluids Sequence WT(s) TE(us) NECHO REPT
from the map and defining a fluid model for quantitative 1 2.4 200 3000 1
2 2.4 2000 300 1
analysis is left to the interpreter. This is analogous to picking
3 2.4 4000 100 1
the shale point, the matrix line, and the gas point in the 4 2.4 6000 100 1
density-neutron crossplot. Table 1– CPMG Pulse sequences in an Asian well.
Log-Mean Diffusion (DLM) Approach At this particular depth taken in the water zone, the rock
In standard MRF analysis, the raw echo data are fit directly possesses substantial internal field gradients. The internal
using the constraints of Eqs. 1 to 3. An alternative approach is gradients add to the tool gradient and cause an additional
to use the maps themselves as input to derive the MRF diffusion relaxation mechanism. Since the inversion uses just
solution. Since the information contained in the maps is the fixed tool gradient, in order to fit the data it must use
essentially the same as that in the original echo data, the two increased diffusion rates.5 Sidewall cores also show a high
methods of solution should be comparable. In practice, the content of clay minerals at this depth. The internal gradient is
NMR data often lack resolution in the diffusion dimension, created by the difference in magnetic susceptibility of the iron-
and the different fluids D-T2 amplitudes are spread over large rich clay minerals and the surrounding quartz mineral in this
areas of the map. The problem then consists of reassigning the shaly interval. For other MRF stations taken in cleaner
amplitude spread in the diffusion dimension to the different sandstone intervals, the water diffusion constant agrees with
formation fluids. A simple approximate way is to use the the theoretical values (data not shown).
logarithmic (i.e., geometric) mean diffusion rate for each T2,
DLM(T2) computed from the maps, and redistribute the Restricted Diffusion Example
amplitude at this T2 according to the chosen fluid model. For Fig. 5 shows an example of restricted diffusion in a known
example, for a model consisting of water and oil, the log-mean water zone of a carbonate reservoir in the Middle East. The
diffusion rate at each T2 is well was drilled with a water-base mud. The CPMG pulse
sequences used in this job are listed in Table 2.
DLM (T 2 ) = DW (T 2 )S X 0 ( T 2 ) DO (T 2 )1− S X 0 ( T 2 ) (4)
Sequence WT(s) TE(us) NECHO REPT
where DW(T2) and DO(T2) are the diffusion rates for water and 1 12 200 5000 1
2 3 2000 300 1
oil, respectively, at each T2 value. The effective water
3 3 200 3000 1
saturation at each T2 is then obtained by rearranging Eq. 4
4 3 4000 200 1
ln ( DLM (T 2 ) / DO ( T 2 )) 5 1 200 3000 1
S XO (T 2 ) = . (5) 6 3 6000 100 1
ln ( DW ( T 2 ) / DO (T 2 )) Table 2 – CPMG Pulse sequences in a Middle East well.

The free water (T2 ~ 300 ms) undergoes unrestricted diffusion

Separate water and oil T2 distributions, FWAT, and FOIL, can now in large pores and agrees with the theoretical response. The
be derived from the total T2 distribution, F(T2i ): bound water is trapped in smaller pores, and experiences
restricted diffusion.6 A model-based inversion processing
FWAT (T 2i ) = S X 0 (T 2i ) × F(T 2i ) , (6) might compute partial oil saturation in this case.

FOIL (T 2i ) = ( 1 − S X 0 ( T 2i )) × F(T 2i ) , (7) Gas Example

F(T 2i ) = ∑∑ F( T 2 ,D ,(T 1 / T 2 )
j k
i j k ).
It is clear that the DLM approach is possible only when the fluid
model has just two components. In the case of a model involving
more than two components, the extra components are
successively eliminated from the D-T2 map. For example, for a
water-oil-gas model, the gas component can be eliminated from
the map using the Visual-Point-and-Click approach as described
above, and the D-T2 map renormalized for the water-oil model.
Examples
Internal Gradient Example
Fig. 4 shows an example of internal gradient effect on NMR
fluid responses. The well was drilled with water-base mud
(WBM) through a series of sand-shale sequences in Asia. The
CPMG pulse sequences used in this job are listed in Table 1.
In another well drilled with WBM through the Marjan
sandstones reservoirs in the Middle East, MRF data were
(8) acquired with the same CPMG pulse sequences shown in
Table 2. Fig. 6 shows an example of gas in high-permeability
sands. The large pore sizes cause the fluids to relax close their
bulk values, resulting in a smeared peak at long T2 as seen in
the left panel. The D-T2 inversion results are normally
integrated over the third dimension (T1/T2) as shown in the
top right panel. However, it is also possible to decompose the
total signal into separate results corresponding to each T1/T2
plane and benefit from T1 information in the fluid
interpretation. For the acquisition sequence used here, there
are three different wait-times. With such limited polarization,
just two elements in T1/T2 space are sufficient to account for
the data. The D-T2 inversion results for T1/T2 = 1 and T1/T2 =
3 are shown in the bottom right panel. It can be seen that the
signal is actually composed of a shorter T2 peak that has a
T1/T2 ~ 1, and a longer T2 peak that has a T1/T2 ~ 3.
4 SPE 84478

Fig. 7 shows the decomposition of the D-T2 map along the Gas Oil OBM Water Total Total Total
T1/T2 dimension. It can be seen that the water-base mud Oil Liquid
filtrate signal has a T1/T2 ~ 1, whereas the gas signal has Visual 13 2.1 2 4.3 8.4 21.4
predominantly a T1/T2 ~ 3. DLM 13 4.5 4.4 8.9 21.9
DMR 12.8 7.9 20.7
Water, Oil, OBM, Gas Condensate Example Table 3 – Comparison of D-T2 maps Visual, DLM and DMR
The Alwyn field in the British sector of the North Sea is results. All volumes are in p.u. The estimated noise from this
now being reappraised for the remaining hydrocarbon data set is less than 1 p.u.
reserves. Hydrocarbon is found in both Jurassic and Triassic
sandstones, but the Triassic reservoirs have historically Medium Oil Example
attracted little attention as the production effort focused on the The following example was acquired in a well drilled
prolific Jurassic sands that lie above. As the larger reservoirs through the Oficina formation in eastern Venezuela. This well
are coming off plateau production, increasing attention is was deviated at 55 deg, and drilled with pure synthetic oil-
being paid to the Triassic where significant reserves remain. base mud, containing no water. The interval of interest is
Productivity of these complex formations is far less known to be a sand of mixed wettability, with strong aquifer
predictable than that of the Jurassic, and new techniques support. MRF stations were selected using conventional
including MRF are being developed to better understand the openhole logs. The objectives were 1) to establish the
production behavior. The well was drilled with oil-base mud repeatability of the measurement, 2) to differentiate between
(OBM), and the same pulse sequences shown in Table 2 were oil and water, and 3) to verify the different types of oil
used to acquire the data. expected in this well. The MRF acquisition sequence was
selected to provide sensitivity both to diffusion and to T1.
Fig. 8 shows the context where the MRF stations were taken. NMR acquisition parameters are listed in Table 4.
For brevity, only the bottom station is reported.
Sequence WT(s) TE(us) NECHO REPT
Fig. 9 shows the quantitative D-T2 map analysis. On the left 1 8 200 5000 1
panel, after delineating and labeling the different fluids 2 3 2000 300 1
following the visual point-and-click approach, the respective 3 3 200 3000 1
4 3 4000 200 1
signal amplitudes are integrated over their respective windows
5 1 200 3000 1
to give the different fluid volumes as seen in Table 3. On the 6 3 6000 100 1
right panel is shown the D-T2 map renormalized after Table 4 – MRF Pulse Sequence used in a Venezuela well.
eliminating the gas following the DLM approach. Note that the
restricted diffusion effect in the very small pores causes DLM Four MRF stations were acquired as shown in Fig. 10.
to plot between the water and oil lines. After correction, the Absence of any long T2 OBM filtrate signal indicates
DLM results show good agreement with the visual point-and- negligible invasion over the oil interval. The D-T2 map at the
click results. lowermost station, taken in a water zone at x526 ft, shows that
this water-bearing sand is clean, very porous, and permeable.
It is interesting to compare the D-T2 map results with the It has very large pores, allowing the water to relax at near its
established Density-Magnetic Resonance (DMR) 6 results. The bulk rate, corresponding to T2 ~ 5 s at formation temperature.
DMR calculation with the following parameters (bulk density The weak signal just below T2 = 100 ms on the D-T2 map is
= 2.25 g/cm3, CMR porosity = 18 p.u., matrix density = 2.65 consistent with traces of residual oil.
g/cm3, liquid properties: density = 1 g/cm3, hydrogen index =
1, gas condensate properties: T1= 3s, hydrogen index = 0.8, Three stations were also acquired over the oil zones, in close
density = 0.55 g/cm3) gives a gas-corrected porosity of 20.7 progression, at x490, x471 and x450 ft. These stations were
p.u., and a gas condensate volume of 12.8 p.u.. These results meant to establish the repeatability of the measurement, as no
are close to the results of the quantitative analyses from the D- significant variation in viscosity was expected in this sand.
T2 map and are summarized in Table 3. Note that, depending Instead, these stations suggest an oil viscosity profile, with
of the methodology used, different breakdowns of the fluids viscosity decreasing vs. depth, from 16 cp at x490 ft, to 11 cp
are obtained. The DMR technique solves for gas condensate at x471 ft, and less than 10 cp at x450 ft. While this result was
and “liquid” that comprises water, OBM, and native oil. The a little surprising, it was also reflected in the general trend of
DLM technique solves for gas condensate, water, and “total oil” increasing T2 seen in the T2 distribution recorded vs. depth
that comprises OBM and native oil. The visual point-and-click over the entire oil-bearing interval. The viscosity estimated by
technique solves for all the different fluids that the eye can the MRF method compares well with PVT results on the same
see. producing interval in nearby wells, which indicated viscosities
ranging from 5.3 to 17.9 cp, with a median value of 11.9 cp.
This confirms that oil-base mud filtrate contamination was
minimal in the oil leg.
SPE 84478 5

As the drilling fluid contained no water, a comparison Dg Gas diffusion constant, cm2/s
between saturations from a conventional petrophysical Do Oil diffusion constant, cm2/s
evaluation and saturations from MRF yields additional Dw Water diffusion constant, cm2/s
information. A significant discrepancy in saturation is FOIL Oil T2 distribution
observed at x490 ft, with the MRF station showing a FWAT Water T2 distribution
measurably higher water percentage. The upper two points P Pressure, psi
show virtually no free water, and are in agreement with the Sw SXO Flushed zone water saturation
calculated from resistivity. We believe that the MRF SW Virgin zone water saturation
measurement is accurate in all three cases, and that it reflects T Absolute temperature, degK
the onset of water movement in the lowermost part of the T1 Longitudinal relaxation time, s
sand, which is known to have a strong waterdrive. The D-T2 T2 Transverse relaxation time, s
map quantitative results at x526 ft, x490 ft, and x450 ft are T2LM Log-mean T2, s
shown in Fig. 11. It shows the presence of water at x490 ft just T1/T2 T1 to T2 ratio
above the OWC. The signal just below T2 = 100ms, which is TE Echo spacing, ms
consistent throughout the series of stations, including the WT Wait-time, s
water zone, may be attributed to heavy-oil components, λ Diffusion to T2 relaxation ratio, cm2/s2
possibly having high asphaltene content. Cores taken there
show oil stains, reflecting the continuous water encroachment References
over time into what were once oil-bearing sands. The result at 1. Slijkerman, W.F.J. et al.: “Processing of Multi-Acquisition
the top station (x450 f) confirms the absence of free water. An NMR Data,”paper SPE 56768 presented at the 1999 SPE
MDT sample was also retrieved at that depth, and found to be Annual Technical Conference and Exhibition, Houston, Texas,
water free. The interpretation is consistent with a mixed wet 3-6 October.
reservoir in which heavy (asphaltene-rich) components in the
oil are bound to the rock grain surface. 2. Freedman, R. et al.: “A New NMR Method of Fluid
Characterization in Reservoir Rocks: Experimental
Confirmation and Simulation Results,”paper SPE 75325
Conclusion published in the SPE Journal, 2001.
Successful NMR fluid characterization logs require a three-
stage process: 3. Hurlimann, M. D. et al.: “Diffusion-Editing: New NMR
Measurement of Saturation and Pore Geometry,” Transactions
1) Planning and acquisition – For high-viscosity oils, we of the 43rd SPWLA Annual Logging Symposium, Oiso, Japan,
recommend increasing the SNR of the long echo spacing June 2-5, 2002, paper FFF.
CPMG sequences by repeating these measurements, or to use
DE sequences. For low-viscosity oils, OBM, and gas, we 4. Freedman, R. et al.: “Wettability, Saturation, and Viscosity
Using the Magnetic Resonance Fluid Characterization Method
recommend multi-wait sequences in addition to diffusion
and New Diffusion-Editing Pulse Sequences”, paper SPE 77397
sequences to capture T1 information. A signal-to-noise ratio of presented at the 2002 SPE Annual Technical Conference and
20 or higher is recommended for robust inversion of multi- Exhibition, San Antonio, Texas, September 29 - October 2.
measurement NMR data.
5. LaTorraca, G.A. et al.: “Magnetic Susceptibility Contrast
2) Processing and interpretation – We recommend Effects on NMR T2 Logging,” Transactions of the 38th SPWLA
supplementing D-T2 maps with model-based fluid inversion Annual Logging Symposium, Paris, France, 1995, paper JJ.
schemes. D-T2 maps are preferably obtained from a 3D
model-independent inversion (D-T2-T1/T2). The maps are 6. Hurlimann, M.D. et al.: ”Restricted Diffusion in Sedimentary
Rocks, Determination of Surface-area-to-volume Ratio and
used to guide model-based inversion in cases where the fluids
Surface Relaxivity,” Journal of Magnetic Resonance, Series A
deviate from existing correlations, or when the assumptions (1994) 111, 169-178.
about diffusion regime and wettability are not valid.
Alternatively, D-T2 maps can be used directly to give 7. Freedman, R. et al.: “Combining NMR and Density Logs for
qualitative and quantitative fluid interpretations. Petrophysical Analysis in Gas-Bearing Formations,”
Transactions of the 41st SPWLA Annual Logging Symposium,
3) Validation and crosschecking – As with any analysis, it is Keystone, Colorodo, 1998, paper II.
advisable to crosscheck NMR fluid interpretation results with
measurements from other tools such as resistivity, density, 8. Cao Minh, C. et al.: “Integration of NMR with Other Openhole
Logs for Improved Formation Evaluation,” paper SPE 49012
dielectric tools, etc. In particular, NMR-derived oil viscosity
published in SPE Formation Eval. & Eng. 3(6), December 2000.
may require laboratory calibration to account for possible 9. Jaynes, E.T.: “Information Theory and Statistical Mechanics”,
deviations. Once validated, NMR-derived Sxo might be used to Phys. Rev. (1957) 106, 620.
derive other petrophysical answers in conjunction with other
logs. 8 10. Catalano, D. et al.: “A Maximum-Entropy Analysis of the
Problem of the Rotametric Distribution for Substituted
Nomenclature Biphenyls Studied by 1H Nuclear Magnetic Resonance
D Diffusion constant, cm2/s Spectroscopy in Nematic Liquid Crystals”, J. Chem. Phys.,
DLM Log-mean diffusion constant, cm2/s (1991) 94, 3928.
6 SPE 84478

Acknowledgements where S is a diagonal matrix whose elements are the singular

We are grateful to the operators for releasing the data used in values of H. The columns of U and V are known as the left and
this paper. We are also indebted to the anonymous reviewers right singular vectors, respectively, and are orthonormal.
in many organizations who have provided us with constructive
remarks. To express H as a two-dimensional matrix, we define a
general measurement coordinate, τ, (eg. τ = WT, TE, NE) and
Appendix – Model-Independent Multidimensional a distribution coordinate, x, (eg. x = T2, D, R12). From Eq.
Inversion (A.2), we can now write
Given the limited quantity and signal-to-noise ratios typical of
downhole NMR data, multidimensional distributions of
relaxation times and diffusion rates are grossly under-
determined. The problem of inversion is then one of deriving a
∑U
τ
T
n (τ ) H ( τ,x ) = sn VnT ( x ) = v n ( x ) (A.3)

distribution with no extraneous information content and that

simultaneously accounts for the measured data. Many
where vn are un-normalized orthogonal vectors expressed in the
inversion schemes use regularization to smooth the
space (x) of the distribution function. Combining Eqs. A.1 and
distributions and reduce noise artifacts. The algorithm
A.3 leads to the useful expression
presented here adopts a different approach, based on the
maximum entropy principle (MEP). 9, 10

The response, A, from a three-dimensional (3D) measurement ∑U

τ
T
n (τ ) A( τ ) = ∑ v ( x ) f( x )
x
n + δn
can be expressed as
= vn + δ n
(A.4)
A( τ 1 ,τ 2 ,τ 3 ) = ∑∑∑
i j k
f ( i, j,k )H( τ 1 ,τ 2 ,τ 3 ;i, j,k )
Since the UTn(x) are available from SVD analysis of H, and the
+δ ( τ 1 ,τ 2 ,τ 3 ) A(τ) are just the measured echo amplitudes, it is straightforward
(A.1) to compute the distribution moments, v n .
where τ1, τ2, τ3 are three acquisition parameters which define a
particular measurement, f(i,j,k) is the amplitude of the From information theory 10, the least biased distribution, f(x), is
component i, j, k in a 3D distribution (i.e., each dimension given by
corresponds to a particular property); H(τ1, τ2, τ3; i, ,j, k) is the
 
response of that component to the measurement specified by
parameters τ1, τ2, τ3;, and δ is a random noise term. To place Eq.
f ( x ) = Z -1 exp 


∑α v ( x ) 
n
n n (A.5)

A.1 in context, τ1, τ2, and τ3 could, for example, be wait-time

(WT), interecho spacing (TE) and echo number (NE) in a CPMG
measurement. Similarly, index i might refer to a particular T2
value, T2(i); j may refer to a diffusion rate, D(j); and k might
correspond to a particular T1 value or T1/T2 ratio, R12(k).
Solution of Eq. A.1 consists of determining f(i,j,k) given a series
of measurements, A(τ1, τ2, τ3), assuming that the form of H(τ1,
τ2, τ3; i, ,j, k) is known. The problem is grossly under-
determined because the response kernels H(τ1, τ2, τ3; i, ,j, k) are
substantially linearly dependent.
where αn, are adjustable parameters which are chosen such that
the measured moments, v n , are reproduced, and Z is a
normalization constant. The inversion is optimized in the sense
that the number of variable parameters is identical to the number
of data points to be fit (i.e., the moments, v n ). Unlike
conventional inversion algorithms commonly used for deriving
T2 distributions, this scheme does not require any explicit
regularization. Instead, by using equation A.5, it ensures that the
resulting distribution has maximum entropy,

The first step in applying MEP is to identify a suitable set of S = −k ∑ ln( f ( x )) f ( x ) .

i
i i (A.6)
basis functions defined in the space of the distribution (eg. in T2
– D – T1/T2 space), whose mean values can be measured. One
such set of functions can be obtained by Singular Value SI Metric Conversion Factors
ft X 3.048 E-01 = m
Decomposition (SVD) of the measurement kernel, H(τ1, τ2, τ3; i,
psi X 6.894757 E+03 = Pascal
,j, k). The SVD method decomposes the kernel matrix, H, into
the product of three separate matrices
H = USV T (A.2)
SPE 84478 7

Fig. 1 Diffusion (left panel), T1 (center panel) and T2 (right panel) of water, oil, and gas as a function of temperature.

Fig. 2 NMR spin echo data are presented as amplitudes versus diffusion (D) and relaxation (T2) as shown in the left panel
(perspective view). The right panel provides a more practical representation of D-T2 map. Theoretical responses of water, oil, and gas
are overlaid on the maps to help the interpretation. D-T2 map is analogous to a density-neutron crossplot. The crossplot itself is
model-independent, but in order to interpret the data, an overlay of theoretical responses of sandstone, limestone, dolomite, shale, and
gas is necessary.
8 SPE 84478

Internal Gradient
High GOR

Mixed
Wettability
Restricted
Diffusion

Fig. 3 Some exceptions to ideal models: 1) Internal gradient pulls data in the up-arrow direction (such that water could be mistaken
for gas), 2) Restricted diffusion pulls data in the other direction as indicated by the down arrow (such that water could be mistaken for
oil), 3) Mixed wettability pulls the oil data to the left (such that the oil volume/saturation are computed too low), and 4) high GOR
cause oils to shift along the North-East trend (such that oils could be mistaken for water or gas).

Fig. 4 Example of internal gradient effect on NMR fluid response. The internal gradient adds to the tool gradient and hence, diffusion
is higher than expected. The sample is a clay-rich sandstone in the water-bearing zone of a well drilled with water-base mud in Asia.

Unrestricted
diffusion

Restricted
diffusion

Fig. 5 Example of restricted diffusion in Middle East carbonates. The free water undergoes unrestricted diffusion in large pores and
agrees with the theoretical response. The bound water is trapped in smaller pores and therefore experiences restricted diffusion.
SPE 84478 9

Fig. 6 Example of gas in high-permeability sandstone reservoirs in the Middle East. The large pore sizes cause the water to relax
close to its bulk value overlapping the gas signal, and resulting in a smeared peak at long T2.

Fig. 7 Decomposition of D-T2 map along the T1/T2 dimension. It can be seen that the water-base mud filtrate signal has a T1/T2 ~ 1
(right panels), whereas the gas signal has predominantly a T1/T2 ~3 (left panels). The dashed T2 distribution is the total T2
distribution shown as reference.
10 SPE 84478

Fig. 8 North Sea OBM gas condensate example. -Two MRF stations were acquired in the Alwyn North N46 at the points shown. The
objective was to help determine if both sands contained similar hydrocarbon types, as the poorly communicating Triassic sands have
sometimes displayed slightly varying hydrocarbon densities across the reservoir.

Fig. 9 North Sea example. Quantitative D-T2 map analyses from two methods. On the left panel, after delineating and labeling the
different fluid types using the visual point-and-click approach, the respective signal amplitudes are integrated over their respective
windows to give the different fluid volumes. On the right panel is the D-T2 map renormalized after eliminating the gas using the DLM
approach. Note that restricted diffusion effect in the very small pores causes DClm of the bound water to plot between the water and
oil lines.
SPE 84478 11

Fig. 10 Venezuela example. Four MRF stations were taken at depths shown by the yellow circles. Three D-T2 maps are shown on the
right. They correspond to the bottom station in the water zone, the middle station in the oil zone near the OWC, and the top station
near the top of the oil zone.
12 SPE 84478

Fig. 11 Venezuela D-T2 map quantitative results from the DLM method. The oil distributions in green show a consistent asphaltene
signal just below T2 = 100 ms in all stations, including the bottom station (x526 ft) in the water zone. It is visible from the two stations
in the oil zone (x490 ft and x450 ft) that the asphaltene content in the oil decreases higher up in the reservoir (decreasing amplitude
ratio of the short T2 vs. long T2 components of the oil distribution) that leads to varying oil viscosity in the reservoir. The x490 ft
station near the OWC also shows water encroachment (small water peak at T2 ~ 200 ms).