property of interest - discovery group to shale gas log... · 2012-01-04 · figure out or guess...
TRANSCRIPT
Property of interest Core data Log data
Porosity Crushed dry rock He porosimetry
Density (mostly)
TOC LECO or RockEval GR, density, resistivity
Water saturation As-received retort or Dean-Stark
Resistivity + kerogen corrected porosity
Mineralogy XRD, FTIR, XRF Density, neutron, Pe, ECS-type logs
Permeability Pulse decay on crushed rock
This is tough…………
Geomechanics Static moduli DTC, DTS, RHOB, & synthetic substitutes
Geochemistry Ro, S1-S2-S3, etc. Resistivity (sort of…)
Global stochastic Global deterministic Local deterministic
› aka direct calibration to core Each has distinct advantages &
disadvantages Each has its staunch defenders &
proponents None is clearly “the best” for all shales Industry has not settled on one approach or
best practices
Global/stochastic › Figure out or guess what components are
present in the shale (mineral, organic, fluid) › Characterize end member properties of
each › Invert the logs for properties that best fit the
observed log character › Implementations include ELAN, Multimin,
Statmin, etc.
Guess at mineral and fluid volumes
Compute log responses from a forward model
Compare computed log responses with actual log data
Satisfactory match?
DONE
Yes
No
Solutions are very Sensitive to the “total error” bars
Actual log curves vs. forward models – convergence is the criteria for a satisfactory model.
How do you know, or guess at, the end member points? › What are the kerogen endpoints (GR, RhoB,
Nphi, etc.)? › What is the inorganic grain density of the
clay components? › What are the error functions associated with
those end points AND the log data (that is, the total error bar)?
Global/deterministic › Characterize properties using broad
collections of shale samples, either globally or basin/play specific
› Schmoker 1979, 1981 equations › deltaLogR (Passey, 1990) › Schlumberger SpectroLith is a service
company specific method
TOC (v/v) = (ρgray sh – ρb) / 1.378 (1979 eqn) TOC (v/v) = (GRgray sh – GR)/(1.378 * A) (1981 eqn) TOC (v/v) = WTOC * RHOma/RHOTOC
0
2
4
6
8
10
12
14
16
18
2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9
Bulk density (g/cm3)
tota
l org
anic
car
bon
(wt %
)
New Albany Shale, Illinois basin EGSP cores (1976-1979), all blue logs
Published Schmoker relation
Basically a sonic F overlay Two universal eqn’s published by Passey
et al. (1990, AAPG Bull) › ∆logR = log10 (R/Rbl) + 0.02 (∆T-∆Tbl) This defines deltaLogR
› TOC = ∆logR * 10^(2.297 – 0.1688 LOM) This relates deltaLogR to TOC (wt%)
Most of us are concerned about the LOM parameter, but the second two constants were empirically determined
Passey et al, 2010, SPE 131350
TOC (wt %) = ∆logR * 10^(2.297 – 0.1688*LOM) Passey et al, 1990, AAPG 74 (12) 1777-1794
Deterministic suite of regression eqn’s to compute mineral volumes and kerogen-free grain density
RhoMecs = a + b Si + c (Ca,Na) + d (Fe,Al )+ e S
Regional or global equation does not apply locally & you don’t know that
Variable selected does not really correlate very well with property of interest › e.g. gamma ray, because Vuranium is not a
simple function of TOC Co-linearities
› e.g. using density to predict both porosity and TOC
Local/deterministic › Calibrate our log models in restricted areas,
down to individual wells or single cores › Simple linear & non-linear regressions GR vs. TOC, RHOB vs. TOC
› Multiple linear regression methods › Multiple non-linear regression A common implementation are neural
networks
( * / 1)(1 / )
m b TOC m TOC TOCT
m fl TOC fl m TOC
W WW
ρ ρ ρ ρφρ ρ ρ ρ ρ− − +
=− + −
But, we don’t directly measure the inorganic grain density, nor the kerogen density.
Have to solve for these two properties by comparing model prediction to core measured TOC’s and porosity
SPE 131768
(1 ) (1 )b hc T wT w T wT m T TOC TOC TOCS S V Vρ ρ φ ρ φ ρ φ ρ= − + + − − +
0
2
4
6
8
10
12
14
16
18
2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9
Bulk density (g/cm3)
tota
l org
anic
car
bon
(wt %
)
New Albany Shale, Illinois basin EGSP cores (1976-1979), all blue logs
0% porosity, RhoKer = 1.00, RhoMa inorganic = 2.72 line
Scatter results from variation in porosity & RhoMa from sample to sample
TOC from GR model
Not so different than stochastic method, except it minimizes error in POROSITY only
Colorado Niobrara example (too few core points!)
∆logR = log10 (R/Rbl) + 0.02 (∆T-∆Tbl) TOC = ∆logR * 10^(2.002 – 0.1749 LOM), for LOM = 9
Merge in the core gamma scan
Depth shift, honor the physical breaks
Don’t interpolate Convert core depths
to log depth, then import
Keep an audit trail! › L = C + x
Kill the outliers
Marcellus Sh., Vclay vs. XRD clay model
Everybody thinks they are right and have the best method….. › Of course, we’re all selling our services too!
Each method has its advantages in certain settings or types of wells › Rich suite of modern logs vs. old, public
domain log coverage? › Vertical vs. horizontal well bores? › Do you have core data for calibration?
1. Transparency › Some methods are basically black boxes &
are difficult or impossible to reproduce › Others are totally transparent, you can write
out a set of equations and you may agree or not, but there they are And if you get more information, they are easy
to tweak and re-apply to prior wells
2. Transportability › Some solutions require proprietary software,
or only one vendor can really run it. › Others require a very specific logging
measurement If you ran the “wrong color” logging company,
or didn’t run the right tool: you’re done. If logging conditions prevent running a
particular service: you’re done. If it’s not your well and you don’t have access
to that data: you’re done.
3. Cheap and easy! › This is self-explanatory, everyone likes cheap
and easy. › Some solutions are powerful and accurate,
but may not be cheap (when you consider everything required to implement) .
› Some solutions are difficult. They require a specialist and might be totally non-reproducible – even by the same person! Try a double blind test sometime and see what
you get back…………….
4. Fit for purpose accuracy › Getting water saturation to 3 decimal places
is probably not realistic – so why do we do it? › Sometimes close enough is good enough Log analysis on “the hood of a Chevy” vs. log
inversions on a supercomputer › What will impact your business decisions,
and how different does an answer have to be to make you do something differently?
May your acreage map be yellow, your balance sheet black, And your 2012 outlook rosy! Happy holidays from all of us at The Discovery Group.