Why high-resolution geology gives the best predictions of CO storage dynamics

Classification: Internal
Status: Draft
Ice margin processes, Greenland
Tidal Delta Sedimentary Architecture
Niell Klinter Formation
Why high-resolution geology
gives the best predictions of
CO2 storage dynamics
Philip Ringrose
StatoilHydro, Trondheim, Norway
Svalbard Workshop: Modeling and risk assessment of geological storage of CO2
August 3 - 7, 2009
2
Outline
1. Brief history of the abuse of geology and mathematics in
the oil industry
2. Examples of impact of geology and flow dynamics on CO2
plume development at the In Salah Storage site
3. Some propositions
“Okay, okay, pussy cat – but I can just
add a fudge factor to make it right.”
3
Numerical Flow Simulation Errors
• Gridding Errors: Fine-scale rock variables are discretized into a
“manageable” grid leading to smearing of the true permeability field.
• Numerical Dispersion of the saturation-front occurs due to finitedifference truncations of derivatives (temporal and spatial).
Well data
Mudstone Heterolith
Sandstone
k
Upscaled logs
k
Unfortunately, most
modellers ignore and fail
to minimize these errors
Modelled Flow Front
Upscaled Reservoir model
4
Dry gas
injection into
wet gas /
condensate
2500
Fine Grid
Coarse Grid (no upscaling)
Coarse Grid (one-step upscaling) 2000
Coarse Grid (two-step upscaling)
0.8
0.6
1500
0.4
1000
0.2
500
0
0
Gridding
0.2
0
Volume Gas Injected (fraction of Free GIP)
Upscaling
Full-field
coarse grid has
4 vertical cells
0.4
Gas Oil Ratio (Sm3/Sm3)
Smørbukk Gas
Injection Example
Oil Rate
(fraction of target rate)
1.0
Sector model
fine grid has 25
vertical cells
Pickup et al 2000, SPE 62881
Sector model is
c. 6km long and 20m thick
5
Nice 3D graphic version
6
Scaling Group Theory
u x Δx μo
Viscous
=
Capillary k x (dPc / dS)
Gravity Δρ g Δz
=
Capillary (dPc / dS)
Darcy’s Law
e.g. for typical reservoir
properties and conditions:
= 10-12 x 0.1 x 1 = 10-1
10-14 x 100
System dimension
(grid size)
Pore-throat contrast in
0.1m-spaced layers with
kx = 10mD and viscous
gradient of 1kPa/m
Capillary Pressure
gradient
• Accurate upscaling of multi-phase flow processes requires correct
assessment and scaling of the balance of fluid forces
• Most reservoir flow processes are capillary and gravity dominated
– The main exception is near-wellbore flow which is viscous dominated
• However, capillary forces are commonly neglected (and “packaged”
implicitly in the relative permeability function)
7
Some things are not what they appear to be …
Meeting a friend in a corridor, Ludvig Wittgenstein said:
“Tell me, why do people always say that it was natural for
men to assume that the sun went around the earth rather
than that the earth was rotating?”
His friend said, “Well, obviously, because it just looks as if
the sun is going around the earth.”
To which the philosopher replied, “Well, what would it
have looked like if it had looked as if the earth was
rotating?”
from Jumpers, a play by Tom Stoppard.
And so it is with viscous-dominated, numerically-dispersed,
finite-difference models of CO2, gas or oil displacement into
water-saturated media – they look right for the wrong reasons.
8
Upscaling (the right way)
Rustad et al 2008 (SPE 113005)
2. Calculate multiphase
flow functions (CL)
1. Identify lamina and pore
types from core
Rock
type 2
Pore to Field Workflow, Statfjord
Rock
type 1
3. Apply functions in
lithofacies models
4. Upscale lithofacies
models (VL and VE)
Upscaled curves
1,0E+00
Krgx VE
Krgy VL
Krogx VE
Krogy VL
1,0E-02
5. Apply to full field
(significantly improved
history match)
1,0E-04
1,0E-06
0
0,25
0,5
0,75
1
Sg
Key factor was anisotropic
relative permeability
9
Upscaling in Practice
• The Representative Elementary Volume (REV) concept should give the basis for
model design
• Models should fit the geology (not visa versa)
Scales of measurement
Permeability (md)
Thin section & SEM
1000
Probe Core
Perm. plugs
Pore-scale
model
Seismic data
& well tests
Logging
tools
Lithofacies
model
Geomodel
Lithofacies 1
100
Pore type 1
Stratigraphic
REV
Lithofacies
REV
10
Lamina
REV
1
Lithofacies 2
Pore type 2
0.0001
0.001
0.01
0.1
Lengthscale [m]
1
10
Nordahl & Ringrose, 2007
10
Example: Critical Gas Saturation
• Statfjord Depressurisation study (Lerdahl et al, 2005, SPE 94191)
– Somewhat analogous to CO2 brine systems (?)
• Estimation of effective critical gas saturation for lithofacies-scale model
• Sgcrit is anisotropic
and scale-dependent
Gas flow velocity in
Rannoch thin bed model
(c. 10m by 1m)
11
Propositions (1)
• Our simulations are plagued by gridding and numerical dispersion errors
– “A plague in both your houses” (numerical and geological)
• We historically compensate for this by adjusting the permeability
functions (krel endpoints and exponents, permeability multipliers, etc.)
– “To make it look like the sun goes around the earth”
• Systematic upscaling and rescaling of flow dynamics and rock
architecture is possible and practical
– Anisotropic multi-phase flow functions are probably the most critical
factor in sedimentary systems
– Fracture/fault matrix interactions are probably the most critical factor
in fractured rock systems
12
Examples from the In Salah CO2 Project
Key Field Statistics:
• Five gas producers
Kb-503
• Three CO2 injectors
• Initial reservoir conditions:
Kb-502
Kb-14
• P= 175bars
po
ro
si
ty
m
ap
Kb-11
Se
i
sm
ic
Kb-12
Kb-15
Kb-501
Kb-13
• T = 95oC
• Reservoir is 1880m deep,
20m thick and 20 x 8km2 in
area
• 3 MT CO2 have been injected
(2004 to March 2009)
13
Full-field simulation
• Eclipse 300 simulator used
KB-503
KB-5
KB-502
for full-field gas production
management, well planning
and CO2 injection
• Coarse grid (400x6m cells)
KB-501
adequate for field
management but insufficient
for CO2 forecasting
• No faults included, apart from
permeability corridor needed
to history match KB-502 / KB5 observations
• Suffers from aforementioned
grid dispersion errors
Gas saturation at 1st September 2008
(Courtesy of In Salah Joint Gas Venture team)
14
The KB-502 Story
(First Break, January 2009)
Well observations:
• CO2 from injector KB-502
detected at appraisal well
KB-5 after 2 years injection
(Distance 1.3km)
Zone of surface
deformation
Conductive
fracture
orientation
Observation
well KB-5
Possible CO2
plume
extension
Injection well
KB-502
Historymatched CO2
plume
Reservoir history match (Eclipse):
• 4D permeability corridor
(suspected fault?)
• Fracture flow also important
C
GW
Satellite Data:
• Deformation ellipse detected
around injector
• Larger area than the
breakthrough zone
15
Long-term simulations of CO2 migration
MPath
forecast
• Using
the for
Permedia MPath simulator: Invasion percolation model of
injection
from well
502
gas phase
into
brine-filled pore space controlled by capillary entry
pressure
Capillary-dominated
case (end-member?)
• Based on best estimate fluid and rock properties but neglects multi-
CO2phase
migration
afterand geochemical reactions
mixing
100 years
Colours indicate invasion
time sequence
16
Short-term snapshot of MPath IP model
•
Detection of CO2 at
observation well after
2 years injection from
well 502 gives a
calibration point for the
longer-term forecasts.
•
Preliminary model gives
plausible match (after 2
years)
•
Many other factors
affecting CO2 plume
(multi-phase mixing)
•
Matrix-flow model with
anisotropy but no
explicit fracture flow
2008 unfaulted reservoir model
17
Faults and
Fractures
Subtle seismic
features that are not
seismic artefacts
Modelled Faults:
• No dip information
• Based on seismic
• Further work done
to map and quantify
faults and fractures
(continuing)
• Subtle faults best
detected on low-angle
illumination of time
surface
• Reservoir model faults
digitized using Gocad
• 2009 faulted model
18
Permedia/MPath study using hi-res fault model
• Ongoing study by Permedia Group (Andrew Cavanagh)
• Preliminary result from high resolution mesh (400M cells)
• Next stage aims to compare black-oil simulation with IP approach
Topographic control
Fault control
Fault & fracture control
19
Fracture Modelling
Fault
Fracture
• Discrete Fracture Network (DFN)
models used for fracture flow modelling
(Martin Iding, Fraca+)
• Different cases to capture range of
uncertainty
• Input data from FMI, core, and drilling
data
kx
From core to reservoir model
Effective fracture
permeability
~100-400md
Iding & Ringrose, 2008 (GHGT-9)
20
Faults and Stress Field
KB-502
Structural analysis shows that:
• Fault 12 could be open under the
current-day stress field.
• Main fracture set is also closely
aligned with current-day field
• Other fault and fracture sets more
likely to be closed/cemented/old
Fault 12
n
ai s
M ture
c
fra
Fault 12
All digitised fault data
KB502 fractures
North
σ1
σ1
σ1
East
West
σ1
σ1
South
nd set
o
c
Se ture
ac
Fr
σ1
21
Propositions (2)
• Conventional simulation model (E300) did not predict CO2 breakthough. Was this because of:
– Gridding artefacts
Yes to all three
– Failure to include faults and fractures
– Inappropriate physics (viscous/gravity/capillary ratios)
• Capillary-dominated model (MPath) gives plausible picture of CO2
distribution and forecasts
– But is it credible (simplified physics)?
Not yet convinced the
reservoir engineers
• Current reservoir geological insights include:
– Presence of a fracture corridor around KB-502 (1 to 4 Darcy)
– Diffuse facture flow (fracture-k is c.400md cf c.10md in matrix)
– Fault orientation wrt stress field is important (closed and open faults)
– Effects of sedimentary bedding and channels also important (currently
only represented in the seismic inversion porosity model)
Gives us quite significant reservoir modelling challenges
22
In Salah Satellite Data
PSInSAR™ has been used to record surface deformation related to
subsurface injection (and production):
•
Pioneering work by TRE and LBNL (Vasco et al. 2008)
First TRE dataset (2003-2007)
revealed ~5mm uplift over the
CO2 injectors
Ongoing monthly satellite
surveys and surface
calibration using tiltmeters
KB-5
KB-503
KB-502
KB-501
s
Ga
l
fie
do
utl
ine
0
10 km
-5
0
mm/yr
5
Tiltmeter
Pinnacle Technology
23
Surface Deformation Data
Krechba satellite image
(PSInSAR):
• May 2008
• Pinnacle Technologies
24
Geomechanical Modelling
• Reference case 2D geomechanical Model (Gemmer et al., Trondheim CCS, 2009)
• Shows that surface deformation
Modelled Vertical Strain (m)
is consistent with linear elastic
response to injection pressure
• Ongoing work in many R&D
groups to understand this
process better
Vertical displacement,
Uz (m)
Vertical displacements at surface
0.015
0.010
0.005
0.000
-0.005
-0.010
0
5000
10000
15000
20000
25000
30000
Distance (m )
Line through observed deformation from satellite observations
35000
40000
45000
25
R&D Highlight – Coupled Modelling
Research results from Lawrence Livermore (LLNL) group
Morris, J. et al., 8th Annual CCS Conf. Pittsburgh, May 2009
Sector model – with mechanically reactive faults (LDEC)
Injector
KB-502
Credible surface deformation
with confined fault
compartment scenario
26
Propositions (3)
• Satellite InSAR data at In Salah gives us a unique insight into rock
mechanical responses to CO2 injection
• First order deformation patterns/magnitudes can be explained by simple
linear elastic response to injection pressure
• BUT details of surface response to sub-surface pressure requires higherresolution coupled models (ongoing work at LBNL and LLNL):
– Mechanical layering is clearly important (Don Vasco et al.)
– Coupled hi-res geo-mechanical flow modelling important (Joe Morris et al.)
• Region affected by elevated injection pressures is clearly different from
and larger than CO2 plume
27
Conclusions
• Upscaling is a fact of life – you can’t ignore it
• You need detailed geological models
• Last word to Albert Einstein:
– Two things are infinite:
the universe and human stupidity;
and I'm not sure about the universe.