Parameter uncertainty in geological formations, and its impact on CO - - PowerPoint PPT Presentation
TCCS-9, 13-14 June 2017, Trondheim, Norway Parameter uncertainty in geological formations, and its impact on CO 2 storage capacity estimation 13 June, 2017 Rebecca Allen*, Halvor M. Nilsen, Odd Andersen, Knut-Andreas Lie, Olav Myner SINTEF
TCCS-9, 13-14 June 2017, Trondheim, Norway
13 June, 2017 Rebecca Allen*, Halvor M. Nilsen, Odd Andersen, Knut-Andreas Lie, Olav Møyner SINTEF Digital, Norway
“HOW MUCH?”
[1] Halland et al. (2014) CO2 Storage Atlas: Norwegian Continental Shelf, Norwegian Petroleum Directorate. 2 / 23
Static estimates Dynamic estimate
◮ based on pore volume ◮ M = Vpρco2Seff ◮ Seff based on aquifer
boundaries
◮ based on structural traps ◮ M =
Vtrapφρco2 (1 − Srw)
◮ simulate injection and
migration periods
◮ quantify long-term
leakage In this work, we consider structural trapping and long-term plume migration.
3 / 23
“HOW MUCH?”
[1] Halland et al. (2014) CO2 Storage Atlas: Norwegian Continental Shelf, Norwegian Petroleum Directorate. 4 / 23
“HOW MUCH?”
[1] Halland et al. (2014) CO2 Storage Atlas: Norwegian Continental Shelf, Norwegian Petroleum Directorate.
d i s s
u t i
r a t e t e m p e r a t u r e p r e s s u r e d e n s i t y s a l i n i t y porosity permeability anisotropy caprock shape (macro) caprock rugosity r
k c
p r e s s i b i l i t y f a u l t i n g Many uncertain parameters!
4 / 23
“HOW MUCH?”
[1] Halland et al. (2014) CO2 Storage Atlas: Norwegian Continental Shelf, Norwegian Petroleum Directorate.
d i s s
u t i
r a t e t e m p e r a t u r e p r e s s u r e d e n s i t y s a l i n i t y porosity permeability anisotropy caprock shape (macro) caprock rugosity r
k c
p r e s s i b i l i t y f a u l t i n g ...ones considered in this work
4 / 23
For example...
Impact of some parameters on plume migration
Base case upslope injection point Increased permeability Increased porosity Increased caprock tilt
5 / 23
5 / 23
Representing uncertainty: We generate a Gaussian-type field, and apply it to the (spatially correlated) parameters one-at-a-time. For example... Perturbations applied to top surface elevations of Sandnes geomodel
Base grid Perturbed grid Gaussian-type perturbation (+/- 15 meters) 6 / 23
Model response: We then (i) compute the structural trapping capacity (static estimate) or (ii) simulate long-term plume migration and leakage (dynamic estimate). For example... Model response to perturbations applied to top surface elevations of Sandnes geomodel
500 realizations standard deviation = 50 Mt
7700 7750 7800 7850 7900 7950 Structural Trapping Capacity (Mt) 50 100 150 200 250 300 350 400 450 500 Realization i P90 P10 P50 Base Case
7 / 23
Model response to perturbed caprock (Utsira geomodel)
0 years 100 years 1000 years 5000 years 15 000 years
Initial plume
8 / 23
Model response to perturbed caprock (Utsira geomodel)
0 years 100 years 1000 years 5000 years 15 000 years
Initial plume Plume mis-match quantified using Sørensen-Dice coefficient
Area A Area B Area C
2C A+B
8 / 23
Model response to perturbed caprock (Utsira geomodel)
0 years 100 years 1000 years 5000 years 15 000 years
Initial plume
S = 1
0.7 0.8 0.9 1 5 10 15 20 25 30 35 Frequency 0.2 0.4 0.6 0.8 1 Probability
µ = 0.96 σ = 0.007
S
0.7 0.8 0.9 1 5 10 15 20 Frequency 0.2 0.4 0.6 0.8 1 Probability
µ = 0.93 σ = 0.014
S
0.7 0.8 0.9 1 2 4 6 8 10 Frequency 0.2 0.4 0.6 0.8 1 Probability
µ = 0.85 σ = 0.031
S
0.7 0.8 0.9 1 5 10 15 Frequency 0.2 0.4 0.6 0.8 1 Probability
µ = 0.83 σ = 0.036
S
8 / 23
Model response to perturbed caprock (Utsira geomodel)
0 years 100 years 1000 years 5000 years 15 000 years
Initial plume
5000 10000 15000 Year 0.75 0.8 0.85 0.9 0.95 1 Sørensen-Dice coefficient
µ µ ± σ
8 / 23
Challenge: When dealing with uncertain parameters, we might have to consider 1000s of geomodel realizations... thus evaluating the model response can be computationally demanding. In our work...
◮ Static capacity estimates are relatively cheap (traps are identified by
spill-point analysis, a topography analysis algorithm).
◮ With regards to long-term simulations, Vertical-Equilibrium (VE)
modelling and migration/leakage forecasting reduces our CPU time.
9 / 23
Spill-point analysis: identifies traps, spill points, spill paths, and
spill path trap 1 catchment 1 Out of domain
10 / 23
Spill-point analysis: identifies traps, spill points, spill paths, and
Year 10
10 / 23
Spill-point analysis: identifies traps, spill points, spill paths, and
Year 210
10 / 23
Segregated, migrating plume
z x y g θ CO2 + residual brine brine + residual CO2
H h hmax ◮ From top to bottom: free CO2, residual CO2, brine ◮ Vertical equilibrium: no significant flow in vertical direction ◮ Upscale 3D flow equations onto 2D domain ◮ Reduce CPU time without reducing quality of result
Nilsen, H.M., Lie, KA. & Andersen, O. Comput Geosci (2016) 20: 93. doi:10.1007/s10596-015-9549-9 11 / 23
11 / 23
Utsira-South: model response (plume migration) to perturbations
Base grid Perturbed k ([0.15, 4.96] darcy) Perturbed φ (±0.05) Perturbed zt (±5 meters)
Base grid injection point permeability 0.8 0.9 1 30 2800
Years SDC
base plume
perturbed plume
porosity 0.8 0.9 1 30 2800
Years SDC
caprock elevation 0.8 0.9 1 30 2800
Years SDC
12 / 23
Utsira-South: model response (plume migration) to perturbations
Base grid Perturbed k ([0.15, 4.96] darcy) Perturbed φ (±0.05) Perturbed zt (±5 meters)
Base grid injection point permeability 0.8 0.9 1 30 2800
Years SDC
base plume
perturbed plume
porosity 0.8 0.9 1 30 2800
Years SDC
caprock elevation 0.8 0.9 1 30 2800
Years SDC
Model response is likely sensitive to magnitude of the perturbations...
12 / 23
Sensitivity of model response to magnitude of perturbations
Perturbation level & response Parameter Low Medium High zt (meters) ± 1 ± 5 ± 15
0.8 0.9 1 0.8 0.9 1 0.8 0.9 1
φ (unitless) ± 0.02 ± 0.05 ± 0.10
0.8 0.9 1 0.8 0.9 1 0.8 0.9 1
k (darcy) [0.32, 2.18] [0.15, 4.96] [0.03, 25.67]
0.8 0.9 1 0.8 0.9 1 0.8 0.9 1
Years SDC
13 / 23
The benefit of forecasting migration is we reduce number of simulation years required to capture long-term leakage.
0 years 100 years 1000 years 5000 years 15 000 years
Initial plume
5000 10000 15000 Years since simulation start 100 200 300 400 500 600 Mass (MT)
Structural residual Residual Residual in plume Structural plume Free plume Exited Trapping forecast
Forecast to remain Forecast converged
14 / 23
Perturbation levels versus simulated trapping
1000 2000 3000 Years 200 400 600 Forecast to remain (Mt)
+/- 1 meter
1000 2000 3000 Years 200 400 600 Forecast to remain (Mt)
+/- 5 meter
1000 2000 3000 Years 200 400 600 Forecast to remain (Mt)
+/- 15 meter
1000 2000 3000 Years 200 400 600 Forecast to remain (Mt)
+/- 30 meter
± 1 meter ± 5 meters ± 15 meters ± 30 meters
15 / 23
Perturbation levels versus simulated trapping
1000 2000 3000 Years 200 400 600 Forecast to remain (Mt)
+/- 1 meter
1000 2000 3000 Years 200 400 600 Forecast to remain (Mt)
+/- 5 meter
1000 2000 3000 Years 200 400 600 Forecast to remain (Mt)
+/- 15 meter
1000 2000 3000 Years 200 400 600 Forecast to remain (Mt)
+/- 30 meter
± 1 meter ± 5 meters ± 15 meters ± 30 meters More trapping when larger zt perturbations applied...
15 / 23
Sandnes and Utsira: two different geomodels...different model response
5 10 15 20 25 30 Perturbation (+/-, meters) 800 850 900 950 1000 1050 1100 1150 1200 CO2 (Mt) Base capacity
(b) Utsira
5 10 15 20 25 30 Perturbation (+/-, meters) 9800 9850 9900 9950 10000 CO2 (Mt) Base capacity
(a) Sandnes
Base (161 traps) Perturbed Mean (165 traps)
16 / 23
Sandnes and Utsira: two different geomodels...different model response
5 10 15 20 25 30 Perturbation (+/-, meters) 800 850 900 950 1000 1050 1100 1150 1200 CO2 (Mt) Base capacity
(b) Utsira
5 10 15 20 25 30 Perturbation (+/-, meters) 9800 9850 9900 9950 10000 CO2 (Mt) Base capacity
(a) Sandnes
Base (51 traps) Perturbed Mean (86 traps)
16 / 23
16 / 23
An initial caprock pressure P is computed assuming hydrostatic conditions, with a possible pressure deviation, i.e., P = (ρwgz + Ps) (1 + d/100) ρw : water density (1020 kg/m3) g : gravitational acceleration z : caprock depth Ps : surface pressure (1 atm) d : deviation (e.g., ±50%) An initial caprock temperature T is computed assuming a linear relationship between T and depth below sea bottom, i.e., T = Tb + ∇T (z − zb) Tb : seafloor temperature (7◦C) zb : seafloor depth (100 meters) ∇T : thermal gradient (e.g., 35.6◦C/km ±10◦C)
17 / 23
Aquifer pressure (bar) (press dev. = 0%) below critical press.
300 600 900 1200 1500 1800 CO2 (Mt)
25 50 Pressure deviation (%)
Impact of pressure on structural capacity
e r
r e s s u r e d u n d e r
r e s s u r e d Aquifer temperature (◦C) (temp gradient = 35.6◦C/km) below critical temp.
300 400 500 600 700 CO2 (Mt) 25 30 35 40 45 Temperature gradient (C/km)
Impact of temperature on structural capacity w a r m e r g e
e l c
e r 608 Mt conditions below critical point 1211 Mt 671 Mt
Base geomodel Impact of variation on trapping capacity
18 / 23
18 / 23
Faults treated as sealing (Tf = 0), semi-sealing (Tf = 0.01), or conducting (Tf = 1) Simulate injection through series of wells adjacent to faults; observe migration and pressure buildup
N N
Hammerfest aquifer basin
19 / 23
Year 10
Sealing Semi-sealing None
CO2 saturation Pressure (bars) compartmentalized pressure buildup
20 / 23
Year 50
Sealing Semi-sealing None
CO2 saturation Pressure (bars) (Pressure dissipates and returns to initial conditions by Year 3000) highly compartmentalized pressure buildup
21 / 23
21 / 23
Recap:
◮ Aim was not to provide accurate storage capacities, but to illustrate
sensitivity to parameter variation due to uncertainty
◮ Use of VE modelling and migration forecasting reduced
computational cost of simulations Key findings:
◮ Uncertainties in zt and k can have greater impact on plume
dynamics than φ (e.g., minimal plume mis-match seen in Utsira even after perturbing φ by ±50%)
◮ Geomodels can respond differently to same perturbation level
(e.g., increased trapping in Utsira after perturbing zt by ±8 meters, however not in Sandnes)
◮ CO2 density and its state can change within a matter of meters
(depth); dependent on aquifer conditions
◮ Sealing and semi-sealing fault treatments can yeild noticeably
different results in terms of pressure compartmentalization Future work: Consider cross-correlation effects, not just one-factor-at-a-time
22 / 23
This work was funded in part by the Research Council of Norway through grant no. 243729 (Simulation and optimization of large-scale, aquifer-wide CO2 injection in the North Sea).
https://www.sintef.no/MRST/
23 / 23
23 / 23
[1] Halland, E.K., Mujezinovi´ c, J., Riis, F., (Eds.), (2014) CO2 Storage Atlas: Norwegian Continental Shelf, Norwegian Petroleum Directorate
24 / 23
Sensitivity Analysis: measure a model response (i.e., capacity estimates, plume migration) to changes in model input parameters (i.e., geomodel properties).
25 / 23
Utsira-South: model response (structural trapping) to perturbations
500 550 600 650 700 CO2 (Mt) 20 40 60 80 100 120 140 160 180 Frequency 0.2 0.4 0.6 0.8 1 Probability 520 540 560 580 600 620 640 660 CO2 (Mt) 50 100 150 Frequency 0.2 0.4 0.6 0.8 1 Probability
Structural capacity (Mt) Parameter µ σ cov = σ/µ caprock depth ± 1 meter 608 5.4 0.9% ± 5 meters 607 25.5 4.2% ± 15 meters 614 68.2 11.1% porosity ± 0.02 592 6.1 1.0% ± 0.05 592 15.2 2.6% ± 0.10 592 30.1 5.1%
NB: larger zt perturbations increases the mean trapping capacity µ...
26 / 23
CO2 density (kg/m3)
supercritical
27 / 23
20 40 60 80 T (C) 500 1000 1500 Depth (m) temp grad = 30 C/km temp grad = 35.6 C/km temp grad = 40 C/km 50 100 150 200 P (bar) 500 1000 1500 Depth (m) 200 400 600 800 Density (kg/m3) 500 1000 1500 Depth (m)
31◦C 73.8 bar
20 40 60 80 T (C) 500 1000 1500 Depth (m) 50 100 150 200 P (bar) 500 1000 1500 Depth (m) p deviation = -25% p deviation = 0% p deviation = 25% 200 400 600 800 Density (kg/m3) 500 1000 1500 Depth (m)
31◦C 73.8 bar 28 / 23
Lie et al., (2016) A simulation workflow for large-scale CO2 storage in the Norwegian North Sea, Computational Geosciences 20 (3) 607-622
29 / 23
◮ CO2 reference density: 760 kg/m3 (ρ = ρ(p, T) via Span & Wagner) ◮ water density: 1020 kg/m3 ◮ CO2 viscosity: 6 × 10−5 Pa·s (variable, µ = µ(p, T)) ◮ brine viscosity: 8 × 10−4 Pa·s ◮ rock compressibility (pore volume multiplier): 1 × 10−5 bar−1 ◮ water compressibility: 4.3 × 10−5 bar−1 ◮ residual water saturation: 0.11 ◮ residual CO2 saturation: 0.21 ◮ sea depth: 100 meters ◮ seabed temperature: 7 ◦C ◮ temperature gradient: 35.6 ◦C/km ◮ porosity: from CO2 Atlas ◮ permeability: from CO2 Atlas
30 / 23