This document discusses efforts to model groundwater flow near the Waste Isolation Pilot Plant (WIPP) salt repository using the d3f++ and PFLOTRAN codes. It summarizes work to update an existing coarse-scale model of the WIPP site to include density-driven flow and improve the mesh and parameterization. Challenges included the old mesh's irregularity and aspect ratios as well as representing an evolving water table. Both codes struggled with the original mesh. Simpler 2D benchmark problems were suggested to better compare the codes' capabilities before further work on the full basin-scale model.

1. Basin-Scale Density-Dependent
Groundwater Flow Near a Salt Repository
Anke Schneider
Gesellschaft für Anlagen- und Reaktorsicherheit
Kristopher L. Kuhlman
Sandia National Laboratories
Middelburg, The Netherlands
September 5-7, 2017
Sandia National Laboratories is a multi-mission laboratory managed and
operated by National Technology and Engineering Solutions of Sandia LLC, a
wholly owned subsidiary of Honeywell International Inc. for the U.S. Department
of Energy’s National Nuclear Security Administration under contract DE-
NA0003525. SAND2017-9392 C.

2. WIPP Hydrogeology
 Repository in Salado
bedded salt formation
 >500-m thick salt unit
 Hydrogeology of
formations above salt
 Rustler Formation
 Culebra dolomite
 Magenta dolomite
 Anhydrite
 Mudstone/Halite
 Dewey Lake Red Beds
 Silt/sand stones + clay
 Dockum Group
 Silt/sand stones + clay
2

3. Rustler Conceptual Model
3
West
(Nash Draw)
East  West of WIPP
 Shallow units
 High permeability
 Relatively fresh water
 East of WIPP
 Deeper units
 Low permeability
 Saturated brine
 Regional groundwater
 Flow used in WIPP PA
 Long-term geological
stability of salt

4. 4
Corbet (2000)
Model Domain
WIPP LWB
NM/TX Border
1996 WIPP PA Model
2004 PA Model
2014 WIPP PA Model
Scale [km]
0 10 20 4030

5. Corbet (2000) WIPP Model
5
 Most of Delaware Basin
 Transient Simulation
 Climate variation (dry vs. wet)
 14,000 y → present → 10,000 y
 Model Implementation
 “water table” moving boundary
model
 ~8700 km2 region (78 km × 112 km)
 Coarse mesh (2 km square cells)
 12 model layers (10 geo layers)
 1,500 cells/layer
 ~18,000 elements total
Halite areas (low k)
Dissolution area
(high k)

6. Motivation
6
 Benchmark against existing solution (Corbet, 2000)
 Comparison with original model
 Old mesh, model parameters & boundary conditions
 Include new processes, features & data
 Include density-driven flow (e.g., Davies, 1989)
 Include chemistry & mineral dissolution
 Investigate flow & chemistry boundary conditions
 Test and update hydrogeological conceptual model
 Incoporate current data: 81Kr GW age data, water level data
 Comparison and Development of Models
 PFLOTRAN (SNL)
 Add density dependent flow
 d3f++ (GRS)

7. Update of Corbet model
7
Corbet (2000): Hydraulic conductivity [m/s]
d3f++/PFLOTRAN: Permeability [m2]

8.  density-driven groundwater flow
 salt and heat transport
 fluid density and viscosity depending on salt concentration
and temperature
 porous and fractured media
 free groundwater surface – levelset function
 sources and sinks
 transport of radionuclides
 decay and ingrowth
 equilibrium and kinetically controlled sorption
 precipitation/dissolution
 diffusion into immobile pore water
 colloid-borne transport
 numerics based on UG, G-CSC, Frankfurt University
 finite volume methods
 geometric and algebraic multigrid solvers
 completely parallelized (UG: scaling invest. some 100,000 proc.)
8
d³f++: distributed density-driven flow

9. Applications of d3f++
Applications 9
 Porous media, overburden of host formations
• Gorleben Site: 2D density-driven flow and RN
transport in high saline environment
• Cape Cod: 2D contaminant transport with
pH-dependent sorption
 Low permeable media
• Generic German Site in clay: 3D diffusive transport
in a low permeable anisotropic clay formation
 Fractured media
• Yeniseysky site: Flow and transport in fractured rock
• Äspö (URL): Flow in the repository near field
• Grimsel (URL): Colloid-facilitated transport in clay

10. WIPP Site: „Basin-Scale“ model
 SNL: Data of „Basin-scale“
groundwater model after
Corbet & Knupp 1996
 raster data of 10
hydrogelogic units
source:
SNL, SECOFL3D
10
d³f++
ProMesh
www.promesh3d.com
Dewey Lake/Triassic
Anhydrite 5
Mudstone/Halite 4
Anhydrite 4
Magenta Dolomite
Anhydrite 3
Mudstone/Halite 3
Anhydrite 2
Culebra Dolomite
Los Medanos Member
Forty-Niner Member
Tamarisk Member

11. unit permeability [m²]
Dewey Lake/Triassic 10-14-10-12
Forty-Niner Member 10-20-10-12
Magenta Dolomite 10-18-10-12
Tamarisk Member 10-20-10-12
Culebra Dolomite 10-17-10-11
Los Medanos Member 10-17
WIPP-Site: Prism grid, 6 layers
N
example:
Culebra Dolomite
source: Corbet 2000
182,784 prisms (2x refined)  18,000 hexahedrons SECOFL3D
50x vertical exaggeration
11

12. WIPP-Site: Initial and boundary conditions
N
closed boundaries
c=1
(saturated brine)
12
assumed
recharge rates
source:
Corbet 2000
initial condition:
water table
14,000 years ago
source:
Corbet &Knupp 1996

13. WIPP-Site: Initial and boundary conditions
N
closed boundaries
c=1
(saturated brine)
recharge 2.0 – 0.0/0.1 mm/year, c=0 / seepage
initial condition:
water table
14,000 years ago
source:
Corbet &Knupp 1996
13

14. WIPP-Site: d³f++ simulations
14
density-driven flow, fixed water table (top boundary), permeability const. (layers)
(280,000 prisms)
model time 10,000 years, computing time 15 minutes, timestep 100 years
concentration
Darcy’s velocity

15. WIPP-Site: d³f++ simulations
15
density-driven flow, free water table
level 2 (182,784 prisms)
free water table
salt concentration

16. WIPP-Site: d³f++ simulations
16
density-driven flow, free water table
level 2 (182,784 prisms)
model time 100 years,
timestep 0.005 year (levelset method)
Darcy’s velocity, water table

17. Summary and outlook
Difficulties:
 non steady-state density-driven flow model
 strongly anisotropic (thin layers, jumping coefficients)
 free groundwater surface 8,700 km²
Current work:
BMWi-funded joint project GRUSS (GRS, G-CSC Frankfurt University)
 improve grid generating/refinement
 improve robustness of solvers (convergence, timesteps)
 implement volume of fluid (VOF) method to speed-up free surface handling
Next steps:
 increase timestep levelset method
 simulation 14.000 years past
 reproduction of SECOFL3D results (Corbet & Knupp, 1996)
17

18.  Reactive multiphase flow and transport code for porous media
 Open source license (GNU LGPL 2.0)
 Object-oriented Fortran 2003/2008
 Pointers to procedures
 Classes (extendable derived types with
member procedures)
 Founded upon well-known (supported) open source libraries
 MPI, PETSc, HDF5, METIS/ParMETIS/CMAKE
 Demonstrated performance
 Maximum # processes: 262,144 (Jaguar supercomputer)
 Maximum problem size: 3.34 billion degrees of freedom
 Scales well to over 10K cores
18

19. SNL PFLOTRAN version
19~25x vertical exaggeration

20. SNL PFLOTRAN version
20Original Mesh: 13-layer hexahedral (cuboid) elements (18,000 elements)
100x vertical exaggeration

21. Issues Encountered
21
 Old Mesh is very coarse
 PFLOTRAN and d3f++ have difficulty with mesh
 Mesh violates conventions regarding
 Regularity (Δz varies too much in space)
 Connectivity (must build mesh “by hand”)
 Aspect ratio (2 km × 2 km × 1s-100s m)
 Anke (GRS): re-mesh using modern tools (LARGE)
 Kris (SNL): struggle with old mesh (COARSE)
 Moving water table ≠ Richards equation
 Unsaturated flow parameters are guessed
 Recharge applied at water table vs. applied at land surface
 CFL condition requires very small time steps
 Too few elements to capitalize on parallel
 Smaller elements → smaller time steps!

22. Schedule
22
 SECOFL3D data provided by SNL
 GRS begins building d3f++ model
 SNL begins building PFLOTRAN model
 SNL consults
 GRS builds d3f++ model equivalent to Corbet (2000)
 SNL builds PFLOTRAN equivalent to Corbet (2000)
 GRS ‘includes’ density-driven flow
 SNL includes density-driven flow to PFLOTRAN
 WIPP basin-scale model is:
 Numerically difficult
 Uses non-ideal mesh (pancake elements)
 Has complex boundary conditions
 Try benchmarking simpler (2D) problems:
 Compare processes
 Use PFLOTRAN QA suite problem?
Year1Year2+Year

23. 23
Thank you for your attention!