Presentation from 2008

1. Parametric versus Conceptual
Uncertainty – Frenchman Flat,
Nevada
MODFLOW and More:
Ground Water and Public Policy
May 19-21, 2008
Greg Ruskauff

2. Acknowledgements
 Work performed by the Stoller-Navarro Joint Venture
(and its subcontractors) under DOE Contract DE-AC52-
03NA99205. Project Managers: Bill Wilborn, K.C.
Thompson
 Many contributors, including: Ian Nienheuser, Nicole
Denovio, John Ewing, Edward Kwicklis

3. Introduction
 Underground nuclear testing
conducted on the Nevada Test Site
from 1957-1992
 Testing left behind radionuclides in
the rock
 Underground Testing Area (UGTA)
Project established to ensure the
protection of the public and the
environment

4. Geologic Setting
 Great Basin area of United
States
 Complex geology
 Many major features are
buried
 Drill-hole data sparse,
difficult to acquire
 Indirect means used to
establish the character and
presence of rocks

5. Frenchman Flat Geologic Model
Alluvial
Aquifers
B
Volcanic B’
Aquifers
Aquitards
Lower
Carbonate
Aquifer

6. Frenchman Flat HFM Uncertainty
HFM Alternative Description
DETA – detachment fault This alternative is a no detachment fault model.
alternative
DISP – Displacement fault This alternative is concerned with the locations and
alternative displacement of basin-forming faults.
CPBA – CP Basin The CP basin alternative extends the UCCU beneath
alternative all of CP basin.
BLFA – Basalt Lava-Flow The BLFA HSU is modeled as a single continuous
Aquifer flow, rather than three separate zones
Base BLFA

7. Conceptual Model Uncertainty
 Bredehoeft (2005) has suggested that proper conceptual model
selection is the greatest uncertainty
 Neuman and Weirenga make a similar point (NUREG/CR-6805, 2003)
 Nishikawa (1997) and Harrar et al. (2003) demonstrate the impact of
alternative conceptual models on predictions
 Beven (1993) laid out the argument for considering multiple alternative
models and model structures due to the problem of ‘equifinality’ - the
concept that a unique model with an ‘optimal’ set of parameters is
inherently unknowable. Instead, he argued for a set of acceptable and
realistic model representations that are consistent with the data.
 UGTA Project explicitly recognizes and addresses this issue

8. Uncertainty Components
Decision
Structural
Scenario/Framework Uncertainty
Model
Parameter/Knowledge
Statistical
Variability Uncertainty
Parameter/Knowledge
Model
Structural
Scenario/Framework
Uncertainty
Decision

9. Uncertainty Analysis
 Initial idea was “The differences in geologic framework models cause resulting
differences in unit properties that are at least as large as the uncertainty in the
properties, and thus the uncertainty is captured.”
− Conduct transport parameter uncertainty on a fixed flow field
1500
1400
 Null space Monte Carlo 1300
 sampling allows calibration 1200
After Calibration
 constrained parameter 1100
Calibration Objective Function (-)
Forward Run Only
 uncertainty to be combined
1000
900
 with HFM uncertainty 800
700
 Using the SVD and a sampled 600
500
 parameter project onto space 400
 of small eigenvalue (little 300
 influence on solution) 200
100
0
10
13
19
22
25
28
31
34
37
46
49
73
82
85
88
91
94
97
10
1
4
7
16
40
43
52
55
58
61
64
67
70
76
79
Realization
 What is the impact of flow calibration on the prediction of interest?
− Does the hard data constrain the prediction?

10. First-Order Predictions
 Flow through
test cavity as
a simple
prediction
surrogate

11. Transport Results
 Computed
plume
volume
exceeding
SDWA
radionuclide
standard as
metric

12. Flow Model Calibration
 HFMs have many (more than 100) adjustable parameters
 Frenchman Flat has 26 steady-state head observations, and model boundary
flows (4) estimated from the regional model
 HFMs have many (more than 100) adjustable parameters
 PEST with singular value decomposition and regularization with a Linux
cluster able to effectively calibrate and conduct parameter uncertainty
analysis
 Singular value decomposition of the JtQJ
matrix in order to define the linear
combination of base parameters which
comprises each super parameter –
dominant eigenvalues and their components
(eigenvectors) become “super” parameters
− Idea first proposed by Jacobson (1985)
− 10 to 15 super parameters

13. Conclusions
 Discerning (especially before any model analysis) geologic uncertainty
important to the prediction is difficult in this complex setting
− Some alternatives have obvious consequences relative to the flow model
behavior
− Others have no obvious flow model consequence, but could have
transport consequences not revealed during flow model analysis
 Approached thoughtlessly, having additional HFMs can add an additional
burden to a project and impede true understanding
 Uncertainty approach has so far met with regulatory approval, but has a
significant burden not suitable in its entirely for all applications

14. Parametric versus Conceptual
Uncertainty – Frenchman Flat,
Nevada
MODFLOW and More:
Ground Water and Public Policy
May 19-21, 2008
Greg Ruskauff