Modeling Uncertainty:
Realism vs Conservatism in
Radiological Performance Assessment

John Tauxe, Paul Black, Bruce Crowe, and Don Lee

presented at the 2003 National Ground Water Association Midsouth Focus Conference

Contact: John Tauxe
Neptune and Company, Los Alamos, NM 87544
505-662-2121 • jtauxe(a)neptuneinc.org
(replace (a) with @ to make the address legitimate)

Read the abstract below, or see the on-line presentation:
Tauxe_et_al_NGWA_2003.html
(seems to work only with Microsoft Internet Explorer).
If that does not work, download the PowerPoint Show version.


Modeling Uncertainty:
Realism vs Conservatism in
Radiological Performance Assessment
•Abstract•

John D. Tauxe, PhD, PE, Neptune and Company, Inc.;
Paul K. Black, PhD, Neptune and Company, Inc.;
Bruce M. Crowe, PhD, Los Alamos National Laboratory; and
Donald W. Lee, PhD, PE, Oak Ridge National Laboratory

Introduction

The U.S. Department of Energy (DOE) requires the completion of performance assessments (PAs) for their low-level radioactive waste (LLW) disposal facilities. The PAs provide the basis for establishing with reasonable expectation that disposal sites meet the radiological performance objectives established in DOE M 435.1. Compliance with these requirements has been achieved at multiple sites across the DOE complex through development of deterministic PAs that are formally reviewed and approved by the Low-Level Waste Federal Review Group (LFRG). This approval process by the LFRG is the basis for issuance of a Disposal Authorization Statement (DAS), a major step in the management of LLW disposal facilities.

All PAs have uncertainty in their analytical and numerical models, in their model assumptions, in their input parameters used in the models and in the conceptual model assumptions and scenarios that underlie and support the models. The impact of uncertainty on the conclusions of the PAs must be evaluated systematically over a 1,000-year compliance interval. Uncertainty is not fully assessed in a deterministic PA but is instead presumed to be bounded through a combination of conservative assumptions and application of bounding parameter values in the PA calculations. Uncertainty in the performance of a disposal system should be more explicitly quantified and evaluated and/or reduced during PA maintenance studies that follow PA approval by the DOE. Quantification and reduction of uncertainty can greatly aid the DOE in their long-term management of disposal facilities. Verification of a PA should be viewed as a two-step process. The first step requires demonstration with reasonable expectation of the compliance or safety of a disposal site with respect to the performance objectives established through regulation. The second requires quantification of uncertainty, and use of that knowledge to cost effectively manage disposal facilities. This presentation is concerned with the use of probabilistic PA modeling as an effective tool for the second part of the two-step process.

Background

Low-level radioactive waste from environmental cleanup activities at the Nevada Test Site (NTS) and from multiple sites across the DOE complex is disposed at two operating Radioactive Waste Management Sites (RWMSs) on the NTS. These facilities are managed by the DOE National Nuclear Security Administration Nevada Site Office (NNSA/NSO) and were recently designated as one of two regional disposal centers. Annual volumes of disposed waste exceed 50,000 m3 (> 2 million ft3). To safely and cost-effectively manage the disposal facilities, the Waste Management Division of NNSA/NSO is in the process of implementing decision-based management practices using problem-oriented probabilistic performance assessment modeling. Probabilistic PA models are under development and replace the existing deterministic PAs completed originally for the Area 5 and Area 3 RWMSs located in, respectively, Frenchman Flat and Yucca Flat on the NTS.

Probabilistic PA models use probability distributions to represent significant input parameters and sample through simulation these distributions through application of numerical models. Probabilistic models encompass uncertainty in the inventory, in fate and transport processes and in exposure pathways to potential receptors while attempting to represent and evaluate all components of uncertainty (variability, parameter uncertainty, model uncertainty and scenario uncertainty). The outputs of probabilistic models are probability distributions that, if correctly constructed, represent a best estimate of the performance of a disposal site and the uncertainty associated with that estimate, conditioned on the model assumptions. The development of probabilistic PA models for the RWMSs represents a concerted effort to replace the existing conservative deterministic PA with models that more closely approach the realistic performance of the disposal sites. There is no clearly agreed upon definition of probabilistic modeling, but three components are nearly always at the heart of all probabilistic models: There is continuing confusion surrounding the definitions, advantages, disadvantages and resources required to develop and apply probabilistic models to PA studies of LLW disposal sites. Application of probabilistic modeling is currently undergoing a renaissance in environmental work because this approach more fully encompasses the complex and dynamic system responses and the uncertainty of environmental systems. This renaissance is a natural progression of increased acceptance and understanding of probabilistic modeling, and the continued technological advances in computer hardware and software. Probabilistic models can be perceived as academic rather than practical with unnecessarily complicated models that are approachable only by researchers with strong backgrounds in mathematical modeling and probability theory. To the contrary, these models can and should be highly flexible and easily adaptable to a range of decision problems. The models and the probabilistic methodology underlying the models can be designed to directly respond to regulatory and decision requirements and to be understandable at even a non-technical level. They should be focused at an applied level and should be readily approachable by the standard teams that have successfully conducted LLW PAs for DOE sites.

This paper attempts to overcome some of the resistance to the application of probabilistic models to LLW disposal sites by considering Ongoing model refinement of the Area 5 and Area 3 PAs follows an iterative process with initial development of a fully probabilistic model followed by successive upgrading of the probabilistic model components guided by program decision priorities, and the results of sensitivity and uncertainty analyses. Current activities are focused on development of defensible mean-centered probability density functions for model parameters and evaluating and updating model components shown by sensitivity analyses to be significant contributors to total system variance. Probabilistic models emphasize fate and transport processes associated with shallow trench disposal of LLW in the unsaturated zone of an arid desert setting where the primary release mechanisms are dominated by upward directed advective and diffusive flux of liquid and vapor. Model revisions include reassessment of the magnitude and range of liquid advection, and changes through data acquisition in the biotic transport modules (plant uptake and animal burrow excavation). Additional components under consideration for revision are the range and defensibility of defining inventory uncertainty for variable waste streams and developing more realistic and less conservative source term release models.

The primary benefits of probabilistic PA models for shallow land burial of LLW include reduction in conservatism thereby allowing full utilization of the disposal capability of the sites and streamlined evaluation of new waste streams and problematic waste streams. Furthermore, the use of probabilistic model outputs markedly enhances sensitivity and uncertainty analyses promoting more efficient decision making for model improvements, for uncertainty reduction, and for streamlining of monitoring programs. Finally, iterative probabilistic modeling combined with cost-benefit analyses can be used to establish a cost-effective and defensible strategy for facility closure and long-term stewardship.

reference for this presentation:

Tauxe, J.D., P.K. Black, B.M. Crowe, and D.W. Lee, 2003, Modeling Uncertainty: Realism vs Conservatism in Radiological Performance Assessment, 2003 NGWA Midsouth Focus Conference • Subsurface Monitoring & Modeling Issues, Nashville, TN, September 18-19 2003