This report is the third (and final) report in a series that addresses issues related to hydrologic uncertainty assessment at decommissioning sites. Analyses in the first two reports in this series emphasized the application of relatively simplified models of subsurface flow and transport. Because of their relative computational speed, such simplified models are particularly attractive when the impact of uncertainty in flow and transport needs to be evaluated. These same simplifications, however, have the potential to provide unrepresentative estimates of dose and its uncertainty. Such misrepresentation may have important consequences for decisions based on the dose assessments. The significance of this concern was evaluated by comparing results from uncertainty assessments conducted on a test case using a simplified modeling approach and a more complex/ realistic modeling approach.
The test case used a three-dimensional domain with a U-234 source in the near surface, a 5-m-thick aquifer 7 m below the surface, and a small well pumping directly downstream, on the boundary of the contaminated zone. Exposure was assumed to occur through the drinking water pathway only, with all drinking water originating from the pumped well. A series of Monte Carlo simulations of flow and transport were performed using STOMP as the complex model and RESRAD as the simplified model. Hydraulic conductivity and air entry were modeled as random fields in STOMP using geostatistics from the Las Cruces Trench site. Random distributions for hydraulic parameters in the RESRAD simulations were based on both site-specific data and generic distributions. Sensitivity analyses were conducted on both codes using a combination of Monte Carlo simulation and single parameter variation.
Peak doses predicted by the simplified model were several times higher than peak doses predicted by the complex model. This difference was attributed to the lack of dispersion in RESRAD and differences in aquifer mixing. The RESRAD concentration breakthrough curves exhibited a sharp peak with essentially no contaminant in the well until the time of the peak while STOMP predicted a much earlier arrival of contaminants in the well. Which code provided conservative results thus depended on whether the RESRAD peak occurred before the 1000-year regulatory criterion.
The random field characterization of the subsurface for the complex model used all available site data. Uncertainty in predicted dose was correspondingly small, with the peak dose coefficient of variation being 30%. When the variances of parameters in the simplified model were based on a generic dataset, the uncertainty in predicted peak dose was much larger; the coefficient of variation was 52% in this case. When the variances of parameters in the simplified model were based on the extensive site-specific data, the coefficient of variation for the peak dose was reduced to 22%. In this case, however, the mean peak dose was actually less similar to the STOMP results than the generic case.
For the RESRAD Monte Carlo simulations involving random soil hydraulic properties, the variability of peak dose was entirely attributed to variability in the aquifer hydraulic conductivity. Sensitivity to other parameters was examined by varying one parameter at a time. These results indicated that the recharge rate, the aquifer gradient, and the depth of penetration of the well were significant contributors to uncertainty in peak dose and the time of the peak dose. RESRAD predicted peak dose was more sensitive to the parameter values than was the STOMP predicted peak dose for the aquifer hydraulic conductivity, aquifer gradient, and depth of well penetration. Sensitivity to the recharge rate appeared to be comparable for the two codes.
Stochastic predictions of mean dose over time for the complex model were relatively insensitive to the geostatistical parameters. Ensemble mean peak dose predicted by the complex model was sensitive to the ensemble mean hydraulic conductivity. For a given ensemble mean hydraulic conductivity, however, the complex model showed no correlation between the spatial geometric mean aquifer conductivity of individual realizations and the resulting peak dose. This was in contrast to the simplified model in which the dose from individual realizations was strongly correlated with the aquifer conductivity. This result has implications for the value of hydraulic conductivity data. Adopting the homogeneous parameterization of RESRAD leads to a conclusion that reducing the uncertainty in the value of the aquifer hydraulic conductivity parameter will have a significant impact on the uncertainty in peak dose. Looking at the individual realization results from the STOMP model suggests that characterization of the average aquifer hydraulic conductivity is relatively unimportant in reducing uncertainty in peak dose. Characterizing the pattern of aquifer heterogeneity is likely to be more important than obtaining a value for the spatial mean hydraulic conductivity.
President and CEO, Shlomo Orr, PhD, Peng, has over 33 years of extensive consulting, research, and project management experience in the field of Hydrology and Water Resources. He earned a PhD in Hydrology and Water Resources with a minor in Soil and Water Sciences, and BSc and MSC in Civil Engineering. Dr. Orr's background includes modeling, planning, and controlling complex subsurface flow and transport phenomena. He has a broad background in conceptual and computational-numerical modeling of fluid flow and solute transport in saturated and unsaturated porous and fractured formations, including stochastic models that account for uncertainties and provide the basis for risk assessment.
©Copyright - All Rights Reserved
DO NOT REPRODUCE WITHOUT WRITTEN PERMISSION BY AUTHOR.