Verification and Uncertainty Reduction of Amchitka Underground Nuclear Testing Models

Abstract

The modeling of Amchitka underground nuclear tests conducted in 2002 is verified and uncertainty in model input parameters, as well as predictions, has been reduced using newly collected data obtained by the summer 2004 field expedition of CRESP. Newly collected data that pertain to the groundwater model include magnetotelluric (MT) surveys conducted on the island to determine the subsurface salinity and porosity structure of the subsurface, and bathymetric surveys to determine the bathymetric maps of the areas offshore from the Long Shot and Cannikin Sites. Analysis and interpretation of the MT data yielded information on the location of the transition zone, and porosity profiles showing porosity values decaying with depth. These new data sets are used to verify the original model in terms of model parameters, model structure, and model output verification. In addition, by using the new data along with the existing data (chemistry and head data), the uncertainty in model input and output is decreased by conditioning on all the available data. A Markov Chain Monte Carlo (MCMC) approach is adapted for developing new input parameter distributions conditioned on prior knowledge and new data. The MCMC approach is a form of Bayesian conditioning that is constructed in suchmore » a way that it produces samples of the model parameters that eventually converge to a stationary posterior distribution. The Bayesian MCMC approach enhances probabilistic assessment. Instead of simply propagating uncertainty forward from input parameters into model predictions (i.e., traditional Monte Carlo approach), MCMC propagates uncertainty backward from data onto parameters, and then forward from parameters into predictions. Comparisons between new data and the original model, and conditioning on all available data using MCMC method, yield the following results and conclusions: (1) Model structure is verified at Long Shot and Cannikin where the high-resolution bathymetric data collected by CRESP yield profiles matching those used to construct the Long Shot and Cannikin model cross sections in 2002. (2) Distributions of model input parameters (recharge, conductivity, and recharge-conductivity ratio) used in 2002 for the three sites are verified where the new data indicate distributions with narrower ranges (smaller uncertainty) but within the range employed in the 2002 model. (3) As a conservative approach, distribution of fracture porosity used in 2002 was deliberately skewed toward lower values. New CRESP data indicate that the selected porosity range was overly conservative. In addition, the range of porosity values obtained from the analysis of the MT data is found to generally be about three orders of magnitude lower than range of values used in the 2002 model, though the values themselves are much larger from the MT data. (4) Distributions of the flow model output (head distribution, salinity distribution, groundwater fluxes) resulting from the 2002 model for the three sites are verified where the new model output after conditioning on the data lie within the range of the 2002 model output. (5) Cannikin model output at location of well UAe-1 is not fully verified where the new model results for small salinity values are not fully enclosed by the uncertainty bounds of the original model output. (6) With the new porosities developed from the analysis of MT data, radionuclides require thousands of years to reach the seafloor. No breakthrough resulted for any of the three sites within the 2000 year model timeframe, despite ignoring all retardation mechanisms (sorption, radionuclide trapping in glass, matrix diffusion, and radioactive decay). (7) The no-breakthrough results verify the original model in the sense that this result lies within the uncertainty bounds of the 2002 model expressed as + 2 {sigma}{sub Q} and - 2 {sigma}{sub Q}. The lower bound, - 2 {sigma}{sub Q}, in the 2002 model gave negative values implying that the bound is essentially zero. The current results of no-breakthrough match this lower bound. (8) Significant uncertainty reduction is achieved for model input parameters (recharge, conductivity, and recharge-conductivity ratio) with the R/K ratio experiencing a very dramatic reduction. (9) Uncertainty in groundwater fluxes is also reduced due to the reduction of R/K uncertainty. (10) Groundwater velocities based on new data are orders of magnitude slower than the velocities produced by the 2002 model due to the higher porosity obtained from the analysis of the MT data. (11) Uncertainty reduction in radionuclide mass flux could not be assessed as the velocities are too small to produce radionuclide breakthrough within the model timeframe of 2,000 years.« less