Uncertainty Reduction for Parameters of a Seawater Intrusion Model using Markov Chain Monte Carlo

The Milrow underground nuclear test was one of three tests that were conducted on Amchitka Island, Alaska. A stochastic groundwater flow and contaminant transport model was created for the site which propagated uncertainty in input parameters through flow and transport simulations to yield an output with a wide range of uncertainty. The Consortium for Risk Evaluation with Stakeholder Participation (CRESP) sponsored field efforts in the summer of 2004, which yielded data pertaining to the location of the freshwater lens, derived from geophysical surveys on the island. These data are compared to the groundwater model input distributions for verification, and then the distributions are tightened around the new data for uncertainty reduction using a Markov Chain Monte Carlo approach. The geophysical data resulted in a transition zone location much deeper than that identified from the salinity profile at a borehole drilled on the island near Milrow ground zero. A number of scenarios are investigated in which reduction of parameter uncertainty through the use of Markov Chain Monte Carlo is evaluated using the salinity data alone, the geophysical data alone, and all data together. Due to the inconsistency between the salinity data and the geophysical logs, the use of these data sets combined does not yield a reduction of uncertainty similar to that obtained by using the geophysical logs alone. A hypothetical scenario is used where the geophysical interpretation is assumed to be consistent with the salinity data and the resulting reduction of uncertainty is found to be significant.