We present a methodology that uses pilot and anchor points with
probability distributions for saturated hydraulic conductivity in a
groundwater contaminant transport model. This approach directly links
locations with calibration target data (e.g., water levels and drawdown
at monitoring wells) to the most relevant physical parameter(s) that
drive behavior, in a way that promotes model parsimony. Distributions
for hydraulic conductivity are developed for monitoring well locations
with pumping tests in order to reflect the state of uncertainty in the
local estimates; these locations are called anchor points. Pilot points
are placed between monitoring wells, and because they have more
uncertainty these are generally assigned wider distributions that
reflect plausible hydraulic conductivity values for the geologic
material in which they are located. Scaling issues are considered in the
development of these distributions. Pilot points are not randomly or
uniformly distributed in the domain; rather they are considered
connectors between locations with data (anchor points) and placed
strategically between them. For a given model realization, hydraulic
conductivity values at both pilot and anchor points are sampled from
their respective distributions and all remaining locations are derived
using an interpolation scheme (e.g., kriging). This approach to
hydraulic conductivity assignment honors location-specific data,
geologic heterogeneity, and spatial patterns. Given that inverse
analysis of high-dimensional models tends to be ill-posed and thus
sensitive to initialization of parameters, the distribution development
process plays a critical role in driving the outcome of model
calibration.