Conditioning a Robust Remediation Design Using the Concept of Scaled Likelihood

Optimization modeling has been shown to be an important tool for cost-effective groundwater remediation. Currently most optimal remediation design tools are deterministic, assuming full knowledge of all inputs. Those that include uncertainty, typically utilize a random sampling of possible conductivity fields as input, with each realization presumed equally likely. This study develops and evaluates a scaled likelihood (ScL) approach to robust remediation design. An existing robust optimizer that links a three-dimensional numerical flow and transport simulator to a genetic algorithm (GA) is modified to the ScL approach. The robust optimizer considers different possible conductivity fields in each generation of the GA search and searches for a robust design that performs well across multiple feasible realizations. In the ScL approach, multiple conductivity realizations are still utilized, but greater weight is placed on the conductivity realizations that result in more accurate representations of the initial plume concentration measurements. The ScL approach impacts the estimated initial plume concentrations, the measure of fitness across realizations, and the estimated reliability of a remediation policy. The remediation costs and reliability of the ScL-based solutions will be compared to those found using the existing optimizer. Furthermore, computational effort of the two approaches will be compared.