The Stochastic Cost Optimization Toolkit (SCOToolkit) was developed to provide a practical tool to assess and optimize the design and operation of remediation systems for groundwater contaminated with chlorinated solvents and other DNAPLs. SCOToolkit explicitly considers uncertainty in site characterization, remediation system characteristics, monitoring data, and model prediction by optimizing remediation system design and monitoring variables to minimize the probability-weighted life cycle cost to achieve remediation objectives.

Technical Approach

SCOToolkit uses a robust semi-analytical model to simulate DNAPL source depletion and 3-D dissolved phase transport in response to natural or engineered conditions. Groundwater flow is modeled as a steady-state curvilinear flow field. The transport model is coupled with performance models that consider the following remediation technologies:

  • Monitored natural attenuation
  • DNAPL source excavation
  • Thermal source treatment (electric resistance, conduction, steam)
  • In situ chemical oxidation (pulsed injection or recirculation with reinjection)
  • Electron donor injection for enhanced source and/or plume remediation
Termination criteria for individual remediation system components and for site-wide compliance are defined by statistical rules based on monitoring data and the performance models are coupled with cost functions to compute the time and cost to achieve site-wide compliance. Cost functions are also considered to implement plume containment measures (e.g., barrier walls or pump-and-treat). The performance model is coupled with an inverse solution to estimate model parameters, parameter covariances, and residual prediction error from available field data. These results are used to generate equiprobable realizations of remediation performance and cost. A stochastic cost optimization algorithm is used to determine values for design and operation variables, including monitoring, to minimize the expected (i.e., probability-weighted average) net present value cost to achieve compliance. The program allows users to incrementally assess and optimize remediation performance by periodically recalibrating with new monitoring data, updating performance and cost projections, and reoptimizing design and monitoring variables. Incremental application of SCOToolkit will significant decrease time-to completion and cost-to-completion for most sites. Very large savings will be realized when systems that are likely to fail or significantly underperform are identified early enabling appropriate adjustments are made promptly. However, we have seen that simply optimizing operational and monitoring parameters for a given remediation strategy can reduce life-cycle cost by 20 to 40% or more. Likely savings over a portfolio of sites will far exceed the cost of performing analyses with SCOToolkit.


The idea for this project grew from a 2006 SERDP/ESTCP workshop on Reducing the Uncertainty of DNAPL Source Zone Remediation, which crystallized the need for a comprehensive but practical approach to site remediation that couples an array of simulation models with calibration, error propagation, and stochastic optimization algorithms. SCOToolkit is the outcome of SERDP project ER-2310 led by Jack Parker (University of Tennessee) and Ungtae Kim (Cleveland State University). Dr. Kim was responsible for implementation and integration of the many program components and functions through many iterations as our understanding of the complex stochastic decision process grew to produce the final efficient and robust program that is now available to improve remediation design performance and reliability and reduce life-cycle cost. Initial versions of the coupled calibration and stochastic cost optimization code were written by Mike Cardiff, Xiaoyi Liu and Jonghyun Lee with direction from Peter Kitanidis (Stanford University) under SERDP project ER-1611. Bob Bordon (North Carolina State University) provided guidance for the ISCO model formulation. Alyson Fortune, Steffen Griepke, James Galligan, Amber Bonarrigo and Ralph Baker (TerraTherm Inc) and Greg Beyke (TRS Group) provided advise on thermal model development. Aleisa Bloom (Oak Ridge National Lab), Robert Lyon (URS Corp), James Gillie and Bill Myers (Joint Base Lewis McChord), Mike Singletary, Val Jurka and Arun Gavaskar (NAVFAC), Dave Becker, Mike Riggle, Jean Chytil, Chuck Coyle, Jeff Skogg, Delma Stoner and Lynn Jenkins (USACE) helped identify and implement analyses of field demo sites. James Rayner, Cathy Crea, Dave Reynolds and Michael Kavanaugh (Geosyntec), Bernie Kueper and Kevin Mumford (Queens University), and Peter Kitanidis (Stanford University) undertook a rigorous beta test of SCOtoolkit on “virtual site” data generated with a high resolution numerical model. From the initial idea for SCOToolkit through its final development, Andrea Leeson (SERDP ER program manager) and members of the SERDP Science Advisory Board provided encouragement for the effort in addition to funding support.