The objective of this project was to develop and test a methodology to periodically assess and optimize remediation and monitoring strategies at US Department of Defense (DoD) dense nonaqueous phase liquid (DNAPL) contaminated sites with remedies in place (RIP). Methods were developed to model cost and performance of source zone and dissolved plume remediation technologies—including thermal treatment, chemical oxidation, enhanced bioremediation, and reactive barriers—and to optimize system operation and monitoring to meet user-defined cleanup criteria with minimum life-cycle cost, considering uncertainty in performance predictions using a stochastic optimization approach. Physical, chemical and biological processes expected to significantly affect performance are incorporated in the model, including effects of back-diffusion from low permeability zones, such as clay layers or matrix zones in fractured rock.
The capability of the Stochastic Cost Optimization Toolkit (SCOToolkit) developed under previous DoD funding was greatly extended in this project. The previous 2D contaminant transport model was rewritten simulate 3D transport with steady-state groundwater flow along linear or curvilinear streamlines with multiple DNAPL sources. A rigorous solution for resident and flux concentrations was derived and implemented that prevents physically-impossible counter-flow dispersion (which most solutions allow). In conjunction with an upscaled dispersion model, the solution enables efficient simulation of transport in dual-porosity media and associated back-diffusion phenomena.
Performance and cost functions were developed and stringently tested for thermal source reduction (TSR), source zone in situ chemical oxidation (ISCO), enhanced source zone mass transfer, and enhanced dissolved plume bioremediation involving electron donor injection in multiple galleries. Multiple remediation technologies may be operated concurrently or serially.
Site-wide no-further-action decisions were based on statistical criteria applied to compliance well data. For example, annual average concentrations must be less than a specified probability upper confidence limit of current concentration based on an N-year regression. Termination criteria for individual remediation system components were based on component-specific performance monitoring data. For example, individual injection galleries may be shut off when the contaminant concentration is less than a value that is optimized to meet compliance criteria with minimum cost.
Source zone TSR termination decisions were commonly predicated on soil sampling data and ISCO on dissolved concentration data. A method for estimating average soil concentration during thermal treatment from mass recovery measurements was developed and tested, which was found to be more reliable and less costly than soil sampling. Incorporating soil sampling during ISCO was found to reduce errors associated with slow rebound of groundwater concentrations after treatment termination.
Source zone performance monitoring options were incorporated into SCOToolkit, which also allows source regions to be divided into treatment zones (e.g., with different estimated levels of contamination) and subdivided further into monitoring zones (e.g., for soil or water sampling, cumulative mass recovery for thermal treatment). Statistical criteria were developed to allow termination of individual monitoring zones, treatment zones, or the entire system with equal decision reliability at all scales. The SCOToolkit included an inverse solution to obtain best estimates of model parameters and their uncertainty using available field and laboratory data as well as prior estimates of parameters and their uncertainty. A stochastic optimization technique was used to determine optimum operational and monitoring variables to minimize the expected costs over multiple realizations of uncertain parameters and measurements.
Protocols were developed and implemented to periodically refine model calibration taking into consideration new data from monitoring, to assess the probability of the current operations to meet cleanup objectives, and to reoptimize (or redesign if necessary) remediation and monitoring variables to minimize expected cost-to-complete taking into consideration performance and cost uncertainty. Because prediction uncertainty generally decreases as additional data was used for calibration, predictions became more accurate and less overdesign was required to compensate for uncertainty. The SCOToolkit package also included a number of Excel-based tools to pre-process data for input into calibration and optimization modules, as well as to analyze performance monitoring data to make real-time termination decisions based on the multi-scale statistical decision protocol.
Case studies on hypothetical and field sites demonstrated that incremental re-optimization can greatly improve the likelihood of meeting remediation criteria within a target timeframe while reducing the expected cost by 10 to 20% or more over conventional approaches. Optimization of performance monitoring parameters (e.g., termination criteria, number of treatment zones and monitoring zones, type and number of samples per monitoring zone) was observed to reduce expected (probability-weighted average) cost-to-complete by 5 to 15% and to reduce 95% upper confidence limits of cost by up to 30% compared to conventional approaches.
Dividing thermal treatment areas into multiple zones with different soil concentration ranges and allowing individual zones to terminate early when local statistical criteria were met achieved sitewide criteria with 6% lower expected costs than a single zone. Optimizing confidence limit probability, local-scale cleanup level, and number of monitoring zones per treatment zone with three treatment zones, using mass recovery data instead of soil data, achieved an additional 10% cost reduction.
If confirmation of mass recovery-based results with soil sample data is desired or required, delaying each local termination decision until confirmed by soil sampling will increase cost. Therefore, if confirmatory soil sampling is required, the research team recommend waiting until all heating units have been stopped based on mass recovery data before performing site-wide soil sampling.
An optimized example problem using mass recovery data to make thermal termination decisions had a 16% lower expected cost than using soil concentration data following typical industry practice, while the 95% upper confidence limit of cost was 28% lower. Thus, the proposed methodology not only yields expected cost savings, but also sharply reduces worst case cost overruns.
Using multiple zones that are allowed to terminate independently based on statistical criteria provided similar cost savings for ISCO. Optimization of injected oxidant concentrations, treatment zone-level cleanup criteria, reinjection criteria, and performance monitoring variables yielded a failure-adjusted expected cost for an example problem 11% lower than a non-optimized case approximating best engineering practice. Furthermore, the cost probability distribution for the optimized design eliminated positive skew evident in the “best practice” case such that the worst case cost for the optimized design was 14% lower than that for the non-optimized design.
Following thermal treatment of three identified sources at Joint Base Lewis McChord in Washington State, SCOToolkit identified a fourth DNAPL source that had not been located during site characterization studies. Stochastic optimization with interim calibration results did not favor undertaking thermal treatment of the fourth source. Final calibration results indicated that this decision resulted in an undiscounted cost savings equal to 46% of the total cost.
At Dover Air Force Base in Delaware, an incremental stochastic optimization protocol yielded an expected cost savings of 29% with an 18-20 year earlier expected time-to-complete.
Optimization of an ISCO system at an Atlas missile site in Colorado indicated that increasing injected oxidant concentrations and the duration of annual oxidant injection periods predicted a decrease in the expected operating cost by 24% with a 90% probability of meeting the NFA date.
Although most DoD sites have or should soon have remedial action plans in progress, many will not achieve regulatory closure quickly. Some planned remedies will likely not perform as expected and will require modification or, in some cases, implementation of a different remedial action plan. This project provides tools to periodically assess remediation performance, identify and rectify problems, and optimize remediation operations and monitoring to minimize life cycle costs while meeting remediation objectives. By explicitly optimizing operations to minimize probability-weighted cost-to-complete taking into account uncertainty in site characterization, model predictions, and remediation technology performance, as well as measurement "noise," numerous nonlinear interactions and tradeoffs are taken into account that conventional approaches would never consider. Results indicate that average savings in cost-to-complete across all sites of 10% to 30% can be readily achieved along with substantial decreases in remediation duration.