Optimisation of permeable reactive barrier systems for the remediation of contaminated groundwater

Permeable reactive barriers (PRBs) are one of the leading technologies being developed in the search for alternatives to the pump-and-treat method for the remediation of contaminated groundwater. A new optimising design methodology is proposed to aid decision-makers in finding minimum cost PRB designs for remediation problems in the presence of input uncertainty. The unique aspects of the proposed methodology are considered to be: design enhancements to improve the hydraulic performance of PRB systems; elimination of a time-consuming simulation model by determination of approximating functions relating design variables and performance measures for fully penetrating PRB systems; a versatile, spreadsheet-based optimisation model that locates minimum cost PRB designs using Excel's standard non-linear solver; and the incorporation of realistic input variability and uncertainty into the optimisation process via sensitivity analysis, scenario analysis and factorial analysis. The design methodology is developed in the context of the remediation of nitrate contamination due to current concerns with nitrate in New Zealand. Three-dimensional computer modelling identified significant variation in capture and residence time, caused by up-gradient funnels and/or a gate hydraulic conductivity that is significantly different from the surrounding aquifer. The unique design enhancements to control this variation are considered to be the customised down-gradient gate face and emplacement of funnels and side walls deeper than the gate. The use of velocity equalisation walls and manipulation of a PRB's hydraulic conductivity within certain bounds were also found to provide some control over variation in capture and residence time. Accurate functional relationships between PRB design variables and PRB performance measures were shown to be achievable for fully penetrating systems. The chosen design variables were gate length, gate width, funnel width and the reactive material proportion. The chosen performance measures were edge residence, centreline residence and capture width. A method for laboratory characterisation of reactive and non-reactive material combinations was shown to produce data points that could realistically be part of smooth polynomial interpolation functions. The use of smooth approximating functions to characterise PRB inputs and determine PRB performance enabled the creation of an efficient spreadsheet model that ran more quickly and accurately with Excel's standard non-linear solver than with the LGO global solver or Evolver genetic-algorithm based solver. The PRB optimisation model will run on a standard computer and only takes a couple of minutes per optimisation run. Significant variation is expected in inputs to PRB design, particularly in aquifer and plume characteristics. Not all of this variation is quantifiable without significant expenditure. Stochastic models that include parameter variability have historically been difficult to apply to realistic remediation design due to their size and complexity. Scenario and factorial analysis are proposed as an efficient alternative for quantifying the effects of input variability on optimal PRB design. Scenario analysis is especially recommended when high quality input information is available and variation is not expected in many input parameters. Factorial analysis is recommended for most other situations as it separates out the effects of multiple input parameters at multiple levels without an excessive number of experimental runs.