Protecting groundwater from contamination is an important issue which has been backed up with legislation over the past decade. The Water Resources Act 1991 and Environment Act 1995 outline the duties of the Environment Agency to monitor, protect and enhance the environment, including water resources.
The EC Groundwater Directive (80/68/EEC) requires member states to prevent the introduction of List I substances into groundwater and to restrict the input of List II substances to prevent groundwater pollution.
(See the Department of the Environment Food & Rural Affairs website www. defra. gov. uk/environment/water/ground/guidance. htm).
Part IIa of the Environmental Protection Act requires local authorities to inspect land in their area and to identify contaminated land. This has led to more contaminated land being identified and increased dependence on risk analyses to protect human health and the environment.
In 1999, in an attempt to standardise techniques used in the assessment of soil and groundwater contamination, the Environment Agency issued R&D Publication 20, outlining a methodology for deriving remedial targets for soil and groundwater that are protective of water resources. It outlines a tiered approach where progressively more sophisticated (and less conservative) analyses are made as further data is collected and the site conceptual model is refined.
Many tools are available to aid the implementation of the tiered approach to water resource protection, ranging from the simplest spreadsheet to complex numerical models with large associated data needs.The choice of technique generally comes down to data and time constraints, the project objectives and the sensitivity of the problem.
The simple spreadsheet provided with R&D Publication 20 takes a deterministic approach to predicting contaminant concentrations at the water table or down gradient receptor. Single values are entered for each parameter and single results gained for each prediction. This approach can be difficult to justify in many cases as it does not consider the issues of parameter uncertainty, originating from both natural variability, measurement error and the use of generic values. The supporting user manual recognises this and recommends that sensitivity analysis is undertaken as an integral part of the assessment.
Commercial analytical risk models such as ConSim move ahead of the simple spreadsheet by encompassing more active processes and being based on probabilistic techniques. ConSim's probabilistic approach allows the user to specify a probability density function (PDF) that describes the uncertainty and variability of each input, and results are shown as concentration versus probability charts. ConSim is designed for tiers 1, 2 and 3 soil analyses, tiers 2 and 3 groundwater analyses and for combined soil and groundwater analyses.
ConSim is an ideal tool for assessing the risk to groundwater from contaminated soils, but the user needs to have confidence in its ability to simulate real sites. The report on assigning values to uncertain parameters in subsurface fate and transport modelling (Environment Agency 2001) states that recognising and understanding the uncertainty in fate and transport models is a key issue and involves three areas: conceptual uncertainty, model uncertainty and parameter uncertainty.
Much validation has been published on numerical approaches to groundwater modelling but little on the popular probabilistic risk models.The results from a validation exercise where field data is compared with the results from a numerical model and from ConSim is presented here, in an attempt to start to address the issue of model uncertainty.
Rexco coal carbonisation plant is a well documented contaminated site near Mansfield in Nottinghamshire (Figure 1). It has been extensively characterised with 42 monitoring boreholes and 112 monitoring points generated from multilevel samplers (Figure 2). A detailed numerical flow and solute transport model has been generated for the site and numerous analyses made to understand the active processes (Davison and Lerner, 2000, Broholm and Arvin, 2000, Jones et al, 1998).
The source of contamination was found to be ammonium liquor, a byproduct from the coal carbonisation process.While Rexco was producing coke the ammonium liquor was channelled along a drainage ditch to a small lagoon. The liquor infiltrated to groundwater through permeable parts of the drainage ditch and through the lagoon itself.
The main components of the ammonium liquor are ammonium, with an estimated concentration of 12800mg/l, and phenol, with an estimated concentration of 7700mg/l (Broholm et al, 1998).
Following investigation, the site was found to be contaminated with ammonium but no significant phenol was identified. It is believed that the higher retardation of ammonium relative to phenol has led to the chromatographic separation of the two plumes and hence the absence of phenol. Studies have also shown the potential for the degradation of phenol at the Rexco site (Broholm and Arvin, 2000), which could contribute to the absence of phenol from the monitoring boreholes.
Rexco is an example of a complex contaminated site in terms of flow, the source term and geochemistry.The industrial evolution of the site, with its associated alterations to the pumping requirements, led to four significant changes in the groundwater flow direction over the past 15 years (Davison and Lerner, 1999). One of the pumping schedule alterations resulted in a 90infinity change in local groundwater flow direction.
Understanding the source term is problematic, as the release of pollutants from the industrial process ended more than 30 years ago. Consequently, no direct evidence of its characteristics exist.
There are many examples of the site's complex geochemistry. One is the interaction between the source components and the environment with the product of the oxidation of ammonium (ie nitrate) being used as an electron acceptor during the biological degradation of phenol.
It is contentious whether probabilistic models can truly be validated, as probability versus concentration curves cannot be generated from field data for a single site.However, to test the ability of ConSim to predict concentrations at down-gradient receptors, a comparison was made between the site monitoring data, the ConSim predictions, and the MT3D and Modflow numerical modelling. Simulating this site data is a challenging test for ConSim given the complex system and extensive field data. The conceptual model for this simulation involves a continuous leachate source of phenol and ammonium passing through the unsaturated zone by plug flow.Time of travel to the water table is retarded using a partition coefficient (K d). It is assumed there is no degradation of ammonium in the unsaturated zone, and therefore the concentration at the base of the unsaturated zone eventually equals the input concentration.
The phenol, however, was allowed to degrade within the unsaturated zone. The groundwater flowing beneath the site dilutes the concentrations of phenol and ammonium, and concentrations at monitoring boreholes are calculated based on an extended version of the Domenico advection-dispersion equation (Figure 3).
Inputs for Rexco were entered in the ConSim environment mostly as PDFs covering the parameter uncertainty and spatial variability.Where possible the PDFs were established on field data but some were based on expert judgement as in many site investigations. The exact inputs to the simulations are extensive and will not be reproduced here.
A Monte Carlo analysis with 501 runs was completed. The first set of results from ConSim closely fitted the monitoring data along the flow line from the contaminant source. However, contaminant concentrations in monitoring wells distant from this central flow line were consistently underestimated.
The problem was identified as a consequence of the changing groundwater flow directions which had smeared the plume across the field site. As ConSim does not attempt to account for transient conditions an increased transverse dispersivity was chosen. A uniform distribution of transverse dispersivity with a range from 1m to 10m was selected and this produced a good match with all the field data (total pathway length is approximately 600m).
Figure 4 shows the level of similarity between the observed ammonium concentrations and those produced from ConSim. As ConSim is a probabilistic model, a single output should not be compared with the field data. Figure 5 shows the range of ammonium concentrations at each of the monitoring boreholes from the 50th to the 95th percentile.All the field observed concentrations lie in the range predicted by the model and the concentration trends match.
The sensitivity of a model result to its input parameters is very revealing. A detailed sensitivity analysis was made of the deterministic numerical flow and solute transport model. The results showed the source flux to be the most sensitive parameter affecting the concentrations observed in monitoring boreholes around the site. In ConSim the most sensitive parameter was found to be the infiltration rate, which is a key part of the source flux term in MT3D. This, combined with ConSim's ability to reproduce concentrations at the site, adds considerable confidence to its ability to robustly simulate contaminant concentrations, where the basic conceptual model in ConSim fits the site.
The exercise showed that ConSim overestimated the concentrations near to the source at later times.
This is due to the assumption of a constant source through time, whereas in reality this source has been in decline since the surface input ceased in 1970. Since this conservative assumption has been identified as a limitation in realistically representing site data, a new version of ConSim is under development for the Environment Agency that will include a declining source term.
Applying a half-life derived from literature, ConSim predicts that all of the phenol degrades in the unsaturated zone before it reaches the water table. Studies on the Sherwood Sandstone from the Rexco site have shown that concentrations of phenol above a few hundred milligrams per litre are toxic to bacteria and therefore very little biodegradation would in reality have taken place as the source contained several thousand milligrams per litre (Broholm and Arvin 2000).
This issue should be tackled by the user who should check whether input concentrations are likely to be toxic to bacteria and whether other conditions, such as electron acceptor availability, pH, substrate availability and nutrient availability are all conducive to biological degradation. Site-specific half-lives should be used whenever possible and careful analyses of the site conditions made. If data is not available it would be prudent to assume biodegradation does not occur.
Overall, ConSim provided a good indication of likely contaminant concentrations both spatially and temporally, and this can only serve to increase confidence in the use of such models. The sensitivity of the model output to various input parameters should be tested to ensure that site specific data collection can be focused on the most important parameters, and that the level of uncertainty in the results can be minimised.
Where the conceptual model in ConSim fits the conditions at the subject site, and with careful choice of input PDFs, it is an ideal tool for undertaking lower tiered risk analyses. For high risk sites and for the design of complex remedial systems it may be necessary and more cost-effective to develop a full scale numeric model to gain a better understanding of the controlling processes, investigate transient effects and model different remedial scenarios.
Acknowledgements The development of ConSim is funded by the Environment Agency and Golder Associates. The Rexco study was sponsored by the EU under contract no EV5V-CT94-0529 and by the Environment Agency and was carried out under the direction of David Lerner, University of Sheffield, to whom the authors are grateful for the use of data.The authors would also like to thank Jonathan Smith of the EA for reviewing the text.
Broholm MM and Arvin E (2000). Biodegradation of phenols in a sandstone aquifer under aerobic conditions and mixed nitrate and iron reducing conditions. Journal of Contaminant Hydrology.
Broholm MM, Jones I, Tortensson D & Arvin E (1998). Groundwater contamination from a coal carbonisation plant. In: Lerner DN & Walton NRG (eds), Contaminated land and groundwater: Future directions. Geological Society, London, Engineering Geology Special Publication, 14, 159-165.
Davison RM and Lerner DN (1999). Evaluating natural attenuation of groundwater pollution from a coal carbonisation plant: developing a local scale model using Modflow, MODTMR and MT3D. Water and Environmental Management Journal, Vol 14, No 6 pp 419-426 Davison RM and Lerner DN (2000). Proving natural attenuation at a site contaminated by ammonium and phenolics. In Sililo et al (eds), Groundwater: Past achievements and future challenges. Cape Town. Balkema, Rotterdam. 737-740 Environment Agency (2001). Guidance on assigning values to uncertain parameters in subsurface fate and transport modelling. National Groundwater and Contaminated Land Centre Report NC/99/38/3.
Environment Agency (1999). Methodology for the derivation of remedial targets for soil and groundwater to protect water resources.
Environment Agency R&D Publication 20.
Environment Agency (1999). ConSim: Contamination impact on groundwater: Simulation by Monte Carlo method. Environment Agency R&D Project P2-007. Prepared by Golder Associates, Nottingham.
Jones I, Davison RM and Lerner DN 1998. The importance of understanding groundwater flow history in assessing present day groundwater contamination patterns; a case study. In Lerner DN and Walton NRG (eds), Contaminated land and groundwater: Future directions.
Geological Society, London, Engineering Geology Special Publication 14, 159-165.