Propagation of water pollution uncertainty and risk from the subsurface to the surface water system of a catchment

Summary

This paper investigates the propagation of quantifiable probability and quantification uncertainty of water pollution from local pollutant sources at and below the land surface, through the groundwater system, to downstream surface water recipients. Methodologically, the study shows how the risk and uncertainty of surface water pollution within a catchment may be assessed by a combined methodology of a Lagrangian stochastic advective-reactive modelling approach, which accounts for the quantifiable pollutant transport randomness, and a scenario analysis approach, which accounts for different quantification uncertainties. The results show that, in general, unambiguous risk assessment requires at least a reliable order-of-magnitude quantification of the prevailing relation between the average rate of physical pollutant transport from source to recipient and the average rate of pollutant attenuation. If this average relation can be reliably estimated to fall within two identified, relatively wide open value ranges, the assessment of pollution risk to surface waters from localised sources at or below the soil surface may be unambiguous even under otherwise large quantification uncertainty. For a relatively narrow, closed value range of this average rate relation, however, risk assessment must either rely on conservative assumptions, or else be based on a more detailed and resource demanding quantification of pollutant transport.

Introduction

Various contaminated land sites and groundwater pollutant plumes may be located within the catchment areas of sensitive water recipients, such as drinking water supplies, lakes, streams and coastal waters. The pollutant releases from such sites and plumes, and from possible future releases in the catchments (for instance from planned or present industrial and transportation activities and pollution accidents), pose pollution risks to the downstream surface water environments. These risks need to be assessed, for instance according to the EU Water Framework Directive (WFD; European Commission, 2000), which requires catchment-scale water management for achieving and maintaining good physico-chemical and ecological status in all the waters of the EU member states.

To assess the water pollution risks posed by present and potential future pollutant releases within a catchment area, we need to model the pollutant transport and mass exchange that take place along the transport pathways from the sources to the water recipients within and downstream of the considered catchment area. However, model predictions are associated with uncertainty that needs to be accounted for in this catchment-scale risk assessment. Most model parameters cannot be precisely determined due to large and irregular variability in time and space combined with measurement errors, a general lack of data and site-specific information, and uncertainty about whether the model itself constitutes an adequate mathematical representation of the real-world pollutant transport process (e.g. Kavanaugh et al., 2003, O’Hagan and Oakley, 2004, Beven, 2006, Baresel and Destouni, 2007).

Numerous studies have in the past decades developed stochastic modelling approaches to account for the physical spreading effect of random spatial aquifer heterogeneity on subsurface solute transport (see for instance Dagan, 1989 and Rubin, 2003) for reviews of different approaches to stochastic solute transport modelling in randomly heterogeneous formations). Some of these approaches use probability density functions (pdfs) of advective solute travel times as a basis for deriving the statistics of solute concentrations and mass flows. Such travel time-based, Lagrangian stochastic approaches have been applied to solute transport through subsurface water systems (soil water, groundwater; e.g. Shapiro and Cvetkovic, 1988, Destouni, 1992, Cvetkovic and Dagan, 1994, Destouni and Graham, 1995, Simmons et al., 1995, Yabusaki et al., 1998, Foussereau et al., 2001, Tompson et al., 2002, Malmström et al., 2004) and catchments (Simic and Destouni, 1999, Lindgren et al., 2004, Lindgren and Destouni, 2004, Botter et al., 2005). The travel time pdfs that must be quantified in these approaches are commonly unknown and approximated by assuming some common pdf type (e.g. lognormal, inverse Gaussian) based only on knowledge about the travel time mean and variance (e.g. Cvetkovic et al., 1998, Destouni et al., 2001). The physical solute travel time statistics can further be coupled with relevant pollutant reaction models. This coupled Lagrangian stochastic advective-reactive (LaSAR) modelling approach can be used for quantifying the transport of reactive pollutants in terms of both expected transport (e.g. Destouni and Graham, 1995, Lindgren et al., 2004, Lindgren and Destouni, 2004, Malmström et al., 2004) and transport variance (e.g. Destouni, 1992, Andricevic and Cvetkovic, 1996, Destouni and Graham, 1997, Andersson and Destouni, 2001, Baresel and Destouni, 2007). The expected transport accounts then for the physical solute spreading effect of aquifer heterogeneity in a statistically quantifiable population of random heterogeneity outcomes, and the transport variance is a measure of the uncertainty implied by this randomness with regard to the actual field outcome from the whole statistical population.

However, pollutant transport uncertainty does not only result from the statistically quantifiable randomness of aquifer heterogeneity. Even if the mean and variance of solute travel time can be quantified from some, relatively readily obtainable field statistics of measurable groundwater hydraulics (e.g. Destouni and Graham, 1997, Simic and Destouni, 1999; Destouni et al., 2001), the further derivation of solute concentration and mass flux statistics requires more field knowledge, for example about the spatial correlation structure of groundwater hydraulics, which is more difficult and resource demanding to assess than independent population statistics. Furthermore, also other essential factors, such as pollutant release from the source zone and biogeochemical reaction rates, may be quite difficult and resource demanding to measure in the field. Our uncertainty about all these factors leads to additional pollutant transport uncertainty to that from the statistically quantifiable randomness in aquifer heterogeneity.

In this paper, we develop a combined methodology for quantifying the propagation of both the statistically quantifiable probability and the quantification uncertainty of water pollution through the groundwater system to downstream surface water recipients. The methodology combines a LaSAR modelling approach, which accounts for statistically quantifiable randomness in aquifer heterogeneity, with a scenario analysis approach, which accounts for quantification uncertainty about: (1) the spatial correlation structure of aquifer heterogeneity, (2) the pollutant release from the source zone, and (3) the biogeochemical pollutant attenuation rate. This combined approach is used for investigating the main effects of these different types of quantification uncertainties and statistically quantifiable randomness on the probability of exceeding a given, environmental or health-based, pollutant concentration limit at a surface water recipient boundary downstream of a short-term accidental (e.g. resulting from accidents involving dangerous goods), or a long-term continuous (e.g. resulting from contaminated land sites) pollutant release. Furthermore, the practical usefulness of this approach is finally also demonstrated for assessments of surface water pollution risk posed by existing or possible future local source releases within a catchment area.