Risk assessment models and uncertainty estimation

Risk assessment models and uncertainty estimation of groundwater contamination from point sourcesMads TroldborgPhD Thesis July 2010DTU Environment Department of Environmental Engineering Technical University of Denmark

PrefaceThis PhD thesis is based on research undertaken from October 2006 to May 2010 at the Department of Environmental Engineering, Technical University of Denmark (DTU). The work was done under the supervision of Associate Professor Philip J. Binning (primary supervisor) and Professor Poul L. Bjerg (cosupervisor). The work was funded through a PhD scholarship by DTU. The research was primarily carried out at DTU, but included two external stays at the Institute of Hydraulic Engineering, University of Stuttgart (Sept-Oct 2007, Sept-Dec 2008). The content of this thesis is based on three scientific journal papers and one conference paper. At the time of writing, two of the journal papers have been published and one is under revision: I. Troldborg, M., Lemming, G., Binning, P.J., Tuxen, N., Bjerg, P.L. (2008). Risk assessment and prioritisation of contaminated sites on the catchment scale. Journal of Contaminant Hydrology 101(1-4), 14-28. Troldborg, M., Binning, P.J., Nielsen, S., Kjeldsen, P. Christensen, A.G. (2009). Unsaturated zone leaching models for assessing risk to groundwater of contaminated sites. Journal of Contaminant Hydrology 105, 28-37.II.III. Troldborg, M., Nowak, W., Tuxen, N., Binning, P.J., Bjerg, P.L., Helmig, R. (2010). Uncertainty evaluation of mass discharge estimates from a contaminated site using a fully Bayesian framework. Water Resources Research (in revision). IV. Overheu, N., Troldborg, M., Tuxen, N., Flyvbjerg, J., Østergaard, H., Jensen, C.B., Binning, P.J., Bjerg, P.L. (2010). Concept for risk-based prioritisation of point sources. Proceedings of Groundwater Quality 2010. Zurich, Switzerland. In the thesis, these papers are referred to by author names and Roman numerals.i

The papers are not included in this web-version, but can be obtained from the library at Department of Environmental Engineering, DTU, Miljoevej, Building 113, DK-2800 Kgs. Lyngby, Denmark, library@env.dtu.dk. Additionally, the following reports and publications, related to the topic of the thesis, have been co-authored during the PhD-study, and will also be referred to in the thesis: Troldborg, M., Nowak, W., Binning, P.J., Bjerg, P.L., Helmig, R. (2010). Uncertainty of mass discharge estimates from contaminated sites using a fully Baysian framework. In: Managing Groundwater and the Environment. IAHS Red Book publication. ModelCARE2009 Wuhan, China. Christensen, A.G., Binning, P.J., Troldborg, M., Kjeldsen, P., Broholm, M. Upgrading JAGG to version 2.0 - Vertical transport to the first significant aquifer. Published by the Danish EPA, in press. (in Danish). Jørgensen, P.R., Klint, K.E., Troldborg, M., Binning, P.J. (2008). The JAGG model has been refined to cover risk assessments of aquifers below fractured clay till. VJ Blad. (in Danish) Tuxen, N., Larsen, L.C., Troldborg, M., Bjerg, P.L., Binning, P.J. (2008). Revised assessment of the existing investigations at Rundforbivej 176. Published by Orbicon A/S for the Capital Region of Denmark (in Danish)Kgs. Lyngby, May 2010 Mads Troldborgii

AcknowledgementsThere are a number of people I wish to thank for their help and contribution to this thesis. First of all, I would like to thank my supervisors, Associate Professor Philip J. Binning and Professor Poul L. Bjerg for their superb and competent guidance throughout the entire PhD You are both very inspiring and the main reason I ended up doing a PhD I owe great thanks to the entire staff at the Institute of Hydraulic Engineering, University of Stuttgart, for two pleasant and very educational stays. I am particularly grateful to Professor Rainer Helmig for his support, cheerful disposition and for funding my stays in Stuttgart, and to Junior-Professor Wolfgang Nowak for introducing me to the world of stochastic modelling and for constantly being available with outstanding guidance and advice. For great collaboration and co-authorship I would like to thank Nina Tuxen and Gitte Lemming with whom I also had the pleasure of sharing office. Thanks for many good discussions and your numerous valuable comments on my work. Also the co-authors Anders G. Christensen, Peter Kjeldsen, Signe Nielsen, and Niels Døssing Overheu are acknowledged for their contributions to my work. I also wish to thank Carsten Bagge Jensen, John Flyvbjerg and Henrik Østergaard from the Capital Region of Denmark for co-authorship and for providing the information on the cases studied in the thesis. The Technical University of Denmark and the Otto Mønsted Foundation are gratefully acknowledged for their financial support. The international research school of water resources (FIVA) is greatly appreciated for organising and financing a range of interesting PhD courses. I would also like to thank my colleagues and friends at DTU, especially the PhD club for cosy events and tasty cakes, Anne Harsting for always being kind and helpful, and Torben Dolin and Julie Camilla Middleton for helping out with graphics and illustrations. Last, but not least, I would like to thank my family and friends for all their encouragement. I would especially like to express my gratitude to my mum and dad, Niels, Søren and Jennifer for their unconditional love and support.iii

iv

AbstractA large number of contaminated sites are threatening groundwater resources worldwide. The expected costs for investigation and clean-up at these sites by far exceed the limited resources available. Regulators are therefore faced with the challenge of prioritising remediation efforts in order to ensure that the sites constituting the greatest risk to groundwater are cleaned up first. Risk assessment is therefore required. The conventional practice for assessing whether a point source poses a risk to groundwater has been focused on the measurement or prediction of contaminant concentrations at a local scale. If these concentrations do not comply with the contaminant-specific maximum concentration level (generic standards), the point source is considered a risk. The estimation of the concentration levels in groundwater is typically based on contaminant fate and transport modelling. A large number of modelling tools exist for local scale risk assessment. Most of these models are conceptually simple and designed to estimate the contaminant impact from a point source on groundwater using only few inputs and can therefore be applied to many different types of sites. A literature review has, however, revealed certain limitations of the existing tools for local scale risk assessment of groundwater contamination. It was found that most of the models ignore gas phase transport in the unsaturated zone, although this is known to be a dominant transport mechanism for volatile compounds such as chlorinated solvents and BTEXs. Analytical models accounting for both gas phase and water phase transport as well as sorption and first-order degradation were therefore developed to simulate the contaminant transport through the unsaturated zone. The models were tested on two field sites contaminated with volatile compounds and were found useful for practical risk assessment. The traditional risk assessment tools are also found less suitable for prioritisation purposes, because they consider the contaminated sites individually and focus only on predicting plume concentrations at a local scale. This is useful for assessing whether a particular contaminated site poses a risk to groundwater, but does not permit prioritisation of sites at larger scales. A modelling tool for risk assessment of contaminated sites on the catchment scale was therefore developed. This screening model evaluates the risk associated with each point source in terms of their ability to contaminate abstracted groundwater at thev

water supply in the catchment. It combines site-specific transient mass discharge calculations from all identified sites within the catchment with 3-dimensional catchment-scale contaminant fate and transport modelling. The tool was tested on the groundwater catchment for a water supply located in Denmark and was found to be valuable as a basis for prioritising point sources according to their impact on groundwater quality as well as for identifying the point sources that were most likely to cause the observed contamination at the water supply. Risk assessments of groundwater contamination are generally subject to data limitations and are therefore prone to large uncertainties. In order to ensure that resources are spent in the most cost-effective manner, it is important that these uncertainties are accounted for. In practice, however, the uncertainty is often not considered, but is handled based on a precautionary principle, where input parameter values are chosen so as to ensure conservative model outputs. The uncertainties in risk assessment of contaminated sites at local and catchment scale were investigated, particularly focusing on conceptual model uncertainties. An overview of available methods for evaluating the uncertainties in risk assessments was also provided. At local scale, the uncertainty related to the estimation of contaminant mass discharges from point sources was rigorously evaluated. Such estimates have many useful applications and constitute a key input for the catchment scale risk assessment. A methodology that uses multiple conceptual site models in a Bayesian inverse geostatistical framework was developed for quantifying the uncertainty in mass discharge estimates across multilevel control planes. The method generates an ensemble of flow and transport realisations that all honour the measured data at the control plane from which a mass discharge probability distribution is determined. The method was successfully applied to a TCE contaminated site for which four essentially different conceptual models based on two source zone models and two geological models were formulated. The method also provided a means of testing which of the conceptual site models were most likely to reflect the true site conditions. At catchment scale, the delineation of the capture zone is critical. The capture zone determines which contaminated sites are potential threats to the water supply as well as the travel times from the individual sites to the water supply. The uncertainty related to the catchment delineation depends on the groundwater model used. It was demonstrated how two different conceptual hydrogeological models could influence the location and extent of a capture zone and hereby affect the catchment scale risk assessment of a contaminated site.vi

The uncertainties at both local scale and catchment scale hamper the prioritisation of point sources and need to be accounted for to allow a more robust decision-making process. It was demonstrated how an assessment of the uncertainties related to both the mass discharge estimates and the capture zone delineation can be incorporated into the prioritisation using a scoring system. However, the influence of uncertainty on the prioritisation of contaminated sites should be the target for further research, possibly by including economical considerations. Better validation of the risk assessment models at both local scale and catchment scale should also be an issue for future research.vii

viii

SammenfatningAntallet af forurenede lokaliteter, der truer grundvandsressourcerne verden over, er enormt. Sammenlignet med de forventede udgifter til kortlægning og oprensning af de mange lokaliteter er de tilgængelige midler yderst begrænsede. Myndighederne står derfor overfor en udfordring med at få prioriteret oprydningsindsatsen således at de lokaliteter, der udgør den største grundvandsrisiko fjernes først. Risikovurderinger er derfor påkrævet. Traditionelt har vurderingen af hvorvidt en punktkilde udgør en grundvandsrisiko været fokuseret på målingen eller beregningen af forureningskoncentrationer på lokal skala. Hvis disse koncentrationer overstiger grundvandskvalitetskriteriet vurderes det, at punktkilden udgør en risiko. Estimeringen af koncentrationsniveauerne i grundvandet er typisk baseret på modellering af forureningstransporten. Der findes et stort antal modelværktøjer til risikovurdering på lokal skala. De fleste af disse værktøjer er konceptuelt simple og designet til at estimere forureningsbelastningen fra en punktkilde ud fra kun få input og kan derfor anvendes på mange forskellige typer forurenede grunde. Et litteraturstudium har imidlertid identificeret visse begrænsninger ved de eksisterende værktøjer til risikovurdering af grundvandsforurening på lokal skala. Det er blevet vist, at de fleste af modellerne ser bort fra gastransport i den umættede zone, til trods for at dette kan være en dominerede transportmekanisme for flygtige stoffer såsom klorerede opløsningsmidler og BTEX’er. Analytiske modeller, der inkluderer transport i både gas- og vandfasen samt sorption og førsteordens nedbrydning, er derfor blevet udviklet til at simulere forureningstransporten gennem umættet zone. Modellerne er blevet testet på to lokaliteter forurenet med flygtige stoffer og er på den baggrund blevet fundet anvendelige til risikovurdering. De traditionelle risikovurderingsværktøjer er også fundet mindre velegnede til prioriteringsformål, idet de betragter punktkilderne enkeltvist og kun fokuserer på beregning af koncentrationer i forureningsfanen på lokal skala. Dette er nyttigt til at vurdere hvorvidt en given forurenet lokalitet udgør en grundvandsrisiko, men tillader ikke en prioritering af lokaliteter på større skalaer. Et modelværktøj til risikovurdering af forurenede lokaliteter på oplandsskala er derfor blevet udviklet. Dette screeningsværktøj evaluerer risikoen forbundet med hver lokalitet i forhold til deres forureningspåvirkning af det grundvand, der indvindes ved vandforsyningen i oplandet. Værktøjet kombinerer site-specifikkeix

transiente forureningsfluxbestemmelser fra alle de identificerede lokaliteter indenfor oplandet med 3-dimensionel forureningstransport på oplandsskala. Modellen er blevet testet på grundvandsoplandet til et vandværk placeret i Danmark og er på den baggrund blevet fundet anvendelig til at prioritere punktkilderne ud fra deres forureningsbelastning af grundvandet samt til at identificere hvilke punktkilder, som var de mest sandsynlige årsager til den observerede forurening ved vandværket. Risikovurderinger af grundvandsforurening er generelt genstand for data begrænsninger og er derfor forbundet med store usikkerheder. For at sikre at de tilgængelige midler bruges på den mest omkostningseffektive måde, er det vigtigt at tage højde for disse usikkerheder. I praksis betragtes usikkerhederne ved risikovurderinger imidlertid ikke, men håndteres typisk ud fra et forsigtighedsprincip, hvor inputværdierne vælges således, at der sikres konservative model resultater. Usikkerhederne ved risikovurdering af forurenede lokaliteter på lokal og oplandsskala er blevet undersøgt med særligt fokus på de konceptuelle modelusikkerheder. En oversigt over de tilgængelige metoder til evaluering af usikkerhederne i risikovurdering er desuden blevet præsenteret. Usikkerhederne ved bestemmelse af forureningsfluxe fra punktkilder på lokal skala er blevet omfattende evalueret. Forureningsfluxbestemmelser har mange anvendelsesmuligheder og udgør et væsentligt input til risikovurderingen på oplandsskala. En metode, der benytter flere konceptuelle modeller for en given forurenet lokalitet i et Bayesiansk invers geostatistisk regi, er blevet udviklet til at bestemme usikkerhederne ved forureningsfluxbestemmelser gennem et multi-level kontrolplan. Metoden genererer et ensemble af flow og transportsimuleringer, der alle matcher de målte data i kontrolplanet, hvorudfra en sandsynlighedsfordeling af forureningsfluxen kan bestemmes. Metoden er med succes blevet anvendt på en TCE forurenet lokalitet, hvor fire forskellige konceptuelle modeller baseret på to forureningskildemodeller og to geologiske modeller var blevet opstillet. Metoden gav også mulighed for at teste, hvilken konceptuel model, der bedst repræsenterede de faktiske forhold ved lokaliteten. På oplandsskala er afgrænsningen af indvindingszone kritisk. Indvindingszonen afgør hvilke forurenede lokaliteter, der potentielt udgør en trussel for vandforsyningen samt hvad transporttiderne fra de individuelle lokaliteter til vandværket er. Usikkerhederne forbundet med afgrænsningen af oplandet afhænger af den anvendte grundvandsmodel. Det er blevet demonstreret, hvordan to forskellige konceptuelle hydrogeologiske modellerx

influerede på udbredelsen af indvindingszone og derved også påvirkede risikovurderingen af en forurenet lokalitet på oplandsskala. Usikkerhederne på både lokal og oplandsskala besværliggør prioriteringen af punktkilder og bør håndteres for at sikre en mere robust beslutningstagning. Det er blevet demonstreret hvordan en vurdering af usikkerhederne på både forureningsfluxbestemmelserne og på afgrænsningen af indvindingsoplandet kunne inkorporeres i en prioritering ved hjælp af et scoresystem. Indflydelsen af usikkerheder på prioritering af forurenede lokaliteter bør dog være genstand for yderligere undersøgelser, eventuelt ved at inkludere økonomiske betragtninger. En bedre validering af risikovurderingsmodellerne på både lokal og oplandsskala bør også være et emne for fremtidig forskning.xi

xii

Table of contents1 INTRODUCTION ....................................................................................................... 1 1.1 BACKGROUND AND MOTIVATION ............................................................................. 1 1.2 OBJECTIVES ............................................................................................................. 4 1.3 FRAMING OF THESIS ................................................................................................. 4 1.4 ORGANISATION OF THESIS ....................................................................................... 5 2 RISK ASSESSMENT – DEFINITIONS AND TERMINOLOGY ......................... 7 2.1 GENERAL DEFINITIONS ............................................................................................ 7 2.2 RISK ASSESSMENT OF CONTAMINATED SITES ........................................................... 7 3 RISK ASSESSMENT AT THE LOCAL SCALE .................................................. 13 3.1 GENERIC STANDARDS ............................................................................................ 14 3.2. RISK ASSESSMENT AT DIFFERENT KNOWLEDGE LEVELS ........................................ 15 3.3 MASS DISCHARGE .................................................................................................. 18 3.4 RISK ASSESSMENT TOOLS....................................................................................... 21 3.4 FINDINGS FOR LOCAL SCALE RISK ASSESSMENT ..................................................... 33 4 RISK ASSESSMENT AT THE CATCHMENT SCALE ...................................... 35 4.1. CATCHMENT-SCALE RISK ASSESSMENT APPROACHES ........................................... 36 4.2 FINDINGS FOR CATCHMENT SCALE RISK ASSESSMENT ............................................ 44 5 UNCERTAINTIES IN RISK ASSESSMENTS ...................................................... 45 5.1 UNCERTAINTY TERMINOLOGY AND DEFINITIONS ................................................... 45 5.2 METHODS FOR QUANTIFYING UNCERTAINTY IN RISK ASSESSMENTS....................... 51 5.3 MASS DISCHARGE UNCERTAINTY ........................................................................... 54 5.4 UNCERTAINTIES AT THE CATCHMENT SCALE ......................................................... 57 5.5 FINDINGS FOR UNCERTAINTIES IN RISK ASSESSMENT ............................................. 62 6 CONCLUSIONS AND PERSPECTIVES ............................................................... 63 6.1. FUTURE RESEARCH DIRECTIONS ............................................................................ 65 7 REFERENCES .......................................................................................................... 67 8 APPENDICES............................................................................................................ 89xiii

xiv

1 Introduction1.1 Background and motivationContaminated sites are a significant threat to groundwater resources worldwide. The European Environment Agency estimates that there may be as many as 3 million contaminated sites across the EU, of which about 250 000 sites require clean up (EC, 2006; EEA, 2007). The costs for investigation and clean-up of these sites are excessive and compared to the scope of the problem, the available resources for investigating and cleaning up contaminated sites are very limited. Thus, though considerable efforts have already been made, it will still take decades to clean up the legacy of contamination (EEA, 2007). It is therefore crucial that the resources are allocated to the contaminated sites posing the greatest risk. In order to decide which sites should be given the highest priority and to streamline future activities, risk assessment is a very important and useful tool (Cushman et al., 2001; Ferguson et al., 1998). Contaminated sites are complex environmental problems that require insight into various scientific disciplines such as geology, hydrology, chemistry, microbiology and toxicology. Risk assessments of contaminated sites are useful tools for assembling the concepts and knowledge from these different disciplines in a transparent and scientifically sound way and can thus, if applied sensibly, offer valuable results for decision-making (Ferguson et al., 1998). 1.1.1 Risk assessment at different scales Various approaches exist for assessing whether a contaminated site constitutes a risk to groundwater (e.g. Spence, 2001; Aziz et al., 2000; Davison and Hall, 2003; Newell et al., 1996). Most of these methods focus on a local scale and aim to evaluate if the resulting groundwater concentrations below or downstream of the contaminant source zone are above a certain limit value. If the concentrations do not comply with the regulatory standards the contaminated site is considered a risk. The resulting concentrations can in some case be measured directly, but often need to be calculated from site-specific information regarding released amounts of contaminants, type of contaminant, geology, hydrogeology etc. For the last decade this approach has been common practice in many countries, including Denmark (Bardos et al., 2002). However, the prioritisation of point sources necessitates that the risk is considered not only on the local scale, but also on larger scales. For an initial1

prioritisation of contaminated land aquifer vulnerability mapping methods such as DRASTIC (Aller et al., 1987) are widely used. These methods assign scores to different spatially distributed indicators (e.g. top-layer geology, depth to groundwater, recharge and likely types of contaminants spilled), which subsequently are integrated into an overall risk index. Vulnerability mapping helps identifying the areas most susceptible to contamination, but does not account for the degree and extent of contamination at actual sites. Vulnerability mapping can therefore not be used for a more detailed prioritisation of point sources and to identify at which sites remediation should be initiated. Since the motivation for initiating clean-up is often governed by the possible impact on water supply wells, it has been proposed to conduct risk assessments at catchment scale (e.g. Einarson and Mackay, 2001; Frind et al., 2006), where the risk of a point source is assessed in terms of its ability to contaminate abstracted water at the supply wells in the catchment. In this context the estimation of contaminant mass discharges from the individual point sources within the catchment has been found valuable (Einarson and Mackay, 2001), because such estimates are dynamic measures of the total contaminant impact. 1.1.2 Risk assessment models Risk assessments often involve modelling of contaminant fate and transport. The transport and fate of contaminants in the subsurface is, however, complex and depends on a range of physical, chemical and biological factors that may be subject to great variations both spatially and temporally and are affected by both the contaminant properties and the site-specific settings. It is important to account for these various processes in a risk assessment as they have the potential to affect, for example, the mobility and toxicity of the contaminant (Ferguson et al., 1998). In order to model the complex real-world system several simplifications have to be made. These simplifications should account for the most important site-specific features and dominant processes. Many models for assessing the impact from a contaminated site on groundwater exist and range from simple screening tools to more advanced numerical transport models (ASTM, 1999; Walden, 2005). There is a huge difference in the complexity of these models depending on the specific purpose of the model, the number of processes allowed for, how the processes and site-features are represented mathematically etc. As risk assessments are often subject to data limitations, they usually rely on analytical models, which have minimal data requirements and thus are2

applicable even when only basic site data are available. Although analytical models do not allow for the same level of detail and knowledge as numerical models, they are more straightforward to use and are easily applied to many different types of sites. However, it is essential that the risk assessment models do not overly simplify the system in concern, which is very often the case. Not allowing for dominant processes or important site features can lead to substantial under- or overestimations of the risk. For example, neglecting the effect of degradation of biodegradable compounds, the production of metabolites during sequential degradation, preferential flow paths in fractured clay and/or gaseous diffusion in the unsaturated zone for volatile compounds can potentially result in a very misleading risk assessment. Many screening models for risk assessment of groundwater contamination overly simplify the fate and transport in the subsurface. There is still a need for the development of improved risk assessment tools in order to handle the vast number of point sources worldwide. These tools should be able to deal with many different types of contaminants and allow for as many processes and site-specific features as possible, but at the same time not become too data demanding. 1.1.3 Uncertainties in risk assessments Risk assessments are associated with significant uncertainties. The uncertainties are caused by several factors such as errors in the conceptual model, parameter uncertainty, and errors in applied model algorithm and in data used as input (Beven, 2005; Højberg and Refsgaard, 2005; Walker et al., 2003). It is important that the uncertainties related to risk assessments of groundwater contamination are taken into consideration to ensure that resources are spent in the most cost-effective manner. An evaluation of the uncertainties determines the reliability of the risk assessment and thus helps clarifying whether more investigations are needed or if action can be initiated. However, uncertainties related to risk assessments of groundwater contamination are often not given much attention. Most of the current risk assessment tools do not include uncertainty considerations at all, and those that do, only take input and/or parameter uncertainties into account. Methodologies for the handling of parameter uncertainty are many, whereas little is known about how to deal with lack of conceptual understanding. This is despite the fact that conceptual uncertainties have been recognised as the most significant sources of error (Konikow and Bredehoeft, 1992; Refsgaard et al., 2006). Thus, there is a need for more systemic research about the influence of uncertainties on risk3

本文来源：https://www.bwwdw.com/article/3qoi.html