Introduction and Objectives
Environmental pollution of toxic substances has led governments to develop new laws and regulation that puts constraints on these chemical emissions. These are based on environmental quality standards and environmental/ecological risk assessment. The key question to be answered is: "What is the likelihood (i.e. probability) of adverse effects occurring to exposed ecological systems due to exceedance of a toxicity level by an environmental concentration?". The goal of ecological risk assessment is to estimate this likelihood. It is based on the comparison of a predicted or measured exposure/environmental concentration with a 'no effect concentration' based on a set of (acute or chronic) toxicity test results (i.e. testing species sensitivity).
This PhD dissertation studied and developed a range of statistical techniques needed to answer the key question with a risk probability and an uncertainty or confidence interval rather than with the current black or white "yes, there is potential risk / no risk" answer which the conventional risk assessments provide. Such answers may mislead stakeholders to think that ecological risks are simple black and white issues.
After all, in a Probabilistic Ecological Risk Assessment (PERA), the exposure concentration and species sensitivity are treated as random variables taken from probability distributions (respectively Exposure Concentration Distribution (ECD) and Species Sensitivity Distribution (SSD)) which are combined to give a risk probability (see figure below). In this way, the inherent variability and uncertainty of the environmental concentration and the species sensitivity is accounted for. Variability represents inherent heterogeneity or diversity in a wellcharacterised population. Uncertainty represents partial ignorance or lack of perfect information about poorly characterised phenomena or models (e.g. sampling or measurement error). PERA therefore delivers a more transparent, realistic and nonconservative approach to estimate risks. It is recognised in literature that probabilistic methods would improve the environmental evaluation of chemicals, if appropriate action is taken to address their potential weaknesses.
Some of these current (mainly statistical) weaknesses in probabilistic ecological risk assessment are addressed in this PhD dissertation. Most of them deal with misuse of existing techniques (e.g. Monte Carlo analysis, bootstrap), reliability of statistical techniques at small sample size, the lack of consensus on which method or model or what sample size to use, misinterpretation of probability distributions (e.g. output of Monte Carlo analysis), inappropriately or insufficiently dealing with uncertainty or variability (e.g. one versus twodimensional Monte Carlo analysis), discussions on how to calculate probabilistic risk...
Results and Discussion
It was shown that interpretation of all probability distributions in a PERA framework should be made carefully. In Monte Carlo simulations, separation of uncertainty and variability and the correct application of Monte Carlo analysis simplify the interpretation of a model's output distribution of interest. A case study showed that the NOEC (No Observed Effect Concentration) of Cu will be larger than 75 mg/l for 80% of the time for a lake in Sweden. This is a quite different result than being 80% certain that the NOEC will be larger than 75 mg/l.
A probabilistic risk should be interpreted as the probability that a random selected exposure or environmental concentration will exceed a species sensitivity. Examples have shown that the same risk probability can represent different environmental conditions (e.g. depending on whether the ECD represents spatial or temporal variability). Therefore, it is suggested to always include as much information as possible in the answer to the key question described above: indicate what type of variability the ECD or the SSD represents (geo or timereferenced), what endpoint was used.
Throughout this dissertation, parametric and nonparametric methods were often used in parallel. Results are very sensitive to the choice of the method. Consequently, the importance of a proper use of distribution selection methods should not be underestimated. Statistical tests, graphical exploration and expert knowledge can help in identifying the appropriate distribution. For calculating a lower percentile (as the 5th percentile), it was found that preference should be given to parametric methods when the sample size is below 10 and preference should be given to nonparametric methods when the sample size is very large (e.g. 50). For the intermediate sample sizes, either parametric or nonparametric techniques can be used or maybe a combination of the two could be used.
Several examples and case studies have proven that the probabilistic risk characterisation considers the quantitative information of the full range of the ECD and SSD (including lower SS than its 5th percentile and higher ECs than the 90th percentile) instead of only considering the upper tail of the ECD and the lower tail of the SSD. Consequently, several issues on calculating tail percentiles can be omitted because they are no longer needed in the risk characterisation.
In addition, the probability distributions in probabilistic risk assessment can be wide due to large spatial (and temporal) variability (wide distributions). This can lead to conservative assessments having a higher probability of a large risk. Instead of lumping all the sources of variability into one probability distribution, spatial and/or temporal differences could be explicitly accounted for in a respectively geo and/or timereferenced analysis (or spatialtemporal analysis). Several case studies showed that georeferencing makes the risk assessment more realistic. In one case study, the probabilistic risk was a factor 3 lower than the nongeorisk because geographical information was not accounted for.
Publications
My PhDthesis: Verdonck F.A.M. (2003). Georeferenced probabilistic ecological risk assessment. Ghent University, Faculty of Applied Biological Sciences, Gent, Belgium, 200p. Go to thesis.
Verdonck F.A.M., Jaworska J., Janssen C.R. & Vanrolleghem P.A. (2003) Geography Referencing Probabilistic Risk Of Chemicals In Rivers. Water Sience and Technology, 48(3), 3946.Go to paper
Verdonck F.A.M., Aldenberg, T., Jaworska, J. & Vanrolleghem P.A. (2003) Limitations of current risk characterization methods in probabilistic ecological risk assessment. Environmental Toxicology and Chemistry, 22 (9), 22092213. Go to paper.
Verdonck F.A.M., Jaworska J., Janssen C.R. & Vanrolleghem P.A. (2002) Probabilistic environmental risk assessment framework for chemical substances. In: Proceedings iEMSs 2002, Integrated Assessment and Decision Support, 2427 June 2002, Lugano, Switzerland.
Verdonck F.A.M. & Vanrolleghem P.A. (2002) (in dutch) Milieurisicoanalyse in een onzekere wereld. Het ingenieursblad, 89, 4347.
Verdonck F.A.M., Janssen C.R., Thas O., Jaworska J. & Vanrolleghem P.A. (2001) Probabilistic environmental risk assessment. PhD symposium, Med. Fac. Landbouww. Univ. Gent, 66(4), 1319. 10 oktober 2001, Gent, Belgium.
Verdonck F., Jaworska J. & Vanrolleghem P.A. (2001) Comments on species sensitivity distributions sample size determination based on Newman et al. (2001). SETAC Globe 2(3), Learned Discourse, 2224. Go to paper.
Verdonck F.A.M., Jaworska J., Thas O. & Vanrolleghem P.A. (2001) Determining safe environmental standards using bootstrapping, Bayesian and maximum likelihood techniques: a comparative study. Analytica Chimica Acta, 446 (12), 429438 and presented at CAC2000, 7th International Conference on Chemometrics in Analytical Chemistry, 1620 October 2000, Antwerp, Belgium. Go to abstract. Go to paper.
Verdonck F., Jaworska J., Thas O. & Vanrolleghem P.A. (2000) Uncertainty techniques in environmental risk assessment. Med. Fac. Landbouw. Univ. Gent, 65/4, 2000, 247252. Presented as a poster on the PhD symposium, October 11, 2000, Gent, Belgium. Go to paper.
Oral presentations
Verdonck F.A.M., Rousseau D., Bixio D., Thoeye C. & Vanrolleghem P.A. (accepted). Added value of concentrationdurationfrequency curves of wastewater plant effluent quality. International Congress on Modelling and Simulation, 1317 July 2003, Townsville, Australia.
Verdonck F.A.M., Deksissa T., De Laender F., De Schamphelaere K.A.C., Matamoros D., Vandenberghe, V., Vincke S., Janssen, C.J. & Vanrolleghem P.A. (2003) Uncertainty and variability in spatiotemporal probabilistic risk modelling. 13th Annual Meeting of SETACEurope, 27 April  1 May 2003, Hamburg, Germany.
Verdonck FA.M., Janssen C.R. & Vanrolleghem P.A. (2002) Potential of ecological informatics in probabilistic risk assessment of chemicals in rivers. Proceedings Ecological informatics applications in water management conference, 18, 67 November 2002, Gent, Belgium.
Verdonck F.A.M., Jaworska J., Janssen C.R. & Vanrolleghem P.A. (2002) Geography Referencing Probabilistic Risk Of Chemicals In Rivers. In: Proceedings 1st IWA Young Researchers Conference, 3340, 910 September 2002, Cranfield, UK.
Vanrolleghem P.A. & Verdonck F.A.M. (2002) Appetizer on concepts and state of the art in chemical (ecological) risk assessment. COST meeting, Innsbrück, Austria.
Verdonck F.A.M., Thas O., Jaworska J. & Vanrolleghem P.A. (2002). Added value of a hierarchical bootstrap model in environmental standard setting. In: Proceedings TIES 2002, Annual Conference of The International Environmetrics Society, 1822 June 2002, Genova, Italy.
Verdonck F.A.M., Janssen C.R., Jaworska J., Thas O. & Vanrolleghem P.A. (2002) Accounting for hierachical variability in species sensitivity distributions. Proceedings 12th Annual Meeting of SETACEurope, 1216 May 2002, Vienna, Austria.
Verdonck F.A.M., Deksissa T., Matamoros D. & Vanrolleghem P.A. (2002) Dealing with variability in chemical exposure modelling in rivers. In: Proceedings Seminar on Exposure and Effects, Modelling in Environmental Toxicology. C.5. 48 februari 2002, Antwerp, Belgium.
Verdonck F.A.M., Jaworska J., Janssen C.R., Thas O. & Vanrolleghem P.A. (2001) Comparing statistical techniques for uncertainty assessment of species sensitivity distributions: effect of sample size. Proceedings 11th Annual Meeting of SETACEurope, 610 May 2001, Madrid, Spain.
Verdonck F. & Vanrolleghem P.A. (2000) Statistical tools in risk assessment. COSTmeeting, 2122 September 2000, Aartselaar, Belgium.
This work was also presented as a poster on several conferences.
Project duration: 19991101  20031101
Project home page
No website available
Contact
Frederik Verdonck
Department of Applied Mathematics, Biometrics and Process Control
Coupure Links 653
B9000 Gent
Belgium
Tel: +32(0)9 264.59.37
Fax: +32(0)9 264.62.20
Email: frederik.verdonck@biomath.rug.ac.be
Last update: 01 december 2008, webmaster@biomath.ugent.be
