A fuzzy logic-classification of sediments based on ... - Semantic Scholar

4 downloads 0 Views 443KB Size Report
Jun 10, 2009 - mathematic methods: a normalisation, b normal distribution, c box plot and d empirical method ..... Dwyer FJ, Ingersoll C,. Fairchild JF (1994) Use of benthic invertebrate community structure .... Modell 130:111–119. Singh NP ...
J Soils Sediments (2009) 9:168–179 DOI 10.1007/s11368-009-0087-8

SEDIMENTS, SEC 2 • RISK MANAGEMENT • RESEARCH ARTICLE

A fuzzy logic-classification of sediments based on data from in vitro biotests Steffen Keiter & Thomas Braunbeck & Susanne Heise & Stefan Pudenz & Werner Manz & Henner Hollert

Received: 20 October 2008 / Accepted: 27 March 2009 / Published online: 10 June 2009 # Springer-Verlag 2009

Abstract Background, aim, and scope Ecotoxicological risk assessment of sediments is usually based on a multitude of data obtained from tests with different endpoints. In the present S. Keiter (*) : T. Braunbeck : H. Hollert Department of Zoology, Aquatic Ecology and Toxicology Section, University of Heidelberg, Im Neuenheimer Feld 230, 69120 Heidelberg, Germany e-mail: [email protected] T. Braunbeck e-mail: [email protected] H. Hollert e-mail: [email protected] S. Heise Institut für Biogefahrenstoffe und Umwelttoxikologie, Hamburg University of Applied Sciences (HAW), Lohbrügger Kirchstr. 65, Hamburg 21033, Germany e-mail: [email protected] S. Pudenz Westlakes Scientific Consulting Ltd., Department of Environmental Science, Moor Row, Cumbria CA24 3 LN, UK e-mail: [email protected] W. Manz German Federal Institute of Hydrology, Am Mainzer Tor 1, 56068 Koblenz, Germany e-mail: [email protected] H. Hollert Institute for Environmental Research (Biology V), Department of Ecosystem Analysis, RWTH Aachen University, Worringerweg 1, 52074 Aachen, Germany

study, a fuzzy logic-based model was developed in order to reduce the complexity of these data sets and to classify sediments on the basis of results from a battery of in vitro biotests. Materials and methods The membership functions were adapted to fit the specific sensitivity and variability of each biotest. For this end, data sets were categorized into three toxicity levels using the box plot and empirical methods. The variability of each biotest was determined to calculate the range of the gradual membership. In addition, the biotests selected were ranked according to the biological organisation level in order to consider the ecological relevance of the endpoints measured by selected over- or underestimation of the toxicity levels. In the next step of the fuzzy logic model, a rule-base was implemented using if... and...then decisions to arrive at a system of five quality classes. Results The results of the classification of sediments from the Rhine and Danube Rivers showed the highest correlation between the biotest results and the fuzzy logic alternative based on the empirical method (i.e. the classification of the data sets into toxicity levels). Discussion Many different classification systems based on biological test systems are depending on respective data sets; therefore, they are difficult to compare with other locations. Furthermore, they don‘t consider the inherent variability of biotests and the ecological relevance of these test systems as well. In order to create a comprehensive risk assessment for sediments, mathematical models should be used which take uncertainties of biotest systems into account, since they are of particular importance for a reliable assessment. In the present investigation, the variability and ecological relevance of biotests were incorporated into a classification system based on fuzzy logic. Furthermore, since data from different sites and

J Soils Sediments (2009) 9:168–179

investigations were used to create membership functions of the fuzzy logic, this classification system has the potential to be independent of locations. Conclusions In conclusion, the present fuzzy logic classification model provides an opportunity to integrate expert knowledge as well as acute and mechanism-specific effects for the classification of sediments for an ecotoxicological risk assessment. Recommendations and perspectives In order to achieve a more comprehensive classification, further investigation is needed to incorporate results of chemical analyses and in situ parameters. Furthermore, more discussions are necessary with respect to the relative weight attributed to different ecological and chemical parameters in order to obtain a more precise assessment of sediments. Keywords Classification . Ecological relevance . Fuzzy logic . Hazard assessment . Ranking

1 Background, aim, and scope Ecotoxicological risk assessment requires a multitude of data with different endpoints. In addition, interactions between the variables are usually very complex; therefore, the data are difficult to analyse and interpret (Adriaenssens et al. 2004; Tran et al. 2002). However, especially for decision makers and stakeholders, it is essential to aggregate the results of such complex interrelationships for a straightforward assessment of environmental samples, e.g. contaminated sediments. Most of the data and knowledge concerning ecotoxicological investigations have to deal with uncertainties, e.g. measurement errors and the inherent variability of biological systems (Adriaenssens et al. 2004; Tran et al. 2002). Therefore, Regan et al. (2000) distinguishes two main groups of uncertainties: epistemic and linguistic uncertainty. Epistemic uncertainty is divided further into six main types, e.g. measurement or systematic errors as well as natural variation with respect to the dynamic of an ecosystem. Linguistic uncertainty is defined as the vagueness, context dependence, ambiguity, indeterminacy of theoretical terms, and underspecificity of natural as well as scientific language (Regan et al. 2002). Hence, for a reliable decision support and classification of contaminated sediments, it is of prime importance to handle these uncertainties (Adriaenssens et al. 2004). There are many different approaches to interpret the structure of data from ecotoxicological investigations (Babut et al. 2007, Jorgensen 1999). However, only a few approaches are able to cope with epistemic and linguistic uncertainties as well as with complex issues of integrative assessments (Adriaenssens et al. 2004; Babut et al. 2007).

169

Fuzzy logic offers a way to address these problems and has repeatedly been reported as a methodology for environmental impact assessment as well as classification of contaminated sediments and environmental samples (Babut et al. 2007; Heise et al. 2000; Hollert et al. 2002b; Salski and Kandzia 1996; Shepard 2005; Silvert 1997, 2000). Fuzzy logic was introduced by Zadeh (1965) as a tool to handle vague expert knowledge and uncertain or imprecise information. Non-fuzzy data processing is usually based on methods with yes/no-decisions and zero/one-memberships: ‘one/yes’, if the data belong to the set; and ‘zero/no’, if they do not (also known as ‘crispy values’; (Bojorquez-Tapia and Juarez 2002)). In contrast, in fuzzy rule-based systems, knowledge is represented by if…and…then rules (Adriaenssens et al. 2004), and a variable belongs to a fuzzy set with a defined membership degree between zero and one (Fig. 1). Fuzzy logic thus enables the processing of uncertainties and imprecise data by means of a gradual membership. Adriaenssens and co-workers (2004) concluded that fuzzy logic, in particular by means of fuzzy rulebased models, appears to be a very promising tool for environmental assessment. In several studies, fuzzy-rule based models have been used for the classification of contaminated sediments on the basis of results from in vitro biotests (Babut et al. 2007; Duft et al. 2003; Heise et al. 2000; Hollert et al. 2002b). The present communication is a follow-up to these studies and has been designed to develop fuzzy logic classification concepts by Heise et al. (2000) as well as Hollert et al. (2002b). In the investigations by Hollert and co-workers (2002b), the variability and ecological relevance of the selected test systems were insufficiently implemented into the rule-based classification system. Although Heise et al. (2000) incorporated the variability into the fuzzy logic

Fig. 1 Examples for a zero/one membership function (conventional set; left) and a fuzzy set membership function (right). The fuzzy set exhibit gradual memberships (redrawn from Hollert et al. 2002)

170

classification, the ecological relevance of the selected test systems was not taken into consideration. In general, due to compensation processes, the extent of an effect by a pollutant declines with an increase in biological organisation, whereas the ecological relevance of results from ecotoxicological investigations increases with the level of biological organisation (Braunbeck 1998). Hence, the ecological relevance of the selected biological tests should definitely be given more importance in integrative risk assessment and classification. In the present study, ecological relevance has been implemented into the fuzzy logic as a linguistic uncertainty. The epistemic uncertainty will be considered by the inherent biological variability of the different test systems. With these modifications, a rulebased assessment system was created, which is able to classify sediments on the basis of results from in vitro tests.

2 Bioassays selected for the creation of the database In this study, data from four different bioassays were selected to classify sediments: (1) the neutral red assay (cytotoxicity; (Hollert et al. 2000; Keiter et al. 2006; Kosmehl et al. 2004); as well unpublished data from their own laboratory); (2) the EROD assay (dioxin-like activity; (Keiter et al. 2008) as well unpublished data from their own laboratory); (3) the comet assay (genotoxicity; (Keiter et al. 2006; Kosmehl et al. 2004; Seitz et al. 2008)); and (4) the fish egg assay (embryo toxicity; (Hollert et al. 2000; Hollert et al. 2003; König et al. 2002; Seitz et al. 2008; Ulrich et al. 2002)). The fish egg assay was carried out both with native sediments and acetonic sediment extracts: Native sediments represent the bioavailable fraction of lipophilic, particlebound substances, whereas organic extracts of whole sediments have frequently been used to assess the total (eco-)toxicological potential of sediments (Hollert et al. 2003). In order to avoid redundancy of data, the results of the selected biotest were analysed by Spearman‘s rank correlation (SigmaStat 3.5, Systat, Point Richmond, USA). Thereby, a weak correlation was found between the fish egg and comet assays (rS =0.62, p