Options
2018
Doctoral Thesis
Titel
Analysis and Optimization of Ecotoxicological Models under Uncertainty
Abstract
The problem of assessing the environmental risk of pesticides can be formulated as a robust feasibility problem: A pesticide is only approved by regulatory agencies of Germany and the entire European Union, if it causes no damage to the environment and man in any possible scenarios. Typically, experimental toxicity studies are used to construct a scenario based approximation of an uncertainty set. In doing so, the considered scenarios are possibly unrealistic, but they aim to represent the worst case. Thus, an overestimation of the effects is accepted to be certain that all possible risks are covered. We consider mechanistic models that do not only describe the effect of pesticides but also aim to explain the underlying processes. Such models may build a more realistic approximation of uncertainty set and thus be instrumental in assessing the risk. This allows the consideration of more complex scenarios. Currently, the use of the models is discussed but at most used in addition to conventional standard tests in European risk assessment. One reason is that they are not well documented. There exists a large family of models differing in complexity as well as precision. Furthermore, there are no standardized model test scenarios, yet. Especially, it is not clear how trustworthy the predictions of the models as well as the models themselves are. Mathematically, we consider parametrized systems of nonlinear differential equations and the complex interdependency of two associated problems. On the one hand, we have the inverse problem of identifying the model parameters. On the other hand, we have the direct problem of predicting effects under varying environmental conditions. This step requires an in depth analysis of the discretization error (ordinary differential equations), extrapolation error (situations not covered by calibration), conceptual model error (situation not covered by model) and consequences of changes in parameter values (sensitivity and uncertainty analysis). Furthermore, the characterization of the physically sound parameters is in special interest in view of the construction of model test scenarios. Besides a realistic modeling, the solution of the model system (exact, approximate) and the choice of the initial values of the inverse problem play a prominent role. We study three different models in the main chapters that are connected by the aim of describing mechanistically the effect of active substances in the environment. The first part deals with the unified framework of the model guts (General Unified Threshold model of Survival). This model is based on two different mortality hypotheses: stochastic death (sd) and individual tolerance (it). The second and third part concentrate on models describing the sublethal effects of active substances on growth of the water plant Lemna. Although both models treat the same species and have the same objective, the models differ in concept and complexity. The first model is a logistic growth model with an effective growth rate that is influenced by environmental conditions and dosage of a pesticide. The second model involves Dynamic Energy Budget theory (deb) describing growth in terms of changing the energy budget in time. We analyze and quantify uncertainties and errors in the considered ecotoxicological models. Furthermore, we develop methodologies to reduce the uncertainties such that the prediction of the models is enhanced. Thus, this thesis provides a foundation for the use of sophisticated mathematical models in environmental risk assessment of pesticides.
ThesisNote
Duisburg-Essen, Univ., Diss., 2018
Beteiligt
Verlagsort
Duisburg