Preparation and use of the domain expert knowledge for automated modelling of aquatic ecosystems\ud

This thesis is concerned with automated modelling (AM) of aquatic ecosystems. The method used here integrates the two basic principles of modelling, i.e., empirical or data-driven in theoretical or modelling by using the expert background knowledge. The integration of empirical in theoretical modelling is based on the use of the background knowledge in the procedure of model induction from measured data. The theoretical knowledge that guides the process of model induction includes a knowledge library of generalised knowledge from a specific domain in a task specification of the observed system.\ud \ud The thesis is divided into two parts. The first part deals with elaboration of knowledge library in the domain of modelling of aquatic ecosystems. The library includes knowledge about food web modeling by following the mass conservation principle. The knowledge is formalized in terms of (1) taxonomy of variable types, (2) basic processes that govern the behavior of aquatic ecosystems, (3) alternative models of the basic processes, and (4) knowledge how to combine models of individual processes into a model of the entire ecosystem. We evaluated the generality of the knowledge in the library through reconstruction of three wellknown models of different complexity. Thus, we showed that such formalization of the modelling knowledge provides a solid unifying framework for both handcrafting ecological models as well as their automated induction from measured data.\ud \ud In the second part we applied the developed library in the AM method on four real world domains. Using the measurements and the background knowledge we constructed models for each domain. The models were evaluated according to their accuracy and transparency. We tackled the following domains: Lake Glumsoe (Danmark), Lagoon of Venice (Italy), Lake Kasumigaura (Japan), and Lake of Bled (Slovenia). The quality of the models is above all dependant on (1) the knowledge in the library, (2) the quality of the measurements, (3) ecosystem complexity, and (4) the expert knowledge introduced in the induction procedure