The political interest in fish habitat suitability and, consequently, of fish populations has increased. The fish habitat suitability is a key factor for successful ecological restoration, for example via dam removal and implementation of fish passage. Furthermore, the fish population composition determines the ecological quality of water bodies within the Water Framework Directive (WFD).

The aim of this thesis is to develop an ecological model for fish habitat suitability in the river Rotte basin in the Netherlands that can be used in policy development for ecological restoration. The model serves as a decision support system to explain differences in fish population and evaluate the impact of management actions to modify fish habitat suitability. The river Rotte suits perfectly for this case study, because recently a fish passage has been realised between the Rotte and Nieuwe Maas to facilitate fish migration and expand living conditions. Furthermore, the river Rotte is a designated WFD water body, but the current status of the river varies between "poor" and "moderate" due to an unbalance between plant-loving and benthivorous fish species.

The model developed in this thesis is a Bayesian Belief Network model that predicts habitat factors for food preference and preference for habitat structure. The model is based on machine learning with a set of cases from monitoring data and predicts the probability distribution for fish habitat suitability for plant-loving and benthivorous fish species. The model has been applied to assess the impact of local conditions on fish habitat differentiation and to evaluate the impact of management actions on fish habitat suitability. The research shows that the Bayesian Belief Network model is very useable for policy making and facilitates the participation of various stakeholders. However, the current version of the model shows inadequate prediction accuracy and relies heavily on sampling data. This can be improved by expanding the scope of the model to include other water bodies in the Netherlands and by using metamodels for specific model variables.

Overall, the Bayesian Belief Network model is functional and usable for policy making, but further improvements are needed to enhance its prediction accuracy. This can be achieved through expanding the model's scope, evaluating its performance, and including more habitat factors in the model structure.