Hydrological models can be categorised into three groups: empirical models that are based on simple mathematical functions or being data-driven, conceptual models that rely on abstract combinations of storages and fluxes to depict catchment processes, or physically-based models that are structurally complex and incorporate physical interactions at different scales to simulate processes and system states. To calibrate and evaluate especially physically-based models, the use of multi-criteria evaluation schemes has proven to be effective to find model parameterisations that can reproduce multiple catchment processes and states instead of only the discharge. However, uncertainty in models, originating from different sources, often limits the robust interpretability of simulation results, making it necessary to assess how modelling applications can be improved to reduce uncertainty. In this thesis, the relevance of structural adequacy for the depiction of processes in models was demonstrated for the micro level for the example of dynamic phenology, where spatiotemporal model performance was improved by implementing a dynamic approach to modelling leaf emergence instead of a static one. For model evaluation at macro level, it was shown how a multi-criteria approach combining groundwater dynamics, surface runoff patterns and discharge can identify process-behavioural parameterisations and thus improve process depiction in hydrological models. At the meta level, it was demonstrated how the quasi-coupling of a hydrological and a hydraulic model can combine the different strengths of both models to simulate surface runoff processes taking infiltration into account, whereby the relevance of multi-criteria evaluated hydrological models for the derivation of hydrological variables was shown. In addition, uncertainty was explicitly incorporated into model evaluation at the meta level, where a virtual reality model approach was applied to assess the contribution that different variables can make to model evaluation when the associated measurement uncertainty is taken into account. Based on the individual results, it was possible to conclude that uncertainty and its different sources are a relevant factor in model evaluation at different levels and could be reduced by improving structural model adequacy, adapting model application approaches, and explicitly incorporating uncertainty into model evaluations.