Hydrological modeling is a crucial tool for various tasks, such as operational flood forecasting and water management. Traditional hydrological modeling approaches involve conceptual modeling, which relies on heuristic process descriptions that are encoded in fixed model structures.

However, in recent years, new data-driven techniques have shown to outperform traditional approaches, provided that sufficient training data is available. Our project aims to explore ways to combine both approaches by integrating expert knowledge and physical restrictions into data-driven methods. By incorporating the flexibility of data-driven techniques with the established expert knowledge from literature over the past decades, we expect to achieve faster model training and better generalization capabilities.