Good modelling practice begins before modelling. Data, perception, and model hypotheses

Over the past decades, numerous guidelines have been proposed to improve hydrological modelling practice, with much emphasis placed on model calibration, evaluation, and intercomparison. This includes our own work and that of many others. While these efforts have advanced methodological rigor, they often implicitly assume that the set of candidate models is already well defined. In practice, however, the modelling space is vast, particularly for distributed models where process representations, parameterizations, and spatial variability can differ substantially. Selecting suitable model structures therefore remains a fundamental and often underexplored challenge.

In this contribution, we argue that hydrological modelling should not be the starting point of analysis, but rather the outcome of a structured chain of reasoning. This chain begins with the data: understanding data characteristics, limitations, and information content, and interpreting them in the context of dominant hydrological processes. Such data-driven reflection naturally leads to explicit and testable model hypotheses, which then form a meaningful basis for model selection and comparison. Central to this workflow is the perceptual model, which acts as a conceptual bridge between data interpretation and formal model structures.

Using examples from recent work in distributed hydrological modelling, we illustrate how this process-oriented approach can guide the choice of model complexity and structure, reduce arbitrariness in modelling decisions, and improve the interpretability of results. The contribution emphasizes that good modelling practice requires not only robust calibration and comparison strategies, but also a transparent and data-informed pathway that precedes model application itself.