A Hypothesis-based Approach to Hydrological Model Development: The Case for Flexible Model Structures

Ambiguities in the appropriate representation of environmental processes have manifested themselves in a plethora of hydrological models, differing in almost every aspect of their conceptualization and implementation. This current overabundance of models is symptomatic of insufficient scientific understanding of environmental dynamics at the catchment scale, which can be attributed to difficulties in quantifying the impact of sub-catchment heterogeneities on the catchment’s hydrological response. In this presentation we advocate the use of flexible modeling frameworks during the development and subsequent refinement of catchment-scale hydrological models. We argue that the ability of flexible modeling frameworks to decompose a model into its constituent hypotheses, necessarily combined with incisive diagnostics to scrutinize these individual hypotheses against observed data, provides hydrologists with a very powerful and systematic approach for improving process representation in models. Flexible models also support a broader coverage of the model hypothesis space and hence facilitate a more comprehensive quantification of the predictive uncertainty associated with system and component non-identifiabilities that plague many model analyses. As part of our discussion of the advantages and limitations of flexible model frameworks, we critically review major contemporary challenges in hydrological hypothesis-testing, including exploiting data to investigate the fidelity of alternative process representations, accounting for model structure ambiguities arising from uncertainty in environmental data, and the challenge of understanding regional differences in dominant hydrological processes. We assess recent progress in these research directions, and how such progress can be exploited within flexible model applications to advance the community’s quest for more scientifically defensible catchment-scale hydrological models.