Quality Assurance in Conceptual Hydrologic Models: Developing functional validation tests for ensuring model quality and robustness

Currently, the quality assurance of conceptual hydrological models is primarily based on calibration and validation procedures, such as the validation tests proposed by Klemeš [1986]. These procedures provide insufficient testing of the underlying assumptions of a model structure and their correctness and credibility for specific purposes. While we assume the models we use are implemented physically correct, actual “crash tests” (Andréassian et al. [2009]) or quality assurance procedures do not exist.

This study therefore focuses on the development of a standardized quality assurance procedure for conceptual hydrologic models. A so called functional test scheme is proposed that complements existing calibration and validation procedures. Hereby, expected and unexpected model setups and parameterizations are tested and the model response is evaluated. The applied functional approach involves self-generated time series with synthetic climate data and a synthetic catchment to systematically test individual processes and procedures. We developed a line of test series for the modular modelling framework RAVEN, where several iterative test runs with changing model setups and parameterizations have been conducted in order to gain further insights into the correctness and plausibility of the implemented approaches and equations. We developed an R package that enables the almost automated execution of the repetitive processes in the test application for the RAVEN-based models.

Preliminary results revealed some minor and major problems of model functioning, sometimes related to simple reasons like unclear information in the model documentation. For example, showed the testing that the slope correction for different slope angles is not applied on manually entered PET data, while the documentation does not explicitly mention that slope angles are only affecting internally generated PET data. The conducted experiments prove the potential of readily developed functional tests and provide a basis for further developments in this regard.