Improving model consistency and process realism in conceptual rainfall-runoff models

Hydrological models are used for several purposes, from reservoir management to stormflow prediction. However, their predictive power is often questionable. As most rainfall-runoff models are highly dependent on calibration, models often show deteriorated performance during the validation period. This is caused by too much focus on and trust in the model calibration. But to get the right answers for the right reasons, a different approach is necessary. Earlier research showed that automated model calibration should be balanced with a more expert-knowledge-driven strategy of constraining models. Therefore, the following research question is the main subject of this thesis: “How do model complexity and expert-knowledge influence consistency and realism in a conceptual rainfall-runoff model?” The hypothesis is that more consistency and a more realistic catchment representation can be achieved by increasing the complexity of the model, counter-balanced by adding constraints. Increasing model complexity makes it possible to implement additional dominant hydrological processes in a model. The constraints make it possible to eliminate unrealistic parameter sets and parameter sets leading to unrealistic model behaviour. This is expected to lead to more consistency and improved realism of the processes. Besides evaluating models in terms of calibration objective functions only, hydrological signatures are used. These signatures show the ability of the models to reproduce specific hydrological behaviour of the catchment. Four European catchments are studied in this thesis, of which one catchment is located in Germany, one in Sweden and two in Italy. These catchments all have different characteristics, varying from a relatively flat catchment with large areas of grassland and pastures, to mountainous catchments with large areas of forest. Several models are developed and applied to each catchment. The results show that even though improvements are not always directly clear from the calibrated objective functions, in some cases the models have an increased ability to reproduce specific hydrological signatures. This thesis shows that by increasing model complexity, together with the addition of parameter and process constraints, more plausible model results can be obtained. The approach used in this research makes it possible to obtain a model which behaves well according to the modellers’ perception of the catchment. Besides, the constraints have a constraining effect on the possible parameter space and therefore, reducing the effects of equifinality. The results show that evaluating a model based on calibration performance solely is insufficient. Hydrological signatures give useful information about specific strengths and weaknesses of a model and should be also taken into account during hydrological research.?Water ResourcesWater ManagementCivil Engineering and Geoscience