Added value of distribution in rainfall-runoff models for the Meuse basin

Abstract

Why do equal precipitation events not lead to equal discharge events across space and time? The easy answer would be because catchments are different, which then leads to the second question: Why do hydrologists often use the same rainfall-runoff model for different catchments? Probably because specifying and distributing hydrological processes across catchments is not straightforward. It requires catchment data and proper tools to evaluate the details and spatial representation of the modelled processes. However, making a model more specific and distributed can improve the performance and predictive power of the hydrological model. Therefore, this thesis evaluates the added value of including spatial characteristics in rainfall-runoff models.
Most model experiments in this thesis are carried out in the Ourthe catchment, a subcatchment of the Meuse basin. This catchment has a strong seasonal behaviour, responds quickly to precipitation and has a large influence on peak flows in the Meuse. It has a variety of landscapes, among which steep forested slopes and flat agricultural fields.
This thesis proposes a new evaluation framework (Framework to Assess Realism of Model structures (FARM)), based on different characteristics of the hydrograph (hydrological signatures). Key element of this framework is that it evaluates both performance (good reproduction of signatures) and consistency (reproduction of multiple signatures with the same parameter set). This framework is used together with various other model evaluation tools to evaluate models at three levels: internal model behaviour, model performance and consistency, and predictive power.
The root zone storage capacity (Sr) of vegetation is an important parameter in conceptual rainfall-runoff models. It largely determines the partitioning of precipitation into evaporation and discharge. Distribution of a climate derived Sr-value (i.e., based on precipitation and evaporation) was compared with Sr-values derived from soil samples in 32 New Zealand catchments. The comparison is based on spatial patterns and a model experiment. It is concluded that climate is a better estimator for Sr than soil, especially in wet catchments. Within the Meuse basin, climate derived Sr -values have been estimated as well; applying these newly derived storage estimates improved model results.
Two types of distribution have been tested for the Ourthe catchment: the distribution of meteorological forcing and the distribution of model structure. The distribution of forcing was based on spatially variable precipitation and potential evaporation. These were averaged at different levels within in the model, thereby creating four levels of model state distribution. The model structure was distributed by using two hydrological response units (HRUs), representing wetlands and hillslopes. Eventually, a lumped and a distributed model structure were compared, each with four levels of model state (forcing) distribution. From this, it is concluded that distribution of model structure is more important than distribution of forcing. However, if the model structure is distributed, the forcing should be distributed as well.
Knowing that distribution of model structure is relevant, more detailed process conceptualisations have been tested for the Ourthe Orientale, a subcatchment of the Ourthe. An additional agricultural HRU was introduced for which Hortonian overland flow and frost in the topsoil are assumed to be relevant. In addition, a degree-day based snow module has been added to all HRUs. Adding these process conceptualisations improved the performance and consistency of the model on an event basis. However, the implemented processes and the related signatures are sensitive to errors in forcing and model outliers and should therefore be implemented carefully.
This thesis finishes with two explorative comparisons; one comparing the newly developed model of the Ourthe Orientale catchment with other catchments; the second between the newly developed model and other models, including the HBV configuration currently used for operational forecasting in the Meuse basin. These comparisons were carried out based on visual inspections of parts of the hydrograph. The results show that the newly developed model can be applied in neighbouring catchments with similar performance. The comparison with other models demonstrates that a very quick overland flow component and a parallel configuration of fast and slow runoff generating reservoirs is important to reproduce the dynamics of the hydrograph related to different time scales. Both aspects are included in the newly developed model. As a results, the newly developed model is better able to reproduce most of the dynamics of the hydrograph than the operational HBV configuration, used at the moment of writing.
Distribution and detailed process conceptualisation are very beneficial for rainfall-runoff modelling of the Ourthe catchment. However, they should be applied with care. Conceptual models are a strong simplification of reality. When confronting them only with discharge data, there is a risk of misinterpreting other hydrological processes.
This thesis suggests two possible opportunities to further improve conceptual models. First, catchment understanding could be increased by adding more physical meaning to the models, such as the climate derived root zone storage capacity. And second, remote sensing and plot scale data could be combined to link hydrological processes at different scales. In this way conceptual models can probably be used to get more insight into scaling issues, which occur when moving from hillslope to catchment scale.