On the role of model structure in hydrological modeling: Understanding models

Modeling is an essential part of the science of hydrology. Models enable us to formulate what we know and perceive from the real world into a neat package. Rainfall-runoff models are abstract simplifications of how a catchment works. Within the research field of scientific rainfall-runoff modeling, the focus is shifting from performance to consistency with fundamental principles of hydrological processes. Despite all efforts there are still gaps in our understanding of how real catchments work. Additionally there are enormous endeavors to understand the behavior of rainfall-runoff mechanisms within a catchment. Examples of research items that characterize system behavior are; water residence time, travel time, differences between velocity and celerity, use of tracers and remote sensing techniques. Meanwhile not much attention is given to understanding how all this knowledge should be implemented in a model. This thesis is an attempt to understand the role of model structure and the related assumptions in making better use of the available knowledge on catchment processes. The starting point of this thesis is building a hydrological model based on topographical landscape units, using a newly introduced descriptor called HAND (Height above nearest drainage). The purpose of this part of research was to build a model which contains our perception of how rainfall-runoff mechanisms work for distinct landscapes, classified based on HAND and slope. Three different models have been developed from simple to complex. A simple model has a limited number of constraints while a complex model can have various constraints imposed. To have a realistic behavior of these models a set of constraints has been defined and imposed on the model parameters as well as model fluxes and states in a comparative fashion. After simulations with these models, it is observed that even without calibration and just by constraining, the complex model was able to simulate discharge within an acceptable range. Moreover, without the constraints the complex model shows higher parity with observation compared to the simple model. To refine this method even more, a simple parameter search strategy is proposed to satisfy all imposed constraints (chapters 3, 4 and 5). The fact that adding proper constraints can have such a positive impact on the accuracy of the outcome is the basis for further research into the structural elements and type of constraints imposed on the model. However to study the effect of structural changes in a model, the dominant effect of assumptions on state-discharge relations (parameterizations) had to be diminished. To achieve this, weak parameterization was introduced building on the concept of random functions. Using weak parameterization, it was found that structural complexity can improve themodel’s correspondence with observed data significantly. It was also found that there is a balance between model complexity, imposed constraints and performance of the model. Although increased complexity of a model can improve its performance, an over-constrained model can deviate from observed data. This means that, although a complex model may be structurally more favorable, the constraints, based on expert knowledge, should be set-up in such a way as to minimize the bias of the model outcome. This research can be a basis for the dialog between modelers and experimentalists. This approach can be used to investigate the behavior of the model in relation to its structural configuration, parameterizations and the set of imposed constraints (chapter 6). The last part of this thesis (chapter 7) covers the introduction of the concept of time consistent model parameters. Traditionally a part of the time series is kept for a separate evaluation of the model. However, during so-called SubPeriod calibration (SuPer calibration), parameter sets are evaluated based on their resulting model performance for different sub-periods of the entire observed time series. The parameter sets identified by SuPer calibration are different from the optimal parameter set found by calibration over the same period of data. This thesis concludes with a critical review of this research work together with discussions that took place during the research (chapter 8). This chapter is reflecting the novelty and deficiency of presented work and the battles to be fought in future.Water ManagementCivil Engineering and Geoscience