Projects per year
Six steps can be distinguished in the process of hydrological modelling: the perceptual model (deciding on the processes), the conceptual model (deciding on the equations), the procedural model (get the code to run on a computer), calibration (identify the parameters), evaluation (confronting output with observations), and uncertainty analysis (estimate uncertainty in the model and its output). An engineer conducts these steps different than a scientist, because the goal of an engineer is to solve practical problems, while the goal of a scientist is to increase the understanding of the system. The difference between scientists and engineers is most pronounced in the perceptual modelling step. However, in many of the current hydrologic sciences studies, engineering and scientific approaches are mixed. As a scientist, three common philosophies of science can be adopted: verificationism, falsificationism, and Bayesianism. It was demonstrated that verificationism most closely resembles engineering in the modelling steps, while falsificationism and Bayesianism call for a different practice.
In this thesis, several of the modelling steps have been investigated in more detail. In order to investigate these modelling steps, we applied widely used hydrological models (Chapter 2). These models vary in complexity, and have been applied to catchments with varying temporal and spatial scales.
In Chapter 3 three parameter identification methods and their data requirements were compared for a small (3.3 km2) catchment using a parsimonious two-parameter model. Two methods based on discharge data were employed, Bayesian based automatic calibration (DREAM) and recession analysis, and one physics-based method was employed, Boussinesq theory. Automatic calibration and recession analysis both required five months of discharge data in order to obtain stable parameter estimations. Boussinesq theory, which allows a-priori parameter estimation based on catchment characteristics, showed to lead to highly uncertain parameters due to uncertainty in the catchment characteristics.
Chapter 4 deals with the transferability (and thus sensitivity) of parameters across spatial and temporal resolutions in the Thur catchment (1703 km2). It was shown that parameters were hardly sensitive to the spatial resolution (a high transferability), while the parameters were very sensitive to the temporal resolution (especially from hourly/daily to a monthly time step). This indicates that the spatial variability is substantially underestimated. In this study we adopted common practice for hyper-resolution models applied at a large domain. The results therefore provide a strong motivation to further investigate and improve the representation of spatial and temporal variability in large-domain hydrological models.
Chapter 5 shows that decisions during model configuration, basically subjective decisions of the modeller, significantly impact the simulation of hydrological extremes in the Thur basin (1703 km2) We explored four modelling decisions; the spatial resolution of the model, the spatial representation of forcing, the calibration period, and the performance metric, and investigated if these decisions influenced the simulated flood and drought characteristics in the Thur basin. It was shown that for the flood characteristics, the performance metric was the most influential decision, and for the drought characteristics, the calibration period was most important. Subjective modelling decisions introduce uncertainty in the modelling process. Working with multiple hypotheses of model implementations could help in providing insight in this uncertainty.
In Chapter 6 we explore three sources of uncertainty in a hydrological climate change impact assessment for the period 2070-2100 for 605 basins throughout the contiguous United States; parameter uncertainty, hydrological model structural uncertainty, and uncertainty in General Circulation Model (GCM) forcing. It was demonstrated that the uncertainty introduced by any of the three sources can be thus large that even the sign of the change is unknown in many basins (i.e., an increase or decrease in average annual runoff compared to the period 1985-2008). This uncertainty could be attributed to the snow parameterization in the hydrological models and disagreement among the GCMs on the change in precipitation. Furthermore, it was demonstrated that processes related to aridity and intermittent flow behaviour are not yet well captured in the investigated hydrological models.
In Chapter 7 it was shown, based on a literature study of 192 peer-reviewed publications, that the spatial resolution at which the Variable Infiltration Capacity (VIC) model is applied has increased, while the calibration and validation time interval has remained unchanged. It is argued that the calibration and validation time interval should keep pace with the spatial resolution of the model in order to resolve the processes relevant at the applied spatial resolution. Different processes are relevant at different spatial scales; the calibration and validation time interval should reflect the temporal scale of these processes in order to estimate the credibility of the model.
The results from all the studies can be summarized in three points: Not only the model choice, but also the configuration of the model determines the outcome of the model; sufficient data are needed to constrain and evaluate a model; and the large uncertainty in modelling studies provides a strong motivation to increase our understanding - i.e., to focus on science rather than on engineering. In order to establish this, models should be related to theories (hypotheses), which should be tried to falsified. The model set-up should be considered an element of the tested theory. Novel observation technologies provide the opportunity to test and falsify these theories, and can lead to the formulation of new theories.
|Qualification||Doctor of Philosophy|
|Award date||19 Apr 2017|
|Place of Publication||Wageningen|
|Publication status||Published - 2017|