Analyses and quantification of modelling uncertainties in streamflow simulations with applications to two catchments: the small lowland Kielstau basin in Germany and the mesoscale mountainous XitaoXi basin in China
Models are the primary way to predict the values of various system performance indicators in hydrologic researches. The usefulness of any model depends in part on the accuracy and reliability of its output. This PhD thesis presents the development of a methodological framework to analyse the impacts of three sources of modelling uncertainty (namely model structure error, parameter estimation and input data resolution) on streamflow simulation and to quantify the associated modelling uncertainties. The case study includes two catchments: the small lowland Kielstau catchment (51.5 km²) in Northern Germany and the mesoscale mountainous XitaoXi basin (2271 km²) in Southern China. The river discharge simulation is completed through the KIDS model (Kielstau Discharge Simulation model, Hörmann et al. 2007; Zhang et al. 2007) using PCRaster modelling language (Van Deursen 1995; Wesseling et al. 1996). The main criterion of model output performance is the Nash-Sutcliffe efficiency (Nash & Sutcliffe 1970). The structural uncertainty is assessed by developing a set of model ensembles with increasing model complexity. The modelling uncertainty induced by parameter estimation is investigated through Monte Carlo based sampling strategy in the framework of SUFI-2 analysis routine (Sequential Uncertainty Fitting, ver. 2, Abbaspour et al. 2004). The uncertainty of changing input data resolutions is analysed by aggregating grid cells. For each of them, a method has been developed to quantify the inherent modelling uncertainties with two statistical measures: R factor and P factor (Abbaspour et al. 2004; Schuol & Abbaspour 2006). Considering the two different catchments of Kielstau and XitaoXi, investigating the effects of model structure on model performance helps to identify the most appropriate model adapted to local hydrological features. Also, result comparisons for the parameter estimation and resolution impacts are conducted between the two basins. It is shown that the uncertainties induced by the different model structures tested in this study are much higher than the ones induced by parameter calibration and input data resolutions using a fixed hydrological model structure. However, modelling uncertainties from different sources are not independent of each other, they can interact in various ways and it is hard to calculate them separately. All the uncertainties obtained here refer to the overall modelling uncertainty while focusing on one aspect of influencing sources. It indicates that model output and modelling efficiency highly depend on tradition and empirical assumptions concerning the choice of model structures, parameter estimation, and the selection of appropriate resolution level. This study may provide a methodology to investigate these issues in the study basins.
Use and reproduction:
No CC License (german copyright law applies)