Abstract
Summary:
Hydrological models reflect a modeler's understanding of natural hydrological processes and, as such, are mere approximations of reality. Therefore, uncertainties are inevitable and found in all links of the hydrological modelling chain. To obtain meaningful results, quantification of these uncertainties is necessary and considered good scientific practice. In this PhD thesis, major uncertainty sources associated with a complex, coupled groundwater-surface water model were investigated using a variety of statistical methods. Being part of the Danish HOBE project, the thesis utilized abundant observational data gathered for various hydrological variables during the project period. The main study area of the thesis was the Ahlergaarde subcatchment (1055 km2). The employed hydrological model was based on the MIKE SHE code, featuring an energy-balance module (SW-ET) that simulated actual evapotranspiration at hourly resolution. Due to the distributed nature of the model, special focus was given to an adequate consideration of spatial variability and to including hydrological variables representative of the grid scale into the uncertainty assessment. Regarding input uncertainty, an approach was presented that allowed the joint consideration of several sources of uncertainty in the space-time mapping of rainfall, including the introduction of so far unacknowledged neighborhood uncertainty. The impact of input (through a rainfall field ensemble) and parameter uncertainty (through Latin Hypercube Sampling) on predictive uncertainty was further studied by means of Monte Carlo simulation. Moreover, in recognition of the existence of considerable observational uncertainties, the concept of effective observational uncertainties was brought forward, allowing the integration of knowledge about scale differences in the comparison of simulated (grid scale) and observed values (point scale) into the uncertainty analysis. Finally, a post-processing technique (k-NN resampling) was successfully tested regarding its ability to perform reliable uncertainty analysis while resulting in bias-corrected predictions and quantification of residual (and therefore, also model structural) uncertainty. The Ph.D. thesis was intended to contribute to an improved understanding and treatment of uncertainties in hydrological modelling and further establish uncertainty analysis in operational hydrology.
Hydrological models reflect a modeler's understanding of natural hydrological processes and, as such, are mere approximations of reality. Therefore, uncertainties are inevitable and found in all links of the hydrological modelling chain. To obtain meaningful results, quantification of these uncertainties is necessary and considered good scientific practice. In this PhD thesis, major uncertainty sources associated with a complex, coupled groundwater-surface water model were investigated using a variety of statistical methods. Being part of the Danish HOBE project, the thesis utilized abundant observational data gathered for various hydrological variables during the project period. The main study area of the thesis was the Ahlergaarde subcatchment (1055 km2). The employed hydrological model was based on the MIKE SHE code, featuring an energy-balance module (SW-ET) that simulated actual evapotranspiration at hourly resolution. Due to the distributed nature of the model, special focus was given to an adequate consideration of spatial variability and to including hydrological variables representative of the grid scale into the uncertainty assessment. Regarding input uncertainty, an approach was presented that allowed the joint consideration of several sources of uncertainty in the space-time mapping of rainfall, including the introduction of so far unacknowledged neighborhood uncertainty. The impact of input (through a rainfall field ensemble) and parameter uncertainty (through Latin Hypercube Sampling) on predictive uncertainty was further studied by means of Monte Carlo simulation. Moreover, in recognition of the existence of considerable observational uncertainties, the concept of effective observational uncertainties was brought forward, allowing the integration of knowledge about scale differences in the comparison of simulated (grid scale) and observed values (point scale) into the uncertainty analysis. Finally, a post-processing technique (k-NN resampling) was successfully tested regarding its ability to perform reliable uncertainty analysis while resulting in bias-corrected predictions and quantification of residual (and therefore, also model structural) uncertainty. The Ph.D. thesis was intended to contribute to an improved understanding and treatment of uncertainties in hydrological modelling and further establish uncertainty analysis in operational hydrology.
| Originalsprog | Engelsk |
|---|
| Forlag | Department of Geosciences and Natural Resource Management, Faculty of Science, University of Copenhagen |
|---|---|
| Antal sider | 195 |
| Status | Udgivet - apr. 2018 |