| Popis: |
This thesis explores a range of uncertainty issues within the commonly used hydrological modelling framework. It assesses the extent that choices made during model construction and calibration result in different model outputs and aims to assess whether it is possible to develop a modelling protocol better than the rest. Using the 876.36 km2 Tern catchment, Shropshire, UK, and the physically-based, distributed modelling code, MIKE SHE, the research draws on large volumes of secondary data and provides a comprehensive catchment review and conceptual model. Two hydrological models of differing spatial complexities are developed and subject to different parameterisations, sensitivity analyses, and calibration methods (manual and automatic). Results are assessed at different locations within the catchment. Six models developed with different protocols result in minimal intra-model uncertainty. Nash-Sutcliffe NSE varies between 0.69–0.79 for discharge at the catchment outlet. Differences between spatial representations are more apparent at internal gauging stations; despite this similar performing models are developed for both spatial representations. Multi-objective automatic calibration produces models which provide more balanced representation of observed data as shown by results of validation. However, it is not possible to statistically identify any of the modelling protocols as better than the rest. Results suggest the amount a particular statistic is used within the calibration will influence other performance statistics. Therefore an independent summary score measure is also developed to assess performance. Intra-model uncertainty is assessed for the six models for eight UKCIP02 climate change scenarios. Results suggest increases in intra-model uncertainty at a similar magnitude as potential impacts of climate change. The research suggests careful choices about the modelling protocol need to be addressed at the outset of any hydrological modelling, with attention given to the uncertainties that may result of decisions made by the modeller – especially if using models in impact studies. |
