An analysis of the uncertainty in spatially distributed real-time flow forecasts
- Publication Type:
- Issue Date:
NO FULL TEXT AVAILABLE. This thesis contains 3rd party copyright material. ----- It is recognised that uncertainty is inherent in the production of any real-time hydrological forecasts. However, very few operational systems provide information on this uncertainty. This uncertainty arises from the errors introduced through a variety of sources associated with the application of a catchment modelling system; a catchment modelling system comprises a series of hydrological and hydraulic process models which, when combined with user-defined parameters, result in prediction of the catchment response to a given storm event. These errors can be classified as structural, parameter, and measurement. Model structure errors arise from the fact that the process models are based on a mathematical simplification of the process as it occurs in nature. Furthermore, the structures of catchment modelling systems differ in the degree of detail described, the manner in which processes are conceptualised, the data requirements, and the degree of spatial and temporal considerations. It is impossible to eliminate the structural uncertainties inherit in the systems, even if error-free calibration data is used. In addition, other sources of uncertainty in the application of catchment modelling systems arise from the input parameters and measurement errors. Model input uncertainties may result from the underlining assumptions regarding input parameters. A typical example is that rainfall depth based on a single point measurement is often used as representative of the rainfall within the catchment regardless of its spatial variability. Model input uncertainties are associated also with measurement errors (if input parameters are directly measured) or estimation errors (if input parameters are estimated by calibration). Decision makers now recognise that real-time operational decisions must be based on an understanding of the uncertainties associated with the real-time hydrological forecasts. It is critical, therefore, that hydrological forecast uncertainties are accounted for, to enable effective operational decisions. As a result the focal point of this research was to identify and assess the sources of uncertainty in a real-time hydrological forecast system in order to develop an uncertainty framework. The uncertainty framework classifies the uncertainty sources in the forecast processes based on their characteristics and nature into aleatory and epistemic uncertainty sources. The aleatory uncertainty represents the natural variability of the model input. To account for this source of uncertainty the radar derived rainfall observations were introduced and the ensemble resolutions were tested to select the most appropriate ensemble resolution. The epistemic uncertainty represents the knowledge based uncertainty source that is present in the catchment modelling system. To account for this source of uncertainty, the best modelling approach was addressed with the model rigorous parameter uncertainty assessment. As a result, the dual-model realisation approach was introduced, where both models were configured with the optimal set of parameters. The uncertainty framework was implemented and the Integrated Hydrological Ensemble Prediction System was formed. The forecast system couples the Short Term Ensemble Prediction System with the MIKE SHE hydrological model that accounts for the uncertainty sources in the forecast processes. The proposed research was undertaken on the Nattai River catchment, this river is one of the major tributary flowing into Sydney’s largest drinking water reservoir.
Please use this identifier to cite or link to this item: