In many areas of the world, the absence of streamflow data to calibrate hydrological models limits the ability to make reliable streamflow predictions. Whilst a large and increasing number of regions are insufficiently gauged, there are also many highly monitored catchments. Transferring the knowledge gained in data-rich areas to data-scarce regions offers possibilities to overcome the absence of streamflow observations. In this thesis knowledge is transferred in the form of signatures, which reflect hydrological response characteristics of a particular catchment. Several signatures may be required to capture different aspects of catchment functional behaviour. Using a large dataset of catchments, observed signatures are regressed against physical and climatic catchment descriptors. Signatures for an ungauged location with known descriptors are then estimated utilising the derived relationships. A Bayesian procedure is subsequently used to condition a conceptual model for the ungauged catchment on the estimated signatures with formal uncertainty estimation. Particular challenges related to the Bayesian approach include the selection of signatures, and specification of the prior distribution and the likelihood functions. A methodological development is based on an initial transformation of the commonly adopted uniform parameter prior into a prior that maps to a uniform signature distribution, aimed at cases where limited prior knowledge regarding the model structure adequacy and the parameters distribution exist. The suggested methodology contributes to improved estimation of response signatures, and is particularly relevant when regionalised information is highly uncertain. A further contribution of this thesis refers to the integration of several regionalised signatures into the model, accounting for the inter-signature error covariance structure. By increasing the number and regionalisation quality of signatures in the conditioning process, better predictions are obtained. Additionally, the consideration of the inter-signature error structure may improve the results when correlations between errors are shown to be strong. When regionalised signatures are integrated into the model, it is shown that model structural inadequacy has a strong effect on the prediction quality.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:676767 |
Date | January 2014 |
Creators | Cardoso Lopes de Almeida, Susana Margarida |
Contributors | McIntyre, Neil ; Le Vine, Nataliya ; Buytaert, Wouter ; Butler, Adrian |
Publisher | Imperial College London |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://hdl.handle.net/10044/1/28086 |
Page generated in 0.0026 seconds