Predictions
in Ungauged Basins: Promise and Progress (Proceedings of symposium S7 held during the
Seventh IAHS Scientific Assembly at Foz do Iguaçu, Brazil, April 2005). IAHS
Publ. 303, 2006, 134-144.
The Multi-Model approach to rainfall–runoff
modelling
Adam m. wyatt & stewart w. Franks
School of Engineering, University of
Newcastle, Callaghan 2308, New South Wales, Australia
adam.wyatt@studentmail.newcastle.edu.au
Abstract This paper introduces the Multi-Model approach to rainfall–runoff modelling: a new lumped modelling method that incorporates numerous alternative process descriptions for the dominant processes within a catchment that affect the streamflow response to climatic forcing conditions. An extended GLUE approach is used to calibrate the numerous model structures contained, obtaining the best 100 parameter sets from two million (uniform) randomly sampled sets for each of the 45 model permutations contained. These model/parameter combinations were then used to produce prediction confidence limits for subsequent runoff predictions, including a calibration period, validation period and a synthetically determined extreme event period. Additionally, the ability of the calibration process to constrain the internal dynamics of models is investigated. The results indicate that calibration to simple runoff data alone is insufficient to constrain the saturation excess process description. Further potential conditioning of the models against saturated area is then investigated to refine the extreme event prediction uncertainty envelopes, showing that a single uncertain criteria limitation (on the extreme event flood peak saturated area) does little to improve the uncertainty envelopes. However, when multiple observations are utilized (on the flood peak and the recession saturated areas) the runoff generating process is sufficiently constrained to dramatically reduce the runoff prediction uncertainty envelopes for the extreme period, irrespective of model structure.
Key words
extreme event; rainfall–runoff modelling; saturated area; uncertain
conditioning