HR: 14:55h
AN: H53H-06 INVITED [Abstracts]
TI: Modelling the Catchment via Mixtures: Issues of Model Specification and Validation
AU: * Marshall, L A
EM: lucy@civeng.unsw.edu.au
AF: School of Civil and Environmental Engineering, The University of New South Wales, UNSW
, Sydney, NSW 2052
Australia
AU: Sharma, A
EM: a.sharma@unsw.edu.au
AF: School of Civil and Environmental Engineering, The University of New South Wales, UNSW
, Sydney, NSW 2052
Australia
AU: Nott, D
EM: d.nott@unsw.edu.au
AF: School of Mathematics, The University of New South Wales, UNSW, Sydney, NSW 2052
Australia
AB:
Two of the largest sources of uncertainty in hydrological modelling arise from an inaccurate model structure and from
inadequate ways of describing the errors in our observed data. Recent computational advances have allowed for incorporation
of the effects of uncertainty using a variety of statistical techniques. However, the current approach to hydrological model
specification is to use a fixed structure to model the catchment and a fixed distribution of errors to describe the range of
possible catchment responses. How do we model what is inherently a dynamic system, when the majority of our modelling tools
are deterministic and static? This issue can also be extended to our understanding of the model errors. Is it sensible to
make the assumption that the structure of the errors in our data do not change over the range of model responses? What if we
could specify different error models to represent different sections of the hydrograph?
A way of formally implementing these ideals presents itself in a class of statistical models known as Hierarchical Mixtures
of Experts (HME). HME models provide a method of combining the performance of several models in a single framework. The
approach allows for a more sophisticated method of model aggregation by allowing individual models to be weighted based on
the preceding catchment conditions. The approach gives greater flexibility to the specification of the model errors, by
allowing a combination of different error structures to apply.
An innovation of the HME framework is that it provides a way of assessing the components of our existing models to see which
best describe observable hydrological processes. However the challenge in applying the HME framework to catchments for
predictive purposes lies in determining how to calculate which model should be selected depending on the state of the
catchment. A case study is presented that shows how the method can be used as a way to summarise model and data uncertainty
more effectively than current approaches. We investigate the usefulness of different catchment predictors to weight the
individual models. The study shows that given careful comparison of the possible mechanisms related to a switch in the
catchment `state' the approach can be a useful predictive tool, giving an aggregated model simulation that is better than
any individual model. The implications and opportunities for HME in a PUB context are illustrated through application of the
approach to a number of different catchments of varying hydrological characteristics.
DE: 1805 Computational hydrology
DE: 1847 Modeling
DE: 1873 Uncertainty assessment (3275)
SC: Hydrology [H]
MN: Fall Meeting 2005