HR: 08:30h
AN: H41G-03 INVITED     [Abstracts]
TI: Parameter estimation in distributed and integrated hydrological models using Pareto optimisation
AU: * Madsen, H
EM: hem@dhi.dk
AF: DHI Water and Environment, Agern Alle 5, Horsholm, DK-2970 Denmark
AB: In the calibration of distributed and integrated hydrological models it is generally recognised that multiple sources of information should be applied. Multi-site and multi-variable caliŞbration and validation should be performed if distributed predictions are needed for different state variables. This calls for formulation of the calibration problem using a multi-objective framework. In this context model calibration can, in general, be performed on the basis of: (1) multi-variable measurements, such as groundwater levels, river flows, and concenŞtration measurements, (2) multi-site measurements consisting of several measurement sites distributed within the modelling domain, and (3) multi-response modes, i.e. calibration criteria that measure various responses of the hydrological processes such as the general water balance, peak flows, and low flows. When using multiple objectives, the solution to the calibration problem will not be a single unique set of parameters but will consist of the Pareto set of solutions (non-dominated solutions), according to various trade-offs between the different calibration criteria. An essential component of solving this problem is how to compare parameter sets when there are two or more calibration criteria. The Pareto domination approach does not rely on a single comparative measure but on whether one solution is dominated by another for the different calibration criteria considered. With this method the modeller avoids having to specify preferences to any of the calibration objectives at the calibration stage. Parameter optimisation based on the Pareto domination approach is a powerful method that has several important advantages: (1) It allows a comprehensive evaluation of the trade-offs between different calibraŞtion objectives and hence highlights possible model structural deficiencies, (2) It offers an elaborate framework for comparison of different models or model conceptualisations by considering several performance criteria in a consistent manner, (3) It allows the modeller or decision-maker to choose (or weight) solutions at a later stage according to the specific model application being considered, (4) It provides a better discrimination between model structure and parameter sets and hence a more well-posed optimisation problem by "unfolding" the equifinality problem often observed in hydrological model calibration, (5) It provides a natural solution to the weighting problem in generalised and weighted non-linear least squares regression. The proposed Pareto optimisation framework is demonstrated on a calibration of the MIKE SHE integrated, hydrological modelling system.
DE: 1804 Catchment
DE: 1805 Computational hydrology
DE: 1829 Groundwater hydrology
DE: 1846 Model calibration (3333)
DE: 1847 Modeling
SC: Hydrology [H]
MN: Fall Meeting 2005