HR: 1340h
AN: H43A-0483    [Abstracts]
TI: Strategies for Calibration of Distributed Hydrological Models: From Steady State to Fully Integrated Models
AU: * Blasone, R
EM: rsb@er.dtu.dk
AF: Institute of Environment & Resources, Technical University of Denmark, Bygningstorvet 115 , Lyngby, 2800 Denmark
AU: Madsen, H
EM: hem@dhi.dk
AF: DHI Water & Environment, Agern All‚ 5, Horsholm, 2970 Denmark
AU: Rosbjerg, D
EM: dr@er.dtu.dk
AF: Institute of Environment & Resources, Technical University of Denmark, Bygningstorvet 115 , Lyngby, 2800 Denmark
AB: Local search methods have been widely applied in calibration of distributed groundwater models. However, the complex shape of the response surface of models of increased complexity, such as distributed and integrated hydrological models, might cause these procedures to fail in locating the global optimum. Distributed models give rise to a potentially huge number of parameters to calibrate and require a proper methodology to deal with multi-site and multi-objective measures and objective functions. Global search algorithms have demonstrated to perform well for calibration of complex models, although at a more expensive computational cost. The main purpose of this study is to investigate the performance of a global and a local parameter optimization algorithm, respectively, the Shuffled Complex Evolution (SCE) algorithm and the gradient-based Gauss-Marquardt-Levenberg algorithm (as implemented in the PEST software), when applied to distributed hydrological models of increasing complexity, ranging from a simple recharge steady state groundwater model to a fully integrated hydrological model. The hydrological model used in this work is the MIKE SHE model, which allows to easily change the model formulation. The study site is the Karup catchment in Denmark that has been already extensively investigated and for which comprehensive hydrological data are available. The calibration is conducted in a multi-objective context, where the error measures related to groundwater levels and streamflow are aggregated and simultaneously minimized. The convergence efficiency of the two procedures is compared and the solutions are evaluated according to the Pareto dominance criterion to deal with the trade offs between the objective functions. The results show a strong dependency of PEST solutions on the starting point of the search and the risk for this procedure of being trapped in local regions of attractions, even for the simple steady-state groundwater model. Despite these features, PEST normally guarantees a fast reduction of the objective function reduction (faster than SCE). The SCE algorithm is, in general, more effective and provides a wider range of solutions according to the different trade-offs between objective functions. A combined application of the global and local search techniques has also been investigated. The SCE method is used as a starting screening procedure to approach the regions where the best solutions are located and subsequently PEST is employed to refine the estimation of the optimum.
DE: 1829 Groundwater hydrology
DE: 1836 Hydrological cycles and budgets (1218, 1655)
DE: 1866 Soil moisture
DE: 1869 Stochastic hydrology
DE: 9820 Techniques applicable in three or more fields
SC: Hydrology [H]
MN: Fall Meeting 2005