A conceptual lumped rainfall-runoff flood event model was developed andapplied on the Gardon catchment located in Southern France and varioussingle-objective and multi-objective functions were used for itscalibration. The model was calibrated on 15 events and validated on 14others. The results of both the calibration and validation phases arecompared on the basis of their performance with regards to six criteria,three global criteria and three relative criteria representing volume,peakflow, and the root mean square error. The first type of criteria givesmore weight to large events whereas the second considers all events to be ofequal weight. The results show that the calibrated parameter values aredependent on the type of criteria used. Significant trade-offs are observedbetween the different objectives: no unique set of parameters is able tosatisfy all objectives simultaneously. Instead, the solution to thecalibration problem is given by a set of Pareto optimal solutions. From thisset of optimal solutions, a balanced aggregated objective function isproposed, as a compromise between up to three objective functions. Thesingle-objective and multi-objective calibration strategies are comparedboth in terms of parameter variation bounds and simulation quality. Theresults of this study indicate that two well chosen and non-redundantobjective functions are sufficient to calibrate the model and that the useof three objective functions does not necessarily yield different results.The problems of non-uniqueness in model calibration, and the choice of theadequate objective functions for flood event models, emphasise theimportance of the modeller's intervention. The recent advances in automaticoptimisation techniques do not minimise the user's responsibility, who hasto choose multiple criteria based on the aims of the study, his appreciationon the errors induced by data and model structure and his knowledge of thecatchment's hydrology.
展开▼
