An experimental approach to reduce the parametric dimensionality for rainfall–runoff models
Sobol’ sensitivity analysis has been used successfully in the past to reduce the parametric dimensionality for hydrological models. However, the effects of its limitation, in that it assumes an independence of parameters, need to be investigated. This study proposes an experimental approach to assess the commonly used Sobol’ analysis for reducing the parameter dimensionality of hydrological models. In this approach, the number of model parameters is directly pitted against an efficiency criterion within a multi-objective genetic algorithm (MOGA), thus allowing both the identification of key model parameters and the optimal number of parameters to be used within the same analysis. The proposed approach was tested and compared with the Sobol’ method based on a conceptual lumped hydrological model (HSAMI) with 23 free parameters. The results show that both methods performed very similarly, and allowed 11 out of 23 HSAMI parameters to be reduced with little loss in model performance. Based on this comparison, Sobol’ appears to be an effective and robust method despite its limitations. On the other hand, the MOGA algorithm outperformed Sobol’ analysis for further reduction of the parametric space and found optimal solutions with as few as eight parameters with minimal performance loss in validation.