Author(s): Freddy Soria; So Kazama; Masaki Sawamoto
Linked Author(s): So Kazama
Keywords: Distributed model; Poorly gauged basins; Sensitivity analysis
Abstract: This paper aims to assess the process of parameter estimation and model performance evaluation in terms of spatial and temporal predictive capability. The emphasis of this study is on the issues and consequences of model parameter values estimation and model output sample space uncertainty. Visual and statistical evaluation tools are used as evaluation tools, where the Sobol variance based sensitivity analysis method (Sobol, 1993) and a simple visual evaluation of parameter identifiability are the main resources. Three basins with poorly gauged conditions (limited information available at various spatial and temporal resolutions) are selected considering their different topographic characteristics. The application to the cited scenario demonstrated the importance of sensitivity analysis as a fundamental element in the analysis of any mathematical model and the correspondent outputs. Visual examination of model parameter sample space or any other analysis should only be carried if the previous mentioned analysis is completed. After a screening analysis procedure, a staged procedure was employed to reduce the sample space uncertainty and therefore the model output variance. First, visual parameter identifiability is analyzed based on the objective functions considered. Later, visual analysis of hydrographs variability and empirical considerations were used to restrict even more the sample space. Results reduced model output uncertainty to the minimum possible level (for the study case). Sample space uncertainty was reduced only for identifiable parameters. Further efforts should be concentrated in a more detailed evaluation of the temporal variability of sensibility indices, their link to dominating processes in a catchment and their spatial distribution.