Keywords
conceptual rainfall-runoff model, flow components, sobol’s indices, model variance, dominant processes
Start Date
1-7-2012 12:00 AM
Abstract
This paper deals with a conceptual rainfall-runoff model in which thetotal flow is obtained as the sum of the flow of individual flow components, such assurface flow, interflow and groundwater flow. In contrast to classical sensitivityanalyses, where the sensitivity of the total flow to each parameter is analysed, weshow here the results of variance and sensitivity analyses carried out for each flowcomponent. It was observed that the variance for the total flow spans a range offour orders of magnitude. A comparison with the variances of each flow componentallows identifying the process with the highest variance at each time step, whichcan be regarded as the dominant process. With respect to the first order indices itwas seen that high values are common when the related flow component has ahigh variance, while interactions are predominant when the respective process isnot as important. These interactions often involve parameters that were designedfor describing other processes, illustrating in this way how parameters can have anindirect effect on many processes. It is concluded that such an analysis motivatesthinking in terms of the processes. Specifically, it is possible to structure the periodinto different segments, depending on the most important process and to analysethese segments as a group, facilitating the identification of patterns.
Global sensitivity analysis for the flow components of a conceptual rainfall-runoff model
This paper deals with a conceptual rainfall-runoff model in which thetotal flow is obtained as the sum of the flow of individual flow components, such assurface flow, interflow and groundwater flow. In contrast to classical sensitivityanalyses, where the sensitivity of the total flow to each parameter is analysed, weshow here the results of variance and sensitivity analyses carried out for each flowcomponent. It was observed that the variance for the total flow spans a range offour orders of magnitude. A comparison with the variances of each flow componentallows identifying the process with the highest variance at each time step, whichcan be regarded as the dominant process. With respect to the first order indices itwas seen that high values are common when the related flow component has ahigh variance, while interactions are predominant when the respective process isnot as important. These interactions often involve parameters that were designedfor describing other processes, illustrating in this way how parameters can have anindirect effect on many processes. It is concluded that such an analysis motivatesthinking in terms of the processes. Specifically, it is possible to structure the periodinto different segments, depending on the most important process and to analysethese segments as a group, facilitating the identification of patterns.
