Sensitivity Analysis of Distributed Rainfall-Runoff Models

Abstract

Two Monte Carlo based methods were applied in a mountainous watershed. The methods covered were the Generalized Likelihood Uncertainty Estimation Technique (GLUE) and the variance based Sobol’ technique. The former assess the likelihood of a model to describe a system recognizing that model and data are subject to uncertainty; the latter statistically evaluates the uncertainty in the estimation of model parameter values. Both are commonly used in model predictive uncertainty. This paper analyzed their applicability to distributed rainfall-runoff schemes. The methods were found to be complimentary: GLUE technique contributed the criteria for rejection/acceptance of behavioral models, and Sobol’ indices described the relative variance contribution of model parameters on total discharge. Bayesian updating within GLUE, and the spatial distribution of sensitivity indices were not covered, and can be the key to extend the analysis into fully distributed schemes.