EVALUATION OF PREDICTIVE UNCERTAINTY IN DISTRIBUTED RAINFALL-RUNOFF MODELS

Abstract

This paper aims to raise the growing importance of predictive uncertainty evaluation in distributed modeling. The ideas are illustrated by applying a particular rainfall-runoff model in four catchments with different characteristics. Sensitivity analysis and parameter identifiability, as complimentary uncertainty measures allowed us to individually evaluate the suitability of model components. The Sobol implementation is affected by the sample size and the correlation degree among parameters, observed after the total sum of individual variance contributions exceeded a theoretical total variance of 1.0, in magnitudes that varied from 5% in dry season to 60% in wet season, turning more difficult straightforward interpretations. Topographic index was used to judge the distributed performance of the model, adequately describing patterns of total discharge in homogeneous catchments with slopes ranging from 10% to 30%, and failing in less homogeneous catchments with low slope landscape.