Impact of imperfect rainfall knowledge on the efficiency and the parameters of watershed models

It is crucial to analyze the sensitivity of watershed (rainfall-runoff) models to imperfect knowledge of rainfall input, in order to judge whether or not they are reliable and robust, especially if they are to be used for operational purposes. In this paper, a new approach to sensitivity analysis is proposed, based on a comparison between the efficiency ratings and parameter values of the models and the quality of rainfall input estimate (GORE and BALANCE indexes, assessing the quality of rainfall time distribution and the total depth respectively). Data from three watersheds of increasing size (71, 1120, and 10700 km2), are used to test three watershed models of varying complexity (three-parameter GR3J model and six-parameter modified versions of TOPMODEL and IHACRES). These models are able to cope with imperfect rainfall input estimates, and react to improvements in rainfall input accuracy by better performance and reduced variability of efficiency. Two different types of model behavior were identified: the models either benefit from improved rainfall data by producing more consistent parameter values, or they are unable to take advantage of the improvements. Although the watershed size seems to be immaterial, the smaller watersheds appear to need more precise areal rainfall estimates (a higher concentration of raingages) to ensure good modeling results.