Hydrological model performance comparison through uncertainty recognition and quantification

In this study, a methodology which is capable of identifying the uncertainty of prediction through recognizing and quantifying the different sources of uncertainty in hydrologic models is applied for model comparison. The methodology is developed to recognize and quantify different uncertainty sources through observing hydrologic model behaviour under increasing input uncertainty levels. Based on the methodology, an index, which originates from the Nash-Sutcliffe efficiency named Model Structure Indicating Index (MSII) developed by the authors is applied to evaluate the reliability of model structure. A ranking of the adequacy of the hydrologic models to the watershed can be achieved by applying MSII. The hydrologic models Storage Function Method (SFM), TOPMODEL and KW-GIUH are used for model quantitative comparison in this study. Of these, a parameter-con strained SFM is used as an example of a poor-structured model; and two versions of TOPMODEL with different vertical flux calculation processes are used to demonstrate the behaviour of different model components. The results show that, at small input uncertainties, no distinction can be made between the capabilities of the hydrologic models to adapt themselves to error-contaminated data. As the input uncertainty increased, however, the distinction between the models becomes larger and the accuracy of the model structures could be quantified through MSII. The results prove that the index can be used as a tool for implementing quantitative model comparison/selection. Copyright (c) 2007 John Wiley & Sons, Ltd.