Analysis of the time-step dependency of parameters in conceptual hydrological models

Abstract The estimation of parameters in deterministic hydrological models is highly disturbed by multiple, partly unkown error impacts. Many research activities have been devoted to the analysis of model uncertainty induced by such error impacts. In this context, spatial and temporal scale transfer issues have been emphasized as important sources of model uncertainty. In literature the spatial scale has been analysed in detail. Also variations of temporal scale impacts on parameters have been named, but not analysed in greater detail. Several model applications have proved empirically, however, that systematic linear and nonlinear parameter changes were induced by the application of different time steps. It is the hypothesis of this paper these findings, in particular that data resolution both of climatic model forcing as well as measurements used for parameter estimation including feedback interaction with model structures have a strong impact on the estimation of optimum parameters. In this paper the analysis of parameters depending on time steps is carried out with a physically defined soil moisture model with data from an experimental hydrological catchment in Austria where longterm data at high spatio-temporal resolution was available. The existence of near functional linear and non linear relationships is clearly confirmed empirically. The results might also be used for regionalisation or model application to ungaged basins, when parameters estimated with one time step are used in another basin with a different time step. It must be emphasized that the results and conclusions so far are only valid for the semi-distributed time discrete model dominantly based on infiltration excess runoff formation applied here in small catchments. It would be desirable to test the methodology with other model structures applied to the same catchments or the same model structures applied to different spatial scales. The hypothesis for future research is that the time dependency will occur for any time discrete model independent of its structure.