Reliabilty estimation of rainfall-runoff models

The predictions from rain-fall-runoff models used for flood forecasting are subject to several uncertainties. Flood estimates depend on the rainfall inputs (i.e., rainfall temporal and spatial distribution) and the values assigned to hydrologic model parameters (i.e., soil properties). Both meteorological and hydrologic models are uncertain in the sense that the models incorporate uncertainty with respect to data, parameters, and model structure in generating flood predictions.; The purpose of this research is to develop methods to study the reliability of rainfall-runoff models, and to identify effective measures to enhance model predictions. A Monte Carlo analysis is used to evaluate the effect of the uncertainties of a rainfall model combined with a runoff model on the reliability of the output hydrograph from a candidate rainfall-runoff model and to provide information on the probability of predicted discharge of floods produced by convective storm events. The probability is expressed by the exceedance probability distribution of the magnitude of the predicted peak discharges accounting for the various uncertainties from different contributing factors. The peak discharge probability for the rural portion of the Onondaga Creek is given as an example using a rainfall-runoff model that incorporates the Triangulated Irregular Network distributed hydrological model with a well-known stochastic rain-fall model with parameters derived from local radar observations.; The study of the reliability of prediction from watershed models provides useful information on the stochastic nature of output from rainfall-runoff models subject to uncertainties and identifies the relative contribution of the various uncertainties to unreliability of the model prediction. The work leads to two main conclusions: (1) the main source of the flood forecast uncertainty is due to uncertainties in the rainfall model and not those in the catchment model; and (2) the most sensitive variables on the predicted peak discharges are the spatial variability of the rain cells of the rainfall model and the overland flow process of the catchment model.