Stochastic Evaluation of Rainfall-Runoff Prediction Performance

Abstract

Given a set of realizations of error data (i.e., the difference between model runoff estimates and stream gauge data) from rainfall-runoff hydrologic models, it is possible to generate a set of error transfer function realizations that, when convoluted with a suitable kernel function such as the hydrologic model output, equate to the original error data. In turn, these error transfer function realizations may be used to generate synthetic error data that is convolved from a separate design storm modeled runoff and the generated error transfer function realizations. The synthetic error data set is then added to the design storm modeled runoff to produce a set of equally likely outcomes for the model prediction. The set of equally likely outcomes is statistically analyzed to provide, for instance, a confidence interval for the possible outcomes of the design storm model. A four-section algorithm is presented that performs each of these tasks.