Abstract: It is important to evaluate the effect of forecast uncertainty on the decision of reservoir operation. Previous studies usually use the traditional Copula to describe uncertainty information in the short-term forecasting of streamflow sequences, but such methods are ineffective and unsatisfactory. We adopt Vine Copula to characterize the uncertainty of streamflow forecasting, and realize a quantitative evaluation of the uncertainty in the conditions of different streamflow levels and different lead times. And the effect of prior information on the subsequent forecasting uncertainty can also be analyzed. Application of this new method to the Jinxi Reservoir shows that Vine Copula can pass the hypothesis test and achieve the best fitting effect compared to the traditional Copula model, and statistically its simulation error is the lowest against the measured data. When the prior information in the same planning period is used, the expectation of its relative forecasting errors and its 90% confidence level interquartile range can be significantly reduced, thus lowering forecasting uncertainty.

Key words: streamflow forecasting, forecasting uncertainty, Vine Copula, prior information, multivariate probability transformation