Abstract: To effectively improve the accuracy in formulated operation schemes under the condition of quantifying reservoir runoff forecast errors, a Gaussian mixture model (GMM)- Copula stochastic model is developed to describe runoff forecast errors for multiple forecast periods. This model is based on the characteristic of the GMM that accurately describes the distribution of runoff forecast errors for a single forecast period due to good adaptive performance. It incorporates the advantage of the high-dimensional meta-student t Copula function through dynamically coupling multiple types of marginal distributions. In a case study of the Jinping I hydropower station in the Yalong River basin, we conduct stochastic simulations and analysis of the reservoir runoff forecast errors for forecast periods of 6 h, 12 h, 18 h and 24 h. The results show that for different forecast periods, simulation errors and actual errors share the same changing pattern and similar major statistical eigenvalues, meeting the predetermined accuracy requirement. This verifies that our stochastic model is feasible and effective and it helps formulate and implement reservoir operation schemes.

Key words: runoff forecast error, Gaussian mixture model, high-dimensional meta-student t Copula, stochastic model, Jinping I hydropower station