Effective stochastic streamflow simulation method based on Gaussian mixture model

Abstract

Stochastic simulations of streamflow sequences are essential for water resource planning and management. In this study, a new Gaussian mixture model (GMM)-based method is proposed to generate long-term synthetic streamflow sequences of any time scale. First, the GMM model, which is established based on the K-means algorithm, expectation–maximization algorithm, and Akaike’s information criteria, is employed to describe the temporal dependences of the observed streamflow data. Second, the accuracy and reliability of the constructed GMM are verified by the Kolmogorov–Smirnov test of goodness of fit. Then, daily, 10-day, and monthly streamflow sequences at a single site were generated using the proposed streamflow simulation method. Finally, the performance of the proposed method was verified based on the absolute and relative performance indexes. Yichang gauging stations on the upper Yangtze River were selected as case studies, and two copula methods were selected as the contrast methods to evaluate the proposed method. The results demonstrate that the proposed method can effectively capture the statistical properties of observed streamflow data at low, middle, and high time scales and can comprehensively preserve the lower- and higher-order statistics as well as the linear and nonlinear correlations of observed streamflow data. The proposed method is an effective method for stochastic streamflow simulations..