Copula-based method for stochastic daily streamflow simulation considering lag-2 autocorrelation

Abstract

Daily stochastic streamflow simulation is widely used for the design of reservoirs, evaluation of reservoir operation rules, and risk evaluation of operation of water resources systems. The major difficulties and challenges of daily streamflow are that there are 365 days that need to be simulated, which entails much more calculation than does monthly streamflow simulation. Since lag-2 auto-correlation is usually large, the lag-2 correlations should be considered. This paper therefore proposes a copula-based method for daily stochastic streamflow simulation. The contribution and novelty of this paper are that: (a) the proposed method can consider lag-2 correlations, compared with the currently used copula based method; (b) the conditional copulas are used to build high dimensional copulas, which make calculations easier; and (c) the method can be used for daily streamflow simulation because of the simplified model structure and effective parameter estimation method. Seven gauging stations on the upper Yangtze River and Pearl River in China were selected as case studies. Results demonstrated that the proposed method preserved the basic statistics (including mean daily flow, standard deviation, and coefficient of skewness) of observed data of each day well. Comparison with the currently used seasonal autoregressive model (SAR(2)) and bivariate copula-based method considering lag-1 autocorrelation indicated that the proposed method produced smaller relative errors and was better overall. Therefore, the proposed method can be regarded as an effective way for stochastic daily streamflow simulation, and can be used for the design of reservoirs and risk analysis of water resources systems.