Abstract: Analyzing the variation trends in runoff series is a basis for hydrological forecasting. How to decompose a nonstationary nonlinear runoff series into a stationary series becomes a difficult issue in the current research of hydrology. This paper presents a coupling model of local mean decomposition (LMD) for time-frequency decomposition and maximum entropy spectral estimation (MESE) for the frequencies of each decomposed component. Monthly runoff series of the Xiahui hydrometric station, located upstream of the Miyun Reservoir, is used to verify the decomposition results and extract more effective frequency information from these series. The results show that LMD gives good decomposition results for the series, and MESE reveals one dominant cycle in each of the components: 2 months, 6 months, 18 months, 5 years, 11 years, and 22 years respectively. This demonstrates LMD-MESE coupling is a new useful approach to analysis and prediction of hydrological time series.

Key words: Miyun Reservoir, local mean decomposition, maximum entropy spectrum estimation, runoff analysis