Component-based Reconstruction Prediction of Runoff at Multi-time Scales in the Source Area of the Yellow River Based on the ARMA Model

Improving the accuracy of hydrological prediction for long-term annual runoff series is important for water resources management and planning. In this study, the complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the runoff time series of the Tangnaihai hydrological station in the source area of the Yellow River (SAYR) from 1960 to 2017. The component reconstructions are evaluated with the Nash-Sutcliffe efficiency (NSE) coefficient, and the autoregressive moving average (ARMA) model is used to simulate and predict the runoff components at multi-time scales, which are also evaluated with the NSE coefficient. The results show that the NSE coefficient of the component simulations and predictions by ARMA are relatively high. Moreover, the NSE coefficients increase in step with the fluctuation period of the runoff component. For the runoff component-based reconstruction models at multi-time scales, the mean relative errors of simulation and prediction are low at 8.25% and 8.78%, respectively. In addition, the high-frequency components play an important role in the modal reconstruction as well as the simulation and prediction. Thus, focusing on the high-frequency components can improve the overall accuracy of runoff prediction based on the ARMA model at multi-time scales.