Improving the precision of monthly runoff prediction using the combined non-stationary methods in an oasis irrigation area

Abstract

The optimized agricultural water schedule and management, especially in the large irrigation district along the river basins, were closely related to the spatial and temporal runoff variations. However, the impacts of climate change and human activities leads to non-linear and non-stationary monthly runoffs. Under this condition, effectively capturing the variation of the runoff time series and improving the accuracy of the prediction model are of vital importance. However, some existing monthly data-driven prediction studies mainly focus on the model structure and calculation load, ignoring the influence of each frequency component of runoff and the fluctuation of runoff series caused by meteorological factors, which may not effectively capture the potential change process. In this paper, a novel method (CEEMD-MPE-EMD-GRU) for the non-stationary monthly runoff prediction was proposed. It fully took advantages of the complementary ensemble empirical mode decomposition (CEEMD), multi-scale permutation entropy (MPE), empirical mode decomposition (EMD) and gated recurrent unit (GRU). The combined model in general is a data-driven model, and compared with the traditional mechanism model, its most notable advantage is that it successfully overcomes the redundant information of the prediction model. In addition, atmospheric input factor analysis is added on the basis of fully decomposing and identifying non-stationary pseudo-components. The hydrological runoff data (1956–2014) obtained from the Manas River locating at Xinjiang, China were used for prediction. The results indicated that the new CEEMD-MPE-EMD-GRU model reached higher accuracy, as its Nash-Sutcliffe efficiency coefficient (0.960) was significantly larger than those of the GRU model (0.813) and the CEEMD-GRU model (0.889). Meanwhile, the root mean square error and the absolute relative error of the CEEMD-MPE-EMD-GRU model decreased to 0.279 and 0.195, respectively. The new runoff prediction model established in this paper would provide more precise evaluation of the monthly runoff prediction and better guidelines for high-efficiency agricultural water scheduling in the irrigation district.