Monthly Precipitation Prediction Based on the CEEMDAN-BMA Model

Forecasting rain is essential for the alleviation and management of floods, environmental flows and water demand in different sectors. Precipitation is affected by various meteorological factors and has strong nonlinear characteristics, which significantly hinders its ability to be predicted. To improve the accuracy and robustness of prediction results, this paper proposes a precipitation ensemble forecasting model (CEEMDAN-BMA model) based on complete ensemble empirical modal decomposition (CEEMDAN) and Bayesian model averaging (BMA) methods using monthly precipitation data from Beijing and Guangzhou stations from January 1950 to December 2020 to explore the model’s validity. The ensemble prediction results of the CEEMDAN-BMA model were analysed based on six evaluation indices. The results show that the CEEMDAN-BMA model performs well in terms of monthly precipitation prediction for both the Beijing and Guangzhou stations. The RMSE, MAE, and R² values of the monthly precipitation prediction results for the Beijing station are 22.355 mm, 14.973 mm, and 0.897, respectively, and the RMSE, MAE, and R² values of the monthly precipitation prediction results for the Guangzhou station are 35.86 mm, 28.371 mm, and 0.932, respectively. In addition, the CEEMDAN-BMA model provides a 90% confidence interval (CI) to quantify the uncertainty of the prediction results. The coverage of the 90% CI of the CEEMDAN-BMA model for the Beijing station is 91.67%, the average width is 82.76 mm, and the average offset is 0.009 mm; the coverage of the 90% CI for the Guangzhou station is 96.67%, the average width is 143.84 mm, and the average offset is 0.059 mm. Compared with those of the other models, the prediction results of the CEEMDAN-BMA model are superior.