Incorporating the ICEEMDAN decomposition to improve the accuracy of models for drought prediction

【Objective】Drought is one of the most common abiotic stresses affecting crop production worldwide. Accurate forecasting is essential for improving irrigation management and water use efficiency. This study evaluates a multi-dimensional time series model for drought prediction based on the ICEEMDAN decomposition, aiming to provide a new method for improving drought prediction accuracy.【Method】The Santun River Irrigation District in Xinjiang was chosen as a case study. Monthly precipitation data from the Nianpanzhuang Station (1980—2023) were used, and the standardized precipitation index (SPI) was calculated for time intervals of 1, 3, 6, 9, 12, and 24 months. Six prediction models were compared: the autoregressive integrated moving average (ARIMA) model, gated recurrent unit (GRU) network, long short-term memory (LSTM) network, and their combinations with the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), resulting in ICEEMDAN-ARIMA, ICEEMDAN-GRU, and ICEEMDAN-LSTM models. These models were used to predict the SPI series at multiple time scales. Model accuracy was evaluated using root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R2).【Result】The accuracy of all six models improved as the time interval increased, reaching the highest at the 24-month interval. ICEEMDAN effectively stabilized the time series data and improved model accuracy for drought prediction. The accuracy of the models was ranked as follows: ICEEMDAN-ARIMA＞ICEEMDAN-GRU＞ICEEMDAN-LSTM＞ARIMA＞GRU＞LSTM.【Conclusion】Incorporating ICEEMDAN enhances the accuracy of drought prediction. Among the six models compared, ICEEMDAN-ARIMA was the most accurate and can be used for drought prediction in the studied region.