Study on Runoff and Drought Prediction Based on VMD-BiLSTM

Against the backdrop of global warming, hydrological droughts occur frequently in the Ganjiang River Basin, threatening basin water security and replenishment of water resources in the middle and lower reaches of the Yangtze River. To accurately predict the evolution of runoff and the Standardized Runoff Index (SRI), this study collected monthly runoff and meteorological data from 1980 to 2020, established the "input feature-prediction target" correspondence, constructed Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), and Variational Mode Decomposition-Bidirectional Long Short-Term Memory (VMD-BiLSTM) models, and conducted predictions of monthly runoff and 3-month scale SRI (SRI-3). The prediction performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2). The results show that monthly runoff exhibits significant periodicity, while SRI-3 fluctuates sharply and both exhibit a significant upward trend; the VMD- BiLSTM model, which decomposes non-stationary sequences via Variational Mode Decomposition (VMD), achieves optimal predictive performance. For the test set, the RMSE and R2 for monthly runoff prediction are 305.4 m3/s and 0.96, respectively, and those of SRI-3 prediction are 0.174 and 0.97. Compared with the LSTM and BiLSTM models, it more accurately captures sequence trends and extreme values, effectively avoiding fitting biases. This study provides technical reference for basin drought early warning and optimal water resource allocation.