Improved water level forecasting in inland rivers using CEEMDAN and BiDR-xLSTM: a case study on the Xijiang River

Accurate water level forecasting is essential for managing inland river systems and mitigating hydrological extremes. This study presents a novel hybrid model combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and a bifurcated and double-residual extended long short-term memory (BiDR-xLSTM) network to address challenges posed by non-stationary signals and complex spatiotemporal dependencies. The proposed CEEMDAN-BiDR-xLSTM model was evaluated using the Xijiang River hydrological dataset (XJRHD), demonstrating superior short-term forecasting performance. It achieved a coefficient of determination (R2) of 0.94 for the 6-h forecasting horizon and 0.89 for the 12-h horizon, outperforming benchmark models. These results highlight the model's robustness and effectiveness in enhancing hydrological forecasting accuracy, with promising implications for flood warning and sustainable water resource management in complex inland river environments.