Novel Hybrid Approach for River Inflow Modeling: Case Study of the Indus River Basin, Pakistan

This study introduces a novel hybrid model for predicting daily river inflow, combining the Hampel filter (HF) for outlier correction, local mean decomposition (LMD) for initial signal decomposition, and ensemble empirical mode decomposition (EEMD) for further decomposition into intrinsic mode functions (IMFs) and residue. The innovative aspect of this model lies in its dual decomposition strategy (LMD-EEMD) followed by prediction using the K-nearest neighbor (KNN) algorithm, resulting in the HF-LMD-EEMD-KNN (HLEK) approach. This combination aims to enhance the accuracy and reliability of inflow predictions. The model's performance was evaluated using river inflow data from four rivers in the Indus River Basin, with key metrics including root relative squared error (RRSE). In the training phase, the HLEK model achieved MAE values of 7.072, 5.859, 2.308, and 3.709 for the Indus, Kabul, Jhelum, and Chenab rivers, respectively, significantly outperforming traditional models. The study concludes that the HLEK hybrid model significantly improves prediction accuracy over simpler models, providing a robust tool for forecasting river inflows. This enhanced accuracy is crucial for water resource management and planning in the Indus River Basin and potentially other regions.