Hybrid method for river inflow prediction: an integration of Hampel filter, decomposition techniques, and support vector machine

Hydrological modeling plays an important role in the management of available water resources globally. In this paper, we have developed a new hybrid model to predict daily inflow series. This approach is a novel composition of the Hampel filter (HF), ensemble empirical mode decomposition (EEMD), variational mode decomposition (VMD), and support vector machine (SVM) model. Firstly, the outlier correction is performed using the HF to remove the unusual and randomness in the inflow series. Secondly, the EEMD is employed to remove the noise of the HF-treated inflow series. Thirdly, the VMD decomposes the denoised inflow series into different modes which are fed to the SVM model. The predictions of modes are obtained and aggregated to determine the final predictions of the proposed HEVS (HF–EEMD–VMD–SVM) hybrid model. The performance of the new hybrid model is demonstrated on the main tributaries of the Indus River Basin (IRB) of Pakistan using different performance measures. The tributaries include the Indus River, Chenab River, Kabul River, and Jhelum River. The results showed that for Chenab River the proposed HEVS hybrid model has 57.99, 48.91, 46.07, 38.3, 36.27, 24.68, and 18.35% lower MSE values than the SVM, HF–SVM, EEMD–SVM, VMD–SVM, and EEMD–VMD–SVM models in the testing phase. Similar results are for the remaining three rivers. The Diebold-Mariano test showed that the accuracy of the HEVS hybrid model is higher than all the competing models in the study. The new approach can be helpful in river flow management and avoid issues of droughts, heat waves, and floods.