Prediction of river inflow of the major tributaries of Indus river basin using hybrids of EEMD and LMD methods

Abstract

Reliable prediction of the water flow entering a reservoir is a crucial concern for agricultural-based countries like Pakistan. This study applies a hybrid method to model the daily river inflow of different tributaries of the Indus river basin (IRB), Pakistan. The hybrid method is composed of hybrid decomposition and three data-driven models. The hybrid decomposition (HD) is based on local mean decomposition and ensemble empirical mode decomposition that decompose river inflow series into different components. Next, the HD components are used as inputs to support vector regression (SVR), K-nearest neighbor (KNN), and autoregressive integrated moving average (ARIMA) models. The predictions of the HD-SVR, HD-KNN, and HD-ARIMA models are aggregated. The final prediction is the mean of the predictions of the HD-SVR, HD-KNN, and HD-ARIMA models (combined as HD-SKA). The potential of the HD-SKA model is explored on the Jhelum, Indus, Kabul, and Chenab rivers in the IRB system in Pakistan. The performance of the HD-SKA model is compared to twelve models using different performance measures. For the Chenab river, the root mean squared error (RMSE), mean absolute error (MAE), and root-relative squared error (RRSE) of the HD-SKA model on test data are 7.9314, 3.5315, and 0.2676, respectively, which are smaller than all competing models in the study. Similar findings are achieved for Kabul, Indus, and Jhelum rivers where the HD-SKA model outperformed all other considered models. The results show that the HD-SKA model has a superior capability of capturing the randomness of river inflow.