Water-Quality Prediction Using Multimodal Support Vector Regression: Case Study of Jialing River, China

Abstract

Accurate water quality prediction can support the early warning of water pollution in water resource management. However, it remains challenging because of hydrological uncertainties in the single scale. A multimodal water quality prediction model based on ensemble empirical mode decomposition (EEMD) and support vector regression (SVR) is proposed to address this problem. Based on the idea of decomposition and ensemble, dissolved oxygen (DO), one representation in water quality, is firstly decomposed into several intrinsic mode functions, which are then modeled by the SVR in each mode. According to the reconstruction theory of the EEMD, the subresults predicted by the SVR of each mode are summarized into the final results. The proposed model is evaluated by the weekly DO concentrations during 2014–2015 from one monitoring station of the Jialing River, China. A back propagation neural network and the standard SVR models are used for a comparison study. The results demonstrate that the addressed model, combining the mode representation capacity of the EEMD and the nonlinear mapping of the SVR, has the best prediction performance among the peer models.