The potential of multiscale EWT-DSE-AI techniques for river water quality modeling with considering the uncertainty of the models

Water quality has a great impact on the environment and human life. So far, various regression and non-linear methods have been used in predicting hydrological phenomena. However, due to the nonlinearity and uncertainties of water quality parameters, these methods do not always lead to accurate predictions. Therefore, this study focused on introducing new multiscale methods combining Empirical Wavelet Transform (EWT), Differential Symbolic Entropy (DSE), and artificial intelligence methods to investigate the river water quality in terms of the Electrical Conductivity (EC). The AjiChay River basin located in Iran was selected as study area. For removing the noise of the raw time series, the datasets were first decomposed using EWT. Then, for quantifying the complexity of input time series the DSE amounts of the obtained components were computed and the most important subseries were inserted into the several intelligence approaches to predict the EC. Finally, the modeling uncertainty was investigated via the Lower Upper Bound Estimation (LUBE) approach. Results showed that the developed multiscale methods obtained significantly better outcomes compared to the single methods. From the sensitivity analysis, the Na⁺ concentration was the most important variable in the EC modeling. In general, the proposed methods could be useful models for river water quality investigating due to having desirable degree of reliability.