Application of data-driven methods to predict the sodium adsorption rate (SAR) in different climates in Iran
- PDF / 1,361,792 Bytes
- 19 Pages / 595.276 x 790.866 pts Page_size
- 25 Downloads / 169 Views
ORIGINAL PAPER
Application of data-driven methods to predict the sodium adsorption rate (SAR) in different climates in Iran Elham Rahnama 1 & Omolbanin Bazrafshan 1 & Gholamreza Asadollahfardi 2 Received: 18 May 2020 / Accepted: 19 October 2020 # Saudi Society for Geosciences 2020
Abstract Water quality is a very important factor for drinking, irrigation, and industry usage. Sodium adsorption rate (SAR) is one of the water quality parameters, which has a vital role in irrigation because of decreasing infiltration in the soil. For future planning of irrigation, forecasting of SAR is necessary. We applied radial basis function (RBF), adaptive neuro-fuzzy inference system-grid partitioning (ANFIS-GP), ANIFIS-SC (SC is subtractive clustering), and ANFIS-FCMC (FCMC is fuzzy c-means clustering), to predict SAR of Aras, Sefid-Rud, Karun, and Mond Rivers, which cover part of the north, northwest, southwest, and south of Iran. We used 15 years’ monthly data, including discharge, pH, chloride (Cl), sulfate (SO4), bicarbonate (HCO3), and electrical conductivity (EC), as input to neural networks. The RBF neural network reached proper results of SAR prediction in training and testing in the Aras, Sefid-Rud, Karun, and Mond Rivers, in which the mean square error (MSE) for training was between 0.000255 and 0.0017 and coefficient of determination (R2) of testing varies between 0.902 and 0.96 using normalized data. ANFIS-GP did not reach the proper model performance (R2 were between 0–0.21 and 0.55 and index of agreement were between 0.55 and 0.88) for the four rivers in comparison with ANFIS-SC and ANFIS-FCMC, since ANFIS-GP may need a long time of processing to analyze numerous combinations of the membership function. The normalization of input data of ANFIS-FCMC improved the performance of the model of the four rivers slightly. However, the normalization of input data of ANFIS-FCMC for SAR prediction of the Aras River did not improve the performance of the model (R2 = 0.48 and index of agreement = 0.44). We could state that variable processing via normalization might not necessarily improve ANFIS model performance. The training of ANFIS did not reach appropriate results for the Mond River because of scattering data. We may state that ANFIS is not a proper method for using scatter data. The results of sensitivity in Sefid-Rud River use the RBF neural network indicated that the electrical conductivity had a maximum effect (19.8%) and the discharge (13%) had the minimum effect on SAR prediction. Keywords RBF neural network . ANFIS . SAR . Water quality . Iran
Introduction
Elham Rahnama [email protected]
Water quality management is one of the important sections of water management, and one of the roles of the management is to determine the suitability of water quality for different usage such as drinking, industrial, and irrigation. Considering water quality parameters, sodium adsorption rate (SAR) is one of the important ratios in agricultural proposes. The SAR formula is
Omolbanin Bazrafshan [email protected]
Data Loading...
