Spatial prediction of temporary and permanent hardness concentrations in groundwater based on chemistry parameters by artificial intelligence

• Published in: International journal of environmental science and technology (Tehran) Vol. 20; no. 6; pp. 6665 - 6684
• Main Authors: Mousavi, M., Qaderi, F., Ahmadi, A.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.06.2023
• Subjects: Aquatic Pollution; Earth and Environmental Science; Ecotoxicology; Environment; Environmental Chemistry; Environmental Science and Engineering; Original Paper; Soil Science & Conservation; Waste Water Technology; Water Management; Water Pollution Control; Support vector regression; Neural network; Groundwater; Hardness; Chemical parameters
• ISSN: 1735-1472; 1735-2630
• DOI: 10.1007/s13762-023-04934-5

Considering the role of water in human health, its quality must be inspected. For the development of management programs to control groundwater quality on a large scale, due to the lack of data in physical models, the spatial distribution of permanent and temporary hardness concentrations in groundwater is calculated using geographical information system. Permanent and temporary hardness point concentrations and spatial chemical data as predictors are used to train statistical models. A feedforward neural network (NN) method and support vector regression (SVR) method is employed to predict temporary and permanent hardness concentrations of Amol–Babol plain using parameters of water, such as electrical conductivity (EC), sodium adsorption ratio, total hardness, pH and concentration of sulfate, Cl − , nitrate, K + . For this purpose, the qualitative data collected from 855 wells during two seasons in six years were used. Then, by creating a feedforward NN method and SVR model, the optimum model was selected, trained, and tested. Based on mean square error, coefficient of determination ( R 2 ), and root-mean-square error criteria, the obtained results were 748.76, 0.9303, and 27.36 for feedforward NN at the permanent hardness and 504.78, 0.9688, and 22.46 for temporary hardness, respectively. These results are 231.61, 0.9304, and 15.218 for permanent hardness and 480.74, 0.9833, and 21.92, respectively, for temporary hardness in the SVR. The obtained results showed the capability of created models in modeling, which predicted hardness concentration in the desired area with an acceptable approximation.