Predicting groundwater quality for irrigation suitability on agricultural practices using machine learning, fuzzy logic and GIS techniques

Abstract

The study focuses on assessing the irrigation water quality index using modern techniques applied to 188 groundwater samples of Arjunanadi River basin (ARB) in southern India. Based on World Health Organization standards, 38 % of pre-monsoon samples exhibit higher electrical conductivity values exceeding 1500 μS/cm. Among the machine learning (ML) algorithms used for predicting irrigation water quality variables, the artificial neural network model demonstrated superior performance with 97 % accuracy. For both monsoon seasons, fuzzy logic models were employed to evaluate irrigation water quality parameters, revealing that all samples were suitable based on sodium absorption ratio. The United States Salinity Laboratory (USSL) diagram indicates that 76 % of pre-monsoon and 72 % of post-monsoon samples fall within the C3S1 zone, suggesting suitability for irrigation across diverse soil varieties with a reduced risk of convertible sodium. The Wilcox diagram classifies 62 % of pre-monsoon and 76 % of post-monsoon samples having good water quality for irrigation. Additionally, Doneen's diagram shows that 72 % of pre-monsoon and 44 % of post-monsoon samples are appropriate for agricultural practices. The overall fuzzy analysis indicates that 89 % of region in the study area is appropriate for agricultural practices. The findings of this study will enable government representatives and legislators raise public awareness on application of groundwater irrigation, which will help achieve Sustainable Development Goals (SDGs) 2 and 6.