Reliability evaluation of groundwater quality index using data-driven models.

Abstract

A trustworthy evaluation of the groundwater quality situations for different usages (i.e., drinking, industry, and agriculture) can definitely improve the management of groundwater resources for quality and quantity control, particularly in the arid and semi-arid districts. In the present investigation, GQI values and their typical categories have been yielded by the World Health Organization (WHO) instruction for the Rafsanjan Plain, the central part of Iran, during a 15-year period beginning in 2002. In this study, four robust Data-Driven Techniques (DDTs) based on the evolutionary algorithms and classification concepts have been applied to present formulations for the prediction of groundwater quality index (GQI) values in the case study of Rafsanjan Plain. In this way, monthly groundwater quality parameters (i.e., electrical conductivity, total hardness, total dissolved solid, pH, chloride, bicarbonate, sulfate, phosphate, calcium, magnesium, potassium, and sodium) were taken from 1349 observations. Performance of DDTs indicated that the Evolutionary Polynomial Regression (EPR) demonstrated the most accurate predictions of GQI than a model tree (MT), gene-expression programming (GEP), and Multivariate Adaptive Regression Spline (MARS). Moreover, to investigate all probable uncertainty in the values of groundwater quality parameters for the Rafsanjan Plain, a reliability-based probabilistic model was designed to assess the values of GQI. Hence, the Monte-Carlo scenario sampling technique has been quantified to evaluate the limit state function from DDTs. Moreover, there is a high probability (almost 100%) for the whole region to pass the "Excellent" quality, but it reduces to almost 50% over the "Good" and leads to almost 0% for the "Poor" quality.