Application of Water Quality Index and Multivariate Statistical Techniques to Assess and Predict of Groundwater Quality with Aid of Geographic Information System

In this study, the groundwater quality and spatial distribution of the Basra province in the south of Iraq was assessed and mapped for drinking and irrigation purposes. Groundwater samples (n = 41) were collected from deep wells in the study area to demonstrate, estimate and model the Water Quality Index (WQI). The analysis of water samples integrated with GIS-based IDW technique was used to express the spatial variation in the study area with consideration of WQI. The physicochemical parameters, including pH, sodium (Na+), electrical conductivity (EC), chloride (Cl-), total dissolved solids (TDS). calcium (Ca2+), nitrate (NO3-), sulfate (SO42-), magnesium (Mg2+), and bicarbonate (HCO3-) were identified for groundwater quality assessment. The results of calculated WQI classify groundwater into three sorts. The results of WQI showed that 2.5%, 2.5% and 95% of the groundwater samples were classified as poor/very poor/unsuitable for drinking, respectively. The GIS tools integrated with statistical techniques are utilized for spatial distribution and description of water quality. Correlation analysis of groundwater data revealed that some parameters have actually a relationship that is strong with the other parameters and they share a common source of origin. Multivariate statistical techniques, especially cluster analysis (CA) and factor analysis (FA), were applied for the evaluation of spatial variations of forty-one selected groundwater samples. Cluster analysis confirmed that some different locations of wells have comparable sourced elements of water pollution, whereas factor analysis yielded three factors which are accountable for groundwater quality variations, clarifying more than 72% of the total variance of the data and permitted to group the preferred water quality. Multi-Layer Perceptron (MLP) models were applied in modeling the water quality index. Comparing different result values of the MLP network suggested that the values of MSE and r for the selected model are 0.1940 and 0.9998, respectively. Finally, it can be revealed that the MLP network precisely predicted the output, i.e. the WQI values.