Hydro-Geochemical Evaluation of Glacial-Streams to Find Their Suitability for Drinking and Irrigation Purposes Using Supervised Machine Learning Approach

This research evaluates the suitability of glacially-fed water resources for drinking and irrigation purposes in Gilgit-Baltistan, Pakistan. A total of 40 water samples were collected and their physio-chemical characteristics were analyzed in the laboratory by measuring their electrical conductivity, turbidity, pH, total dissolved solids (TDS), hardness, and ionic concentrations (Ca2+, K+, Mg2+, Na+, Cl-, SO4-, NO3-, Pb, Cu, Cd, Zn, Fe, Mn, Hg, Cr, and As). The ion charge balance error of physico-chemical parameters was found within the permissible limits. Hydro-geochemical analysis by Gibs plot and Piper diagram illustrated that carbonate weathering governs the geochemistry of the study area. Suitability analysis showed that the majority of the samples are suitable for irrigation purposes. Furthermore, the Water quality index (WQI) for drinking purposes classified the water quality as excellent (50%), good (32.5%), and medium (17.5%). For data modeling, this research uses the Logistic regression (LR) and Random Forest (RF) to predict water quality class (WQC) based on predefined water quality classes. Machine-learning results showed that the LR model performed comparatively better than RF model to predict water quality class. LR has 90% accuracy, 90% F-1 score, and 10% classification error. Whereas, RF has 80% accuracy, 71% F-1 score, and 20% classification error. The outcomes of this study can be used as reference by the local government and policy makers to develop guidelines for water quality at domestic and irrigation usages.