An optimization based framework for water quality assessment and pollution source apportionment employing GIS and machine learning techniques for smart surface water governance

Spatial evaluation of the region is associated with the assessment of the quality of water. Dump sites pose a significant threat to surface water resources due to the possibility of leachate contamination. The study investigated the impact of water quality deterioration on surface water quality, where water is utilized for drinking, agricultural, and industrial purposes. The present study was based on the acquisition and analysis of primary field data collected and evaluates the water’s physicochemical properties and appropriateness for drinking in the Mahanadi River, Odisha. The geospatial approaches namely, Inverse Distance Weighted (IDW) has been implemented to draw the interpolated maps for different water quality parameters. Water samples were collected from 19 different sites to evaluate 20 water quality indicators during the pre-monsoon (PRM) season. The duration of assessment of the current research is associated for six years (2017–2023). Surface water quality assessment sought to determine whether surface water was suitable for drinking using integrated approaches such as: Decision-Making Trial and Evaluation Laboratory (DEMATEL/De)-based water quality index (WQI). In addition, the study area's hydro-chemical facies were examined, and machine learning models’ hyperparameters such as Random Forest (RF), Borda Scoring Algorithm (BSA), Decision Tree (DT), Multilayer Perception (MLP), and Naïve Bayes (NB), were executed before, to training and testing the samples of surface water. The drinking suitability of the water was validated by existing standard equations and plots. Obtained results primarily revealed that nitrogen (TKN), and coliform (bacteriological indicator, TC) are at elevated ranges (> WHO prescribed criteria), and this indicate anthropogenic and geogenic processes regulate water quality in the study region. The study determined that computed De-WQI values indicate that 31.58% (n = 6), 15.79% (n = 3) and 5.26% (n = 1) of tested locations indicated as poor/very poor/unsuitable water while, remaining 36.84% of investigated sites referred to the zone of excellent water, that includes 7 survey locations. The pollution levels at these locations were more strongly associated with a wide range of expanding human activities, including excessive water use, fertilizer effects, agricultural runoff, and the expansion of industries in and around the river corridor. Decision-making models, combined with ML models, including BSA mechanism, were implemented to resolve conflicts, including WQI index. The calculated value indicate location Y-(9) as the most polluted place, followed by two locations such as Y-(8 & 19), respectively. This was evident from the greatest De-WQI readings at this location, which were 410, 276, and 281. According to the findings, the main source of pollution in the river is organic debris from homes. Furthermore, the accuracy, specificity, and sensitivity values of the models were used to compare their overall performance. With a sensitivity of 99.87%, a specificity of 74.56%, and an accuracy of 90.50%, the RF algorithm performs better than any other classifier. The machine learning model may be utilized to continuously monitor the water pollution in surface water after the accuracy result was compared to previous studies. Therefore, the urgent measures for restoring and improving the river water quality include regulating the upstream and downstream in compliance with regional spatial planning regulations, implementing conservation efforts, and changing people's behavior to be more considerate of environmental management for catchment areas.