An optimized approach for predicting water quality features and a performance evaluation for mapping surface water potential zones based on Discriminant Analysis (DA), Geographical Information System (GIS) and Machine Learning (ML) models in Baitarani River Basin, Odisha
Globally, ecosystems, and human health are in danger because river water quality is deteriorating. Surface water resources are running low as a result of things like severe climate change and overexploitation. Estimating water output and locating possible surface water zones are therefore crucial. T...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-01-01
|
| Series: | Desalination and Water Treatment |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1944398625000554 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Globally, ecosystems, and human health are in danger because river water quality is deteriorating. Surface water resources are running low as a result of things like severe climate change and overexploitation. Estimating water output and locating possible surface water zones are therefore crucial. This study examines the surface water quality of Baitarani River, Odisha, which faces serious problems with water quality because of a variety of home, industrial, and environmental causes. This study provides a comprehensive analysis that carefully combines scientific data with the opinions of the local people, using ArcGIS tools to map the water quality at every point along the river. Water samples were collected from thirteen sampling locations. On this basis, physicochemical parameters related to water quality, including 19 parameters, were investigated at the chosen sites. To achieve this goal, this investigation relies on Drinking (Dr)-Water Quality Index (WQI) and multivariate statistical techniques including discriminant analysis (DA), to assess the variations of surface water quality in the selected region. Again, this research used a strong methodology by incorporating Machine learning (ML) algorithms, such as: Artificial Neural Network (ANN), Gaussian Process Regression (GPR), Support Vector Machine (SVM), and Linear Regression Model (LRM), were applied to forecast and confirm the quality of the water. The water quality parameters were employed to determine whether surface water was suitable for residential use by contrasting them with the World Health Organization (WHO). The obtained pH value shows fluctuations, that arising maximum at site X-(3) and the minimum value is observed at X-(8). The Ca2 + value ranged between 12.83 and 25.85 mg/L, while the recorded Cl- value, fluctuate in a range of 7.3–22.12 mg/L. Reduced DO at X-(8) reflects the pressing need for actions to safeguard the river and the towns that depend on it, and is in line with community concerns about the quality of the water. In the research, performance of Dr-WQI method, that generates a mean value of 85.36, indicating good to poor water quality. Eight locations represent polluted sources. In addition, DA has been utilized to validate the result obtained from cluster analysis (CA), thereby facilitating the identification of the variables that distinguish the observed groups. By employing Discriminant Analysis, five water quality parameters such as TH, TDS, Na+, DO, and BOD, were successfully identified, with 100 % assignment rate. From the discriminant scores, it indicates site X-(8), and (11-13), as heavily contaminated as a result of agricultural practices, which are distinguished by their variety and intensity of crops. The DA's spatial distribution map is created with the assistance of inverse distance weighted (IDW) method, that is by-default performed by Arc GIS. Proceeding further, ANN model showed exceptional accuracy in water quality prediction, highlighting the dependability and accuracy of errors in the current work, that includes MAE, MSE, MAD, and RMSE, but also record highest in the validation of goodness-of-fit metrics (NSE and R2). In addition, other models such as SVM, GPR and LRM was evaluated using eight criteria. The approaches' correlation value was nearly 0.99, indicating exceptional prediction accuracy. Up to 6 % perturbations were used to test the ANN model's performance, and it maintained its accuracy and robustness. In conclusion, a detailed comprehension of water quality issues is made possible by combining scientific data with local perspectives. This highlights the critical need for focused interventions to protect the river ecosystem and the welfare of the communities that depend on it, and it also offers insightful information for sustainable environmental management. |
|---|---|
| ISSN: | 1944-3986 |