Monsoonal impacts on water quality in the Baitarani River, Odisha: a comprehensive evaluation using water quality indices, multivariate statistics, and machine learning models for sustainable pollution control
Abhijeet Das
Abstract
The Baitarani River in Odisha is a crucial source of water for millions of people across various districts and urban centres in the state. Given its importance, regular monitoring and maintenance of its water quality are essential to ensure the sustainability of this resource. Among the various techniques used to evaluate surface water, the Water Quality Index (WQI) is one of the most widely accepted and effective methods for summarizing and assessing overall water quality. This study employed the Weighted Arithmetic (WA)-Water Quality Index (WQI), Numerow’s Pollution Index (NPI), Overall Index of Pollution (OIP), Synthetic Pollution Index (SPI), Entropy (E); Multivariate analysis such as Cluster analysis (CA) and Discriminant analysis (DA); Multi-Criteria Decision-Making (MCDM) tools namely WASPAS (Weighted Aggregated Sum Product Assessment), TOPSIS (Technique for Order Preference by Similarity to Ideal Solution), and COCOSO (Combined Compromise Solution) approach, across 13 different sampling sites. According to the guidelines set forth by the World Health Organization (WHO), a total of sixteen parameters were examined in water samples. To enhance the precision of water quality, Machine Learning (ML) techniques used such as artificial neural network (ANN) and gaussian boosting regression (GBR) models; developed and tested for the duration of 2012–2024 (n = 12 years) in monsoon season. Turbidity, EC, and TDS readings exceeded the acceptable limits as per WHO, throughout the sampling period. The findings indicated that the quality of the water exhibited as 69.23% (WAWQI), 92.31% (NPI), 92.31% (OIP), and 69.23% (SPI), achieving an excellent to good level, across all sampling sites. The results of these four water quality phenomena showed that industrial production, agricultural practices, and human activity were the main sources of pollution affecting the surface water quality of the river at the affected locations. Spatial analysis utilizing Geographical Information System (GIS), finds that the most contaminated areas are in the north and east, which is made worse by the uncontrolled release of contaminated water. To ensure the safety of water required for agriculture, PI (permeability index) for all examined locations belongs to Class III (Poor) group. In contrast, almost all samples (100%) were considered suitable for irrigation, based on the observations gained from SAR (sodium adsorption ratio), MH (magnesium hazard), KR (Kelley’s ratio), RSC (residual sodium carbonate), and PS (potential salinity). However, (chloro alkaline indices) CAI-I and II indicates five samples that are put under Reverse Exchange zone, and rest eight goes to Direct Exchange zone. Surface water in this area has a wide range of SR (salinization ratio) concentrations ranging from 0.13 to 0.38. According to the CR (corrosivity ratio) criteria, approximately 1 sample is put under unsuitable category for irrigation. Based on the CI (chlorinity index) observations, the sampling locations were distinguished between Class I- Class V, with averaging CI value of 684.62. In contrast, ASI (agriculture stability index) values ranged from 75 to 256 in the study area, averaging 149.15. While observing the results, > 53% of the water samples were suitable for farming activities. The MSI (mineral saturation index) findings revealed that the amount of value in the surface water system was around 23.08%, 46.15, and 30.77%, indicating Mineral dissolution, Stable, and Precipitation interface. The most significant % is from weathering of volcanic and metamorphic rocks, and ion exchange processes, making most of the water samples unsuitable for irrigation. A great majority of SMI (sea water mixing index) results, 61.54% of the water samples obtained from the study area demonstrated under fresh water zone, as indicated from the interpolation map. However, the study also identified a small percentage, approximately 38.46% of the water samples deemed in the category of salt-water mixing. The Piper diagram delineates three distinct types, including Ca2+-Mg2+-Cl-, Ca2+-Cl- and Ca2+-HCO3- types. Notably, almost 90% of surface water samples lie inside the field of rock dominance when examining Gibb’s figure, highlighting the importance of water-rock interactions and rock weathering. Multivariate statistical approach involving CA differentiated physicochemical parameters and sampling sites into three cluster groups that emphasized the main sources of pollution are irrigation systems, agricultural practices, and water-rock interactions. As per DA observations, ten important discriminating water quality indicators (pH, turbidity, EC, DO, BOD, Sulphate, Ca2+, Mg2+, Na+ and K+) are extracted by stepwise method. In terms of EWQI, 30.77% emerged as the poor grade of water quality, accompanied with 4 sites, followed by medium (15.38%), good (38.46%) and excellent water zone (15.38%). According to the analysis using WASPAS (0.45–0.89), TOPSIS (0.17–0.78), and COCOSO (1.38–3.13), it was found that K-(8, 11–13, 2) promotes unsuitability for drinking and reflects diverse pollution levels. The key indicators affecting the water quality are EC, TDS, and turbidity. In these areas, because of the slow water flow, contaminants can accumulate more easily. Consequently, strict control mechanisms must be put in place for these indications. For WQI and irrigational indices, two machine learning models were used: the ANN and GBR. The ANN model outperformed the GBR model. The findings demonstrated that ANN-2F was the most accurate prediction model, with the strongest correlation between EWQI and extraordinary features. For example, this model possesses two attributes that are essential to the EWQI forecast. The outputs’ R2 values for the training and validation sets are 0.97 (RMSE = 2.49) and 0.95 (RMSE = 2.175), respectively. From the adopted techniques, it is determined that this study's uniqueness resides in its all-encompassing strategy, which integrates several approaches to pinpoint and comprehend the causes of surface water contamination. Therefore, this study could be very helpful in setting priorities for the key locations for managing pollution and water resources. Clinical trail number Not applicable. Water Quality Prioritization in the Baitarani River Watershed, Odisha, India. Impact assessment of water pollutants on physicochemical parameters. Identification of acceptable zones utilizing WAWQI- NPI, SPI, and OIP. Watershed prioritization based on EWQI, TOPSIS, WASPAS, and COCOSO method with CA, and DA for data reduction. Machine Learning Approaches such as ANN and GBR are applied for the analysis. In this river analysis, human activities, industrial pollution, and fertilizer application have a negative impact on water quality, introducing an excess of pollutants and nutrients in water. Proposed a standard operating procedure for drinking water suitability analysis. The study’s findings could assist researchers and policy makers in taking comprehensive actions for sustainable water management.