The Mahanadi River, a lifeline for eastern India, is facing growing pressure from urban sprawl, farm runoff, and saltwater intrusion near its delta. A new study led by Abhijeet Das, a researcher at C.V. Raman Global University (CGU), reveals how seasonal shifts in water chemistry are reshaping the river’s suitability for drinking, farming, and industry—and how artificial intelligence could help turn the tide on pollution.
Over a year, Das and his team sampled surface water at nine locations along the Mahanadi in the Paradip region of Odisha. They measured twelve key parameters—from dissolved salts and metals to nutrients like phosphate and nitrate—then used advanced statistical tools to identify what’s driving contamination. The results were sobering: biochemical oxygen demand (BOD), phosphate, electrical conductivity, total dissolved solids (TDS), and fluoride emerged as the main culprits, especially in areas near cities and farmland.
“These pollutants don’t just make water undrinkable—they signal broader ecological stress,” Das notes. “When BOD spikes, it means organic waste is consuming oxygen faster than nature can replenish it. That’s not just bad for fish; it’s a warning sign for any industry relying on stable water supplies.”
The team calculated several water quality indices to classify the river’s health. The Weighted Arithmetic Water Quality Index (WQI) showed values ranging from 40.36 (good) to 176.13 (unsuitable), with over 77% of samples showing signs of pollution. The Overall Index of Pollution (OIP) painted an even clearer picture: only one-third of sites had acceptable water, while two-thirds were poor—often due to microbial risks. Yet, when it came to irrigation, the river fared better. Parameters like Sodium Adsorption Ratio (SAR) and Percent Sodium (%Na) suggested the water could still support crops in many areas, though careful monitoring would be needed.
Using multivariate analysis—correlation, cluster analysis, and principal component analysis—the team traced the pollution back to natural mineral dissolution and human activity. “The Piper diagram showed that magnesium-chloride-sulfate water types dominate,” explains Das. “That points to rock-water interactions, but the high sulfate and nitrate levels? Those are classic footprints of agricultural fertilizers and urban runoff.”
The study then took a leap forward by testing four machine learning models—Support Vector Machine (SVM), Random Forest (RFM), Gradient Boosting (GBM), and Extreme Gradient Boosting (XGB)—to predict water quality and map pollution hotspots. Among them, the Random Forest model stood out. With a Nash-Sutcliffe efficiency of 0.94 and an R² of 0.93, it outperformed the others in forecasting WQI values.
“This isn’t just about better science—it’s about smarter management,” says Das. “If industries, municipalities, and farmers can get real-time forecasts on water quality, they can adjust operations before problems escalate. For example, a thermal power plant drawing cooling water can temporarily reduce intake during high TDS or BOD events to avoid scaling or corrosion in pipes.”
The implications are especially relevant for the energy sector. Thermal and hydropower plants, desalination facilities, and even data centers all depend on reliable water quality. Fluctuations in TDS, chloride, or sulfate can accelerate equipment corrosion, reduce cooling efficiency, or trigger regulatory fines. With climate change intensifying seasonal variability, such AI-driven predictions could become a cornerstone of resilient water management.
Published in *Applied Water Science*—the journal of record for water quality research—the study doesn’t just diagnose problems; it offers a toolkit. By integrating GIS, machine learning, and conventional monitoring, Das and his team have shown how data can guide conservation, prioritize cleanups, and protect investments in water-dependent industries.
“Water isn’t just a resource—it’s infrastructure,” Das reflects. “And in a world where every drop counts, the future belongs to those who listen to what the water is telling us.”