A geospatial study of the Mahanadi River and its distributaries around Paradip, Odisha, has concluded that most surface water sampled during the monsoon seasons of 2020 to 2023 is unsuitable for drinking without treatment. The open access research, published on 12 January 2026 in the journal Green Technology, Resilience, and Sustainability, combined geographic information system mapping with four complementary indexing and ranking methods, including an entropy-weighted index and a machine learning ensemble ranking model, to classify water quality at nine observation sites. The work was led by Abhijeet Das and drew on secondary monitoring data from the State Pollution Control Board of Odisha.
The researchers compiled twelve water quality parameters: pH, dissolved oxygen, biochemical oxygen demand, sulphate, fluoride, chloride, nitrate, phosphate, total hardness, chromium, total dissolved solids, and iron. Samples were collected during the monsoon months of July to September across a three-year monitoring period, a design choice the authors say was intended to reduce the influence of dilution on the measurements. Laboratory and field analyses followed standard procedures from the American Public Health Association, and quality assurance checks using duplicate and blank samples were applied. Most specimens fell within a five percent charge balance error range, a check used to validate the ionic consistency of the chemical analyses.
The first assessment tool, the Weighted Arithmetic Water Quality Index, produced scores ranging from 40.36 to 176.13, with an average of 90.16. Under this scheme, only two sites, H-1 with a score of 45.48 and H-2 with 40.355, achieved good water quality, meaning they would need at most conventional treatment before drinking. Three sites fell into the poor category, requiring extensive treatment, while four sites, including H-8 at 176.13 and H-9 at 133.46, were classified as very poor or unsuitable for human consumption. The authors attribute the worst scores to growing urban areas, household wastewater, and runoff from farms, with H-8 showing the highest readings of any location.
The entropy-weighted index, which assigns parameter weights objectively from the statistical variability of the data rather than expert judgment, told a similar story. Its values ranged from 67 to 266, averaging 145.89, spanning categories from good to extremely poor. None of the nine samples reached the excellent class, two sites were rated good, and seven sites scored above 100. By the percentage distribution reported in the study, 22.22 percent of samples fell into the good class while the remaining 77.78 percent were unfit for drinking purposes. The highest entropy scores, 266 at H-6, 182 at H-9, and 177 at H-8, were linked by the authors to urban waste and human activity along the river course.
The third method, fuzzy synthetic evaluation, a multi-criteria decision-making technique designed to handle ambiguity in environmental data, produced index values between 111 and 246. Under this classification no sample achieved excellent or good status, two samples were average, four were poor, and three were unsuitable for drinking. In percentage terms, 22.22 percent of sites were rated average, 44.44 percent poor, and 33.33 percent unsuitable. The authors report that elevated concentrations of biochemical oxygen demand, phosphate, total hardness, iron, fluoride, and total dissolved solids drove the poor classifications at seven sites, and that roughly 77.78 percent of the study area would require observation before the water could be used.
The machine learning component, an Ensemble Ranking model, consolidated the different ranking schemes into a single consensus score. Values ranged from 0.50 to 5.61, with lower numbers indicating better water. The lowest ranking, recorded at H-2, indicated exceptionally good water quality, while H-6 scored 5.61, H-8 scored 4.87, H-9 scored 3.56, and H-7 scored 3.50, marking these as the most degraded locations. The ensemble analysis identified H-6 as the most affected site during the study period, a finding consistent with its high entropy index and elevated levels of biochemical oxygen demand, fluoride, iron, total dissolved solids, and total hardness. According to the study, more than half of the samples were deemed unfit for human consumption under this evaluation because of elevated contaminants.
The underlying chemistry helps explain why the indices were so severe. Biochemical oxygen demand, a measure of the oxygen consumed by microorganisms decomposing organic matter, ranged from 4.04 to 5.48 milligrams per liter, exceeding the guideline of 3 milligrams per liter at every one of the nine stations, with the study attributing this to sewage loads and organic pollution. Iron concentrations spanned 0.31 to 1.35 milligrams per liter against a threshold of 0.3, and fluoride ranged from 0.71 to 0.87 milligrams per liter, above the 0.6 milligram per liter limit cited in the study. Phosphate values of 3.2 to 4.12 milligrams per liter also exceeded the reference level, with the highest readings at H-4 and H-5 attributed to fertilizer application, animal waste, and municipal sewage.
Not every parameter signaled contamination. pH values of 6.96 to 7.20 remained within the recommended range for surface water, dissolved oxygen between 6.5 and 7.06 milligrams per liter met the applicable guidelines, nitrate stayed between 4.66 and 5.22 milligrams per liter, well below relevant limits, and all chromium measurements fell under the allowable level. Total dissolved solids ranged from 90.18 to 145 milligrams per liter, and total hardness from 72.54 to 155 milligrams per liter, with only site H-4 exceeding the ideal hardness threshold. Chloride was within limits everywhere except H-4, which recorded 265 milligrams per liter against a 250 milligram per liter benchmark.
The study is candid about its constraints. Because sampling was restricted to the monsoon season, the work cannot capture seasonal fluctuations in water levels and quality, and the reliance on secondary monitoring data from the pollution control board means the analysis does not distinguish between different causes of contamination. The authors also note that the study provides regional averages without the ability to differentiate localized hydrological behavior, and they recommend that future research incorporate additional variables such as land-use change and atmospheric conditions, alongside isotope-based tracing to identify contaminant sources.
Despite those limitations, the authors argue the integrated framework offers practical value for water managers. By combining entropy weighting, fuzzy evaluation, and ensemble ranking within a GIS environment, the approach distills many parameters into interpretable maps and scores that can pinpoint pollution hotspots and prioritize remediation. The team recommends continuous monitoring at the most degraded sites, treatment of surface water before use, and measures such as rainwater recharge and storage, check dam construction, and modern farming practices to improve water quality. The findings are presented as baseline data for researchers and local governing bodies in the Paradip area as they develop integrated, sustainable water management plans for the Mahanadi basin.
Subject of Research: Geospatial and machine learning assessment of monsoon-season surface water quality in the Mahanadi River basin near Paradip, Odisha
Article Title: Geospatial assessment of surface water quality on integrating entropy – machine learning techniques to predict water contamination in the Paradip area of Mahanadi River Basin, Odisha
Article References: Das, A. (2026). Geospatial assessment of surface water quality on integrating entropy – machine learning techniques to predict water contamination in the Paradip area of Mahanadi River Basin, Odisha. Green Technology, Resilience, and Sustainability, 6(1), Article 1. https://doi.org/10.1007/s44173-025-00023-7
Image Credits: AI Generated
DOI: 10.1007/s44173-025-00023-7
Keywords: water quality, Mahanadi River, Paradip, Odisha, entropy-weighted water quality index, machine learning, GIS, fuzzy synthetic evaluation, ensemble ranking, drinking water, water pollution, monsoon
News Source: Teresa Odom. (October 29, 2026). Machine Learning and Entropy Indices Reveal Most Mahanadi Water Near Paradip Unfit to Drink. Scienmag.
