New River Water Quality Index Merges Pollution Data and Ecological Risk Into One Score

A river can look clean to a swimmer and still be dangerous to drink. That gap between what water is fit for and what it is used for lies at the heart of a new study published in the journal Environmental Management, in which researchers at Islamic Azad University’s Ardabil Branch in Iran have built a unified index that judges river water quality and ecological risk at the same time. The work, led by Pouria Rafiee with Hossein Saadati, Ebrahim Fataei and Fatemeh Nasehi, was tested on the Gharehsoo River in northwestern Iran, a waterway that drains a landscape of farms, towns and industry and has long been under pressure from untreated discharges. The result is a tool the authors call the River Water Quality Index, or RWQI, which for the first time folds sixteen physicochemical and microbial parameters together with the risks posed by heavy metals into a single, interpretable score.

Water quality indices are not new. Since the 1960s, environmental agencies have compressed long tables of laboratory measurements into single numbers that managers and the public can grasp quickly. The trouble, as the authors note, is that most existing indices answer only one question at a time. A drinking water index says nothing about whether the same water is safe for fish, and a pollution index says nothing about the cumulative toxicity of metals accumulating in sediments and food webs. For a river that serves drinkers, swimmers, farmers, factories and wildlife simultaneously, that fragmentation forces managers to juggle several incompatible scores, each with its own scale, weighting scheme and thresholds. The RWQI was designed to end that juggling act by producing one integrated assessment that still respects the very different standards that apply to each use.

The study’s first building block is the Enhanced River Pollution Index, or ERPI, a holistic monitoring framework developed in earlier work by Gupta and Gupta that the team adapted for five distinct water use categories: drinking, recreation, wildlife and fisheries, industry, and agriculture. Water samples were collected at seven stations along the Gharehsoo River across two seasons, yielding fourteen station-season samples that were analyzed for the full parameter suite following standard methods for water and wastewater examination. For each sample and each use category, the ERPI compares measured concentrations against the relevant regulatory benchmarks, classifying the water on a scale that runs from excellent down to unsuitable. This produced the reference dataset against which the new index and the machine learning models would later be judged.

The findings for the Gharehsoo River are stark. When the drinking water category was evaluated using the DD classification, which corresponds to water that can be consumed without any treatment, 64.28 percent of the samples fell into the unsuitable class. In other words, nearly two-thirds of the river water sampled could not be drunk even before any consideration of treatment costs, a clear signal of how heavily the river is burdened by pollution along its course. The picture changes dramatically, however, when the same samples are judged against standards for wildlife and fisheries. Using benchmarks from India’s Central Pollution Control Board within the ERPI-WF model, every single sample was rated good to excellent for supporting aquatic life. That contrast is itself informative: it tells managers that the river’s principal contaminants are ones that threaten human consumers rather than the ecosystems themselves, at least under the parameters measured.

With the reference classifications in hand, the researchers turned to prediction. Monitoring every river station continuously is expensive, so a model that can estimate index values from a reduced set of measurements would make routine surveillance far cheaper. The team compared two approaches: multiple linear regression, the classical statistical workhorse, and support vector regression, a machine learning method that maps inputs into a high-dimensional space where complex nonlinear relationships become linear. Two SVR kernel functions were tested, the polynomial kernel and the radial basis function, or RBF kernel, which allows the model to fit highly flexible decision surfaces. When the models were trained to reproduce the ERPI values for each water use category, the SVR variants proved exceptionally accurate, with coefficients of determination approaching one in many cases, meaning nearly all of the variance in the reference index values was captured by the predictions.

The centerpiece of the study is the RWQI itself, which combines the use-specific water quality assessments with an ecological risk evaluation of heavy metals, drawing conceptually on the sedimentological risk framework introduced by Hakanson in 1980. Rather than asking separately whether water is clean enough for a given purpose and whether its metal load endangers ecosystems, the RWQI merges both dimensions into a single value that flags locations where either dimension, or both, is compromised. Applied to the Gharehsoo dataset, the index successfully identified critical points along the river. The Samian station emerged as a critical location for recreational use, meaning that swimming there carries elevated concern and that this reach of the river should be a priority for pollution control and public health warnings.

One of the study’s most methodologically interesting contributions is its treatment of uncertainty. Every index and model output carries some degree of doubt, arising from measurement error, natural variability and the assumptions baked into the formulas, yet most water quality studies simply report point values as if they were exact. The researchers quantified uncertainty using Markov Chain Monte Carlo, or MCMC, a Bayesian computational technique that samples thousands of plausible parameter combinations to build a full probability distribution of results rather than a single number. The MCMC analysis revealed that the RWQI exhibits higher uncertainty than the simpler models, a consequence the authors attribute to the inherent complexity of the integrated framework and to the way combined pollutant risks propagate through the calculation. Importantly, this is presented not as a flaw but as honest bookkeeping: a composite index that blends many parameters and risk terms naturally carries more uncertainty than a single-purpose score, and knowing the size of that uncertainty is essential for defensible decisions.

The practical implications reach well beyond one Iranian river. Because the RWQI delivers an integrated verdict on quality and ecological risk, it can direct scarce remediation resources to the stations and seasons where they matter most, and it can reveal conflicts that single-purpose indices hide, such as water that is safe for fish but unsafe for drinking. The near-perfect performance of the SVR models suggests that monitoring programs could predict index values at unmeasured locations or times from a smaller panel of indicators, cutting laboratory costs while preserving decision-relevant information. The authors position the RWQI as a powerful tool for integrated monitoring of water quality and ecological risk assessment in river management, enabling more targeted protection strategies, and the framework is transferable to other rivers provided the appropriate local standards and metal risk benchmarks are substituted.

The study also sits within a broader shift in water science toward machine learning and uncertainty-aware assessment. Recent years have seen support vector machines, neural networks and hybrid optimization methods applied to everything from streamflow prediction to lake ecosystem health diagnosis, and reviews of water quality index models have repeatedly called for frameworks that handle multiple uses and risk dimensions coherently. By coupling a holistic pollution index with ecological risk evaluation, validating the result with high-performing regression models, and then stress-testing the whole edifice with Bayesian uncertainty analysis, the Ardabil team has offered a template for what rigorous, integrated river assessment can look like. For the Gharehsoo River itself, the message is urgent: drinking water quality is critically compromised along much of its length, and the tools now exist to pinpoint exactly where intervention will do the most good.

Subject of Research: Development of an integrated river water quality and ecological risk index combining physicochemical parameters, heavy metal risk, and machine learning prediction

Article Title: Developing an Innovative River Water Quality Index Model for Ecological Risk Assessment in Rivers

Article References: Rafiee, P., Saadati, H., Fataei, E., & Nasehi, F. (2026). Developing an Innovative River Water Quality Index Model for Ecological Risk Assessment in Rivers. Environmental Management, 76(10), Article 333. https://doi.org/10.1007/s00267-026-02615-w

Image Credits: AI Generated

DOI: 10.1007/s00267-026-02615-w

Keywords: river water quality index, ecological risk assessment, heavy metals, support vector regression, multiple linear regression, Markov Chain Monte Carlo, uncertainty analysis, Gharehsoo River, water quality monitoring, machine learning, environmental management, drinking water