Accelerate EHS with AI: Tracking, Monitoring, and Managing Chemical Contamination in Water

In today's data-driven world, AI isn't just on the map - it's the fast lane. From mining critical insights and tracking regulatory shifts to forecasting risk, monitoring employee health and safety metrics, and streamlining reporting, AI is steering the future of EHS, sustainability, and compliance. But as professionals take the wheel, what roadblocks and green lights lie ahead on the journey? This series of articles will help you navigate the future of AI in EHS.

Academics and scientists are exploiting artificial intelligence (AI) to help governments around the world measure chemical contamination in real time and track down polluters. In doing so, regulators and policymakers hope to respond to and mitigate pollution challenges more rapidly and with greater precision, something that they have hitherto struggled with.

Traditional environmental monitoring usually relies on several agencies working together across different jurisdictions to take field samples, carry out laboratory analyses, and manage remote sensors. That has frequently left governments on the backfoot. “These methods often fall short in dynamic, high-density pollution events - precisely the kinds of phenomena that now dominate the global pollution landscape,” says Ganna Pogrebna, a behavioral data scientist at the Alan Turing Institute, a national center of excellence for AI and data science, based in London.

AI models - with their sophisticated image recognition, machine learning capabilities, and data crunching powers - offer regulators the chance to get ahead of the game. “The appeal of AI lies not just in its analytical power, but in its ability to generate actionable intelligence,” says Pogrebna. “AI offers the potential to transition environmental governance from a retrospective and reactive model to one that is anticipatory, adaptive, and decentralized.”

AI is currently better at tracking pollution in water than in air. That's because water pollution can be more easily gauged through photo analysis and AI is especially skilled at quickly and accurately analyzing large amounts of visual data. Discerning gases in the atmosphere is more complicated, and those efforts are a few years behind what is already underway with AI and water contamination.

Aquatic AI

A recent review, carried out by researchers at Hangzhou Dianzi University's College of Artificial Intelligence, detailed how AI can learn much more from a photo than a human ever could. AI models have been trained to use specialist spectroscopy techniques that detect the vibrational and rotational states of molecular systems, which AI can use to compute the exact chemical composition of materials. Effectively, the Hangzhou researchers concluded, AI can turn videos and photos into a molecular fingerprint - identifying aerosols, liquids, and dissolved gasses within waterways.

But even with all that AI power, it still takes time because humans need to teach the models about the molecular behavior of each chemical that we want to identify. The models then need to screen photos and images repeatedly for each different chemical. As 3E recently reported, each piece of microplastic can contain thousands of different chemicals.

Xiaojing Li is a research fellow at the University of Birmingham's Centre for Environmental Research and Justice. Along with her colleagues, she is looking for a way to measure the effects of several different chemical contaminants at once.

Searching chemical by chemical is simultaneously time consuming and “oversimplified,” says Li, because chemicals don't necessarily stay isolated once they're released into the environment; they mix and interact with each other, which can change their toxicity. “The end result is a cocktail of toxicity,” she says.

This cocktail is what environmental agencies ought to be looking out for, she argues, and one way to do that is by teaching AI to monitor the genetic health of water fleas.

Problem Solving

Water fleas are microscopic crustaceans, and they are highly sensitive to chemical fluctuations in water. If changes in the activity of their genes suggest early signs of stress or harm, AI can help Li to reverse engineer the problem by screening for substances that could be to blame. “By using AI like this, we can identify subsets of chemicals that might be harming aquatic life. This can reveal that chemicals are harmful even at low concentrations that we wouldn't normally be concerned about. It can also reveal novel chemicals that aren't currently regulated but may be worth looking into,” says Li.

AI is also guiding pollution remediation efforts. In 2022, Pogrebna took part in a Bangladeshi project aimed at tackling “plastic islands” that float in and clog up the country's vast river deltas. “These islands are nasty things. You can stand on them and kids play football on them, it's incredibly dangerous,” says Pogrebna.

The plastic islands frequently move with river currents, and in the time between locating one and mounting a response, the island may have drifted somewhere completely different. The delta network is so complex and intertwined, that it's not as simple as just looking further downstream.

“We deployed cameras on boats to capture video footage,” says Pogrebna. “With an AI that knows how the water typically moves, you can predict where the plastic island is going to be a few days in advance.” That information is cherished in a resource-poor environment, says Pogrebna, because it enables targeted cleanups and reduces wasted effort.

Regulatory Enforcement

AI's ability to identify pollution and boost remediation efforts is a great help, says Pogrebna, but it's the chance to use the technology to pinpoint where pollution is coming from that makes it a potentially potent regulatory enforcement tool.

In a scientific paper published in 2024, researchers in China showed how pairing data such as water flow patterns with land use information can create “pollution path tracking models.” When those models detect the location of pollution clusters in rivers and analyze their chemical composition, they can simulate pollution dispersal pathways upstream and estimate the pollution source.

“By integrating data from satellite imagery, oceanography, and field samples, these systems can trace microplastics back to specific industrial activities or product categories,” says Pogrebna. “That's critical information for regulatory enforcement as they increasingly look to curb microplastics and other emergent pollutants.”

There is a world in the near future, says Pogrebna, where AI models play a key role in supporting investigators by helping to link pollutants to their sources. Although any findings would require human verification, they will form part of the evidence that leads to lawsuits against chemical companies - and the industry should ready itself for that heightened scrutiny.

Environmental Irony

There's also a sustainability elephant in the room that those in the AI world have yet to properly address. “If you think about it, it's a bit weird to use AI as a solution because it isn't a sustainable thing. It uses a lot of energy,” says Pogrebna. “The energy that you use in a single ChatGTP query is not insubstantial, and I don't think that people really think about that.”

AI also demands an enormous amount of water to cool its hardware. AI infrastructure may soon consume six times more water than Denmark, according to the United Nations Environment Programme.

Then there's the issue of supplying the rare earth materials that are needed to build the computers and data centers that underpin AI. The pace of AI's technological improvement also poses an environmental quagmire in the form of electronic waste left by machines that are rendered obsolete by unrelenting progress. That waste often contains toxic substances, which can ironically further pollute water bodies if proper recycling processes are not followed.

AI may well offer regulators a powerful tool for environmental protection, but it also presents them with a series of challenges that do not yet have obvious solutions.

Researchers have been aware of AI's setbacks for a while now. Matthew Lowe, a wastewater engineer at Stony Brook University, published a paper back in 2022 on AI's ability to monitor water quality, and he pointed out that scientists have yet to address several limitations if AI is to become widely used by regulators.

“Poor data management, low explainability, poor model reproducibility and standardization, as well as a lack of academic transparency are all important hurdles to overcome in order to successfully implement these intelligent applications,” Lowe concluded. In other words, it may already be possible for governments to track pollution and manage it with the help of AI, but researchers still need to perfect the process.

Given the speed of progress in recent years, however, Pogrebna says it's only a matter of time before researchers can iron out some of these kinks. “We have case studies and cities where AI pollution tracking has been very successful. That shows proof of concept, and the widespread adoption of this technology is the next logical step,” says Pogrebna. “It's a question of when, not if.”

Photo Caption: Island of plastic floats in a river in Bangladesh (Ganna Pogrebna).