For years, people living in the shadow of refineries, coke ovens and chemical plants have complained about the air that burns their lungs, blackens their window sills and coats their gardens in soot. Yet on paper, many of those same facilities appeared to be within legal limits for toxic emissions. That illusion is now collapsing. A detailed ProPublica investigation reveals that when regulators finally forced a group of major polluters to measure what was really in the air at their fence lines, the results were far worse than the self-reported numbers the Environmental Protection Agency had been using for decades.

Instead of relying on continuous monitoring, the EPA has long allowed industrial plants to estimate how much benzene, vinyl chloride and other hazardous chemicals they release by plugging generic “emission factors” into spreadsheets. These factors, some developed in the 1970s and 1980s, often come from a small number of tests and assume that a particular process always leaks roughly the same amount of pollution. In reality, sprawling facilities like coke plants and chemical complexes contain miles of pipes and thousands of valves, flanges and tanks, any one of which can leak unpredictably. The agency’s own technical documents acknowledge that many of these factors are rated as “poor” or “below average” in accuracy, yet they still underpin national emissions inventories and even the government’s 2020 AirToxScreen risk maps.

To see how those estimates hold up against reality, the EPA in 2023 told 20 large facilities – including coke works, chemical plants, and iron and steel mills – to install temporary fence-line monitors and report the results. ProPublica obtained and analyzed that data. In 97 percent of the comparisons, measured pollution at the fence line was higher than what the company’s own estimates would have predicted, often by staggering margins. At U.S. Steel’s Clairton Coke Works near Pittsburgh, benzene levels were about 37 times higher than the modeled figure based on the plant’s reported emissions. At a Dow Chemical complex in Plaquemine, Louisiana, monitors found vinyl chloride roughly 156 times higher than expected, and levels of ethylene dichloride – another cancer-causing chemical – a thousand-fold higher than the model suggested. Several other plants, including facilities operated by Indorama and Formosa in Texas, would have exceeded the EPA’s new health standards for key toxics by wide margins.

Nowhere is the human cost more visible than in Clairton, Pennsylvania. The coke plant there has been baking coal into coke for steelmaking since 1916. It emits benzene, fine particles and sulfur compounds into a narrow river valley where the geography traps polluted air during temperature inversions, leaving residents walking through a brownish haze. Families in nearby towns describe a long list of illnesses: asthma attacks that send children to the emergency room, older residents who can no longer climb a flight of stairs without stopping for breath, and clusters of cancers that are hard to prove scientifically but impossible for locals to ignore. County records show more than $10 million in fines against U.S. Steel over alleged air-pollution violations in recent years, and a 2018 fire at the facility blanketed the valley in fumes and coincided with a spike in emergency visits for asthma.

The core problem runs through the math the government has relied on. Emission factors were originally developed as a rough tool to understand national trends and support models like AirToxScreen. Over time, facilities began using them as the main way to report their own emissions to regulators. Watchdog reports and even the EPA’s inspector general have repeatedly warned that these factors can produce wildly misleading results when used as a substitute for real-world measurements, especially at complex sites full of potential leaks. Yet those same estimates are still used to calculate how much cancer risk is acceptable for people living downwind.

The new investigation shows how much difference direct monitoring can make. A decade ago, under President Barack Obama, the EPA forced oil refineries to install benzene monitors around their fence lines and take action when the annual average concentration exceeded a set “action level.” Monitors quickly uncovered refineries emitting far more benzene than their models had indicated, including one in New Mexico where real-world levels were over 100 times higher than predicted. After companies were required to find and fix leaks, benzene readings around many refineries fell sharply. According to an Environmental Integrity Project analysis, the number of U.S. refineries above the federal benzene action level dropped by about half between 2020 and the end of 2023, a sign that public reporting and fence-line rules can drive real reductions.

Building on that success, the EPA under President Joe Biden finalized new rules in 2024 that extended fence-line monitoring to coke plants, large chemical facilities and some metal plants. The coke-oven rule required facilities like Clairton to keep annual average benzene at the fence line below a strict limit, on the logic that cutting benzene – which often leaks alongside other toxics – would also reduce overall emissions. Additional rules targeted more than 120 major chemical plants and a handful of iron and steel mills, especially in regions like Louisiana’s “Cancer Alley,” where clusters of plants have pushed lifetime cancer risks to the very edge of what the EPA labels acceptable. The agency estimated that about 90,000 people living near these chemical plants faced an unacceptable risk of developing cancer, and that stronger limits and monitoring could cut that number to around 3,000, a 97 percent reduction. Millions more – more than 5 million people in total – live near facilities that would have been covered by the new standards.

At the same time, EPA projections and outside analyses suggested that expanding fence-line monitoring to hundreds of plants could reduce toxic air emissions by thousands of tons per year. The Environmental Integrity Project and Environmental Health News note that the 2015 benzene rule has already pushed many refineries to clean up, and that the 2024 expansion of monitoring to more than 200 chemical plants is expected to further cut emissions of benzene, ethylene oxide, chloroprene and other dangerous substances. For frontline communities that have waited years for stricter oversight, these rules looked like the first real chance to see, in hard numbers, what they were breathing – and to force companies to stop the worst leaks.

That progress is now in jeopardy. Shortly after President Donald Trump returned to office, his administration announced it would reconsider or roll back 31 environmental protections, including the three rules that require fence-line monitoring at coke, chemical and metal plants. While that long regulatory process plays out, the EPA has offered two-year exemptions that let companies postpone compliance. Roughly 50 plants, including the Dow facility in Plaquemine and chemical complexes owned by Indorama and Formosa in Texas, have received or requested such extensions. The EPA acknowledges there is a “discrepancy” between self-reported emissions and the fence-line monitors, but says it has not yet determined why. Industry groups and several senators argue that the standards are too strict and too costly for facilities that face global competition, even though the EPA estimates that fence-line monitoring itself would cost a typical coke plant about $107,000 per year – a small sum compared to the billions in revenue of multinational steel and chemical companies.

For people living near these plants, the delay feels like yet another sacrifice of their health in the name of deregulation. In communities around Clairton, residents who once proudly worked for U.S. Steel now wonder whether decades of exposure contributed to cancers that killed relatives and the breathing problems that restrict their own lives. Some have moved away after losing multiple pets to cancer and falling ill themselves, describing the decision as “escaping with our lives.” In Louisiana’s industrial corridor, where EPA’s own AirToxScreen mapping tool shows some of the nation’s highest estimated cancer risks from air toxics, neighbors who already rely on inhalers and air purifiers say pausing the new rules feels like being told they must keep living inside an uncontrolled experiment.

Experts who have inspected industrial plants for years say the pattern uncovered by the fence-line data is unsurprising. Facilities with sprawling networks of pipes, pumps and tanks are notoriously hard to model accurately without direct measurements, they note. Even if companies are not intentionally low-balling their emissions, using generic emission factors for such complex sites is, in the words of one former EPA inspector, “little better than fantasy.” Radiation, noise and water pollution are routinely measured directly at many industrial sites. For hazardous air pollutants that can cause leukemia, lymphomas and other cancers, critics ask, why should regulators settle for rough guesses when better tools now exist?

The new monitoring data has turned that technical debate into a moral one. When fence-line monitors reveal that real-world emissions are dozens or hundreds of times higher than what the spreadsheets said, the “acceptable” cancer risks used to justify the status quo start to look like a dangerous mirage. Communities from Pittsburgh to Port Neches and Point Comfort have spent years testifying, filing lawsuits and pushing for basic transparency about what they are forced to breathe. The longer the EPA leans on unreliable estimates and delays fully implementing fence-line monitoring, the longer millions of people will live with an invisible burden of toxic exposure they never agreed to bear. In towns already scarred by illness and loss, the question now is how many more years of breathing miscounted pollution regulators – and politicians – are willing to accept.