Industries / Energy, water and waste / Environmental services / Air quality monitoring
Testing at the stack
Stack testing measures what comes out of a chimney, smokestack or ship's funnel, by eye, by drawing a sample through a probe, or with analyzers that run all the time.
7
Milestones achieved, from 1897 to 2026.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1897
A gray card for smoke
Maximilian Ringelmann, a professor of agricultural engineering in Paris, devised a card of gray shades, made from black grid lines on white, that an observer holds up beside a smokestack to grade its smoke from No. 0 (clear) to No. 5 (black). William Kent introduced it to the United States in Engineering News on 11 November 1897. The US Geological Survey printed copies from 1908, the Bureau of Mines took the job over in 1910, and many city, state and federal rules came to set smoke limits in Ringelmann numbers.
LessonThe Bureau warned that a reading depends on the light, the sun's direction, the depth of the plume and the color of the particles, and that results should not be given in fractions "because of variations in physical conditions and in the judgement of the observers.".
Credit
Maximilian Ringelmann; William Kent; the US Geological Survey and Bureau of Mines
1971
Stack tests written into law
On 23 December 1971 the US Environmental Protection Agency (EPA) published performance standards for new plants together with the test methods for checking them. Method 5 draws stack gas through a probe at the same speed as the gas flows (isokinetic sampling), catches the dust on a glass-fiber filter kept at about 120 °C and weighs it. Method 9 has a trained observer record a plume's opacity, the share of light it blocks, every 15 seconds; observers must pass a "smoke school" test of 25 black and 25 white plumes from a smoke generator, with no reading off by more than 15% opacity and an average error of no more than 7.5%, and retest every six months.
LessonThe method says readings against a low-contrast background tend to come out low, and that such a low bias makes it less likely that a plant is cited because of observer error.
Credit
US Environmental Protection Agency
2012
Cameras replace the eye
On 15 May 2012 EPA approved ASTM D7520, the digital camera opacity technique, in place of Method 9 at Hill Air Force Base, and listed the approval (ALT-082) among its broadly applicable alternative methods. Software reads opacity from digital photographs of the plume. The approval held the camera system to the human test: at least four independent users had to read 300 certification plumes with it, within the same 15% and 7.5% limits.
LessonA 2010 presentation from Virginia's Department of Environmental Quality said the state spent about $9,000 a year certifying its inspectors as smoke readers, and called the commercial camera software "poorly supported.".
Credit
ASTM International; EPA's Office of Air Quality Planning and Standards; Hill Air Force Base, which asked for the approval
2014
India puts stacks online
In 2014 India's Central Pollution Control Board (CPCB) directed 17 categories of heavily polluting industry to install continuous emission monitoring systems (CEMS), and in 2017 the Supreme Court ordered every state to show the readings on public web portals. A March 2024 study by the Council on Energy, Environment and Water (CEEW), a New Delhi research group, found that only 20 of the 32 state boards that must run monitoring web portals made them public and only 9 let users download past data. In six states, only 19% of the stacks it analyzed in 2019, and 23% in 2020, met CPCB's rule of 85% data availability; fewer than 7% of the gaps lasted more than 72 hours, but those made up more than 92% of the missing hours.
LessonCEEW notes that periodic manual tests remain the only legally approved way to measure industrial emissions in India, so the continuous readings still cannot by themselves prove a violation.
Credit
Central Pollution Control Board; study by Hemant Mallya, Sankalp Kumar and Sabarish Elango of CEEW
2017
Sniffing ships' exhaust
At the end of 2014 the limit on sulfur in ship fuel in northern Europe's Sulphur Emission Control Area fell from 1.00% to 0.10%. A Chalmers University of Technology team led by Johan Mellqvist measured ships' plumes with "sniffers" on the Öresund and Great Belt bridges and the approach to Gothenburg, and from an aircraft. In October 2017 it reported that 87% to 98% of ships complied: the fixed stations found 2% to 5% of passing ships burning non-compliant fuel, while the aircraft found 13% in violation in the western English Channel in September 2016. In 2019 and 2020 drones supplied by the European Maritime Safety Agency (EMSA) flew through ships' plumes north of the Great Belt and sent the readings straight to the Danish Environmental Protection Agency.
LessonMellqvist said ships expect to be measured near fixed stations, so checks from the air show "how much the vessels actually emit when they are out at sea.".
Credit
Johan Mellqvist and colleagues, Chalmers; the Danish Environmental Protection Agency; the Danish Maritime Authority; EMSA
2018
Satellites question stack data
China required coal power plants to meet tighter sulfur dioxide limits by July 2014. Comparing the plants' own CEMS readings with satellite data, Valerie Karplus, Shuang Zhang and Douglas Almond found that the stack data showed SO2 concentrations falling 13.9% after the deadline, while the satellite data showed on average about half as large a decline, and the two matched worst in the regions facing the toughest limits.
LessonThe authors called for "increased scrutiny of emissions data quality and monitoring practices.".
Credit
Valerie Karplus, Shuang Zhang and Douglas Almond
2026
Continuous dust monitors dropped
A 2024 update to EPA's Mercury and Air Toxics Standards required coal- and oil-fired power plants to prove they met the limit on filterable particles with particle CEMS only. A final rule published on 24 February 2026, in force from 27 April 2026, repealed that requirement and the tighter limits, and restored the 2012 choice of quarterly stack tests, monitors that track a related operating measure (parametric monitoring), or particle CEMS. EPA estimated the saving from dropping the CEMS requirement at $2.9 million a year, and said it no longer believed that the CEMS' benefits for pollution control and transparency outweighed their cost.
LessonWhether a plant's dust is measured all the time or four times a year can change with the rule in force.
Credit
US Environmental Protection Agency
Watch for
Software reads smoke opacity
Fay Huang, Hong-Yi Jang, Yuan-Jen Ou-Yang and Bao-Jen Pong describe a system that finds the smoke in an ordinary photo, sorts it as black or white smoke, and estimates its opacity with neural networks. On their own collected images, whose opacity had been measured by instrument, 90.6% of black-smoke estimates and 96.0% of white-smoke estimates fell within 10 percentage points of the measured value.
Achieved whenan outside party runs the system at a smoke-school certification session and it meets Method 9's limits, or a regulator approves it as an alternative to Method 9.
Claimed by
Fay Huang and colleagues, in Sensors and Materials, October 2026
Open
01
Pass smoke school by software
An open-source program that reads plume opacity from ordinary video, frame by frame, and sits the same certification test a human smoke reader sits: 25 black and 25 white plumes from a smoke generator, scored against the generator's built-in opacity meter.
The briefCredit · unclaimed02
A stack-data report card
A monthly public report for one Indian state that downloads every stack's readings from the state pollution control board's public portal and flags, plant by plant, missing data, readings stuck at one value for hours, and readings above the limit.
The briefCredit · unclaimed
Open 01 · the brief
Pass smoke school by software
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source program that reads plume opacity from ordinary video, frame by frame, and sits the same certification test a human smoke reader sits: 25 black and 25 white plumes from a smoke generator, scored against the generator's built-in opacity meter.
Why it is still open
EPA's approved camera method (ASTM D7520, ALT-082) reads still photographs with commercial software, and a 2010 Virginia review called that software "poorly supported." Huang and colleagues (Sensors and Materials, October 2026) estimated opacity from photos within 10 percentage points in most cases, but report no test run to the certification rules and publish no code. Vendors selling AI opacity cameras publish no smoke-generator results, and one says such systems are "a continuous supplement rather than a regulatory replacement" for certified observers. Hobby programs on GitHub, such as monitor-ringelmann (July 2026), label their output as not certified. No open model has published a pass of the Method 9 certification test.
Built with what exists
A phone or action camera; Python, OpenCV and PyTorch; a small vision model fine-tuned on labeled frames from smoke-generator runs; the Method 9 text and EPA's Visible Emissions Field Manual for the scoring rules; a smoke school that will let the camera film a certification session.
First step
Ask a smoke-school operator for permission to film one certification session from beside the class, together with the generator's meter log, and hand-label 200 frames.
Done looks like
The software passes three separate smoke-school sessions in different light, judged against the generator's meter with Method 9's limits (no reading more than 15% opacity off, average error no more than 7.5% for black and for white plumes), with the code, model and labeled video published.
Open 02 · the brief
A stack-data report card
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A monthly public report for one Indian state that downloads every stack's readings from the state pollution control board's public portal and flags, plant by plant, missing data, readings stuck at one value for hours, and readings above the limit.
Why it is still open
CEEW's March 2024 study analyzed 2019 and 2020 data from six state portals once. The Save Ennore Creek Campaign's report "Breathless in Ennore," published in July 2025, analyzed three Chennai power plants for one winter and found them either over the limit or without valid data about 49% of the time. Neither runs as a standing check, and CEEW found that only 9 of 19 public portals let users download past data.
Built with what exists
Python with requests or Playwright to download portal data; pandas; plain rules for gaps, stuck values (for example the same reading for six hours or more) and exceedances; CPCB's 85% monthly data-availability rule; a static website for the reports.
First step
Pick one of the six portals CEEW found offering downloadable yearly data (Assam, Chhattisgarh, Goa, Himachal Pradesh, Madhya Pradesh or Meghalaya), read its terms of use, and download one month of readings for one plant.
Done looks like
Twelve consecutive monthly reports covering every stack on one state's portal, each stack scored for data availability against the 85% rule, hours of stuck values and hours over the limit, with code public, and a hand check of 50 randomly chosen stack-days that matches the portal in at least 49.