How this study was done, and its limits

How we check facts

• Every number has a source. Each fact on this site is followed by a small number that links to its source. Our own results link to the open data we used.
• Primary sources first. We prefer official documents, peer-reviewed studies and the original data. Where these were not available online, we used news reports of them and say so.
• Checked against the source. Key figures were checked by opening the cited page or document. Scanned government PDFs were read page by page.
• Cautious when sources disagree. Where reports gave different numbers, we used the more cautious figure or gave the range, and named the method or period.
• Claims we could not confirm were removed. In the November 2026 review, several widely repeated shares with no traceable source were taken off the site.
• Open images only. Photos come from Wikimedia Commons under public-domain or Creative Commons licences, credited under each picture.

Known limits

• The model under-reads. Compared with 63 ground monitors, CAMS PM2.5 was about 1.68× lower (median), with a median hourly correlation of 0.22. We use CAMS only for patterns and seasons, never for exact levels.
• The model misses firecrackers and local dust. These causes rely on ground monitors.
• Satellites see once a day. Sentinel-5P passes at about 1:30 pm, when traffic is lighter and the air is well mixed.
• Trajectories are simplified. They use resolved winds without turbulence. Confirm with NOAA HYSPLIT before using them in research.
• Source contributions. The Atlas maps show where roads and kilns are located and where satellite NO₂ columns are elevated. Estimating how much each sector contributed to ground-level PM₂.₅ during 1–10 November 2026 requires a separate source-apportionment analysis. The UrbanEmissions review compares earlier studies and provides historical context for interpreting their estimates.³⁹
• Correlation is not cause. A link between two maps does not prove one causes the other. Regression results use autocorrelated hourly data and are indicative only.
• Licences. SentinelKilnDB is CC BY-NC 4.0 (non-commercial). Check this before using the kiln layer commercially.

Glossary

AQI

Air Quality Index. India's scale runs from 0 (Good) to 500 (Severe).²⁴

PM2.5 / PM10

Particles smaller than 2.5 or 10 micrometres. PM2.5 is the more harmful one.

Secondary particles

Particles that form in the air from gases such as SO₂, NOx and ammonia.

Mixing height

The depth of air near the ground in which pollution can spread.

Ventilation index

Daily maximum mixing height × wind speed in that layer. Below about 6,000 m²/s, pollution spreads poorly.³⁸

Back-trajectory

The path that air took to reach a place, traced backwards in time.

Sentinel-5P / TROPOMI

A European satellite and sensor that measures gases such as NO₂ and SO₂ across the whole Earth every day.

VIIRS

A sensor on NASA/NOAA satellites that detects fires at 375 m resolution.

FCBK / Zigzag kiln

Older fixed-chimney brick kilns, and the cleaner zigzag design that burns fuel more completely.

GRAP

Graded Response Action Plan, Delhi-NCR's step-by-step emergency plan for bad air.

FGD

Flue-gas desulphurisation. Equipment that removes SO₂ from power plant smoke.