Where does Delhi's pollution come from?

The two big studies of where Delhi's particles come from agree on one thing and disagree on almost everything else: no single source is “the” cause, and the answer changes with the season, the method and the edge of the map you draw.

How do scientists work out where pollution comes from?

The work is called source apportionment, and there are two broad methods.

Receptor modelling starts from the air. Researchers collect particles on filters, measure their chemistry (carbon, metals, ions), and match that fingerprint against the known fingerprints of sources such as road dust, diesel exhaust or burning crop residue. Dispersion modelling starts from the sources: build an inventory of what is emitted where, then simulate the weather to see how much of it reaches each point. Delhi's two landmark studies used both, in different proportions, and that alone explains part of why their numbers differ.¹²

What did the IIT Kanpur and TERI–ARAI studies find?

IIT Kanpur (2016). Mukesh Sharma and Onkar Dikshit's study was submitted in January 2016 to the Delhi government's environment department and the Delhi Pollution Control Committee.¹ Its receptor modelling (chemical mass balance) gives PM2.5 shares by source for winter and summer.¹

TERI–ARAI (2018). The Automotive Research Association of India and The Energy and Resources Institute prepared their study for the Department of Heavy Industry and released it in August 2018.²³ Its headline sector shares for PM2.5 come from dispersion modelling over a 2016 emissions inventory.² Side by side, for Delhi's PM2.5:

The categories do not line up one-to-one. IIT Kanpur reports “secondary particles” as their own category, and these form in the air from gases emitted by vehicles, power plants, industry and farms. TERI–ARAI's dispersion model assigns the gases back to the sectors that emitted them, so its “industry” and “transport” shares already include the secondary particles they cause. Shares are rounded and may not sum to 100.

Why do the two studies disagree?

They answer different questions. A receptor model says what the particles in Delhi's air are. A dispersion model says which sector's emissions put them there. The IIT Kanpur winter figure of 30% for secondary particles is not a source at all in TERI–ARAI's framing; it is redistributed across sectors.¹²

They cover different places. TERI–ARAI modelled Delhi inside the whole NCR and beyond, and found that Delhi's own emissions accounted for about 36% of its PM2.5 in winter and 26% in summer, which leaves roughly two-thirds or more arriving from outside the city's boundary.² IIT Kanpur estimated, more loosely, that about a third of Delhi's particle levels came from outside Delhi.¹ The difference is partly what counts as “outside”: a factory in Gurugram is outside Delhi but inside the NCR.

They cover different months. TERI–ARAI's monitoring ran from April 2016 to February 2017, and its modelling did not include October, when crop burning in Punjab and Haryana usually peaks.² Its own press release notes that averaged sector shares cannot capture short-term episodes of agricultural burning.³

Why is “X% of Delhi's pollution is stubble” a misleading sentence?

Because a percentage needs a time window, and stubble burning is the most time-bound source Delhi has. Averaged over TERI–ARAI's winter, agricultural burning comes out at 4% of PM2.5.² IIT Kanpur, looking at the November episode, found biomass-burning particles in the air at about 120 µg/m³ of PM2.5, against a winter mean of 86 µg/m³, and attributed the excess to crop-residue burning in Punjab and Haryana carried in from the north-west.¹ On individual days the share can be far higher. A decision-support system run for Delhi estimated stubble's share of the city's PM2.5 in November 2024 at 20% on average and 38% at its peak; in November 2025 the figures were 7% and 22%.⁴

All of these can be true at once. Stubble is a large part of the problem on some weeks of the year and a small part across the year. A sentence that gives one number without its window is not wrong so much as unanswerable. The useful questions are “on which days?” and “compared with what else?” The chapter Why is Delhi's air worse in winter? takes those up.

What is still unknown?

Both landmark studies are now years old: IIT Kanpur reported in 2016, and TERI–ARAI's monitoring ended in February 2017.¹² Delhi has changed since. The day-by-day source shares quoted in the news, like the stubble figures above, come from a decision-support model rather than from new chemical sampling.⁴ Treat any single percentage, including the ones on this page, as a snapshot from one method in one period.