Delhi’s air
Every reading against its published limit. Every gap named.
Year round
Live — the source publishes hourly; the observation time is printed below
309
AQI · 24-hour · worst of eight
Very Poor
CPCB safe limit 100. Limit broken.
Anand Vihar, Delhi - DPCC. Observed 17:00 IST, 23 August 2026. How this number is made.
India, right now
261 cities report to CPCB. Delhi is ranked 4 of them, and that changes between hours.
74 of 261 above the limit India set for itself. 58 read “Good”. Manesar reports from one station and Delhi from 45 — a city with one monitor is measured less, not better.
All nine figures were read together, and none of them moves while you are here. The eight cities above and Delhi’s row beside them come from one snapshot, taken at 23-08-2026 17:00:00 — the same hour as the reading at the top of this page. That is what makes them comparable, and it is why the order is printed as a reading of one hour rather than as a standing claim: by the time you read it, CPCB has published another.
Who is in it?
The consequence comes before the measurement, because the measurement is not the point. Every figure here is measured against a limit India has not adopted.
Counted or measuredModelled
1.5
million deaths a year
about 1.5 million, against the WHO guideline — 5 µg/m³ annual. Lancet Planetary Health, December 2024.
5.0%
of all mortality
against India’s own standard — 40 µg/m³ annual. Same study, same deaths.
India’s standard is eight times the WHO guideline, so the same harm counts twice over.
29.4%
of Delhi adolescents aged 13–17 showed spirometry-defined asthma or airflow obstruction. Counted
The study’s own strongest association was obesity, not air pollution — 39.8% overweight against 16.4%. Lung Care Foundation with Pulmocare Research, Lung India, September 2021.
8.2 years
of life expectancy lost in Delhi-NCR. 3.5 years across India. Modelled
Nearly twice the toll of childhood and maternal malnutrition. Air Quality Life Index, EPIC, University of Chicago, 2025.
Three of those four figures are models. The one that was counted is the one about children’s lungs, because somebody put real adolescents in front of a spirometer. A model is not a lesser thing than a count, but it is a different thing, and a dashboard that sets them at the same weight is telling you they are the same.
How the number is made
One number stands in for eight, and it is not their average — it is the worst of them. Measured and modelled are set differently, on purpose.
AQI
One number for eight poisons
Eight pollutants folded into one 0–500 figure that reports whichever is worst. It is not a concentration. 309 today; the limit is 100.
PM2.5
Small enough to enter blood
Particles under two and a half microns pass the lung wall into the bloodstream. 131 µg/m³ today. The Indian daily standard is 60.
PM10
Dust you can feel
Coarser particles from roads, construction and soil, stopping in the upper airway. 213 µg/m³ today. The standard is 100.
“The worst sub-index determines the overall AQI.” CPCB, About National Air Quality Index.
PM2.5
309
131 µg/m³
governing
PM10
175
213 µg/m³
NO₂
46
37 µg/m³
SO₂
38
30 µg/m³
O₃
4
4 µg/m³
NH₃
3
10 µg/m³
Eight pollutants, 6 in this number. Pb is not reported at this station. CO is reported — 113 — but left out: the feed states no unit for it, CPCB’s CO breakpoints are in mg/m³ where everything else here is µg/m³, and on either reading the value is not credible. Read as mg/m³ it alone would put almost every station in the top band.
AQI 100 is not a rule of thumb. The boundary sits at PM2.5 60 µg/m³ and PM10 100 µg/m³ — exactly the 24-hour standards India set for itself. Above 100 is above the law.
| Figure | Kind | Source | Cadence |
|---|---|---|---|
| AQI, 309 | Derived | computed from CPCB concentrations | Hourly |
| Station concentrations | Measured | CPCB, 45 Delhi stations | Hourly |
| Published limit, 100 | Standard | CPCB, NAAQS 2009 | Fixed |
| Source split | Modelled | published apportionment study | Per study |
| Farm-fire counts | Measured | NASA FIRMS, per sensor | Daily |
| Attention | Measured | Wikipedia pageviews | Daily |
| Forecast | Modelled | WAQI’s model, not CPCB’s | Daily |
Not CPCB’s published AQI. The feed returns concentrations and no index, so the number at the top of this page is computed here using CPCB’s own breakpoint table.
Where does it come from?
A source apportionment is a model: it takes measured concentrations, wind and an emissions inventory and works backwards to a plausible split. Biomass burning is one segment, and it has a season.
Modelled, not measured
This is the only complete split anyone has published for Delhi, and its newest measurement is from February 2017. Six sectors, summing to a hundred, from the study the Government of India commissioned to settle the question.
Dust more than doubles in summer and industry stays put. The season does not just change how much there is — it changes what it is. A control measure aimed at the winter split is aimed at a different problem in June.
TERI-ARAI, August 2018 — PM2.5, Delhi, dispersion model (WRF–CMAQ), monitoring April 2016 to February 2017. Commissioned by the Department of Heavy Industry, Ministry of Heavy Industries and Public Enterprises. The report, Table E.2, page E13, executive summary.
Four things the study says about itself. Agricultural burning at 4% is a floor, not an estimate — the monitoring never covered October, when burning peaks, and the report says so in as many words. The shares are averaged across the whole period, so they cannot describe a bad day. The model reproduces only 82–87% of the mass actually measured. And “industry” includes biomass burned as industrial fuel, which the report itself calls an overestimate.
Modelled, not measured
The other government study will not give you a single number. IIT Kanpur sampled six sites across Delhi and reported every source as a range, because the answer was different under every monitor. The width of each line below is how much the two studies, and the six sites, fail to agree.
Vehicles are somewhere between 6% and 29%. That is the range in which every argument about Delhi’s cars, buses and trucks is actually being conducted. TERI-ARAI puts transport at 17–28% and says plainly that its figure is higher than IIT Kanpur’s because it counted secondary particles alongside primary ones — the two studies are not quite measuring the same thing.
IIT Kanpur, January 2016 — PM2.5, receptor modelling (chemical mass balance) at six Delhi sites, monitoring Winter 2013-14 and summer 2014. Commissioned by the Government of NCT of Delhi, and the Delhi Pollution Control Committee. The report, Section 4.6, 'Overall Summary and Source Apportionment at a Glance', page 241.
And the categories do not line up. IIT Kanpur’s “coal and flyash” is a fuel signature; TERI-ARAI’s “industry” is a sector that bundles power plants, brick kilns and stone crushers. They are not two estimates of one quantity. They do not average, and neither is a measurement of today’s reading.
Modelled, not measured
Inside the 28% that transport contributes, cars are 3.4. Trucks are more than twice that and two-wheelers twice that. Winter PM2.5, Delhi.
Of the car share, diesel is 67–74% and the oldest cars dominate — BS-II and earlier account for 31–50% of it. Run the study’s own arithmetic and every BS-IV diesel car in Delhi comes to about 0.5–0.9% of PM2.5. That is not an argument for more cars. It is an argument about where a policy rupee buys the most air, and the answer in this table is trucks, brick kilns and generators, not hatchbacks.
Also inside the totals: within industry’s 30%, brick kilns 8%, power stations 6%, stone crushers 2%, and 14% from other industries burning coal, biomass, pet-coke and furnace oil. Within “others”, diesel generators 5% and refuse burning 3%. And within dust’s 17%, road dust is 4 and construction 1 — the report attributes much of the remainder to dust that did not originate in Delhi at all. TERI-ARAI, section E7.3.
The question neither study answers. What is in the air today. Both are seasonal averages whose last measurement is eight years old. There is a system that does answer it — Decision Support System (DSS), run by the Indian Institute of Tropical Meteorology, Pune, which publishes a daily split across 29 sectors with the stubble share taken from the previous evening’s satellite fire counts. It has no public API and its host was unreachable from the machine that built this page, so it is named and linked and never restated.
0
MODIS · 1 km
last 5 days
24
VIIRS · 375 m
MODIS saw none, same fires
The disagreement is itself a measurement. Out of season MODIS resolves none of them at all. At peak — 2025-11-05 +5d — 1,170 against 2,222, only 1.9:1. MODIS misses almost everything when fires are small and catches most of it when they are large, so the gap measures fire size, not instrument error.
A detection is a thermal anomaly, not a confirmed crop fire. The two VIIRS satellites see the same fires at different overpass times and must not be added.
MODIS, five days from 5 November each year, same region. A fixed five-day window is a sample, not a season total — burning dates shift with monsoon withdrawal and harvest timing, so a fixed window can miss a peak. 2024 and 2025 run at roughly a sixth of 2021.
Where it has been, and where it is going
The record starts today; the forecast reaches seven days ahead. For now this page sees further forward than back — that inverts in a week.
One square. The record begins today and fills as the job runs. It draws no square it does not have — an empty cell is absence, not zero. There is no retrospective series: the CPCB feed publishes the latest hour only.
Attention is not air. Searches peak every November at 39,084 and fall to 3,633 in summer — a 10.8× swing — while the readings stay above the limit all twelve months. And the swing is widening: 2022 4.7×, 2023 5.6×, 2024 7.2×, 2025 8×.
Wikipedia pageviews, 55 complete months. It measures what people seek, not what outlets publish. A quiet month is not a clean month. The current month is incomplete and excluded, never plotted.
PM2.5 daily mean, 6 days.
Modelled, not measured
Somebody is forecasting this. Not us, and not the government. The curve is WAQI’s own model. India’s official forecaster is SAFAR, which publishes a 72-hour Delhi forecast with no public API — so it is named and linked, never restated.
Which part of the city, and where the city sits
There are 45 monitors reporting for Delhi and they do not agree with each other. Every figure in this band was read at 17:00 IST, 23 August 2026 — one hour, all 45 stations together, which is what makes them comparable — and the same hour as the reading at the top of the page, which is read from this band's worst station.
Above the limitWithin it
309 and 84, 3.9 km apart. Anand Vihar and Patparganj are the same hour of the same city and they differ by 225 points — Patparganj sits inside the limit while Anand Vihar is very poor. Whichever one is nearest to you is the only one describing your air.
45 monitors, true positions, equal scale on both axes; the 40 × 39 km frame is the box the monitors describe, not Delhi’s boundary. Median spacing between neighbouring monitors 2.3 km — but inside that same box you can stand 14.8 km from the nearest one. Red is above the limit.
Two layers are missing on purpose. There is no ward layer — the boundary file is not published in a usable form, and a hand-drawn one would be a claim about where you live. There is no plume — a plume is a model, and this page does not draw a model on top of measurements without saying so.
… 37 more …
45 stations, 24 above the limit. The quietest reads 16, the loudest 309 — one city, a fivefold spread. A city average hides it. Red is above the limit.
Delhi is 4th of 261 cities, and 7 of the next 12 are its neighbours — Gurugram, Panipat, Noida, Pali, Bhiwadi, Jind, Charkhi Dadri. This is not a city problem. It is an airshed.
74 of 261 cities are above the limit India set for itself, and 58 are “Good” — the country is not uniformly polluted, which is what makes the cluster around Delhi legible. Computed from 499 stations on CPCB’s scale, CO excluded.
The cost of inaction is more than the action
Allocated is the claim. Utilised is the record. Both are dwarfed by the damage — and all three numbers are the government’s own.
The damage, each year
₹7 lakh crore
about $95 billion, roughly 3% of GDP — the report’s own comparisons are 50% of all tax collected annually, and 150% of India’s health budget. Dalberg with Clean Air Fund and CII, 2021.
Released for it since 2019
₹13,415 crore
NCAP and the 15th Finance Commission air-quality grants, cumulative.
Actually spent
₹9,929 crore
74% of what was released. The 82 NCAP cities used ₹831 crore of ₹1,615 crore — 51%.
One year of damage costs about fifty times everything released for it since 2019, and about seventy times what has been spent. Both periods are stated because the comparison only holds if they are: the damage figure is annual, the spending figures are cumulative.
What this is not. It is not a cost-benefit study — nobody has published a costed abatement plan for Delhi-NCR, so this page cannot tell you what fixing the air would cost. It is the narrower claim, and the only one the figures support: what the damage costs each year, beside what has actually been released and spent against it. Sources: PIB releases, CPCB’s PRANA funding guidelines, CREA’s Tracing the Hazy Air, Dalberg with Clean Air Fund and CII. No inference is drawn beyond the arithmetic.
What you can do
Nobody visits a record every morning. This is the part that asks something of you.
Watch your monitor
One message when your monitor’s band changes for the worse. Not when it is merely over the limit — it is over the limit most of the year, and an alert every hour is not an alert. Nothing else, ever, and no address is shared with anybody.
Find the monitor nearest you
Loading the 45 monitors…
Your monitor
Where to write
Why it asks for a monitor and not a pin code. India Post’s own All India Pincode Directory publishes 562 post offices for Delhi and no latitude or longitude column at all, so there is no official way to turn a Delhi pin code into a point on the ground. The alternative was a third-party centroid file of unknown provenance — on this page, of all pages. A monitor is the better question anyway: two of them 3.9 km apart read 392 and 110.
The campaign
Delhi I Can’t See You is Swechha’s campaign on this. This page is the why; the campaign is the what.
Five more situations
Yamuna · Heatwave · Forest fires · Forest loss · Climate event
Named, not linked, until their pages exist.
Build a monitor, and then distrust it
A classroom-grade PM sensor costs a few thousand rupees — an optical particle counter, a microcontroller, a power supply. It will give you a PM2.5 number within minutes of switching on. That number means nothing yet.
Low-cost optical sensors read high in humidity and drift as their chamber fouls. To turn a reading into a measurement you co-locate it beside a reference station for a fortnight and fit your device against theirs. That is the whole difference between the 45 instruments on this page and a gadget on a windowsill, and it is why this page will not accept a crowd-sourced reading as equivalent.
File a complaint that lands somewhere
Open burning, a construction site without sheeting, a generator running in a residential block, a road left unswept — each is already prohibited under rules somebody is meant to enforce.
Sameer, CPCB’s own app, publishes the same station data this page reads and takes a geotagged photograph as a complaint. Green Delhi, the Delhi government’s app, routes the same thing to the responsible department with a trackable number. Neither is Swechha’s and neither is endorsed — they are named because a complaint with a reference number is harder to lose than a complaint without one.
Read the objection window
Environmental clearances carry a public-consultation stage with a fixed window for written objections. Most close unopposed because nobody was watching.
This page does not yet track those windows for Delhi-NCR. It is named here because it is the highest-leverage thing on the list, not because it is built.
Most recent order, as reported
Delhi-NCR Pollution: CAQM Announces Phased Ban On New Petrol, Diesel, CNG Light Goods Vehicles
ETV Bharat · Thu, 20 Aug 2026. Reported, not filed. This page keeps no docket and attaches no judgement — it records that a court was reported to have said something, and links whoever reported it.
100 items from 45 publishers, via Google News. Reporting is tagged as reporting. Never presented as Swechha’s finding. A headline is evidence that something was said, not that it is true.
Every reading, kept
Each day’s reading keeps its own address, with the station that produced it, the hour it was observed, and the limit it was judged against. Nothing is overwritten when it improves and nothing is quietly restated when it gets worse. An empty day stays empty — a gap in the record is a gap in the record, never a zero.
Cite this page
Reuse freely — CC BY 4.0. Every figure carries its source and its cadence in how the number is made, and every figure carries whether it was counted or modelled on the rule beneath it. If you quote a number from here, quote the kind with it.
This is what we are doing.
It is not enough.
The work against this reading
One campaign with a name on it, and a slower thing underneath. Delhi I Can’t See You is the public push; air is also a module in the schools curriculum, and the CityScapes walks take people to the places these readings are taken.
And why it is not a reply
None of it moves an AQI of 307. A campaign changes what a city is willing to say about its air; it does not change what the city burns. The reading at the top of this page is the honest measure of that distance, which is why it is printed before any of this.