Delhi's air over a decade
Daily PM2.5 at the US Embassy in New Delhi, every day from late 2016 through early 2026. Each row is a year. Each column is a day. Darker red means worse air. Small chevrons below each year mark Diwali night (the Hindu calendar shifts the date by about ten days every year).
The monitor goes blind on Diwali. Six of the ten Diwalis on record are followed by a missing day in the data — on the night of the festival or the morning after. Fireworks smoke is so thick that the embassy's PM2.5 sensor drops out. Look for the small black ✕ marks on Diwali-week rows.
Every winter, the air gets 5 to 10 times worse than it does in summer. Cool air traps smoke near the ground, stubble fires across Punjab and Haryana send plumes south, and Diwali fireworks land on top. The one clean break in the pattern is spring 2020, when India's COVID lockdown pushed daily readings down to monsoon-summer levels for the first time on record. By winter 2021 the smog was back.
Hover any day for the exact reading. Hatched bands mark stretches where the sensor was offline — including the four-month gap visible across mid-2022.
Why winter?
Delhi's air is bad year-round, but November through January is on a different scale. Three things stack in the cold months. Rice stubble burning in Punjab and Haryana, from mid-October through November, sends smoke plumes south on the prevailing winds. Cold air sits below warmer air aloft, so the smoke and Delhi's own emissions get trapped in a shallow layer near the ground. Diwali fireworks and biomass cooking and heating layer on top. By March the trap breaks and the westerlies (with the occasional dust storm) push the haze out — until the cycle restarts in October.
The cycle is so consistent that month-by-month averages line up neatly across every year on record. November and January both average over 190 µg/m³ — already deep into India's "very poor" band — while July and August stay under 40, which is "satisfactory." Same shape, every year.
Average PM2.5 by month, 2016–2026
Each dot is one year's monthly average. The thick line is the ten-year average. Notice how the spread in November and January is huge — a bad winter is twice as bad as a clean one — while summer months cluster tightly. The summer floor is whatever is always there; the winter peak is what year-to-year choices and weather add on top.
Annual average and number of severe days
Bars: number of days each year that crossed India's "severe" threshold (PM2.5 above 250 µg/m³). Line: the year's average. The 2020 lockdown clearly pulls the line down. The decade as a whole is roughly flat — not getting noticeably better, not getting noticeably worse. Partial years (2016, 2022, 2026) are dimmer because the sensor was offline for part of those years.
The Diwali sensor outage
Six of the ten Diwalis on record — 2017, 2019, 2021, 2023, 2024, 2025 — show a missing reading on Diwali day or the day after. The surrounding days are fine. The most plausible explanation is that the fireworks plume saturates the embassy's PM2.5 instrument: hours of dense smoke at ground level overload the optical sensor, and the daily aggregate drops out. Diwali air doesn't just get bad — it gets bad enough that the instrument refuses to commit to a number.
On the years where the sensor does report a value, Diwali itself usually sits in the top 50 to 100 worst days of its year — bad, but not record-bad. The year's single worst day is almost always a deep-January cold-snap day or a stagnant late-November smog episode that builds after Diwali, not on it. Fireworks add a sharp local bump; stubble-fire timing and the depth of the inversion drive the multi-day peaks.
| Year | Diwali | PM2.5 that day | Sensor on day +1? | Year's worst day |
|---|