How has air quality varied across India’s cities and towns over the last decade?

air quality
PM10
PM2.5
CPCB
India
An exploratory look at CPCB air quality monitoring data for India’s cities and towns, 2015 to 2024, plus winter 2025–26 week by week.
Author

Srijith Balan

Published

November 12, 2026

Reading this carefully

Shows
Most of India’s monitored locations have had annual PM10 above the national standard throughout 2015–2024: 323 of 433 mapped locations in 2024, and 130 of the 247 locations with at least 8 years of data in every single year. The problem is not confined to the northern plains; West Bengal, Maharashtra and Odisha also have many always-over locations. Last winter, the worst weeks were concentrated in the Delhi region from November into January.
Doesn’t show
Whether air quality has improved. Median PM10 across the 206 long-monitored locations fell about 13% from 2015 to 2024, but CPCB changed how location values are calculated in 2021 and the network grew sharply, so the two can’t be separated. It also doesn’t describe the air between monitored locations, or whole states, and the six weeks of this winter are too short to call a change.
Connected to
Weather, especially the winter inversion over the northern plains; crop-residue burning; industry and mining in Jharkhand and Odisha; road and construction dust; where stations are sited (kerbside, industrial or background); and the growth of the monitoring network itself. None of these were tested in this work.
Annual PM10 across monitored locations in India, 2015 to 2024. Spike height is the annual average; colour is the multiple of the national standard.

Most of India’s monitored locations have recorded annual PM10 above the national standard throughout the last decade, and the median PM10 across them (the middle value: half the locations above it, half below) is only modestly lower, by an amount that can’t be separated from a 2021 change in method. In 2024, 323 of the 433 mapped locations had annual PM10 above 60 µg/m³, and of the 247 locations with data in at least 8 of the 10 years, 130 exceeded the standard in every single year.

The work began as an attempt at a simple state-wise air quality map and turned into something a bit different. This piece covers why the analysis maps monitored locations rather than states, how CPCB’s annual tables were brought into a usable shape, what the decade looks like, how states can be compared without overstating what the data says, and what last winter looked like week by week. It is an exploratory look, not research, and the text tries to say clearly where the data stops supporting a conclusion.

Why monitored locations and not states

The first plan was a state-wise map of India with a weekly AQI value for each state, animated through a year. A look at the locations of the monitoring stations showed why that wouldn’t hold up. Nearly all ambient air quality monitoring stations are located in urban areas, mostly in cities and larger towns, so a state value would really be the value of a handful of monitored cities and towns, extended over large areas of farmland, forest and smaller towns where no measurements are taken. A state like Uttar Pradesh would take the colour of Ghaziabad, Noida and Lucknow, even though the western and eastern ends of the state experience quite different conditions.

The maps therefore show the monitored locations instead. Each point is a place where measurements were actually taken, and the space between them is left empty, because that is what the data can support. The cost is that the regional picture, for example the northern plains behaving as one airshed in winter, has to be read from the pattern of points rather than from a filled map. States return later in the piece, but only in a form that rates the monitored locations of a state, and only where those locations cover enough of the state to say something.

The data

The decade view uses the Central Pollution Control Board’s annual tables from the National Air Quality Monitoring Programme, one PDF per year from 2015 to 2024. They give annual averages of SO₂, NO₂, PM10 and PM2.5 (and from 2024 also NH₃ and benzene), but they are not consistent across the decade. From 2015 to 2020 the tables list individual manual monitoring stations, with the number of monitoring days and the 24-hour minimum and maximum for each. From 2021 onwards they list one integrated value per location, combining manual and continuous stations, with no station detail. PM2.5 is also thin in the early years: in 2015 only a handful of states reported it at all.

These are annual averages, so they can’t give weekly or seasonal patterns, and they can’t be turned into an AQI either, because India’s AQI is defined on 24-hour and 8-hour concentrations. For that reason the decade maps show each location’s annual average as a multiple of the national annual standard (60 µg/m³ for PM10 and 40 µg/m³ for PM2.5) rather than as an AQI.

For the recent winters, daily PM2.5 was taken from CPCB’s Continuous Ambient Air Quality Monitoring Stations (CAAQMS), downloaded through the OpenAQ API because CPCB’s own dashboard couldn’t be accessed for bulk download at the time. That covers 1 October 2025 to 28 February 2026 and 1 October to 11 November 2026. No 2025 annual table from CPCB could be found at the time of writing, so 2025 appears only through these winter weeks.

Getting the tables into usable shape

Most of the effort went into this step. Copying text out of the PDFs ran numbers together (Vapi’s 2022 row came out as “379 119 69” when the cells were 37, 9, 119 and 69), and page numbers attached themselves to the last value on a page. The tables were therefore extracted cell by cell rather than as text, which resolved every fused number found. Seven 2022 rows split across page breaks were matched back to their locations manually and checked against the text.

The 2015 to 2020 station tables had a different problem. State and place names occupy merged cells, vertically centred across several station rows, so simply copying a label down to the rows below placed stations in the wrong location (Ankleshwar’s stations would have ended up under Ahmedabad). Each station was instead assigned to its location and state using the geometry of the table cells, giving 12,903 station-pollutant rows. Every row was checked that its annual mean lies between its own minimum and maximum. Nine rows failed, mostly 2019 rows for Ghaziabad, Hapur and Noida where several stations share one cell, and these were excluded rather than corrected by guesswork. Another 42 rows report monitoring days, minimum and maximum but no annual mean, and those were left blank.

Place names were the next issue. The same place is spelled differently from year to year (Gurgaon and Gurugram, Bareily and Bareilly, three spellings of Byrnihat), and some names were split across lines inside a cell. After harmonisation there are 545 distinct monitored locations. A location here is whatever CPCB lists as a separate place in its tables, which it labels as city, town or village; most are cities and towns, but some are parts of larger urban areas (Jharia, Sindri and Jorapokhar, for example, all lie within Dhanbad Municipal Corporation), some are industrial areas such as Byrnihat, and a few in Punjab are village monitoring sites. The station tables also had 579 rows assigned to the wrong state, for example West Bengal towns tagged as Uttarakhand, and these were corrected using each location’s state in the 2021 to 2024 tables. For 2015 to 2020 a location’s value is the mean of its stations; from 2021 it is CPCB’s own integrated value for that location.

Finally, the locations needed coordinates. These were taken, in order of preference, from the median position of a location’s CPCB stations on OpenAQ, a GeoNames-derived place list, Natural Earth places, and as a last resort a district centroid, always matching within the same state. Fuzzy name matches were checked manually and the wrong ones discarded (Amlai had matched Amla, Anpara had matched Nanpara). That placed 476 of the 545 locations. The 69 that couldn’t be placed are mostly village monitoring sites in Punjab and small towns in Goa, Sikkim and Mizoram, and they are left off the maps rather than located by guesswork. A final check found 27 location-years counted twice because two place names matched the same location; these were removed before any of the calculations below.

What the decade looks like

In 2024, 323 of the 433 mapped locations with PM10 data were above the annual standard of 60 µg/m³. PM2.5 is somewhat better but still poor: 171 of the 365 mapped locations with PM2.5 data exceeded its annual standard of 40 µg/m³. On the map the locations are drawn as spikes, with height showing the annual average and colour showing the multiple of the standard, and the tallest and darkest spikes run along the northern plains and around the mining and industrial towns of Jharkhand and Odisha.

Spike map of annual PM10 in 2024 for 433 monitored locations in India

Annual PM10 in 2024. Spike height is the annual average; colour is the multiple of the 60 µg/m³ standard.

Checking whether things changed over the decade needed some care, because the monitoring network grew a lot. The number of locations with PM2.5 data jumped from 113 in 2020 to 263 in 2021 and reached 378 in 2024, which reflects new monitoring stations rather than new pollution. The comparison therefore uses only the 206 locations with PM10 data in at least 9 of the 10 years. Their median PM10 was 85 to 90 µg/m³ from 2015 to 2019, fell to 67 in 2020, and has been 78 to 88 since. That puts 2024 about 13% below 2015. The method also changed in 2021, from manual stations to CPCB’s integrated values, so a real improvement can’t be separated from the change in how location values are calculated. The 2020 dip coincides with the COVID-19 lockdown, which is the obvious explanation, but this data alone can’t show it was the cause.

Line chart of median PM10, 2015 to 2024, for 206 monitored locations

Median PM10 of the 206 locations monitored in at least 9 of the 10 years. The dashed vertical line marks the 2020 to 2021 change of method.

The persistence of the problem is clearer than the trend. Of the 247 locations with PM10 data in at least 8 of the 10 years, 130 were above the standard in every single year, and only 30 never were. Those 30 are hill towns and cities in Kerala, the Northeast, Sikkim and parts of Karnataka and Tamil Nadu. The highest ten-year averages were Jharia, a coal-mining area within Dhanbad (4.7 times the standard), Ghaziabad (3.7), Delhi (3.7), Bhiwadi (3.6) and Dhanbad (3.4). One might expect the northern plains to dominate the always-over list. They don’t: West Bengal (21 locations) and Uttar Pradesh (17) lead it, followed by Maharashtra (13), Odisha (11) and Madhya Pradesh (11).

Map of monitored locations by share of years their annual PM10 exceeded the standard

Share of each location’s monitored years above the PM10 standard; point size is its average multiple of the standard. Locations with fewer than 5 years of data are left out.

Major cities

The national figures can hide differences between the largest cities, so the two figures below show 16 major cities on their own: the largest metropolitan areas and several state capitals with long monitoring records. In 2024 every one of them was at or above the annual PM10 standard, from Chennai at the standard to Delhi at 3.7 times it. For PM2.5 they divide more clearly: Hyderabad, Bengaluru, Chennai, Mumbai, Surat and Pune were below the annual standard of 40 µg/m³, while Delhi was close to three times it, and a similar division appears in winter 2025–26, when only Hyderabad, Bengaluru, Chennai and Guwahati stayed near or below 50 µg/m³. Values up to 2020 are means of each city’s manual monitoring stations and values from 2021 are CPCB’s integrated values, and several cities had only two or three stations in the earlier years, so the lines show each city’s record rather than a measure of improvement or deterioration.

Dot plot of 2024 PM10 and PM2.5 as multiples of the annual standard, and bar chart of winter 2025-26 mean PM2.5, for 16 major Indian cities

2024 annual PM10 and PM2.5 as multiples of their annual standards (left), and mean PM2.5 over winter 2025–26 coloured by AQI band (right), for 16 major cities.

Small-multiple line charts of annual PM10 and PM2.5 for 16 major Indian cities, 2015 to 2024

Annual PM10 and PM2.5 for the same 16 cities, 2015 to 2024. The dashed line marks the 2020 to 2021 change of method; gaps are years without data.

States, only where coverage allows

A state comparison is still useful, so the analysis rates a state by the share of its monitored locations above the standard, and only where those locations are spread widely enough to represent the state. Coverage is measured as the share of a state’s districts (Census 2011 boundaries) containing at least one monitored location, because a measure such as locations per square kilometre penalises large states and favours small ones. A state is rated if it has at least 3 monitored locations in at least 30% of its districts. Delhi and Chandigarh are single-city territories, so they are treated as fully covered, which also means their share can only be 0% or 100%.

In 2024, 24 states and territories qualified, and 9 of them had every monitored location above the PM10 standard. Uttar Pradesh only just qualifies at 31% coverage even with 26 monitored locations, because its monitoring stations are concentrated in the western part of the state, while Chhattisgarh (28%) and Jharkhand (8%) don’t qualify at all. The table below gives the full picture for 2024 PM10, with the number of continuous monitoring stations in each state in winter 2025–26 for comparison. Station counts for 2021 to 2024 aren’t available, because CPCB’s tables for those years give one value per location without the number of stations behind it. This rates the monitored locations of a state, not the whole state.

Map of states coloured by share of monitored locations above the PM10 standard in 2024

Share of each rated state’s monitored locations above the PM10 standard in 2024. Hatched states have too little coverage to rate.
State Monitored locations, 2024 Above PM10 standard Share above District coverage Rated Continuous stations, winter 2025–26
Rajasthan 34 34 100% 100% Yes 46
Maharashtra 32 32 100% 57% Yes 78
Odisha 25 25 100% 53% Yes 17
Bihar 24 24 100% 58% Yes 34
Haryana 22 22 100% 90% Yes 27
Gujarat 20 20 100% 58% Yes 15
Telangana 9 9 100% 80% Yes 13
Chandigarh 1 1 100% 100% Yes (city territory) 3
Delhi 1 1 100% 100% Yes (city territory) 39
West Bengal 38 37 97% 95% Yes 22
Uttar Pradesh 26 25 96% 31% Yes 53
Punjab 20 18 90% 75% Yes 8
Madhya Pradesh 18 15 83% 32% Yes 24
Jammu & Kashmir 6 5 83% 30% Yes 1
Andhra Pradesh 16 10 62% 100% Yes 9
Assam 14 8 57% 44% Yes 8
Goa 8 4 50% 100% Yes 0
Himachal Pradesh 9 4 44% 58% Yes 1
Karnataka 26 5 19% 87% Yes 20
Kerala 12 2 17% 79% Yes 5
Tamil Nadu 28 3 11% 78% Yes 14
Mizoram 10 0 0% 100% Yes 0
Sikkim 5 0 0% 100% Yes 1
Meghalaya 4 0 0% 57% Yes 2
Jharkhand 4 4 100% 8% No 1
Uttarakhand 3 3 100% 15% No 3
Nagaland 2 2 100% 18% No 1
Dadra and Nagar Haveli and Daman and Diu 1 1 100% 33% No 0
Tripura 1 1 100% 25% No 1
Chhattisgarh 9 8 89% 28% No 14
Puducherry 2 0 0% 50% No 1
Andaman & Nicobar 1 0 0% 33% No 0
Arunachal Pradesh 1 0 0% 6% No 1
Manipur 1 0 0% 11% No 0

Winter 2025–26, week by week

The annual tables hide the season, and in north India the season is most of the story, so winter 2025–26 was analysed at weekly resolution. Daily PM2.5 from CPCB’s continuous monitoring stations was used, dropping days with less than 67% of hours reported and removing 77 daily values stuck at 985 µg/m³. That value is a sensor ceiling, not a measurement, and before its removal Hassan and Tirunelveli appeared as the most polluted locations in India. A station counts only if it has valid data on at least half the days of the season, which left 462 stations, and a weekly value needs at least 4 valid days. A location’s weekly value is the mean of its stations. The weeks are coloured with CPCB’s AQI bands for PM2.5, but a weekly mean is not an official AQI, so the colours serve only as a guide.

As a check on the method, the winter averages were compared with CREA’s published winter snapshot, which uses the same CPCB station data with a stricter 80% coverage rule. The values here were Ghaziabad 167, Noida 166 and Delhi 162 µg/m³, against CREA’s 172, 166 and 163 (CREA, March 2026), close enough to suggest the processing here adds little error of its own.

Weekly mean PM2.5 across monitored locations in India, 1 October 2025 to 28 February 2026. Spike height is the weekly mean; colour is the CPCB AQI band, used only as a guide.

The worst stretch ran from November into January, and the most polluted locations were almost all in the Delhi region: Ghaziabad, Noida, Delhi, Greater Noida and Gurugram top the list. Delhi’s worst week began on 12 November 2025 and averaged 248 µg/m³, roughly four times the 24-hour standard of 60 (which strictly applies to daily values, not weekly ones). There are no readings at all for 11 to 20 January 2026, so that week is missing from every city.

Heat strip of weekly PM2.5 AQI bands for the 20 most polluted locations

Weekly mean PM2.5 for the 20 most polluted locations with data in at least 18 of the 21 weeks, coloured by AQI band. Grey cells have no data.

Within Delhi-NCR the differences between stations are large. Of the 57 stations in the region, the worst over the winter were all in Delhi: Jahangirpuri (209 µg/m³), Wazirpur (207), Rohini (203), Anand Vihar (202), Bawana (200) and Vivek Vihar (199). Even the cleanest, New Industrial Town in Faridabad at 92 µg/m³, was above 60 on every valid day, and in Delhi itself the cleanest were IGI Airport (115) and Lodhi Road (117).

Map of Delhi-NCR continuous monitoring stations coloured by winter 2025-26 mean PM2.5

Mean PM2.5 at each continuous monitoring station in Delhi-NCR, 1 October 2025 to 28 February 2026. Delhi is shaded; dashed lines are district boundaries.

Winter 2026–27 so far (to 11 November)

Winter 2026–27 has data only up to 11 November 2026, about six weeks, so it is compared with exactly the same six weeks of 2025 rather than with the whole of winter 2025–26. To keep the comparison fair, only stations that reported in both years are used, which gives 229 locations. Nationally there is essentially no change: the median change across locations was a 0.6% fall, and 48.5% of locations were higher than in 2025. In the latest week with data, starting 5 November 2026, 59.5% of locations were above the 24-hour PM2.5 standard.

The NCR locations read lower than in 2025, with Delhi down 13%, Ghaziabad 23%, Noida 19% and Gurugram 18%, while some locations elsewhere went the other way (Ahmedabad was up 15%). It is too early to read anything into this. Six weeks is a short window, weather alone could plausibly shift these numbers by that much, and the worst weeks of winter 2025–26 came later in the season.

What this is, and what it isn’t

This is an exploratory look at public monitoring data, not research, and a few limits matter for how far the results can be taken.

  • The method changes in 2021. Up to 2020 a location’s value is the mean of its manual stations; from 2021 it is CPCB’s integrated value. Any change across that line is partly the network, not the air, which is why no decade-long improvement is claimed here.
  • Coverage is uneven. Monitoring stations are concentrated in urban areas, many states have only a few of them, and 69 small towns and village sites couldn’t be placed on the map. The state view only rates monitored locations, and only where they cover at least 30% of districts.
  • Annual averages and weekly means are not AQI. The decade view shows multiples of the annual standard, and the weekly colours borrow AQI bands as a guide.
  • Winter 2026–27 is covered for only six weeks. That is too short to call a trend in either direction.
  • Data gaps. There are no readings at all for 11 to 20 January 2026 and 24 to 26 October 2026, and some stations report only part of the season.

Turning this into something defensible would need a consistent station-level series across the whole decade, station siting information (kerbside, industrial or background), a way to fill the space between monitored locations, such as satellite-derived PM2.5, and a proper check of how much weather explains the year-to-year differences.

Data downloads

The cleaned data behind every map and number here is available to download:

Method

Data
CPCB National Air Quality Monitoring Programme annual tables, 2015 to 2024; CPCB continuous ambient air quality monitoring stations, accessed through OpenAQ, to 11 November 2026; state boundaries from Datameet following India’s official boundary; district boundaries from Census 2011; location positions from OpenAQ, a GeoNames-derived place list and Natural Earth
Processing
PDF tables extracted cell by cell; station rows assigned to locations from cell geometry; place names harmonised (545 locations, 476 placed on the map); 2015–2020 location value = mean of manual stations, from 2021 CPCB’s integrated value. Winter PM2.5: days under 67% hourly coverage dropped, 985 µg/m³ ceiling values removed, stations kept with valid data on at least half the season’s days, weekly values need at least 4 valid days
Checks
Annual mean within each row’s own minimum and maximum (9 rows failed and were excluded); fuzzy place matches checked by hand; duplicate location-years removed; winter averages compared with CREA’s winter 2025–26 snapshot. No correction for weather, and no consistent station-level series across the 2021 method change
Credits
The data extraction, processing, maps and initial draft of this write-up were produced with Claude AI (Claude Opus 5.5), working from the author’s questions and decisions, and reviewed by the author.
Code
Available on request.

A question for readers

If you know of station siting information for CPCB stations, or have worked with this data in another way, I would like to hear about it.

Respond to this exploration

Thoughts, corrections, or questions are welcome. No account needed. Selected responses are added below the post with my reply.