Bengaluru: On December 13, the air quality index (AQI) in Delhi peaked at 448. This means that air quality in the national capital was in the ‘severe’ category, a stage when toxic air seriously affects not only those with existing diseases but also healthy people.And it’s not just Delhi.Many parts of north India are witnessing toxic air; smog grounded more than 350 flights and delayed 400 others on Monday alone, per a report by The Tribune.According to the Centre for Research on Energy and Clean Air’s monthly air quality update for November this year, 20 of 29 cities in the National Capital Region (NCR) recorded higher pollution levels than they did during the previous year.Ghaziabad was the most polluted city in India during the month, recording a monthly average fine particulate matter (PM2.5) concentration of 224 µg/m³. Uttar Pradesh accounted for six of the ten most polluted cities, followed by Haryana with three cities, along with Delhi.The problem extends into the eastern part of the Indo-Gangetic plains where Patna, the capital of Bihar, is giving Delhi competition in this regard: on December 10, Patna was more polluted than Delhi was, with an AQI of 335; Bhagalpur followed at 280.Scientists at the Indian Institute of Technology Kanpur deployed 538 low-cost sensors at every block across all 38 districts in Bihar as part of the first project under the Union government‘s AMRIT (Ambient air quality Monitoring over Rural areas using Indigenous Technology) scheme. This covered rural areas, small towns, industry locations and urban areas, to generate data on PM2.5 concentrations across the state.The government’s existing monitoring system – the CAAQM or the Continuous Ambient Air Quality Monitoring system – in the state comprises only 35 monitoring stations.The team’s results, published as a peer-reviewed study on November 25, show that Bihar has five airsheds. The most polluted was the airshed over northwestern Bihar, consisting of nine districts including Sheohar, Gopalganj and Muzaffarpur. Brick kilns, sugar mills and thermal power stations contributed most to the pollution levels here. Pollution dispersion due to meteorological conditions was least in this airshed when compared to the others across Bihar – which meant that polluted air remained in the atmosphere for longer.An airshed is a geographical area in which the air functions as a single or homogenous unit especially in terms of dispersion of pollutants, due to reasons including meteorological factors.The team also found that there were clear seasonal patterns in the movement of air masses, and pollutant transport. Across all seasons, most air masses were generated locally and within airsheds, suggesting that a huge portion of PM2.5 concentrations are produced regionally, the study noted.During the winter and post-monsoon seasons, air masses mostly originated from the northwestern regions of India, including Punjab, Haryana, Delhi and Uttar Pradesh – suggesting a “significant influx of transboundary pollution into Bihar”.“The analysis further affirms that Bihar’s considerable exposure to PM2.5 concentrations was both local and long-range pollution primarily driven by seasonal northwesterly (westerly) winds, aligning with broader transboundary patterns observed across the Indo-Gangetic Plain,” the study noted.The Wire spoke to the corresponding author of the study, atmospheric scientist S.N. Tripathi, who is a professor at the Indian Institute of Technology Kanpur and dean of the Kotak School of Sustainability housed at the institute.India needs policy interventions that are overarching and not siloed – at the level of airsheds as well as adjoining airsheds, said Tripathi. In the case of tackling air pollution in the Delhi-NCR, bringing in machine learning will help identify the right clusters and airsheds for the area, and deploying low cost sensors here would decrease costs by at least 40 times, he told The Wire. Deploying more sensors would also mean that we can obtain data at a completely different, more detailed scale, he added.Here are edited excerpts from the interview.Q: What is the airshed approach and how is your study different in this regard?A: The concept of an airshed was primarily pioneered in California. People have been alluding to using the airshed approach for better air quality management in India. But taking ground-based dense measurements the way we have, and bringing in rigorous machine learning approaches to identify the airshed, and combining this with key meteorological features which are responsible for dispersing the pollutants – this has not been done to our knowledge, globally.So certainly, it is novel.Q: Your team installed 538 Sensor Ambient Air Quality Monitoring Network (SAAQM) monitors across every block in Bihar’s 38 districts, versus just 35 government-installed monitors – which is an exponential increase, as you say in your paper. You mention that this is an AMRIT project, tell us more about it?Image: Location of the SAAQM at every block across Bihar (blue dots) and the existing CAAQM (represented by the red dots). Source: Anandh et al 2025.A: It is the first government project of its kind under the centre, established under a PPP model at IIT Kanpur, under the aegis of the Principal Scientific Advisor’s Office of India, primarily supported by private industries and philanthropies.Q: You speak about using machine learning to identify airsheds – how did you use this in the study?A: Our primary idea was that if you have a bunch of data, can we see that the data can form some groups, or clusters. Machine learning and AI are very good at identifying such clusters, either similar groups of people or objects, or the differences between them. That was the key idea behind this, which has not been applied to such an extensive spatial or temporal dataset anywhere.Since we had a fairly long time series from 500-odd blocks in Bihar to measure PM2.5 – they have to be the key basis to identify airsheds or micro-airsheds. That will be determined, again, most importantly, by meteorological factors (specifically the ventilation coefficient in the study, which is derived from the wind and mixing height).So at a similar grid, we had the data, we also used the meteorological parameters from other data sources and we tried to combine them and start looking at those patterns using a pipeline of AI-driven models.We were surprised that very systematic and repetitive patterns started emerging over the state of Bihar. This was intuitive, and aligned with the way pollution sources are distributed across Bihar.Q: The study clearly identifies five airsheds in the state. The data also shows seasonal patterns – a mix of both local and regional pollution sources. You have identified brick kilns, thermal power plants and agricultural activity (such as stubble burning) as sources.Why is the identification of airsheds important in this regard? What was the kind of data that you were able to get through such a huge deployment of sensors across the block level?A: People have been trying to create an airshed over India. The problem is when you try to do this over a large country like India, which has an area of 32,00,000 sq km. That’s such a large area – different geographies, orographies, [pollution] sources, population densities. All these have a huge interplay. And let’s not disregard the weather.All of these will give rise to a pollution pattern which will vary across scale. There will be regional patterns (at the level of one or two states), at district or city level, and sub-district … you need data at these resolutions, and with sufficient or reasonable accuracy. So when people used modelled outputs, they might bring the resolution to 20 or 15 km, but the problem is that accuracy is suspect.When they use emission inventories, these are not complete in India, this is not a secret; second, resolution is again a problem. And third, emission inventories are only one part of understanding pollution. It only gives us the source. Because eventually the ambient pollution levels will be guided by meteorology and complex chemistry.Therefore, the airshed, in my understanding, has to be a combination of all these, but necessarily guided by the measured value of pollution. Now this is one thing that distinguishes this study from others.There are five airsheds over Bihar. This file photo (Credit: PTI) shows a layer of smog over the Patna airport on December 8, 2025.Q: Talking about Bihar specifically: what are the actions we would need to take at the level of these five airsheds? For instance, at the level of Airshed 1 (northern Bihar), which is the most polluted: what would we need to implement? We do have some restrictions in place to limit emissions from thermal power plants but these are not often implemented, enforced. What kind of actions are we looking at in terms of Bihar specifically to bring pollution down?A: Over Bihar, there are five airsheds. There are two ways we can look at it, at the high level, and in-depth interventions. The idea of an airshed is that instead of working in silos, you have to work in a cooperative fashion. In Bihar, a single airshed has around seven to nine districts under it. If these districts work together, they will achieve a lot more than the sum of their individual action.Sharing information, coordinated actions, sharing resources – will achieve a lot more per capita than doing it otherwise. That is true within airsheds, and also true between airsheds. That is the idea of the airshed. That it is a step ahead in having a very effective air quality management recognising the role that nature plays and this has to be kept in mind. Recognise this and then start acting at where we contribute to this.North Bihar is infested with brick kilns. What can we do? Can we transition to better technologies? Can we bring other kinds of incentives? Can we do something which ensures that brick kilns are not fired at the most critical time of air pollution? Because we have that kind of data available.Bihar’s air quality is also driven by biofuel consumption at the household level and otherwise. It seems that despite a lot of efforts as part of increasing the use of LPG and its distribution, it is still not used at the level that is expected. What can we do to ensure people do not use biofuel? That can dramatically make the air quality much better in Bihar.So we have to look at the airshed level, sharing information … this can inform the neighbouring airshed management too. It is a win-win situation. Where you say I am cutting down on this much, and by doing so, I am cutting down on a portion that contributes to your share of pollution. The other airshed management does the same.This will create a positive, synergistic cycle where eventually we will not only be cutting down on local emissions but also the transport of pollutants and therefore regional pollution into airsheds.Q: One of the other very interesting findings of the study is that there is a lot of transboundary pollution into Bihar as well, and you have identified that this comes from Delhi, Haryana, Punjab and Uttar Pradesh. This is how meteorology affects pollution levels.If we were to apply this to a region like the Delhi-NCR, which also witnesses both local and long-range pollution – how many airsheds does Delhi have and what would we need to do? Right now we have around 40 continuous air quality monitoring stations in Delhi-NCR. Do we need to scale that up to understand more details of the many factors that are contributing to the huge spike in AQI in the city in the winter months?A: We have this network [of sensors] in Bihar, but it also goes into Uttar Pradesh as part of AMRIT. This covers every block in Uttar Pradesh too. This is an ongoing study. But when we combine the data, we see a common airshed between Bihar and Uttar Pradesh. Very interestingly they run in the northern parts of UP and Bihar and the southern parts of these states, cutting across the boundaries of the states.Basically they are two airsheds in the north and south, having ten to 12 districts each; this shows why Uttar Pradesh and Bihar – certainly western Bihar and eastern Uttar Pradesh – need to work together. Depending on movement and meteorology and sources that are operating.Likewise, this will be the same for Delhi.In Delhi-NCR, sometime back, they established the CAAQM – to manage Delhi’s air quality adopting an airshed approach, it also includes districts from Haryana, Uttar Pradesh and Rajasthan.But the fact is that this [the CAAQM] is based more on heuristics. How do we know that we will not have more districts from Uttar Pradesh or Rajasthan [being part of the Delhi-NCR airshed]? Even districts in Punjab may be part of Delhi’s airshed, and there might be seasonality like we see in Bihar.All this requires frugal (because these are indigenous designs), cost-effective (a government monitor costs Rs 20-25 lakh, these sensors cost Rs 70-80,000 and if we scale it up it will reduce further) monitoring … we can then have more measurements so that we can create more credible micro-airsheds above Delhi-NCR and thereby cooperate better when there is very clear, undeniable, irrefutable evidence. Then parties, agencies, states [have to] start cooperating and that is what we need. That is where there is a lot of scope to do work.Q: One thing you stressed on is the importance of data and evidence. There is now emerging evidence that authorities may be trying to fudge data to some degree – such as spraying water near monitoring equipment in an effort to reduce the pollutants in its immediate vicinity and thus record lower values of air pollutants. How important would it be to make sure that such data cannot be tampered with?A: Any machine can be tampered with. You can try to bring security and data-related privacy inbuilt in the machine; the intent of people who are part of it also matters. There are now definitely enough technological advancements so that it can be made difficult to tamper with data. We need to have all those things in place when we monitor such important things.I totally agree with you that data sanctity is critical to have an efficient management of air quality. If data is not correct we will not be able to bring the right kind of interventions. And we will not be able to overcome and manage these things, so it [correct data] is very very important.When you create a good amount of good data, that also will tell a lot more. Democratisation of data is another way to have better systems in place. So that is how we need to go forward.Q: Our emissions inventories are still incomplete, you said. How important is it to complete this, update this, for the bigger picture? We do see higher emissions in the Indo-Gangetic plains for instance, also due to a lot of thermal power plants in these areas. So how important are emission inventories in understanding air pollution and do we need to get the ball rolling on this?A: Any country that has been enjoying good air quality has all these processes, technologies, data, inventories in place. They are all complementary, they do not make each other redundant. Models always require better and better emission inventories, improving the physics in the model, and machine learning is coming in. All these are part of very good and effective air quality management.