How AI is making affordable air pollution sensors more accurate

Clean air is a fundamental right. However, every day, 100 children under the age of five tragically lose their lives in east Asia and the Pacific due to a silent killer – air pollution.

In response to this crisis, huge investments have been made in outdoor air pollution monitoring systems. These fridge-sized monitoring stations are expensive costing at least £10,000 each, so scaling this up everywhere isn’t financially viable.

Now, a new generation of small, roaming air sensors could better inform people about pollution levels in their local area. Currently, these sensors just aren’t precise enough. Our recent research shows that AI could enhance their accuracy by up to 67%.

These compact devices are the size of a thick mobile phone and cost just a few thousand pounds. They can easily be attached to vehicles such as buses, courier vans and bin lorries that already cruise through our streets. By gathering data wherever people live, work or play, these roaming sensors can build a real-time air quality map that reflects the local environment much more accurately.

The main contributor to air pollution is the burning of fossil fuels – this also produces greenhouse gas emissions that contribute to climate change. Alarmingly, air pollution is responsible for 7 million deaths every year. Children are especially vulnerable due to their developing lungs, weaker immune system, and faster breathing rate.

Last summer, we spent 12 weeks collecting air pollution data from both inexpensive gas sensors and reference instruments at a UK national facility, the Weybourne Atmospheric Observatory in Norfolk.

Every minute, the roaming air sensors transmit pollutant levels to a data centre – that includes particulate matter (such as fine particles of soot), carbon monoxide, ozone, nitrogen dioxide and sulphur dioxide. But we found inconsistencies between data from sensors and data from the national reference instruments.

Before this tech can be scaled up, the accuracy of readings from these air sensors needs improving. We have been studying how the problem-solving ability of AI can be used to improve air pollution monitoring. With a clearer understanding of the complex relationships between different gases, pollutants and environmental conditions, AI can correct any measurement errors.

Currently, measurements are influenced by the presence of other pollutants and environmental factors, such as temperature and humidity. Understanding how a dozen parameters simultaneously affect a specific gas measurement is a real jigsaw puzzle. Even the manufacturers of these sensors have not managed to crack it yet.

During our experiments, we identified the main causes of measurement inaccuracies for each gas sensor. With this information, our AI-driven solution slashed errors by up to 67%. Data science turned flawed, yet promising, sensors into precision tools that can help people seeking cleaner air. With breakthroughs like this, air quality insights could be more quickly scaled up.

From stations to sensors

Better monitoring will improve our understanding of local air pollution sources, their effects on residents, and help pinpoint the sources responsible, plus lead to more tailored warning systems.

In February 2024, the US Environmental Protection Agency committed US$83 million (£64.5 million) to expand and upgrade its air pollution monitoring network. This tech can be used to better understand the threat, shape policies and shape emergency measures. But monitoring stations are expensive, typically £10,000 to £30,000 per unit – and that’s without considering installation and maintenance charges.

Bangladesh, one of the hardest-hit countries in terms of air pollution, has only 11 monitoring stations. Even a wealthy city like London has fewer than one station per 100,000 residents. This is inadequate because pollution levels may significantly vary between neighbouring streets.

Remote sensors offer a more affordable and practical solution, if they can be sufficiently accurate. To be commercially feasible, AI mustn’t make these sensors more expensive. Transparency about how this system makes decisions is also critical. We used simple AI technology that operates on a microcontroller — a tiny computer within the device — to keep the additional cost of incorporating AI to under a few pounds and minimise its energy cost.

Imagine joggers checking local readings before choosing a route, or parents scanning the latest updates to find the safest playground for their children. Suddenly, air quality becomes more than a distant worry: it’s a practical guide to better health.

Gone is the guesswork, replaced by knowledge that helps people make healthier choices. When every breath matters, this tech ensures people no longer have to gamble with their health based on where they live.

Don’t have time to read about climate change as much as you’d like?

Get a weekly roundup in your inbox instead. Every Wednesday, The Conversation’s environment editor writes Imagine, a short email that goes a little deeper into just one climate issue. Join the 40,000+ readers who’ve subscribed so far.