There are few issues as important as keeping the air we breathe clean. When air is free of toxins, it’s almost like we don’t notice. When air quality is less than stellar, it’s — also — almost like we don’t notice (except in the most extreme of examples).

In countries like Paraguay, the air quality problem is often an invisible one and ranges from fine one part of the year to dangerous other parts of the year. Broaden the scope globally and issues around the destructive greenhouse gas that is methane continues to be a part of the climate change conversation.

This year’s Mozilla Technology Fund spotlights the intersection of tech and sustainability. AI is known for carrying a hefty environmental footprint but there are many reasons to use AI – sometimes, even, to help solve the climate change problem. Two Mozilla Technology Fund awardees, Fernanda Carles and Satish Kumar are working to make an invisible problem visible, and they’re using technology to do it.

What Do Cars, Farms and Trash Have In Common? Methane.

It’s no secret that our reliance on fossil fuels is a major roadblock as we move toward a green future. “Fossil fuels constitute over 30% of total methane emissions and the next 30% come from the agriculture industry from cows and dairy farms,” says Satish Kumar. Satish is a PhD candidate at UC Santa Barbara in the US and works with the group EyeClimate on tracking methane emissions using the tool MethaneMapper. Satish notes that MethaneMapper is the largest publicly available methane dataset. “The third main source of methane emissions, in addition to fossil fuels and agriculture, is mishandled trash. These three sources lead to methane emissions which are responsible for 30% of global warming. If we could address the methane problem, then 30% of the climate change problem would be solved,” says Satish.

We see how the effects of fossil fuels can play out on a smaller scale. While more affluent countries can more easily make the switch to electric cars, developing nations are getting left behind. “In Paraguay, we don’t have good public transportation and most of our cars use diesel fuel, the worst fuel from an emissions standpoint,” says Fernanda Carles. Fernanda is a mechatronic engineer studying air quality in her country with the help of the National University of Asunción. Her project, soon to be open source, uses AI to track air quality in her home region. “In addition to fossil fuels, farmers here will burn grassland in an attempt to essentially ‘re-do’ the soil as a quick way to clean it,” says Fernanda. “This can cause air pollution to spike. Additionally, a big source of low air quality in Paraguay is wildfires. In the Global North, air pollution is linked to industry but in Paraguay, high rates of pollution come from wildfires, which can make air quality harder to predict and track.”

What’s The Climate Impact Of Using AI To Measure Climate Impact?

You can’t improve what you don’t measure. Fernanda and the National University of Asunción’s marks one of the first groups to track air quality in Paraguay, along with groups like Aire Libre. AI Air Quality’s goal, much like MethaneMapper’s goal, is to improve data around air quality in Asunción and raise awareness around an invisible problem.

Both projects use a combination of affordable hardware and AI-powered software to collect data about the air quality in Asunción and around the world. If you’ve read other climate stories on this blog, by now you’re likely thinking, “what’s the climate impact of the tools they’re using to collect this data?”

Fernanda notes the goal was always to use lightweight tools, but not for reasons you’d assume. “In our case, we were thinking more about processing time,” says Fernanda. “We wondered, ‘how could we run experiments all night long, from 8pm until 8am the next day.’ A secondary result of this meant our AI models were pretty light with low processing time and, as a result, low energy.” Because the low energy software was quicker, it had the dual benefit of being quick to use and better for the planet than something more energy-intensive.

As for MethaneMapper, Satish notes that the project did gobble up resources, but only at first. “The training of the AI model is expensive,” says Satish, “but once the training is complete, inference-making is never expensive.” Making inferences is how the software arrives at collecting data.

In sum, the two may be using AI, but lightweight versions or uses that are now less energy intensive. And, at the very least, this energy is being put toward climate solutions and not generating new cat pics like we were warned against.

Clearing The Air

Both Mozilla Technology Fund awardees have a similar goal — to track air quality and create lasting impact beyond each project. For Satish, it’s to allow others to create air quality tools that stand on the shoulders of MethaneMapper and, ultimately, reach new heights. Not only is MethaneMapper the largest publicly available dataset accessible by anyone, Satish and the folks at EyeClimate are allowing others to train their machine learning tools using the MethaneMapper. In six months, the team has seen more than 160 downloads of their data.

As for Fernanda and the National University of Asunción, their goal is more analog. “The idea of this project is to keep the conversation going,” says Fernanda. “We have the research, we have the tools. We can show you how the air quality is and we can tell you how this is impacting your health. We want people to start talking about this.”

Two Projects That Put the “AI” in “Air Quality”

Written By: Xavier Harding

Edited By: Audrey Hingle, Kevin Zawacki, Tracy Kariuki

Art By: Shannon Zepeda