Artificial intelligence tracks spread of urban pollution

Engineers at Cornell University have used artificial intelligence to improve models that can track the soot, dust and exhaust fumes that go into urban air pollution and human lungs.

The Cornell engineers say city planners and government health officials can obtain a more precise picture of the well-being of urban dwellers and the air they breathe using this new research.

“Infrastructure determines our living environment, our exposure,” says senior author Oliver Gao, the university’s Howard Simpson Professor of Civil and Environmental Engineering in the College of Engineering. “Air pollution impact due to transportation – put out as exhaust from the cars and trucks that drive on our streets – is very complicated. Our infrastructure, transportation and energy policies are going to impact air pollution and hence public health.”

Previous methods to gauge air pollution were cumbersome and reliant on extraordinary amounts of data points. “Older models to calculate particulate matter were computationally and mechanically consuming and complex,” says Gao, a faculty fellow at the Cornell Atkinson Centre for Sustainability. “But if you develop an easily accessible data model, with the help of artificial intelligence filling in some of the blanks, you can have an accurate model at a local scale.”

New York coughs up the data

Lead author Salil Desai and visiting scientist Mohammad Tayarani, in collaboration with Gao, developed four machine learning models using data gathered in New York City's five boroughs, which have a combined population of 8.2 million people and daily-vehicle miles travelled of 55 million miles.

Ambient air pollution is a leading cause of premature death worldwide. According to a study by The Lancet, in 2015, over 4.2 million annual fatalities – as a result of cardiovascular disease, ischemic heart disease, stroke and lung cancer – were attributed to air pollution.

The team's equations use a few inputs, including traffic data, topology, and meteorology, in an AI algorithm to learn simulations for a wide range of traffic-related air pollution concentration scenarios. The best-performing model was the Convolutional Long Short-term Memory (ConvLSTM), which trained the algorithm to predict many spatially correlated observations.

“Our data-driven approach – mainly based on vehicle emission data – requires considerably fewer modelling steps,” says Desai.