Microsoft's artificial intelligence predicts global air pollution in minutes

A new artificial intelligence (AI) model developed by Microsoft can accurately predict global air pollution levels in less than a minute. The model, called Aurora, is also capable of generating 10-day global weather forecasts.

Aurora is one of several AI weather forecasting tools being developed by tech giants, including GraphCast from Google DeepMind in London and FourCastNet from Nvidia, based in Santa Clara, California. But Aurora's ability to quickly predict global air pollution is groundbreaking, researchers say.

"For me, this is the first big step in atmospheric chemistry and machine learning," says machine learning researcher Matthew Chantry of the European Center for Medium-Range Weather Forecasts (ECMWF) in Reading, UK.

Traditional weather forecasting uses mathematical models of physical processes in the atmosphere, land and sea. To predict air pollution levels, researchers have previously used machine learning along with traditional mathematical models, Chantry says. Aurora appears to be the first purely artificial model to produce a global pollution forecast -- a task much more difficult than forecasting the weather, Chantry says.

"Wow, this is a really great result," he says. One advantage of AI models is that they often require less computing power to make predictions than traditional models, Chantry says.

Artificial intelligence researcher Paris Perdikaris of Microsoft Research AI4Science in Philadelphia, Pennsylvania, and colleagues found that Aurora can predict levels of six major air pollutants worldwide in less than a minute: carbon monoxide, nitrogen oxides, nitrogen dioxide, sulfur dioxide, ozone , and solid particles. His forecasts cover five days. It can do this "with less computational cost" than the traditional model used by ECMWF's Copernicus Atmospheric Monitoring Service, which predicts global levels of air pollution, the team wrote in a preprint [1] published on arXiv on May 20.

Aurora's predictions were as good as those of the traditional model. Policymakers use such forecasts to track air pollution and protect against associated health damage. Air pollution is linked to an increased risk of asthma, heart disease and dementia.

The researchers trained Aurora on more than a million hours of data from six weather and climate models. After training the model, the team tuned it to forecast pollution and weather around the world. The model generates a ten-day global weather forecast along with an air pollution forecast.

In some tasks, the team says, Aurora can outperform other AI weather forecasting models, such as GraphCast, which can outperform conventional models and make global weather forecasts in minutes. But it's too early to make a definitive comparison, says Chantry. says

Further research will show whether "fundamental" AI models trained on diverse data sets, such as Aurora, perform better than those trained on a single data set, such as GraphCast. "There's a lot of interesting science that can be done," he says. .