Making the Invisible Visible through Super Learning

Mount Sinai scientists are reconstructing air pollution exposure at a 50-meter by 50-meter spot at more than 200 million locations in urban U.S. areas, going back 20 years

Two Mount Sinai researchers have developed new environmental exposure risk models capable of measuring environmental toxins at an unprecedented scale and scope. By layering complex statistical algorithms, known as ‘super learning,’ the models can predict health risk and reconstruct past risk from air pollution exposure at the granularity of a single house, up to two decades ago, across the contiguous United States.

Itai Kloog, PhD, Professor in the Department of Environmental Medicine and Public Health at the Icahn School of Medicine at Mount Sinai, and Heresh Amini, PhD, Associate Professor in the same department, integrate dozens of variables from satellites such as land surface temperature, distance to factories, major roads, open water and green spaces, as well as from sensitive monitors placed across the U.S. to assess a person’s health risk from air pollution at a more precise level than ever before achieved. 

“We’re like composers, bringing different pieces of data from satellites, distance to power plants, and more,” says Amini. “Then we orchestrate across all these different pools of data we have to take out something that’s meaningful for population studies, and more.”

Through collaborations with other Mount Sinai researchers, Kloog and Amini have published consequential findings in recent months, including in the emerging field of environmental justice.

In the past, environmental risk models merely estimated a person’s exposure risk based on limited air pollution data and temperature from a large nearby city. By contrast, Kloog and Amini’s models can pinpoint exposures at a 50-meter by 50-meter spot at more than 200 million locations in urban U.S. areas, going back 20 years. Their groundbreaking approach to calculating environmental health risk incorporates many disciplines, including satellite technology, machine learning, the science of air pollution, epidemiology and geography.

“Managing these large data sets isn’t something many other people have expertise or resources to do,” Kloog says.

The pair, who have collaborated together for the past decade, have also greatly expanded the number of sources of possible toxins in their studies and vastly broadened the number and type of pollutants considered. Their research includes toxins newly classified by The Environmental Protection Agency, such as black carbon and ultrafine particles, and gases previously overlooked by other researchers, including ozone.

Amini highlighted a surprising recent finding in which he and collaborators were able to pinpoint rocket launching sites as one of the sources of a higher density of metals in the air of certain residential areas. Amini also found that there are huge plumes of iron in the air around large U.S. airports that circulate in backyards miles away.

In a study published in December, Kloog, Amini, and other researchers found that as air pollution rises, humans face greater risk of dementia, psychiatric hospitalizations, and death. 

Kloog and Amini have also found evidence that poor communities face the greatest health risks from environmental exposures. In Los Angeles, for example, stricter environmental regulations over the years have reduced traffic-related air pollution for the general population. But during the same time period, poor communities in certain areas experienced a worsening of air pollution caused by an increase in the number of warehouses and ships used to transport goods.

The environmental health research pioneers say they hope their findings can help steer urban planners and regulators to choose optimal locations of new daycares, schools, nursing homes and air-quality monitors. Kloog and Amini have publicly shared their data through NASA.

“Through these techniques, we’re able to see effects we couldn’t see before,” says Amini. “We’re making the invisible visible.”