The new artificial intelligence tool uses a combination of machine learning, satellite imagery, and weather data to independently find places with high air pollution all the way down to a city block.
Scientists at Duke University have developed an artificial intelligence tool that will help researchers identify and reduce sources of hazardous emissions. In addition, it will be useful in studying the impact of emissions on human health in a specific location in the city.
The authors of the new AI tool are particularly interested in detecting PM2.5 particle levels.
PM2.5 is particulate matter less than 2.5 microns in size. Their diameter is 30 times smaller than that of human hair. These include a mixture of particles of dust, ash, soot, as well as sulphates and nitrates suspended in the air. It is these substances that cause air turbidity, which is typical for the centers of the largest metropolitan areas.
PM2.5 particles are able to travel deep into the airways and settle in the lungs. Inhalation of these particles can cause irritation to the eyes, nose, throat, or lungs, as well as bouts of coughing, runny nose and choking. But this does not exhaust the danger of their impact. The World Health Organization’s PM2.5 Particle Concentration Rate is 25 micrograms per cubic meter. Exceeding this limit can disrupt the normal functioning of the lungs and cause the development of many dangerous diseases such as lung cancer, respiratory tract infections and cardiovascular diseases.
A 2020 Global Burden of Diseases report reveals that 90% of the world’s population lives in areas where PM2.5 is hazardous to health. At the same time, in most cities there are no ground-based air monitoring stations due to the high cost.
In addition, they only give a general idea of the air pollution conditions in a particular region, but for residents of different areas of the city, these data are useless. To solve the problem, scientists created an instrument to measure PM2.5 in the 300-meter range (city block).
Using satellite data, weather indicators and machine learning, the researchers taught the algorithm to automatically find hot and cold spots of air pollution. The developers used a residual learning technique. The algorithm first estimates PM2.5 levels using only weather data. It then measures the difference between these estimates and the actual particle levels. Eventually, the algorithm learns to use satellite imagery to improve predictions.