Harmful air pollution could soon be identified using satellite images, which may open doors for quarries to tackle exposure of workers to high levels of dust and particulate matter emissions in their operations.
Researchers from Duke University in North Carolina have discovered a way to estimate air quality on small land areas by using only satellite imagery and weather conditions.
The results, which were published online in the Atmospheric Environment journal, used micro-satellite images to determine ground-level air pollution.
This could allow researchers to identify hidden areas of air pollution, along with improving studies of how human health is affected by pollution.
“We’ve used a new generation of micro-satellite images to estimate ground-level air pollution at the smallest spatial scale to date,” Duke University professor of civil and environmental engineering Mike Bergin said. “We’ve been able to do it by developing a totally new approach that uses AI/machine learning to interpret data from surface images and existing ground stations.”
The study focused on predicting levels of microscope airborne particles known as PM2.5. – which have a diameter of less than 2.5 microns (µm). These particles have caused health concerns for humans due to their ability to travel into the lungs.
PM2.5. is ranked as the fifth mortality risk factor globally, according to the 2015 Global Burden of Disease study. More recently, it has been linked to an increased death rate due to COVID-19 by Harvard University’s TH Chan School of Public Health.
While the study was not specifically conducted for quarries, Bergin said that due to the method’s ability to estimate air quality (particulate matter concentrations) down to 100 metres, it could be used to measure particulate matter concentrations for surface industrial activity.
Bergin said the influence of quarry vehicles, operations and equipment on local air quality in a quarry also have the possibility to be measured.
Random forest algorithms
Current methods for detecting the amount of PM2.5 in an area require satellites that measure the amount of sunlight scattered into orbit by the ambient particulates. However, the study notes that clouds and shiny surfaces can cause mismeasurements through this method.
By using machine learning, Duke University assistant professor of civil and environmental engineering David Carlson assisted the research teamwith the development of an algorithm that combines images of Beijing with meteorological data from Beijing’s weather station.
The machine learning process – also known as random forest algorithms – is used in combination with a convolutional neural network. This allows for common features in images to be grouped together to find the most common aspects, which are then added to the random forest algorithm with weather data.
“High pollution images are definitely foggier and blurrier than normal images, but the human eye can’t really tell the exact pollution levels from those details,” Carlson said. “But the algorithm can pick out these differences in both the low level and high level features — edges are blurrier and shapes are obscured more — and precisely turn them into air quality estimates.”
The study used a total of 10,400 images to predict local levels of PM2.5 for the model used. A further 2622 images were tested to see how efficient the model was at predicting PM2.5, which was found to predict levels down to 40,000m², which is a huge jump from the one million square metres that current methods can predict.
“We think this is a huge innovation in satellite retrievals of air quality and will be the backbone of a lot of research to come,” Bergin said. “We’re already starting to get inquiries into using it to look at how levels of PM2.5 are going to change once the world starts recovering from the spread of COVID-19.”