|Ozone is a highly reactive gas that is a combination of both man-made and natural products in the upper and lower atmosphere of the Earth. / Photo by: studio23 via Shutterstock|
Researchers from the University of Houston have developed an ozone forecasting system that uses artificial intelligence, allowing it to predict the ozone levels in certain areas 24 hours in advance. The team’s research was recently published in the journal Science Daily.
Why the need to predict ozone level
Ozone or O3 is a highly reactive gas that is a combination of both man-made and natural products in the upper atmosphere of the Earth as well as the lower atmosphere. The ground-level ozone or tropospheric ozone is formed from photochemical reactions between these two air pollutants: nitrogen oxides and volatile organic compounds. Depending on the presence of sunlight and heat, the photochemical reactions of the two air pollutants result in higher ozone concentrations during summer months.
The unstable gas in the atmosphere causes respiratory problems, especially among individuals who are susceptible to it, such as young children, the elderly, and those with asthma. The study’s first author Alqamah Sayeed said that the current ozone forecasting systems available require several hours to determine or predict the ozone levels. The systems are also less accurate compared to using artificial intelligence.
Using convolutional neural networks
Sayeed and the team’s model, on the other hand, correctly predicted ozone levels in the local area a day in advance and with an accuracy rate of 85 to 95 percent. The authors explained that the key difference between past ozone forecasting methods versus using AI is the convolutional neural networks. CNN is a deep learning algorithm, which takes an input image, assigns the learnable weights and biases to different objects or aspects in the image, and then differentiates them. During their test, they proved that CNN is “capable of sweeping” the data and utilizing it to generate assumptions, depending on what it has previously learned.
This is why the convolutional neural networks are generally utilized to enhance image resolution. The team also shared that they used air pollution and meteorological data gathered at 21 different stations in Houston and other areas in Texas as provided by the Texas Commission on Environmental Quality. All the data represent the conditions in the area from 2014 to 2017. First, the team programmed the CNN using the meteorological information, which includes wind speed, barometric pressure, and temperature as variables. They then included the ozone measurements from every station from 2014 to 2016.
|CNN takes an input image, assigns the learnable weights and biases to different objects or aspects in the image, and then differentiates them. / Photo by: Zapp2Photo via Shutterstock|
Testing the technology
To test the technology’s accuracy in forecasting the ozone levels based on the meteorological conditions of the past day, they also used the 2017 weather data. This is when the AI-based ozone forecasting system was able to reach 90 percent accuracy. They also believe that the technology will become more intelligent over time because the network continuously learns the data.
Can it be applied to other countries?
The group said that although they only used the Texas data, the ozone forecasting system can be used in other areas of the world. They said that even though the United States is not geographically the same as East Asia, the chemistry and physics of ozone creation are not different. Sayeed pointed out that they are currently expanding the AI model by including predictions for other pollutants, such as particulate matter. They also aim for their system to forecast beyond 24 hours.
Their study titled "Using a deep convolutional neural network to predict 2017 ozone concentrations, 24 hours in advance" also states that they used a five-layer deep CNN architecture to produce the real-time predictions of ozone concentrations. Among the inputs fed into the AI were the wind speed, wind direction, temperature, dew point temperature, solar radiation or net radiation, pressure, precipitation, pressure, NOx, and ozone.
Other researchers were Yunsoo Choi, Jia Jung, Anirban Roy, Yannic Lops, and Ebrahim Eslami. All of them are from the Department of Earth and Atmospheric Sciences. They concluded that concentrations in the ozone steadily rose during fall until September and is at its lowest level in December.
In 2017, the regression in self-organizing map (REGiS) was also introduced by Penn State’s Nikolay Balashov. His work forecasts the surface ozone level 48 hours in advance and he used weather patterns and statistical air quality models. “If we can predict the level of ozone ahead of time, then it’s possible that we can do something to combat it,” he said.
AI environmental application and its impact on worldwide employment
Artificial intelligence on environmental applications also impacts net employment worldwide. Database company Statista believes that East Asia will witness the most job gains by 2030 from using AI for environmental purposes, by increasing its workforce by 2.5 percent. The equivalent of such a percentage gain would be about 25.1 million added jobs. Europe’s net employment growth rate because of AI environmental applications would be 1.4 percent, Middle East and North America 1.1 percent, North America 1 percent, Central and South America 1 percent, Sub-Saharan Africa 0.5 percent, and Indo-Pacific 0.4 percent.
Aside from forecasting ozone concentrations, there are also other uses of AI that are related to the environment. For instance, it is used to monitor endangered species, optimize crops, and track diseases as well.