Many people are considering air quality and wondering what to anticipate in the coming days as a result of the intense fire season in Canada.
Small particles and gaseous substances are present in all air. But as air quality deteriorates, these gases and particles can enter the nose, throat, and lungs, as well as circulate in the bloodstream, causing asthma attacks and aggravating heart and respiratory issues. Asthma emergency department visits surged when wildfire smoke turned New York City’s skies orange in early June 2023.
It’s simple to locate a daily air quality index score in most cities that indicates when the air is deemed unhealthy or even dangerous. It is more difficult to forecast the quality of the air in the coming days.
In my capacity as a professor of civil and environmental engineering, I forecast air quality. These forecasts have been enhanced by artificial intelligence, but research indicates that it is far more effective when used in conjunction with conventional methods. This is why:
How experts forecast air quality
A chemical transport model or a machine-learning model are often the two main approaches used by scientists to forecast air quality in the near future, usually a few days ahead or longer. The output produced by these two models is entirely different.
Chemical transport models compute the production and presence of air contaminants using a variety of well-known chemical and physical formulas. They make use of data from meteorology, which gives atmospheric information such as wind, precipitation, temperature, and solar radiation, as well as data from emissions inventory published by local authorities that include pollutants from known sources, such as wildfires, traffic, or industry.
These simulations represent the movement and chemistry of air contaminants. However, their simulations contain numerous variables that have a great deal of uncertainty. For instance, cloudiness modifies the photochemistry by altering the solar energy that enters the atmosphere. The results may become less precise as a result.
Machine learning is used in the EPA’s AirNow air pollution forecasts. A smoke-transport and dispersion model aids in simulating the movement of smoke plumes during wildfire outbreaks. The forecast for August 9 is shown on this map. Orange represents poor air for sensitive individuals, while yellow denotes moderate danger. AirNow.gov
Instead, machine-learning models gather historical data over time to identify patterns that may be applied to present conditions to forecast future air quality for any particular place.
Contrary to chemical transport models, machine learning models do not take into account any chemical or physical mechanisms. Additionally, if the models weren’t trained on data related to extreme circumstances, such as heat waves or wildfire outbreaks, the accuracy of machine-learning estimates may be incorrect. Consequently, while machine-learning algorithms can indicate the locations and times where high pollution levels are most likely to occur, such as during rush hour close to motorways, they typically cannot handle more sporadic occurrences, such as wildfire smoke coming in from Canada.
What is superior?
Scientists have found that neither model is sufficient for predicting air quality on its own, but combining the best features of both models can improve accuracy.
The machine learning and measurement model fusion, or ML-MMF, is a combination approach that can make predictions based on research that are more than 90% accurate. Based on well-established physical and chemical processes, it can replicate the entire process, from the source of air pollution to your nose. They can more accurately educate the public about air quality safety levels and the direction contaminants are moving by adding satellite data.
Recent comparisons between the three models’ predictions and actual pollution measurements were made. The findings were startling: the combined model performed 12% better than the machine-learning model alone and 66% better than the chemical transport model in terms of accuracy.
Despite the fact that applications using machine-learning algorithms are growing in popularity, the chemical transport model is still the most often used technique for forecasting air quality today. The U.S. Environmental Protection Agency’s AirNow.gov regularly forecasts weather using machine learning. Additionally, the site gathers air quality forecast outcomes from regional and local organizations, the majority of which employ chemical transport models.
The combined models will become more accurate ways to forecast hazardous air quality as information sources grow more trustworthy, especially during erratic events like wildfire smoke.