When an air pollution episode starts in a city, it’s important to understand the expected scale of the problem as early as possible to respond most effectively. As part of our internal research projects, our mathematical modeling team is training neural networks to classify air pollution episodes by their potential size and possible cause. This is based on an array of data from about 500 events that occurred in the same city over a year.
We think of a city as a space with a large pool of air — a part of the atmosphere where impurities are collected and mixed, which all citizens breathe.
However, certain areas of this pool may have local pollution sources, resulting in air quality that can be considerably worse than the average. Pollution episodes may also arise from the transfer of pollutants across city/community borders (typically from wildfires) or from the buildup of harmful substances in the air (often due to weather conditions like cold and still weather).
The sooner an episode is identified, the quicker and more effective actions can be taken. For example, if the pollution came from outside, city authorities may not be able to address the source directly but can minimize community’s exposure by warning people and organizations. These might include advice to stay indoors, close windows, turn on ventilation systems, and requests for enterprises to reduce emissions (as the air is already dirty).
Our model analyzes measurements from an entire network of monitoring stations and anticipates the scale and dynamics of pollution dispersion. Currently, it classifies an episode with 90% accuracy within 24 hours of its onset. The classification accuracy is nearly 100% for common events and lower for rare ones, which is attributed to limited data for neural network training.
We continue our research, as we still have a lot of questions, such as "how to train a model to predict the pollution episode type and scale prior to its start."
We will keep you updated on the results.
*Photo taken on November 2, 2023 in Gurugram (India)
However, certain areas of this pool may have local pollution sources, resulting in air quality that can be considerably worse than the average. Pollution episodes may also arise from the transfer of pollutants across city/community borders (typically from wildfires) or from the buildup of harmful substances in the air (often due to weather conditions like cold and still weather).
The sooner an episode is identified, the quicker and more effective actions can be taken. For example, if the pollution came from outside, city authorities may not be able to address the source directly but can minimize community’s exposure by warning people and organizations. These might include advice to stay indoors, close windows, turn on ventilation systems, and requests for enterprises to reduce emissions (as the air is already dirty).
Our model analyzes measurements from an entire network of monitoring stations and anticipates the scale and dynamics of pollution dispersion. Currently, it classifies an episode with 90% accuracy within 24 hours of its onset. The classification accuracy is nearly 100% for common events and lower for rare ones, which is attributed to limited data for neural network training.
We continue our research, as we still have a lot of questions, such as "how to train a model to predict the pollution episode type and scale prior to its start."
We will keep you updated on the results.
*Photo taken on November 2, 2023 in Gurugram (India)
