How is Air Quality Predicted ?

Predicting air quality makes it possible to anticipate measures to reduce polluting emissions.

Many cities around the world (Beijing, Ulaanbaatar, Bombay, New Delhi, Lagos, Mexico City, Paris, etc.) regularly experience peaks in air pollution. Air pollution is characterized by the concentration of ozone on the ground. The inhabitants sometimes live there almost all year round under a thick fog which does not allow them to see beyond a hundred meters. In fact, they are forced to limit, sometimes very drastically, their outings by covering their mouths with a handkerchief or by putting on a mask in order to avoid the harmful effects of the air they breathe.

It is therefore important to know and monitor the quality of the outdoor air in your region in order to adapt your behavior. In France, the Prev’Air site, the air quality forecasting platform in France, provides data and forecast maps comparable to those of the weather forecast. In the event of a pollution peak and episode, the authorities can initiate measures to reduce polluting emissions in advance.

So how is the air quality forecast developed?

How is air quality modelled?

Before predicting any phenomenon using numerical and quantitative methods, it is very important to model it. Thus, the National Institute for the Industrial Environment and Risks tells us that “air quality modeling is based on digital tools that simulate the chemical and physical processes responsible for the evolution of pollutant concentrations in the air. ‘air. These models are fed as input with data and variables such as:

Pollutant emissions from several sectors of activity (road traffic, industrial and agricultural activities, residential heating, maritime transport, etc.),

Meteorological conditions that influence the dispersion of pollutants and also the intensity of chemical processes

Boundary conditions that provide information on pollution contributions from distant sources”.

More details on the forecasting methodology presented on the INERIS website

What is air quality forecasting?

“Economic activity in France and Europe will experience a marked slowdown in 2023, then pick up again in 2024 and 2025.

From meteorological measurements, such as wind speed and temperature, it is possible to predict whether ground-level ozone will be at a high enough level in the coming days to trigger a public air pollution alert. , as can be seen on road and highway billboards, especially around major cities such as Paris, Lyon, Bordeaux and other cities. Air quality forecasting is a process aimed at predicting, with acceptable confidence intervals, the concentrations of atmospheric pollutants over a very short term (one to a few days). It is generally applied to regulated pollutants, such as ozone, nitrogen dioxide and PM10 and PM2.5 particles, to anticipate the arrival of critical situations in which concentrations are likely to exceed regulatory values.

The forecasts are based on air quality models whose outputs can be optimized with additional processing (statistical correction) which notably involves observations from previous days in order to improve the capabilities of the models for detecting episodes. pollution.

For example, using a set of data describing meteorological observations over a fairly long history (3 to 5 years) in a given region on ozone levels, we train a machine learning model (see our Machine Learning training course and our articles) to identify the structure of the data and the various trends observed in the history. It will therefore be able to predict what the pollution trend will be in the days to come.

Air pollution prediction methods can be divided into three categories: statistical prediction methods, artificial intelligence methods, including machine learning, and numerical prediction methods. New so-called “hybrid” methods and models have been developed and have improved the accuracy of forecasts. This classification is not, strictly speaking, specific to this field. A detailed study “ « Air Pollution Forecasts: An Overview ” presents a synthesis of forecasting models in this field.

An example of a forecasting model: the CHIMERE model

The INERIS website presents an example of the CHIMERE forecasting model. It is a chemistry-transport model developed and implemented by the atmospheric modeling and environmental mapping unit of INERIS in collaboration with the CNRS.

“It is a deterministic model. Unlike statistical models, it is not calibrated on observations but is based on the physics and chemistry equations of the atmosphere that govern the transport and transformation of pollutants. The input data it requires are the emission fluxes induced mainly by human activities, and also the meteorological fields that influence the accumulation of pollutants.

This is a so-called “regional” model developed to simulate air quality over areas extending from a few tens (agglomeration) to a few thousand kilometers (continent). But the model has also been deployed as part of a proof of concept throughout the northern hemisphere. For this, significant work has been done to modify geographical projections, take into account the specificity of land use and the cycle of vegetation on a global scale and adapt inventories of emissions on a global scale. . A breakthrough made possible thanks to the computing resources of the Research and Technology Computing Center (CCRT) and which has resulted in the production of high-resolution air pollution maps over the entire northern hemisphere: fine particles, ozone, carbon dioxide, nitrogen, plumes of desert dust”.

 More details on the CHIMERE model on the INERIS website which also presents a 5-minute video on air quality modeling.

In conclusion, the extent of air pollution, especially in large cities, has prompted public authorities, regulatory bodies and cities to support research in the field of forecasting to anticipate pollution peaks and adapt the essential alerts to best protect the health of people and especially those who are the most vulnerable such as children, the elderly and those suffering from health problems.

In terms of forecasting methods and models, the contribution of data science, big data and artificial intelligence (Deep Learning and Machine Learning) is undeniable in improving the accuracy of pollution peak forecasts.