Atmospheric aerosols affect weather, climate and health – applications of neural networks are the next step in atmospheric science

Michael Boy’s group is aiming at building neural networks that help to include complex atmospheric chemistry and aerosol dynamics in large-scale models.

What are your research topics?

Atmospheric aerosols are particles that are 100–10,000 times smaller than the diameter of a hair, and their concentration in the atmosphere is in the range of hundreds to hundreds of thousands per cubic centimeter.

They originate either from direct emissions, like black carbon from combustion (we call them primary particles), or they are formed directly in the atmosphere when certain gas molecules form a cluster that can then grow through the condensation of other gases (secondary particles).

My research group investigates the reason why and when these particles are formed and how they impact weather, climate and health.

Where and how does the topic of your research have an impact?

Our planet is being threatened by many hazards, such as rising temperatures and severe air pollution in urban areas.

In the atmosphere, aerosols influence the formation of cloud droplets. The more aerosols there are, the more cloud droplets there will be – and the smaller the size of the droplets. Smaller cloud droplets will endure longer than bigger ones as they have a lower probability to grow to the size of a rain droplet and precipitate. They will thus affect the amount of radiation reaching the surface of the earth and have a cooling effect on our climate.

In urban areas with high emission of anthropogenic pollutants, tiny particles are harmful to our health. They can penetrate deep into our respiratory system and, if concentrations are high for a long period, cause severe health problems.

If we want to provide recommendations to decision makers on how to reduce emissions of anthropogenic pollutants like nitrogen oxides, sulphur dioxide or aromatics, it is crucial that we achieve a complete understanding about the aerosol formation processes and the complex atmospheric chemistry. Only then can we be sure that the decrease of emissions of compounds, such as those mentioned above, will not imply unwanted negative feedback on the concentrations of aerosols.

What is particularly inspiring in your field right now?

Currently we apply different techniques from machine learning and artificial intelligence to create and solve complex atmospheric chemistry schemes and aerosol dynamics. The first aim is to tackle certain chemical systems that haven’t so far been included in atmospheric chemistry models but have had a crucial impact on the aerosol formation.

Secondly, we want to create neural networks for combined chemistry and aerosol dynamics. This neural network could later be applied to regional weather forecasts, as well as air quality and climate predictions.

These large-scale models require immense computer resources and normally apply simplified parameterizations, which could involve additional uncertainties. To succeed with our neural networks would be groundbreaking for atmospheric science, as such networks would be more than 1000 times faster compared to detailed models.


Michael Boy is Professor in Atmospheric Sciences and Applied Mathematics at the Faculty of Science.

Read about the other newly appointed professors.