Researching, learning and adapting

One of the greatest unknowns in climate change is the question of how particulate matter affects clouds. Yu Wang is using machine learning and satellite data to investigate the surprising role of these tiny particles in the atmosphere.
“Clouds act like an umbrella for the Earth, cooling it down.” Yu Wang investigates how precisely aerosols and the cooling effect of clouds work. (Photograph: ETH Zurich / Nicola Pitaro)

In autumn 2014, Iceland’s Holuhraun volcano erupted, spewing daily about 120,000 tonnes of sulphur dioxide into the air at its peak. A thick belt of tiny particles, also known as aerosols - a notorious air pollutant, formed in the atmosphere above this otherwise virtually unspoiled region. This volcanic eruption served as a very good natural experiment that allowed climate researchers to study how the sudden upwelling of particulate matter affected clouds. “Since aerosols can promote the formation of cloud droplets, they are an important factor in projecting climate change but we still know very little about it,” Wang explains. Since September 2021, the 30-year-old environmental scientist has been an ETH Zurich Fellow at the university’s Institute for Atmospheric and Climate Science, working as a member of the group run by Ulrike Lohmann, Professor of Atmospheric Physics.

Wang talks enthusiastically about a study – published recently in Nature Geoscience – that she, her husband Ying Chen and Ulrike Lohmann co-authored along with other researchers from the British Met office, the Universities of Exeter, Cambridge, Leeds (UK), and Munich (Germany) and NASA (US). She laughs warmly from time to time, clearly delighted by the interest being shown in her results. “I’m really excited about my work,” she says. “Emissions that affect the climate essentially fall into two groups: greenhouse gases and aerosols.” Greenhouse gases heat up the planet, while aerosols counteract this effect mainly through cloud formation.

“Clouds act like an umbrella for the Earth, cooling it down,” Wang says, spreading her arms wide to illustrate her point. But the problem, she adds, is that we are unable to quantify with precision how aerosols and the cooling effect of clouds work. According to the Intergovernmental Panel on Climate Change (IPCC), aerosols are the primary source of uncertainty when it comes to understanding how humanity has impacted the current climate.

So when Iceland had its volcanic eruption, climate researchers jumped at the chance to study the effects of the aerosols released during this event: they compared the clouds over the North Atlantic in autumn 2014 with the situation in the years before and after. But this comparison proved inconclusive because cloud formation also depends largely on the weather, which was different during the eruption from that in the other years.

Machine-made meteorologists

“We also used the volcanic eruption in our work,” Wang says. “But what we did was to apply a machine learning method that can tell us what the clouds are like under certain weather conditions.” This makes it possible to use data from the “clean” years to determine what the cloud situation would have been in 2014 had there been no eruption. “It’s like having a weather forecast,” Wang says. By comparing the machine learning forecast for the cloud situation minus the Holuhraun eruption with data of clouds in the same months in years before and after the eruption, it’s possible to say that the difference is due entirely to the aerosols.

The result of this study surprised the researchers because it contradicts previous notions. “It’s also important to know,” Wang says, “that interactions between aerosols and clouds produce two different effects.” An increase in emissions results in a higher number of cloud droplets, but these are smaller. This makes the clouds brighter, which means they reflect more sunlight away from the Earth. A higher number of smaller droplets also means that the clouds can retain more water before it rains, meaning the clouds last longer. “People used to think that it was cloud brightness that dominated the cooling effect, but we discovered that a cloud’s lifespan or the formation of new clouds is more important,” Wang says. Overall, the aerosols released by the volcanic eruption increased cloud cover by around 10 percent.

«In addition to being air pollutants, aerosols encourage cloud formation and thus influence the climate»      Yu Wang

Wang became interested in particulates long before she became a climate researcher. “I was born near Beijing, where the air is very polluted,” she says. “I wanted to know why the air quality in my hometown was so much worse than in Europe or the United States.” She studied environmental sciences in Changchun and Beijing and decided to use her Master’s thesis to find out why the pollutant concentration responsible for Beijing’s air pollution is so high. “During my field observations, I noticed that the situation in the real atmosphere was so complex that gaining a better understanding would mean working in the lab,” Wang says.

From China to the United Kingdom

For her doctoral studies, Wang was accepted at the University of Manchester; she moved from China to the UK in 2017. “A massive step,” she notes with a sigh, before beaming again and adding: “I’m always excited to discover new things.” In Manchester she worked with an experiment chamber, into which she pumped gas to observe the formation of aerosols. “It was then that I realised that, in addition to being air pollutants, aerosols encourage cloud formation and thus influence the climate,” she says. “That was the moment I started doing climate research.”

Wang points out that her recently published study on interactions between aerosols and clouds was a departure from her previous work because it was based on machine learning methods rather than on climate models. As input, the research team used satellite observations of cloud cover. They fed the machine with data collected by instruments on board two NASA satellites over a period of more than 20 years. NASA handled both data processing and analysis. “To use machine learning, we require a massive dataset,” Wang says. “The observations made between 2000 and 2020 make us very confident that our method works.”

The team’s next step will be to try to channel their new findings into existing climate models. “We want to encourage the entire research community to adapt their models to accommodate our observations,” Wang says. She hopes that this will yield better climate models capable of providing more reliable forecasts.

But Wang admits that this was just a pilot study and that a single volcanic eruption is not an adequate foundation. Therefore, the researchers are also working on other events that triggered an increase or decrease in aerosol emissions, such as observations made before and during the coronavirus pandemic. “We hope our efforts will provide more evidence in near future and make the findings more precise,” Wang says.

Mentioning the coronavirus has a sobering effect on Wang. Before the pandemic, her parents and friends could visit her in the UK, and she would travel to China during the holidays. “It’s now been three years since we’ve seen each other,” she says. “I find that tough.” She is happy to plan the trip to meet them again soon now that China has eased the restrictions. But for the time being, she and her husband – a climate researcher at the Paul Scherrer Institute – feel at home in Europe.

Drawing inspiration on the move

To get new inspiration, Wang likes to go hiking or take trips with her husband. It was on a trip to the seaside of Teignmouth near Exeter that they came up with the idea of using machine learning as part of this exciting climate research.

As a cloud specialist, people often ask Wang if it would be possible to slow global warming by artificially creating clouds. “This falls under geoengineering,” she says, and names two proposals currently being discussed: the first is to inject aerosols into the stratosphere; the second involves pumping sea salt particles into clouds over the oceans. “But these would be more like giving the world a painkiller rather than an actual cure.” What’s more, the Earth is such a complex system that these interventions could prove very dangerous. “That’s why all geoengineering projects have been shelved,” she says.

But Wang remains optimistic and believes that almost every situation has its silver lining – even the extreme floods, droughts and heatwaves that are becoming more and more frequent. “Even global warming sceptics are now starting to see how important this issue is,” she says, adding that her motto is “we research, we learn and we adapt”.


Chen Y, Haywood J, Wang Y, Malavelle F, Jordan G, Patridge D, Fieldsend J, De Leeuw J, Schmidt A, Cho N, Oreopoulos L, Platnick S, Grosvenor D, Field P, Lohmann U., Machine learning reveals forcing from aerosols is dominated by increased cloud cover. Nature Geoscience 15, 609–614 (2022), doi: 10.1038/s41561-022-00991-6