Using machine learning to tackle climate change

Philip Stier

Philip Stier, Professor of Atmospheric Physics

Official Fellow (AI & Machine Learning)

Blog Series: #reubenpeople

As a climate scientist, my work involves dealing with big (huge!) data sets and complex models. In the Climate Processes Group, which I lead in the Department of Physics, we’ve been studying the effects of aerosol pollution on clouds. We know that atmospheric aerosols play an important role in the global climate system, and even small changes in clouds in response to global warming or air pollution can have a varying impact on global climate, serving to accelerate or dampen the greenhouse gas effect.

However, owing to the complexity of clouds and aerosols in the climate system – spanning scales from nanometers to thousands of kilometres – even approaches that combine theory, modelling and observation cannot fully explain or address the uncertainties in their interactions. Increasingly, we are employing machine learning techniques to analyse and emulate complex big climate datasets and shed light on the underlying causes of climate change and air pollution. However, one of the challenges we frequently encounter is the gap between access to technology and the skills required to harness it effectively.

Climate and AI

We were recently awarded an Amazon Research Award, which helps academics advance the frontiers of machine learning to tackle some of the world’s biggest problems. This has contributed to iMIRACLI (innovative MachIne leaRning to constrain Aerosol-cloud CLimate Impacts), an EU-funded Innovative Training Network that brings together leading climate and machine learning experts across Europe with non-academic partners, such as Amazon and the MetOffice, to educate a new generation of climate data scientists.

The network funds 15 PhD students across Europe. They will develop machine learning solutions to deliver a breakthrough in climate research, by tracing and quantifying the impact of aerosol-cloud interactions from the microscale to large-scale climate. Each student will have an interdisciplinary supervisory team, combining academic climate and machine learning supervisors as well as a non-academic advisor. International secondments to co-supervisors as well as to the non-academic partners will enrich student experience and training.

In September 2020, we launched our first summer school for these students, and host the 10th international conference on Climate Informatics. The machine learning techniques we’re developing in climate science are applicable to addressing a whole host of 21st century challenges.

I am looking forward to working with the incoming cohort of Reuben College students to tackle some of these pressing science questions, in particular across the AI/ML and Environmental Change themes.


Philip Stier is Professor of Atmospheric Physics and leads the Climate Processes group in the Department of Physics