Home / Projects / Deepcloud

Deepcloud

NOMIS Project 2025

— 2030

The Question

Clouds play a critical role in climate change. By reflecting sunlight back into space, they cool the planet, while simultaneously trapping infrared radiation in the atmosphere, contributing to warming. These opposing effects shift as the climate warms, making clouds a key factor in determining how much the planet will ultimately respond to increasing greenhouse gas concentrations.

Yet, despite decades of intensive research, quantifying how clouds respond to global warming remains a challenge. This lack of understanding of clouds is the greatest unknown in climate projections. The difficulty of representing clouds in atmospheric models leads to significant uncertainties in weather forecasts. In particular, accurately forecasting extreme precipitation events associated with deep convective clouds remains a major hurdle.

At the heart of this challenge is the complexity of cloud formation, which depends on both microscopic processes, like droplet and ice crystal formation, and large-scale atmospheric processes, such as low- and high-pressure systems spanning hundreds to thousands of kilometers. Accurately representing the physical laws that govern all these processes is currently infeasible, even for the most advanced supercomputers.

“With Deepcloud, we are breaking one of climate science’s biggest barriers: understanding and representing cloud processes in a warming world. By uniting advanced AI, open data and community-driven research, we’re transforming how clouds are modeled — and with them, the future of climate prediction.”

— Markus Rex

The Approach

The Deepcloud project aims to overcome this challenge by using machine learning to model cloud processes. Recent progress in artificial intelligence approaches, combined with the recent quantum leap in the availability of detailed cloud data, open up new possibilities for understanding and simulating clouds.

Rather than explicitly describing all physical processes within clouds, the machine-learning model will learn the most efficient representation of how small-scale processes influence cloud behavior. To train the model, the project will leverage unique observational data from two key sources: a recent year-long Arctic expedition and a cloud-observing satellite launched in 2024. This model of cloud processes will be used to explore how Arctic clouds respond to global warming and how increases in small particles in the atmosphere influence cloud properties. The researchers hope to uncover the extent to which cloud ice is replaced by cloud liquid in a warmer climate and how changes in cloud condensation nuclei and ice nucleating particles affect cloud processes and cloud properties. These insights will advance our understanding of cloud-related climate feedback at high latitudes, transforming the future of climate prediction and policy, ultimately safeguarding the planet for future generations.

Deepcloud: Tearing Down a Long-Standing Barrier in Climate Research by Pioneering a New AI-Based Model Approach is being led by Markus Rex at the Alfred Wegener Institute, Helmholtz Center for Polar and Marine Research (AWI) in Potsdam, Germany.

Feature image: Convective clouds in the clean air of the tropical West Pacific. (Photo: AWI)

Loading...

NOMIS Researcher(s)

Head of Atmospheric Physics and full professor
Alfred Wegener Institute
Loading...

Project News

Loading...