Clouds play a critical role in Earth's climate system and are a major source of uncertainty in climate projections. The vertical distribution of ice and water particles in clouds impacts their radiative properties and with that Earth's energy balance. Recent research also showed that the internal properties of clouds in tropical cyclones influence how storms intensify. Yet satellites face a fundamental trade-off: systems that measure vertical structure lack continuous coverage, while those with continuous coverage cannot see inside clouds.
Now, research conducted through the Earth Systems Lab research programme, funded through the ESA Φ-lab and involving former ESA research fellow, Dr Anna Jungbluth, has developed a breakthrough machine learning framework that translates two-dimensional geostationary satellite imagery into detailed three-dimensional cloud maps in near real-time. Published in November 2025, the study demonstrates for the first time the ability to create global instantaneous 3D cloud reconstructions, with particular success in mapping the internal structure of intense tropical cyclones.
The missing dimension in cloud observation
The vertical structure of clouds profoundly influences Earth's energy balance: Clouds at different altitudes either trap outgoing infrared radiation, warming the atmosphere, or reflect incoming solar radiation, thereby cooling it. Disentangling these competing effects requires detailed knowledge of cloud structure. Satellites can observe the vertical structure of clouds, but these observations are limited in space and time. Ideally, we would need global and continuous monitoring of clouds in 3D.
Vertical cloud measurements that scientists need can be obtained from NASA's CloudSat mission. However, it observes each location only once approximately every 16 days. ESA's EarthCare mission, which has been in orbit for less than two years, despite carrying advanced radar and lidar instruments for atmospheric profiling, also has a long revisit period, exceeding 20 days. Geostationary satellites such as EUMETSAT’s MSG, Japan's Himawari-8, and the American GOES-16 offer the opposite trade-off: continuous monitoring every 10 to 15 minutes, but only of cloud-top properties with no vertical structure information.
"If clouds are heating or cooling the atmosphere has profound implications for our climate system," says Anna Jungbluth, former internal research fellow and part of ESA’s climate team and a researcher on the project. "This is particularly critical for tropical cyclone forecasting, where understanding the vertical cloud structure can mean the difference between accurate early warnings and catastrophic surprise. But no current satellite can measure this continuously in real time, which is exactly the gap this research addresses."
A new window into atmospheric dynamics
The new machine learning model resolves this dilemma. Trained on paired observations from geostationary satellites and CloudSat, it infers three-dimensional cloud properties from two-dimensional geostationary imagery. Building on previous work initiated in 2024, the expanded framework combines data from three geostationary satellites (MSG, Himawari-8, and GOES-16), achieving unprecedented diversity in viewing angles, cloud types, and geographic coverage.
The model predicts three key cloud properties throughout the atmospheric column, which provide insight into the physics of clouds. A case study on Hurricane Dorian, a category 5 storm that devastated the Bahamas in 2019, demonstrated the model's capability to reconstruct the three-dimensional structure of intense storms.
"We combined the continuous monitoring from geostationary satellites with vertical profiling capabilities and used machine learning to understand how these measurements relate to each other," says Jungbluth. "This allows us to predict cloud structure where we have no direct measurements and create 3D cloud maps that show what a profiling satellite would have observed – if one had been there at that exact time and location.” The researchers’ hope is that this work will support climate model develop, in turn aid the forecasting of tropical cyclones that help save lives, protect infrastructure and reduce economic losses.
Enhancing climate data and models
The research demonstrates how combining observations from multiple satellites can address critical gaps in climate monitoring, an approach central to ESA's Climate Change Initiative (CCI). Since 2008, CCI has worked with researchers across ESA Member States to generate long-term, global satellite datasets for Essential Climate Variables - the key components of the climate system identified by the Global Climate Observing System as critical for understanding climate change.
Cloud properties represent one of these Essential Climate Variables. Looking forward, this work and that of the CCI could benefit from synergies in both directions: long-term CCI climate records could serve as additional model input and validation data, whilst the 3D maps generated by the team could help improve CCI data and climate model accuracy.
The study, led by Shirin Ermis from the University of Oxford, involved contributions from ESA, the National Observatory of Athens, Environment and Climate Change Canada, Universitat de València, and Harvard Medical School.
This work was presented at the Conference on Neural Information Processing Systems (NeurIPS 2025) – one of the world's leading interdisciplinary conferences on machine learning and artificial intelligence – where it won the “Best ML Innovation” award at the “Tackling Climate Change with AI” workshop.
