11/12/2024
57 views
2 likes
Launched in May 2024, ESA’s EarthCARE satellite is nearing the end of its commissioning phase with the release of its first data on clouds and aerosols expected early next year. In the meantime, an international team of scientists has found an innovative way of applying artificial intelligence to other satellite data to yield 3D profiles of clouds.
This is particularly news for those eagerly awaiting data from EarthCARE in their quest to advance climate science.
Clouds play a critical role in Earth’s climate system by reflecting sunlight back into space, known as the albedo effect, and by trapping heat radiating from Earth’s surface, part of the greenhouse effect.
For example, high, thin clouds tend to warm the atmosphere because a high proportion of energy from the Sun can pass through and they are also efficient at trapping heat radiating from Earth’s surface. Low, thick clouds on the other hand, tend to have a cooling effect as they reflect a high proportion of the incoming sunlight back out to space.
While scientists know that clouds play an extremely an important role in both cooling and warming our atmosphere, there remains uncertainty when it comes to accounting for the exact influence they have on Earth’s energy balance.
Moreover, given the ongoing climate crisis, there is an urgent need to understand if changes to clouds will exert an overall cooling or warming effect in the future.
Global, realtime 3D cloud data would help reduce these uncertainties, improving climate predictions and helping decision-making.
Over the last decades NASA’s CloudSat mission has provided valuable vertical cloud profiles but was limited by infrequent revisits. Geostationary missions, such as Europe’s Meteosat Second Generation (MSG), on the other hand, take images over Europe every 15 mins, but only obtain a ‘top-down’ view, without directly probing cloud profiles.
Using advanced machine learning techniques, an international team of scientists, coordinated by ESA Φ-lab and FDL Europe, has leveraged advanced machine learning techniques to develop a method for generating ‘3D cloud profiles everywhere, all at once’.
In their proof-of-concept study, they analysed a year’s worth of archived CloudSat and MSG data from 2010. The resulting paper, which was presented this week at the Neural Information Processing Systems conference in Canada, demonstrates how artificial intelligence can extract new insights from existing satellite observations.
Anna Jungbluth from ESA’s Climate and Long-Term Action Division, explained, “We carefully aligned the measured CloudSat profiles with images from MSG. This helped us understand how the ‘view from top’ and the corresponding cloud profiles were related.
“We then trained machine learning models to understand this mapping and derive cloud profiles from the 2D imagery. This allowed us to extend the CloudSat profiles in both space and time.”
The integration of cutting-edge AI techniques and Earth observation expertise exemplifies how innovative approaches can enhance the value of existing and future satellite missions.
The first animation in the body of this article shows how AI was used on an MSG image (infrared channel) with a co-aligned CloudSat track. The model learns from the limited overlap of the MSG image and CloudSat track, and is able to extend the vertical cloud profile in space.
The second animation (also featured in the top banner) shows how after the model is trained, predictions can be made for MSG images without corresponding CloudSat tracks, and 3D cloud maps can be created across space and time.
Michael Eisinger, from the EarthCARE project team and also from ESA’s Climate and Long-Term Action Division, added, “EarthCARE has already given us some very promising preliminary data and we are expecting great science from this new satellite mission. Our work generating these 3D cloud profiles lays the foundation for exploiting EarthCARE from a different angle.
“These new AI methods promise to maximise EarthCARE’s scientific potential and integrate its data into comprehensive global models that will push the boundaries of climate science.”
Stay tuned for more updates as EarthCARE data is harnessed to refine and expand this pioneering approach.
Note: This research has been enabled by FDL Europe Earth Systems Lab a public–private partnership between ESA, Trillium Technologies, the University of Oxford and leaders in commercial AI and supported by Google Cloud, Scan AI and NVIDIA Corporation.