3D volumetric tomography of clouds using machine learning for climate analysis.

The prediction of climate has been a long-standing problem in contemporary science. One of the reasons stems from a gap in the ability to obtain 3D mapping of clouds, especially shallow scattered clouds. These clouds are strongly affected by mixing processes with their surroundings, rendering their internal volumetric structure highly heterogeneous. These heterogeneous clouds modulate the incoming solar energy and the outgoing long-wave radiation, thereby having a crucial role in the climate system. However, their 3D internal mapping is a major challenge. Here, we combine machine learning and space engineering to enable, for the first time, 3D mapping of scatterers in clouds. We employ ten nano-satellites in formation to simultaneously view the same clouds per scene from different angles and recover the 3D internal structure of shallow scattered clouds, from which we derive statistics, including uncertainty. We demonstrate this on real-world data. The results provide key features for predicting precipitation and renewable energy.