GESTAR II Seminar Series, Tuesday, February 25th, 1:00pm

Join us for a hybrid seminar by Mr. Sean Foley (616/MSU), Data Scientist (GESTAR II/Morgan State University). His talk is titled "Retrieving 3D Cloud Structure from Multi-angle Observations with Neural Rendering."

Date and Time: Tuesday, February 25, 2025 at 1:00pm EST

Join us via Teams or join us in person at NASA GSFC, Bldg. 33, Rm. E108.

"Multi-angle sensors capture unique information useful for studying clouds, whose 3D radiative effects have traditionally been difficult to characterize. An increasingly popular approach to 3D cloud retrievals is supervised learning, but this requires an external source of data, e.g., radar or lidar. Unsupervised approaches include stereo / disparity-based methods, but these assume all objects are 100% opaque, which is untrue of clouds. An emerging unsupervised approach to 3D cloud reconstruction is neural rendering, in which a 3D scene is represented by a small neural network, and light transport (forward radiative transfer) is modeled with a simple, differentiable renderer. The differentiability of the renderer allows the neural network to be directly optimized to accurately reproduce the observed imagery. The trained neural network can be queried at a dense set of 3D locations to extract the distribution of the attenuating mass in the scene, which maps directly to the extinction parameter. While neural rendering of clouds is an emerging area of interest, applications are currently limited to simulated data. We have applied neural rendering to real data from the Hyper Angular Rainbow Polarimeter (HARP2) aboard PACE. I will summarize some related work, describe our methods, showcase some results, and comment on future challenges in this emerging research area."

Sean is a data scientist in the Ocean Ecology Lab (OEL) at GSFC who applies machine learning techniques to remote sensing, with a particular interest in clouds and multi-angle sensors. He received his master’s degree at Georgia Tech in Computer Science with a focus on Machine Learning. Before pivoting to remote sensing, his research focused on object recognition in the autonomous vehicle industry. His lifelong fascination with space and the Earth system motivated him to find an overlap between his background and the contemporary needs at NASA. This led to an internship at OEL, which turned into a second internship, and then a full-time job. At OEL, Sean provides a unique skillset in support of the Plankton, Aerosol, Cloud-ocean Ecosystem (PACE) mission, as well as contributing to the Atmosphere Observing System (AOS) mission. Outside of work, Sean plays D&D and board games with his friends, reads sci-fi, and plays piano."

For more information on the GESTAR II Seminar Series, click here.