Amortized Probabilistic Retrieval of Atmospheric CO2 from OCO-2 Spectra Using Deep Learning with Laplace Approximations and Normalizing Flows
Researchers have developed a novel deep learning framework to more efficiently and accurately retrieve atmospheric carbon dioxide (CO2) data from NASA's Orbiting Carbon Observatory-2 (OCO-2) satellite. This new method utilizes Laplace approximations and normalizing flows to achieve inference speeds orders of magnitude faster than current operational algorithms, while also providing more robust uncertainty quantification. The framework is trained on high-fidelity simulations that account for realistic forward model errors, enabling it to handle systematic errors often overlooked by standard inversion techniques and model non-Gaussian posterior distributions. AI
IMPACT Accelerates real-time processing of satellite data for climate monitoring and carbon budget analysis.