Simulation-Based Uncertainty Quantification for Level 2 Retrievals: Tools and Applications
Presenting Author: Jonathan Hobbs
Organization: Jet Propulsion Laboratory
Co-Author(s): Ali Behrangi, Amy Braverman, Eric Fetzer, Kelsey Foster, Kyo Lee, Hai Nguyen, Joaquim Teixeira
Complex retrieval algorithms for converting Level 1 radiances into Level 2 estimates of geophysical quantities are ubiquitous in NASA's Earth-observing constellation. Each retrieval strategy presents unique challenges and requirements, but a common approach involves a physical and/or statistical model with a computational inverse method. Variability in atmospheric and surface processes, as well as in the satellite measurements themselves, contribute to the statistical properties of retrieval algorithms. Specific algorithm implementation choices can also impact the algorithm's accuracy and precision. This probabilistic and computational pipeline can be studied in depth with Monte Carlo simulation experiments of the satellite observing system using ensembles of realistic true states. This work develops methodology, software, and practices that facilitate design and post-processing of these simulation experiments. Retrieval algorithm error distributions are illustrated through several algorithm figures of merit. We illustrate the implementation for multiple remote sensing retrievals, including the Orbiting Carbon Observatory-2 (OCO-2) and Atmospheric Infrared Sounder (AIRS). The experiments can provide retrieval uncertainty estimates for use in applications using Level 2 products.