IMAGe Brown Bag- Bayes in Space! — NASA’s CO2 Measurements and Uncertainty Quantification

07/20/2016 - 12:00pm to 1:00pm

Jenny Brynjarsdottir
Case Western University

NASA's Orbiting Carbon Observatory-2 (OCO-2) mission is now actively collecting space-based measurements of atmospheric carbon dioxide (CO2). Data are collected with high spatial and temporal resolution and the data product includes both an estimate of column averaged CO2 dry air mole fraction (XCO2) and an estimate of uncertainty. In this talk we will take a look at how these estimates are obtained. As with any remote sensing method, the measurements are indirect. The OCO-2 instrument measures reflected sunlight in three spectral bands that make a single "sounding”. These are then used to estimate the XCO2 using a physical forward model for how a given CO2 concentration, and other atmospheric properties, affect a sounding. This estimation is called “retrieval” and is done by applying Bayes Theorem. Due to the amount of data collected, some computational shortcuts are taken to obtain an estimate of the posterior mode (X.hat) and posterior variance (S.hat), using the so-called “optimal estimation” method of Rodgers (2000). Even thought the forward model is not linear, users usually treat the posterior distribution as Gaussian with mean X.hat and variance S.hat. Also, uncertainty due to several uncertain parameter inputs and model discrepancy is not taken into account. A UQ group within the OCO-2 mission has developed a test-bed, where a surrogate forward model (simplified, but physically realistic) can be used to study various aspects of the retrieval. In this talk we will discuss the discuss a few test-bed experiments, such as full exploration of the posterior via Markov Chain Monte Carlo (MCMC) methods. This is joint work with Amy Braverman and Jonathan Hobbs – Jet Propulsion Laboratory, California Institute of Technology.


Wednesday, July 20, 2016
12:00 - 1:00pm
Mesa Lab, Damon Room
(Bring your lunch)