Our approach for OCO-2 is simulation-based uncertainty quantification. This means that we create ensembles of synthetic true state vectors, and use them drive a Monte-Carlo experiment that provides a corresponding ensemble of retrieved state vector estimates. Elements of these two ensembles are paired, resulting in a joint sample that we use to model the relationship between the true state and its retrieved counterpart. We fit a statistical model to the joint sample, and derive regression-style equations to predict the distribution of the true state given the retrieved estimate. Then, when we receive an actual OCO-2 retrieval from the mission's data processing chain, we plug it in to the predictor to obtain the distribution of the true state as a function of the estimated state.
|Comparison among simulation-based pdf (red), retrieved pdf (blue), and true CO\(_2\) concentration (green) for one OCO-2 footprint at Lamont, OK.||Comparison among simulation-based pdf (red), retrieved pdf (blue), and true CO\(_2\) concentration (green) for one OCO-2 footprint at Tsukba, Japan.|
Braverman, A., Hobbs, J., Teixeira, J., and Gunson, M. (2021). “Post hoc Uncertainty Quantification for Remote Sensing Observing Systems,” SIAM/ASA Jour. Uncert. Quantification. Accepted. Download (CL 21-0606).
Hobbs, J., Braverman, A., Cressie, N., Granat, R., and Gunson, M. (2017). “Simulation-Based Uncertainty Quantification for Estimating Atmospheric CO2 from Satellite Data,” SIAM/ASA Journal on Uncertainty Quantification, 5(1), pp. 956-985. Download.