WOMBAT: A fully Bayesian global flux-inversion framework

Abstract:

WOMBAT, the WOllongong Methodology for Bayesian Assimilation of Trace-gases, is a framework currently being developed for flux inversion of trace gases from remote sensing data. It is built around a fully Bayesian hierarchical statistical model that accounts for instrument bias, correlated measurements, and transport-model errors. Fluxes are modeled using a classical basis-function decomposition, and the framework uses source-receptor relationships that are obtained empirically through forward simulation of an atmospheric transport model. For computational efficiency, WOMBAT only takes into account a few months of source-receptor relationships, and it treats the background field separately. It uses an efficient and fully Bayesian Markov chain Monte Carlo scheme to obtain posterior estimates and uncertainties (including joint uncertainties) for all quantities of interest. Using GEOS-Chem and XCO₂ values derived from the OCO-2 Version 7 retrospective (V7r) dataset, we show the utility of WOMBAT in inferring non-fossil-fuel CO₂ fluxes by comparing our results to those in the Model Intercomparison Project (MIP) reported in Crowell et al. (2019, Atmos. Chem. Phys., vol. 19). We find a substantial correlation structure spanning thousands of kilometers in the posterior residuals and, by accounting for this, we obtain posterior means and uncertainties on the CO₂ fluxes that are comparable to the MIP ensemble means and spreads, respectively.