Introduction¶
This tool is developed to estimate fluxes of greenhouse gases (GHG) into the atmosphere based on observations of the atmospheric composition. A priori estimates of GHG fluxes into the atmosphere and atmospheric transport models can be used to predict the concentration at the location of observations. Using the difference of model prediction and observations as a signal, one can compute a correction to the a priori fluxes.
Different methods of inverse modeling can be used to estimate fluxes. Here, we focus on the synthesis inversion or scaling inversion method [1]. In this method, the considered fluxes are grouped into flux categories and the transport of each category is computed independently. The inversion assigns a scaling factor to each flux category and adjusts these scaling factors to optimize the agreement of the model prediction with the observations. DUBFI uses the model prediction for each flux category and the observations as input and computes the scaling factors.
A central part of the inversion problem are the uncertainties in the model-observation comparison. Atmospheric transport models have uncertainties and model errors that need to be considered to obtain realistic flux estimates [2][3]. For continuous observations or spatially close observation sites, the model uncertainties are correlated. DUBFI’s complexity arises from the problem of an accurate estimation and utilization of uncertainties in the model-observation mismatch. These uncertainties and their correlations are estimated using a transport ensemble, as it is typically done in numerical weather prediction. This approach is similar to the works by Steiner et al.[4] (using CTDAS [5]) and Ghosh et al.[6].
DUBFI uses a transport ensemble to provide dynamic uncertainty estimates in the inversion. By this we mean that the uncertainties depend on the fluxes and are adapted to the posterior fluxes in the inversion. In the section on Bayesian Inversion Problem, we derive this self-consistent uncertainty estimate from the assumption of a Gaussian transport model error. This generalizes a method presented by Bruch et al.[7].
This system is designed for estimating emissions of long-lived GHG like CH4, N2O or CO2. Currently, the implementation is limited to observation sites with fixed spatial coordinates, but an extension to other observation types – such as satellite observations – is possible. Another possible extension may address the option to use a flux ensembles similar to Steiner et al.[8].
DUBFI is developed with funding by the German Federal Ministry for Research, Technology and Space (BMFTR) in the ITMS project (grant 01LK2102B), see https://www.itms-germany.de/en.