We show results of the background error covariance matrix representation of the sensitivity of a given pressure level CO concentration to the CO surface fluxes using an EDA of 10 perturbed members (SPPT and fluxes).

Open questions:

• How many members the EDA require to provide robust emission estimates, global and sectoral?
• Is 12 hour assimilation window reasonable? Do we need to reduce to 6 hours to account correctly for the diurnal cycle?
• Using B climatological enough (fires?), can we afford 30+ members EDA? Can we use ensembles for previous days (24,48,72 hours)?