4D-Var inversion of atmospheric methane fluxes by assimilating SCIAMACHY and AIRS satellite retrievals

The adjoint-GEOS-Chem was further developed to assimilate SCIAMACHY and AIRS satellite retrievals for the inversion of methane fluxes from various sources/sinks. The standard finite difference tests were conducted to check the validity of the observational operators and their adjoints. The results indicate that the gradient of a cost function with respect to the emissions, calculated using the adjoint-GEOS-Chem, agrees well with that derived using the first order forward/backward and the second order central difference approximations. The linear regression between the adjoint-based gradient and the finite difference based gradient for two-month simulations is adj=0.975fd (R2=0.99) for the SCIAMACHY observation operator and adj=0.981fd (R2=0.99) for the AIRS observation operator. A one-year inversion for twelve different fluxes, of which five fluxes are varying monthly, and seven fluxes are invariant throughout the year, was then conducted for year 2004 by taking pseudo observations generated from the SCIAMACHY and AIRS observation operators. The results indicate that the inclusion of off-diagonal elements in the background covariance matrix is critical to constrain the fluxes. Incorporating the interval constraint helped constrain the fluxes over their feasible ranges. Also, it is found that even using both the standard L3 AIRS product and the L2 SCIAMACHY product did not constrain well the grid-cell-based inversions at the 4° (longitude) by 5° (latitude) grid. An inversion experiment with true satellite retrievals is now undergoing to obtain spatially-explicit high-resolution methane fluxes for year 2004.