HR: 1340h
AN: A13A-0097    [Abstracts]
TI: Grid-Scale CO$_{2}$ Flux Error Estimates from a Variational Data Assimilation Approach Processing Satellite-Based CO$_{2}$ Measurements
AU: * Baker, D F
EM: dfb@ucar.edu
AF: National Center for Atmospheric Research, 1850 Table Mesa Dr., Boulder, CO 80305 United States
AB: Inversion of atmospheric CO$_{2}$ concentration measurements, using atmospheric transport models, to solve for surface CO$_{2}$ fluxes has proven to be a useful `top-down' check of the carbon fluxes produced by `bottom-up' process models -- the comparison of flux integrals at fairly coarse time and space scales can reveal model mis-tunings, and can even suggest processes missing from the models. For this comparison to be most useful, it should be performed at the finest time/space scales possible (not only to help localize possible model deficiencies, but also to best represent the available fine-scale data, particularly over the continents). The time/space density of carbon observations currently limits the effective resolution at which this comparison may be performed. Should the atmospheric CO$_{2}$ measurements from two satellites scheduled for launch in 2007 prove useful, this resolution will be greatly increased, but then computational limits may become important. To help evaluate the usefulness of the up-coming satellite observations, we have built a variational data assimilation system (4-D Var) that solves for the surface CO$_{2}$ fluxes at resolutions as fine as the horizontal resolution of the underlying global transport model (2.0 x 2.5 degrees) and the model time step (\~15 minutes), using measurements localized up to an equally fine scale. The performance of the method has been assessed using simulated observations (i.e., with an observing system simulation experiment, or `OSSE'). As the size of the flux vector to be estimated is too great to permit a direct estimate, an iterative method is used (in the form of an unconstrained optimization problem). The final estimation accuracy is limited then not only by the number and accuracy of the observations, but also by the number of iterations computationally feasible. With this OSSE set-up, we compute the flux estimation error obtained for measurement sets similar to those planned for the upcoming satellite missions, as well as those from the current flask network augmented with planned flux towers. This error is given as a function of the number of optimization iterations, measurement span considered, and measurement correlation time/length scale assumed. The estimation error obtained by comparing the true and estimated fluxes for the single OSSE case (averaged over time/area to obtain meaningful statistics) is compared to the corresponding error from the leading terms in the covariance matrix built up by the descent method. For those cases in which the estimation errors for the fluxes at the grid scale are too high to be of use, we will discuss at which (coarser) scales the estimates may be believed.
DE: 1615 Biogeochemical processes (4805)
DE: 0312 Air/sea constituent fluxes (3339, 4504)
DE: 0322 Constituent sources and sinks
DE: 0368 Troposphere--constituent transport and chemistry
DE: 0400 Biogeosciences
SC: Atmospheric Sciences [A]
MN: 2004 AGU Fall Meeting