Our purpose in this study is to estimate surface CO2 fluxes as well as atmospheric CO2 concentrations with data assimilation. We applied Local Ensemble Transform Kalman Filter in OSSEs using the SPEEDY-C model which we modified from the original SPEEDY (Molteni, 2003) by adding a carbon tracer. We first assumed the atmospheric CO2 concentration is observed every four grid points in the horizontal and only two levels in the vertical which are the surface layer (σ=0.95) and the level of σ=0.34, while the observations of atmospheric variables are located at rawinsonde locations. Observation are obtained every six hours.

Under the perfect model assumption, we have tried three types of data assimilation: One is the uncoupled data assimilation in which atmospheric CO2 concentration and surface CO2 flux are updated by CO2 observations and not affected by other atmospheric variables. Another is the one-way multivariate data assimilation in which the atmospheric CO2 concentration and surface CO2 flux are updated by these two variables as well as the wind fields, while the wind field, in addition to other atmospheric variables such as specific humidity and temperature, is not affected by these two carbon-related variables. The other is the multivariate data assimilation so that all the dynamical variables are included in one vector. With a constant forcing of surface CO2 flux, both multivariate systems performed well. The one-way multivariate assimilation resulted in the optimal performance for the carbon fluxes, but the fully multivariate system performed almost as well near the surface and had better estimations of u, v, T and q. By contrast, the univariate assimilation of carbon had larger errors and failed after two weeks.

We plan to apply the same methods for the OSSEs with the time-varying flux of surface CO2 which is obtained by the coupled system of atmosphere-vegetation. In addition, the experiments having only daily observation of CO2 concentration and time varyingwill be also tested and reported at the conference.