2.10. THREE-DIMENSIONAL INVERSE MODELING
In order to reconcile the contemporary budget of CO, and to understand by which processes this gas is absorbed by the ocean and/or by the terrestrial biosphere, it is of first importance to identify which regions are gaining or losing CO. In past years, several inverse calculations were performed to estimate the north/south gradient of net sources from the observed zonal mean concentrations [Tans et al., 1989; Enting et al., 1991; Ciais et al., 1995a,b; Bousquet et al., 1996; Law et al., 1996]. The observational network has grown significantly in the last decade, and the present framework is designed to go beyond the interpretation of the north/south gradient. In the present work, the goal is to study the feasibility of a 3-D inversion to assess the net fluxes over large regional scales. For this, about 25 continental and oceanic regions were defined. For each of them a forward simulation was made using a normalized source and the TM2 model [Heimann, 1995]. The linear combination of separate sources providing the best fit to the observations is calculated using the singular value decomposition technique.
Before applying this method to real data, modeled data were defined from a forward simulation using a global flux distribution as realistic as possible [Ramonet, 1994]. Different subsets of these modeled data were used to recalculate the initial flux distribution over each postulated source. This methodology allows study of the influence of the surface and/or tropospheric network resolution and the contribution of new monitoring sites. The first results showed that the annual fluxes were resolved at ±0.2 GtC yr when a network of as few as 144 regularly spaced surface sites was used. The errors of the deduced fluxes strongly increase when the number of regularly spaced surface sites is reduced to 50. Using the same locations as the CMDL network, there are additional problems due to a lack of an observational site over large continental areas, especially over South America. Preliminary results show that the error for this region can be largely reduced by adding a surface site in Central Brazil.
In the work described previously, the implicit assumption of perfectly simulated
transport is made. It is known that simulation of the transport is one of the
sources of uncertainty in inverse calculations [Rayner and Law, 1995].
To take into account the error induced by the transport models, the NCAR Community
Climate Model (CCM2) will be used to calculate the modeled data and the TM2
for the normalized sources. There will then be an estimate of how sensitive
the results are to the transport calculations of different models. In the near
future plans are to use the additional information provided by isotopic ratios
and to apply the inverse method to the measurements smoothed in time.