The whole understanding of the behaviour of carbon dioxide (CO2) sources and sinks at the surface, needed to accurately simulate future evolution of CO2 concentration in the atmosphere, is presently very limited by the inadequate spatio-temporal sample of the current surface network. Observation from space may allow to overcome this difficulty. The ability of retrieving CO2 from vertical sounders has been proven with TOVS instruments (18 infared channels with a spectral resolution of 120, and 4 microwave channels) [Chedin et al. 2002
In this thesis, we present the retrieval of atmospheric CO2 distribution from space observations made by the infrared AIRS (2378 channels) and the microwave AMSU (15 channels) instruments, flying onboard the Aqua satellite launched by NASA in May 2002. This work represents one of the first use of these instruments of second generation. Thanks to the very high spectral resolution of AIRS, a set of channels presenting optimal properties to retrieve different trace gases has been selected using the OSP method introduced in [Crevoisier et al. 2003]. Cloud detection tests have been designed to work under clear-sky conditions.
The availability of simultaneous observations of the atmosphere by infrared and microwave sounders is an advantage. Indeed, although sensitive to CO2 concentration variations, the infrared observations are mainly sensitive to atmospheric temperature variations. Hence, the simultaneous use of microwave observations made in the oxygen absorption bands, which are only sensitive to temperature, and of infrared observations allows separating CO2 from temperature variations.
It must be noted that infrared channels are only sensitive to CO2 variations in the mid to high troposphere (roughly above 5km/600 hPa) [Crevoisier et al. 2003]. This is well seen on the CO2 Jacobians plotted on figure 2. Therefore, to be able to link the retrieved CO2 concentration to surface fluxes, a good modelisation of atmospheric transport is needed.
Using a non linear inference scheme based on neural networks, seven months of observations (April-October 2003) have been inverted in terms of CO2 mixing ratio in the troposphere (5-15 km), in the tropical region [20°S ; 20°N], nightime, with a resolution of 15°×15° [Crevoisier et al. 2004]. The estimated precision is 2.5 ppmv (0.7 %).
In situ aircraft observations, made at an altitude close to the one seen by AIRS and AMSU channels, as well as surface observations have shown very good agreement with our retrievals in terms of seasonal and latitudinal variations. The retrieved spatial variations have highlighted strong latitudinal and longitudinal CO2 variations. Some of them have been related either to known behaviour of the tropospheric CO2 or to transport phenomena (biomass burning).