The feasibility of retrieval of CO2 vertical profile and its column-averaged value from the reflected sunlight, measured by sensor TANSO-FTS onboard GOSAT is demonstrated. The model spectra in 0.76-µm O2 A-band and in CO2 bands at 1.6 and 2.06 µm were used to train neural networks for CO2 retrieval. A separate neural network was developed for each of 4 scanning angles and for continuous range of solar zenith angles. Mean retrieval errors better than 1 ppm for column-averaged value and better than 4 ppm for CO2 concentration at surface were obtained for 1:300 level of noise for all bands involved into the retrieval procedure.
GOSAT; vertical profile; neuron nets