HR: 1340h
AN: A13D-1506    [Abstracts]
TI: Optimization of GOSAT Atmospheric Retrieval of CO₂ in Presence of Atmospheric Particles Using Empirical Orthogonal Function Representation
AU: * Desbiens, R
EM: raphael.desbiens@nies.go.jp
AF: National Institute for Environmental Studies, 16-2 Onogawa, Tsukuba, Ibaraki, 305-8506, Japan
AU: Aoki, T
EM: aoki.tadao@nies.go.jp
AF: National Institute for Environmental Studies, 16-2 Onogawa, Tsukuba, Ibaraki, 305-8506, Japan
AU: Yokota, T
EM: yoko@nies.go.jp
AF: National Institute for Environmental Studies, 16-2 Onogawa, Tsukuba, Ibaraki, 305-8506, Japan
AB: As part of the international effort to promote greenhouse-gases observation and to understand furthermore the carbon cycle, the Ministry of Environment of Japan (MOE), the National Institute for Environmental Studies (NIES), and the Japan Aerospace Exploration Agency (JAXA) plan to launch GOSAT (Greenhouse gases Observing SATellite) in 2008 for the global monitoring of CO₂ and CH₄ from space. The inverse problem that must be solved for GOSAT level 2 retrieval algorithms is very complex, involving thousands of measurement parameters (wavenumber channels) for 3 of the spectral bands of the TANSO Fourier transform spectrometer. Potentially more than one hundred state parameters would need to be retrieved, depending on the vertical resolution of the volume mixing ratio (VMR) profiles required for the retrieval of each species, to which we must add aerosols and cloud parameters. The most computationally expensive part of current signal processing comes from the computation of the forward model itself, based on HSTAR radiative transfer code, and also from the calculation of the Jacobian (derivative of the forward model relative to state parameters). The Jacobian is calculated using numerical derivatives, which implies computing perturbed states of the forward model. Furthermore, iterative algorithms selected to solve non-linear inverse problems require computing a new Jacobian at each iteration. We explored the benefit of using an optimal representation for the state parameters based on empirical orthogonal functions (EOF). This compact representation allows reducing the number of state parameters near to the level of the number of degrees of freedom in the signal, which is significantly smaller than the actual number of state parameters required for the accurate computation of the forward model in nadir observation mode. We present EOF based on the singular vectors of the normalized Jacobian, taking into account the prior covariance matrix of the state parameters. Such optimal representation proved to be efficient also for non-linear retrieval involving light scattering by atmospheric particles like thin clouds and aerosols. Using EOF derived from clear sky conditions, we developed a simple way to approximate efficiently the Jacobian when considering atmospheric particles, without need to change the EOF basis when particles parameters are updated. This optimization requires evaluating the forward model a number of times that is only the number of optimal parameters (number of EOF) instead of computing it a number of times equal to the number of parameters in full representation. For example, CO₂ retrieval in 1.6 μ m band must be performed for more than 25 atmospheric layers, while the number of degrees of freedom of the signal is less than 3 for the vertical profile. When particles parameters are retrieved simultaneously (leading to a non-linear inverse problem), this advantage may becomes even more significant since iterative methods require to recalculate the Jacobian at each iteration. This aspect will be discussed at the meeting.
DE: 0322 Constituent sources and sinks
DE: 0394 Instruments and techniques
DE: 1640 Remote sensing (1855)
SC: Atmospheric Sciences [A]
MN: 2007 Fall Meeting