HR: 0800h
AN: A21D-0912    [Abstracts]
TI: A Fast Radiative Transfer Model for Hyperspectral Remote Sensing
AU: * Timucin, D A
EM: timucin@email.arc.nasa.gov
AF: NASA Ames Research Center, Mail Stop 269-3, Moffett Field, CA 94035 United States
AB: The spectral profile of the Earth's atmosphere is a combination of smooth features due to aerosol and cloud particles and of rough variations due to gas molecules. The solar spectral flux incident at the top of the atmosphere is of a similar nature, comprising fine molecular emission lines superimposed on a smooth blackbody spectrum. (The other ``boundary condition,'' provided by the reflectance of the Earth's solid surface, has only a smooth spectral variation.) For passive remote sensing of the Earth's atmosphere and/or surface in the solar region of the spectrum, the main consequence of these characteristics is that a high-fidelity radiative transfer model must employ a very dense wavelength grid for its numerical solution. In attempting to retrieve unknown geophysical parameters from a data set, such a model calculation constitues the first step. The simulation of the sensor output is then completed by convolving the result of the radiative transfer calculation with the spectral response function of the electro-optic instrument that collected the data. This second step leads to a ``coarse-graining'' of the radiation spectrum incident on the sensor, and effectively throws away much of the spectral detail that was painstakingly computed in the first step. Finally in the third step, one has to minimize the disagreement between the observed and the simulated data sets by searching over the space of unknown parameters -- a process that entails many iterations of the ``forward model'' comprising the first two steps. Moreover, in the case of an hyperspectral data set, this procedure has to be carried out, rather independently, for each spectral band and for each spatial pixel of the sensor. Two techniques are well established in the community for performing forward-model calculations -- the line-by-line (LBL) and the correlated-k distribution (CKD) methods -- whose pros and cons are widely known. In this paper, I present a new approach that integrates the two distinct steps of the forward-model calculation into a single scheme. This is achieved by Fourier-transforming the radiative transfer equation and its boundary conditions, and subsequently applying the instrument spectral response. This results in a set of model equations that is specialized to the particular sensor being used, and whose solution directly yields the sensor output. The form of the resulting set is identical to that of the traditional one, and therefore well-known numerical techniques such as discrete ordinates may be employed here as well. Furthermore, by way of a specific example, I demonstrate that the density of the new numerical grid is roughly two orders of magnitude less than that of the original wavelength grid. Thus, a significant speed-up in forward-model calculation is seen to be possible by incorporating into the formulation of the radiative transfer problem one's own sensor characteristics.
DE: 0300 ATMOSPHERIC COMPOSITION AND STRUCTURE
DE: 0360 Radiation: transmission and scattering
DE: 0394 Instruments and techniques
DE: 3359 Radiative processes
DE: 3360 Remote sensing
SC: Atmospheric Sciences [A]
MN: Fall Meeting 2005