A solely radiance-based spectral anisotropic distribution model and its application in deriving clear-sky spectral fluxes

Anisotropic distribution model (ADM) plays a uniquely central role in converting broadband radiance measurement to broadband flux. Scene type classifications are usually needed for such ADM and such classifications are usually done with auxiliary measurements and information since broadband radiance does not contain detailed information about temperature, humidity, and clouds. Recently Huang et al. (2008 and 2010) has developed spectral ADM based on such scene type classifications and successfully derived spectral flux from spectral radiance measurement. Unlike broadband radiances, the spectrally resolved radiances indeed contain rich information about temperature, humidity, and clouds. Therefore, it is meaningful to explore whether it is possible to develop scene-type classification solely based on spectral radiance and consequently to construct spectral ADM solely base on radiances measurement. Using AIRS spectrum as an example, here we develop a clear-sky scene classification algorithm solely based on AIRS radiances. The definitions of scene types are similar to those of clear-sky scene types used in CERES SSF algorithm, which are discrete intervals based on surface skin temperature, lapse rate (temperature change of the first 300 mb above the surface), and the total precipitable water (TPW). Brightness temperature of AIRS channel at 963.8 cm^-1 are used for determine corresponding discrete intervals of surface skin temperature. This channel is also used in conjunction with a channel at 748.6 cm^-1 for categorizing the lapse rate. Given the slow varying of water vapor continuum in the window region and the dominant weight of lower tropospheric humidity in TPW, a double-differential technique is used to categorize the TPW. By choosing two pairs of AIRS channels with similar frequency intervals, the technique can classify the TPW without any a priori information about continuum absorption since double differencing largely remove the slow-varying continuum contribution. One month of 6-hourly ECMWF interim-reanalysis data are used to train the algorithm. Then data from different month and different years are selected for validation. It turns out the accuracy of classifying the surface skin temperature is more than 99% and that of classifying the lapse rate is over 92%. The accuracy of grouping TPW is about 80% with less satisfactory results for profiles having an inversion boundary layer. Next we apply this set of radiance-based spectral ADM algorithm to synthetic radiances computed from a different set of temperature and humidity profiles and compare with the synthetic flux directly computed from the same data set. The differences are further examined and stratified with respect to the groups of surface temperature, lapse rate, and TPW. The inclusion of land surface spectral emissivity in the algorithm is also explored.