Neural network analysis of the radiative interaction between neighboring pixels in inhomogeneous clouds

Abstract

In this study, we analyzed the effect of the radiative interaction between neighboring pixels on the high-resolution radiant flux of bounded cascade inhomogeneous clouds by using a one-layer mapping neural network as generalized regression analysis. The analysis was done for reflectance, transmittance, and absorptance at different wavelengths under different conditions of illumination. The sign and magnitude of output coefficients indicate how neighboring pixels contribute to the radiant flux of a target pixel. We found that the variation of output coefficients with the distance from the target pixel changes significantly in its shape and horizontal extent not only with the type of radiant flux we consider but also with the wavelength and solar zenith angle. The mapping neural network clearly reveals the asymmetric feature of radiative interaction between neighboring pixels under oblique illumination, which illustrates the shadowing and enhancing effects of local cloud inhomogeneity. The present analysis shows that the mapping neural network is a flexible method of analysis when used as a generalized regression analysis. Copyright 2001 by the American Geophysical Union.