Applicability of a Neural Network Approach to Retrieving the Optical Thickness and Effective Radius of Droplets in Single-Layer Horizontally Inhomogeneous Cloudiness

Abstract

Liquid-drop clouds play a significant role in the evolution of cloud systems and the formation of the Earth’s radiation balance. Determination of their optical and microphysical characteristics is one of the most important problems of optics and atmospheric physics. The paper is devoted to assessing the applicability of an artificial neural network to processing synthetic data of passive satellite measurements of reflected solar radiation of low and medium spatial resolution in the visible and short-wave infrared spectral regions in order to simultaneously retrieve the optical thickness and effective radius of droplets of horizontally inhomogeneous cloudiness. The network is trained using the Monte Carlo calculated values of radiance in marine stratocumulus clouds generated by a fractal model. Through a nonlinear approximation of the dependence of optical and microphysical parameters of clouds on radiation characteristics, the tested algorithm allows taking into account the effects of horizontal radiative transfer, unlike classical IPA/NIPA (Independent Pixel Approximation/Nonlocal Independent Pixel Approximation) schemes. It is shown that the errors in solving the inverse problem can be reduced by assimilating data in adjacent pixels, reducing spatial resolution, and using radiance data received at small solar zenith angles. The high correlation between the test and retrieved optical thickness and effective radius indicate the possibility of using a neural network approach to interpreting satellite measurement data.