Use of Artificial Neural Networks to Retrieve TOA SW Radiative Fluxes for the EarthCARE Mission

Abstract

The Earth Clouds, Aerosols, and Radiation Explorer (EarthCARE) mission responds to the need to improve the understanding of the interactions between cloud, aerosol, and radiation processes. The fundamental mission objective is to constrain retrievals of cloud and aerosol properties such that their impact on top-of-atmosphere (TOA) radiative fluxes can be determined with an accuracy of 10 W · m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-2</sup> . However, TOA fluxes cannot be measured instantaneously from a satellite. For the EarthCARE mission, fluxes will be estimated from the observed solar and thermal radiances measured by the Broadband Radiometer (BBR). This paper describes an approach to obtain shortwave (SW) fluxes from BBR radiance measurements. The retrieval algorithms are developed relying on the angular distribution models (ADMs) employed by Clouds and the Earth's Radiant Energy System (CERES) instrument. The solar radiance-to-flux conversion for the BBR is performed by simulating the Terra CERES ADMs us ing a backpropagation artificial neural network (ANN) technique. The ANN performance is optimized by testing different architectures, namely, feedforward, cascade forward, and a customized forward network. A large data set of CERES measurements used to resemble the forthcoming BBR acquisitions has been collected. The CERES BBR-like database is sorted by their surface type, sky conditions, and scene type and then stratified by four input variables (solar zenith angle and BBR SW radiances) to construct three different training data sets. Then, the neural networks are analyzed, and the adequate ADM classification scheme is selected. The results of the BBR ANN-based ADMs show SW flux retrievals compliant with the CERES flux estimates.