Application of an Artificial Neural Network Simulation for Top-of-Atmosphere Radiative Flux Estimation from CERES

Abstract

The Clouds and the Earth's Radiant Energy System (CERES) provides top-of-atmosphere (TOA) radiative flux estimates from shortwave (SW) and longwave (LW) radiance measurements by applying empirical angular distribution models (ADMs) for scene types defined by coincident high-resolution imager-based cloud retrievals. In this study, CERES ADMs are simulated using a feed-forward error back-propagation (FFEB) artificial neural network (ANN) simulation to provide a means of estimating TOA SW and LW radiative fluxes for different scene types in the absence of imager radiance measurements. In all cases, the ANN-derived TOA fluxes deviate from CERES TOA fluxes by less than 0.3 W m?2, on average, and show a smaller dependence on viewing geometry than TOA fluxes derived using ADMs from the Earth Radiation Budget Experiment (ERBE). The ANN-derived TOA SW and LW fluxes show a significant improvement in accuracy over the CERES ERBE-like fluxes when compared regionally.