Estimation of CERES TOA Fluxes using Artificial Neural Network Algorithms

Continuous monitoring of the Earth radiation budget (ERB) is critical to our understanding of the Earth’s climate and its variability with time. The Clouds and the Earth’s Radiant Energy System (CERES) instrument onboard NASA Terra and Aqua satellites provide radiation budget measurements by combining broadband radiance measurements with radiances from a high-resolution, multispectral imager like MODIS. To estimate TOA radiative fluxes, CERES broadband radiances are converted to fluxes using empirical angular distribution models (ADMs). However, in the absence of imager coverage or unavailability of imager data (e.g., due to malfunction of the instrument), accurate scene identification and subsequent estimation of TOA fluxes are difficult. Another approach to estimate the TOA fluxes involve using the CERES ERBE-like algorithms which does not use any MODIS imager information and is based purely on standalone CERES broadband data. One advantage of using the CERES ERBE-like data product is its simplicity while standalone nature of the product allow user quick access to CERES radiance and flux data. Since CERES ERBE-like data is based on a 30-year old ERBE algorithm, the estimated TOA fluxes are prone to large uncertainty (due to Scene ID and ADM errors) compared to CERES TOA fluxes. In order to improve the standalone ERBE-like TOA fluxes, these two deficiencies must be corrected. This abstract describes a new methodology for estimating CERES TOA fluxes without using coincident MODIS data.

The new algorithm methodology can be split in to two steps. In the first step, CERES broadband radiances are first classified in to clear and cloudy scenes using the Random Forests (RF) method. Random Forests (RF) are an ensemble learning methodology (Breiman, 2001) which use decision tree classifiers as the base learner. RF scene classification is carried out using CERES TOA radiances (TOA LW and SW) and ancillary variables (eg., precipitable water content, skin temperature, etc.) in to clear and cloudy scenes without using any imager information. In the second step, RF classified clear-sky and all-sky broadband radiances are converted to corresponding TOA fluxes using artificial neural network (ANN) algorithms. For this purpose, a modified ANN algorithm of Loukachine and Loeb (2003) is used to produce CERES clear-sky and all-sky ADMs that can be used for converting CERES TOA radiances to TOA flux. Using this method, TOA clear-sky and all-sky fluxes are calculated for each CERES broadband radiance values without using any MODIS imager information. An inter-comparison of TOA fluxes estimated using the ANN method with that using CERES and ERBE method is carried out. Studies show that the ANN clear-sky and all-sky method produce better determination of TOA clear-sky flux, all-sky flux and cloud radiative forcing compared to the ERBE method.