Improved cloud phase retrieval approaches for China's FY-3A/VIRR multi-channel data using Artificial Neural Networks

Retrieving cloud phase accurately is important for cloud parameter studies, weather forecasting, and climate change research. Consequently, the purpose of this study is to develop better and more accurate cloud phase retrieval approaches to upgrade the current threshold technique used for China's second-generation polar-orbit meteorological satellite FengYun-3A (FY-3A). In this paper, improved cloud phase retrieval approaches using a supervised Back-Propagation Neural Network (BP-NN), and an unsupervised Self-Organizing Feature Map Neural Network (SOFM-NN) were proposed and investigated. The results of this study indicated that the two ANN approaches are satisfactory in discriminating cloud phase using FY-3A/Visible and InfRared Radiometer (VIRR) multi-channel data, and the average accuracy rates for the BP-NN approach are 93.50%, 93.81%, 94.25%, and 93.38% for the winter, spring, summer, and fall season categories, respectively, while for the SOFM-NN approach, rates are 91.93%, 92.08%, 92.63%, and 91.97%, respectively. The BP-NN approach performs slightly better than the SOFM-NN approach. Moreover, the two ANN approaches are found to perform more accurately than the current FY-3A operational product. Therefore, our work demonstrated that the ANN approaches provide an attractive alternative for cloud phase retrieval that could potentially be used to upgrade the current threshold technique used for the FY-3A operational product.