A Comparative Study on Inversion Algorithms of Aerosol Extinction Coefficient Based on BP and Elman Neural Network

Abstract

Lidar, as an important technical means for detecting atmospheric aerosols, has been widely used in fields such as atmospheric radiation and light propagation. The aerosol extinction coefficient, as a physical parameter for quantitatively analyzing the light scattering characteristics of atmospheric aerosols, was of great significance for further research on the characteristics of atmospheric aerosols. However, the current use of traditional methods to retrieve aerosol extinction coefficient with disadvantages such as many assumptions, complicated calculations, and uncertain measurement results. In the paper, the neural network model, which was widely favored by the majority of scientific and technological workers, was used to invert the aerosol extinction coefficient. After extensive training using the BP and Elman network, it was found that the network could directly predict the extinction coefficient of atmospheric aerosols by detecting the returned lidar signals, which effectively improves the retrieval efficiency of atmospheric aerosols. The experimental results showed that after comparing the two networks used, it was found that Elman has a better effect than BP in retrieving atmospheric aerosol extinction coefficient, and the accuracy advantage was more obvious. Therefore, the practical value and application prospect of using Elman neural network were better when inverting aerosol extinction coefficient.