Noise reduction and retrieval by modified lidar inversion method combines joint retrieval method and machine learning

Abstract

To address the problem in which the signal-to-noise ratio of a raw atmospheric lidar signal decreases rapidly as the range increases, which has a tremendous effect on the accuracy and the effective range of lidar retrieval, many de-noising algorithms have been proposed. Among these methods, those based on the ensemble Kalman Filter (EnKF) exhibit good performance. EnKF-based methods can simultaneously denoise lidar signals and yield accurate retrieval results. However, due to poor forecasting in the EnKF step, biases exist in the results of these methods. In this study, a modified lidar inversion method was proposed for horizontal aerosol characteristic retrieval, which combines the joint retrieval method and Gaussian processing machine learning. This method compensates for the poor forecasting in the EnKF step in the joint retrieval method through the Gaussian processing machine learning algorithm, which can reduce the biases in the retrieval results. The modified lidar inversion method was applied to both simulated and real lidar signals, and the results show that the modified lidar inversion method is effective and practical in aerosol extinction characteristics’ analysis.