Coherent high-spectral-resolution lidar for atmospheric temperature remote sensing with a convolutional neural network

Abstract

A 1550-nm coherent high-spectral-resolution lidar (CHSRL) for detecting the atmospheric Rayleigh-Brillouin scattering spectrum (RBS) is proposed, and atmospheric temperature is retrieved by employing a convolutional neural network (CNN). The homodyne detection scheme is adopted to reduce the bandwidth requirement. The impact of the system parameters on the measured RBS line shape is analyzed and simulated, and a molecular scattering data correction method is introduced. Based on the system’s power spectrum characteristics, a convolutional neural network is used to extract temperature information from the measured RBS, thereby avoiding the initial value sensitivity and the high signal-to-noise ratio requirement of the conventional nonlinear least squares fitting algorithm. The feasibility of the proposed CHSRL integrated with the CNN for atmospheric temperature measurement is validated by continuous experimental observations with a temperature uncertainty as low as 2.2 K.