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

Area of Science:

• Atmospheric physics
• Optical remote sensing
• Spectroscopy

Background:

• Accurate atmospheric temperature profiling is crucial for weather forecasting and climate modeling.
• Traditional lidar methods for temperature retrieval often require high signal-to-noise ratios and are sensitive to initial parameters.

Purpose of the Study:

• To propose and validate a 1550-nm coherent high-spectral-resolution lidar (CHSRL) system for atmospheric temperature measurement.
• To integrate a convolutional neural network (CNN) for enhanced temperature retrieval from Rayleigh-Brillouin scattering (RBS) spectra.

Main Methods:

• Development of a 1550-nm CHSRL system utilizing homodyne detection to reduce bandwidth requirements.
• Analysis and simulation of system parameters impacting RBS line shape, including a molecular scattering data correction method.
• Application of a CNN to extract temperature information from RBS power spectrum characteristics, bypassing conventional fitting algorithm limitations.

Main Results:

• The CHSRL system effectively detects the atmospheric RBS spectrum.
• The CNN-based method demonstrates robustness, avoiding sensitivity to initial values and high signal-to-noise ratio requirements.
• Experimental validation confirmed the feasibility, achieving a low temperature uncertainty of 2.2 K.

Conclusions:

• The proposed 1550-nm CHSRL integrated with a CNN offers a reliable and precise method for atmospheric temperature measurement.
• This approach presents a significant advancement over conventional lidar techniques for atmospheric remote sensing.
• The system's performance was validated through continuous experimental observations, demonstrating its practical applicability.