Denoising coherent Doppler lidar data based on a U-Net convolutional neural network.

Area of Science:

• Atmospheric science
• Remote sensing
• Signal processing

Background:

• Coherent Doppler wind lidar (CDWL) is ideal for atmospheric boundary layer (ABL) wind sensing but struggles with variable aerosol concentrations, limiting its detection range.
• Traditional spectral centroid algorithms for CDWL data processing are unreliable in low signal-to-noise ratio (SNR) conditions, affecting wind speed accuracy.

Purpose of the Study:

• To develop and validate a novel algorithm for accurate radial wind velocity estimation from CDWL data, particularly in low-SNR environments.
• To enhance the performance and extend the operational range of CDWL for atmospheric wind profiling.

Main Methods:

• A convolutional neural network (CNN) U-Net architecture was trained and tested using simulated lidar spectrum data.
• The proposed CNN U-Net algorithm focuses on denoising and accurate Doppler shift estimation for radial wind velocity retrieval.
• Performance was evaluated against the traditional spectral centroid method using simulated and joint observation data with radiosondes.

Main Results:

• The CNN U-Net algorithm demonstrated superior accuracy and a greater detection range compared to the spectral centroid method in low-SNR conditions.
• Numerical simulations confirmed the effectiveness of the U-Net for denoising and Doppler shift estimation.
• Joint observations with radiosondes showed excellent agreement, validating the algorithm's real-world performance.

Conclusions:

• The developed CNN U-Net-based algorithm significantly improves the accuracy and detection range of CDWL for wind remote sensing in the ABL.
• This deep learning approach offers a robust solution for overcoming the limitations of traditional methods in challenging low-SNR conditions.
• The findings suggest a promising advancement for atmospheric boundary layer wind measurement technologies.