TL;DR
This paper introduces the Adaptive Noise-Reduction Peak-Retrieval (ANRPR) algorithm to improve the accuracy and detection range of wind-field measurements from Coherent Doppler Lidar (CDL). By combining adaptive noise reduction with 2D spectral filtering and an adaptive peak-retrieval strategy, ANRPR significantly enhances wind speed and direction estimation, especially in low signal-to-noise ratio (SNR) conditions.
What Problem Does This Paper Address?
Accurately measuring atmospheric wind fields is crucial for weather forecasting, climate modeling, pollution monitoring, and renewable energy development. Coherent Doppler Lidar (CDL) is a powerful tool for this purpose due to its high precision and long detection range. However, CDL signals are susceptible to various noise sources, particularly instrumental noise and atmospheric turbulence, which degrade the signal-to-noise ratio (SNR) as the detection distance increases. This noise makes it challenging to accurately extract the Doppler shift from the lidar echo signal, leading to biased wind speed estimations and reduced detection capabilities, especially in low-SNR regions. Existing methods often struggle with instability, require manual parameter tuning, or are less effective in noisy environments.
Methods
The proposed ANRPR algorithm tackles these challenges through a multi-stage approach:
- Adaptive Noise Reduction (airPLS): The core of the noise reduction is the adaptive iterative reweighted penalized least-squares (airPLS) method. Unlike traditional methods that might subtract a fixed noise profile or use simpler fitting, airPLS iteratively fits a penalized least-squares model to the power spectrum of each range gate. This allows for a more adaptive and robust removal of background noise, even in the presence of signal fluctuations.
- 2D Gaussian Low-Pass Filtering: To leverage the spatial continuity of wind fields, the algorithm applies a 2D Gaussian low-pass filter to the spectral data across adjacent range gates. This step smooths out small fluctuations and enhances the correlation between range gates, further reducing the impact of noise on peak detection.
- Adaptive Peak-Retrieval: Finally, an adaptive peak-retrieval algorithm is employed to extract the Doppler shift. This method is designed to be automatic, requiring no manual parameter adjustments, and is optimized for the characteristics of coherent wind measurement lidar echo signals. It aims to identify the true peak even in noisy spectra, improving robustness and accuracy.
The overall process involves dividing the lidar echo signal into range gates, applying airPLS for noise removal, splicing the processed spectra into a 2D matrix, applying the 2D Gaussian filter, and then using the adaptive peak-retrieval to extract Doppler shifts for wind speed calculation.
Datasets
The study utilizes observational data collected by a pulsed CDL system (model KC-WL-2006) manufactured by Hefei Tainuo Lidar Hi-Tech Ltd. The experiments were conducted on 27 November 2022, at Shandong Normal University. The paper also references and compares its results with data from a meteorological gradient tower for validation. No new benchmark datasets were introduced; rather, real-world lidar data were used for evaluation.
Key Insights and Ideas
- airPLS for Robust Noise Reduction: The adaptive nature of airPLS proves significantly more effective than traditional methods (like Periodogram Method or Least-Squares Method) in reducing background noise and improving SNR, especially in later range gates where signals are weaker.
- 2D Filtering for Spatial Coherence: Exploiting the spatial continuity of wind fields through 2D Gaussian filtering is crucial for improving peak detection accuracy by reducing spurious peaks caused by noise.
- Parameter-Free Peak Retrieval: The development of an adaptive peak-retrieval algorithm that doesn't require manual parameter tuning is a significant step towards automating and simplifying lidar data processing.
- Improved Performance in Low SNR: The ANRPR algorithm demonstrates superior robustness and accuracy in low-SNR conditions, which are common at greater detection distances, a key limitation for many existing methods.
Key Results
- Enhanced Detection Distance: The ANRPR algorithm showed improvements in detection distance by 26.3% and 85.1% compared to traditional methods.
- Higher Accuracy: Correlation coefficients with meteorological gradient tower data exceeded 0.90 for both horizontal wind speed and direction at various heights, indicating high accuracy and reliability.
- Improved SNR: The airPLS method alone demonstrated an SNR increase from -12.38 dB to -10.23 dB compared to the LSM method in specific test cases.
- Better Continuity: ANRPR significantly improves the continuity of wind speed estimations over range gates and time, reducing outliers and providing results more aligned with actual atmospheric conditions.
- Robustness in Low SNR: The algorithm exhibits superior stability and more accurate inversion results in low-SNR scenarios compared to traditional methods.
Limitations and Future Work
- Limitations: While ANRPR shows significant improvements, the paper notes that as SNR decreases with increasing height, the error rate of peak-finding can still increase. The study also mentions that the wind speed estimation accuracy is within ±0.5 m/s for vertical wind speed, which is within the permissible range, but further improvements might be desired.
- Future Work: The authors suggest that future research should focus on further enhancing the algorithm for a wider range of practical application scenarios, including vehicle-mobile settings for on-the-move detection. They also aim to explore more advanced methods to further improve accuracy and detection capabilities.
Summary updated