LWFNet: Coherent Doppler Wind Lidar-Based Network for Wind Field Retrieval

TL;DR

This paper tackles the challenge of high-resolution tropospheric wind field retrieval using coherent Doppler wind lidar, where traditional methods falter in low-SNR, high-altitude regimes. It introduces LWFNet, a novel end-to-end wind field retrieval network that combines a Line Transformer for preserving spectrogram integrity with a KAN decoder for interpretable wind extraction, trained on classical spectral centroid targets and validated against radiosonde data. Across high-SNR and full-range evaluations, LWFNet outperforms the conventional spectral centroid method and other SOTA deep models, revealing a surprising super-accuracy where the network exceeds the precision of its labels. The work demonstrates a practical advance in lidar-based wind retrieval, offering extended detection range, improved accuracy, and a benchmark architecture that leverages physical characteristics of CDWL power spectra for high-resolution wind monitoring and forecasting.

Abstract

Accurate detection of wind fields within the troposphere is essential for atmospheric dynamics research and plays a crucial role in extreme weather forecasting. Coherent Doppler wind lidar (CDWL) is widely regarded as the most suitable technique for high spatial and temporal resolution wind field detection. However, since coherent detection relies heavily on the concentration of aerosol particles, which cause Mie scattering, the received backscattering lidar signal exhibits significantly low intensity at high altitudes. As a result, conventional methods, such as spectral centroid estimation, often fail to produce credible and accurate wind retrieval results in these regions. To address this issue, we propose LWFNet, the first Lidar-based Wind Field (WF) retrieval neural Network, built upon Transformer and the Kolmogorov-Arnold network. Our model is trained solely on targets derived from the traditional wind retrieval algorithm and utilizes radiosonde measurements as the ground truth for test results evaluation. Experimental results demonstrate that LWFNet not only extends the maximum wind field detection range but also produces more accurate results, exhibiting a level of precision that surpasses the labeled targets. This phenomenon, which we refer to as super-accuracy, is explored by investigating the potential underlying factors that contribute to this intriguing occurrence. In addition, we compare the performance of LWFNet with other state-of-the-art (SOTA) models, highlighting its superior effectiveness and capability in high-resolution wind retrieval. LWFNet demonstrates remarkable performance in lidar-based wind field retrieval, setting a benchmark for future research and advancing the development of deep learning models in this domain.

Figures (7)

• Figure 1: LWFNet framework overview and spectrogram analysis. (a) illustrates the backbone structure used for LWFNet signal processing, here median filter serves as an optional block, the use of which will be discussed later. (b) presents a detailed view of the Transformer encoder architecture. Finally, (c) shows the vector embedding splitting strategy, emphasizing the relative independence of backscattering signals across different range gates and showcasing the effectiveness of the unique vector embedding layer.
• Figure 2: The ground-based coherent Doppler lidar employed for data collection and analysis.
• Figure 3: Scatter plots for the spectral centroid estimator and LWFNet, compared against radiosonde measurements, are presented for all 32 test instances in March 2024. The subtitles at the top of the columns indicate the specific region of interest tested. The first and second rows display the scatter plots for horizontal wind speed and horizontal wind direction, respectively. The $R^{2}$ values and least squares fitting line formula are illustrated at the bottom right corner of each plot. Note: (1) The speed resolution of radiosonde measurement data is 1 $m/s$, thus the scatter plot for horizontal wind speed takes the form of discrete columns. (2) Azimuth angles range from 0 $^\circ$ to 360$^\circ$ within a closed loop. To address situations where small differences in wind direction correspond to large absolute angular differences (e.g., 1 $^\circ$ and 359 $^\circ$), here we expand the conventional wind direction range from 0$^\circ$ to 360$^\circ$ to the range from -180$^\circ$ to 360$^\circ$.
• Figure 4: Wind field retrieval results of LWFNet and the spectral centroid estimator (denoted as "Profile"), compared with radiosonde ground truth. The subtitles at the top of the columns indicate the testing dates. The first and second rows present the evaluation results for horizontal wind speed and horizontal wind direction, respectively. Note: CST refers to China Standard Time and will be used as such throughout the text.
• Figure 5: Wind field retrieval results of LWFNet and the spectral centroid estimator on March 14th, 2024, from 10:00 to 14:00 CST, are presented. The first and second rows display the retrieval results from the spectral centroid algorithm and LWFNet, respectively.