Review of “Comparison of the Performance between Three Doppler Wind Lidars and a Novel Wind Speed Correction Algorithm”

General comments:

The manuscript presents a comprehensive comparative analysis of three Doppler wind lidars (DWLs) under varying environmental and technical conditions, alongside a novel machine learning-based correction algorithm. The systematic evaluation of three DWLs under diverse environmental conditions (aerosol concentration, planetary boundary layer height (PBLH), cloud base height (CBH)) provides actionable insights for instrument deployment and data interpretation. The discussion of aerosol concentration as a critical factor for low-altitude wind profiling is well-supported and relevant for urban air quality and aviation studies. The novel integration of IF and RF algorithms to correct lidar data is a notable contribution, particularly the validation against Aeolus satellite retrievals. However, several aspects require clarification, methodological justification, or refinement to strengthen the manuscript’s impact. Below are specific comments and suggestions.

The language is generally ​​good, technical, and appropriate for a scientific preprint​​. It effectively communicates complex research in atmospheric science and remote sensing. The logic is ​​sound and well-structured​​, following a standard scientific paper format. The fluency is ​​mostly good​​, though there are some minor grammatical issues and slightly awkward phrasing that could be improved for enhanced clarity and readability.

Major comments:

1. Ambiguity in the finding on aerosol impact:​ The paper repeatedly states that higher PM2.5 concentrations (L3: 35-50 μg/m³) lead to improved Lidar performance (e.g., "Particle mass concentration loading has positive correlation on the Lidar performance", "MSD Lidar exhibited the highest wind speed correlation(R2= 0.82)... when PM2.5 ranges from 35 to 50μg/m³", "the wind direction accuracy... is enhanced with the increase of aerosol concentration", "indicating that particle loading is the critical factor").

• My concerns are follows:​​ This finding appears counter-intuitive and contradicts fundamental Lidar physics and established literature. While aerosols are necessary as scattering targets for coherent DWL, ​​excessively high aerosol concentrations (like 35-50 μg/m³, typical of polluted conditions) cause significant signal attenuation.​​ This attenuation should degrade signal-to-noise ratio (SNR) and measurement accuracy, especially at higher altitudes, not improve it. The observed correlation improvement under L3 conditions needs a much more robust and physically sound explanation.

1. Insufficient discussion of vertical resolution mismatch in Aeolus validation:​​ The validation of the RF-corrected CUIT Lidar data against Aeolus satellite data (showing R²=0.83) is presented as a key result demonstrating the effectiveness of the correction algorithm. However, a critical methodological limitation is glossed over.

• My concern is the Aeolus L2B product has a ​​relative coarse vertical resolution​​ (0.25 km to 2 km bins, Table 1), while the ground-based Lidars (especially CUIT/MSD with 30m/50m/60m resolution) provide high-resolution profiles. Comparing a point measurement (radiosonde) or a high-resolution profile (Lidar) to a vertically averaged satellite product (Aeolus) is inherently problematic. The reported R² value might be significantly influenced by this mismatch rather than solely reflecting the accuracy of the RF correction or the Lidar itself. Averaging the high-resolution Lidar data to match Aeolus bins before calculating R² is essential for a fair comparison.
• My recommandations are 1. Explicitly state the vertical resolution mismatch as a significant limitation in the Aeolus validation section. 2. Discuss how the coarse Aeolus resolution might impact the validation result (e.g., smoothing out small-scale features the Lidar might resolve, potentially inflating agreement). Acknowledge that the R²=0.83 reflects agreement on the scale resolvable by Aeolus, not necessarily the full resolution of the Lidar.

1. The RMSE values for wind speed and direction (e.g., MSD: 1.11 m/s, CUIT: 4.45 m/s) are presented without explicit context. Are these values aggregated across all altitudes (<2 km) or stratified by specific height ranges? Clarify whether these RMSEs are altitude-dependent.
2. The "high-dimensional and incomplete datasets" processed by the IF-RF algorithm require clarification. Specify the input features (e.g., altitude, PM₅, PBLH) and the proportion of missing data imputed.
3. The R² = 0.83 between RF-corrected CUIT and Aeolus satellite data is promising, but the manuscript does not discuss potential biases in satellite retrievals (e.g., Aeolus’s own uncertainties under cloudy vs. clear-sky conditions). A brief comparison of satellite and radiosonde error profiles would strengthen this section.
4. The manuscript needs to improve writing quality.

Minor comments:

1. Pay attention to the use of "the". Like abstract: "Lidars data comparison focus on the low altitudes"
2. Coordinates need a degree sign and a space when naming the direction. e.g., (39.80°N, 116.32°E).
3. ..like m/s and m s-1...should be serious in writing scientific paper.
4. Check the numbers of section headings. Results and discussions should be 3, and. 1.3.1 Isolation tree should be 2.2.1….
5. A range of numbers should be specified as "a to b" or "a...b"(e.g., L1 (PM₅ = 0–15 µg/m³), L2 (PM_2.5 = 15–35 µg/m³), and L3 (PM_2.5 = 35–50 µg/m³)).
6. The statement "the measurement accuracy decreases with altitude increase (<2 km)" is redundant. Rephrase to "measurement accuracy decreases with increasing altitude (up to 2 km)."
7. The phrase "particle mass concentration loading has positive correlation on the Lidar performance" is ambiguous. Specify whether higher aerosol concentrations improve or degrade
8. The "optimal performance at lower altitudes (500–750 m)" conflicts with the earlier statement that accuracy decreases with altitude. Clarify whether "optimal" refers to relative performance across lidars or absolute performance.
9. Summary: "MSD Lidar exhibited the most accurate performance... which is the closest to the radiosonde" → "MSD Lidar exhibited the highest accuracy... closest to radiosonde measurements."
10. Avoid passive voice where possible (e.g., "It has been demonstrated that PBLH..." → "PBLH significantly influences lidar performance...").
11. "the accuracy of the three ground-based Lidars is evaluated" -> "...Lidars are evaluated".
12. "Lidars data comparison focus" -> "Lidar data comparison focuses".
13. "Comparison of results shows the root mean square errors(RMSE) for wind speed were..." -> "Comparisons show the root mean square errors (RMSE) for wind speed were..."
14. "the wind direction accuracy observed by the three Lidars is enhanced" -> "...observed with the three Lidars is enhanced".
15. "Machine learning was used to remove anomalies and complement the missing values" -> "...to remove anomalies and impute missing values"
16. "the random forest(RF) demonstrated superior performance with the Area Under the Curve(AUC) of 0.93(CUIT) and 0.90(WP)" -> "...superior performance, with AUC values of 0.93 (CUIT) and 0.90 (WP)".
17. "RF-based correction of CUIT enhanced R 2from 0.42 to 0.65." -> "RF-based correction of CUIT data enhanced the R² value from 0.42 to 0.65.".
18. "The R 2between RF-based CUIT and Aeolus satellite is shown as 0.83" -> "The R² between the RF-corrected CUIT data and Aeolus satellite data was 0.83"
19. "Fig.1. Location and instruments..." -> "Figure 1. Location of the Nanjiao Observatory and instruments deployed during the campaign."
20. "Author contributions: YZ performed the data collection and analysis, wrote the manuscript, HH performed the data presentation..." ("Performed the data presentation" is vague).

General Remarks:

The manuscript demonstrates a rigorous comparative analysis of three DWLs under different atmospheric conditions and integrates novel ML techniques to address data anomalies and gaps. The findings on PBLH-dependent lidar performance have direct implications for urban air quality studies and wind energy applications, where boundary layer dynamics are paramount. The proposed IF-RF pipeline offers a scalable solution for operational lidar networks, particularly in polluted regions with high aerosol variability. By evaluating lidar performance against radiosonde and satellite datasets, the study provides actionable insights into the interplay between aerosol dynamics, boundary layer processes, and instrument accuracy. While the work is methodologically robust, several areas require refinement to enhance clarity, statistical rigor, and physical interpretability. I consider this manuscript adequate for publication in Atmospheric Measurement Techniques once my comments are addressed.

Major Comments:

1. The manuscript should explicitly acknowledge the significant disparity in sample sizes across the analyzed PBLH strata, particularly the notably small sample (N = 41) within the 1500–1750 m stratum. This limited representation inherently reflects the rarity of PBLH events in the study region, as discussed in Section 3.3 ('PBLH’s impact on Doppler wind Lidars'). Consequently, any statistical interpretations or conclusions drawn specifically from this stratum require considerable caution and should be clearly qualified. We recommend discussing the limited statistical power and the inherent challenge of capturing sufficient events.
2. The study links PBLH and CBH to DWL performance but does not address how cloud microphysical properties (e.g., hydrometeor phase, particle size distributions, liquid/ice water content) modulate signal attenuation. High humidity under coupled PBLH-CBH conditions likely promotes hygroscopic growth, altering aerosol scattering properties. This microphysical-aerosol interaction mechanism is especially critical to consider in complex, polluted urban environments like Beijing, where aerosol loading and composition are highly variable and can profoundly influence lidar performance. We strongly recommend the authors discuss this potential mechanism and its implications for their findings, or explicitly acknowledge this limitation for future work.
3. Ambiguity in High-Aerosol Performance: The assertion that peak lidar accuracy occurs at PM_2.5=35-50 μg/m³ (L3) contradicts conventional lidar attenuation models. The conclusion that higher PM_2.5 concentrations (35-50 μg/m³) enhance lidar performance (e.g., Sec. 3.2, Fig. 5) merits further discussion. Given that dense aerosol layers are known to cause signal attenuation (reducing penetration depth and SNR), the observed performance improvement under polluted conditions (L3) appears paradoxical. We recommend expanding the physical interpretation to address how attenuation effects were overcome in this study.

(a) Lidar fundamentals: High aerosol loads increase attenuation, degrading SNR.

1. It is necessary to discuss the limitations of the temporal resolution of ERA5 reanalyzed data. Analyzing ERA5 data’s accuracy is not as precise as the radiosonde, please discuss the issue of time resolution.

Minor Comments:

1. Pay attention to the use of plural, like DWLs and DWL…
2. Line 12, Units must be written exponentially (e.g. W m–2).
3. Line 75-77, please add references.
4. Line 77-78, please check whether the reference is correct.
5. Line 92, Coordinates need a degree sign and a space when naming the direction (e.g. 30° N).
6. Line 96, Common abbreviations to be applied: hour as h (not hr), kilometre as km, metre as m.
7. Line 145, what is the IF
8. Results and discussions should be 3, and. 1.3.1 Isolation tree should be 2.2.1….
9. Line 317, what is ROC?
10. Line 339, what is AUC?
11. Line 382, ranges need an endash and no spaces between start and end (e.g. 1–10, Jan–Feb).
12. Avoids accusatory language (e.g., "contradicts" → "appears counter to," "warrants deeper analysis")
13. Line 419, “RF-based correction of CUIT enhanced R² from 0.42 to 0.65” → “RF correction significantly enhanced R² from 0.42 to 0.65”.
14. Simplify conclusions and avoid redundant expressions.