Summary

The paper describes an instrument designed to continuously profile water vapor in the lower atmosphere at short time scales (10s averaging for Nyquist frequency of 0.05Hz).  The instrument design and relevant specifications are described.  An hour of high resolution data in the PBL and above are shown as examples.  The goal is to resolve turbulence and an analysis of turbulence spectra are highlighted.  The instrument is compared to radiosondes and state-of-the-art Raman lidar systems.  Overall the paper achieves its objectives but there are some concerns and issues that need to be addressed.

Specific comments: major issues and concerns

1. Calibration of the DIAL needs to be discussed further.  The thinking and explanation are not very scientific
1. (line 87) "The DIAL technique is advantageous for measuring water vapor for several reasons, most important because it is inherently self-calibrating by its working principle".  Then on line 205: The Water vapor DIAL is found to have a bias at low ranges. Then Line 210, the  biases are calibrated away using radiosonde data.  This removes one of the most useful benefits of DIAL, so Please discuss the magnitude of the problem, and calibration procedure further.
2. Line 206 "However, due to the DIAL principle and the instrumental setup, this cannot be a classic overlap issue." Why not?
3. Line 207  "Therefore, we assume that there has been an issue with a detector overload which leads to this artifact."  Shouldn't you be able to tell if the detector is saturating?  It later is indicated that the feature that would allow the authors to tell if there are issues with clouds was turned off, is that correct?
4. The calibration makes many of the comparisons not very compelling, such as at Line 604.  “4 systems show good agreement in the lowest 2 km above ground”.  Have not all of these systems been calibrated to the radiosonde below 1.1 km?
2. Rayleigh-Doppler errors in DIAL 
1. In the simplified version of the DIAL equation (Eq 1) it seems the outgoing and return absorption coefficients being additive is not correct.   [See Bösenberg 1998 Eq. 10 and 11]  Furthermore it is suggested to have the G term in Eq 1 written out or referenced.
2. The authors discuss that DIAL is subject to RD errors under two regimes, (molecular backscatter higher than aerosol, and at strong aerosol gradients), then proceed to not apply a correction due to its difficulty (and/or its introducing more uncertainty).  At this point, it seems relevant to note several papers in the recent literature that solve the Rayleigh Doppler problem in DIAL by simultaneously measuring the molecular to aerosol scattering ratio (backscatter ratio).  This was done with a high spectral resolution lidar (HSRL) channel.  This seems particularly applicable to the ATMONSYS instrument as it is well positioned to measure the backscatter ratio at 355 nm from the Raman lidar, or at 532 nm with an I2 HSRL channel (likely better as closer in wavelength).  Suggest that the authors acknowledge this method as a possible means to have a reliable RD correction.  This would also allow ATMONSYS to offer robustly calibrated aerosols measurements and remove reliance on Klett inversions (one of the many drawbacks of this technique shows up in Figures 6 and 7 - and noted in the caption of Figure 7)
3. The above is relevant as discrepancies with the radiosondes and Raman lidars may be  due to gradients (RD error) as mentioned in line 630. And the proper implementation of the correction term calls for further research (line 634).  The “shreds of clouds” mentioned in line 635 as the most probable cause of discrepancy seems also to be gradient problems (RD errors) but this could be addressed with an HSRL channel. 

1. Line 39 Numerical Weather Prediction.  There are confusing science drivers for this instrument application.  
1. This instrument is well-suited for short time scales –  the science driver mentioned in the opening sentences of the abstract.  And again in line 27 understanding the humidity transport process (process studies) is a very reasonable science objective. 
2. But the science driver given on line 41 needs clarification.  The WMO OSCAR requirements have uncertainty, temporal, vertical and horizontal requirements (note these are listed as ‘goal’, ‘breakthrough’,and ‘threshold’).  So is this uncertainty requirement for process studies, or for regular observations?  This instrument is not well suited to improve the numerical weather prediction for routine monitoring as it would be impractical to meet the horizontal requirements.  In the same sense that it is not economically feasible to make the radiosonde observations at sufficient horizontal spatial scales to improve weather forecasts.  Line 391 again references the WMO criteria <5%.   And finally, line 660 references this criteria again, and alludes to monitoring and data assimilation.  This does not appear to be a realistic science driver for this instrument. 

1. Line 197 Spectroscopic T and P dependency 
1. Line 197.  Please explain the rationale behind using radiosondes to inform the water vapor spectroscopic line parameters.  The data is short, presented from 1 hr of a single day.  Yet the radiosonde becomes uncorrelated in time (the radiosonde apparently was at 10.75 UTC and used to evaluate the period 11.6 to 12.6 UTC, correct?).  Would not surface measurement of the T and P (assuming a lapse rate and hydrostatic equation to get profiles) provide better results?  Reanalysis data would yield even higher quality data, if that was required.
2. Planetary Boundary Layer heights
1. Throughout the measurement section (lines 272 - 354) the planetary boundary layer heights (PBLH) are discussed.  Moisture and aerosol gradients are used synonymously with PBLH.  But, as is well known, these methods are proxies for the PBLH and can fail for a variety of reasons.  The aerosol lidar community often overlooks this issue.  The authors have the means to measure the PBL height directly using thermodynamic buoyancy (virtual potential temperature from radiosondes) or kinematics (Doppler lidar).  For example, the vertical wind velocity in Figure 7 provides evidence that the PBL is around 1 km above ground level at 12 UTC (automated methods to derive the top of the PBLH from this data based on the bulk Richardson number exist).   At minimum, why not use these direct methods as proof that the proxy gradient methods are correct for the time shown?  This is important to provide more confidence for claims as in Line 396 “This is again an indication for the position of the PBL top”  and the analysis that follows.
2. In cases where this is not possible (perhaps the overview section around Figure 5), state that PBL heights were assumed using gradient methods and, as such, might not be the actual PBL.
3. Line 370.  The rationale for the negative water vapor sounds reasonable but likely incomplete. Would not RD error be expected at the steep gradient?  How about the effect of cloud heterogeneity?  Furthermore, quality controlling the data by masking out negative water vapor might introduce problems from binning/smoothing.  Why not use a gradient method to remove clouds before retrieval of the DIAL to avoid any smoothing issues? 
4. Section 4.3 Turbulence spectra 
1. As a suggestion, since the frequency response doesn’t have much overlap with the expected trend, perhaps plotting this data as an Allan Variance (two sample variance vs integration time) would be easier to interpret.  In this case the Kolmogorov constant is +⅔.  
2. Of course it is possible the frequency rolloff at longer integration times may be due to instrument instability beyond 1 minute or longer.  But the most compelling rationale for the DIAL accuracy is the similar lack of low frequencies seen in Doppler winds.  The rest of the discussion regarding the reasons for the non-Kolmogorov atmosphere is tangential.
3. Line 465 Do you mean altitude 973 m AGL and not 500 m?  It is hard to tell which altitude is deviating from which.  But it is clearly different at all altitudes  from the Doppler wind spectrum.
4. Line 662.  “ A spectrum analysis of the DIAL showed good agreement with Kolmogorovs…”  This conclusion is unjustified.  What was shown was good agreement with the Doppler lidar frequency spectrum.  And that the Kolmogorov inertial subrange rolled off at low frequencies for some reason or other (which is not really necessary to explain) 

Technical corrections: minor grammar, misspellings, or strange word choices

Line 10. ‘Evaded’ perhaps 'explained' would be better?

Line 16.  Shreded is misspelled, but suggest changing to ‘broken clouds’

Line 46.  Deeply requested.  Suggest changing to ‘often requested’

Line 140. ‘Begin of the lidar range’.  Suggest ‘start of the lidar range’

Line 166. ‘Renowned DIAL equation’.  Suggest changing to ’well-known DIAL equation’ 

Line 281. ‘Surpassing the lidar’.  Suggest “passing over the lidar”

Line 336 and 633.  Straight forward should be one word, straightforward

Line 370 ‘Supersaturated’.  Not a good word choice in English for this condition.  Is it something in the electronic gain saturated or perhaps a non-linear response of the detector (perhaps some combination of both?).  Suggest ‘saturated non-linear response’  

Line 379  ‘Spread is weaker’.  Suggest ‘spread is reduced’

Line 621. 40 min hour.  It seems the word hour is unintended 

Line 636 and line 675. ‘Shreds of clouds’.  Suggest ‘wisps of clouds’

Line 639 ‘Such flags have not been set in the corresponding time’,  Unclear what meaning is desired here.

Line 644 ‘Proof’ is the wrong word.  Use ’prove’ 

Line 663 ‘Interludes’:  Suggest ‘portions’.  Actually the spectrum doesn’t agree well beyond at time scales longer than approximately 1 minute  

Many thanks for the paper “The ATMONSYS water vapor DIAL: Advanced measurements of short-term variability in the planetary boundary layer”.

This paper consists of two main ideas:

a) the description of the system ATMONSYS water vapor DIAL and

b) “claiming” of resolving turbulent latent energy fluxes (in different sections of the paper).

I would withdraw the paper and would rewrite it containing part a) as

part a) needs more information and

part b) is not sufficient. The unit of the turbulent latent energy flux is W/m2. There are no (height-dependent) latent energy flux values presented in the paper. 

I would need some more explanation regarding the topic “the paper present novel concepts, ideas, tools, or data”.

There are already published water vapor DIAL systems and (even in more detail) latent energy flux measurements with DIAL systems.

Please, check

Christoph Senff, Jens Bösenberg, and Gerhard Peters, Measurement of Water Vapor Flux Profiles in the Convective Boundary Layer with Lidar and Radar-RASS,

https://doi.org/10.1175/1520-0426(1994)011<0085:MOWVFP>2.0.CO;2

Behrendt, A., Wulfmeyer, V., Senff, C., Muppa, S. K., Späth, F., Lange, D., Kalthoff, N., and Wieser, A.: Observation of sensible and latent heat flux profiles with lidar, Atmos. Meas. Tech., 13, 3221–3233, https://doi.org/10.5194/amt-13-3221-2020, 2020.

In more detail:

Lines 40-42: It seems that a permanently measuring system for the WMO is the goal of the ATMONSYS developers? Otherwise it makes no sense to refer to the requirements of the WMO.

Only absolute values of the required uncertainties are given at the web page of the WMO. But, the authors indicate 5% which is a relative uncertainty. The truck in Fig. 1 looks like a mobile system which could also be used at different areas. In more dry areas one would need a system with an uncertainty of less than 2-10 g kg^-1 (marked/breakthrough, threshold values of the WMO for altitudes at near surface). Aside, I didn’t find requirements defined for air specific humidity at larger altitudes at this web page. I know that in arid regions the water vapor content can easily be below 1 g kg^-1 (at ground and at lofted altitudes). Given the presented lidar ATMONSYS, I don’t understand the argument of line 40-42.

Sec. 2.2:

Why is no beam expander installed? Is the emitted light polarized? If yes, how is the transmission of the different polarization directions and the retardation (of all used channels) in the receiver?

Sec. 2.3:

Line 158: The digitization range of the transient digitizer 12 bit, yielding to the dynamic range of the DAQ of 1/4096. The detector coupling to the digitizer needs to be explained (compare line 640). The AOD-dependent maximum range of water vapor measurements (lambda_on) should be given with this number and a standard atmosphere at mid-latitudes.

Lines 189-190: How stable are the emitted wavelengths? How is the seeding working? Which control is installed in the Ti-Sa that it follows DL1 and DL2 from shot to shot? How is the wavelength of the emitted light monitored?

Lines 208-210: The “logistic function” may be necessary, but needs more explanation. There must be a “physical motivation”! Otherwise, the authors should avoid discussing data below 1.1 km altitude. As I read, the median value is fitted to the “closest available radiosonde”. Hence, it seems to be a height-dependent calibration? What is done in case of a variable water vapor profile (known and also later shown). What is done if radiosonde data are not available?

At detailed error analysis of the ATMONSYS data is missed! This is an absolute prerequisite for a newly constructed system.

Table 2: More information is needed to the errors, the measuring range (under certain conditions). The authors didn’t show data until 4.5 km altitude. That’s why, I’m a bit anxious about this value in the table. Beam divergence of the laser: please, indicate that the given value is the full angle beam divergence (some scientists think about a half angle when using the words “laser beam divergence”). 

Sec. 3:

Lines 230-244, RAMSES lidar: I didn’t find the measurement errors and the vertical resolution in the description.

Lines 245-252, ARTHUS lidar: I didn’t find the measurement errors in the description.

Lines 253-265, Doppler wind lidar: I didn’t find the measurement errors in the description.

Lines 259-260: I didn’t get the content of the sentence “We removed the data with a high noise level by filtering with a relatively low Signal-to-Noise Ratio (SNR) + 1 threshold of 1.000 to keep the data availability high.”

Lines 266-270, Radiosonde: I didn’t find the measurement errors in the description.

Lines 271 and follows: Measurement day: 18 July 2021.

It seems to me that a new air mass has moved over the old one during this day as the depolarization of the particles (lines 284-286) and the wave structure in Fig. 7b indicate. Hence, it might have not been the best day for presenting the data.

Fig. 5:

There is a need to specify the local time relation to UTC since the local time indicates when the sun is at its highest and one may expect the largest height of the PBL. It seems that the largest height of the PBL is at 16 UTC which equals to 18 local time. Please, correct me and explain the details

It would also be possible and helpful to indicate the PBL height in one of the charts.

The rainbow scale of the colors in charts b and c is not suited for the given values (especially not in the PBL).

Why do some charts show clouds and others not as all charts are from one system?

I missed a chart of the ATMONSYS data in this Fig.!

There might be differences between the values measured by ATMONSYS and RAMSES (comparing Figs. 5a and 6a; time between 12 and 12.6 UTC; at least in the altitude range between 2.5 and 3.5 km)?

Sec. 4:

Lines 306-308: Never trust colored plots when looking for data out of the colors. But its okay, PBL top is ca. 1.2 km.  

Lines 317-319: The same explanation is necessary for the wind.

Line 322: delete “drastically” as it is “only” the factor of 2. 

Lines 325-327: The wind direction changed at the PBL top. Could this indicate a different air mass?

Lines 350-353: What a pity. 

Fig. 6:

Could you please confirm that the data are larger than the system noise nevel?

Could the “wave-like” structure of the particle backscatter coefficient between ca. 1.8 and 2.5 km altitude be caused by the slightly existing wind sheering?

Fig. 7:

a) looks much noisier than the other charts especially > 1.2 km. Why should the water vapor show different structures? Would it be possible to smooth the water vapor data in altitudes > 1.2 km?

Sec. 4.2:

This section is not straightforward. There are no formulas given at all.

For me it would be helpful to have first insights into the errors of the measurements (together with an error propagation analysis) and afterwards a second subsection with observations of the atmospheric variability.

Fig. 8 and all the explanations regarding this Fig. make no sense with the sentence at lines 425-427.

Line 423: Usually, the ascent speed of a radiosonde is 5 m/s. This means, that the radiosonde reaches an altitude of 5x60x5 m = 1500 m in 5 min (compare also line 587).

Sec. 4.3, Turbulent spectra:

Lines 430-431: There is not only the mechanically caused turbulence but also the thermally.

Line 432, “strong winds in the free troposphere”: the wind speed is almost the same in the upper PBL and the lower free troposphere (< 2.5 km)? 

Lines 432-433: Please, numbers of eddies diameters and PBL heights => for crosschecking.

Lines 438-439: Please, refer to the two given references in more detail.

Line 448: How long did you average? With other words, how long did you assume a “frozen atmosphere” (Taylor hypothesis). The same equation needs to be defined for the quantity “wind vertical velocity”.

Line 463: “good agreement” => how good?

Lines 474-475: The selected measurement day seems to be not optimal? Why not taking another day?

Line 480: “Thus, the co-spectra can be seen as frequency spectra of the latent heat flux.” => but the latent heat flux is more than spectra. It is a height-dependent value in W/m2. The error analysis (including error propagation) is missed.

Lines 506-507: The measurement equipment is only useful if its noise is less the “noise” of the observations. I feel, that the white noise from the instruments in the frequency spectra must be observable at higher frequencies and/that the -5/3 dependency is resolvable at lower frequencies than the white noise. 

Fig. 10:

The results are not understandable. Why is the co-spectral power largest at the lowest frequencies (largest eddies) at 358 m AGL?

Why differ the results so much from the observations of Senff et al. (from 10 July 1991, PBL top at 1100 m)? Both observations (Senff et al. and Speidel et al.) are made during summer time and the boundary layer heights are almost the same.

The section is titled “Turbulent spectra”. This represents the presented results.

The unit of the turbulent latent energy flux is W/m2. There are no (height-dependent) latent energy flux values presented in the paper. 

Sec. 5

I would like to see more detailed discussed inter-comparisons with other measurements to proof the performance of the ATMONSYS.

Sec. 5.1, Radiosonde:

Please avoid long explanations (lines 546-574), as the well-known fact “A spatial mismatch due to radiosonde drifting would be a more plausible cause.” replaces the discussion before.

There are many inter-comparisons published, but references to them are very rare (for instance to the campaign COPS; https://projekte.uni-hohenheim.de/cops/).

Sec. 5.2, Other water vapor lidars:

The error bars of all systems are missed.

Lines 633-635: It is only possible to present inter-comparisons which are in detail discussed including the correction term G. There is no sense for presentations of “half-inter-compared” results.

Line 640, full saturation of the channel: I didn’t get this idea. Usually, this should be avoided by the proper design of the detector coupling to the digitizer.