(please see the supplement pdf)

Summary:

The manuscript by Poltera et al. describes a different method for analyzing water vapor measurements using frostpoint hygrometers. The idea of their paper is that equilibrium points are defined as measurement points during which the reflectivity measured by the instruments does not change. They name these points golden points and use them to correct measurements by other instruments, such as the Vaisala RS41. After that, they use these corrected measurements to improve the CFH measurements between the equilibrium points.

This paper provides a novel idea and should help improve the analysis of frostpoint hygrometer observations. The authors recommend making measurements of the reflectivity public in quality controlled final data, most importantly the NDACC data set. This is a good suggestion.

However, the paper suffers from a lack of focus and a generally verbose style, which makes it hard to follow. Th wordiness also hides some of the limitations of this approach and thereby its true value.

This manuscript requires major revisions, for which I detail my reasons below. Nevertheless, the idea is a significant progress for measurements of tropospheric and particularly stratospheric water vapor. If the authors can agree to my recommendations, this study should become a significant contribution to this important observing technique.

Major comments:

1) In my opinion, the authors overstate their fundamental idea. Dew-point and frost-point hygrometry relies on the assumption that the temperature of the condensate layer is a measure of the vapor pressure in the overlying gas phase if a condensate layer does not change. This is also the definition of their golden points.

It has long been recognized that this principle is difficult to implement and that an active controller must regulate the mirror temperature.  In the early days, this was done manually (see for example the work by Brewer and colleagues) and later electronic implementations of that principle were built. In all implementations (manual or electronic readings), averaging is done to minimize the deviation of the amount of condensate from a pre-determined or manually observed amount condensate. The averaging then implies that the condensate temperature readings do not reflect the instantaneous vapor pressure, but rather a value averaged over the same time interval.

Their term “Golden Points” and the graphic representation implies that this averaging is not needed and that moments in time exist, when the instantaneous reading can be used to calculate the vapor pressure. However, in practice they find that this is not straight forward, and they again rely on averaging to smooth out the reflectance signal. They attribute the source of noise to electronic noise. However, it is more likely that this is noise in the overall measurement system, which includes the condensate layer. It should be possible to evaluate the contribution of electronic noise when no condensate is present, i.e. prior to filling cryogen and possibly during the mirror clearing cycles. However, this does not solve the problem that averaging is needed to determine so called golden points.

Since their method requires averaging as well, the term “points” is not quite appropriate and theirs is just another method of looking at averaged dew-points or frost-points. Calling this method “Golden” is somewhat overstated.

Any actively controlled system such as a chilled mirror hygrometer tries to minimize the error function (Um – Uset), i.e. tries to find the point where Um = Uset. Any deviation from that is used to counteract the disturbance. In mathematical terms, Um≠Uset is almost always true for instantaneous readings, but averaged over time, =Uset=const and therefore d/dt = 0. This is equivalent to the smoothing of the detector signal done by the authors to find the points where d/dt goes through zero. It is also true for any PID controller, which minimizes the error signal. Their misconception may lie in confusing instantaneous readings (either of the reflectance or of the mirror temperature) and readings after “sufficient” smoothing.

Brewer (A W Brewer et al 1948 Proc. Phys. Soc. 60 52), has an excellent description of the details of the instruments they used to make the first stratospheric frostpoint measurements. Brewer described the challenge of visually deciding, whether a condensate remains constant, when time constants were on the order of a minute. Their electric detector was a significant improvement and allowed a better determination of constancy of the frost-layer. In effect, “Golden Points” is only a new name for the chilled mirror principle, which had certainly been around since the 1920s (see some references in Brewer et al, 1948), although not yet suitable for stratospheric dryness. That achievement is clearly that of Brewer, Dobson, and colleagues. However, they are not the inventor of the chilled mirror principle or any “Golden Points” by a different name. Previous instruments worked on the same principle, but at warmer temperatures. To all scientists of that generation, it seems to have been clear that the necessary condition was a condensate that remains constant. The challenge was how to achieve that.

Having said that, there is great value in looking at the deviations of the reflectance from the expected value and to try to estimate a correction of the mirror temperature based on that error signal. This has to my knowledge not been done and should help particularly for instruments, which have larger uncertainties than others such as some CFH instruments or the SkyDew instruments.

2) In several meteorological and metrological communities, establishing well justified uncertainties is as important as the measurement itself. This aspect of the measurement process is ignored by the authors, and they do not estimate any uncertainties in their method. This leaves the reader, who has no other information than the claims of the authors, without a metric whether their method is better than previous work. I see great potential in this method and believe it may improve what has been done so far. But without uncertainties it is difficult to compare their results with previous efforts. The authors ignored the work that was done in the NDACC data set of frostpoint based stratospheric water vapor measurements, even though they are clearly aware of it and use these profiles. In that data set, the data have been filtered and smoothed using a variable Gaussian filter, which the authors acknowledge or averaged into 250 m layers. However, the authors ignored the uncertainty estimates provided in this data set and not having any uncertainty estimates of their own, it is hard to quantify the benefit of their work.

Plots of differences for example should include uncertainty estimates that were provided to bring the differences into context. That would then allow them to discuss the details and potential shortcomings of the existing uncertainty estimates and propose a better uncertainty treatment. It could well be that their method is superior to previous work, but just saying so is not sufficient.

3) I strongly urge the authors to reduce the length of their manuscript. I estimate that the total page count can be cut in half without any significant loss of information. This should help the reader get a better insight into the details of this method.

The majority of the manuscript focuses on the Cryogenic Frostpoint Hygrometer (CFH). They also discuss Skydew as another frostpoint instrument, and FLASH as another reference. To strengthen their discussion, I would suggest completely removing the discussion of FLASH (FLASH could be mentioned in one sentence in the conclusion section). I also suggest moving the Skydew discussion to the very end or potentially to delete it as well. It adds little to the value of their approach as a whole.

To further reduce text, the authors should look at repeated discussions of the same point and consolidate all of these into one single discussion. I will provide examples below but cannot list all.

Detailed comments:

Abstract, first sentence. Here the authors claim that they “introduce  new retrieval protocol for … measurements in the upper troposphere/lower stratosphere of unprecedented accuracy”. This claim is repeated in line 418 and again in line 218, here at least with the caveat “at the equilibrium points”. However, nowhere in the manuscript do they actually discuss the accuracy of their method. They only show differences from the (I assume NDACC) data that without referring to the uncertainties provided in these files. Their discussion should be refined by deriving an accuracy of their method and comparing it with the existing estimates. This is clearly doable but has not been done.

What they term as accuracy is the difference between the raw CFH measurements and corrected CFH measurements based on a comparison with another sensor, where the correction of both sensors has been combined into one higher order algorithm. This is not an accuracy estimate.

Lines 29-31: This sentence represents the lack of understanding of the NDACC frostpoint data, which report uncertainty estimates. This statement should be made within the context of the uncertainties and vertical resolution reported therein. Are the large deviations that they find, covered by the larger uncertainties reported in the data files? I assume the large deviations also go along with large uncertainties in these files. If their method can reduce this uncertainty, or if their method shows that the existing uncertainty estimates appear to be incorrect, then this would be a very valuable contribution.

Lines 185ff: This would be the place to introduce the uncertainties that are explicitly provided in the NDACC data files, rather than just citing upper estimate values from publications.

Line 206: Equating the amount of condensed material with “layer thickness” is not valid. The authors should not give the impression that variations in reflectance are equal to variations in “layer thickness”.  As they are aware, condensate can grow along preferred nucleation sites (mirror imperfections) or along small temperature gradients within the mirror. How well operators clean the mirror is also likely to play a role. The current frostpoint hygrometers are unable to derive a layer thickness from the reflectance. The term in parenthesis should be deleted here.

Line 230ff: Do the authors imply, that the time lag correction applied by Vaisala is incomplete and hence apply a new correction? Is that a correction on top of a correction or a correction using the Vaisala raw data?

Line 298: Vömel et al (2007) quote a range “between 0.1 K for well-behaved instruments to 1.0 K” for slightly unstable instruments.

Line 379ff: It may be worth mentioning again, that Vömel et al  (2007) quote a value of “0.1 K for well-behaved instruments”, implying that this may be a rough uncertainty for “well behaved” instruments, much less than the several K referred to here. While this is clearly not stable in a mathematical sense, it would meet WMO criteria. The authors show an example for such a profile in Figure 7. The statement “a chilled mirror hygrometer is therefore unable to keep the condensate in static equilibrium…” is too strong. Deviations of 0.1 K seem to be close enough to be able to call this “equilibrium” using the WMO criteria for measurement accuracy.

Line 382ff: The term “transient equilibrium” feels like a contradiction in terms. Please note my discussion above, that the authors use averaging to define the equilibrium. Figure 10B shows that the equilibrium condition is not always perfect. In this case the authors explain the deviation by the insufficient temporal resolution and shift the profile by a fraction of a second. A similar effect can happen within the CFH at slower and well resolved oscillations if they are large enough. That may point towards the limitation of the assumption of equilibrium within the mirror system. While this does not contradict the basic idea, which I fully support, it warrants caution calling such points “golden”. Here, the authors need to smooth the reflectivity signal by 15 s, which is within the range of oscillations by the CFH. This means that instantaneous measurements of the equilibrium (points) are insufficient. This entire discussion should be re-written.

Lines 262ff: Is it a good assumption that the reflectance is directly given by a change in the frost coverage in ug/cm2 if it also depends on how the frost coverage is distributed? It is possible that a redistribution of frost-coverage without a net loss of ice might lead to a change in the detector signal. Why else would there be an influence by the ice morphology?

Line 315: It could probably be added that the instrument can fly in the wake of the parachute and balloon.

Lines 422f: Please define what you mean by “fine structure”. Is that structure on the scale of meters or many 10s to 100s of meters?

Lines 424ff: This discussion compares instruments, which do not have quantified uncertainties (likely large and understood biases) with an instrument that has quantified uncertainties. This sentence should somehow reflect that.

Line 434: Is that uncertainty taken from the data file (actually estimated) or taken from a publication (nominally)? The NDACC data files do provide these numbers.

Section 3.2: It is a good approach to correct a slower responding instrument such as the CFH using data from a faster responding instrument (FLASH) after it has been bias corrected. The minor challenge is that it is not absolutely clear where the FLASH biases come from, but as long as that instrument has sensitivity, this should work. For the RS41, this may be a little tricker to implement. The RS41 itself is a slow instrument in the region of interest and has been time lag corrected. That may introduce unreal structures in the vertical profile that are a result of the correction. Furthermore, the RS41 does lose sensitivity in the stratosphere. The manuscript gives the impression that the RS41 data could be used up to burst altitude. Please add a discussion up to what altitude this instrument could be used.

Lines 482: This statement points at the true value of their work, how to minimize the uncertainties away from the equilibrium points using additional information coming from the detector signal.

Line 505: Why do the authors assume an area density to describe the amount of condensate? That implies a homogeneous condensate layer and wouldn’t properly describe a patchy or coarse condensate layer, which is mentioned at other places in the manuscript. Describing the condensate layer by the number of molecules could be more appropriate.

Line 508 and many other equations: The authors mix equations that are written on their own line and equations that flow within the text. It would be easier to follow if all equations were written on their own and given an equation number. Why do the authors introduce A’ before A? Can they introduce another letter to avoid confusion between both?

Lines 511ff: The assumption that A’ is constant over time is probably true over short time periods, but probably not over the time of an entire ascent, or even just stratospheric ascent. How will that affect their correction algorithm?

Section 4.1.2: The thickness of the diffusion layer depends on the speed of the air flowing across the mirror. Due to the pendulum motion of the payload, the air flow through the tube is likely the vary somewhat. During descent, the airflow is likely to be much different, changing the thickness of the diffusion layer significantly. Therefore, this parameter and others that include it cannot be taken as a constant. How much does it vary within a sounding?

Lines 565ff: The number of equilibrium layers depends largely on the controller. The performance of the controller depends on the last three items in this list. It is an active controller, and the performance is not independent of these. The variation of the airflow through the tube is a simple external disturbance, it is neither “good” nor “poor” for the balloon instruments. The morphology is not just fine versus coarse frost, it is also the distribution of frost on the mirror (patchy versus homogeneous).

Line 583: Where does the estimate of 250 um come from? It is not justified in section 4.1.2.

Line 589: I would not call A constant, when the authors devote an entire section to discuss its variability. It’s a parameter.

Lines 606ff: This is a key issue with the so called “golden points”. I agree that the noise in the measurement system must be smoothed. A 21 s filter is about the time period of the controller response, i.e. a fair amount of the noise in the system is filtered out in this step. Filtering is important, but that makes “points” into “averages” and points no longer represent instantaneous measurements but rather time intervals. This is what happens in all frost-point or dew-point hygrometers.

Line 618: Where does this uncertainty (+/- 0.2 K) come from? In the caption of Figure 3, a value of +/- 0.3 K is stated. By their definition, the uncertainty at the golden points should be exactly 0 K. There is of course the calibration uncertainty; however, it is not relevant here. Since they average around the equilibrium point, the accuracy of that point has to be slightly degraded. Is that the source of their uncertainty? As pointed out earlier, the estimate of that uncertainty needs to be discussed in greater detail.

Line 624f: This has even been done away from the equilibrium points through averaging.

Line 631: Vaisala is operationally correcting for time lag. Is there a reason not to use that? Do the authors know, what time constant Vaisala is using? If they are correcting the Vaisala RS41 data based on raw data, do they use the relative humidity at sensor temperature or that at ambient temperature? What smoothing is applied to the RS41 time lag corrected data? How does their time lag correction compare to that applied by Vaisala?

Line 648ff and Figure 3: The data in this figure were taken prior to the launch of the balloon with very different ventilation, with liquid water at condensate, and with constant humidity and constant temperature. This never happens in flight. This figure is good for showing the basic idea, but at the same time may hide the complications in flight, when ambient humidity and temperature are constantly changing. Can they show a similar example from a typical flight segment? Can the parameter A be derived as cleanly? What would be the uncertainty and time dependence in flight? In the same context, it is probably not appropriate to talk of “monolayers”, since there is no monocrystal of H2O on the mirror. Furthermore, how can the authors translate this parameter into a “absolute sensitivity” of the hygrometer? And how can they compare this “absolute sensitivity” to a precision during one lab campaign? Absolute sensitivity and precision are very different things.

Line 687: Do the authors use the measured sensor temperature or, as written, an approximate ambient temperature plus 5 K?

Section 4.2.3: This section indeed contains a new and interesting aspect. Here, they make use of information (the reflectance) that has previously been ignored.

Line 703: How many flight segments are there? From the previous descriptions I assume 3, i.e. liquid condensate, ice between -15 C and -53 C, and again ice below -53 C. However, that may not be correct. In particular, in the stratosphere, the RS41 loses all sensitivity, and this approach can no longer be used. I’m sure the authors considered that and filtered the data accordingly. However, I could not find that part explained. 

Line 704: Here, the manuscript mentions a 31-point second order Savitsky Golay filter. The caption to Figure 5 mentions a 15-point and 21-point (half width) filter. Are these indeed different? Does the filter width vary between soundings?

Line 805: The suggestion of using pre-launch data to estimate the instrument parameters is not appropriate. This requires that the sensor is ventilated, which it normally isn’t. Furthermore, the instrument is not yet in full thermal equilibrium, which can create additional artifacts. Using in-cloud data also seems to be risky. CFH operations typically avoid liquid clouds, where 100% is a good assumption. In cirrus clouds, this assumption of ice saturation longer holds.

Lines 860ff: The discussion about the temporal resolution is important and should be expanded. The method by the authors may indeed improve the temporal resolution over the smoothing that is done by other groups. Just using equilibrium points limits the resolution to the controller oscillations. Using what they call non-equilibrium correction may indeed improve that. It would be very beneficial, if they could define and quantify the temporal resolution for their method.

Lines 882: This is an interesting discussion. Clearly, the frostpoint hygrometer shows some oscillations. As they point out, for example in the region between 11.1 km and 11.2 km, the mirror temperature appears too cold. They should also point out, that in this region, the reflectance seems to decrease (growing condensate), which is consistent. It also appears, that the rate of condensate growth is related to the level of cold bias, which is also consistent. They point this out 20 lines later, but this seems to be the main benefit of looking at the reflectance, which should be highlighted at the beginning of this section (or even much earlier in the paper). 

Section 5.3: This is an important point but should be shortened to bring out the true value of their correction approach.

Line 1050: Where does the number of 0.2 K come from? Does it refer to the WMO criteria?

Lines 1053ff: Here they compare the benefits of their method to the raw frostpoint measurements. However, it would be more instructive, if they compared that to the NDACC data product. What is the benefit their method could bring to that data set?

Figure 7: Here they end the correction at the hygropause because of the higher reliability of the RS41 below. This point should be expanded, since it refers to the altitude up to which this correction may be used. Can the authors add the actual water vapor profile and indicate the tropopause altitude?

Section 5.5.1: This is an important section, since it indicates the strength of their approach for stratospheric measurements. It can be shortened and strengthened at the same time.

Line 1085: This is the first time that an altitude limit for the RS41 is mentioned. The limit of the RS41 and how they extend the approach using tropospheric data is important and should be highlighted earlier.

Figure 8, Caption: Move all commentary into the main text. In particular, justify the RS41 limit in the main text in its separate discussion.

Lines 1137ff: This entire paragraph is an interesting discussion, but not relevant here. What is missing is how the authors applied their correction in the absence of any RS41 or other reference data. That is explained in the following section (5.6) for Skydew. However, that section ends with the sentence that for the CFH it is better to use a second reference instrument. It is not explained how they did the correction shown in Figure 10. The non-equilibrium discussion of section 5.6 for Skydew disrupts the flow of the discussions for the CFH and does not contribute there. This could be addressed in two ways. Finish the entire discussion for the CFH (including section 6) and then elaborate on Skydew and its differences to CFH. Alternatively, expand section 5.6 to explain how this works for the CFH and how the data shown in section 5.5.2 were created. In that case, it would be better to move section 5.6 before of section 5.5.

 Lines 1180ff: Since the mirror temperature of the Skydew oscillates faster and with a stronger amplitude, have the authors considered that the basic equilibrium assumption may not be satisfied. Because of the fast and large oscillations, the entire system may not be in sufficient equilibrium and even measuring the temperature at much higher frequency may not solve that challenge. The ad-hoc time shift makes it work here, but this shift has been arbitrarily selected to make it work.

Lines 1195: Here the authors say that there is “good agreement”, yet that there are “finite errors, whose quantification are beyond the scope of this paper”. At the end of the paragraph, they state that for the CFH another sonde as reference is the best option. These statements are made without justifications and do require quantifications of errors. I don’t believe they are beyond the scope of this paper; they should be at the core of this paper.

Their example for the Skydew correction is in the upper troposphere, where the Vaisala radiosonde is probably still a very good reference not needing much correction. How does that work in the stratosphere 10 km higher at much colder mirror temperatures? Their section leaves the impression that it will always work without needing a second reference instrument.

 Line 1231: It is just a hypothesis that there are patches of coarse ice crystals on the mirror causing the large oscillations. There could also be a slew of other reasons causing this behavior causing controller instabilities. I would suggest deleting that hypothesis. It would be interesting to know how often the large changes are related to sharp gradients.

Lines 1233ff: Can the authors compare that to the smoothing done in the NDACC data set? Such a statement would clearly point to an improvement compared to what has been done so far.

Line 1275: The authors should clarify, that RS41 data have several corrections applied. These may also contribute to the differences between the instruments. Faulty sensors is a poor choice of words here.

Lines 1278: Do the authors imply here, that only tropospheric data are shown in the statistics? Line 1284 seems to imply that also data in the lower stratosphere are used. Figure 12 shows stratospheric frostpoint temperatures in regions where the RS41 is likely not to be sensitive. Please clarify the exact data range that was used.

Lines 1295ff: See earlier comment about the lack of equilibrium conditions in the mirror system of the Skydew. This applies for CFH as well.

Line 1297: The noise in the resistance measurement can be evaluated prior to filling the cryogen. It is unlikely to contribute to the uncertainty.

Figure 12 seems to indicate that the by far largest improvement of the corrections is for large outliers. This is an important contribution. Figure 12B is a little hard to interpret. It seems to imply that about 99% of all data have only a small improvement. But for the remaining 1% of all data, the improvement can be quite significant. How does that compare with earlier statements about the importance of the correction? Maybe I’m interpreting that figure incorrectly.

Line 1311ff: Here they implicitly repeat the basic chilled mirror assumption. Averaged over many points, a working hygrometer measures the averaged dew-point or frost-point temperature. The method of smoothing has only little influence. However, here they could make the same comparison using the NDACC data and evaluate, which method is better using this metric.

Lines 1318ff: “The remaining …” This is highly speculative. I would suggest just deleting this sentence.

Line 1321: Stray light should not be an issue here, since the authors only used nighttime soundings. If they indeed included daytime soundings, it would be good to see the analysis separated by day and night. That would truly show the influence of stray light.

Line 1327: Can the authors please quantify “occasional” in terms of percentage of soundings, not percentage of data points?

Line 1336: Here is another repetition of the assumption that the so called “golden points” are instantaneous measurements, where in fact, they rely on averaging, similar to other averaging methods.

Lines 1353ff: This is only stated here. Only one short profile section was used to make this argument (which could have been fortuitous), but no robust comparison with existing data was done. The paper would benefit greatly, if they supported this statement with a solid statistics using for example the NDACC data set.

Lines 1364: The uncertainty estimate of 4% is stated here but was not shown in the paper. What is the basis for this number? I must clarify that I agree that the uncertainty may well be reduced but understanding what the basis for their uncertainty estimate is will better support their claim. It may also point towards what are fundamental limitations and where could this technique be improved.

Line 1367: This is the first time they mention down sampling to 500m. That is extensive averaging, and the resulting uncertainty may be substantially lower than their estimate. How did they calculate their number?

Lines 1385ff: What do they mean by “a priori profiles”. This sentence needs rephrasing, since I don’t believe they want to say that “CFH a-priori profiles derived from the hygrometer’s Golden Points alone are not useful”. 

Text length:

In the following, I give an incomplete listing of text that can be combined and/or deleted. There are more and the authors should make an effort shortening their text and providing more focus.

Abstract: The entire abstract could be cut in half if some of the discussion therein would be limited to the main text and not repeated here.

I fully agree that water is a very important trace gas in the atmosphere. Yet, in the context of this lengthy manuscript submitted to a technology focused journal, I would suggest deleting section 1.1. 

Section 2.1.1 can be combined with section 2.2, which goes into further detail, and the combined section can be shortened significantly by referring to the appropriate original work.

Section 2.1.3:  The FLASH section appears to be out of place here and interrupts the flow of argument of the frostpoint hygrometers. Similarly, the FLASH section 4.3 interrupts the flow of the argument and contributes little overall. FLASH instruments are probably no longer available and may no longer play a role in observations of stratospheric water vapor. Since FLASH contributes little to the argument, it could be mentioned in a single sentence in the conclusion section of the paper or completely deleted. The argument of the paper on chilled mirror hygrometers would become stronger.

Section 2.2.2 can probably be deleted and replaced by a reference to the relevant previous publications. The few points that are relevant here could be combined with section 2.1.1.

Lines 491: Another example of text that can be deleted, since it has been discussed and referenced earlier.

The introduction to section 4.1, lines 496 to 502 are a repeat from earlier.

Line 510: Here they defined ΔU for liquid and ice using different scaling factors but make no use of that distinction later in the manuscript. In fact, other uses of ΔU are without the scaling factor. This line could just be deleted.

Lines 622f: This is one example of unnecessary cross-reference and preview, that can be deleted.

The entire section 4.4 can be deleted. All points are discussed in detail later.

Lines 867ff: These are repetitions from earlier and can be deleted.

Lines 992 to 1006 are repeats and can be deleted.

Lines 1060ff: This is a repeat from earlier and can be deleted.

Line 1081: Delete “This is further …”. This point is made in more detail two paragraphs later. There, the detailed discussion of the COBALD data is distracting and can be shortened.

Lines 1101: Since the ECC failed, this discussion adds nothing to the water vapor measurement. The ECC discussion can be deleted

Lines 1124ff: The chemistry discussion is distracting and not relevant for the water vapor measurements. It can be deleted.

Line 1231: “… deviations of more than 5 K are possible …” says the exact same thing as the previous half sentence. This can be deleted.

Lines 1269fff: This has been described before and can be shortened/deleted here.

Figure captions: In most figure captions, the text should be reduced to the minimum needed to explain the graphs. Any interpretation of these graphs should be moved to the main text and consolidated with the explanations there.

Technical details:

Line 16: Delete “true”

Line 360: Delete “an instrument similar to”. This was after all a chilled mirror hygrometer that was built specifically for that purpose. The introduction to section 3 could be shortened and combined with the previous discussions in sections 2.1.1 and 2.2.

Lines 399fff: Can you provide an example for this misconception? Otherwise, don’t use the term “common misconception”.

Throughout the manuscript, please make sure to refer to figures either using Figure X or Fig. X, but not both. In addition, make sure that the Figures are printed soon after their first reference (Figure 3 is referenced on page 23, but shown on page 30, Figure 4 is referenced on page 27 and printed on page 32, Figure 5 is referenced on page 33 and printed on page 35).

Line 652: Better use “average”, “mean”, or “median” value instead of “canonical”.

Throughout the manuscript the authors use the terms “accuracy”, “precision”, “error”, “uncertainty” and “sensitivity” almost interchangeably. In the context of their work, they need to be more careful, in their wording, since they have different implications (systematic, random, combined uncertainties).

Line 785: “As in this sounding” refers to which sounding?

Line 943: Change “thickness” to “reflectance”

Line 1046: Change “whether” to “which”