see attached pdf file

OVERVIEW

The airborne measurements of atmospheric turbulence performed during the MAESTRO field campaign are certainly unique and of great value for the studies of boundary layer dynamics, atmospheric convection and turbulence in general. The manuscript describes the motivation behind the campaign, aircraft instrumentation, flight strategies, corrections, calibrations, segmentation and classification of the data included in the companion dataset.

The main advantage is the simultaneous in-situ measurement of the turbulent fluctuations of the three wind velocity components, temperature and humidity. Temperature and humidity have been even measured with a few alternative instruments. The researchers using the data will appreciate that the results from different instruments are provided in a consistent framework and a few basic turbulence statistics have been already derived.

However, I believe some important aspects of data processing and quality assessment need major revisions before publication. These issues concern the calibration methodology, choice of best sensors for temperature and humidity, flight segmentation and heterogeneity score. The composition of the manuscript can be improved to clearly indicate the order of the processing steps applied to the data. Also, the key properties of the involved instrumentation and the discussion of related uncertainties are missing. Eventually, the content of the dataset is not entirely consistent with its description given in the manuscript. Those major problems are explained in detail below. Minor specific comments and technical suggestions are given in further sections of the review.

MAJOR ISSUES

1. Calibrations of the temperature and humidity instruments

It is a common practice to calibrate a  potentially drifting fast-response instrument against a stable slow-response reference instrument. However, in my opinion the selection of the reference based on the highest correlation with the fast-response one is questionable. In principle, this choice should not involve the results from the instrument being calibrated but rather be directed by a priori known characteristics, reliability or laboratory tests of the possible reference sensors.

a) The three reference hygrometers exhibit substantial systematic differences in the absolute values of water vapor mixing ratio. These offsets are carried on to the calibrated high-frequency record, depending on the choice of the reference sensor. Your correlation criterion does not take into account the absolute accuracy of the reference sensors. Yet, the accurate temperature and humidity are crucial if a user of the data wants to analyze relative humidity, saturation deficit or buoyancy effects.

b) Because the calibration is applied separately for each horizontal segment, the calibration of the instrument sensitivity (i.e. the slope of the linear calibration fit) might not work correctly in horizontally homogeneous thermodynamic conditions, i.e. where there are no significant variations in humidity/temperature along the segment. This can occur in weakly turbulent/laminar layers at higher altitudes and even in strongly turbulent flow without sources, sinks or mean gradients of these scalars. Then, the calibrated sensitivity is burdened with massive uncertainty and it might be better to recall the calibration from other segments. Please consider whether the problem is relevant for your measurements.

2. Choice of the best sensors

Scientific approach is to test theories against experiments, not the other way round. I believe the agreement of the measurements with the Kolmogorov theory must not be used to asses their quality. Ideally, your measurements should be used to test whether the theory holds. In fact, there is much evidence that either the Kolmogorov assumptions (e.g. homogeneity, isotropy) are sometimes not met in atmospheric turbulence or the predictions of the theory (I.e. universal scaling) do not agree with observations.

3. Segmentation algorithm

a) The algorithm to optimize segment placement should be thoroughly explained in the manuscript. In my opinion, the diagram in Fig. A1 is alone insufficient without proper description. In addition, even the diagram lacks the definition of some variables (e.g. "duramin") and implies some problems in the method. For instance, heterogeneity score seems to be computed over a data series of length "durati" whereas the segment stored afterwards has the length “seg_length”.

b) Please specify which temperature and humidity instruments are involved in the computation of heterogeneity score and hence in the segmentation procedure. It is also unclear, whether the segmentation involved already calibrated wind, temperature and humidity because, on the other hand, the calibrations themselves were performed on the segments.

4. Heterogeneity score

I suggest revising the design and description of the heterogeneity score. If it has been introduced in literature before, please provide appropriate references.

a) The definition provided in Eq. (1) can be simplified to make your idea comprehensible.

• Use either integration with respect to time (dt) or distance (dx), consistently.
• Do not use the same symbol T for time period and temperature.
• The factor 100 is actually irrelevant for segment definition and classification. It can be removed or placed in front of the whole expression.
• Where appropriate, you could write an average over a segment denoted by overbar (as in Eq. (2)) instead of the integrals.

b) As far as I understand the definition, the design of the heterogeneity score seems to be deficient in some aspects.

• Due to employing the “cumulative variance”, the score is not invariant to time reversal or shuffling the order of data points inside a series, which I would recognize as desirable properties for such a heterogeneity measure. A simple alternative, which can serve your purpose, might be the standard deviation of F² divided by the variance of F: std(F²)/std²(F).
• Due to the last integration over time, the score depends on segment length, which should rather be avoided. I guess you want to normalize by ΔT as in Eq. (3).
• The second term inside the brackets representing steady linear accumulation of variance also needs to be normalized by ΔT. Otherwise it has different units than the first term in the brackets.
• Try to examine whether the range of possible heterogeneity values is bounded.

5. Manuscript composition and clarity

a) I suppose it would be natural if the composition of the manuscript reflects the order of the processing operations applied to the data, for instance: segmentation, calibrations, quality assessment, derivation of turbulence statistics and their errors. The dependencies among the processing steps should be explained clearly in the text. I got confused because they might have been convoluted, e.g. calibrated wind, temperature and humidity are needed to define the segments (through the heterogeneity score) but the calibrations are performed on the segments, implying the segments are defined previously.

b) The volume of the presented material, in particular the number of plots, is large. For better readability I suggest focusing on the information which is most relevant for dataset users and removing or transferring to the appendix the secondary details.

• As far as I know the methodology, the wind calibration accounting for the biases of the attack and sideslip angles is a standard practice for 5-hole probes. I expect the same or similar had to be executed in the past so that you get your initial wind signals. Thus, to my mind you basically updated such a calibration. If so, I see it rather as a detail of pre-processing. It can be briefly mentioned in the main part of the manuscript but described and evaluated in the appendix. The corrected wind can be directly used to derive all the dependent quantities in the dataset.
• Fig. 4 illustrating example time series for a random segment is rather superfluous because it does not convey any important information about the dataset. 
• The analysis of the scaling of the inertial range (Figs. 9, 11, 13) is relevant only as far as it was important for the choice of the best sensors. Otherwise, only an example would be sufficient to illustrate the derived parameter included in the dataset, i.e. inertial range spectrum slope.
• Table 2 is not entirely consistent with the dataset content. Instead, the notations introduced there are rather confusing. Therefore, I suggest removing it.

6. Instrument properties

The uncertainties of the directly measured quantities are not discussed. In sec. 2.2, I expected information about accuracy and response time for each sensor, not only the acquisition frequency. Please provide the complete model names and manufacturers of the instruments if they are commercially sold or relevant references documenting their properties if they have been developed exclusively for your research. It could be also helpful to state whether the sensors have been calibrated or tested in a lab, whether the slow-response ones (e.g. dew point hygrometer) have been corrected for their inertia or fast-response ones (e.g. spectroscopic hygrometers) for the impact of the flow around the fuselage.

7. Dataset composition

A public dataset should be designed to serve users’ needs and with their convenience in mind. This requires simple and meaningful variable names, file names, folder structure and internal consistency across the dataset.

a) For easy searching and loading of the data, I suggest simplifying file names. Instead of combining three different conventions - MAESTRO (RFXX), ORCESTRA (2024-MM-DD-a/b) or SAFIRE (as2400XX) - please select one of them. Otherwise, a user needs to prepare a dictionary of these nomenclatures to guess a particular file name. For MAESTRO nomenclature, choose one of: RFX (no leading zero for a single digit), RFXX (two digits) or RF_X. Currently, all three are mixed in the manuscript and the dataset.

b) The keywords denoting three types of files - CALIBRATED, FLUCTUATIONS, MOMENTS for leg time series, segment time series and segment statistics, respectively - do not match their content. First, I understood all the files contain calibrated data, not only leg time series. Second, the segment time series contain not only the fluctuations but also the non-decomposed time series. Third, segment statistics involve not only turbulent moments but also integral length scales, inertial range slopes etc. You may consider LEG_TIMESERIES, SEGMENT_TIMESERIES, SEGMENT_STATISTICS or similar.

c) Please use the same variable names and units for a given quantity consistently across the dataset, I.e. in the three types of files. 

• In leg and segment time series, there are: ALTITUDE, LONGITUDE, STATIC_PRESSURE, T (for potential temperature), U_L, V_T. In segment statistics there are: alt, lon, lat, PS, THETA, U, V.
• In segment statistics, most of the variables follow the pattern [statistics]_[input_variable], e.g. MEAN_TS. However, there are a few which do not comply with it: UWE_mean, VSN_mean, WS_mean, W_mean, THETA_mean, MR_mean. This brings unnecessary disorder.
• T in some places denotes temperature, in other potential temperature. On top of that, potential temperature is also sometimes denoted with THETA. I suggest a unified convention, e.g. T to denote potential temperature everywhere.
• In leg and segment time series, time is given in milliseconds; in segment statistics as a string.

d) Horizontal wind velocity components are considered in three coordinate systems related to: aircraft, mean wind and geographic directions. Wherever you mean a given component in a given coordinate system, please always use the same variable name and provide both the component name and coordinate system in the attributes. Noting only “streamwise”, “longitudinal” or “transverse” without a coordinate system can be confusing because these can be understood to refer to the aircraft-oriented as well as mean-wind-oriented coordinate systems.

8. Dataset content

The set of variables provided in the files do not agree with the description given in the manuscript (lines 430-441 and Table 2).

a) Segment statistics files include more variables than listed in Table 2, e.g. turbulent moments derived from detrended time series. To my mind, Table 2 is confusing and I consider it rather superfluous. The notations given there are not used neither in the manuscript nor in the dataset. In some cases, the notation goes against the convention adopted in variables names.

b) Segment time series files are supposed to contain 19 time series according to the text: 9 non-decomposed series (two horizontal wind components in three coordinate systems, vertical wind, water vapor mixing ratio and potential temperature), 5 fluctuation series from high-pass filtering (three wind components in one coordinate system, water vapor mixing ratio and potential temperature), 5analogous fluctuation series from detrending. I found only 14 and some of them lack the attributes defining the coordinate system or derivation method (filtering or detrending). It looks like U_L_fluc, V_T_fluc are equal to U_c_fluc, V_c_fluc which makes me even more confused.

c) Leg time series files are not precisely described in the manuscript. The text suggests they should include the same variables as for the segment time series but this is not the case.

d) For horizontal wind velocity, I recommend providing at least non-decomposed and high-pass filtered time series in all three coordinate systems (as the detrended can be easily recreated by a user). If there is a storage limitation involved, you can provide them in one selected coordinate system but then in the appendix please give the explicit equations to convert into other two coordinate systems.

MINOR ISSUES

Manuscript

• Lines 28-39. Literature review extensively reports the results on shallow trade wind cumulus clouds. However, among the given targets of the MAESTRO campaign the equally important are deep convective clouds. If this is indeed the case, those could get a proportionate attention in the discussion of the past studies. In total, the review of previous works does not have to be long in such a paper describing a dataset.
• Lines 42-46. Please provide suitable references describing the listed campaigns if such exist.
• Sec. 2.3 (Flight sampling strategy). This subsection could get a simpler, logical structure if all the six types of horizontal transects are introduced together. Later you can discuss, which types were actually executed in which flights and why. It would be convenient to organize the subsequent figures so that the types appear in the order of their typical altitude, I suppose S, B, L, T, M, H. The division of the legs into homogeneous segments deserves a separate subsection.
• Sec. 2.3 (Flight sampling strategy). How does the segmentation described here for turbulence measurements relate to the datasets from other instruments onboard ATR-42? Is the segmentation universal or specific for this one dataset, how can one match other observations with the turbulence data?
• Lines 170-187. Are there any particular arguments behind the exact thresholds (30 and 60) of the heterogeneity score dividing the segments into homogeneous, intermediate and heterogeneous?
• Fig. 5. The caption should describe the content of the panels, e.g. for the last panel what the dots denote and what “std” in the legend means.
• Line 226. Please justify why you decided to compute dissipation rate from the vertical velocity component.
• Line 227. The inertial range can be resolved only up to an angular wave number corresponding to the Nyquist frequency, which is half of the sampling frequency, hence 0.8 rad/m.
• Line 230. I expect σ_f is the square root of: the definite integral of the vertical wind spectrum or the variance of the filtered vertical velocity time series; not the “variance of the integrated vertical wind spectrum”.
• Line 231. Please provide a source for the value of the Kolmogorov constant you used.
• Line 270-273. You wrote earlier (line 246) that the average vertical wind was already zero in straight legs before applying the wind correction. How can you then obtain the biases of the attack and sideslip angles by minimizing the average vertical wind in stabilized legs even further? In general, the sideslip angle controls mostly the coupling between the horizontal velocity components and does not significantly influence the vertical wind retrieval in horizontal stabilized legs. How the minimization of the average vertical wind can provide the sideslip angle correction then? This aspect is actually indicated by your Fig. 12, taking into account the negligible bias obtained for the attack angle and the measurable bias for the sideslip angle.
• Line 392. What exactly is meant by “only statistically significant population of segments”?
• Sec. 3.3.7 (Systematic error). Using the term “systematic error” is misleading here because what you computed is not the systematic error as introduced by Lenschow et al. (1994). They considered the inherent systematic error of turbulent moment resulting from approximating the ensemble mean by the temporal average of a finite segment. Yours is a consequence of removing a range of scales by the particular choice of cutoff frequency in high-pass filtering. I would not call it “error” but rather “filtering effect” or similar.
• Line 473. The cited raw dataset seems to include only temperature from the Rosemount and humidity from the FAST-WAVE, not the entire campaign data involved in the preparation of the turbulence dataset (i.e. raw wind, aircraft state, temperature and humidity from other sensors). Please clarify that.

Dataset

• Please explain what it means when data records contain NAN values. Is it due to the failure of instrumentation, unacceptable quality of measurement or failure of derivation method?
• (segment statistics) Together with the inertial range spectrum slopes, you can also provide the corresponding correlation coefficients in log-log space, which you analyze in the manuscript.
• (segment statistics) It seems the variables [VAR/M3/SKEW]_[U_L/U_L_DET/V_T/V_T_DET] are exactly equal to [VAR/M3/SKEW]_[U/U_DET/V/V_DET]. If so, one of these sets can be removed.
• (segment time series) The records are actually longer (5.5 min) then given in the text and indicated by the time bounds in the corresponding segment statistics (4.5 min). Please explain why.
• (leg time series) The files apparently involve additional leg types (D, A, V), which are not mentioned in the manuscript. Please correct the file names or define the extra types.

TECHNICAL ISSUES

Manuscript

• Line 4. Mesoscale cloud organization is already mentioned as a goal in the previous sentence.
• Line 8 and 81. The correct form in this meaning is probably “consists of”.
• Line 35. Please give details on the cited reference (Bony et al., in prep.) in the bibliography if it is already available as preprint or submitted to a journal.
• Line 172ff. The heterogeneity scores are given in % in the text but with no units in the plots.
• Line 176. Probably, you mean orange hashed area in Fig. 3.
• Line 247. Correlation of what?
• Line 251. Probably, you mean “from the air velocity with respect to the aircraft”.
• Line 323. Probably, you mean “wavelengths smaller than the integral length scale”.
• Fig. 11. Probably, you mean temperature sensors in the caption.
• Line 385. “Valid” might be better in this context than “relevant”.
• Line 432. Probably, you mean segment instead of flight.

Dataset

• The access via OpenDAP protocol through Thredds server does not work for the files containing leg time series and segment time series.
• (segment statistics) Probably, MEAN_WD is given in radians instead of degrees.
• (segment statistics) THETA_mean is given in K instead of Celsius.
• (segment statistics) MEAN_TS has a wrong unit given in the attributes.
• (segment statistics) Time is given in different string pattern than specified in the attributes.
• (segment statistics) There are variables lacking attributes: date, VAR_U_L …. SKEW_V_T .
• (segment time series) There are variables lacking attributes: U_c, V_c.
• (leg time series) For the T, long_name disagrees with standard_name.