Machine Learning and Deterministic Methods for Detection Meteorological Phenomena from Ground Measurements: Application for Low-Level Jet and Sea-Breeze Identification in Northern France
Round 1
Reviewer 1 Report
See attached document.
Comments for author File: 
Comments.pdf
Author Response
Dear Reviewer,
The authors would like to thank the referee for his valuable comments. Below, all the addressed points are answered separately. The comments of the reviewers are given in normal letters and the answers in italic. In the revised manuscript, the revisions are marked up using the track changes function.
The manuscript “Machine learning and deterministic methods…” by Sayahnya Roy et al. refers to the important issue of how interesting meteorological phenomena can be detected using ground measurements. The methodology proposed by the authors is innovative, and - in spite of minor weaknesses in their application - the paper deserves been published, after a minor revision.
Major comments
- In their introduction, the authors correctly write that sea breeze may affect air quality “through its ability to efficiently mix different pollutants and therefore to favor the formation of secondary products”. Their references [9-12] describe only a subset of potential secondary products, namely processes associated with particulate matter. In the introduction one would have expected to read that sea breezes are also known to lead to photochemical smog formation, and it would be appropriate to include one or two references on previous work in this context.
As suggested by the reviewer 2, we modified the manuscript by removing the topic of aerosols in order to make the manuscript clearer.
- As the present paper includes results only on PM10, it may be misleading for the reader to find in the introduction a lot about toxicity, inflammation processes of lung cells and health effects in general, as the latter are attributed to finer aerosol fractions. The authors should add one or two sentences clarifying this, and possibly write that additional information on gaseous pollutants and finer particulate matter fractions, for instance PM2.5, would be required to characterize better the impact of sea breeze phenomena on air quality.
As suggested by the reviewer 2, we modified the manuscript by removing the topic of aerosols in order to make the manuscript clearer.
- It is incorrect to use the term “thermal instability” for a local circulation system (in this case the sea breeze), line 185 on p. 6. Horizontal temperature gradients evolve everywhere in the world (mountain and valley wind systems are another example), and - following Le Chatelier’s principle - compensating secondary flows emerge which are by all means stable.
We corrected.
Minor comments
- The authors were lazy regarding the numbering of their figures and the associated referencing in the text. They should carefully correct the large number of mistakes, as otherwise most readers would be seriously confused.
We corrected.
- The references are numbered twice in the list at the end of the manuscript.
We corrected.
- The authors use a large number of abbreviations, and it would be very helpful for the reader if a list of these abbreviations would be included in the manuscript.
We included in the manuscript the following list:
Abbreviations:
NLLJ: Nocturnal Low-Level Jet
SB: Sea Breeze
AGL: Above Ground Level
RNN: Recurrent Neural Network
IOP: Intensive Observation Period
TKE: Turbulence Kinetic Energy
HWTT: Haar Wavelet Threshold Technique
SWT: Symlets Wavelet slope Technique
SCSBC: Sign Change of Sea Breeze Component
LSTM: Long Short-Term Memory
ADAM: ADAptive Momentum estimator
- The last paragraph on p. 7 should start as follows: “Figure 10 shows the time average of wind speed for all NLLJ days”.
We corrected.
- Apart from the English language editing recommended, the authors should take care that non-linguistic mistakes are also removed (line 31 on p. 1: “meteorological” instead of “metrological”; line 162 on p. 5: HWTNLLJ instead of HWTTNLLJ; Figure 4: “Decompose” instead of “Dicompose”; lines 271 and 272 on p. 8: μg/m3 instead of ug/m3).
We corrected.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Review of the manuscript „Machine learning and deterministic methods for detection meteorological phenomena from ground measurements: Application for low-level jet and sea-breeze identification in northern France“
By Sayahnya Roy et al.
The article introduces an interesting method, namely the determination of meteorlogical phenomena based on ground-based sonic anemometer observations. While the definition of sea breeze can be related to groud-based observations, the low-level jet (LLJ) is a phenomenon that is defined by an increase of wind speed with altitude. This cannot be measured by ground-based observations, therefore an increase in turbulence during night time is used as proxy.
In my opinion, this assumption requires a detailed validation with a statistically relevant data base. It seems that lidar data are available for a longer time period, so it would be possible to show that this method is actually valid.
I have several additional concerns about the content and methods, and therefore suggest rejecting the manuscript in the current form.
Major concerns
A general question is why you focus on nocturnal LLJ for a coastal site? The phenomenon of LLJ development as described in the article is typical for sites on land, with strong radative cooling of the surface during night. However, this is not the case in coastal areas. On the contrary, it is much more likely to have stable stratifications, which can result in LLJ formation, during the day: In summer the sea surface temperature of the ocean is relatively constant around 15-20°C. Above land, the temperature is typically lower during the night and warmer during the day. So if the air masses are coming from land and are advected above the ocean, then there is warmer air above colder surface. This temperature inversion leads to low vertical mixing, and potential formation of LLJ.
As a good overview of phenomena at the coast I would recommend
https://www.schweizerbart.de/papers/metz/detail/prepub/101184/Coastal_impacts_on_offshore_wind_farms__a_review_f
An analysis of LLJ and stability for coastal and offshore locations can be found here:
https://www.mdpi.com/2073-4433/13/5/839
They show that LLJ at a coastal location occur around as frequently at noon as at midnight.
Further, it is not clear how the topic of aerosols is mixed into the manuscript. I would suggest making a second publication out of it. L. 47-54 is somehow out of the context. It is unclear what it hast o do with sea breeze. It seems that for this specific location, there is more pollution for wind directions coming from sea to land, however, this has nothing to do with the phenomenon of sea breeze, but with the location of the aerosol sensors downstream of industry. In this paragraph, it sounds like there is toxic aerosol coming from the open water, which is quite surprising.
l. 35: this statement is true for sites above land, not necessarily for the coast. It also depends on seasons, and is much more complex than described here.
l. 56: Explain more the importance of sea breeze, include a thorough review of the literature on this, give examples
l.65: The study of Greene et al. is based on LLJ in the Great Plains. These are very strong LLJ cases, also introduced by orography. It is probably not appropriate to compare these to coastal LLJ in Europe.
l. 79-80: the LLJ classification based on surface wind measurements raises some questions, if this is statistically relevant. The focus of the article should be the validation with lidar data. The different applied methods may agree well, but they both have not been validated against real LLJ data. It says in l. 96 that you have 86 days of wind lidar data (up to which altitude?). So please use them!
l. 129: why are the „extreme events“ not considered? How do you define such extreme events?
l. 134: what is meant by „slope of temperature gradient“? Where is it measured? Why is it important?
l. 176-177: please explain „if the slope is positive from morning to midnight and negative from idnight to early morning“. You are talking about TKE here. So you are assuming the TKE increases during the day and decreases during this night. This is a general statement, indicating that turbulence is normally higher during day than during night. What does it have to do with LLJ development and your classification?
l. 218/219: Again, why do you limit LLJ to night time? What about interaction with sea breeze?
l. 245: why do you only compare your results to 3 lidar measurement days of LLJ? Why are all examples during the night?
l. 257/258: Power calculation. Why do you only take the wind speed at hub height for calculating power output? You should use at lest something like an equivalent wind speed, taking into account the rotor area, which is state of the art:
https://iopscience.iop.org/article/10.1088/1742-6596/524/1/012108
l. 261: 10 MW: Do you have validation data from a nearby wind turbine?
l. 265-267: Please provide a definition of sea breeze wind directions, maybe already in the introduction. A change from E to NE or E to SE is not much. Please define what changes in wind direction over what time period are necessary to classify the event as sea breeze.
Conclusions: I think it is not relevant that both of your methods agree to 99%. It would be much more important to compare them to LLJ statistics of the wind lidar
Conclusion 3 is a statement without any proof, and that is certainly not the outcome of your analyses. It is based on 1 simple calculation (which is wrong, in my opinion).
Conclusion 4 has no statistical proof either. I would be very careful to derive from 1 example that the wind speed at a certain altitude is 2 times larger than at ground. This depends strongly on many parameters and requires a valid statistics, which is not provided in the manuscript.
Minor issues
In the abstract it remains unclear which data sets are used for which analysis.
l. 63 please explain „SB events are very effective to fulful the power requirements“. What requirements are these?
Fig. 1: Text is hard to read. Please indicate the distance between lidar and sonic. The influence of this distance should also be discussed in the validation.
Numbering of figures not correct
Typos/grammar
l. 44 remove comma after „both“
l. 52: favor (remove s)
l. 141: remove comma after „since“
l. 149: output (not in capital letters)
l. 255: estimated
l. 271: use correct symbol for microgram
l. 276: within
l. 278: has, not have
l. 281: only place in the text that „we“ is used. I recommend to avoid it.
l. 296: with the aim
Author Response
Dear Reviewer 2,
The authors would like to thank the referee for his valuable comments. Below, all the addressed points are answered separately. The comments of the reviewers are given in normal letters and the answer in italic. In the revised manuscript, the revisions are marked up using the track changes function. You will find, in attached file, the response.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The manuscript has significantly improved. The statistics comparing the LLJ observations with the retrieval algorith is very useful. All comments have been taken into account appropriately, so the manyuscript is ready for publication.
