the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Nov 2024
Measurement report: Can Zenith Wet Delay from GNSS "see" atmospheric turbulence? Insights from case studies across diverse climate zones
Abstract. Global Navigation Satellite Systems (GNSS) microwave signals are almost unaffected by clouds but are delayed as they travel the troposphere. The hydrostatic delay accounts for approximately 90 % of the total delay and can be well modeled as a function of temperature, pressure, and humidity. On the other hand, the wet delay is highly variable with space and time, making it difficult to model accurately. A zenith wet delay (ZWD) can be estimated as part of the GNSS positioning adjustment and is proportional to the specific humidity in the atmospheric boundary layer (ABL). Whereas its average term can describe mesoscale events, its small-scale component is associated with turbulent processes in the ABL and the focus of the present contribution. We introduce a new filtering and estimation strategy to analyze small-scale ZWD variations, addressing questions on daily or periodic variations of some turbulent parameters, and the dependence of these parameters on climate zones. Five GNSS stations were selected for case studies, revealing promising specific daily and seasonal patterns depending on the estimated turbulence at the GNSS station (buoyancy or shear). This research lays the groundwork for more accurate models and prediction strategies for integrated WV turbulence. It has far-reaching applications, from nowcasting uncertainty assessments to the stochastic modeling for Very Large Baseline Interferometry or GNSS.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(1265 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1265 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Atmospheric delays affect global navigation satellite system (GNSS) signals. This study analyses the wet delay, a variable component caused by atmospheric water vapor, using a novel filtering method to examine small-scale turbulent variations. Case studies at five global stations revealed daily and seasonal turbulence patterns. This research will improve water vapour and cloud models, enhance nowcasting, and refine stochastic modelling for GNSS and very long baseline interferometry.
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-2680', Anonymous Referee #2, 19 Nov 2024
General Comments
The manuscript describes a method to derive turbulence parameters from observations by ground-based GNSS stations. These observations are processed to obtain time series of the equivalent total zenith propagation delay. After a filtering process a time series of estimated variations in the equivalent zenith wet delay are used to characterise turbulence. I find the work interesting and unique, but have some thoughts concerning a critical assessment of the results and the structure and the clarity of presentation.
A first reflection is why the manuscript was submitted to ACP rather than AMT (Atmospheric Measurement Techniques). I find that the method, rather than the estimated turbulence parameters, is the motivation for publishing (and the title also starts with "Measurement Report:". I assume this is a question for the editor.
The time series of the Zenith Wet Delay (ZWD) from the GNSS data processing has a temporal resolution of 30 s. It is stated that no constraint is applied (line 122) but the estimates of the ZWD are averages of the air in the sampled volume. This volume depends on the elevation cutoff angle and which GNSS that were used. I miss information about this. If a frozen flow is assumed (as you do in the manuscript) and an elevation cutoff angle of say 5 or 10 degrees is used, quite a high wind speed is required in order to have independent air volumes with a sample period of 30 s. This issue should be discussed and how the results are affected by this fact. Depending on the conclusions you may want to reduce the temporal resolution.
A related issue is that there is a need to validate the GNSS wet delay time series obtained with the method you describe. One possibility is the use of microwave radiometry. The main advantage of the radiometer is that it samples a much smaller volume and it can measure in the same direction continuously. (Also in this case, however, the temporal resolution for independent samples is limited by the sampled air volume, which in turn is determined by the antenna beam width(s).) Probably none of the sites you study is equipped with a radiometer. However, there are several GNSS/VLBI sites that offer this possibility, see e.g. Teke et al. (2013) (Troposphere delays from space geodetic techniques, water vapor radiometers, and numerical weather models over a series of continuous VLBI campaigns, J. Geod.,87:981–1001, DOI 10.1007/s00190-013-0662-z).
You present results for one summer and one winter day for each GNSS station. I find this to be insufficient. If I may be a bit provokative, but a data point in a climate time series is often an average over 30 years. I have not visited all the sites studied but I have been in similar climates and my experience is that the weather conditions may change considerably from one day to another (possibly excluding the tropical site). I do not argue that you shall use data from several years, but because you spend much effort on comparing the results from the different sites there is a high risk for overinterpretation. One month, or at least one week, of data from each site and season will make more sense. I think it will also be of interest to compare the parameters estimated for adjacent days.
A related matter is that the argument that the spring and the autumn may be slightly different for the tropical station (SEY2) is definitely valid for all the stations. Consequently I recommend to handle all the stations in the same way. Furthermore, to me it does not make sense to present the "extra" results for RIO2 and SEY2 in appendices. It will be easier for the reader to follow if all the results for each station are presented together. (See more details in the section with specific comments below)
Specific comments
In the abstract (and in line 30) you state that "microwave signals are almost unaffected by clouds but are delayed as they travel the troposphere". In many geodesy applications the induced effect on the delay by clouds are ignored because thy are comparable to the uncertainties. In your case however, you surpress the large variations in the estimated propagation delay by filtering and the question is then how the small scale variability in the delay caused by clouds affect your interpretations of the turbulence parameters. For example, cumulus clouds may cause delays of several millimetres. Solheim et al. (1999)( Propagation delays induced in GPS signals by dry air, water vapor, hydrometeors, and other particulates, J. Geophys. Res., 104, 9663–9670) state that "A cloud droplet concentration of 1 g/m^3 for a distance of 1 km has an integrated liquid value of 1 mm and would therefore induce a radio path delay of 1.45 mm." In Line 36 it is stated that the topic is on WV turbulence. Perhaps it will be more correct to refer to turbulence due to WV and liquid water clouds?
In Section 2.1 you mention briefly that the ZTD can be estimated from GNSS observation. I think this is the appropriate place to give the details on how this is done rather than in the paragraph starting on Line 120.
Line (L) 219: It is not clear to me why satellite orbit and clock errors can be ignored. Do you mean that the products from GFZ are free of error? This cannot be true. If you believe that these errors are small enough to be ignored, it needs to be motivated.
Table 1: "Gravity waves" is not a type of climate. You may instead call it characterisation of GNSS station. Additionally, it is not really needed to present this in a table (with only one line as an entry). Each site can be described/characterised in the running text.
L 238-240: When checking the supplementary material I find that there is only one additional day for each site and season. From just reading the information available one cannot conclude that this material support your conclusions. As mentioned above you need more than one day of data to characterise the turbulence at a site.
Another question related to the supplementary material is that when I randomly checked some of the data files I find that the ground pressure is constant over the entire day in all cases. Does that mean that when you subtract the hydrostatic delay from the ZTD it has a constant value and that any variations in the hydrostatic delay will alias with the wet delay variability? This needs to be explained / discussed.
L 400-404: You have not presented any results for the PAYN station, so there is no need to discuss such results. If it is of relevance for this study I think it shall be included rather than referring to the "next contribution".
L 420: Also here the "next contribution" is mentioned. It is sufficient say that confirmation is needed because the reader will not know where the next contribution is to be found.
Technical CorrectionsA general comment: At many places you use the wording "in this contribution" (or something similar). In most cases these words should be deleted. It is obvious, e.g. L 8, 55, 178, 184, 215-216, 240, and 380.
L 3: 90% --> 90 % (Also at many other places: insert a "space" between the value and the unit (SI recommendation). Also there shall be no dash between the value and the unit.)
L 32: You probably mean hydrostatic delay (not dry), which is the term you use elsewhere?
L 48: Why is the slope not mentioned here?
Figure 1: Correct the labels on the y axis in the ZWD' graph (too close for a good readability).
L 104: Better to write the mathematical expression in one line, or make it a numbered equation.
L 106: Units shall not be in italics.
L 109: 1 hour --> 1 h
L 121: 30-second rate --> 30 s rate
L 126: 4h --> 4 h UT
L 170: Better to write the mathematical expression in one line, or make it a numbered equation.
Figure 2: The time series graphs are too small, make them bigger or delete them (they are not really needed)?
L 220, 221, 222, 223, 226: shall read 30 s, 1 h, 24 h
L 284: Do not start a sentence with a symbol.
L 318 & 326. RIA2 --> RIO2
L 411: 375 unit?
L 422: automn --> autumnCitation: https://doi.org/10.5194/egusphere-2024-2680-RC1 -
AC1: 'Reply on RC2', gael kermarrec, 23 Nov 2024
General Comments
The manuscript describes a method to derive turbulence parameters from observations by ground-based GNSS stations. These observations are processed to obtain time series of the equivalent total zenith propagation delay. After a filtering process a time series of estimated variations in the equivalent zenith wet delay are used to characterise turbulence. I find the work interesting and unique, but have some thoughts concerning a critical assessment of the results and the structure and the clarity of presentation.
>>>> Many thanks for your positive comments on the manuscript, which help us improve it.
A first reflection is why the manuscript was submitted to ACP rather than AMT (Atmospheric Measurement Techniques). I find that the method, rather than the estimated turbulence parameters, is the motivation for publishing (and the title also starts with "Measurement Report:". I assume this is a question for the editor.
>>>> The article was intentionally submitted to ACP because it offers the opportunity to publish 'measurement reports.' In our opinion, a publication in AMT would require a longer review process. By publishing here, we have the chance to present preliminary results and outline the methodology in one paper, ensuring we don’t need to repeat these steps in future publications.
The time series of the Zenith Wet Delay (ZWD) from the GNSS data processing has a temporal resolution of 30 s. It is stated that no constraint is applied (line 122) but the estimates of the ZWD are averages of the air in the sampled volume.
>>>> By 'constraint,' we mean that the ZWD is not assumed to follow a specific noise structure (e.g., a random walk). Assuming such a structure could bias our results, which are based on identifying slopes in the power spectrum.
This volume depends on the elevation cutoff angle and which GNSS that were used. I miss information about this. If a frozen flow is assumed (as you do in the manuscript) and an elevation cutoff angle of say 5 or 10 degrees is used, quite a high wind speed is required in order to have independent air volumes with a sample period of 30 s.
>>>> The ZWD values are mapped as 'slant wet delays' from all GNSS satellites to the vertical. While the frozen approximation is clearly an approximation but well documented in Wheelon (2001), we are confident in its validity for two main reasons: (i) we observe the expected slope in real data using both visual and statistical methods, and (ii) further investigations using machine learning strategies—combining real data from radiometers and other sensors at PAYN—indicate that the atmospheric layers sensed by the ZWD are at an altitude of 1500–2000 m, where the geostrophic wind predominates (study to be published in a next step). This point is further discussed in the corresponding section on the Taylor Frozen hypothesis and in the introduction. Additionally, we have enhanced the description of how the ZWD is computed.
This issue should be discussed and how the results are affected by this fact. Depending on the conclusions you may want to reduce the temporal resolution.
>>>>> This could be done, but not with the stations we selected (frequency is not available). Using a lower temporal resolution introduces further noise, which would require more advanced filtering strategies. We plan to address this in a future contribution on the topic. Once more, we derive the methodology with some examples.
A related issue is that there is a need to validate the GNSS wet delay time series obtained with the method you describe. One possibility is the use of microwave radiometry. The main advantage of the radiometer is that it samples a much smaller volume and it can measure in the same direction continuously. (Also in this case, however, the temporal resolution for independent samples is limited by the sampled air volume, which in turn is determined by the antenna beam width(s).) Probably none of the sites you study is equipped with a radiometer.
>>>> Our aim is not to validate the ZWD, as this has been extensively addressed in many other contributions. Instead, we focus on the short-term variations related to turbulence. As you correctly noted, the ZWD is a 'particular' parameter, representing an integrated quantity along the signal path and a time series, whereas satellite images, such as those from OLCI, derive spatial quantities (e.g., 'outer scale length'). In our view, it is unlikely that an instrument will be developed that allows direct comparison and that is the reason why we would carry on with machine learning (gradient boosting) strategies to combine several instruments for validation/insvestigations. This uniqueness makes the parameters we estimate both novel and innovative. The article highlights the need for further investigation to better understand these parameters, such as through large eddy simulations or the aforementioned gradient boosting. While we do not claim to hold the ultimate truth, our goal is to spark curiosity and foster further exploration. We have add a section "note" in the taylor frozen section to comment on that and further related topics.
However, there are several GNSS/VLBI sites that offer this possibility, see e.g. Teke et al. (2013) (Troposphere delays from space geodetic techniques, water vapor radiometers, and numerical weather models over a series of continuous VLBI campaigns, J. Geod.,87:981–1001, DOI 10.1007/s00190-013-0662-z).
>>>> thanks for point out these stations. We will keep that in mind and are currently making further work on PAYN.
You present results for one summer and one winter day for each GNSS station. I find this to be insufficient. If I may be a bit provokative, but a data point in a climate time series is often an average over 30 years.
>>>> I also work on climate time series and completely agree with your point. However, we are not claiming in the article to draw climatological conclusions. Instead, we demonstrate that the parameters we estimate appear to be related to the climate zone, potentially offering new insights. We are also not suggesting that the same parameters or identical patterns will emerge every day. The values naturally vary, as turbulence changes from minute to minute. However, a periodical patterns for some stations seem evident.
Although I have not visited all the studied sites, I have been in similar climates, and in my experience, weather conditions can vary significantly from one day to the next (with the possible exception of the tropical site).
I fully understand your concern about the need for more extended data collection. While we are not arguing against the value of multi-year data, given the considerable effort we put into comparing results across sites, we recognize the risk of overinterpretation. Collecting at least one month—or at minimum, one week—of data from each site across different seasons would indeed provide a more robust basis for analysis. Additionally, comparing parameters estimated for adjacent days could offer valuable insights.
>>>> We did conduct such investigations but chose not to include them in this article. The reason is that presenting multiple days could be confusing, as readers might focus on finding 'exactly the same' patterns (or not), leading to additional questions such as 'why only two days,' and so on. Instead, we restricted our analysis to one day to highlight the differences in patterns between seasons and stations. This approach also makes it easier to present the results (figure are not overloaded), as the goal of this work is to demonstrate the potential of the parameters (as a report) rather than conduct a deep investigation.
For a more comprehensive analysis, we would consider data from a single station over 30 years, incorporating techniques like moving averages and other methods. Understanding these parameters is a complex and challenging task that requires significant effort. However, in the revised version of the article, we have added some comments regarding adjacent days. We have add futher days in the repository.
A related matter is that the argument that the spring and the autumn may be slightly different for the tropical station (SEY2) is definitely valid for all the stations.
>>>> you are right but we are intented to think that the differences should be more extrem for a tropical station than for the other one.
Consequently I recommend to handle all the stations in the same way. Furthermore, to me it does not make sense to present the "extra" results for RIO2 and SEY2 in appendices. It will be easier for the reader to follow if all the results for each station are presented together. (See more details in the section with specific comments below)
>>>> we would like to keep these two particular cases in appendices so that the main body of the article focuses on two days. We have the feeling that it would break the dynamic of the article and make it more difficult to follow. We thank you for your comprehension.
Specific comments
In the abstract (and in line 30) you state that "microwave signals are almost unaffected by clouds but are delayed as they travel the troposphere". In many geodesy applications the induced effect on the delay by clouds are ignored because thy are comparable to the uncertainties. In your case however, you surpress the large variations in the estimated propagation delay by filtering and the question is then how the small scale variability in the delay caused by clouds affect your interpretations of the turbulence parameters. For example, cumulus clouds may cause delays of several millimetres. Solheim et al. (1999)( Propagation delays induced in GPS signals by dry air, water vapor, hydrometeors, and other particulates, J. Geophys. Res., 104, 9663–9670) state that "A cloud droplet concentration of 1 g/m^3 for a distance of 1 km has an integrated liquid value of 1 mm and would therefore induce a radio path delay of 1.45 mm." In Line 36 it is stated that the topic is on WV turbulence. Perhaps it will be more correct to refer to turbulence due to WV and liquid water clouds?
<>>>> Thank you for pointing out this reference; it is indeed very interesting. We have added it to the abstract/introduction. While we would like to retain the term 'WV,' we acknowledge that it may be more accurate to refer to 'WV and liquid water clouds.' This insight has inspired us to explore the parameters further, specifically comparing days with and without clouds or precipitation.
In Section 2.1 you mention briefly that the ZTD can be estimated from GNSS observation. I think this is the appropriate place to give the details on how this is done rather than in the paragraph starting on Line 120.
Line (L) 219: It is not clear to me why satellite orbit and clock errors can be ignored. Do you mean that the products from GFZ are free of error? This cannot be true. If you believe that these errors are small enough to be ignored, it needs to be motivated.
>>>>> We have changed it accordingly. Further, the GFZ satellite orbit and clock are estimated using a global GNSS network comprising approximately 140 stations. While these products are not error-free, the large number of stations ensures that high-frequency signals caused by turbulence have no significant impact on the satellite orbit and clock estimates. In other words, the orbit and clock products are very smooth, making them suitable for detecting high-frequency signals observed by individual stations.
Table 1: "Gravity waves" is not a type of climate. You may instead call it characterisation of GNSS station. Additionally, it is not really needed to present this in a table (with only one line as an entry). Each site can be described/characterised in the running text.
>>>> We have corrected it but would like to keep the table for clarity, i.e., as a summary
L 238-240: When checking the supplementary material I find that there is only one additional day for each site and season. From just reading the information available one cannot conclude that this material support your conclusions. As mentioned above you need more than one day of data to characterise the turbulence at a site.
>>>> We have corrected the text accordingly. This is not the topic to "characterise" the turbulence but rather to show that there is evident differences in pattern and range of values. In the future, we will need to be more specific about which kind of turbulence can be monitored with GNSS, using, e.g., large eddy simulation or gradient boosting to identify the main contributions (and potentially dependencies with height). It is only a measurement report in our opinion, giving an idea about what can be done.
Another question related to the supplementary material is that when I randomly checked some of the data files I find that the ground pressure is constant over the entire day in all cases. Does that mean that when you subtract the hydrostatic delay from the ZTD it has a constant value and that any variations in the hydrostatic delay will alias with the wet delay variability? This needs to be explained / discussed.
>>>>> The pressure is from GTP2 model. The daily variation of the pressure is slow and usually within several hpa. The constant pressure can introduce an error with low frequency in ZWD, but should have no impact by detecting turbulent signal. We have added this information in the text describing the computation.
L 400-404: You have not presented any results for the PAYN station, so there is no need to discuss such results. If it is of relevance for this study I think it shall be included rather than referring to the "next contribution".
>>>> you are right. We have added a more general sentence.
L 420: Also here the "next contribution" is mentioned. It is sufficient say that confirmation is needed because the reader will not know where the next contribution is to be found.
>>>> we have corrected it accordingly. Technical Corrections
A general comment: At many places you use the wording "in this contribution" (or something similar). In most cases these words should be deleted. It is obvious, e.g. L 8, 55, 178, 184, 215-216, 240, and 380. >>>> We have deleted most of them.
L 3: 90% --> 90 % (Also at many other places: insert a "space" between the value and the unit (SI recommendation). Also there shall be no dash between the value and the unit.) L 32: You probably mean hydrostatic delay (not dry), which is the term you use elsewhere?
>>>> in our opinion, this is the same L 48: Why is the slope not mentioned here?
>>> because the slope is not estimated in this contribution. Figure 1: Correct the labels on the y axis in the ZWD' graph (too close for a good readability). L 104: Better to write the mathematical expression in one line, or make it a numbered equation. L 106: Units shall not be in italics. L 109: 1 hour --> 1 h L 121: 30-second rate --> 30 s rate L 126: 4h --> 4 h UT L 170: Better to write the mathematical expression in one line, or make it a numbered equation. Figure 2: The time series graphs are too small, make them bigger or delete them (they are not really needed)?
>>>> we would like to keep them as they servie as illustration. We mention it in the caption now. L 220, 221, 222, 223, 226: shall read 30 s, 1 h, 24 h L 284: Do not start a sentence with a symbol. L 318 & 326. RIA2 --> RIO2 L 411: 375 unit? L 422: automn --> autumn
>>>> corrected when appropriate
Citation: https://doi.org/10.5194/egusphere-2024-2680-AC1
-
AC1: 'Reply on RC2', gael kermarrec, 23 Nov 2024
-
RC2: 'Comment on egusphere-2024-2680', Anonymous Referee #1, 25 Nov 2024
The paper describes a new method for analyzing atmospheric turbulence using ZWD estimated from GNSS. The method first filters the ZWD time series, removing the long-period variations. Then parameters related to atmospheric turbulence are fitted to the remaining short-term fluctuations. The method is demonstrated by applying it to two days (one summer and one winter day) of data from five different GNSS stations in different climate regions. I think the method is rather interesting and can be a useful tool for studying atmospheric turbulence.
It would have been nice to have some kind of external validation of the method. Right now, there are no definitive evidence presented that the method actually works, i.e., that it senses atmospheric turbulence.
A weakness of the paper is that the method is tested only on five stations, and two days of data per station. This is a too small sample to draw any definitive conclusions. I could understand that the authors wanted to look at individual days and compare the ZWD variations seen to the turbulence parameters and see if any expected correspondence could be found. However, I think no definitive conclusion could be drawn from that comparison. Perhaps it would be better to (also) look at longer time series and mor stations (several from the same climate region).
In several figures, e.g., Figs, 4, 6, and 8 there are negative ZWD, while we would expect ZWD to be positive. Looking at the raw data (https://doi.org/10.25835/HCC01FRE) it seems that what is plotted as ZWD is not actually the ZWD, but the estimated correction to the a priori (COR_WET). Since the a priori ZWD and ZHD seems to be constant, this is actually the correction to the ZTD, not only the ZWD. I guess for the derived turbulence parameters it does not matter since variations in ZHD will be mostly for longer time scales. However, when the authors comment on the variations of the ZWD for the various stations and days, it might be important to know if the variations seen are in ZHD (i.e. pressure) or ZWD (water vapour) to draw the correct conclusions. Nevertheless, the quantity should be labelled correctly (i.e. correction to ZTD instead of ZWD).
In figure A1 it seems strange that the “ZWD” at the end of doy 134 is quite different from the value at the beginning of doy 135. The values at the end of doy 135 and beginning of doy 136 also differ significantly.
In the GNSS data processing, what GNSS were used? I find no information about this.
What software was used. The paper says Bernes GNSS Software, the supplementary material (https://doi.org/10.25835/HCC01FRE) says EPOS software.
Line 30: “..high-rate GNSS are unaffected by clouds”. In fact, this is also true for non-high-rate GNSS. Furthermore, the GNSS signals are actually affected by clouds, but this effect is small and typically not explicitly considered.
Line 33: “dry delay” -> “hydrostatic delay”
Line 78: “κ′ a factor depending on the surface temperature and specific gas constant”. It should perhaps be pointed out that the Bevis formula is only an approximation.
Line 218-220: “In the GNSS data processing we used the GeoForschungs Zentrum multi-GNSS satellites orbit and clock product in Precise Point Positioning (PPP) mode. Thus, the error of satellites orbit and clock can be ignored in this study.” I do not agree that errors in satellite orbits and clocks can be ignored. Since PPP are used, the errors in orbits and clocks will affect the estimated ZWD. I guess the GFZ products are of high quality, hence the error will be small, but maybe not negligible.
Line 230-232: “Effects from, e.g., nontidal loading, multipath, or antenna center phase variations are expected to be present in the residuals of the positioning adjustment and not in the ZWD, which is estimated as an independent delay.” I do not agree. Perhaps non-tidal loading will mostly go into the position estimates (although there might be variations on periods shorter than one day which the daily estimates will not catch). However, I think multipath and antenna phase center variations will affect both ZWD and coordinates.
Line 306: “contrary to what we observed for the station UNB3”. What is observed is not exactly contrary to UNB3. On UNB3 a ZWD increase and sigma2 decrease was observed, for NYA2 there is a ZWD decrease and a sigma2 increase. Seems to be consistent to me.
Line 410: “For comparison, we also computed the days DOY 135 and 136”. Should it be 134 and 136 (135 was the day with gravity waves)?
Line 414: “DOY 169”. This day is not show in the figure.
Line 414-415: “nearly constant with a strong sinus-like shape”. Strange formulation. It is not constant if it has a sinus shape.
Citation: https://doi.org/10.5194/egusphere-2024-2680-RC2 -
AC2: 'Comment on egusphere-2024-2680', gael kermarrec, 27 Nov 2024
The paper describes a new method for analyzing atmospheric turbulence using ZWD estimated from GNSS. The method first filters the ZWD time series, removing the long-period variations. Then parameters related to atmospheric turbulence are fitted to the remaining short-term fluctuations. The method is demonstrated by applying it to two days (one summer and one winter day) of data from five different GNSS stations in different climate regions. I think the method is rather interesting and can be a useful tool for studying atmospheric turbulence.
>>>> Thank you for your positive comments on our manuscript. It indeed introduces a novel parameter for studying atmospheric turbulence, and we hope it will inspire further exploration of its dependencies and potential as a new metric for atmospheric studies. This paper serves as an introduction, aiming to provide insights and spark curiosity. It is not intended as an in-depth analysis of the parameter; such work will follow in future research, where it will be compared (rather than validated, as it is a new quantity and may not be comparable) with real measurements, such as those from lidars or radiometers. Initial results are already very promising and will be shared in due course.
It would have been nice to have some kind of external validation of the method. Right now, there are no definitive evidence presented that the method actually works, i.e., that it senses atmospheric turbulence.
>>>>The fact that we observe the spectrum predicted by turbulence theory, particularly the slopes from both visual and statistical perspectives, is not definitive evidence but a strong indication that our approach is likely correct. Wheelon's book supports this understanding. While validation will be very challenging, the use of machine learning could help uncover key dependencies in the data—for instance, determining the height of WV turbulence or the role of clouds. There are many exciting topics to explore, and this article represents a first step in that direction. In our opinion, introducing measurements at this stage would have been nearly overwhelming. Furthermore, it is submitted as a 'measurement report.' If you are interested, we also conducted an initial comparison with OLCI data in a subsequent contribution, which shows excellent agreement between both approach (spatial/temporal):
G. Kermarrec, X. Calbet, Z. Deng, C. Carbajal Henken, R. Preusker, "Retrieval of water vapor in the atmosphere and its spectral content: from OLCI to GPS," Proc. SPIE 12730, Remote Sensing of Clouds and the Atmosphere XXVIII, 127300F (19 October 2023); https://doi.org/10.1117/12.2678381
A weakness of the paper is that the method is tested only on five stations, and two days of data per station. This is a too small sample to draw any definitive conclusions.
>>>> You are correct, and we also avoid drawing definitive or general conclusions. We have slightly rewritten the end of the introduction to clarify this point as well as the conclusion.
I could understand that the authors wanted to look at individual days and compare the ZWD variations seen to the turbulence parameters and see if any expected correspondence could be found. However, I think no definitive conclusion could be drawn from that comparison. Perhaps it would be better to (also) look at longer time series and mor stations (several from the same climate region).
>>>> We agree with you, but addressing this fully would likely require a book rather than an introductory article. The fact that you raise this question shows that we have achieved our goal: the reader is engaged, wants to know more, and has understood the problem.
In several figures, e.g., Figs, 4, 6, and 8 there are negative ZWD, while we would expect ZWD to be positive. Looking at the raw data (https://doi.org/10.25835/HCC01FRE) it seems that what is plotted as ZWD is not actually the ZWD, but the estimated correction to the a priori (COR_WET). Since the a priori ZWD and ZHD seems to be constant, this is actually the correction to the ZTD, not only the ZWD. I guess for the derived turbulence parameters it does not matter since variations in ZHD will be mostly for longer time scales. However, when the authors comment on the variations of the ZWD for the various stations and days, it might be important to know if the variations seen are in ZHD (i.e. pressure) or ZWD (water vapour) to draw the correct conclusions. Nevertheless, the quantity should be labelled correctly (i.e. correction to ZTD instead of ZWD).
>>>>Yes, in the GNSS data analysis we estimated the ZWD corrections (COR_WET) to the a priori constant ZWD. Those ZWD corrections are plotted in the figures and are used to derive the turbulence parameters in this work. The slow variation ZHD and ZWD should have no impact. We have a note in the corresponding (2.2) as well as in 3.1 and all captions for clarity.
In figure A1 it seems strange that the “ZWD” at the end of doy 134 is quite different from the value at the beginning of doy 135. The values at the end of doy 135 and beginning of doy 136 also differ significantly.
>>>>>The 24-hour GNSS observations are used to estimate the 30-second ZWD. Since the ZWD is highly correlated with other estimated parameters (such as coordinates and clock error), boundary effects can be observed in the estimated ZWD parameters. To mitigate these boundary errors, a sliding window analysis could be considered. However, this lies beyond the scope of our paper and will be addressed in future work. We have correspondingly accounted for that fact in the processing to compute our parameters by excluding the first batches. In section 3, we have added a note to point out these effects.
In the GNSS data processing, what GNSS were used? I find no information about this.
>>>>>Thank you for the questions. We used all the GNSS satellites from the GFZ multi-GNSS products, which is includes GPS, GLONASS, Galileo, BeiDou and QZSS. It is specifed in the text now (section 3)
What software was used. The paper says Bernes GNSS Software, the supplementary material (https://doi.org/10.25835/HCC01FRE) says EPOS software.
>>>> it is the EPOS software from Gfz Potsdam, sorry for the mistake.
Line 30: “..high-rate GNSS are unaffected by clouds”. In fact, this is also true for non-high-rate GNSS. Furthermore, the GNSS signals are actually affected by clouds, but this effect is small and typically not explicitly considered.
>>>> You are correct, and we have become even more specific thanks to Reviewer 2, who pointed out the reference: Solheim et al. (1999) ("Propagation delays induced in GPS signals by dry air, water vapor, hydrometeors, and other particulates," J. Geophys. Res., 104, 9663–9670). The paper states that "A cloud droplet concentration of 1 g/m³ for a distance of 1 km has an integrated liquid value of 1 mm and would therefore induce a radio path delay of 1.45 mm." In the first version of the article, we mentioned that the topic focuses on WV turbulence. In the introduction, we now clarify that it refers to turbulence caused by both WV and liquid water clouds, but we "simplify" it to WV for readability.
Line 33: “dry delay” -> “hydrostatic delay”
>>>> in our opinion, this should be the same.
Line 78: “κ′ a factor depending on the surface temperature and specific gas constant”. It should perhaps be pointed out that the Bevis formula is only an approximation.
>>>> ok, we have added that it is an approximation in the text.
Line 218-220: “In the GNSS data processing we used the GeoForschungs Zentrum multi-GNSS satellites orbit and clock product in Precise Point Positioning (PPP) mode. Thus, the error of satellites orbit and clock can be ignored in this study.” I do not agree that errors in satellite orbits and clocks can be ignored. Since PPP are used, the errors in orbits and clocks will affect the estimated ZWD. I guess the GFZ products are of high quality, hence the error will be small, but maybe not negligible.
>>>>The GFZ satellite orbit and clock are estimated using a global GNSS network of approximately 140 stations. While not error-free, the large number of stations ensures that high-frequency signals caused by turbulence do not significantly affect the satellite orbit and clock estimates. In other words, these products are very smooth, making them suitable for detecting high-frequency signals observed by individual stations.
Line 230-232: “Effects from, e.g., nontidal loading, multipath, or antenna center phase variations are expected to be present in the residuals of the positioning adjustment and not in the ZWD, which is estimated as an independent delay.” I do not agree. Perhaps non-tidal loading will mostly go into the position estimates (although there might be variations on periods shorter than one day which the daily estimates will not catch). However, I think multipath and antenna phase center variations will affect both ZWD and coordinates.
>>>>>The antenna center phase variations are corrected with the absolute antenna calibrations in the IGS20 frame during the GNSS data processing. The multipath can cause usually a error with slow variation in the ZWD estimations. We have added this in section 3. One should mention that we focus on a very specific high frequency noise, and that the fact that we search for a very particular value of the psd slope should prevent us from multipath effect.
Line 306: “contrary to what we observed for the station UNB3”. What is observed is not exactly contrary to UNB3. On UNB3 a ZWD increase and sigma2 decrease was observed, for NYA2 there is a ZWD decrease and a sigma2 increase. Seems to be consistent to me.
>>>> you are right, it is corrected accordingly in the new version.
Line 410: “For comparison, we also computed the days DOY 135 and 136”. Should it be 134 and 136 (135 was the day with gravity waves)?
>>>> you re right, thanks for pointing that out.
Line 414: “DOY 169”. This day is not show in the figure.
>>>> sorry for the mistake, it has been corrected.
Line 414-415: “nearly constant with a strong sinus-like shape”. Strange formulation. It is not constant if it has a sinus shape.
>>>> we have corrected to "a light wavy shape". indeed, this was not consistent.
Citation: https://doi.org/10.5194/egusphere-2024-2680-AC2
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-2680', Anonymous Referee #2, 19 Nov 2024
General Comments
The manuscript describes a method to derive turbulence parameters from observations by ground-based GNSS stations. These observations are processed to obtain time series of the equivalent total zenith propagation delay. After a filtering process a time series of estimated variations in the equivalent zenith wet delay are used to characterise turbulence. I find the work interesting and unique, but have some thoughts concerning a critical assessment of the results and the structure and the clarity of presentation.
A first reflection is why the manuscript was submitted to ACP rather than AMT (Atmospheric Measurement Techniques). I find that the method, rather than the estimated turbulence parameters, is the motivation for publishing (and the title also starts with "Measurement Report:". I assume this is a question for the editor.
The time series of the Zenith Wet Delay (ZWD) from the GNSS data processing has a temporal resolution of 30 s. It is stated that no constraint is applied (line 122) but the estimates of the ZWD are averages of the air in the sampled volume. This volume depends on the elevation cutoff angle and which GNSS that were used. I miss information about this. If a frozen flow is assumed (as you do in the manuscript) and an elevation cutoff angle of say 5 or 10 degrees is used, quite a high wind speed is required in order to have independent air volumes with a sample period of 30 s. This issue should be discussed and how the results are affected by this fact. Depending on the conclusions you may want to reduce the temporal resolution.
A related issue is that there is a need to validate the GNSS wet delay time series obtained with the method you describe. One possibility is the use of microwave radiometry. The main advantage of the radiometer is that it samples a much smaller volume and it can measure in the same direction continuously. (Also in this case, however, the temporal resolution for independent samples is limited by the sampled air volume, which in turn is determined by the antenna beam width(s).) Probably none of the sites you study is equipped with a radiometer. However, there are several GNSS/VLBI sites that offer this possibility, see e.g. Teke et al. (2013) (Troposphere delays from space geodetic techniques, water vapor radiometers, and numerical weather models over a series of continuous VLBI campaigns, J. Geod.,87:981–1001, DOI 10.1007/s00190-013-0662-z).
You present results for one summer and one winter day for each GNSS station. I find this to be insufficient. If I may be a bit provokative, but a data point in a climate time series is often an average over 30 years. I have not visited all the sites studied but I have been in similar climates and my experience is that the weather conditions may change considerably from one day to another (possibly excluding the tropical site). I do not argue that you shall use data from several years, but because you spend much effort on comparing the results from the different sites there is a high risk for overinterpretation. One month, or at least one week, of data from each site and season will make more sense. I think it will also be of interest to compare the parameters estimated for adjacent days.
A related matter is that the argument that the spring and the autumn may be slightly different for the tropical station (SEY2) is definitely valid for all the stations. Consequently I recommend to handle all the stations in the same way. Furthermore, to me it does not make sense to present the "extra" results for RIO2 and SEY2 in appendices. It will be easier for the reader to follow if all the results for each station are presented together. (See more details in the section with specific comments below)
Specific comments
In the abstract (and in line 30) you state that "microwave signals are almost unaffected by clouds but are delayed as they travel the troposphere". In many geodesy applications the induced effect on the delay by clouds are ignored because thy are comparable to the uncertainties. In your case however, you surpress the large variations in the estimated propagation delay by filtering and the question is then how the small scale variability in the delay caused by clouds affect your interpretations of the turbulence parameters. For example, cumulus clouds may cause delays of several millimetres. Solheim et al. (1999)( Propagation delays induced in GPS signals by dry air, water vapor, hydrometeors, and other particulates, J. Geophys. Res., 104, 9663–9670) state that "A cloud droplet concentration of 1 g/m^3 for a distance of 1 km has an integrated liquid value of 1 mm and would therefore induce a radio path delay of 1.45 mm." In Line 36 it is stated that the topic is on WV turbulence. Perhaps it will be more correct to refer to turbulence due to WV and liquid water clouds?
In Section 2.1 you mention briefly that the ZTD can be estimated from GNSS observation. I think this is the appropriate place to give the details on how this is done rather than in the paragraph starting on Line 120.
Line (L) 219: It is not clear to me why satellite orbit and clock errors can be ignored. Do you mean that the products from GFZ are free of error? This cannot be true. If you believe that these errors are small enough to be ignored, it needs to be motivated.
Table 1: "Gravity waves" is not a type of climate. You may instead call it characterisation of GNSS station. Additionally, it is not really needed to present this in a table (with only one line as an entry). Each site can be described/characterised in the running text.
L 238-240: When checking the supplementary material I find that there is only one additional day for each site and season. From just reading the information available one cannot conclude that this material support your conclusions. As mentioned above you need more than one day of data to characterise the turbulence at a site.
Another question related to the supplementary material is that when I randomly checked some of the data files I find that the ground pressure is constant over the entire day in all cases. Does that mean that when you subtract the hydrostatic delay from the ZTD it has a constant value and that any variations in the hydrostatic delay will alias with the wet delay variability? This needs to be explained / discussed.
L 400-404: You have not presented any results for the PAYN station, so there is no need to discuss such results. If it is of relevance for this study I think it shall be included rather than referring to the "next contribution".
L 420: Also here the "next contribution" is mentioned. It is sufficient say that confirmation is needed because the reader will not know where the next contribution is to be found.
Technical CorrectionsA general comment: At many places you use the wording "in this contribution" (or something similar). In most cases these words should be deleted. It is obvious, e.g. L 8, 55, 178, 184, 215-216, 240, and 380.
L 3: 90% --> 90 % (Also at many other places: insert a "space" between the value and the unit (SI recommendation). Also there shall be no dash between the value and the unit.)
L 32: You probably mean hydrostatic delay (not dry), which is the term you use elsewhere?
L 48: Why is the slope not mentioned here?
Figure 1: Correct the labels on the y axis in the ZWD' graph (too close for a good readability).
L 104: Better to write the mathematical expression in one line, or make it a numbered equation.
L 106: Units shall not be in italics.
L 109: 1 hour --> 1 h
L 121: 30-second rate --> 30 s rate
L 126: 4h --> 4 h UT
L 170: Better to write the mathematical expression in one line, or make it a numbered equation.
Figure 2: The time series graphs are too small, make them bigger or delete them (they are not really needed)?
L 220, 221, 222, 223, 226: shall read 30 s, 1 h, 24 h
L 284: Do not start a sentence with a symbol.
L 318 & 326. RIA2 --> RIO2
L 411: 375 unit?
L 422: automn --> autumnCitation: https://doi.org/10.5194/egusphere-2024-2680-RC1 -
AC1: 'Reply on RC2', gael kermarrec, 23 Nov 2024
General Comments
The manuscript describes a method to derive turbulence parameters from observations by ground-based GNSS stations. These observations are processed to obtain time series of the equivalent total zenith propagation delay. After a filtering process a time series of estimated variations in the equivalent zenith wet delay are used to characterise turbulence. I find the work interesting and unique, but have some thoughts concerning a critical assessment of the results and the structure and the clarity of presentation.
>>>> Many thanks for your positive comments on the manuscript, which help us improve it.
A first reflection is why the manuscript was submitted to ACP rather than AMT (Atmospheric Measurement Techniques). I find that the method, rather than the estimated turbulence parameters, is the motivation for publishing (and the title also starts with "Measurement Report:". I assume this is a question for the editor.
>>>> The article was intentionally submitted to ACP because it offers the opportunity to publish 'measurement reports.' In our opinion, a publication in AMT would require a longer review process. By publishing here, we have the chance to present preliminary results and outline the methodology in one paper, ensuring we don’t need to repeat these steps in future publications.
The time series of the Zenith Wet Delay (ZWD) from the GNSS data processing has a temporal resolution of 30 s. It is stated that no constraint is applied (line 122) but the estimates of the ZWD are averages of the air in the sampled volume.
>>>> By 'constraint,' we mean that the ZWD is not assumed to follow a specific noise structure (e.g., a random walk). Assuming such a structure could bias our results, which are based on identifying slopes in the power spectrum.
This volume depends on the elevation cutoff angle and which GNSS that were used. I miss information about this. If a frozen flow is assumed (as you do in the manuscript) and an elevation cutoff angle of say 5 or 10 degrees is used, quite a high wind speed is required in order to have independent air volumes with a sample period of 30 s.
>>>> The ZWD values are mapped as 'slant wet delays' from all GNSS satellites to the vertical. While the frozen approximation is clearly an approximation but well documented in Wheelon (2001), we are confident in its validity for two main reasons: (i) we observe the expected slope in real data using both visual and statistical methods, and (ii) further investigations using machine learning strategies—combining real data from radiometers and other sensors at PAYN—indicate that the atmospheric layers sensed by the ZWD are at an altitude of 1500–2000 m, where the geostrophic wind predominates (study to be published in a next step). This point is further discussed in the corresponding section on the Taylor Frozen hypothesis and in the introduction. Additionally, we have enhanced the description of how the ZWD is computed.
This issue should be discussed and how the results are affected by this fact. Depending on the conclusions you may want to reduce the temporal resolution.
>>>>> This could be done, but not with the stations we selected (frequency is not available). Using a lower temporal resolution introduces further noise, which would require more advanced filtering strategies. We plan to address this in a future contribution on the topic. Once more, we derive the methodology with some examples.
A related issue is that there is a need to validate the GNSS wet delay time series obtained with the method you describe. One possibility is the use of microwave radiometry. The main advantage of the radiometer is that it samples a much smaller volume and it can measure in the same direction continuously. (Also in this case, however, the temporal resolution for independent samples is limited by the sampled air volume, which in turn is determined by the antenna beam width(s).) Probably none of the sites you study is equipped with a radiometer.
>>>> Our aim is not to validate the ZWD, as this has been extensively addressed in many other contributions. Instead, we focus on the short-term variations related to turbulence. As you correctly noted, the ZWD is a 'particular' parameter, representing an integrated quantity along the signal path and a time series, whereas satellite images, such as those from OLCI, derive spatial quantities (e.g., 'outer scale length'). In our view, it is unlikely that an instrument will be developed that allows direct comparison and that is the reason why we would carry on with machine learning (gradient boosting) strategies to combine several instruments for validation/insvestigations. This uniqueness makes the parameters we estimate both novel and innovative. The article highlights the need for further investigation to better understand these parameters, such as through large eddy simulations or the aforementioned gradient boosting. While we do not claim to hold the ultimate truth, our goal is to spark curiosity and foster further exploration. We have add a section "note" in the taylor frozen section to comment on that and further related topics.
However, there are several GNSS/VLBI sites that offer this possibility, see e.g. Teke et al. (2013) (Troposphere delays from space geodetic techniques, water vapor radiometers, and numerical weather models over a series of continuous VLBI campaigns, J. Geod.,87:981–1001, DOI 10.1007/s00190-013-0662-z).
>>>> thanks for point out these stations. We will keep that in mind and are currently making further work on PAYN.
You present results for one summer and one winter day for each GNSS station. I find this to be insufficient. If I may be a bit provokative, but a data point in a climate time series is often an average over 30 years.
>>>> I also work on climate time series and completely agree with your point. However, we are not claiming in the article to draw climatological conclusions. Instead, we demonstrate that the parameters we estimate appear to be related to the climate zone, potentially offering new insights. We are also not suggesting that the same parameters or identical patterns will emerge every day. The values naturally vary, as turbulence changes from minute to minute. However, a periodical patterns for some stations seem evident.
Although I have not visited all the studied sites, I have been in similar climates, and in my experience, weather conditions can vary significantly from one day to the next (with the possible exception of the tropical site).
I fully understand your concern about the need for more extended data collection. While we are not arguing against the value of multi-year data, given the considerable effort we put into comparing results across sites, we recognize the risk of overinterpretation. Collecting at least one month—or at minimum, one week—of data from each site across different seasons would indeed provide a more robust basis for analysis. Additionally, comparing parameters estimated for adjacent days could offer valuable insights.
>>>> We did conduct such investigations but chose not to include them in this article. The reason is that presenting multiple days could be confusing, as readers might focus on finding 'exactly the same' patterns (or not), leading to additional questions such as 'why only two days,' and so on. Instead, we restricted our analysis to one day to highlight the differences in patterns between seasons and stations. This approach also makes it easier to present the results (figure are not overloaded), as the goal of this work is to demonstrate the potential of the parameters (as a report) rather than conduct a deep investigation.
For a more comprehensive analysis, we would consider data from a single station over 30 years, incorporating techniques like moving averages and other methods. Understanding these parameters is a complex and challenging task that requires significant effort. However, in the revised version of the article, we have added some comments regarding adjacent days. We have add futher days in the repository.
A related matter is that the argument that the spring and the autumn may be slightly different for the tropical station (SEY2) is definitely valid for all the stations.
>>>> you are right but we are intented to think that the differences should be more extrem for a tropical station than for the other one.
Consequently I recommend to handle all the stations in the same way. Furthermore, to me it does not make sense to present the "extra" results for RIO2 and SEY2 in appendices. It will be easier for the reader to follow if all the results for each station are presented together. (See more details in the section with specific comments below)
>>>> we would like to keep these two particular cases in appendices so that the main body of the article focuses on two days. We have the feeling that it would break the dynamic of the article and make it more difficult to follow. We thank you for your comprehension.
Specific comments
In the abstract (and in line 30) you state that "microwave signals are almost unaffected by clouds but are delayed as they travel the troposphere". In many geodesy applications the induced effect on the delay by clouds are ignored because thy are comparable to the uncertainties. In your case however, you surpress the large variations in the estimated propagation delay by filtering and the question is then how the small scale variability in the delay caused by clouds affect your interpretations of the turbulence parameters. For example, cumulus clouds may cause delays of several millimetres. Solheim et al. (1999)( Propagation delays induced in GPS signals by dry air, water vapor, hydrometeors, and other particulates, J. Geophys. Res., 104, 9663–9670) state that "A cloud droplet concentration of 1 g/m^3 for a distance of 1 km has an integrated liquid value of 1 mm and would therefore induce a radio path delay of 1.45 mm." In Line 36 it is stated that the topic is on WV turbulence. Perhaps it will be more correct to refer to turbulence due to WV and liquid water clouds?
<>>>> Thank you for pointing out this reference; it is indeed very interesting. We have added it to the abstract/introduction. While we would like to retain the term 'WV,' we acknowledge that it may be more accurate to refer to 'WV and liquid water clouds.' This insight has inspired us to explore the parameters further, specifically comparing days with and without clouds or precipitation.
In Section 2.1 you mention briefly that the ZTD can be estimated from GNSS observation. I think this is the appropriate place to give the details on how this is done rather than in the paragraph starting on Line 120.
Line (L) 219: It is not clear to me why satellite orbit and clock errors can be ignored. Do you mean that the products from GFZ are free of error? This cannot be true. If you believe that these errors are small enough to be ignored, it needs to be motivated.
>>>>> We have changed it accordingly. Further, the GFZ satellite orbit and clock are estimated using a global GNSS network comprising approximately 140 stations. While these products are not error-free, the large number of stations ensures that high-frequency signals caused by turbulence have no significant impact on the satellite orbit and clock estimates. In other words, the orbit and clock products are very smooth, making them suitable for detecting high-frequency signals observed by individual stations.
Table 1: "Gravity waves" is not a type of climate. You may instead call it characterisation of GNSS station. Additionally, it is not really needed to present this in a table (with only one line as an entry). Each site can be described/characterised in the running text.
>>>> We have corrected it but would like to keep the table for clarity, i.e., as a summary
L 238-240: When checking the supplementary material I find that there is only one additional day for each site and season. From just reading the information available one cannot conclude that this material support your conclusions. As mentioned above you need more than one day of data to characterise the turbulence at a site.
>>>> We have corrected the text accordingly. This is not the topic to "characterise" the turbulence but rather to show that there is evident differences in pattern and range of values. In the future, we will need to be more specific about which kind of turbulence can be monitored with GNSS, using, e.g., large eddy simulation or gradient boosting to identify the main contributions (and potentially dependencies with height). It is only a measurement report in our opinion, giving an idea about what can be done.
Another question related to the supplementary material is that when I randomly checked some of the data files I find that the ground pressure is constant over the entire day in all cases. Does that mean that when you subtract the hydrostatic delay from the ZTD it has a constant value and that any variations in the hydrostatic delay will alias with the wet delay variability? This needs to be explained / discussed.
>>>>> The pressure is from GTP2 model. The daily variation of the pressure is slow and usually within several hpa. The constant pressure can introduce an error with low frequency in ZWD, but should have no impact by detecting turbulent signal. We have added this information in the text describing the computation.
L 400-404: You have not presented any results for the PAYN station, so there is no need to discuss such results. If it is of relevance for this study I think it shall be included rather than referring to the "next contribution".
>>>> you are right. We have added a more general sentence.
L 420: Also here the "next contribution" is mentioned. It is sufficient say that confirmation is needed because the reader will not know where the next contribution is to be found.
>>>> we have corrected it accordingly. Technical Corrections
A general comment: At many places you use the wording "in this contribution" (or something similar). In most cases these words should be deleted. It is obvious, e.g. L 8, 55, 178, 184, 215-216, 240, and 380. >>>> We have deleted most of them.
L 3: 90% --> 90 % (Also at many other places: insert a "space" between the value and the unit (SI recommendation). Also there shall be no dash between the value and the unit.) L 32: You probably mean hydrostatic delay (not dry), which is the term you use elsewhere?
>>>> in our opinion, this is the same L 48: Why is the slope not mentioned here?
>>> because the slope is not estimated in this contribution. Figure 1: Correct the labels on the y axis in the ZWD' graph (too close for a good readability). L 104: Better to write the mathematical expression in one line, or make it a numbered equation. L 106: Units shall not be in italics. L 109: 1 hour --> 1 h L 121: 30-second rate --> 30 s rate L 126: 4h --> 4 h UT L 170: Better to write the mathematical expression in one line, or make it a numbered equation. Figure 2: The time series graphs are too small, make them bigger or delete them (they are not really needed)?
>>>> we would like to keep them as they servie as illustration. We mention it in the caption now. L 220, 221, 222, 223, 226: shall read 30 s, 1 h, 24 h L 284: Do not start a sentence with a symbol. L 318 & 326. RIA2 --> RIO2 L 411: 375 unit? L 422: automn --> autumn
>>>> corrected when appropriate
Citation: https://doi.org/10.5194/egusphere-2024-2680-AC1
-
AC1: 'Reply on RC2', gael kermarrec, 23 Nov 2024
-
RC2: 'Comment on egusphere-2024-2680', Anonymous Referee #1, 25 Nov 2024
The paper describes a new method for analyzing atmospheric turbulence using ZWD estimated from GNSS. The method first filters the ZWD time series, removing the long-period variations. Then parameters related to atmospheric turbulence are fitted to the remaining short-term fluctuations. The method is demonstrated by applying it to two days (one summer and one winter day) of data from five different GNSS stations in different climate regions. I think the method is rather interesting and can be a useful tool for studying atmospheric turbulence.
It would have been nice to have some kind of external validation of the method. Right now, there are no definitive evidence presented that the method actually works, i.e., that it senses atmospheric turbulence.
A weakness of the paper is that the method is tested only on five stations, and two days of data per station. This is a too small sample to draw any definitive conclusions. I could understand that the authors wanted to look at individual days and compare the ZWD variations seen to the turbulence parameters and see if any expected correspondence could be found. However, I think no definitive conclusion could be drawn from that comparison. Perhaps it would be better to (also) look at longer time series and mor stations (several from the same climate region).
In several figures, e.g., Figs, 4, 6, and 8 there are negative ZWD, while we would expect ZWD to be positive. Looking at the raw data (https://doi.org/10.25835/HCC01FRE) it seems that what is plotted as ZWD is not actually the ZWD, but the estimated correction to the a priori (COR_WET). Since the a priori ZWD and ZHD seems to be constant, this is actually the correction to the ZTD, not only the ZWD. I guess for the derived turbulence parameters it does not matter since variations in ZHD will be mostly for longer time scales. However, when the authors comment on the variations of the ZWD for the various stations and days, it might be important to know if the variations seen are in ZHD (i.e. pressure) or ZWD (water vapour) to draw the correct conclusions. Nevertheless, the quantity should be labelled correctly (i.e. correction to ZTD instead of ZWD).
In figure A1 it seems strange that the “ZWD” at the end of doy 134 is quite different from the value at the beginning of doy 135. The values at the end of doy 135 and beginning of doy 136 also differ significantly.
In the GNSS data processing, what GNSS were used? I find no information about this.
What software was used. The paper says Bernes GNSS Software, the supplementary material (https://doi.org/10.25835/HCC01FRE) says EPOS software.
Line 30: “..high-rate GNSS are unaffected by clouds”. In fact, this is also true for non-high-rate GNSS. Furthermore, the GNSS signals are actually affected by clouds, but this effect is small and typically not explicitly considered.
Line 33: “dry delay” -> “hydrostatic delay”
Line 78: “κ′ a factor depending on the surface temperature and specific gas constant”. It should perhaps be pointed out that the Bevis formula is only an approximation.
Line 218-220: “In the GNSS data processing we used the GeoForschungs Zentrum multi-GNSS satellites orbit and clock product in Precise Point Positioning (PPP) mode. Thus, the error of satellites orbit and clock can be ignored in this study.” I do not agree that errors in satellite orbits and clocks can be ignored. Since PPP are used, the errors in orbits and clocks will affect the estimated ZWD. I guess the GFZ products are of high quality, hence the error will be small, but maybe not negligible.
Line 230-232: “Effects from, e.g., nontidal loading, multipath, or antenna center phase variations are expected to be present in the residuals of the positioning adjustment and not in the ZWD, which is estimated as an independent delay.” I do not agree. Perhaps non-tidal loading will mostly go into the position estimates (although there might be variations on periods shorter than one day which the daily estimates will not catch). However, I think multipath and antenna phase center variations will affect both ZWD and coordinates.
Line 306: “contrary to what we observed for the station UNB3”. What is observed is not exactly contrary to UNB3. On UNB3 a ZWD increase and sigma2 decrease was observed, for NYA2 there is a ZWD decrease and a sigma2 increase. Seems to be consistent to me.
Line 410: “For comparison, we also computed the days DOY 135 and 136”. Should it be 134 and 136 (135 was the day with gravity waves)?
Line 414: “DOY 169”. This day is not show in the figure.
Line 414-415: “nearly constant with a strong sinus-like shape”. Strange formulation. It is not constant if it has a sinus shape.
Citation: https://doi.org/10.5194/egusphere-2024-2680-RC2 -
AC2: 'Comment on egusphere-2024-2680', gael kermarrec, 27 Nov 2024
The paper describes a new method for analyzing atmospheric turbulence using ZWD estimated from GNSS. The method first filters the ZWD time series, removing the long-period variations. Then parameters related to atmospheric turbulence are fitted to the remaining short-term fluctuations. The method is demonstrated by applying it to two days (one summer and one winter day) of data from five different GNSS stations in different climate regions. I think the method is rather interesting and can be a useful tool for studying atmospheric turbulence.
>>>> Thank you for your positive comments on our manuscript. It indeed introduces a novel parameter for studying atmospheric turbulence, and we hope it will inspire further exploration of its dependencies and potential as a new metric for atmospheric studies. This paper serves as an introduction, aiming to provide insights and spark curiosity. It is not intended as an in-depth analysis of the parameter; such work will follow in future research, where it will be compared (rather than validated, as it is a new quantity and may not be comparable) with real measurements, such as those from lidars or radiometers. Initial results are already very promising and will be shared in due course.
It would have been nice to have some kind of external validation of the method. Right now, there are no definitive evidence presented that the method actually works, i.e., that it senses atmospheric turbulence.
>>>>The fact that we observe the spectrum predicted by turbulence theory, particularly the slopes from both visual and statistical perspectives, is not definitive evidence but a strong indication that our approach is likely correct. Wheelon's book supports this understanding. While validation will be very challenging, the use of machine learning could help uncover key dependencies in the data—for instance, determining the height of WV turbulence or the role of clouds. There are many exciting topics to explore, and this article represents a first step in that direction. In our opinion, introducing measurements at this stage would have been nearly overwhelming. Furthermore, it is submitted as a 'measurement report.' If you are interested, we also conducted an initial comparison with OLCI data in a subsequent contribution, which shows excellent agreement between both approach (spatial/temporal):
G. Kermarrec, X. Calbet, Z. Deng, C. Carbajal Henken, R. Preusker, "Retrieval of water vapor in the atmosphere and its spectral content: from OLCI to GPS," Proc. SPIE 12730, Remote Sensing of Clouds and the Atmosphere XXVIII, 127300F (19 October 2023); https://doi.org/10.1117/12.2678381
A weakness of the paper is that the method is tested only on five stations, and two days of data per station. This is a too small sample to draw any definitive conclusions.
>>>> You are correct, and we also avoid drawing definitive or general conclusions. We have slightly rewritten the end of the introduction to clarify this point as well as the conclusion.
I could understand that the authors wanted to look at individual days and compare the ZWD variations seen to the turbulence parameters and see if any expected correspondence could be found. However, I think no definitive conclusion could be drawn from that comparison. Perhaps it would be better to (also) look at longer time series and mor stations (several from the same climate region).
>>>> We agree with you, but addressing this fully would likely require a book rather than an introductory article. The fact that you raise this question shows that we have achieved our goal: the reader is engaged, wants to know more, and has understood the problem.
In several figures, e.g., Figs, 4, 6, and 8 there are negative ZWD, while we would expect ZWD to be positive. Looking at the raw data (https://doi.org/10.25835/HCC01FRE) it seems that what is plotted as ZWD is not actually the ZWD, but the estimated correction to the a priori (COR_WET). Since the a priori ZWD and ZHD seems to be constant, this is actually the correction to the ZTD, not only the ZWD. I guess for the derived turbulence parameters it does not matter since variations in ZHD will be mostly for longer time scales. However, when the authors comment on the variations of the ZWD for the various stations and days, it might be important to know if the variations seen are in ZHD (i.e. pressure) or ZWD (water vapour) to draw the correct conclusions. Nevertheless, the quantity should be labelled correctly (i.e. correction to ZTD instead of ZWD).
>>>>Yes, in the GNSS data analysis we estimated the ZWD corrections (COR_WET) to the a priori constant ZWD. Those ZWD corrections are plotted in the figures and are used to derive the turbulence parameters in this work. The slow variation ZHD and ZWD should have no impact. We have a note in the corresponding (2.2) as well as in 3.1 and all captions for clarity.
In figure A1 it seems strange that the “ZWD” at the end of doy 134 is quite different from the value at the beginning of doy 135. The values at the end of doy 135 and beginning of doy 136 also differ significantly.
>>>>>The 24-hour GNSS observations are used to estimate the 30-second ZWD. Since the ZWD is highly correlated with other estimated parameters (such as coordinates and clock error), boundary effects can be observed in the estimated ZWD parameters. To mitigate these boundary errors, a sliding window analysis could be considered. However, this lies beyond the scope of our paper and will be addressed in future work. We have correspondingly accounted for that fact in the processing to compute our parameters by excluding the first batches. In section 3, we have added a note to point out these effects.
In the GNSS data processing, what GNSS were used? I find no information about this.
>>>>>Thank you for the questions. We used all the GNSS satellites from the GFZ multi-GNSS products, which is includes GPS, GLONASS, Galileo, BeiDou and QZSS. It is specifed in the text now (section 3)
What software was used. The paper says Bernes GNSS Software, the supplementary material (https://doi.org/10.25835/HCC01FRE) says EPOS software.
>>>> it is the EPOS software from Gfz Potsdam, sorry for the mistake.
Line 30: “..high-rate GNSS are unaffected by clouds”. In fact, this is also true for non-high-rate GNSS. Furthermore, the GNSS signals are actually affected by clouds, but this effect is small and typically not explicitly considered.
>>>> You are correct, and we have become even more specific thanks to Reviewer 2, who pointed out the reference: Solheim et al. (1999) ("Propagation delays induced in GPS signals by dry air, water vapor, hydrometeors, and other particulates," J. Geophys. Res., 104, 9663–9670). The paper states that "A cloud droplet concentration of 1 g/m³ for a distance of 1 km has an integrated liquid value of 1 mm and would therefore induce a radio path delay of 1.45 mm." In the first version of the article, we mentioned that the topic focuses on WV turbulence. In the introduction, we now clarify that it refers to turbulence caused by both WV and liquid water clouds, but we "simplify" it to WV for readability.
Line 33: “dry delay” -> “hydrostatic delay”
>>>> in our opinion, this should be the same.
Line 78: “κ′ a factor depending on the surface temperature and specific gas constant”. It should perhaps be pointed out that the Bevis formula is only an approximation.
>>>> ok, we have added that it is an approximation in the text.
Line 218-220: “In the GNSS data processing we used the GeoForschungs Zentrum multi-GNSS satellites orbit and clock product in Precise Point Positioning (PPP) mode. Thus, the error of satellites orbit and clock can be ignored in this study.” I do not agree that errors in satellite orbits and clocks can be ignored. Since PPP are used, the errors in orbits and clocks will affect the estimated ZWD. I guess the GFZ products are of high quality, hence the error will be small, but maybe not negligible.
>>>>The GFZ satellite orbit and clock are estimated using a global GNSS network of approximately 140 stations. While not error-free, the large number of stations ensures that high-frequency signals caused by turbulence do not significantly affect the satellite orbit and clock estimates. In other words, these products are very smooth, making them suitable for detecting high-frequency signals observed by individual stations.
Line 230-232: “Effects from, e.g., nontidal loading, multipath, or antenna center phase variations are expected to be present in the residuals of the positioning adjustment and not in the ZWD, which is estimated as an independent delay.” I do not agree. Perhaps non-tidal loading will mostly go into the position estimates (although there might be variations on periods shorter than one day which the daily estimates will not catch). However, I think multipath and antenna phase center variations will affect both ZWD and coordinates.
>>>>>The antenna center phase variations are corrected with the absolute antenna calibrations in the IGS20 frame during the GNSS data processing. The multipath can cause usually a error with slow variation in the ZWD estimations. We have added this in section 3. One should mention that we focus on a very specific high frequency noise, and that the fact that we search for a very particular value of the psd slope should prevent us from multipath effect.
Line 306: “contrary to what we observed for the station UNB3”. What is observed is not exactly contrary to UNB3. On UNB3 a ZWD increase and sigma2 decrease was observed, for NYA2 there is a ZWD decrease and a sigma2 increase. Seems to be consistent to me.
>>>> you are right, it is corrected accordingly in the new version.
Line 410: “For comparison, we also computed the days DOY 135 and 136”. Should it be 134 and 136 (135 was the day with gravity waves)?
>>>> you re right, thanks for pointing that out.
Line 414: “DOY 169”. This day is not show in the figure.
>>>> sorry for the mistake, it has been corrected.
Line 414-415: “nearly constant with a strong sinus-like shape”. Strange formulation. It is not constant if it has a sinus shape.
>>>> we have corrected to "a light wavy shape". indeed, this was not consistent.
Citation: https://doi.org/10.5194/egusphere-2024-2680-AC2
Peer review completion
Journal article(s) based on this preprint
Atmospheric delays affect global navigation satellite system (GNSS) signals. This study analyses the wet delay, a variable component caused by atmospheric water vapor, using a novel filtering method to examine small-scale turbulent variations. Case studies at five global stations revealed daily and seasonal turbulence patterns. This research will improve water vapour and cloud models, enhance nowcasting, and refine stochastic modelling for GNSS and very long baseline interferometry.
Data sets
Zenithwetdelay Gael Kermarrec and Zhiguo Deng https://doi.org/10.25835/HCC01FRE
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 181 | 101 | 81 | 363 | 14 | 12 |
- HTML: 181
- PDF: 101
- XML: 81
- Total: 363
- BibTeX: 14
- EndNote: 12
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 161 | 44 |
| Germany | 2 | 55 | 15 |
| Sweden | 3 | 14 | 3 |
| China | 4 | 13 | 3 |
| Romania | 5 | 11 | 3 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 161
Gael Kermarrec
Xavier Calbet
Zhiguo Deng
Cintia Carbajal Henken
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1265 KB) - Metadata XML