Balloon drift estimation and improved position estimates for radiosondes

Voggenberger, Ulrich; Haimberger, Leopold; Ambrogi, Federico; Poli, Paul

doi:https://doi.org/10.5194/gmd-2023-215

Preprints

https://doi.org/10.5194/gmd-2023-215

Preprints

Submitted as: development and technical paper

22 Nov 2023

Submitted as: development and technical paper |

| 22 Nov 2023

Status: a revised version of this preprint was accepted for the journal GMD and is expected to appear here in due course.

Balloon drift estimation and improved position estimates for radiosondes

Ulrich Voggenberger, Leopold Haimberger, Federico Ambrogi, and Paul Poli

Abstract. When comparing model output with historical radiosonde observations, it is usually assumed that the radiosonde has risen exactly above its starting point and has not been displaced by the wind. This has changed only relatively recently with the availability of Global Navigation Satellite System (GNSS) receivers aboard the radiosondes in the late-1990s, but even then the balloon trajectory data were often not transmitted, although this information was the basis for estimating the wind in the first place. Depending on the conditions and time of year, radiosondes can sometimes drift a few hundred kilometres, particularly in the mid-latitudes during the winter months. The position errors can lead to non-negligible representation errors when the corresponding observations are assimilated.

This paper presents a methodology to compute changes in the balloon position during its vertical ascent, using only limited information, such as the vertical profile of wind contained in the historical observation reports. The sensitivity of the method to various parameters is investigated, such as the vertical resolution of the input data, the assumption about vertical ascent speed of the balloon, and the departure of the surface of the Earth from a sphere. The paper considers modern GNSS sonde data reports for validation, for which the full trajectory of the balloon is available, alongside the estimated wind. Evaluation is also conducted by comparison with ERA5 and by conducting low-resolution data assimilation experiments. Overall, the results indicate that the trajectory of the radiosonde can be accurately reconstructed from original data of varying vertical resolution and that the more accurate balloon position reduces representation errors, and, in some cases, also systematic errors.

Received: 30 Oct 2023 – Discussion started: 22 Nov 2023

Download & links

Ulrich Voggenberger, Leopold Haimberger, Federico Ambrogi, and Paul Poli

Status: closed

RC1:
'Comment on gmd-2023-215', Anonymous Referee #1, 21 Dec 2023

Review of "Balloon drift estimation and improved position estimates for radiosondes" by Voggenberger et al
General
I am pleased to see efforts to calculate and treat historical radiosonde drift in reanalyses.

This is a step is the right direction but needs tidying up.
It is known that radiosondes generally drift further in winter than summer (eg Seidel et al, 2011).

This is mentioned in the abstract but nowhere else and I find it a little strange that most

of the examples presented are for the summer eg Figure 5 and the assimilation experiments.
Quality control is an essential part of any data assimilation system (and is still messier than

we would like). What quality control (if any) was applied in this study to remove 'bad' reported data?
Complication at the poles:
The issue is stated in the Provisional 2023 edition of WMO No 8 (see below) and mainly affects the

South pole station (which still reports in TAC). If dispacements are calculated for this station

then the wind direction should also be adjusted.
https://community.wmo.int/en/activity-areas/imop/wmo-no.8/wmo-no-8-provisional-2023-edition

Chapter 13 Measurement of upper wind

Particular care is needed in reporting wind direction near to the North or South Pole especially

with the reporting of position at each level. A radiosonde crossing the North Pole must be in a

southerly airflow just before crossing the Pole and in a northerly airflow just afterwards. BUFR

reports should give wind direction relative to the longitude at each level. (The older

alphanumeric codes used a special coordinate system for stations within 1° latitude of the pole.)

Detailed comments
19 'alongside the estimated wind' - 'alongside the reported wind'?
29 'so-called representativeness errors'

There is a recent trend to call them 'representation errors', eg Hodyss and Nichols (2015, Tellus),

Janjic et al, (2018, QJRMS) although I would regard these references as optional here.
36 'Even later, when GNSS sensors became available, the information collected was often not transmitted'

The drift information could not be transmitted in the older alphanumeric codes.

It was only with the (ongoing) migration to BUFR that this became possible

along with the reporting of many more levels in the vertical (Ingleby et al., 2016)
Ingleby, B., P. Pauley, A. Kats, J. Ator, D. Keyser, A. Doerenbecher, E. Fucile, J. Hasegawa, E. Toyoda, T. Kleinert, W. Qu, J. St James, W. Tennant, and R. Weedon, 2016: Progress toward High-Resolution, Real-Time Radiosonde Reports. Bull. Amer. Meteor. Soc., 97, 2149-2161, https://doi.org/10.1175/BAMS-D-15-00169.1.
37 'Only since the mid-2010s is the balloon drift taken into account'

This needs some rewriting. NCEP have estimated and used radiosonde drift since 2000 (according to

Laroche and Sarrazin, 2013), but I haven't seen the documentation for the NCEP method.

Laroche and Sarrazin (I think) implemented it in the Canadian NWP system.

Ingleby and Edwards (2014, ASL) at the Met Office were early users of the BUFR

drift positions and quoted an earlier study: 'Macpherson (1995) found little benefit from treating balloon drift and concluded that for simple advection without wind shear the errors due to neglecting the exact time and position of each level cancelled.'

Suggested text: 'In the 2010s there was a move towards using the balloon drift information

in modern observation processing of GNSS sondes, with beneficial results (e.g. Laroche and Sarrazin, 2013; Ingleby et al., 2018).'
60-62 'High-resolution radiosonde data ... in BUFR ... University of Wyoming'

For information ECMWF BUFR radiosonde data is available (loosely associated

with IGRA) at the following link:

https://www.ncei.noaa.gov/data/ecmwf-global-upper-air-bufr/archive/

There is a separate archive (slightly less comprehensive) of data received by NCEP.
73 [GNSS sondes] 'can track the horizontal and vertical position of the balloon and the sensor at high frequency'

Delete 'the balloon and'. The high frequency data, especially the raw data

(before smoothing), sometimes clearly shows the effect of pedulum motion

under the balloon (eg Ingleby et al, 2022 https://doi.org/10.5194/amt-15-165-2022 and references, the results there raise questions about how optimal the smoothing is) - worse because

of the long suspensions often used now (55 m is recommended for the RS41).

I feel that pendulum motion should be briefly mentioned.
76 'if a single theodolite was used' I am sure that the vast majority of

theodolite ascents only use a single theodolite.
89 '[BUFR] has a much higher vertical resolution (1 second frequency'

This is potential resolution. Many European ascents report with 2 second

frequency, there area also many reports from other areas (eg China) with

fewer than 300 levels.

NB. Perhaps slightly clearer to replace 'the latter' with 'BUFR'.
91 '5-degrees'
103 '2.3 Estimation of the balloon trajectory'

There is some overlap with the work of Laroche and Sarrazin (2013).

Mention this in the text and explain where there are differences.
119-120 ', but temporal information is not available for all the reported pressure levels.' - ': time information is not available for any of the reported pressure levels' (slightly clearer)
130 'a piecewise polytropic atmosphere' I had to look up 'polytropic' and I am still not quite

sure what it means in this context. I suspect that you mean that temperature is a piecewise linear

function of height - please say this if so. My preference is to avoid the word polytropic.
155-156 'This is true for the differences within a single ascent, but also for the differences between different ascents.' - 'This is true both within a single ascent and also between different ascents.'
192 'The software necessary for the creation of calculated balloon trajectories can be found in:'

Is it the same software in each of the three repositories?
215 '4. Verification with GNSS radiosondes'

'Validation' might be better than 'Verification'
216 'indisputable reference' I suggest replacing indisputable with 'good'.

I agree that GNSS data is usually very good quality, but there can be problems with interference or

only a few GPS satellites above the horizon. As mentioned above pendulum motion complicates a detailed

view of the positions/raw winds.
230 'Figure 5 ...'

The longitude displacements (in degrees) are larger than the latitude displacements - presumably

the longitude displacements are dominated by high latitude ascents.

I recommend showing values in km instead and ideally show summer and winter displacements separately.

At the lowest level there is a local maximum in latitude RMS - probably caused by a minor error in

the reported launch position, I have seen such errors.

Y-axis - personally I would prefer to see more levels labelled.
246 'Figure 8' Again, I think that displacements in km would be better.
333-334 'The second experiment assimilates the radiosonde observations as slanted profiles'

Is each level given a separate latitude/longitude/time or is it done (for efficiency reasons) in

15-minute sub-profiles as in Ingleby et al (2018).
334 'the balloon trajectory, when it is available' ie when there is a reasonably complete wind profile?
364 'Laroche, 2013' - 'Laroche and Sarrazin, 2013
370-371 'we also expect that any positive impact of sonde displacement be amplified when

satellite observations are used alongside radiosondes' I'm not convinced of this - more satellite

data usually means less impact of radiosonde data (especially tempertures).
384 'experiments, covering a two-month period in summer 1980'

There is usually more drift in winter.
419-420 'reconstructing displacements based on the reversed calculation of wind speed and direction'

'reconstructing displacements based on the wind profile' would be better in my opinion.

With GNSS the winds are sometimes derived from the Doppler effect rather than differentiating the

displacements.
Figure 1. Perhaps clarify that the wind rose gives the direction from which the wind is coming.
Figure 9. At first I didn't understand what the green dashed line was. I now think it is the

RMS difference between ERA5 temperatures 'vertical' and 'slanted' - I presume that there are time

differences included too (please confirm). Does the dashed green line use the upper or lower axis?
Figures 11. The 10 hPa differences from ERA5 in 1970 are very large, presumably due to the

radiosonde type used at the time and possibly a lack of quality control.

I wondered if figures 10 and 11 add much or would be better omitted.
Figure 15. The 10 hPa obs-bg RMSE values are huge (15 degrees).

My guess is that there has been little or no attempt to quality control the reported values.
573 'Favà et al' drop 'Discuss.' ?

Citation: https://doi.org/10.5194/gmd-2023-215-RC1
- AC1:
  'Reply on RC1', Ulrich Voggenberger, 16 Jan 2024
  
  Thank you for your review and comments. We appreciate your input.
  Regarding the summer month examples, at least two plots will be replaced by winter data. We will also plot the total displacements (in km) once for clarification but will otherwise stick to lat/lon as units of displacements to keep consistency with BUFR encoding.
  As for the assimilation experiment:
  
  As is often the case with data assimilation experiments, which are computationally expensive undertakings, the exact choice of time period was influenced by considerations of technical nature. In the present case, the availability of a number of pre-existing experiments for summer 2006 with the same version of the IFS as used for the present investigations presented an important advantage. This time period is indeed important for climate reanalysis as summer 2006 is the first season of well-constrained stratosphere thanks to GNSS radio occultation observations, following the launch of the COSMIC constellation in April that same year. This pre-existing material enabled us to run the radiosonde drift data assimilation experiment for longer than would have been the case otherwise, if baseline experiments had been needed prior to establishing a control experiment. We also note that Choi et al. (2015) also ran experiments in the summer season (2013 in their case), albeit for a much shorter time period (under a month). All that said, we agree that, under ideal circumstances, one should run data assimilation experiments over several time periods. Nevertheless, the present choice of season is important, even if it is at the lower end of the expected impact. A negligible impact would indeed indicate that the effects are not worth to consider in data assimilation for this particular season. This is not what we find, and for this reason we believe the results are worth presenting. We have added a statement in the revised paper indicating the following:
  
  “Given previous results indicating a larger effect of the balloon drift during winter seasons (e.g. McGrath et al., 2006), and given the much greater number of radiosonde stations in the Northern Hemisphere as compared to the Southern hemisphere (e.g., see Figure 13), the present choice of the data assimilation season (Northern hemisphere summer) represents a conservative approach. An impact of larger magnitude may be expected at different time periods, in particular during Northern hemisphere winter.”
  Quality control is performed at a basic level by threshold checking the input temperature and wind data. The temperature should be between 172 and 372 K, and wind components should be below 150 m/s. It could be argued that narrowing down the checks based on ERA5 background departure threshold could further clean the data. However, this would make the calculated displacements not independent anymore. The main issue would be erroneous wind values since temperature values are only used for layer height calculation. We will investigate whether further quality control will improve ERA5 RMSE differences.
  You are right it is important to exercise caution when using trajectories from sondes that are likely to cross the poles. This point will also be highlighted in the paper. We will not change the original wind encodings, however, we only calculate the displacements.
  @192: All repositories contain identical code.
  @230: Longitudinal displacements are greater than latitudinal displacements, not only in high latitude ascents. The westerlies and trade winds are predominantly zonal. We will investigate and address the low-level local maximum in the latitude RMS.
  The displacement is presented in its latitude and longitude components because it is encoded in BUFR in this form. The plots should clearly show the difference in magnitude between the two components.
  Additionally, we will include a plot of the total distance in km, in addition to the lat/lon plots, in at least one figure.
  @334: the trajectory will be calculated if there are no gaps in the profile and the lowest available level is not more than 1500 m above the station height. This will be mentioned in the paper.
  @370-371: Also, the background for comparison with earlier data is of lower quality. Therefore, removing the location representation error may show less improvement.
  @Figures 10 and 11: We want to keep Figure 11 to demonstrate ^, while Figure 10 can be omitted.
  @Fig 15: The legend states that the obs-bg RMSE values are 10^-3, indicating their small size.
  We will consider all other comments and make additions to the paper as necessary.
  
  Citation: https://doi.org/10.5194/gmd-2023-215-AC1
  - AC3: 'Reply on AC1', Ulrich Voggenberger, 16 Jan 2024
    
    In our previous comment the first three sentences referring to the data assimilation experiment should read:
    "As is often the case with data assimilation experiments, which are computationally expensive undertakings, the exact choice of time period was influenced by considerations of technical nature. In the present case, the availability of a number of pre-existing experiments for summer 1980 with the same version of the IFS as used for the present investigations presented an important advantage. This time period is indeed important for climate reanalysis as summer 1980 is the first season of well-constrained troposphere/stratosphere thanks to starting availability of two TOVS satellites."
    
    Citation: https://doi.org/10.5194/gmd-2023-215-AC3
RC2:
'Comment on gmd-2023-215', Anonymous Referee #2, 02 Jan 2024

I am glad to see this work being done so that rawinsonde data can be better assimilated into re-analyses. One thing that should be clarified is the two different datasets (TAC and BUFR), that one contains time and three-dimensional position information and one does not, and the fact that perhaps not all data are yet transmitted in BUFR. Therefore, this technique could also be useful in operations until the time that all data are available in BUFR.
I note that similar work has been done for downsondes (as opposed to the current upsondes), https://doi.org/10.1175/JTECH-D-17-0023.1, and that this should be referenced.
Some of the writing is confusing and could use some editing for improvement. ines 240-242 are a good example.
Some additions to the current study would be helpful:
1. Mean statistics with height over all sondes in both TAC and BUFR of the differences between the calculated and actual trajectories. Figure 5 shows the statistics for the "summer months" (undefined) for one year, which is helpful, but incomplete. In addition, it isn't clear why the displacements are separated into latitudinal and longitudinal components instead of total distance. If there is a physical reason to do so, this should be explained. If there is no physical reason, then total differences should be shown.
2. Perhaps examples of particularly accurate and particularly inaccurate displacement calculations (trajectories) can be shown to show the user how this may work in practice.
Additional comments:
Figure 1 is not referenced in the text.
Significant levels are defined starting on line 92. There are also significant levels for thermodynamic variables, and these should also be defined here for completeness.
The text says that PILOT/PIBAL have height data. All observations with mandatory levels have height data (if there are thermodynamic data), but it seems that these values are not used in the calculations. If this is correct, why not? If this is incorrect, clarify in the text.
Line 198: Is the station height also necessary for the calculation?

Citation: https://doi.org/10.5194/gmd-2023-215-RC2
- AC2: 'Reply on RC2', Ulrich Voggenberger, 16 Jan 2024
  
  Thanks a lot for these comments!
  The paper will provide a more elaborate description of the difference between TAC and BUFR to ensure clarity.
  @240-242: To enhance readability, we will begin with this explanation and utilise simpler sentences with a logical progression of information.
  @1. We will plot the total displacements once for clarification but otherwise use lat/lon as units of displacement to maintain consistency with BUFR encoding.
  @2. Figure 6 and 7 will be updated to provide examples, which show the range of accuracy of the calculated trajectories. This will also be mentioned in the text.
  @height data: It is used to calculate observation time for PIBAL ascents. It is not used when temperature is available, since this ensures consistent calculation of height across all radiosondes. In historical ascents the geopotential has been often calculated differently, using different values for gravity acceleration etc. This was one reason why since the 1980s geopotential height has no longer been assimilated in most NWP centers. Therefore we avoid using it as well.
  @198: station height is not necessary for the calculation but it is used for quality control purposes. Profiles that begin more than 1500m above the station height are discarded as it is assumed that too much information is missing.
  All the other comments will be considered and the paper will be updated accordingly.
  
  Citation: https://doi.org/10.5194/gmd-2023-215-AC2
EC1: 'Comment on gmd-2023-215', Juan Antonio Añel, 16 Jan 2024

Dear authors,
Many thanks for your replies. This is a comment simply to make clear that I welcome a revised version of your manuscript.
You will received Instructions for it after the Discussions period is closed.
Juan A. Añel
Geosci. Model Dev. Executive Editor

Citation: https://doi.org/10.5194/gmd-2023-215-EC1

Status: closed

RC1:
'Comment on gmd-2023-215', Anonymous Referee #1, 21 Dec 2023

Review of "Balloon drift estimation and improved position estimates for radiosondes" by Voggenberger et al
General
I am pleased to see efforts to calculate and treat historical radiosonde drift in reanalyses.

This is a step is the right direction but needs tidying up.
It is known that radiosondes generally drift further in winter than summer (eg Seidel et al, 2011).

This is mentioned in the abstract but nowhere else and I find it a little strange that most

of the examples presented are for the summer eg Figure 5 and the assimilation experiments.
Quality control is an essential part of any data assimilation system (and is still messier than

we would like). What quality control (if any) was applied in this study to remove 'bad' reported data?
Complication at the poles:
The issue is stated in the Provisional 2023 edition of WMO No 8 (see below) and mainly affects the

South pole station (which still reports in TAC). If dispacements are calculated for this station

then the wind direction should also be adjusted.
https://community.wmo.int/en/activity-areas/imop/wmo-no.8/wmo-no-8-provisional-2023-edition

Chapter 13 Measurement of upper wind

Particular care is needed in reporting wind direction near to the North or South Pole especially

with the reporting of position at each level. A radiosonde crossing the North Pole must be in a

southerly airflow just before crossing the Pole and in a northerly airflow just afterwards. BUFR

reports should give wind direction relative to the longitude at each level. (The older

alphanumeric codes used a special coordinate system for stations within 1° latitude of the pole.)

Detailed comments
19 'alongside the estimated wind' - 'alongside the reported wind'?
29 'so-called representativeness errors'

There is a recent trend to call them 'representation errors', eg Hodyss and Nichols (2015, Tellus),

Janjic et al, (2018, QJRMS) although I would regard these references as optional here.
36 'Even later, when GNSS sensors became available, the information collected was often not transmitted'

The drift information could not be transmitted in the older alphanumeric codes.

It was only with the (ongoing) migration to BUFR that this became possible

along with the reporting of many more levels in the vertical (Ingleby et al., 2016)
Ingleby, B., P. Pauley, A. Kats, J. Ator, D. Keyser, A. Doerenbecher, E. Fucile, J. Hasegawa, E. Toyoda, T. Kleinert, W. Qu, J. St James, W. Tennant, and R. Weedon, 2016: Progress toward High-Resolution, Real-Time Radiosonde Reports. Bull. Amer. Meteor. Soc., 97, 2149-2161, https://doi.org/10.1175/BAMS-D-15-00169.1.
37 'Only since the mid-2010s is the balloon drift taken into account'

This needs some rewriting. NCEP have estimated and used radiosonde drift since 2000 (according to

Laroche and Sarrazin, 2013), but I haven't seen the documentation for the NCEP method.

Laroche and Sarrazin (I think) implemented it in the Canadian NWP system.

Ingleby and Edwards (2014, ASL) at the Met Office were early users of the BUFR

drift positions and quoted an earlier study: 'Macpherson (1995) found little benefit from treating balloon drift and concluded that for simple advection without wind shear the errors due to neglecting the exact time and position of each level cancelled.'

Suggested text: 'In the 2010s there was a move towards using the balloon drift information

in modern observation processing of GNSS sondes, with beneficial results (e.g. Laroche and Sarrazin, 2013; Ingleby et al., 2018).'
60-62 'High-resolution radiosonde data ... in BUFR ... University of Wyoming'

For information ECMWF BUFR radiosonde data is available (loosely associated

with IGRA) at the following link:

https://www.ncei.noaa.gov/data/ecmwf-global-upper-air-bufr/archive/

There is a separate archive (slightly less comprehensive) of data received by NCEP.
73 [GNSS sondes] 'can track the horizontal and vertical position of the balloon and the sensor at high frequency'

Delete 'the balloon and'. The high frequency data, especially the raw data

(before smoothing), sometimes clearly shows the effect of pedulum motion

under the balloon (eg Ingleby et al, 2022 https://doi.org/10.5194/amt-15-165-2022 and references, the results there raise questions about how optimal the smoothing is) - worse because

of the long suspensions often used now (55 m is recommended for the RS41).

I feel that pendulum motion should be briefly mentioned.
76 'if a single theodolite was used' I am sure that the vast majority of

theodolite ascents only use a single theodolite.
89 '[BUFR] has a much higher vertical resolution (1 second frequency'

This is potential resolution. Many European ascents report with 2 second

frequency, there area also many reports from other areas (eg China) with

fewer than 300 levels.

NB. Perhaps slightly clearer to replace 'the latter' with 'BUFR'.
91 '5-degrees'
103 '2.3 Estimation of the balloon trajectory'

There is some overlap with the work of Laroche and Sarrazin (2013).

Mention this in the text and explain where there are differences.
119-120 ', but temporal information is not available for all the reported pressure levels.' - ': time information is not available for any of the reported pressure levels' (slightly clearer)
130 'a piecewise polytropic atmosphere' I had to look up 'polytropic' and I am still not quite

sure what it means in this context. I suspect that you mean that temperature is a piecewise linear

function of height - please say this if so. My preference is to avoid the word polytropic.
155-156 'This is true for the differences within a single ascent, but also for the differences between different ascents.' - 'This is true both within a single ascent and also between different ascents.'
192 'The software necessary for the creation of calculated balloon trajectories can be found in:'

Is it the same software in each of the three repositories?
215 '4. Verification with GNSS radiosondes'

'Validation' might be better than 'Verification'
216 'indisputable reference' I suggest replacing indisputable with 'good'.

I agree that GNSS data is usually very good quality, but there can be problems with interference or

only a few GPS satellites above the horizon. As mentioned above pendulum motion complicates a detailed

view of the positions/raw winds.
230 'Figure 5 ...'

The longitude displacements (in degrees) are larger than the latitude displacements - presumably

the longitude displacements are dominated by high latitude ascents.

I recommend showing values in km instead and ideally show summer and winter displacements separately.

At the lowest level there is a local maximum in latitude RMS - probably caused by a minor error in

the reported launch position, I have seen such errors.

Y-axis - personally I would prefer to see more levels labelled.
246 'Figure 8' Again, I think that displacements in km would be better.
333-334 'The second experiment assimilates the radiosonde observations as slanted profiles'

Is each level given a separate latitude/longitude/time or is it done (for efficiency reasons) in

15-minute sub-profiles as in Ingleby et al (2018).
334 'the balloon trajectory, when it is available' ie when there is a reasonably complete wind profile?
364 'Laroche, 2013' - 'Laroche and Sarrazin, 2013
370-371 'we also expect that any positive impact of sonde displacement be amplified when

satellite observations are used alongside radiosondes' I'm not convinced of this - more satellite

data usually means less impact of radiosonde data (especially tempertures).
384 'experiments, covering a two-month period in summer 1980'

There is usually more drift in winter.
419-420 'reconstructing displacements based on the reversed calculation of wind speed and direction'

'reconstructing displacements based on the wind profile' would be better in my opinion.

With GNSS the winds are sometimes derived from the Doppler effect rather than differentiating the

displacements.
Figure 1. Perhaps clarify that the wind rose gives the direction from which the wind is coming.
Figure 9. At first I didn't understand what the green dashed line was. I now think it is the

RMS difference between ERA5 temperatures 'vertical' and 'slanted' - I presume that there are time

differences included too (please confirm). Does the dashed green line use the upper or lower axis?
Figures 11. The 10 hPa differences from ERA5 in 1970 are very large, presumably due to the

radiosonde type used at the time and possibly a lack of quality control.

I wondered if figures 10 and 11 add much or would be better omitted.
Figure 15. The 10 hPa obs-bg RMSE values are huge (15 degrees).

My guess is that there has been little or no attempt to quality control the reported values.
573 'Favà et al' drop 'Discuss.' ?

Citation: https://doi.org/10.5194/gmd-2023-215-RC1
- AC1:
  'Reply on RC1', Ulrich Voggenberger, 16 Jan 2024
  
  Thank you for your review and comments. We appreciate your input.
  Regarding the summer month examples, at least two plots will be replaced by winter data. We will also plot the total displacements (in km) once for clarification but will otherwise stick to lat/lon as units of displacements to keep consistency with BUFR encoding.
  As for the assimilation experiment:
  
  As is often the case with data assimilation experiments, which are computationally expensive undertakings, the exact choice of time period was influenced by considerations of technical nature. In the present case, the availability of a number of pre-existing experiments for summer 2006 with the same version of the IFS as used for the present investigations presented an important advantage. This time period is indeed important for climate reanalysis as summer 2006 is the first season of well-constrained stratosphere thanks to GNSS radio occultation observations, following the launch of the COSMIC constellation in April that same year. This pre-existing material enabled us to run the radiosonde drift data assimilation experiment for longer than would have been the case otherwise, if baseline experiments had been needed prior to establishing a control experiment. We also note that Choi et al. (2015) also ran experiments in the summer season (2013 in their case), albeit for a much shorter time period (under a month). All that said, we agree that, under ideal circumstances, one should run data assimilation experiments over several time periods. Nevertheless, the present choice of season is important, even if it is at the lower end of the expected impact. A negligible impact would indeed indicate that the effects are not worth to consider in data assimilation for this particular season. This is not what we find, and for this reason we believe the results are worth presenting. We have added a statement in the revised paper indicating the following:
  
  “Given previous results indicating a larger effect of the balloon drift during winter seasons (e.g. McGrath et al., 2006), and given the much greater number of radiosonde stations in the Northern Hemisphere as compared to the Southern hemisphere (e.g., see Figure 13), the present choice of the data assimilation season (Northern hemisphere summer) represents a conservative approach. An impact of larger magnitude may be expected at different time periods, in particular during Northern hemisphere winter.”
  Quality control is performed at a basic level by threshold checking the input temperature and wind data. The temperature should be between 172 and 372 K, and wind components should be below 150 m/s. It could be argued that narrowing down the checks based on ERA5 background departure threshold could further clean the data. However, this would make the calculated displacements not independent anymore. The main issue would be erroneous wind values since temperature values are only used for layer height calculation. We will investigate whether further quality control will improve ERA5 RMSE differences.
  You are right it is important to exercise caution when using trajectories from sondes that are likely to cross the poles. This point will also be highlighted in the paper. We will not change the original wind encodings, however, we only calculate the displacements.
  @192: All repositories contain identical code.
  @230: Longitudinal displacements are greater than latitudinal displacements, not only in high latitude ascents. The westerlies and trade winds are predominantly zonal. We will investigate and address the low-level local maximum in the latitude RMS.
  The displacement is presented in its latitude and longitude components because it is encoded in BUFR in this form. The plots should clearly show the difference in magnitude between the two components.
  Additionally, we will include a plot of the total distance in km, in addition to the lat/lon plots, in at least one figure.
  @334: the trajectory will be calculated if there are no gaps in the profile and the lowest available level is not more than 1500 m above the station height. This will be mentioned in the paper.
  @370-371: Also, the background for comparison with earlier data is of lower quality. Therefore, removing the location representation error may show less improvement.
  @Figures 10 and 11: We want to keep Figure 11 to demonstrate ^, while Figure 10 can be omitted.
  @Fig 15: The legend states that the obs-bg RMSE values are 10^-3, indicating their small size.
  We will consider all other comments and make additions to the paper as necessary.
  
  Citation: https://doi.org/10.5194/gmd-2023-215-AC1
  - AC3: 'Reply on AC1', Ulrich Voggenberger, 16 Jan 2024
    
    In our previous comment the first three sentences referring to the data assimilation experiment should read:
    "As is often the case with data assimilation experiments, which are computationally expensive undertakings, the exact choice of time period was influenced by considerations of technical nature. In the present case, the availability of a number of pre-existing experiments for summer 1980 with the same version of the IFS as used for the present investigations presented an important advantage. This time period is indeed important for climate reanalysis as summer 1980 is the first season of well-constrained troposphere/stratosphere thanks to starting availability of two TOVS satellites."
    
    Citation: https://doi.org/10.5194/gmd-2023-215-AC3
RC2:
'Comment on gmd-2023-215', Anonymous Referee #2, 02 Jan 2024

I am glad to see this work being done so that rawinsonde data can be better assimilated into re-analyses. One thing that should be clarified is the two different datasets (TAC and BUFR), that one contains time and three-dimensional position information and one does not, and the fact that perhaps not all data are yet transmitted in BUFR. Therefore, this technique could also be useful in operations until the time that all data are available in BUFR.
I note that similar work has been done for downsondes (as opposed to the current upsondes), https://doi.org/10.1175/JTECH-D-17-0023.1, and that this should be referenced.
Some of the writing is confusing and could use some editing for improvement. ines 240-242 are a good example.
Some additions to the current study would be helpful:
1. Mean statistics with height over all sondes in both TAC and BUFR of the differences between the calculated and actual trajectories. Figure 5 shows the statistics for the "summer months" (undefined) for one year, which is helpful, but incomplete. In addition, it isn't clear why the displacements are separated into latitudinal and longitudinal components instead of total distance. If there is a physical reason to do so, this should be explained. If there is no physical reason, then total differences should be shown.
2. Perhaps examples of particularly accurate and particularly inaccurate displacement calculations (trajectories) can be shown to show the user how this may work in practice.
Additional comments:
Figure 1 is not referenced in the text.
Significant levels are defined starting on line 92. There are also significant levels for thermodynamic variables, and these should also be defined here for completeness.
The text says that PILOT/PIBAL have height data. All observations with mandatory levels have height data (if there are thermodynamic data), but it seems that these values are not used in the calculations. If this is correct, why not? If this is incorrect, clarify in the text.
Line 198: Is the station height also necessary for the calculation?

Citation: https://doi.org/10.5194/gmd-2023-215-RC2
- AC2: 'Reply on RC2', Ulrich Voggenberger, 16 Jan 2024
  
  Thanks a lot for these comments!
  The paper will provide a more elaborate description of the difference between TAC and BUFR to ensure clarity.
  @240-242: To enhance readability, we will begin with this explanation and utilise simpler sentences with a logical progression of information.
  @1. We will plot the total displacements once for clarification but otherwise use lat/lon as units of displacement to maintain consistency with BUFR encoding.
  @2. Figure 6 and 7 will be updated to provide examples, which show the range of accuracy of the calculated trajectories. This will also be mentioned in the text.
  @height data: It is used to calculate observation time for PIBAL ascents. It is not used when temperature is available, since this ensures consistent calculation of height across all radiosondes. In historical ascents the geopotential has been often calculated differently, using different values for gravity acceleration etc. This was one reason why since the 1980s geopotential height has no longer been assimilated in most NWP centers. Therefore we avoid using it as well.
  @198: station height is not necessary for the calculation but it is used for quality control purposes. Profiles that begin more than 1500m above the station height are discarded as it is assumed that too much information is missing.
  All the other comments will be considered and the paper will be updated accordingly.
  
  Citation: https://doi.org/10.5194/gmd-2023-215-AC2
EC1: 'Comment on gmd-2023-215', Juan Antonio Añel, 16 Jan 2024

Dear authors,
Many thanks for your replies. This is a comment simply to make clear that I welcome a revised version of your manuscript.
You will received Instructions for it after the Discussions period is closed.
Juan A. Añel
Geosci. Model Dev. Executive Editor

Citation: https://doi.org/10.5194/gmd-2023-215-EC1

Ulrich Voggenberger, Leopold Haimberger, Federico Ambrogi, and Paul Poli

Viewed

Total article views: 344 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
249	69	26	344	13	10

HTML: 249
PDF: 69
XML: 26
Total: 344
BibTeX: 13
EndNote: 10

Views and downloads (calculated since 22 Nov 2023)

Month	HTML	PDF	XML	Total
Nov 2023	50	15	4	69
Dec 2023	54	12	5	71
Jan 2024	68	18	7	93
Feb 2024	31	7	7	45
Mar 2024	37	17	3	57
Apr 2024	9	0	9

Cumulative views and downloads (calculated since 22 Nov 2023)

Month	HTML	PDF	XML	Total
Nov 2023	50	15	4	69
Dec 2023	54	12	5	71
Jan 2024	68	18	7	93
Feb 2024	31	7	7	45
Mar 2024	37	17	3	57
Apr 2024	9	0	9

Viewed (geographical distribution)

Total article views: 323 (including HTML, PDF, and XML) Thereof 323 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 09 Apr 2024

Short summary

The paper presents a method for calculating balloon drift from historical radiosonde ascent data. This drift can reach distances of several hundred kilometres and is often neglected. Verification shows the beneficial impact of the more accurate balloon position on model assimilation. The method is not limited to radiosondes, but would also work for drop sondes, ozone sondes, or any other in-situ sonde carried by the wind in the pre-GNSS era, provided the necessary information is available.


Total:	0
HTML:	0
PDF:	0
XML:	0