Characterizations of Europe's integrated water vapor and assessments of atmospheric reanalyses using more than two decades of ground-based GPS
- 1Geodetic Institute, Karlsruhe Institute of Technology, Karlsruhe, 76131, Germany
- 2KMI-IRM, Royal Meteorological Institute of Belgium, Brussels, B-1180, Belgium
- 3Riverside Technology Inc, Asheville, NC 28801, USA
- 4Karlsruhe Institute of Technology, IMK-IFU, Garmisch-Partenkirchen, 82467, Germany
- 5Curtin University, School of Earth and Planetary Sciences, Perth, WA 6845, Australia
- 1Geodetic Institute, Karlsruhe Institute of Technology, Karlsruhe, 76131, Germany
- 2KMI-IRM, Royal Meteorological Institute of Belgium, Brussels, B-1180, Belgium
- 3Riverside Technology Inc, Asheville, NC 28801, USA
- 4Karlsruhe Institute of Technology, IMK-IFU, Garmisch-Partenkirchen, 82467, Germany
- 5Curtin University, School of Earth and Planetary Sciences, Perth, WA 6845, Australia
Abstract. Ground-based Global Positioning System (GPS) has been extensively used to retrieve Integrated Water Vapor (IWV) and has been adopted as a unique tool for the assessments of atmospheric reanalyses. In this study, we investigated the multi-temporal-scale variabilities and trends of IWV over Europe by using IWV time series from 108 GPS stations for more than two decades (1994–2018). We then adopted the GPS IWV as a reference to assess six commonly-used atmospheric reanalyses, namely CFSR, ERA5, ERA-Interim, JRA55, MERRA2, and NCEP2. The GPS results show that the diurnal cycles peak within 16:00–24:00 local time with peak-to-peak amplitudes accounting for 2 %–18 % of the daily mean. The diurnal 1-hourly anomalies can be much more intensive with a range of −100 % to 200 %. The annual cycles peak in July and August with maximum values of 17–32 kg m−2. The interannual variations of IWV over Europe are found to be mainly linked to the North Atlantic Oscillation (NAO) and the East Atlantic (EA) patterns. The IWV continues to increase over Europe during the last two decades at 0–0.4 kg m−2 decade−1 in the north and 0.4–1 kg m−2 decade−1 in the south. Regarding the assessments of the reanalyses, the intercomparisons with respect to GPS reveal a general superiority of the newly-released ERA5 IWV product. For instance, ERA5 only has a slight wet bias with a median value of 1 %, whereas the median bias for MERRA2 is 4 %. ERA5, MERRA2, and NCEP2 are the best, second best, and worst performers respectively in modelling the variability of daily IWV time series, with standard deviations of daily IWV differences against GPS by 0.5–1.6, 0.7–2.3, and 1.2–3.0 kg m−2, respectively. Moreover, the daily GPS IWV time series is best correlated with the ERA5 IWV with a median Pearson correlation coefficient of 0.996, whereas the second strongest and weakest median correlations are observed in MERRA2 and NCEP2 with values of 0.991 and 0.971, respectively. Furthermore, the correlations between the IWV trends from the reanalyses and GPS are strongest for ERA5 (0.82), a bit weaker for MERRA2 (0.72), and weakest for NCEP2 (0.52).
- Preprint
(2603 KB) -
Supplement
(740 KB) - BibTeX
- EndNote
Peng Yuan et al.
Status: final response (author comments only)
-
CC1: 'Comment on acp-2021-797', Olivier Bock, 03 Dec 2021
Dear Peng Yuan and co-authors,
Thank you for releasing this interesting study. I am happy to see that you used the representativeness statistic that we proposed in a previous publication and that you confirm and extend our results to other reanalyses.
Below I submit a few questions and comments about your manuscript. Thank you in advance for your answers.
Best regards,
Olivier BOCK1. Please comment on the choice and on the quality of the used GPS data set (NGL), as other data sets exist for Europe (e.g. the EPN repro2, Pacione et al., 2017).
2. Please provide more details on the homogenization method and results (e.g. the number and magnitude of detected breaks) and comment on their uncertainty. Explain also how the offsets in the GPS series are corrected, knowing that the breaks are detected in the GPS – reanalysis series and not in the GPS series directly.
Regarding the homogenization method, I checked your earlier paper (Yuan et al., 2021), and was wondering why you used a manual segmentation method when many statistical methods exist, which have been assessed by Van Malderen et al., 2020. Can you comment on that choice?
I also understand that in your segmentation method, you select only breaks which are confirmed by known equipment changes from the IGS log files. As you may have experienced: i) not all breaks are easy to detect (the example illustrated in Yuan et al., 2021, is a very optimistic case); ii) the IGS metadata may be incomplete and iii) the reanalysis may also have breaks. These limitations should be acknowledged in the paper.
Moreover, regarding the first two points, I think the manual approach is very subjective and also probably too conservative. You mention in the former paper that you detected 21 breaks from 108 stations over 21 years, i.e. an average of 1 break per station every 108 years. This number is very small compared to other studies, e.g. Ning et al., 2016, and Nguyen et al., 2021, using statistical methods. Overall, Nguyen et al., 2021, detected 1 break per station every 5.8 years (after screening) considering all breaks, among which the validated cases represent 1 break per station every 16 years. Both studies also show some obvious examples of undocumented breaks (namely for HERS) and breaks attributed to the reanalysis. Regarding the last point, you write that no obvious breaks were found in the reanalysis. What are your criteria to detect breaks in the reanalysis?
3. The analysis of the diurnal cycle is interesting. However, to make a fair intercomparison, the reanalyses should be analysed at the smaller common resolution which is 6-hourly, and not interpolated to a higher resolution (1-hourly). For the two reanalyses which have higher resolution (ERA5 and MERRA-2), you may show both the native and under-sampled (6-hourly) results.
4. In section 3.2, you may mention that the moist bias of ERAI over Europe was also reported by Parracho et al. 2018.
5. Please explain how you compute the trends.
6. In Section 6, you may mention that the trend results are also in line with the findings of Parracho et al. 2018, and Nguyen et al., 2021.
7. What is MERRA2’ in Figure 5?
Nguyen KN, Quarello A, Bock O, Lebarbier E. Sensitivity of Change-Point Detection and Trend Estimates to GNSS IWV Time Series Properties. Atmosphere. 2021; 12(9):1102. https://doi.org/10.3390/atmos12091102
Pacione, R., Araszkiewicz, A., Brockmann, E., and Dousa, J.: EPN-Repro2: A reference GNSS tropospheric data set over Europe, Atmos. Meas. Tech., 10, 1689–1705, https://doi.org/10.5194/amt-10-1689-2017, 2017. -
RC1: 'Comment on acp-2021-797', Anonymous Referee #1, 09 Jan 2022
Title: Characterizations of Europe’s integrated water vapor and assessments of atmospheric reanalyses using more than two decades of ground-based GPS
Authors: Peng Yuan et al.
Journal: Atmospheric Chemistry and Physics
This manuscript investigated the multi-temporal-scale variabilities and trends of IWV and assessed six commonly-used atmospheric reanalyses (CFSR, ERA5, ERA-Interim, JRA55, MERR2, and NCEP2) over Europe using IWV time series from 108 GPS stations for more than two decades. I have the following comments:
Main comments:
- The authors have taken into account of vertical IWV adjustment. However, the height system of GPS is different from that of the reanalyses. I’m not sure if the authors have considered the unification and differences of the different height systems?
- In the manuscript, the authors used the difference time series between ERA5 IWV and GPS IWV to visually detect the breaks in GPS IWV, so the potential significant differences may be eliminated since the homogenization, also this may be the reason why the ERA5 outperforms than other reanalyses. Are these breaks based on ERA5 IWV still significant, are there any other reanalyses used for the homogenization process?
- The spatial resolution contributes to most of the representativeness differences, such as the ERAI provides the products with higher spatial resolution (i.e. 0.25°) than the product used in this paper (0.75°). The conclusion that ERA5 has the best performance on the representativeness differences is questionable. This needs more clarification or convincing statements.
- Line 202: “The 3- and 6-hourly IWVs are linearly interpolated into 1-hourly time series.” Have the authors assessed the accuracy of the interpolated IWV? For IWV which changes in a high frequency, linear interpolation seems to be not a good choice.
- Line 157: There seems to be a missing full stop between “reanalyses” and “Compared”. Please check it.
-
RC2: 'Comment on acp-2021-797', Anonymous Referee #2, 23 Mar 2022
General Comments
The work presented in the manuscript gives an overall summary of applications of ground-based GPS observations in Europe of estimated time series of integrated water vapour (IWV), which to my knowledge is unique. It is broad in the sense that it deals with temporal scales from sub-daily to decades, while many previously published results often focus on one particular "signal", e.g. diurnal, annual, trends. As far as I can tell there are no new results in the manuscript, i.e. results that are different from what is already published. Three times it is stated that the results are "in line" with previously published results (lines 222, 323, and 399). Of course, it is also an important part of research to verify earlier findings, but if possible, I would appreciate if there was more emphasis on noted differences compared to earlier results. I am afraid I cannot help with the details. It is an impressive reference list and for me it is impossible to get a reasonably complete overall knowledge during the time allowed for the review.
Specific commentsL108: I do not understand the meaning of "integration rate of 95 %"? Can you explain what is being integrated?
L112: You report that the observations were weighted based on the elevation angle. Is it not important how the weighting was done (a weighting function including sine and cosine terms)?
L192: It is mentioned that homogenisation was done as described by Yuan et al. (2021). I think such a process is critical and it deserves some more detail in your paper instead of having to go through the reference. For example, do you allow breaks to be inserted in the GPS IWV time series at a specific time epoch even if there has been no change noted in the log file for the hardware or the environment at the site?
L262: My interpretation is that you determine the amplitudes of the diurnal signal as the peak-to-peak value regardless of when the peaks occur. This makes me wonder if the results will be different if instead the phase and amplitud of the sine wave with a 24 h period is estimated, e.g, through the method of least squares. (In some studies also a semidiurnal term, a period of 12 h, is estimated.) It will be of interest if you comment on this, at least for a couple of sites in different climate zones?
L268: You find a correlation between the diurnal amplitude and the station height. Since station height (I guess) correlate with the site's distance to the ocean, another approach would be to correlate the amplitude with this distance. It is well known that the ocean (as long as there is no ice) acts like a low pass filter on daily variations in temperature and humidity.
L315: This whole section seems questionable if it is worth to be published? Do the GPS IWV data yield any new findings? Given the very high correlation between IWV from GPS and from the reanalyses, it seems as all the reported patterns, and their time dependences, will be seen by using reanalyses data only?
Technical Corrections
Line (L)1+: You use the American spelling of vapour, although ACP is a European journal?
L97: ... IWV -using ... ?
L17: 2%-18% --> 2 %–18 % (similar changes to be carried out many times in the manuscript)
L154: IWVs --> The IWV values ?
L157: reanalyses Compared --> reanalyses. Compared
L203: IWVs are --> IWV for all sites and days are ?
L398: (29.5°E, 40.8°N), --> (29.5 °E, 40.8 °N), (see also L447-448)
L444: 0-0,4 --> 0.0 – 0.4 ?
L446: 0,4-1 --> 0.4 – 1.0 ?
L480+: doi links are missing for almost all references and the established standard acronyms for journals are not used.
Figure 2: The yellow colour is not ideal. I suggest to use cyan or magenta instead. You may also consider to use darker colours in Figures 5 and 8. Different colours in these figures are not really needed for clarity, although it may look nicer compared to have it all in black.
Peng Yuan et al.
Peng Yuan et al.
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
536 | 113 | 18 | 667 | 37 | 10 | 10 |
- HTML: 536
- PDF: 113
- XML: 18
- Total: 667
- Supplement: 37
- BibTeX: 10
- EndNote: 10
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1