the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 May 2022
Can state-of-the-art infrared satellite sounders and reanalyses detect moisture inversions in the Arctic?
Abstract. Moisture inversions, i.e. layers in the troposphere where specific humidity increases with height, are extremely frequent in the Arctic. They are strongly intertwined with cloud processes, as well as the energy budget, by affecting the downward longwave radiation. In this study, the capability of two benchmark satellite sounders, the Infrared Atmospheric Sounding Interferometer (IASI) and the Atmospheric Infrared Sounder (AIRS), to detect moisture inversions is systematically assessed based on radiosonde data from the Arctic site Ny-Ålesund. In particular for IASI, such an analysis has not been done before. The frequency of occurrence of moisture inversions at Ny-Ålesund based on radiosoundings is above 95 % in all seasons, with multiple inversions in the same profile occurring most of the time (in 82 % of the profiles). We first performed a sensitivity analysis which revealed that the chosen vertical grid of the specific humidity profiles has a distinct impact on the inversion frequency: in general, the lower the grid resolution, the lower also the detected inversion frequency. However, even when reducing the vertical resolution of the radiosonde profiles used in the comparison to match that of the IASI and AIRS retrievals, a large underestimation in both inversion frequency and inversion strength can be found in the satellite products. While observed inversion frequency in any 100-hPa-deep layer between 1000 and 400 hPa is typically between 10 and 20 % (50 and 80 %) in the regridded (original) radiosonde data, inversion frequency in satellite data is in most cases below 10 %. A better agreement has been found for IASI below the 900-hPa level and in particular for winter, while AIRS does not detect any inversions below the 700-hPa level in this season. In addition to satellite data, reanalysis data also have the potential to provide an Arctic-wide view on moisture inversion characteristics. Here, we thus also assessed the capability of the latest reanalysis developed by ECMWF, ERA5, to detect moisture inversions at Ny-Ålesund, by performing a similar comparison with radiosoundings. We found that ERA5 represents Arctic moisture inversion characteristics very well, if the radiosonde profile resolution is reduced to that of the reanalysis.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(2587 KB)
Interactive discussion
Status: closed
-
RC1: 'Comment on amt-2022-22', Anonymous Referee #2, 02 Jun 2022
Chellini and Ebell compare statistics of moisture inversions in the Arctic between datasets of the nadir thermal IR sounding instruments IASI and AIRS, the ECWMF ERA5 reanalysis data and radiosoundings as reference measurements. They draw as their main conclusions (a) that the nadir-sounders strongly underestimate inversions in specific humidity and, (b) that the reanalysis datasets characterize such profiles quite well when compared to radiosonde profiles with adapted vertical resolution.
In my opinion, this is a valuable and clear study expressing the limitations of satellite moisture data with a focus on specific features of moisture vertical profiles in Arctic regions. Thus, it helps also to prevent inappropriate use of those datasets. My main criticism concerns the method applied to adapt the vertical resolution of radiosonde data to that of the nadir-sounders. As the authors state themselves, the correct method would have been the application of the satellite instruments’ averaging kernels to the radiosonde profiles. In this study, only a vertical interpolation and averaging is performed which does by no means simulate typical nadir sounding averaging kernels. That this is wrong can be seen by the fact that the statistical analysis of humidity inversions when applying their averaging to the radiosoundings (Fig. 4) is largely different from the analysis on the satellite data themselves (Fig. 5). If the real averaging kernels were applied, the differences should be much smaller. Thus, this exercise only shows that the wrong ‘resolution-reduction’ has been applied to the radiosonde dataset. Therefore, I would draw the conclusion that the related sections (3.2,4.2) should better be omitted since they do not really contribute any new insights and rely on a very inaccurate method.
Similarly, this would apply to the comparison of resolution reduced radiosoundings and ERA5. However, this may be more appropriate due to the presumably better behaved vertical ‘kernel’ and resolution of ERA5 compared to nadir-sounding instruments.
In summary I would recommend this work for publication in AMT, if the major comment, as expressed above, is taken into consideration.
Specific comments:
L265:
Please specify if IASI/AIRS data is also assimilated in ERA5.
L375, ‘Standard Product (7 between 1000 and 400 hPa) is close to the typical number of degrees of freedom of the IASI retrievals (e.g., ranging between 2.5 and 7.2, according to Ebell et al. (2013))’:
It is not close but at the upper, optimistic end of the typical degrees of freedom reached by IASI retrievals.
L451, ‘IASI seems to be able to capture better inversion frequency below the 800-hPa level in winter.’, also L15 (abstract) ‘A better agreement has been found for IASI below the 900-hPa level and in particular for winter’:
Considering the fact, that thermal IR nadir sounders generally have a very low sensitivity on the lowermost atmospheric layers (due to the small thermal contrast between surface and lower atmospheric temperature), I doubt that this ‘better’ achievement is due to real information from the measurement but may stem from retrieval artefacts. At least this should be a part of the discussion.
L566, ‘… If ERA5 showed a similar good performance, the data could also be used as a reference to be compared to the satellite retrievals across the whole Arctic region’:
It should be emphasized here that this argument is valid only in case the radiosoundings used for comparison are really independent from the reanalysis, have not been assimilated and are locally distinct from radiosoundings used in the assimilation.
Citation: https://doi.org/10.5194/amt-2022-22-RC1 -
RC2: 'Comment on amt-2022-22', Anonymous Referee #1, 16 Jun 2022
The mauscript of Chellini and Ebell illustrates the results of a study aiming to answer the question of whether humidity profiles retrieved from infrared satellite (nadir) sounders and ECMWF reanalyses can be used for the detection of moisture inversion layers in the Arctic.
The authors have substantially ignored my initial comments, therefore, I am sorry that I have to repeat them here and, still, I cannot recommend the manuscript for publication in the Atmospheric Measurement Techniques journal. The motivations will be clear from the comments included hereafter.
General comments
1) Independently of the vertical grid that may have been used for profile retrievals, at low altitudes (0 – 10 km), the vertical resolution (i.e. the full width at half maximum of the averaging kernels) of IASI water vapour profiles is of the order of 3 – 5 km. For this sole reason, one would not even try to use the profiles retrieved from these missions to detect moisture inversions in 100 hPa layers. A more sound question would be: which is the minimum size (in terms of strength and depth) versus altitude, of moisture inversions that can be detected from IASI and AIRS retrieved profiles? To my opinion the work could be re-organized in the attempt to answer this question. As it is now, the manuscript only shows that the trivial answer (no) to the question set in the title (at least the part regarding satellite sounders) is really true.
2) By ignoring the averaging kernels of IASI and AIRS measurements, the authors do not allow for the strong correlations existing between retrieved profile grid points. The peaks of the averaging kernels of nadir sounders are overlapping each other, therefore, adapting the radiosonde vertical resolution with simple, non-overlapping, “box” kernels is not adequate. This is the reason why the authors find a substantial disagreement between the moisture inversion statistics determined from satellite and from adapted-radiosonde profiles. This makes very critical (and to my opinion unacceptable) the content of Section 4.3 of the manuscript.
3) As far as the presentation form is concerned, in scientific papers there is no need for lengthy text descriptions of the behavior of the lines shown in the plots: plots, on their own, are a tool to summarize the results. The text could be significantly shortened and should rather focus on the interpretation of the observed results (discrepancies, in this case).
Specific comments
L. 231: Which is the vertical resolution (FWHM of AKs) of AIRS retrieved humidity profiles? How many are the degrees of freedom?
L. 233: the profile inversion in the AIRS support product shown in Fig. 1C could be an artifact easily originated by the regularization used in the retrieval. Do you use the retrieval diagnostic to assign a “confidence level” to the detected humidity inversion layers?
L. 242: different statistics of AIRS and IASI measured profiles also due to the different strategy to handle cloudy measurements in AIRS and IASI retrievals. The two instruments / retrievals also smooth differently the horizontal structures in the atmosphere: the averaging area is 12 km for IASI and 39 km for AIRS. How do you account for these differences in the statistics presented later?
L. 289: why don’t you apply the same criterion based on measurement error also to IASI and AIRS profiles? Note that, while radiosonde profile grid points are mostly independent from each other, IASI and AIRS retrieved profile grid points are strongly correlated to each other, thus the correlations should be properly taken into account in this case.
L. 293: is this part of the analysis useful? Note that the procedure described is not equivalent to degrading the radiosonde profiles to the same vertical resolution of the satellite nadir sounders. To do that, averaging kernels of the related nadir retrievals should be used.
L. 294: “vertical grid” or “vertical resolution” ? I guess the vertical resolution is relevant in this context. Of course, vertical sampling must be adequate to the actual vertical resolution (see the Nyquist theorem).
L. 336 – 360: this lengthy descriptive text is not useful, a look at Fig. 3 already conveys all the information given in the text. The text is useful only when provides a physical interpretation of the results shown in the figure.
L. 367: Note that radiosonde profiles should be adjusted both to the vertical resolution and to the vertical grid of the IASI and AIRS Support retrievals. These adaptations can be achieved by convolving the radiosonde profiles with the averaging kernels of the retrieved IASI and AIRS profiles. Adaptation of the vertical grid only (as described in Sect. 3.2) is useless and the subsequent analysis highlights exactly this problem.
L. 372: If convolution with the averaging kernels is out of the scope of this work, then I suggest to remove this comparison: in general it is better to omit something rather than presenting something wrong.
L. 374 – 375: the number of 7 retrieval grid points (constant) is not similar to 2.5 degrees of freedom that IASI and AIRS profiles may show in polar regions. In case of 2.5 degrees of freedom, the 7 averaging kernels of the retrieved grid points may easily show overlapping peaks, thus they are not similar to the box functions that you are using to degrade the radiosonde profiles (see sect. 3.2).
L. 394 – 400: if the discrepancy in frequencies of inversion detection between HRRS and IGRA depends on how you processed HRRS measurements, why don’t you correct the procedure such that the same algorithm can be applied to profiles from all the considered sources?
Figure 4: the disagreement between HRRS and IGRA radiosondes (due mainly to different processing for inversion detection, see above) is rather concerning and casts doubts about the suitability of the two datasets to detect moisture inversions. Conversely, the other behaviors are rather obvious: as you degrade the profile sampling grid and vertical resolution, the sharp profile features become harder to detect and, even if detected, these show anyhow a reduced intensity due to the smoothing applied. Panel c) for p > 950 hPa seem to show and exception (see also line 387 of your paper) to this general, expected, behavior: do you have an explanation for it?
L. 421 – 446: the wordily description of the curves presented in Fig. 5 is not useful, one can just look at the figure and rapidly learn the results.
Section 4.2: Again, there is no point to compare IASI and AIRS statistics to that of “IGRA adjusted to AIRS” radiosonde profiles. First, the adjustment is not done with the correct kernels. Secondly, considering that the presence of clouds and humidity inversions are likely to be correlated, the comparison of IASI / AIRS statistics should be made towards that of the “matching and adjusted” radiosonde profiles and in similar cloud coverage conditions. I suggest to remove this section from the paper if you are not willing to correct these issues.
L. 450: the low vertical resolution of IASI and AIRS is not linked to the fact that they are passive sensors. It is rather linked to the nadir sounding technique used.
L. 498 – 500: to see if ERA5 is a suitable dataset to evaluate the statistics of humidity inversions, I would have started by comparing, at the various pressure levels, the width of the distribution of profile inversion depths (as obtained from radiosondes) to the actual vertical resolution of the ERA5 model. The fact that ERA5 performs better than IASI and AIRS (because of higher vertical resolution) is not surprising as the ECMWF model assimilates both IASI (MetOp-A/B) and AIRS measurements.
L. 525 and ff: this sentence is like saying that sharp vertical profile features cannot be detected using measurements with low vertical resolution. This is largely expected, it is not a “discovery” as presented here.
L. 537, “low vertical humidity information content of the passive remote sensing instruments”: this is not true, please delete this sentence. The fact that IASI and AIRS measurements are not suitable for your study is uniquely due to the coarse vertical resolution of IASI and AIRS profiles as compared to the vertical extension of the features you want to observe. The low vertical resolution of IASI and AIRS is related to the nadir sounding technique employed, e.g. passive limb sounding remote sensing measurements have a much finer vertical resolution.
L. 552 – 553: Does this mean that you can infer the correct statistics of humidity inversions if you correct for the effect of the insufficient vertical resolution of ERA5? I don’t think so. I think it would be better to re-phrase the sentence.
Technical / minor corrections
p. 5, fig. 1: please specify if the time is UTC or local.
L. 411: “rather large” with respect to what ?
Citation: https://doi.org/10.5194/amt-2022-22-RC2 -
RC3: 'Comment on amt-2022-22', Anonymous Referee #3, 23 Jun 2022
Review of "Can state-of-the-art infrared satellite sounders and reanalyses detect moisture inversions in the Arctic?", Chellini et Ebelll, AMTD 2022.
The manuscript presents an important systematic work in view of radiative budget and climate studies. The work analyses and documents the strengths and limitations of numerical models and of satellite products complementarily to sondes which, thought extremely accurate, have very sparse coverage and cannot support global climate studies.I find the manuscript very well written: the structure is clear and the explanation are sufficiently detailed, with concise but yet complete illustrations.
I recommend the publication of the manuscript with minor revisions, considering the general and specific comments below.
General comments:
The hyperspectral IR Nadir sounders don't have the resolving power necessary to accurately represent the inversions like sondes do, especially surface-based and on short vertical scales. However, sounding products often contain at least inflexions where not-too-small temperature inversions occur. If observed with humidity too, even though not accurately describing the inversions, could this be interesting information for climate studies? I feel that this discussion is missing in the paper and could be more interesting than making the (expected) experience that those sounders lack vertical resolution, especially near surface.Also, it would be useful to be specific that the assessment is really with IASI and AIRS products of a given version - as opposed to IASI and AIRS as a mission. It seems a bit pedantic, but the measurements may or may not be exploited at their best, especially in the context of polar sounding with humidity inversions. Even the title should perhaps be explicit about an assessement of IASI and AIRS L2 (versions TBC). Recent work by Prange et al, seem to indicate that IASI sounding could detect elevated moisture inversions in some instances.
Confirm wether the IASI L2 products have been used correctly: all-sky sounding and quality information seem to have been overlooked. Same with ignoring scene-dependent or at least typical averaging kernels to convolve with radiosonde profiles. This could be important to this study.
Evaluate and discuss if the numerical model is as skilled away from sonde stations as it is in Ny Alesund in representing humidity inversions. At least from a statistical stand-point, do we get numbers close to those of Ny-Alesund?
Specific comments:
L9-L11: I understand the point, but I feel it could advantageously be presented from a different perspective. The point is not so much the vertical grid sampling the retrieval (and analyses) may have been performed on, but more the vertical resolution intrinsic to the passive IR sounders - assuming the measurements are exploited to the maximum extent. The actual vertical resolution being usually significantly smaller than the vertical grid sampling (which is true for numerical models too). So the question is more about what are the vertical scales that matter for climate and radiative budget studies -which obviously super-resolved radiosonde profiles can represent- wrt how much of it can be captured by the passive satellite IR sounders. Perhaps the title could be revised a bit along these lines.You are finding that, as expected, IR sounders can catch some (stronger) inversions but usually miss most (medium and small) and underestimate their intensity. Similar findings were made regarding surface-base temperature inversion in Antarctica. See paper, which I suggest to include in the reference.
Boylan P. et al., "Identification and intercomparison of surface-based inversions over Antarctica from IASI, ERA-Interim, and Concordiasi dropsonde data", 2016, https://doi.org/10.1002/2015JD024724L14-15: consistent with Boylan et al for temperature.
L17-19: It is essential to state whether the radiosondes used in your assessment have been assimilated in the ERA-5 reanalyses. Also to note somewhere if you happened to have used radiosonding away from Ny-Alesund. Indeed, because of the regular synoptic sounding from this station, it can be expected that the numerical model is particularly well constrained in that area. I suggest testing the model abilities away from the regular radiosonding stations, e.g. by computing similar statistics on inversions and see if number are close to those found in Ny Alesund.
Fig.1 and throughout: Use "Metop" official case after EUMETSAT standards (not MetOp)
L93: yes, IASI measures through a broader spectral range than AIRS, but more importantly for water-vapour sounding, with higher spectral resolution (a little over twice higher in the WV band).
L97: same as before, important to mention somewhere (the earlier the better I think) whether these sondes have been assimilated by the numerical models.
L113-114: well in fact, very often the grids are oversampled, so really we should be smoothing the radiosondes with the averaging kernels of the retrievals, which are normally provided along with the products, or with proxy AK (e.g. static seasonal AK typical for this region). This would be more of a validation exercise of the given IASI and AIRS products. But it remains more interesting to explore what these instruments can contribute (or not) to the monitoring of humidity inversions given their instrinsic vertical resolution/sensitivity. E.g. what Fig.4 shows, would that be sufficient?
L155: strictly speaking, NWP assimilates radiances, not profiles. I suggest rephrasing: "is to provide such _thermodynamic information_ with high resolution..."
L156: "the nominal accuracy" ==> the 1K/1km was the end-user requirements spelled at the beginning of the mission. IASI L2 Temperature do actually perform better than these objectives in the mid troposphere.
L159: the actual resolution of the IASI L1C products (radiances) after homogeneisation (i.e. removing instrument signature and wavelength dependency) is atually 0.5cm-1 (sampling is 0.25cm-1).
L159: MetOp ==> Metop + suggest reference Klaes et al. on the EPS programme. "The EUMETSAT Polar System - 13+ Successful Years of Global Observations for Operational Weather Prediction and Climate Monitoring", March 2021, Bulletin of the American Meteorological Society 102(6):1-34, DOI: 10.1175/BAMS-D-20-0082.1
L185 and L186 is a confusion: OEM attempted in clear-sky as per cloud mask. Double-check.L217: Actually, AIRS V6 includes a combined MW+IR retrieval scheme, which was the nominal strategy. However it became not functional after the loss of AMSU-A2 in 2016 (https://docserver.gesdisc.eosdis.nasa.gov/repository/Mission/AIRS/3.3_ScienceDataProductDocumentation/3.3.5_ProductQuality/V6_Test_Report_Supplement_Performance_of_AIRS+AMSU_vs_AIRS-Only_Retrievals.pdf)
It also includes a statistical first retrieval, implementing ANN - first guess to the physical retrieval, similarly to IASI. (https://docserver.gesdisc.eosdis.nasa.gov/repository/Mission/AIRS/3.3_ScienceDataProductDocumentation/3.3.4_ProductGenerationAlgorithms/V6_L2_Product_User_Guide.pdf)
L238-243: "IASI retrieval under specific sky conditions" and "No quality control is provided" ==> this is wrong. IASI L2 from EUMETSAT include all-sky sounding, nominally exploiting the microwave in synergy with IASI in the statistical retrievals (this is actually the regional service). The OEM usually only improves little over the first retrieval. The operational products also include quality indicators and error estimates, which have been used in other studies. The cloud mask in v6.5 (2020-) showed improvements over v6.4 (used here), which could play a role in the results here.
Why have all-sky and QC info been discarded? I suggest liaising with the products developers if that has just been omitted in your study.L265: yes, indeed.
L372: "out of scope": why has the utilisation of AK been excluded from this study? Out of simplicity? I think it is limiting somewhat the conclusiveness and should be explained.
L372: "We argue...": the syntax reads weird, not sure to get the point here.
L375: I see the rationale, but the 2.5-7.2 hinge points might be placed differently in the vertical. Also, I would expect DoF to be on the lower side in the Arctic anyway, I suggest commenting on quick typical DoF figures in the Arctic.
L405-408: you would get more match-ups with IASI all-sky retrievals.
L411: a word of cautious with the footprint size: IASI 12km is at Nadir. It goes up to 40km on swath edge. Similarly AIRS 50km at Nadir would flirt with 200km on swath edge. Would consequence could it have on humidity inversion statistics and in particular compared to point-measurements with radiosondes?
L412-413: Here again, IASI L2 contain all-sky retrievals.
L414-420: I find the difference between IASI-A and IASI-B puzzling/interesting. Big enough to be noted and it seems statistically significant. This was not expected, why is that?
L446: interesting result if genuine retrieved information, which would suggest that this should be at reach for IASI - but not yet in this product version.
L450: yes, indeed. For example, for IASI, a parallel to the 1K/1km figure for temperature was 10%RH/2km for humidity (in good conditions, how many DoF over Polar regimes?)... I thnk this rough figure should be accounted for in the preliminary discussions and in introducing the objective of this work.
4.4: I think we're hitting the same concept as with AIRS/IASI: the vertical sampling of ERA-5 is higher than the actual vertical resolution of the physical processes modelled and information assimilated in general. Are there Kernel functions for the models? that notion could be recalled in this paragraph too.
L497: yes, the radiosonde assimilated on this very point could well explain the good fit. The ERA-5 mid-range forecasts before an inversion occurs could be looked into to test this assumption. Also checking the frequency of low-level inversion in other places empty of radiosonde to constrain the model near surface, compared to the stats in Ny-Alesund is felt missing and would be very informative in this discussion.
L545: the difference IASI-A and IASI-B is unexplained while the IR+MW measurements are very comparable in nature (Metop-A having slightly less MW channels than -B at that time). It is certainly worth highlighting, feed-back to the products developpers and good information to other products users overall.
L576: yes, why hasn't it be done in this study? "mostly based on MW" ==> that is not correct. The first guess exploits MW and "constraints" the OEM. But MW only really add value over IASI for the part of the atmosphere affected by clouds.Citation: https://doi.org/10.5194/amt-2022-22-RC3 -
EC1: 'Comment on amt-2022-22', Martin Riese, 28 Jun 2022
Dear authors,
The interactive public review and discussion revealed major deficiencies. I can only encourage you to resubmit a revised manuscript, if you can satisfactorily address all comments upon resubmission deadline, e. g. by applying satellite averaging kernels to the radiosonde profiles and by reorganizing the work in order to address the question of detection limits etc. as requested by the reviewers. If this is not feasible in a timely manner, I would recommend a new submission at a later date.
With best regards,
Martin Riese
Citation: https://doi.org/10.5194/amt-2022-22-EC1 -
AC1: 'Reply on EC1', Giovanni Chellini, 15 Jul 2022
Dear Dr. Riese,
we would like to thank the three referees for the time they put in to review our manuscript, as well as you for taking on the role of editor. We are especially grateful to the reviewers for highlighting the need for major changes in our study, and we intend to adjust our approach and methods based on their comments. Unfortunately the time we need to apply said changes far exceeds the time given by the journal for the revision. We have thus decided to withdraw the manuscript, and resubmit a revised version at a later date.
Kind regards,
Giovanni Chellini and Kerstin EbellCitation: https://doi.org/10.5194/amt-2022-22-AC1
-
AC1: 'Reply on EC1', Giovanni Chellini, 15 Jul 2022
Interactive discussion
Status: closed
-
RC1: 'Comment on amt-2022-22', Anonymous Referee #2, 02 Jun 2022
Chellini and Ebell compare statistics of moisture inversions in the Arctic between datasets of the nadir thermal IR sounding instruments IASI and AIRS, the ECWMF ERA5 reanalysis data and radiosoundings as reference measurements. They draw as their main conclusions (a) that the nadir-sounders strongly underestimate inversions in specific humidity and, (b) that the reanalysis datasets characterize such profiles quite well when compared to radiosonde profiles with adapted vertical resolution.
In my opinion, this is a valuable and clear study expressing the limitations of satellite moisture data with a focus on specific features of moisture vertical profiles in Arctic regions. Thus, it helps also to prevent inappropriate use of those datasets. My main criticism concerns the method applied to adapt the vertical resolution of radiosonde data to that of the nadir-sounders. As the authors state themselves, the correct method would have been the application of the satellite instruments’ averaging kernels to the radiosonde profiles. In this study, only a vertical interpolation and averaging is performed which does by no means simulate typical nadir sounding averaging kernels. That this is wrong can be seen by the fact that the statistical analysis of humidity inversions when applying their averaging to the radiosoundings (Fig. 4) is largely different from the analysis on the satellite data themselves (Fig. 5). If the real averaging kernels were applied, the differences should be much smaller. Thus, this exercise only shows that the wrong ‘resolution-reduction’ has been applied to the radiosonde dataset. Therefore, I would draw the conclusion that the related sections (3.2,4.2) should better be omitted since they do not really contribute any new insights and rely on a very inaccurate method.
Similarly, this would apply to the comparison of resolution reduced radiosoundings and ERA5. However, this may be more appropriate due to the presumably better behaved vertical ‘kernel’ and resolution of ERA5 compared to nadir-sounding instruments.
In summary I would recommend this work for publication in AMT, if the major comment, as expressed above, is taken into consideration.
Specific comments:
L265:
Please specify if IASI/AIRS data is also assimilated in ERA5.
L375, ‘Standard Product (7 between 1000 and 400 hPa) is close to the typical number of degrees of freedom of the IASI retrievals (e.g., ranging between 2.5 and 7.2, according to Ebell et al. (2013))’:
It is not close but at the upper, optimistic end of the typical degrees of freedom reached by IASI retrievals.
L451, ‘IASI seems to be able to capture better inversion frequency below the 800-hPa level in winter.’, also L15 (abstract) ‘A better agreement has been found for IASI below the 900-hPa level and in particular for winter’:
Considering the fact, that thermal IR nadir sounders generally have a very low sensitivity on the lowermost atmospheric layers (due to the small thermal contrast between surface and lower atmospheric temperature), I doubt that this ‘better’ achievement is due to real information from the measurement but may stem from retrieval artefacts. At least this should be a part of the discussion.
L566, ‘… If ERA5 showed a similar good performance, the data could also be used as a reference to be compared to the satellite retrievals across the whole Arctic region’:
It should be emphasized here that this argument is valid only in case the radiosoundings used for comparison are really independent from the reanalysis, have not been assimilated and are locally distinct from radiosoundings used in the assimilation.
Citation: https://doi.org/10.5194/amt-2022-22-RC1 -
RC2: 'Comment on amt-2022-22', Anonymous Referee #1, 16 Jun 2022
The mauscript of Chellini and Ebell illustrates the results of a study aiming to answer the question of whether humidity profiles retrieved from infrared satellite (nadir) sounders and ECMWF reanalyses can be used for the detection of moisture inversion layers in the Arctic.
The authors have substantially ignored my initial comments, therefore, I am sorry that I have to repeat them here and, still, I cannot recommend the manuscript for publication in the Atmospheric Measurement Techniques journal. The motivations will be clear from the comments included hereafter.
General comments
1) Independently of the vertical grid that may have been used for profile retrievals, at low altitudes (0 – 10 km), the vertical resolution (i.e. the full width at half maximum of the averaging kernels) of IASI water vapour profiles is of the order of 3 – 5 km. For this sole reason, one would not even try to use the profiles retrieved from these missions to detect moisture inversions in 100 hPa layers. A more sound question would be: which is the minimum size (in terms of strength and depth) versus altitude, of moisture inversions that can be detected from IASI and AIRS retrieved profiles? To my opinion the work could be re-organized in the attempt to answer this question. As it is now, the manuscript only shows that the trivial answer (no) to the question set in the title (at least the part regarding satellite sounders) is really true.
2) By ignoring the averaging kernels of IASI and AIRS measurements, the authors do not allow for the strong correlations existing between retrieved profile grid points. The peaks of the averaging kernels of nadir sounders are overlapping each other, therefore, adapting the radiosonde vertical resolution with simple, non-overlapping, “box” kernels is not adequate. This is the reason why the authors find a substantial disagreement between the moisture inversion statistics determined from satellite and from adapted-radiosonde profiles. This makes very critical (and to my opinion unacceptable) the content of Section 4.3 of the manuscript.
3) As far as the presentation form is concerned, in scientific papers there is no need for lengthy text descriptions of the behavior of the lines shown in the plots: plots, on their own, are a tool to summarize the results. The text could be significantly shortened and should rather focus on the interpretation of the observed results (discrepancies, in this case).
Specific comments
L. 231: Which is the vertical resolution (FWHM of AKs) of AIRS retrieved humidity profiles? How many are the degrees of freedom?
L. 233: the profile inversion in the AIRS support product shown in Fig. 1C could be an artifact easily originated by the regularization used in the retrieval. Do you use the retrieval diagnostic to assign a “confidence level” to the detected humidity inversion layers?
L. 242: different statistics of AIRS and IASI measured profiles also due to the different strategy to handle cloudy measurements in AIRS and IASI retrievals. The two instruments / retrievals also smooth differently the horizontal structures in the atmosphere: the averaging area is 12 km for IASI and 39 km for AIRS. How do you account for these differences in the statistics presented later?
L. 289: why don’t you apply the same criterion based on measurement error also to IASI and AIRS profiles? Note that, while radiosonde profile grid points are mostly independent from each other, IASI and AIRS retrieved profile grid points are strongly correlated to each other, thus the correlations should be properly taken into account in this case.
L. 293: is this part of the analysis useful? Note that the procedure described is not equivalent to degrading the radiosonde profiles to the same vertical resolution of the satellite nadir sounders. To do that, averaging kernels of the related nadir retrievals should be used.
L. 294: “vertical grid” or “vertical resolution” ? I guess the vertical resolution is relevant in this context. Of course, vertical sampling must be adequate to the actual vertical resolution (see the Nyquist theorem).
L. 336 – 360: this lengthy descriptive text is not useful, a look at Fig. 3 already conveys all the information given in the text. The text is useful only when provides a physical interpretation of the results shown in the figure.
L. 367: Note that radiosonde profiles should be adjusted both to the vertical resolution and to the vertical grid of the IASI and AIRS Support retrievals. These adaptations can be achieved by convolving the radiosonde profiles with the averaging kernels of the retrieved IASI and AIRS profiles. Adaptation of the vertical grid only (as described in Sect. 3.2) is useless and the subsequent analysis highlights exactly this problem.
L. 372: If convolution with the averaging kernels is out of the scope of this work, then I suggest to remove this comparison: in general it is better to omit something rather than presenting something wrong.
L. 374 – 375: the number of 7 retrieval grid points (constant) is not similar to 2.5 degrees of freedom that IASI and AIRS profiles may show in polar regions. In case of 2.5 degrees of freedom, the 7 averaging kernels of the retrieved grid points may easily show overlapping peaks, thus they are not similar to the box functions that you are using to degrade the radiosonde profiles (see sect. 3.2).
L. 394 – 400: if the discrepancy in frequencies of inversion detection between HRRS and IGRA depends on how you processed HRRS measurements, why don’t you correct the procedure such that the same algorithm can be applied to profiles from all the considered sources?
Figure 4: the disagreement between HRRS and IGRA radiosondes (due mainly to different processing for inversion detection, see above) is rather concerning and casts doubts about the suitability of the two datasets to detect moisture inversions. Conversely, the other behaviors are rather obvious: as you degrade the profile sampling grid and vertical resolution, the sharp profile features become harder to detect and, even if detected, these show anyhow a reduced intensity due to the smoothing applied. Panel c) for p > 950 hPa seem to show and exception (see also line 387 of your paper) to this general, expected, behavior: do you have an explanation for it?
L. 421 – 446: the wordily description of the curves presented in Fig. 5 is not useful, one can just look at the figure and rapidly learn the results.
Section 4.2: Again, there is no point to compare IASI and AIRS statistics to that of “IGRA adjusted to AIRS” radiosonde profiles. First, the adjustment is not done with the correct kernels. Secondly, considering that the presence of clouds and humidity inversions are likely to be correlated, the comparison of IASI / AIRS statistics should be made towards that of the “matching and adjusted” radiosonde profiles and in similar cloud coverage conditions. I suggest to remove this section from the paper if you are not willing to correct these issues.
L. 450: the low vertical resolution of IASI and AIRS is not linked to the fact that they are passive sensors. It is rather linked to the nadir sounding technique used.
L. 498 – 500: to see if ERA5 is a suitable dataset to evaluate the statistics of humidity inversions, I would have started by comparing, at the various pressure levels, the width of the distribution of profile inversion depths (as obtained from radiosondes) to the actual vertical resolution of the ERA5 model. The fact that ERA5 performs better than IASI and AIRS (because of higher vertical resolution) is not surprising as the ECMWF model assimilates both IASI (MetOp-A/B) and AIRS measurements.
L. 525 and ff: this sentence is like saying that sharp vertical profile features cannot be detected using measurements with low vertical resolution. This is largely expected, it is not a “discovery” as presented here.
L. 537, “low vertical humidity information content of the passive remote sensing instruments”: this is not true, please delete this sentence. The fact that IASI and AIRS measurements are not suitable for your study is uniquely due to the coarse vertical resolution of IASI and AIRS profiles as compared to the vertical extension of the features you want to observe. The low vertical resolution of IASI and AIRS is related to the nadir sounding technique employed, e.g. passive limb sounding remote sensing measurements have a much finer vertical resolution.
L. 552 – 553: Does this mean that you can infer the correct statistics of humidity inversions if you correct for the effect of the insufficient vertical resolution of ERA5? I don’t think so. I think it would be better to re-phrase the sentence.
Technical / minor corrections
p. 5, fig. 1: please specify if the time is UTC or local.
L. 411: “rather large” with respect to what ?
Citation: https://doi.org/10.5194/amt-2022-22-RC2 -
RC3: 'Comment on amt-2022-22', Anonymous Referee #3, 23 Jun 2022
Review of "Can state-of-the-art infrared satellite sounders and reanalyses detect moisture inversions in the Arctic?", Chellini et Ebelll, AMTD 2022.
The manuscript presents an important systematic work in view of radiative budget and climate studies. The work analyses and documents the strengths and limitations of numerical models and of satellite products complementarily to sondes which, thought extremely accurate, have very sparse coverage and cannot support global climate studies.I find the manuscript very well written: the structure is clear and the explanation are sufficiently detailed, with concise but yet complete illustrations.
I recommend the publication of the manuscript with minor revisions, considering the general and specific comments below.
General comments:
The hyperspectral IR Nadir sounders don't have the resolving power necessary to accurately represent the inversions like sondes do, especially surface-based and on short vertical scales. However, sounding products often contain at least inflexions where not-too-small temperature inversions occur. If observed with humidity too, even though not accurately describing the inversions, could this be interesting information for climate studies? I feel that this discussion is missing in the paper and could be more interesting than making the (expected) experience that those sounders lack vertical resolution, especially near surface.Also, it would be useful to be specific that the assessment is really with IASI and AIRS products of a given version - as opposed to IASI and AIRS as a mission. It seems a bit pedantic, but the measurements may or may not be exploited at their best, especially in the context of polar sounding with humidity inversions. Even the title should perhaps be explicit about an assessement of IASI and AIRS L2 (versions TBC). Recent work by Prange et al, seem to indicate that IASI sounding could detect elevated moisture inversions in some instances.
Confirm wether the IASI L2 products have been used correctly: all-sky sounding and quality information seem to have been overlooked. Same with ignoring scene-dependent or at least typical averaging kernels to convolve with radiosonde profiles. This could be important to this study.
Evaluate and discuss if the numerical model is as skilled away from sonde stations as it is in Ny Alesund in representing humidity inversions. At least from a statistical stand-point, do we get numbers close to those of Ny-Alesund?
Specific comments:
L9-L11: I understand the point, but I feel it could advantageously be presented from a different perspective. The point is not so much the vertical grid sampling the retrieval (and analyses) may have been performed on, but more the vertical resolution intrinsic to the passive IR sounders - assuming the measurements are exploited to the maximum extent. The actual vertical resolution being usually significantly smaller than the vertical grid sampling (which is true for numerical models too). So the question is more about what are the vertical scales that matter for climate and radiative budget studies -which obviously super-resolved radiosonde profiles can represent- wrt how much of it can be captured by the passive satellite IR sounders. Perhaps the title could be revised a bit along these lines.You are finding that, as expected, IR sounders can catch some (stronger) inversions but usually miss most (medium and small) and underestimate their intensity. Similar findings were made regarding surface-base temperature inversion in Antarctica. See paper, which I suggest to include in the reference.
Boylan P. et al., "Identification and intercomparison of surface-based inversions over Antarctica from IASI, ERA-Interim, and Concordiasi dropsonde data", 2016, https://doi.org/10.1002/2015JD024724L14-15: consistent with Boylan et al for temperature.
L17-19: It is essential to state whether the radiosondes used in your assessment have been assimilated in the ERA-5 reanalyses. Also to note somewhere if you happened to have used radiosonding away from Ny-Alesund. Indeed, because of the regular synoptic sounding from this station, it can be expected that the numerical model is particularly well constrained in that area. I suggest testing the model abilities away from the regular radiosonding stations, e.g. by computing similar statistics on inversions and see if number are close to those found in Ny Alesund.
Fig.1 and throughout: Use "Metop" official case after EUMETSAT standards (not MetOp)
L93: yes, IASI measures through a broader spectral range than AIRS, but more importantly for water-vapour sounding, with higher spectral resolution (a little over twice higher in the WV band).
L97: same as before, important to mention somewhere (the earlier the better I think) whether these sondes have been assimilated by the numerical models.
L113-114: well in fact, very often the grids are oversampled, so really we should be smoothing the radiosondes with the averaging kernels of the retrievals, which are normally provided along with the products, or with proxy AK (e.g. static seasonal AK typical for this region). This would be more of a validation exercise of the given IASI and AIRS products. But it remains more interesting to explore what these instruments can contribute (or not) to the monitoring of humidity inversions given their instrinsic vertical resolution/sensitivity. E.g. what Fig.4 shows, would that be sufficient?
L155: strictly speaking, NWP assimilates radiances, not profiles. I suggest rephrasing: "is to provide such _thermodynamic information_ with high resolution..."
L156: "the nominal accuracy" ==> the 1K/1km was the end-user requirements spelled at the beginning of the mission. IASI L2 Temperature do actually perform better than these objectives in the mid troposphere.
L159: the actual resolution of the IASI L1C products (radiances) after homogeneisation (i.e. removing instrument signature and wavelength dependency) is atually 0.5cm-1 (sampling is 0.25cm-1).
L159: MetOp ==> Metop + suggest reference Klaes et al. on the EPS programme. "The EUMETSAT Polar System - 13+ Successful Years of Global Observations for Operational Weather Prediction and Climate Monitoring", March 2021, Bulletin of the American Meteorological Society 102(6):1-34, DOI: 10.1175/BAMS-D-20-0082.1
L185 and L186 is a confusion: OEM attempted in clear-sky as per cloud mask. Double-check.L217: Actually, AIRS V6 includes a combined MW+IR retrieval scheme, which was the nominal strategy. However it became not functional after the loss of AMSU-A2 in 2016 (https://docserver.gesdisc.eosdis.nasa.gov/repository/Mission/AIRS/3.3_ScienceDataProductDocumentation/3.3.5_ProductQuality/V6_Test_Report_Supplement_Performance_of_AIRS+AMSU_vs_AIRS-Only_Retrievals.pdf)
It also includes a statistical first retrieval, implementing ANN - first guess to the physical retrieval, similarly to IASI. (https://docserver.gesdisc.eosdis.nasa.gov/repository/Mission/AIRS/3.3_ScienceDataProductDocumentation/3.3.4_ProductGenerationAlgorithms/V6_L2_Product_User_Guide.pdf)
L238-243: "IASI retrieval under specific sky conditions" and "No quality control is provided" ==> this is wrong. IASI L2 from EUMETSAT include all-sky sounding, nominally exploiting the microwave in synergy with IASI in the statistical retrievals (this is actually the regional service). The OEM usually only improves little over the first retrieval. The operational products also include quality indicators and error estimates, which have been used in other studies. The cloud mask in v6.5 (2020-) showed improvements over v6.4 (used here), which could play a role in the results here.
Why have all-sky and QC info been discarded? I suggest liaising with the products developers if that has just been omitted in your study.L265: yes, indeed.
L372: "out of scope": why has the utilisation of AK been excluded from this study? Out of simplicity? I think it is limiting somewhat the conclusiveness and should be explained.
L372: "We argue...": the syntax reads weird, not sure to get the point here.
L375: I see the rationale, but the 2.5-7.2 hinge points might be placed differently in the vertical. Also, I would expect DoF to be on the lower side in the Arctic anyway, I suggest commenting on quick typical DoF figures in the Arctic.
L405-408: you would get more match-ups with IASI all-sky retrievals.
L411: a word of cautious with the footprint size: IASI 12km is at Nadir. It goes up to 40km on swath edge. Similarly AIRS 50km at Nadir would flirt with 200km on swath edge. Would consequence could it have on humidity inversion statistics and in particular compared to point-measurements with radiosondes?
L412-413: Here again, IASI L2 contain all-sky retrievals.
L414-420: I find the difference between IASI-A and IASI-B puzzling/interesting. Big enough to be noted and it seems statistically significant. This was not expected, why is that?
L446: interesting result if genuine retrieved information, which would suggest that this should be at reach for IASI - but not yet in this product version.
L450: yes, indeed. For example, for IASI, a parallel to the 1K/1km figure for temperature was 10%RH/2km for humidity (in good conditions, how many DoF over Polar regimes?)... I thnk this rough figure should be accounted for in the preliminary discussions and in introducing the objective of this work.
4.4: I think we're hitting the same concept as with AIRS/IASI: the vertical sampling of ERA-5 is higher than the actual vertical resolution of the physical processes modelled and information assimilated in general. Are there Kernel functions for the models? that notion could be recalled in this paragraph too.
L497: yes, the radiosonde assimilated on this very point could well explain the good fit. The ERA-5 mid-range forecasts before an inversion occurs could be looked into to test this assumption. Also checking the frequency of low-level inversion in other places empty of radiosonde to constrain the model near surface, compared to the stats in Ny-Alesund is felt missing and would be very informative in this discussion.
L545: the difference IASI-A and IASI-B is unexplained while the IR+MW measurements are very comparable in nature (Metop-A having slightly less MW channels than -B at that time). It is certainly worth highlighting, feed-back to the products developpers and good information to other products users overall.
L576: yes, why hasn't it be done in this study? "mostly based on MW" ==> that is not correct. The first guess exploits MW and "constraints" the OEM. But MW only really add value over IASI for the part of the atmosphere affected by clouds.Citation: https://doi.org/10.5194/amt-2022-22-RC3 -
EC1: 'Comment on amt-2022-22', Martin Riese, 28 Jun 2022
Dear authors,
The interactive public review and discussion revealed major deficiencies. I can only encourage you to resubmit a revised manuscript, if you can satisfactorily address all comments upon resubmission deadline, e. g. by applying satellite averaging kernels to the radiosonde profiles and by reorganizing the work in order to address the question of detection limits etc. as requested by the reviewers. If this is not feasible in a timely manner, I would recommend a new submission at a later date.
With best regards,
Martin Riese
Citation: https://doi.org/10.5194/amt-2022-22-EC1 -
AC1: 'Reply on EC1', Giovanni Chellini, 15 Jul 2022
Dear Dr. Riese,
we would like to thank the three referees for the time they put in to review our manuscript, as well as you for taking on the role of editor. We are especially grateful to the reviewers for highlighting the need for major changes in our study, and we intend to adjust our approach and methods based on their comments. Unfortunately the time we need to apply said changes far exceeds the time given by the journal for the revision. We have thus decided to withdraw the manuscript, and resubmit a revised version at a later date.
Kind regards,
Giovanni Chellini and Kerstin EbellCitation: https://doi.org/10.5194/amt-2022-22-AC1
-
AC1: 'Reply on EC1', Giovanni Chellini, 15 Jul 2022
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 600 | 177 | 53 | 830 | 45 | 46 |
- HTML: 600
- PDF: 177
- XML: 53
- Total: 830
- BibTeX: 45
- EndNote: 46
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1