Characterizing the lower troposphere by combining GNSS radio occultation and nadir sounding observations in a tomographic approach

Wang, Kuo-Nung; Ao, Chi O.; Hajj, George A.; Morris, Mary G.; Moore, Angelyn W.

doi:10.5194/egusphere-2026-4427

Preprints

https://doi.org/10.5194/egusphere-2026-4427

Preprints

11 Aug 2026

| 11 Aug 2026

Characterizing the lower troposphere by combining GNSS radio occultation and nadir sounding observations in a tomographic approach

Kuo-Nung Wang, Chi O. Ao, George A. Hajj, Mary G. Morris, and Angelyn W. Moore

Abstract. Characterizing the moist thermodynamic structure of the lower troposphere (LT) from spaceborne observations is challenging. This is largely due to the LT's highly variable moisture distribution which requires high-resolution observations in both the vertical and horizontal dimensions. Two of the most impactful spaceborne remote sensing technologies in numerical weather prediction (NWP) are (1) the global navigation satellite system – radio occultation (GNSS-RO) and (2) nadir sounding instruments which include microwave radiometers (MWR) and infrared sounders (IR). These two technologies provide complementary information of atmospheric temperature and water vapor. GNSS-RO provides high vertical resolution (∼200 m) with near zero interference from hydrometeor, but its horizontal resolution along the ray path is coarse (>100 km). On the other hand, nadir sounders measure brightness temperature (TB) that can be related to the temperature and water vapor structure in the atmospheric column with higher horizontal resolution (∼25 km). However, retrievals from passive sounders have lower vertical resolution (>2 km), and are complicated by precipitation, clouds, and surface emissivity uncertainty over land.

In this study we combine these two complementary types of observations by use of a novel tomography method to improve the 3D characterization of water vapor in the LT. This is done by adjusting the water vapor density along the RO links to match the RO observations, while preserving the horizontal water vapor variability based on information derived from the nadir sounders. We test this method using simulations based on mesoscale model outputs from the Weather Research and Forecasting Model (WRF), as well as real observations from RO and Cross-track Infrared Microwave Sounder Suite (CrIMSS). We demonstrate that the tomographic retrieval can resolve the complex moisture structure better than what is possible from either measurement alone.

Received: 23 Jul 2026 – Discussion started: 11 Aug 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Kuo-Nung Wang, Chi O. Ao, George A. Hajj, Mary G. Morris, and Angelyn W. Moore

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-4427', Anonymous Referee #1, 15 Sep 2026
Review of: Characterizing the lower troposphere by combining GNSS radio occultation and nadir sounding observations in a tomographic approach; Author(s): Kuo-Nung Wang et al. MS No.: egusphere-2026-4427
Comments:
General:
an interesting manuscript / idea on how to make more use of simultaneous / collocated measurements for improved lower troposphere and PBL sampling

Major Comments:
the abstract is unclear on which instruments are actually combined. Please clearly state in the abstract that it looks at RO and MWR only.

the assumption of unbiased refractivity data, not only from RO, but also from MWR, needs to be further discussed. The proposed method on RO by Wang 2017 requires precipitable water (PW) info. Is that as well determined from the MWR? Or a combination of MWR/IR in the future? On uncertainties of the MWR temperature and water vapour data (nadir data) that is used in input to the initial refractivity field, more info should be provided regarding the retrieval issues/limitations, e.g., over land and sea. Surface emissivity is not a problem over sea, but land ones introduce uncertainties. You mention on L195 that the data needs to be “sufficiently accurate”. What is exactly sufficient?

Related to potential problems of MWR retrievals over land, can you include a little validation of the results obtained by combining COSMIC-2 and CrIMSS? Is the correction e.g., aligned with global NWP data? Also, aren’t there any other RO observations in that region available, so that more RO data is used? Or focus on a different period, e.g., when ROMEX data was available in large numbers?

I suggest to point out e.g., in the conclusions, that this might be an interesting approach for a regional NWP model, using refractivity. It would allow 2/3D use of RO data, potentially in an efficient implementation. Can you actually also add a word about computing times/efficiency of the algorithm?

ECMWF already uses 2D bending angle assimilation with a highly efficient algorithm, and more global NWP models move that way. Any comments on whether this global NWP approach, or your proposed refractivity approach, have any advantages / disadvantages?

Minor Comments:
Eq 2: mention more accurate representations are available, even if for a simulation SW53 is good enough. It might though be problematic for the real data investigation, can you clarify this further in the discussion, also in Section 3.2?

L145: is the ionosphere included in the model? I suspect not, as you’re only looking into the lower troposphere. But maybe mention this.

L197: this “negligible” of course assumes the a priori and estimated field is of high quality already. Thus the retrieval needs to work very well under all conditions, up to near ducting. Do you need an additional QC to assure this assumption is valid, even for extremer cases? Assess if the 1m per voxel stated is not violated with the a priori and the final solution?

Figure 3: are RO observations assumed to be vertical? Which is mostly not the case.

Figure 3: are RO observations assumed to be vertical? Which is mostly not the case, in particular at lower altitudes. Please clarify this, and also potential limitations when using real data.

Figure 4: is the MWR retrieval realistic? It appears to have a vertical resolution rather fine scale, in the hundreds of meters. And the smoothing (as well as actual instrument data) will lead to strong correlated data. How is that impacting the retrieval solution? Being no MWR expert, I wonder if a boxcar does lead to the same result as an actual retrieval process, with an averaging kernel that has a much wider smoothing. In other word: is a boxcar on the retrieved variable leading to the same result as a retrieval process from brightness temperatures?

L310: I assume all RO observations have the same uncertainty, independent on constellation or observation geometry. Suggest to mention it.

Figure 7: the occultations appear to still align quite well with the 33 Degree plane. Are the satellites (MWR, RO constellation) in similar orbit planes?

Figure 8 and prior ones: the dashed circle of interest is about 200km, but the RO observation is rather a line of 200km, so maybe do a narrow ellipsoid instead? This would also show the observation direction more clearly.

L358: are these 2 instrument observations taken exactly at the same time?

Editorial:
L403: editorial - the CRIMSS acronym is mostly a combination of lower/upper case letters in the manuscript
Citation: https://doi.org/10.5194/egusphere-2026-4427-RC1
RC2:
'Comment on egusphere-2026-4427', Anonymous Referee #2, 22 Oct 2026
The paper describes a tomographic approach to use complementary information from the nadir looking sounders and gnss radio occultation. The paper describes the methodology used to obtain the refractivity fields in 3D, and they evaluate their solutions using simulated data using WRF, and then they apply the methodology to a real case.
The paper is very interesting, they describe a methodology that is relatively simple and yet they show, using simulated data, that the information brought by the RO in the vertical effectively improves the vertical structure of the nadir sounders’ water vapor. The paper is well written, well structured, and I believe it deserves publication.
I also have some comments that I think they deserve, at least, some more discussion. And then a list of other minor comments.
To discuss further:
- Most of the current RO inversion methods strongly rely on spherical symmetry: however, here the cases you want to “improve” here are cases that show strong horizontal gradients, situations where the spherical symmetry is broken. It could be worth including a short discussion, or some references, that assess the errors in the retrievals affected by these strong horizontal gradients. Basically, to state whether the expected inverted refractivity from RO in these situations is still usable or suffers from large uncertainty (besides the biases you already mention), and if there is something that can be done about it.
- In the methodology, you use a circle of 2º to describe the area of influence of the RO (e.g. L200-L210). This is true for the along-ray direction, but may not be true for the direction perpendicular to the direction of the ray. If I understand correctly, you notice this in L322-323. I think it deserves a more in depth discussion, since it may be relevant in the cases with multiple RO. Also, probably it would be more correct to define an ellipse following the ray, rather than a circle, although I understand that as a first approximation and to show the point, it may be enough as it is.
- Section 3.2 (real case). I find several issues with the case example:
--- (1) First of all, the image (inset of Figure 10) shows a collection of rays, which I assume represent rays at different heights, not vertical because of the tangent point drift. I don’t recall a discussion about this in the paper. It would be good for the unfamiliarized reader a better explanation of why there are several rays depicted, and why these don’t form a vertical plane. Also, in L355 you refer to tangent points (in plural) while in L357, L365, L367, L368, etc. you talk about tangent point (in singular). Is the latter a “reference” tangent point? How is it defined? This may have an effect on the rest of the paper, because you define a circle of 2º around the tangent point, but you don’t seem to account for the drift. For example, in L201 you talk about “distance from the tangent point location”, in singular. If the drift is taken into account (for example, the ROPP 2D ray tracer takes that into account), then I would recommend mention it explicitly and more clearly.
--- (2) There si no “truth” validation for the case: it is shown how the RO changes the water content fields from the MWR, but we don’t know if this change is improving or making the estimate worse. Also, you argue in the abstract and introduction that MWR can suffer from issues related to land emissivity processes, and yet you have chosen a case that is mainly over land. Is there an option to double check with other datasets on whether the information from RO is actually “improving” the retrieval? I understand that reanalyses may be affected by the fact that are assimilating both MWR and RO, but may be worth a check?
--- (3) Have you checked for multiple RO collocated with MWF cases using all RO missions together? And within the ROMEX dataset? Since you show the multiple RO collocation case in the simulated results, it would be great to see a real case of such occurrences. If you conclude that with the current RO capability (even in ROMEX), these are almost impossible to find in real situations, then you could add a short discussion about this, and maybe hypothesize what should happen for these to occur (only more RO in global would help? ROs obtained in certain orbital configuration? Etc.).
Minor:
About the platforms, sensors, technology: It is not clear if you use microwave, infrared, both, or whatever is available. However, sometimes in the text seems that you generalize everything to microwave, but then IR appears again. I suggest a careful look about this and ensure consistency through the whole text. For example:
L349 – L359: You start talking about RO and MWR, then you mention ATMS collocations, but later on you say you use CrIMSS. Please, be consistent. Are you using passive microwave, infra-red, or both? Also, the actual platform you are collocating with (in addition to sensor) may help (e.g. NPP? NOAA-20? NOAA-21?)

L42: This sentence suggests that you will be only using passive microwave. Is that true? If true, I suggest to modify the abstract so the IR appears less relevant. Also, this is related to previous minor comment in L349-L359.

L15: Here you only mention CrIMSS, and no ATMS.

L88, L94, L95: Here it seems that you are going to use only MWR.

L50: “radio holographic” : a sentence or a reference explaining what it is, and why is it relevant in this context.
L87: Here you define the grid to be 0.1º x 0.1º (“to capture the horizontal resolution of MWR”). Since the refractivity you end up “correcting” is the N_{nadir}, I wonder why you don’t use the exact MWR footprint (in kilometers), so that you avoid one interpolation step when using real data.
L100: The N_{nadir} and a share the same vertical coordinate, z_k. However, N_{nadir} is only provided at another z which is of worse vertical resolution than z_k (since z_k is determined by the RO). Then, my question is: is N_{nadir} vertically linearly interpolated into z_k levels? Or is it just mapped, so that several consecutive z_k levels have exactly the same N_{nadir} value, like steps? Could this have an influence to the final result?
L242-246: A few more details on the WRF simulation are needed. For example, (1) is the simulation run in one domain? Or multiple nested domains? (2) which is the resolution (in km) of each domain?; (3) which is the vertical resolution of the model output?; Even though I don’t think it is relevant for the study itself, in general, WRF simulations come with the description of the basic parametrizations and schemes that have been used.
L289-296: A suggestion: I would include, in each panel, the distance to each tangent point, so that the panel in the left shows dist_TP1 = ~0 and dist_TP2= XXkm, the panel in the right shows dist_TP1 = XXkm and dist_TP2= 0km, and the panels in between show something mixed. I think it could help the reader to fully understand this paragraph, and the image.
Figures 4, 6, 7, 8, 9: A suggestion: Could these figures benefit from a few additional panels showing the difference with respect to the truth? I think a panel with the differences would highlight the improvement brought in by the RO. Probably not a new figure, but just an additional row with a couple panels showing the differences. Just a suggestion, consider it or just disregard the comment.
L314: Syntax error around “... Figs. 7 …” ?
L323: Does this relate to my comment above about the along-ray / perpendicular-to-the-ray radius of influence?
L358: I don’t understand why and where to are you performing the interpolation. Could this be avoided by my comment on L87, above?
Citation: https://doi.org/10.5194/egusphere-2026-4427-RC2
RC3: 'Comment on egusphere-2026-4427', Anonymous Referee #3, 30 Oct 2026

General comments:
The manuscript presents a new method to combine radio occultations with nadir soundings using a tomographic approach with some clever modifications. Focus is on improving the information about the spatial distribution of water vapor in the lower troposphere. The approach is mostly used on simulated data, and it is difficult to verify the improvement when using real data. However, I find the approach ingenious and the results seem very promising. The paper is well structured, and fairly well written, though with some improvements to be made. In my opinion, the manuscript can be published after minor revision.
Specific comments:
line 23: I don't understand "non-linearly connected processes". What would it mean for processes to be linearly conected? Maybe revise text.
line 44: "GNSS-RO measures the received excess phase delay ... and computes ...". The sentence should be revised. GNSS-RO is a technique, which can't measure or compute as such. Is 'delay' necessary? It is just 'excess phase' in most other places.
Figure 1: It is unclear what 'd' is in the figure. Only much later (in Eq. 8) it becomes clear that it is the length of the ray path within the voxel. It should be clarified already when discussing Fig. 1.
line 97: Perhaps use "proportional" instead of "linearly related".
line 117: I'm not sure what "linearly smoothed" means (also in line 266). Could it be re-phrased?
Figure 2: The figure caption should be revised. The figure does not illustrate "the difference between total excess phase and partial excess phase". At least more explanation is needed.
line 138: "... can be applied when doing simulations and in the processing of real observations.". It is unclear how this can be done for real observations. Later it seems that the domain is only in the troposphere, but here it is indicated that the domain is where N > 0 (which is not only below 8 km). The text should be more clear on what the domain is and how the same approach can be used for real observations where L_c in Eq. 5 is not known.
line 144 and 202: What is the upper boundary of the domain set to?
line 163: "The exact positions of the GNSS transmitter or receiver are not needed. This is true when processing simulated or real data.". What is meant by "processing" here? Maybe use another word. When processing real data from excess phase to bending angle and refractivity, exact positions are needed.
line 170: How are partial ray path samples handled when a ray path crosses the voxel boundaries? Is 'd' in Eq. 8 always a multiple of 20 m? Is that precise enough? (excess phase precision in real data can be a few millimeters)
Eq. 11 and after: Should there be a subscript k on a(z)? So a(z_k) as in Eq. 1? If not, then it should be clarified why not.
Eq. 12: The mathematical formulation here needs improvement. In previous equations, Phi_p was given for a single ray, but in Eq. 12 it seems to be a vector. The authors should write Eq. 11 and 12 in vector-matrix form, and describe what the elements of the matrix and vectors in Eq. 11 are. Perhaps write parts of the matrix explicitly to make it clear.
line 191 : "... is the pseudo inverse operator [see, e.g., Hajj et al. (1994)]." It is unclear if the formalism is new, or if it was already proposed before. I could not find mention of a pseudo inverse operator in Hajj et al. (1994).
line 198: "subsequent iterations" is not previously described, nor later. I assume it refers to numerical iterations when solving Eq. 12, but it is not clear.
line 202: "ray paths within the domain does not exceed 3°". In Fig. 3 it looks like the domain is 8 km vertically, which would mean that the length of ray paths within the domain is about 6° horizontally.
Eq. 16: I assume the subscript T stands for 'Total' (should be roman then, not italic), but it is not really defined in the text. Perhaps mention it in the sentence before.
line 245: "The WRF domain is linearly interpolated to the 3D grid defined in the previous section.". I assume it is interpolated to the mid-point of voxels and constant within each voxel, but it is not clear.
line 263: "The resulting refractivity field can then be used to derive the water vapor solution by using Eq. 2 and assuming the temperature a priori to be true.". This sentence needs revising because the equation relates specifically to the nadir (a priori) field in all variables. And there is no mention about the pressure in Eq. 2. It needs to be more clear how the water vapor is derived from the refractivity solution. I suggest to add another equation and explain the different terms.
Fig 5: The last sentence in the caption should be removed. Discussions should be only in the text (which it is). The same can be said about Fig. 9.
line 301: "At some locations such as -150 to -140° longitude and 32° latitude ...". Perhaps say "... such as along the 32° latitude ...".
line 303: I think it should be "between -140 and -146° longitudes", and "cyan solid line" (not orange).
line 310: "... that does not limited to the single occultation plane condition." Needs revision.
line 314: "The 3D tomography results are shown in Figs. 7 that showing the 33° latitude vertical cross section, and 8 that showing a 2km altitude horizontal cross section." Needs revision.
line 317: "... the concave size of each parabolic curve.". Would "... the width of each parabolic curve" be better?
Fig 7 caption: "The tomography solution of water vapor with 5 ROs on the same plane.". Could "on the same plane" be removed?
line 336: "This RMSE is computed over the entire simulated domain". In three dimensions? What are the boundaries of the domain?
line 339: "the a priori in this case is smoothed horizontally first, and vertically next which creates a much smoother field". I suppose it is the 'truth' that is smoothed to obtain the a priori? Besides that, I don't understand what was done exactly. More details of the smoothing should be provided.
line 352: Perhaps it is not so unlikely to have two ROs in the same plane as in the simulations. Could there be cases from Sentinel-6 where A and B satellites are flying in tandem? Maybe discuss the possibility.
Fig 12: Why are the '1 deg' RO+MWR solutions in the right panel not on each side of the 'tangent' solution in the middle panel? It seems that the '+1 deg' solution is almost the same as the tangent solution, whereas the '-1 deg' solution is considerably larger. Is it consistent with the difference in water vapor in the left panel?
line 380: "The dotted curves in color green and red". I think 'color' could be removed from the sentence.
Figure 10: Why is the tangent point trajectory so irregular? Why is there an apparent gap in the ray paths? It should be discussed in the text.
line 392: "In this article we introduced ..." could be just "We have introduced ..."
line 399: "which can land further insight on optimal methods that ...". Do you mean "lend further insight"?
line 414: "... ray tracing described in Radio Occultation Processing Package". Perhaps better to say "... ray tracing based on the work in the Radio Occultation Processing Package". It doesn't seem to be exactly the same since a term is neglected in Eq. A4. Maybe this should be mentioned specifically.
line 415: "The height of each point". I think you mean "The change of height of each point".
line 429: "This can be achieved by interpolating the vertical refractivity gradient calculated from the atmospheric column where the specific point in the ray path located within the 3D domain." Needs revision.
Technical corrections:
line 8: The acronym TB can be left out. There is no further reference to it.
Figure 3: "path" in caption should be "paths". Remove 'g/m^3' from the title of the plot and put it on top of the color bar. Units should be written exponentially. Titles above figures are not needed, instead (a), (b), (c), ... in panels are preferred (see the AMT guidelines).
Figure 5: Missing label on x-axis (perhaps the label could be only below the left-most panel).
line 20: "between 0-6 km" -> "between 0 and 6 km". When using the word between, it should be paired with the word and.
line 27 (and elsewhere): There should be a space between numbers and units.
Math notation: Only variables should be italic. For example 'nadir' as a subscript should be roman (see the AMT guidelines).
There are many undefined acronyms/abbreviations throughout the text. All acronyms should be defined. Acronyms in plural should be followed by the suffix -s (see the AMT guidelines).
line 326: "between 2 to 4 km" -> "between 2 and 4 km".
line 367: "As expected, that the changes are mainly located ...". Skip 'that'.
Fig A1 caption: "texts" -> "text"

Citation: https://doi.org/10.5194/egusphere-2026-4427-RC3

Kuo-Nung Wang, Chi O. Ao, George A. Hajj, Mary G. Morris, and Angelyn W. Moore

Viewed

Total article views: 387 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
217	135	35	387	34	23

HTML: 217
PDF: 135
XML: 35
Total: 387
BibTeX: 34
EndNote: 23

Views and downloads (calculated since 11 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	66	16	4	86
Sep 2026	52	41	13	106
Oct 2026	86	70	14	170
Nov 2026	13	8	4	25

Cumulative views and downloads (calculated since 11 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	66	16	4	86
Sep 2026	52	41	13	106
Oct 2026	86	70	14	170
Nov 2026	13	8	4	25

Viewed (geographical distribution)

Total article views: 331 (including HTML, PDF, and XML) Thereof 331 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 04 Nov 2026

Short summary

In this study we combine two of the most impactful spaceborne remote sensing technologies for weather prediction: the global navigation satellite system – radio occultation (GNSS-RO) and nadir sounding instruments such as microwave radiometer (MWR). We demonstrate that using a novel tomography method to combine these two observations significantly improve the characterization of the complex 3D water vapor structure in the lower troposphere than what is possible from either measurement alone.


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