the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 27 May 2026
Impact of Increased GNSS Radio Occultation Data Coverage on Tropical Cyclogenesis Prediction
Abstract. This study investigates the impact of increased global navigation satellite system (GNSS) radio occultation (RO) data coverage on the prediction of tropical cyclogenesis in the North Atlantic Ocean. RO data from the Radio Occultation Modelling Experiment (ROMEX) are used to construct three datasets with varying numbers of profiles and spatiotemporal coverage. The impacts of these datasets on the genesis forecasts for six selected tropical cyclones during the 2022 Atlantic hurricane season are assessed. Results show that assimilating RO datasets with higher horizontal data density and more homogeneous spatiotemporal distribution leads to improved detection of tropical cyclogenesis. Additional model diagnosis shows that the improved prediction of cyclogenesis is associated with increased specific humidity and relative vorticity, which support stronger upward motion and create a more favourable environment for tropical cyclone development. It is noted that despite the increased RO observation and data coverage, genesis was not predicted for Hurricane Lisa and Hurricane Ian. Further analysis shows that increasing the number of RO profiles and having the data more evenly distributed help improve the pre-genesis environment, with increased specific humidity and relative vorticity. These findings offer guidance for the design of future satellite observing systems: (1) increased RO profile density, beyond what is currently available operationally, is crucial for improving the representation of the pre-genesis environment, especially in data-sparse regions like the tropical North Atlantic Ocean, and (2) homogeneous data distribution minimizes data gaps and improves the accuracy of the observed atmospheric state, by minimizing the sampling errors.
- Preprint
(33569 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 10 Jul 2026)
- RC1: 'Comment on egusphere-2026-2704', Anonymous Referee #1, 01 Jul 2026 reply
-
RC2: 'Comment on egusphere-2026-2704', Anonymous Referee #2, 05 Jul 2026
reply
Review Comments
Impact of Increased GNSS Radio Occultation Data Coverage on Tropical Cyclogenesis Prediction
By Hsiao-Chun Lin, Ying-Hwa Kuo, Jan-Peter Weiss, John Braun, William Gullotta
General comments
This manuscript investigates the impact of increased GNSS radio occultation (RO) data coverage on tropical cyclogenesis prediction over the North Atlantic using WRF 3DVAR experiments. Three RO datasets are compared: an 8k baseline dataset, a 14k best-distributed dataset, and a 14k poorest-distributed dataset. The topic is important and timely, particularly for future GNSS RO constellation design and for improving tropical cyclone genesis forecasts over data-sparse oceanic regions.
The manuscript has scientific merit. The comparison between profile density and spatiotemporal distribution is potentially useful, and the focus on tropical cyclogenesis is relevant to both operational forecasting and observing-system design. However, the current version requires revisions before publication. Some of the main conclusions appear stronger than what the current evidence can support. Specific comments are provided below.
Specific comments
- The manuscript defines a successful genesis prediction as one in which the model-predicted genesis time falls within ±24 h of the observed genesis time. However, this criterion does not appear to be applied consistently in Table 2 and the related discussion. For example, Fiona in 14kBestRO is predicted at +36 h, which is outside the stated ±24 h window. Therefore, Fiona should not be described as a successful case unless the definition is changed. The authors may instead present Fiona as a delayed-genesis case with improved pre-genesis structure, or select another case that satisfies the stated success criterion. This revision is important for maintaining consistency between the methodology and the case-study interpretation.
- In addition, the treatment of cases marked by triangles also requires clarification. If these short-lived or non-persistent genesis signals are counted as 0.5 hits, the authors are suggested to clearly distinguish among genesis detected within the ±24 h window, genesis detected outside the verification window, and short-lived genesis-like signals that do not persist or intensify.
- The study includes six tropical cyclones from the 2022 Atlantic hurricane season. The best-performing experiment, 14kBestRO, detects only a limited number of genesis events, and two cases, Ian and Lisa, are not predicted by any experiment. Therefore, broad statements that increased and more homogeneous RO coverage improves tropical cyclogenesis prediction should be tempered. If increasing the sample size is beyond the scope of the present study, the authors may provide uncertainty estimates for the probability of detection or revise the wording to emphasize that the results are suggestive rather than conclusive.
- The manuscript states that increased RO data coverage helps improve the pre-genesis environment, with increased specific humidity and relative vorticity. This statement should be qualified. It is unclear whether the assimilation of RO observations always leads to increases in moisture and vorticity. If so, what is the physical explanation for this systematic increase? If not, please revise the statement to more accurately reflect the case-dependent impact of RO assimilation.
- Fig. 3 appears to illustrate the tracking algorithm's criteria, but its role in the main argument is not sufficiently clear. If the figure is only illustrative, it could be moved to the supplement. Similarly, Fig. 4 shows Gaston in the GFS analysis and in the three experiments, but the manuscript does not clearly explain what this figure is meant to demonstrate. Please clarify the purpose of the two figures.
- Fig. 5 shows that the number of assimilated RO profiles in 14kPoorRO is only slightly larger than that in 8kRO in the mid- and lower troposphere, and is sometimes even smaller at certain DA cycles. Within the 500-km radius of the vortex center, the number of RO profiles in 14kPoorRO is also comparable to that in 8kRO. It is therefore unclear why Fig. 7 shows broadly positive impacts, as indicated by negative relative RMSE differences, for 14kPoorRO. Could the authors further elaborate on this?
-
In addition, Fig. 7 shows that 14kPoorRO performs better than 14kBestRO in several upper-level variables. However, the profile counts shown in Fig. 5 do not clearly demonstrate that 14kPoorRO has more upper-level RO data than 14kBestRO. The manuscript attributes this improvement to increased RO sampling at high latitudes and possible improvements in upper-level jets, but this explanation is not sufficiently supported by the current diagnostics. The authors are encouraged to provide additional evidence to support this interpretation.
- In relation to the Fiona case, Fig. 10 shows the evolution only up to the observed genesis time, whereas the model-predicted genesis in 14kBestRO occurs 36 h later. Therefore, Fig. 10 does not directly diagnose the simulated genesis process in the model. The authors are encouraged to extend the analysis to the model-predicted genesis time, or at least include additional diagnostics around that time, to better support the interpretation of why genesis occurs in the 14kBestRO experiment.
- Some cases, especially Ian, Julia, and Lisa, occur near the Caribbean Sea and Central America. Terrain interaction, coastline proximity, sea-surface temperature representation, and model resolution may affect low-level circulation, moisture convergence, and cyclogenesis. The 15-km WRF domain may also limit the ability to represent convective organization during genesis. Could the authors discuss how these geographical and model-configuration factors may influence the failed or delayed genesis forecasts?
- The differences among the experiments in Figs. 11 and 16 are not easy to identify with the current color scales. It is encouraged to adjust the colorbar ranges or use more appropriate color intervals to enhance the contrast among experiments. This would make the differences in relative vorticity more visually distinguishable and help readers better assess the physical interpretation discussed in the text.
Citation: https://doi.org/10.5194/egusphere-2026-2704-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 40 | 10 | 10 | 60 | 10 | 8 |
- HTML: 40
- PDF: 10
- XML: 10
- Total: 60
- BibTeX: 10
- EndNote: 8
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study by H.-C. Lin et al, titled "Impact of Increased GNSS Radio Occultation Data Coverage on Tropical Cyclogenesis Prediction," investigates the impact of assimilated GNSS radio occultation dataset size and temporal distribution on WRF model tropical cyclone (TC) genesis forecasts. Cycled WRF data assimilation (DA) experiments are run for six 2022 Atlantic TCs in three configurations to evaluate (1) the impacts of nearly doubling the daily assimilated RO profile count using commercial observations provided by the Radio Occultation Modeling EXperiment (ROMEX) and (2) the temporal sampling of the ROMEX-enhanced observation dataset. Results show that both increased sampling density and homogeneous temporal distribution of RO observations across 6-hourly DA cycles (to mitigate impacts of some satellite receiver platforms only supplying data over the western Atlantic during 00 and 12 UTC cycles) are needed to provide the greatest improvement in TC cyclogenesis probability of detection. Diagnostic analysis of successful Hurricane Fiona (2022) and failed Hurricane Ian (2022) forecasts further show that the additional mid-tropospheric water vapor information gained from assimilating the ROMEX RO observations (at least over a 72-h cycled DA period) can in some cases not be enough to get the model to predict genesis if the model representation of vortex winds is too poor. Overall, this study validates both the benefits of augmenting current government RO satellite missions with commercial RO satellite data and the need to optimize temporal RO sampling continuity to improve model predictability of TC cyclogenesis in data-sparse regions such as the tropical North Atlantic Ocean.
Overall impression: This is a well-designed study with clearly presented results that builds on previous work evaluating impacts of RO DA on TC cyclogenesis forecasts. A notable new and valuable contribution of this study is its investigation of cyclogenesis forecast sensitivity to temporal continuity of RO observation sampling, which can inform the design of future RO receiver satellite observing systems. Additionally, the model diagnostic work showing the model water vapor and vorticity fields' response to the assimilated ROMEX observations, which includes additional sensitivity testing with analysis fields swapped in from the operational GFS, is nicely done. However, some improvements and clarifications could be made to the results interpretations, and the study could also benefit from additional description of the DA methodology and some caveats when extrapolating these results to an operational model. I recommend accept after Minor Revision.
Comments on the substance of the paper:
1. It would be worth mentioning some aspects of the WRF/WRFDA model configuration that differ from operational global and regional TC forecasting models. For example, the 3DVar algorithm uses just a static background error covariance matrix without ensemble-based "errors of the day." Also, impacts of the relatively coarse 15-km horizontal resolution: Might this explain why some cases experienced TC genesis but quickly dissipated, such as Gaston in the 14kBestRO experiment? I realize that this resolution is not much coarser than current operational global models that the forecasters use for TC cyclogenesis prediction. But if global forecast model horizontal resolution continues to increase in the future, might we expect to see greater value in utilizing RO data for cyclogenesis prediction, and how would this study's results apply to those future models?
2. Could you please provide some information regarding the prescribed RO refractivity observation errors? Do they vary according to satellite mission, latitude, altitude, or atmospheric conditions?
3. Have you explored sensitivity of your results to variations in the vortex tracker settings, such as the 50-km radius used for defining the circulation?
4. Probability of Detection statistics shown in Table 2. It appears that you are not counting the events where genesis briefly occurred but the TC quickly dissipated, shown as triangles in Table 2, in the POD values listed for each experiment. For example, the POD for 8kRO is listed as 1/6, which would, among all cases, just be counting Julia as a hit within the +/- 24-h time tolerance criterion. Therefore, the triangle for Fiona must not be counted, even though in lines 180-181 you say that such cases are counted as a “0.5 hit”. Similarly, the 14kBestRO experiment POD is listed as 2/6, which presumably comes from Julia and Nicole being successful hits, but the 0.5 hit from Gaston is not being counted here. Maybe I’m missing something?
5. Figure 9 and related discussion of the Hurricane Fiona RH and vorticity fields at t=48 h (lines 284-294). Even though 14kBestRO shows the “strongest and broadest moistening” among the three experiments, as you write in line 286, there are some notable differences between this experiment’s 48-h forecast RH/vorticity fields and the GFS analysis shown in Figs. 9a, 9b. For example, whereas the GFS analysis shows 75% RH extending above 350 hPa (Fig. 9a), the 14kBestRO experiment’s 75% RH contour extends only up to around 500 hPa, with very dry air in the upper troposphere all the way to the TC center - more so than in the other two experiments. Additionally, even though you point out the difference between the “very narrow vorticity structure of 14kPoorRO” compared to the GFS (line 291), it’s worth noting that despite its being confined within the 0.5-degree radius, 14kPoorRO’s azimuthally-averaged vorticity profile (Fig. 9h) more closely resembles that of the GFS (Fig. 9b) in terms of its greater depth, compared to the other two experiments, which forecast a shallow vortex that extends up to around 500 hPa (Figs. 9f,9g). Could it be possible that the GFS and 14kPoorRO are correctly capturing a tighter and deeper circulation for Fiona at this time and that 8kRO and 14kBestRO are failing to capture the deeper vortex, despite their doing a better job than 14kPoorRO in capturing the wider vortex closer to that of the GFS in the lower troposphere? Could you provide some additional comment on these structural differences shown in Figure 9 that seemingly contradict your main point about 14kBestRO forecasting inner-core RH fields and vortex structure for Fiona that are closest to the GFS analysis?
6. When describing the failed Ian genesis case forecast by 14BestRO, you write in lines 340-341: “However, the modest moisture increase was insufficient to counteract the ventilation effects of vertical wind shear.” Do you have evidence showing that the drier vortex shown in the three simulations, compared to the GFS analysis, was due to vertical wind shear-induced ventilation effects, as opposed to a dry bias in the WRF analysis for the eastern Caribbean region (Fig. 13; perhaps due to model physics or the DA algorithm) that could not be corrected enough in any of the simulations to enable cyclogenesis to happen? One way to demonstrate the vertical wind shear-induced ventilation effects could be to compare the radial wind and moisture anomaly covariances for Ian’s vortex between the GFS analysis and the three experiments at different levels and compute their environmental vertical wind shear vectors. Otherwise it might be best to rephrase this sentence to suggest that ventilation effects are just one possible explanation for the drier WRF forecast Ian vortices, compared to the GFS.
Minor/editorial comments:
Line 50: Recommend re-phrasing “GNSS RO observations can more accurately capture the thermodynamic environment which are favorable for the development” to
“GNSS RO observations can more accurately capture thermodynamic environments favorable for the development” or something similar
Line 95: “The GFS analyses is” —> “The GFS analyses are”
Lines 96-97: Probably better to rephrase “the best track data from the National Hurricane Center (NHC), National Oceanic and Atmospheric Administration (NOAA)” to
“the best track data from the National Oceanic and Atmospheric Administration (NOAA)’s National Hurricane Center (NHC)"
Line 103: “Individual DA experiment is” --> “Individual DA experiments are”
Lines 204-205: In the Figure 5 caption the first sentence is long and cumbersome. Recommend placing a period after “14kPoorRO (orange) experiments”. Then for the next sentence, use something like “The left panels (a, d, g) show the average among all cases, middle panels (b, e, h) show results for Hurricane Fiona, and right panels (c, f, i) show results for Hurricane Ian.
Lines 211-217: For the RMSE analysis shown in Figure 7, are the data points collected for generating RMSE statistics collected from the full 120-h free forecast period? Are data also collected from the DA cycling period? It would be worth clarifying this in the text.
Lines 240-240: Might the relatively coarse 15-km model horizontal resolution also be a possible factor contributing to the model’s failure to capture cyclogenesis for two cases? Even though (in lines 106-107) you mention the Li and Pu (2014) study’s finding limited improvement in TC cyclogenesis prediction with increased model resolution, it appears plausible to this reviewer that since moist convection generally plays an important role in TC cyclogenesis, your model’s inability to resolve convective updrafts at 15-km grid spacing could also hinder its ability to capture TC cyclogenesis for some cases.
Line 277: I think you meant to write “noted” after “It should be..”, right? Please correct this.
Line 334: “In contrast, none of the three DA experiments reproduced the observed increase in relative humidity…”. Recommend replacing “observed” with “GFS-analyzed” or something similar since the GFS gridded analysis cannot be considered to be an observation.
Lines 334-335: “In contrast, none of the three DA experiments reproduced the … intensification of vorticity and its upward extension” It’s worth noting that all three experiments do show transient upward development of weak cyclonic relative vorticity all the way to 100 hPa around t=0 h, although this feature only lasts for about 12 hours for all three experiments. Perhaps the more notable difference between the three experiments’ vortex structures and the GFS is their shallower cyclonic relative vorticity field after t=12 h continuing through the end of the evaluation period at t=48 h.
Figure 15: Could you please mark the center location of developing TC Ian on the panels as forecast by the different experiments? At first I was confused looking at the strong TC located near the Bahamas in the t=1 h and t=24 h panels, until I realized that I was looking at Fiona. Also, it might be worth plotting the observed storm position on Figs. 15e-h since you contrast its motion with that of the GFS_q14kBestRO storm in lines 378-380. Interestingly, the GFS_init storm shown in Fig. 15d also appears to have a southward bias in its motion that brings it into a less favorable environment, but unlike the GFS_q14kBestRO storm this storm can still intensity.
Line 414: ini_GFS —> init_GFS
Lines 414-417: By convective and non-convective precipitation, you are referring to precipitation generated by the cumulus physics and microphysics schemes, right? And not from a convective versus stratified partitioning algorithm, right? It would be good to mention this in the text or Figure 18 caption.