Assimilation of FY-3D and FY-3E Hyperspectral Infrared Atmospheric Sounding Observation and Its Impact on Numerical Weather Prediction during Spring Season over the Continental United States
Round 1
Reviewer 1 Report
Review of “Assimilation of FY-3D and FY-3E Hyperspectral Infrared Atmospheric Sounding Observation and Its Impact on Numerical Weather Prediction during Spring Season over Continental United States”
General comments:
This is an interesting study that compared the assimilation impact of HIRAS1 onboard FY-3D and HIRAS2 onboard FY-3E on the numerical weather prediction over the US. The manuscript is logically organized and well written. I have some minor comments before the manuscript can be accepted for publication.
1. Lns 115-117, the authors stated that “the off-diagonal values of HIRAS2’s R matrix are smaller than those from HIRAS1”. Why does that mean the former has more accurate observations that the latter? I think the variance (diagonal) rather than the covariance (off-diagonal) can reflect the accuracy of observations.
2. Lns 162-168, the experimental setup was not clearly clarified. Is the cycling data assimilation implemented within each day and restarted at the beginning of each day?
3. Why was the model resolution of 13 km used? Why were the forecasts integrated for 24 hours? The weather at the 13-km resolution is predictable for a few days.
4. Fig. 7, why are the biases lower at upper levels than at lower levels?
5. Fig. 10, the differences for temperature among the experiments are quite small at initial time. If the forecast lead time is extended beyond one day, would the differences become larger?
6. Lns406-408, “(2) except …”, the sentence is confused. Please reword.
The manuscript is logically organized and well written. Only minor editing of English language is required.
Author Response
Dear Reviewer,
Deeply appreciated for your time effort on reviewing our manuscript, our point-to-point replies to your comments are listed below.
1. Lns 115-117, the authors stated that “the off-diagonal values of HIRAS2’s R matrix are smaller than those from HIRAS1”. Why does that mean the former has more accurate observations that the latter? I think the variance (diagonal) rather than the covariance (off-diagonal) can reflect the accuracy of observations.
RE: Explanation “In most cases, off-diagonal values in the observational error covariance are assumed to be zero in data assimilation system and the diagonal values have the highest significance in utilizing observation, that is to say, decreasing the off-diagonal value in observational error covariance can improve the agreement between operational practice and theoretical assumption. ” added to manuscript (line 117-121).
2. Lns 162-168, the experimental setup was not clearly clarified. Is the cycling data assimilation implemented within each day and restarted at the beginning of each day?
RE: we rewrite the workflow description like this “The initial condition for the first analysis cycle and boundary condition for analysis and forecast cycles are provided by the NOAA GFS operational analysis product. The system kick off the first analysis cycle at 00:00 UTC every (cold start), the initial condition for the forecast cycle (final analysis) is generated from 23 hourly-basis continuous analysis cycles (warm start). It is important to understand that not all analysis cycles have the observation assimilated due the polar-orbiting satellite platform’s revisiting time vacancy. For the analysis cycles which do not have available observation to assimilate, the 1-hour lead-time forecast will be conducted without initiating the DA process. After retrieving the final analysis, the forecast cycle will start generating a 24-hours lead-time forecast. ” (line 166 - 174).
3. Why was the model resolution of 13 km used? Why were the forecasts integrated for 24 hours? The weather at the 13-km resolution is predictable for a few days.
RE: The reasons are listed below:
- To avoid Representativeness error, assimilating coarse horizontal resolution observation to a fine horizontal resolution initial condition can introduce the representativeness error issue.
- Lowering down the horizontal resolution in data assimilation system to accommodate observation’s horizontal resolution does not appear to decrease the system’s performance, as an living example, the ECMWF operates the 9-km global prediction system but its data assimilation system is operated at a coarser resolution. While the interpolation will bring extra error (interpolation error) when converting initial condition from coarse to fine horizontal resolution, this issue can be partly solved by digital filtering.
- We choose 24 hours as the desired forecast lead-time length based on the NOAA Rapid Refresh system (RAP) design. The NOAA RAP system provides 21-hours lead-time forecast at 00:00, 06:00, 12:00, 18:00 UTC (forecast initialization time) and 18-hours lead-time forecast at the rest of initialization time. To this point, a 24-hours lead-time forecast could be a reasonable choice.
4. Figure 7, why are the biases lower at upper levels than at lower levels?
RE: As can be seen from the figure below (from another under review manuscript), the hyperspectral infrared observation lacks of the capability of detecting low-troposphere atmospheric information, since its spectral weighting function peak level mostly stays above 800hPa. This figure shows the temperature weighting function of MTG-IRS and the y-axis stands for the pressure level.
5. Figure 10, the differences for temperature among the experiments are quite small at initial time. If the forecast lead time is extended beyond one day, would the differences become larger?
RE: We could not confirm whether the departure will be larger or smaller in this experiment, but generally speaking, the ACC departure among different results will expand when extra forecast lead-time time is added upon, please see the reference here: https://confluence.ecmwf.int/display/FUG/Anomaly+Correlation+Coefficient#:~:text=At%20ECMWF%20the%20anomaly%20correlation,to%20reference%20model%20climate%20and
6. Lns406-408, “(2) except …”, the sentence is confused. Please reword.
RE: We rewrite the sentence to “compared to CrIS onboard JPSS1, HIRAS2’s performance is gnerally comparable and better in temperature and 11 - 23 hours lead-time forecast. The performance improvement can be attributed to HIRAS2’s new spectrum design, improved observation quality ( decrease), as well as the FY-3E satellite’s early morning orbit selection”.
Best regards,
Qi Zhang
Reviewer 2 Report
Dear Editor,
This paper evaluates the impact of data assimilation of two hyperspectral infrared atmospheric sounding observations on numerical weather forecasts over the United States in the spring. This paper is well written and organized. It includes valuable information about the applicability of HIRAS2 assimilation in NWP.However, the presentation of the data (including HIRAS1 and 2) needs to be improved. The paper requires minor changes for it to be accepted
- Page4- Sec.2-1. The reason to choose 30 PCs to calculate the error and correlation is not clear. Is there any critical value to choose the number of PCs?
- Page 6- Sec.2-2. The reason for using such physics option should be clarified? Is it based on other references or is there any sensitivity analysis?
- Page 8- Fig.7. The interpretation of underestimation/overestimation (throughout the atmosphere) of each parameter should be added to the text.
- Page 10-Fig.9. In fig. 9.(b and f) all experiments have similar performance at the upper atmosphere. It should be discussed.
- Page 10-Fig.9. The probable cause of variable V wind mean bias below 900mb, should be discussed.
- Page 12- Sec. 3-3. POD and FAR are recommended to evaluate.
- Fig. 12& 14 and Fig. 13&15 have the similar captions. It is recommended to revise the captions and add the date.
Author Response
Dear Reviewer,
Deeply appreciated for your time effort on reviewing our manuscript, our point-to-point replies to your comments are listed below.
1. Sec.2-1. The reason to choose 30 PCs to calculate the error and correlation is not clear. Is there any critical value to choose the number of PCs?
RE: This is a great question, but we are afraid that we could only partly answer it. In conclusion, the specific number of how many PCs should be used in unclear. But based on the radiance simulation comparison between HTFRTC and RTTOV, we can confirm that the differences are low (mean bias smaller than 0.15K of brightness temperature) if the PCs amount is larger than 60, and the accuracy improvement increment slows down when the PCs amount is higher than 50 (approximately). In addition, the first 15 PCs carry approximately 99% of the observational information content. To this point, the PCs amount being assimilated should be a number between 15 and 50. The consideration of choosing 30 as the PCs amount is based on the computational cost. In this case, 30 PCs appears to be the best choice, because the performance evaluation differences among experiments do not increase as fast as they used be if we add more than 30 PCs into the assimilation, but this does not mean the rest of the PCs will not generate positive impact on NWP. In the end, for different scenario, e.g. running system at a different region, the PCs amount used in the DA system needs to be reconsidered.
2. Page 6- Sec.2-2. The reason for using such physics option should be clarified? Is it based on other references or is there any sensitivity analysis?
RE: the microphysics scheme selection in the experiments follow the scheme settings from NOAA EMC’s Rapid Refresh Forecast System (RRFS). This system is going to the replacement of NOAA’s 13km rapid refresh (RAP) forecast system. This is the default setting for UFS-SRW’s Continental US (CONUS) 13km forecast application.
3. Page 8- Fig.7. The interpretation of underestimation/overestimation (throughout the atmosphere) of each parameter should be added to the text.
RE: We added the discussions into the manuscript (line 229 - 236): From Figure 7, we can detect that the temperature field in the GFS initial condition is colder than the observation, which can be partly correced by hyperspectral infrared sounding assimilation; however, the added information content fails to decrease the random error (standard deviation) in lower troposphere (below 750hPa) because the hyperspectral infrared sounders are more sensitive to atmospheric changes above 800hPa. Besides, assimilating hyperspectral infrared sounding observation enlarges (decreases) the embedded dry bias in GFS initial condition from 950hPa to 650hPa (at 1000hPa) while the standard deviation seldom changes after the assimilation process.
4. Page 10-Fig.9. In fig. 9.(b and f) all experiments have similar performance at the upper atmosphere. It should be discussed. Page 10-Fig.9. The probable cause of variable V wind mean bias below 900mb, should be discussed.
RE: Discussion added “ It is interesting that the vertical performance vibration from 1000hPa to 950hPa exists in wind evaluation, especially the v-component wind (Figure 9h). This is mainly due to the amount vertical distribution of aircraft observation: according to the report from World Meteorological Organization (WMO) [46], the aircraft observation amount at near surface (100025 hPa) is extremely small, and the observation amount starts climbing up during the landing and taking-off process. Generally, the first available observation’s pressure level locates between 975hPa and 925hPa. The vertical distribution inequality of aircraft observation also attributes to the specific humidity’s similar performance at the upper atmosphere: aircraft generally observes the temperature and wind over upper troposphere while water vapor observations are gathered over middle and low troposphere. Due to the relatively homogeneous observation amount in vertical scale, the problem listed above is less detectable in temperature evaluation” (line 299 - 310).
5. Page 12- Sec. 3-3. POD and FAR are recommended to evaluate.
RE: the POD and FAR evaluations are added as Appendix B.
6. Fig. 12& 14 and Fig. 13&15 have the similar captions. It is recommended to revise the captions and add the date.
RE: captions are changed.
Best regards,
Qi Zhang
Reviewer 3 Report
A set of sensitive DA numerical experiments were setup to evaluate the impact of the PC scores derived from the FY3D/3E and CrIS. The results are quite impressive. The experiments and the analysis are exquisite. Do not have any major questions for this study. Here are some minor comments:
1. The tables are very blurred in the review version. It would be better to make the display clearer.
2. The Figure 6 is too complex. It would be better to use the simple workflow such as the figure 8 since the major difference of the experiments is the PC data derived from different satellites.
3. The author used the hourly DA cycle to assimilate the PC scores and 24-hour forecast are evaluated. But the reviewer did not find any details about the simulation periods. Are these based on one particular cases? Or the simulation periods covered all the “tornado” case time?
Author Response
Dear Reviewer,
Deeply appreciated for your time effort on reviewing our manuscript, our point-to-point replies to your comments are listed below.
1. The tables are very blurred in the review version. It would be better to make the display clearer.
RE: we adjusted the font size in each table.
2. The Figure 6 is too complex. It would be better to use the simple workflow such as the figure 8 since the major difference of the experiments is the PC data derived from different satellites.
RE: Figure 6 is modified based on reviewer’s suggestion.
3. The author used the hourly DA cycle to assimilate the PC scores and 24-hour forecast are evaluated. But the reviewer did not find any details about the simulation periods. Are these based on one particular cases? Or the simulation periods covered all the “tornado” case time?
RE: We added a paragraph in the manuscript to explain the simulation periods. “The following part consists of 4 sections: section 3.1 focuses on initial condition evaluation; section 3.2 focuses on upper level atmospheric variable evaluation in the forecast product; section 3.3 focuses on hourly precipitation forecast evaluation; section 3.4 focuses on tornado outbreak prediction accuracy evaluation. For section 3.1, 3.2, 3.3, the evaluation results derive from the 24-hours lead-time forecast from Mar. 15th 2022 to Apr. 22nd 2022 (missing forecast results on Mar. 22nd and Mar. 23rd, due to HIRAS2 observation unavailability). Results shown in section 3.4 come from selected case studies, case selection reasons will be mentioned in that section” (line 193 - 200).
Best regards,
Qi Zhang
Reviewer 4 Report
Journal: Atmosphere-
Manuscript ID: Manuscript Number 2364544
Title: "Assimilation of FY-3D and FY-3E Hyperspectral Infrared Atmospheric Sounding Observation and Its Impact on Numerical Weather Prediction during Spring Season over the Continental United States"
Authors: Qi Zhang , Min Shao.
General Comments to the Authors:
In this study, the authors present the results of the data assimilation experiment with observation radiance from HIRAS1, HIRAS2, and CrIS sensors, using the UFS-SRW system over the Continental United States. There are some general comments that need to be taken into consideration by the authors:
1- Abstract and conclusion: The study is interesting, it brings a practical application to data assimilation techniques, but I believe it should be treated as a case study. The title and the abstract seek to give a generality to the theme, however, the content of the article is a specific analysis of some days, it is not possible to carry out a statistical analysis with several cases and, consequently, it is not possible to work with the results in a generic way.
2- Methodology: The referenced material is ok, but some general information about the PC scores needs to be presented. The reason to be CONUS has been chosen by authors as dominium needs to be expressed. Some figures need to be corrected or some information needs to be presented or completed.
3- Results: the data coverage from different satellites used in this study need to be better presented to make clearer the difference between the data in the period of the day, which can influence the data assimilation process and consequently the generated results. The tornado forecast skill needs to be evaluated by objective score and more events would be considered.
I think that the paper can be published after the inclusion of some additional comments about the mentioned points and make other minor improvements.
See my comments in the "Specific Comments to the Authors" section for more details.
Comments for author File: 
Comments.pdf
Author Response
Dear Reviewer,
Deeply appreciated for your time effort on reviewing our manuscript, our point-to-point replies to your comments are listed below.
Introduction:
a. Line 51: I suggest placing a new figure highlighting the difference between the spatial coverage between Hiras1 and Hiras2, and the CONUS dominion needs to be highlighted in this figure too.
RE: Figure added in Appendix A
(b) Figure 1: I suggest treating the wavelength with frequency and putting the unit of measurement in Hertz
RE: Changed
(c) Figure 2: I suggest adding a paragraph explaining why J01-Cris was added as a reference in this study topic. I think that the reasons are obvious, but need to be mentioned by authors.
RE: we add this sentence “ Apart from the control experiment, in which no data assimilation process is conducted, we add another one, CrIS experiment, to demonstrate the baseline performance of hyperspectral infrared sounding observation assimilation in principal component space” to the introduction part (line 67 - 70).
Materials and Methods:
(a) Line 78: I suggest putting a definition of the PC Score is. It does not seem to me to be something trivial for researchers in the area. The referenced material is ok, but some general information about the PC score needs to be presented in this section.
RE: we add some explanations following line 78 “Specifically, principal component analysis (PCA) converts the hyperspectral infrared observations from spectrum space to “imaginary” orthometric space — principal component scores, using pre-calculated Empirical Orthogonal Functions (EOFS). The pre-calculated EOFS is equivalent to the coefficients in channel-based radiative transfer models, e.g. the Community Radiative Transfer Model (CRTM) and the Radiative Transfer for TOVS ( RTTOV ) .This process can retain crucial independent information content, decrease information redundancy and improve observation’s signal-to-noise ratio” (line 86 - 93).
(b) Figure 4: It would be interesting if the author brought what the reference parameters would be for the six graphs in Figure 4. For the covariance error, it seems to me that the closer to zero, the better, but for the correlation coefficients, what would be the best result?
RE: Specifically for this question “For the covariance error, it seems to me that the closer to zero”, the answer can be found in this publication “NWP SAF IRSPP User Manual” (https://nwp-saf.eumetsat.int/site/download/documentation/irspp/NWPSAF-MO-UD-053-IRSPP_User_Manual.pdf (page #17)), this issue is mainly due to “the leading PC scores are much smaller than would normally be the case (typically less than 1.0)”, which is also the reason why we use the first 30 PCs instead of all 300 PCs (maximum allowed amount by HTFRTC). In data assimilation, it is the observational error covariance that is used, however, we plot the correlation coefficients to accommodate potential reader’s specific demand (we encountered this issue a while ago). In scientific aspect, the correlation coefficients reveal that assimilating PC scores can potentially eliminate the “cross-channel correlation” issue in selected channel radiance assimilation scheme.
(c) Line 149: Please, include the reason to be CONUS has been chosen by authors to present yours results. I know some obvious reasons, but their selection or preference needs to be expressed.
RE: We made the target area selection based on the reasons listed below: 1. accuracy difference over different regions in GFS, the GFS forecast product’s accuracy is higher in US, Europe compared to East Asia, which could potentially hinder the demonstration of hyperspectral infrared observation assimilation’s positive impact on regional numerical weather prediction (this theory hasn’t been verified, we hope that we can conduct such research targeting at a or several different regions when future supports are secured); 2. UFS-SRW readiness uncertainty, the Unified Forecast System -Short Range Weather application has only been tested over CONUS domain and there is a suggested microphysics scheme combination for 13km horizontal resolution CONUS domain, extra resources is going to be needed to select a stable microphysics scheme combination for a different region, worst scenario, we will have to modify the forecast model then test it before initiating our desired research.
(d) Figure 6. I seeing this figure, it is not clear for me which day this 23 cycle referee. If the data assimilation cycle is done every day in the period (from 16th Mar. 2022 to 12th Apr. 2022, 64 lack of 21st Mar. 2022 and 22nd Mar. 2022, due to HIRAS2 data unavailability) totalize 26 cycles and not 23. Please make this information clearer or correct this figure.
RE: we changed the figure for clearance, also we add some explanations from line 171 to line 179 “The initial condition for the first analysis cycle and boundary condition for analysis and forecast cycles are provided by the NOAA GFS operational analysis product. The system kick off the first analysis cycle at 00:00 UTC every day from 15th Mar. 2022 to 11th Apr. 2022 (cold start), the initial condition for the forecast cycle (final analysis) is generated from hourly-basis continuous analysis cycles (24 cycles in total). It is important to understand that not all analysis cycles have the observation assimilated due the polar-orbiting satellite platform’s revisiting time vacancy. For the analysis cycles which do not have available observation to assimilate, the 1-hour lead-time forecast will be conducted without initiating the DA process. After retrieving the final analysis, the forecast cycle will start generating a 24-hours lead-time forecast”.
(e) Figure 6. I understand from seeing this figure that only in the 23 cycle (last day of period: 12th Apr. 2022) the forecast is done, but in the section 3.4 forecast tornado cases are used during 30th March and 05th Please correct this figure to avoid misunderstanding by readers.
RE: This issue can be solved with the modification in the abstract (reply to question “abstract #2”).
Results:
(a) Figure 8: Apparently, the assimilation window of the sensors (HIRAS1, HIRAS2 and CRIS) has observations at different times. I believe that this should be made clearer in the text. This aspect from different sensors are prevalent in this data assimilation process and consequently in the obtained results.
RE: We add the following sentences “To summarize, the performance fall-behind in HIRAS1 experiment mostly results from its inferior observation quality, since FY-3D shares almost the same local overpass time with JPSS1. However, the performance improvement in HIRAS2 experiment may come from different ways: 1. FY-3E operates in early-morning orbit, which means HIRAS2 experiment can assimilate the closest-to-forecast-initialization-time observation while the other experiments cannot; 2. HIRAS2’s is smaller than HIRAS1 and has a more comprehensive spectral coverage over thermal infrared region” (line 253 -260).
(b) Figure 12, 13, 14 and 15: what would be the reference standard in this figure? I mean, I think that it is more interesting to change these results for an objective analysis. Additionally, what other days are not taken into consideration in this analysis? For example, 06th and 12th April carried out a total of 50 tornado cases and this day are not evaluated.
RE: The reference standard follows the significant tornado parameter’s sensitivity: based on the the National Weather Service’s definition, the STP is sensitive to predict the outbreak of tornadoes with EF scale higher or equivalent to 1. Based the case selection parameter “there must be at least one tornado with EF scale higher than 1 during the 24 prediction lead hours”, only 23nd Mar., 30th Mar., 05th Apr. are qualified for case study.
Conclusions:
(a) The study is interesting, it brings a practical application to data assimilation techniques, but I believe it should be treated as a case study. The title and the abstract seek to give a generality to the theme, however the content of the article is a specific analysis of some days, it is not possible to carry out a statistical analysis with several cases and, consequently, it is not possible to work with the results in a generic way.
RE: We add the following sentences to the conclusion part “Although positive impact on NWP can be detected when HIRAS2 observation is assimilated, the limitations are negligible: 1. the experiments are conducted at a short-range lead-time and regional scale, more experiments at different lead-time range, different region, or global scale are still needed; 2. the experiments last for roughly one month, which could be treated as an extended-range case study, the conclusions demonstrated in this article may be less valid under different scenarios, additional studies in different seasons (e.g. summer) and different atmospheric situations (e.g. hurricane) are still needed.” (line 474 - 480).
Best regards,
Qi Zhang
