the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Nov 2023
Incremental Analysis Update (IAU) in the Model for Prediction Across Scales coupled with the Joint Effort for Data assimilation Integration (MPAS-JEDI 2.0.0)
Abstract. In a cycling system where data assimilation (DA) and model simulation are executed consecutively, the model forecasts initialized from the analysis (or data assimilation) can be systematically affected by dynamic imbalances generated during the analysis process. The high-frequency noise arising from the imbalances in the initial conditions can impose constraints on computational stability and efficiency during subsequent model simulations and can potentially become the low-frequency waves of physical significance. To mitigate these initial imbalances, the incremental analysis update (IAU) has long been utilized in the cycling context. This study introduces our recent implementation of the IAU in the Model for Prediction Across Scales for the Atmospheric component (MPAS-A), coupled with the Joint Effort for Data assimilation Integration (JEDI), through the cycling system called MPAS-Workflow. During the integration of the compressible nonhydrostatic equations in MPAS-A, analysis increments are distributed over a predefined time window (e.g., 6 hours) as fractional forcing at each time step. In a real case study with the assimilation of all conventional and satellite radiance observations every 6 h for one month, starting from mid-April 2018, model forecasts with IAU show that the initial noise illustrated by surface pressure tendency becomes well constrained throughout the forecast lead times, enhancing the system reliability. The month-long cycling with the assimilation of real observations demonstrates the successful implementation of the IAU capability in the MPAS-JEDI cycling system. Along with the comparison between the forecasts with and without IAU, several aspects on the implementation in MPAS-JEDI are discussed. Corresponding updates have been incorporated into the MPAS-A model (originally based on version 7.1), which is now publicly available in MPAS-JEDI and MPAS-Workflow Version 2.0.0.
- Preprint
(1967 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-2299', Anonymous Referee #1, 21 Nov 2023
This manuscript introduced the three-dimensional IAU implemented in the MPAS-JEDI 2.0.0. Previous studies have shown that IAU can effectively remedy the imbalance caused by intermittent data assimilation. It is worthwhile to investigate the performances and potential issues of IAU for a global model with varying horizontal meshes, which would provide guidance for future seamless predictions. The manuscript is pleasant to read. Please see my specific comments as below.
- l58-60, this sentence needs be clarified. What’s the difference between full fields and prognostic variables? Are the prognostic variables subsets of the full fields? If yes, why transform to the prognostic variables imposes more imbalances than that to full fields?
- l62-64, it would be nice to add some references for the imbalances mentioned here.
- l118, what is phi?
- l112-122, since the transformation from grid point to mesh grid is not linear, it is not equivalent to transform the increment or the analysis. For the MPAS-IAU, is the native increment or the analysis used as the input for MPAS simulations?
- l127-130, how much the error could be introduced by this hydrostatic assumption?
- l176, please spell 4DIAU out at the first time.
- l173-177, it is interesting to know the IAU terms for hydrometer variables. Are they the same as Eq. 12?
- l195-210, if a restart file is used for cycling, how the analysis and analysis tendency are computed for multiple time slices?
- l229, what is YAML?
- l225-235, is UFO an independent module outside of minimization or filtering? If so, how the bias correction (VarBC) is performed for radiance observations? How’s the inter-channel correlations handled by the UFO?
- l241, it is interesting to know whether the IAU functions well with inhomogeneous grids?
- Section 4, it would be more convincing to have the statistical significance of the error differences between CTRL and IAU. It would be nice to have the verifications of CTRL and IAU relative to ECMWF or NCEP analysis, especially for water vapor.
- l280-285, please give some explanations for the error differences between the CTRL and IAU. Why IAU helps over the tropics but not over the polar regions. Is it possible this is due to the moving systems over the tropics (Ge et al. 2023 JAMES)?
Citation: https://doi.org/10.5194/egusphere-2023-2299-RC1 -
AC2: 'Reply on RC1', Soyoung Ha, 29 Dec 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-2299/egusphere-2023-2299-AC2-supplement.pdf
-
CEC1: 'Comment on egusphere-2023-2299', Juan Antonio Añel, 20 Dec 2023
Dear authors,
Unfortunately, after checking your manuscript, it has come to our attention that it does not comply with our "Code and Data Policy".
https://www.geoscientific-model-development.net/policies/code_and_data_policy.html
You have archived your code on GitHub. However, GitHub is not a suitable repository for scientific publication. GitHub itself instructs authors to use other alternatives for long-term archival and publishing, such as Zenodo, and it provides an integration to create Zenodo repositories directly. Therefore, please, publish your code in one of the appropriate repositories, and reply to this comment with the relevant information (link and DOI) as soon as possible, as it should be available before the Discussions stage.In this way, if you do not fix this problem, we could have to reject your manuscript for publication in our journal. I should note that, actually, your manuscript should not have been accepted in Discussions, given this lack of compliance with our policy. Therefore, the current situation with your manuscript is irregular.
Also, you must include in a potentially reviewed version of your manuscript the modified 'Code and Data Availability' section, the DOI of the code.
Juan A. Añel
Geosci. Model Dev. Executive EditorCitation: https://doi.org/10.5194/egusphere-2023-2299-CEC1 -
AC1: 'Reply on CEC1', Soyoung Ha, 27 Dec 2023
Dear Chief Editor,
Thank you for bringing to our attention your Code and Data Policy, which we had not acknowledged at the time of submission.
Our organization - NCAR/UCAR - manages multiple community system development projects and releases all official versions through designated GitHub repositories because this platform facilitates cross-organizational collaborations and support. While our contributions in this study significantly affected the latest release and the entire code structure, it would not be appropriate to claim ownership of the whole system or scripts. Also, it is not straightforward to list all contributors as team members continue to change.
We now understand your policy regarding alternatives for long-term archival and publication to preserve the exact version used in the presented study. However, decisions regarding this matter are beyond our immediate control and are up to the policymakers of our own institution.
For this particular draft, we mirrored the official versions in Zenodo, following the instructions provided by this journal, and added them in our References. We also modified 'Code and Data Availability' section as follows:
Code and data availability.
The exact version of MPAS-JEDI, including its Python-based post-processing package, is archived on Zenodo (The MPAS-JEDI team: Jung et al. (2023)) and can be also accessed via the project website (https://github.com/JCSDA/mpas-bundle/tree/release/2.0.0). The current version of MPAS-Workflow that provides all the scripts and configurations to run cycling experiments detailed in this study, is archived on Zenodo (Guerrette et al. (2023a)) and is also available at https://github.com/NCAR/MPAS-Workflow. More general information about the JEDI system can be found on the project website (https://jointcenterforsatellitedataassimilation-jedi-docs.readthedocs-hosted.com/en/latest/overview/index.html).}
Added in References are:
Guerrette, J. J., Abdi-Oskouei, M., Ban, J., nos, I. H. B., Bresch, J., Ha, S., Jung, B.-J., Liu, Z., Snyder, C., Schwartz, C., Wu, Y., and Yu, Y.: MPAS-Workflow, https://doi.org/10.5281/zenodo.10433323, 2023a.
The MPAS-JEDI team: Jung, B. J., Vahl, S., Ban, J., Abdi-Oskouei, M., Trémolet, Y., Liu, Z., Snyder, C., Olah, M. J., Herbener, S., Holdaway, D., Bresch, J., Yu, Y., and Ménétrier, B.: mpas-bundle, https://doi.org/10.5281/zenodo.10433668, 2023.
===
Hope this can fix the issue. If there's anything else we should do, please let us now.
Thank you for your consideration,
Soyoung Ha
Citation: https://doi.org/10.5194/egusphere-2023-2299-AC1
-
AC1: 'Reply on CEC1', Soyoung Ha, 27 Dec 2023
-
RC2: 'Comment on egusphere-2023-2299', Anonymous Referee #2, 23 Dec 2023
This paper focuses on implementing Incremental Analysis Update (IAU) in MPAS-JEDI and evaluating its impact on model forecasts. IAU helps reduce initial imbalances in model forecasts caused by dynamical and physical balance issues during data assimilation, so model forecasts using IAU exhibit improved control over initial noise.
It's recommended to specify the control variables used in data assimilation and how to transform them into analysis increments.
I believe that an ensemble size of 20 is not sufficient for global analysis, despite using an ensemble with 100% background error covariance. It would be good to mention localization and ensemble spread inflation methods.
Regarding the ensemble Kalman filter used for the initial conditions of the ensemble forecast in the 3DEnVar system, clarification on its specifics would be beneficial for a comprehensive understanding.
In line 21 of the manuscript, there is an expression “It does not consider dynamical or physical balances across model grids or variables, nor does it account for the conservation of mass, momentum, or energy. Hence, the initial balance of the atmospheric flow can by disrupted by data assimilation when the initial state is replaced by the analysis state.” While acknowledging that analysis increments from data assimilation might not fully reflect the model's balance, it's important to note that data assimilation does account for dynamical or physical balances across model grids or variables using background error covariance. For instance, temperature observations impact surrounding grids, affecting wind and humidity variables.
The authors notes at line 177 “it is easily extended for 4DIAU with varying weights over the IAU time window”, However, transitioning from 3DIAU to 4DIAU may not be straightforward in the current version of MPAS-JEDI. Expanding to 4DIAU would necessitate multiple analysis increments over different times, requiring adjustments to the 4DEnVar and cycle suite.
The following are minor correction requests.
L21: can by disrupted by data assimilation
-> can be disrupted by data assimilation
L30: the incremental analysis update (IAU) method was introduced in Bloom et al. (1996)
-> the incremental analysis update (IAU) method was introduced by Bloom et al. (1996).
L47: stability in the the fine mesh region.
-> stability in the fine mesh region.
L74: Details of the implementation is described
-> Details of the implementation are described
L110: and the second term in the right-hand side
-> first and the second term on the right-hand side
L195: while ensuring the model forecasts reproducible
-> while ensuring the model forecasts are reproducible
or while ensuring reproducibility the model forecasts are
L240: global analysis and forecast cycling was conducted
-> global analysis and forecast cycling were conducted
Citation: https://doi.org/10.5194/egusphere-2023-2299-RC2 -
AC3: 'Reply on RC2', Soyoung Ha, 29 Dec 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-2299/egusphere-2023-2299-AC3-supplement.pdf
-
AC3: 'Reply on RC2', Soyoung Ha, 29 Dec 2023
-
EC1: 'Comment on egusphere-2023-2299', Shu-Chih Yang, 04 Feb 2024
Although the authors claim that the manuscript is categorized as a “technical” paper, the associated implementation is successful with comprehensive justification. I agree with the reviewer’s comments about showing the results related to water vapor since it is vital to know whether implementing the IAU method affects the moisture accuracy of MPAS-A.
In addition, it is worth addressing the issue of the degradation of the northern polar region with IAU. According to Fig. 7b, the rms difference is about 20% in the north polar region and 10% near 25-degree S during the first-day forecast!
If the authors attributed the difference to their own error in GFS analysis, I suggest showing the same figure with other analyses as the reference, like the EC analysis.
Citation: https://doi.org/10.5194/egusphere-2023-2299-EC1 -
AC4: 'Reply on EC1', Soyoung Ha, 14 Mar 2024
Thank you very much for your own review. In response to your comments, we provide four figures in the supplement as Figs. S1-S4, adding a few paragraphs in the manuscript (Page 10). Please find our point-by-point response to your valuable comments below. Our responses are written in Italic with bullets.
Although the authors claim that the manuscript is categorized as a “technical” paper, the associated implementation is successful with comprehensive justification. I agree with the reviewer’s comments about showing the results related to water vapor since it is vital to know whether implementing the IAU method affects the moisture accuracy of MPAS-A.
- We totally agree with you that any new implementation should be thoroughly investigated for correctness. For that, we have cross-checked all mathematical expressions and their corresponding codes multiple times. However, we want to take this opportunity to clarify that ensuring successful development is distinct from providing comprehensive justification for the impact study, which involves various scientific aspects. We also want to note that the main motivation and purpose of the IAU implementation are to effectively suppress initial noise resulting from dynamic imbalances, as already demonstrated in Figs. 2 and 3.
- But we agree that it would be interesting to examine the impact of IAU on moisture as part of the control variables. In response to your comment, we provide two figures in the supplement as Figs. S1 and S2.
- In Fig. S1, we compare water vapor mixing ratio (Qv [g/kg]) in 6-h forecasts between CTRL and IAU against ERA-5 analysis globally over a total of 92 6-hourly cycles from April 21 to May 13, 2018 (e.g., after a one-week spin-up period). Although root-mean-square errors are the same as 0.7 in both experiments (not shown), the systematic bias indicates that IAU produces slightly better agreement with ERA-5 analysis than CTRL, especially in the boundary layer (< 2 km).
- We also present a panel plot in Fig. S2, where we verify the analysis and the first-day forecast against sounding observations between the surface and 250 hPa for the entire month. Compared to the CTRL run, RMS errors are slightly worse at the analysis time (which corresponds to a 3-h forecast in the IAU run, as opposed to the analysis in CTRL) by up to 2% in (a). But in (b), as forecasts start from the analysis, the errors decrease during the first 6 h, and exhibit a statistically significant reduction of approximately 4% at 18 h in the troposphere. Hence, we cannot claim that IAU always improves moisture analysis and forecasts, but it is fair to say that it does lead to some improvements.
- As we believe that it would be best to stay focused on surface pressure to represent the entire column, we have decided to include them in the Supplemental material, with the comment in the text (L282) as “The moisture verification for the 6-h forecast is provided in the supplement.”. Thank you for your understanding.
In addition, it is worth addressing the issue of the degradation of the northern polar region with IAU. According to Fig. 7b, the rms difference is about 20% in the north polar region and 10% near 25-degree S during the first-day forecast! If the authors attributed the difference to their own error in GFS analysis, I suggest showing the same figure with other analyses as the reference, like the EC analysis.
- Thank you for your careful review on Figure 7. We admit that it is challenging to explain the forecast degradation in a specific region, especially given that we did not find anything particularly suspicious in Fig. 3. We fully understand the editor’s concern, though, and we would like to examine the issue through the observation-space diagnostics, rather than the verification against any particular analysis. This work was meant for the analysis cycling (with our own data assimilation), so it is legitimate to check the performance with respect to observations. In Fig. S3, where we verified our background forecasts for surface pressure against measurements, we noticed one red dot near the North Pole (to the north of Greenland) in the bottom panel. And there are more red dots to the north of Russia, all of which seemed to contribute to the larger errors shown in Fig. 7b.
- Also, you correctly captured the red area near 25°S, which seems associated with the red dots in the southern part of the tropics, as depicted in Fig. S3b. But as you can see through the global distribution, we do not have enough observations near the Poles and over the ocean to sufficiently constrain the model state. The month-long error statistics reveal almost no colors (e.g., little deviations) over the CONUS domain with a dense observing network, indicating that IAU itself does not degrade forecast errors when the model states are well constrained through data assimilation. It remains unclear whether the impact of IAU has intertwined with model errors in certain data-sparse regions. Note that model errors are not accounted for in this hybrid 3D-EnVar framework.
- It is worth noting that we implemented IAU on the model's unstructured mesh in a generic manner, not specifically tied to geographic regions. For clarity, following the statement “Compared to the GFS analysis, MPAS forecasts in CTRL exhibit the largest (or the fastest) error growth in the Southern Hemisphere. Forecasts in the IAU run, on the other hand, tend to reduce errors in the tropics while increasing errors near the North Pole region.”, we added a paragraph in L289-295, stating “This aligns with the findings of Ha et al. (2017), where forecast errors were significantly reduced over the tropics in a variable-resolution mesh, including both resolution-transition and high-resolution parts. Because IAU is implemented on the model's unstructured mesh (which is in a random order), it is not associated with particular geographic locations or mesh configurations. Given its time filtering feature, IAU might be more effective in simulating low-frequency modes dominant over the tropics. It is also noted that the impact of IAU may be nonlinearly intertwined with model errors in data-sparse regions, such as the Poles. However, model errors are not accounted for in the hybrid 3DEnVar system used in this study. Additional area-specific features in the verification are provided in the supplemental material.”.
- Fig. S4 further supports our point that the IAU significantly improved forecast errors in most regions, except for the North Pole area and the Southern Hemisphere Ocean, in a statistically significant manner. As illustrated in Fig. 4 in the manuscript, the IAU run assimilated slightly more observations (by 1-2%) throughout the month-long cycling, which is a good indicator that the DA cycling system work more effectively. Overall, it is our belief that our IAU implementation was successfully completed, with positive impacts on the analysis and short-term forecasts.
-
AC4: 'Reply on EC1', Soyoung Ha, 14 Mar 2024
Status: closed
-
RC1: 'Comment on egusphere-2023-2299', Anonymous Referee #1, 21 Nov 2023
This manuscript introduced the three-dimensional IAU implemented in the MPAS-JEDI 2.0.0. Previous studies have shown that IAU can effectively remedy the imbalance caused by intermittent data assimilation. It is worthwhile to investigate the performances and potential issues of IAU for a global model with varying horizontal meshes, which would provide guidance for future seamless predictions. The manuscript is pleasant to read. Please see my specific comments as below.
- l58-60, this sentence needs be clarified. What’s the difference between full fields and prognostic variables? Are the prognostic variables subsets of the full fields? If yes, why transform to the prognostic variables imposes more imbalances than that to full fields?
- l62-64, it would be nice to add some references for the imbalances mentioned here.
- l118, what is phi?
- l112-122, since the transformation from grid point to mesh grid is not linear, it is not equivalent to transform the increment or the analysis. For the MPAS-IAU, is the native increment or the analysis used as the input for MPAS simulations?
- l127-130, how much the error could be introduced by this hydrostatic assumption?
- l176, please spell 4DIAU out at the first time.
- l173-177, it is interesting to know the IAU terms for hydrometer variables. Are they the same as Eq. 12?
- l195-210, if a restart file is used for cycling, how the analysis and analysis tendency are computed for multiple time slices?
- l229, what is YAML?
- l225-235, is UFO an independent module outside of minimization or filtering? If so, how the bias correction (VarBC) is performed for radiance observations? How’s the inter-channel correlations handled by the UFO?
- l241, it is interesting to know whether the IAU functions well with inhomogeneous grids?
- Section 4, it would be more convincing to have the statistical significance of the error differences between CTRL and IAU. It would be nice to have the verifications of CTRL and IAU relative to ECMWF or NCEP analysis, especially for water vapor.
- l280-285, please give some explanations for the error differences between the CTRL and IAU. Why IAU helps over the tropics but not over the polar regions. Is it possible this is due to the moving systems over the tropics (Ge et al. 2023 JAMES)?
Citation: https://doi.org/10.5194/egusphere-2023-2299-RC1 -
AC2: 'Reply on RC1', Soyoung Ha, 29 Dec 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-2299/egusphere-2023-2299-AC2-supplement.pdf
-
CEC1: 'Comment on egusphere-2023-2299', Juan Antonio Añel, 20 Dec 2023
Dear authors,
Unfortunately, after checking your manuscript, it has come to our attention that it does not comply with our "Code and Data Policy".
https://www.geoscientific-model-development.net/policies/code_and_data_policy.html
You have archived your code on GitHub. However, GitHub is not a suitable repository for scientific publication. GitHub itself instructs authors to use other alternatives for long-term archival and publishing, such as Zenodo, and it provides an integration to create Zenodo repositories directly. Therefore, please, publish your code in one of the appropriate repositories, and reply to this comment with the relevant information (link and DOI) as soon as possible, as it should be available before the Discussions stage.In this way, if you do not fix this problem, we could have to reject your manuscript for publication in our journal. I should note that, actually, your manuscript should not have been accepted in Discussions, given this lack of compliance with our policy. Therefore, the current situation with your manuscript is irregular.
Also, you must include in a potentially reviewed version of your manuscript the modified 'Code and Data Availability' section, the DOI of the code.
Juan A. Añel
Geosci. Model Dev. Executive EditorCitation: https://doi.org/10.5194/egusphere-2023-2299-CEC1 -
AC1: 'Reply on CEC1', Soyoung Ha, 27 Dec 2023
Dear Chief Editor,
Thank you for bringing to our attention your Code and Data Policy, which we had not acknowledged at the time of submission.
Our organization - NCAR/UCAR - manages multiple community system development projects and releases all official versions through designated GitHub repositories because this platform facilitates cross-organizational collaborations and support. While our contributions in this study significantly affected the latest release and the entire code structure, it would not be appropriate to claim ownership of the whole system or scripts. Also, it is not straightforward to list all contributors as team members continue to change.
We now understand your policy regarding alternatives for long-term archival and publication to preserve the exact version used in the presented study. However, decisions regarding this matter are beyond our immediate control and are up to the policymakers of our own institution.
For this particular draft, we mirrored the official versions in Zenodo, following the instructions provided by this journal, and added them in our References. We also modified 'Code and Data Availability' section as follows:
Code and data availability.
The exact version of MPAS-JEDI, including its Python-based post-processing package, is archived on Zenodo (The MPAS-JEDI team: Jung et al. (2023)) and can be also accessed via the project website (https://github.com/JCSDA/mpas-bundle/tree/release/2.0.0). The current version of MPAS-Workflow that provides all the scripts and configurations to run cycling experiments detailed in this study, is archived on Zenodo (Guerrette et al. (2023a)) and is also available at https://github.com/NCAR/MPAS-Workflow. More general information about the JEDI system can be found on the project website (https://jointcenterforsatellitedataassimilation-jedi-docs.readthedocs-hosted.com/en/latest/overview/index.html).}
Added in References are:
Guerrette, J. J., Abdi-Oskouei, M., Ban, J., nos, I. H. B., Bresch, J., Ha, S., Jung, B.-J., Liu, Z., Snyder, C., Schwartz, C., Wu, Y., and Yu, Y.: MPAS-Workflow, https://doi.org/10.5281/zenodo.10433323, 2023a.
The MPAS-JEDI team: Jung, B. J., Vahl, S., Ban, J., Abdi-Oskouei, M., Trémolet, Y., Liu, Z., Snyder, C., Olah, M. J., Herbener, S., Holdaway, D., Bresch, J., Yu, Y., and Ménétrier, B.: mpas-bundle, https://doi.org/10.5281/zenodo.10433668, 2023.
===
Hope this can fix the issue. If there's anything else we should do, please let us now.
Thank you for your consideration,
Soyoung Ha
Citation: https://doi.org/10.5194/egusphere-2023-2299-AC1
-
AC1: 'Reply on CEC1', Soyoung Ha, 27 Dec 2023
-
RC2: 'Comment on egusphere-2023-2299', Anonymous Referee #2, 23 Dec 2023
This paper focuses on implementing Incremental Analysis Update (IAU) in MPAS-JEDI and evaluating its impact on model forecasts. IAU helps reduce initial imbalances in model forecasts caused by dynamical and physical balance issues during data assimilation, so model forecasts using IAU exhibit improved control over initial noise.
It's recommended to specify the control variables used in data assimilation and how to transform them into analysis increments.
I believe that an ensemble size of 20 is not sufficient for global analysis, despite using an ensemble with 100% background error covariance. It would be good to mention localization and ensemble spread inflation methods.
Regarding the ensemble Kalman filter used for the initial conditions of the ensemble forecast in the 3DEnVar system, clarification on its specifics would be beneficial for a comprehensive understanding.
In line 21 of the manuscript, there is an expression “It does not consider dynamical or physical balances across model grids or variables, nor does it account for the conservation of mass, momentum, or energy. Hence, the initial balance of the atmospheric flow can by disrupted by data assimilation when the initial state is replaced by the analysis state.” While acknowledging that analysis increments from data assimilation might not fully reflect the model's balance, it's important to note that data assimilation does account for dynamical or physical balances across model grids or variables using background error covariance. For instance, temperature observations impact surrounding grids, affecting wind and humidity variables.
The authors notes at line 177 “it is easily extended for 4DIAU with varying weights over the IAU time window”, However, transitioning from 3DIAU to 4DIAU may not be straightforward in the current version of MPAS-JEDI. Expanding to 4DIAU would necessitate multiple analysis increments over different times, requiring adjustments to the 4DEnVar and cycle suite.
The following are minor correction requests.
L21: can by disrupted by data assimilation
-> can be disrupted by data assimilation
L30: the incremental analysis update (IAU) method was introduced in Bloom et al. (1996)
-> the incremental analysis update (IAU) method was introduced by Bloom et al. (1996).
L47: stability in the the fine mesh region.
-> stability in the fine mesh region.
L74: Details of the implementation is described
-> Details of the implementation are described
L110: and the second term in the right-hand side
-> first and the second term on the right-hand side
L195: while ensuring the model forecasts reproducible
-> while ensuring the model forecasts are reproducible
or while ensuring reproducibility the model forecasts are
L240: global analysis and forecast cycling was conducted
-> global analysis and forecast cycling were conducted
Citation: https://doi.org/10.5194/egusphere-2023-2299-RC2 -
AC3: 'Reply on RC2', Soyoung Ha, 29 Dec 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-2299/egusphere-2023-2299-AC3-supplement.pdf
-
AC3: 'Reply on RC2', Soyoung Ha, 29 Dec 2023
-
EC1: 'Comment on egusphere-2023-2299', Shu-Chih Yang, 04 Feb 2024
Although the authors claim that the manuscript is categorized as a “technical” paper, the associated implementation is successful with comprehensive justification. I agree with the reviewer’s comments about showing the results related to water vapor since it is vital to know whether implementing the IAU method affects the moisture accuracy of MPAS-A.
In addition, it is worth addressing the issue of the degradation of the northern polar region with IAU. According to Fig. 7b, the rms difference is about 20% in the north polar region and 10% near 25-degree S during the first-day forecast!
If the authors attributed the difference to their own error in GFS analysis, I suggest showing the same figure with other analyses as the reference, like the EC analysis.
Citation: https://doi.org/10.5194/egusphere-2023-2299-EC1 -
AC4: 'Reply on EC1', Soyoung Ha, 14 Mar 2024
Thank you very much for your own review. In response to your comments, we provide four figures in the supplement as Figs. S1-S4, adding a few paragraphs in the manuscript (Page 10). Please find our point-by-point response to your valuable comments below. Our responses are written in Italic with bullets.
Although the authors claim that the manuscript is categorized as a “technical” paper, the associated implementation is successful with comprehensive justification. I agree with the reviewer’s comments about showing the results related to water vapor since it is vital to know whether implementing the IAU method affects the moisture accuracy of MPAS-A.
- We totally agree with you that any new implementation should be thoroughly investigated for correctness. For that, we have cross-checked all mathematical expressions and their corresponding codes multiple times. However, we want to take this opportunity to clarify that ensuring successful development is distinct from providing comprehensive justification for the impact study, which involves various scientific aspects. We also want to note that the main motivation and purpose of the IAU implementation are to effectively suppress initial noise resulting from dynamic imbalances, as already demonstrated in Figs. 2 and 3.
- But we agree that it would be interesting to examine the impact of IAU on moisture as part of the control variables. In response to your comment, we provide two figures in the supplement as Figs. S1 and S2.
- In Fig. S1, we compare water vapor mixing ratio (Qv [g/kg]) in 6-h forecasts between CTRL and IAU against ERA-5 analysis globally over a total of 92 6-hourly cycles from April 21 to May 13, 2018 (e.g., after a one-week spin-up period). Although root-mean-square errors are the same as 0.7 in both experiments (not shown), the systematic bias indicates that IAU produces slightly better agreement with ERA-5 analysis than CTRL, especially in the boundary layer (< 2 km).
- We also present a panel plot in Fig. S2, where we verify the analysis and the first-day forecast against sounding observations between the surface and 250 hPa for the entire month. Compared to the CTRL run, RMS errors are slightly worse at the analysis time (which corresponds to a 3-h forecast in the IAU run, as opposed to the analysis in CTRL) by up to 2% in (a). But in (b), as forecasts start from the analysis, the errors decrease during the first 6 h, and exhibit a statistically significant reduction of approximately 4% at 18 h in the troposphere. Hence, we cannot claim that IAU always improves moisture analysis and forecasts, but it is fair to say that it does lead to some improvements.
- As we believe that it would be best to stay focused on surface pressure to represent the entire column, we have decided to include them in the Supplemental material, with the comment in the text (L282) as “The moisture verification for the 6-h forecast is provided in the supplement.”. Thank you for your understanding.
In addition, it is worth addressing the issue of the degradation of the northern polar region with IAU. According to Fig. 7b, the rms difference is about 20% in the north polar region and 10% near 25-degree S during the first-day forecast! If the authors attributed the difference to their own error in GFS analysis, I suggest showing the same figure with other analyses as the reference, like the EC analysis.
- Thank you for your careful review on Figure 7. We admit that it is challenging to explain the forecast degradation in a specific region, especially given that we did not find anything particularly suspicious in Fig. 3. We fully understand the editor’s concern, though, and we would like to examine the issue through the observation-space diagnostics, rather than the verification against any particular analysis. This work was meant for the analysis cycling (with our own data assimilation), so it is legitimate to check the performance with respect to observations. In Fig. S3, where we verified our background forecasts for surface pressure against measurements, we noticed one red dot near the North Pole (to the north of Greenland) in the bottom panel. And there are more red dots to the north of Russia, all of which seemed to contribute to the larger errors shown in Fig. 7b.
- Also, you correctly captured the red area near 25°S, which seems associated with the red dots in the southern part of the tropics, as depicted in Fig. S3b. But as you can see through the global distribution, we do not have enough observations near the Poles and over the ocean to sufficiently constrain the model state. The month-long error statistics reveal almost no colors (e.g., little deviations) over the CONUS domain with a dense observing network, indicating that IAU itself does not degrade forecast errors when the model states are well constrained through data assimilation. It remains unclear whether the impact of IAU has intertwined with model errors in certain data-sparse regions. Note that model errors are not accounted for in this hybrid 3D-EnVar framework.
- It is worth noting that we implemented IAU on the model's unstructured mesh in a generic manner, not specifically tied to geographic regions. For clarity, following the statement “Compared to the GFS analysis, MPAS forecasts in CTRL exhibit the largest (or the fastest) error growth in the Southern Hemisphere. Forecasts in the IAU run, on the other hand, tend to reduce errors in the tropics while increasing errors near the North Pole region.”, we added a paragraph in L289-295, stating “This aligns with the findings of Ha et al. (2017), where forecast errors were significantly reduced over the tropics in a variable-resolution mesh, including both resolution-transition and high-resolution parts. Because IAU is implemented on the model's unstructured mesh (which is in a random order), it is not associated with particular geographic locations or mesh configurations. Given its time filtering feature, IAU might be more effective in simulating low-frequency modes dominant over the tropics. It is also noted that the impact of IAU may be nonlinearly intertwined with model errors in data-sparse regions, such as the Poles. However, model errors are not accounted for in the hybrid 3DEnVar system used in this study. Additional area-specific features in the verification are provided in the supplemental material.”.
- Fig. S4 further supports our point that the IAU significantly improved forecast errors in most regions, except for the North Pole area and the Southern Hemisphere Ocean, in a statistically significant manner. As illustrated in Fig. 4 in the manuscript, the IAU run assimilated slightly more observations (by 1-2%) throughout the month-long cycling, which is a good indicator that the DA cycling system work more effectively. Overall, it is our belief that our IAU implementation was successfully completed, with positive impacts on the analysis and short-term forecasts.
-
AC4: 'Reply on EC1', Soyoung Ha, 14 Mar 2024
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 204 | 95 | 26 | 325 | 9 | 7 |
- HTML: 204
- PDF: 95
- XML: 26
- Total: 325
- BibTeX: 9
- EndNote: 7
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1