the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Stochastically perturbed physics-tendencies based ensemble mean approach in the WRF model: a study for the North Indian Ocean tropical cyclones
Abstract. Tropical cyclones (TCs) are among the catastrophic natural hazards over the North Indian Ocean (NIO), and they are expected to become more frequent in the upcoming years. TCs occur primarily in the pre-monsoon (April–June) and post-monsoon (October–December) seasons, wreaking havoc on South Asian regions. For reliable alerts and disaster warnings ahead of time, better forecasting of TC features such as track, landfall, intensity, rainfall, and so on is crucial. The present study uses the stochastically perturbed physics-tendencies (SPPT) ensemble-mean approach along with digital filter initialization (DFI) to the initial and boundary conditions for the high-resolution Weather Research and Forecasting (WRF) model. The model’s sensitivity has been investigated for the two NIO TCs, Tauktae (in May 2021) and Nivar (in November 2020), by performing a large number of experiments. Compared with control runs, the track simulations in terms of the reduction in along-track (cross-track) errors for Tauktae and Nivar were improved by 68.8 % (23.4 %) and 28.2 % (40.7 %), respectively, in the DFI experiment. Further improvements were found in the SPPT-based ensemble mean experiments (DFI+SPPT) as the along-track (cross-track) errors, compared to control simulations, were reduced by 65.3 % (27.7 %) and 37 % (54.1 %), for Tauktae and Nivar, respectively. However, the DFI simulations showed a potential to improve the TCs’ track simulation but failed to reduce the error in intensity simulation. On the other hand, DFI+SPPT experiments improved the model's reliability in simulating TCs’ intensity (maximum sustained wind speed and minimum sea-level pressure) considerably. Thus, the DFI+SPPT experiments showed higher skills in simulating the TCs’ characteristics.
- Preprint
(2965 KB) - Metadata XML
-
Supplement
(1714 KB) - BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on wcd-2022-10', Anonymous Referee #1, 11 Jul 2022
General assessment
The authors analyse the ensemble forecast performance of tropical cyclones when a stochastic disturbance is added to the physics trends (SPPT) in the WRF model. They also add a numerical filter initialization (DFI) to the initial state of the forecast. The performance of the DFI+SSPT ensemble forecast is evaluated against a control simulation and against a simulation that uses only DFI. These various model set ups are use to forcast two tropical cyclones which developed in the North Indian Ocean. The authors show that for a number of parameters evaluating the transverse and longitudinal position error, intensity, precipitation associated with CT, DFI+SSPT performs better. Therefore, they recommend that it be used in future forecasts made with WRF.
The paper and especially the result section describe the figures (which are of good quality) in great detail: there are many numbers and acronyms which sometimes make the reading difficult. The text could sometimes be simplified. While the conclusion merely summarises the main results, what is striking is the absence of any discussion of the results in the paper, which I think is necessary.
From a scientific point of view, the study of the impact of the SPPT parameterisation is relevant, since its usefulness for forecasting is debated. On the other hand, the evaluation made here is questionable: the DFI+SSPT ensemble is compared to a deterministic forecast of the CNTL and DFI, which should not allow to conclude (I develop this concern below). The difference between DFI+SSPT and the other experiments is always interpreted as due to the introduction of SPPT but could be equally due the size of the ensemble. Similarly, the difference between CNTL and the other experiments is always interpreted as the introduction of DFI, but could be equally due to the retuning that has been performed.
I therefore recommend that the paper be reconsidered for publication after a major revision.
Major concerns
[1] I have a major concern about the comparison between DFI and DFI+SSPT. This comparison is essential because it supports the only major conclusion of the paper : “The SPPT based ensemble mean approach with digital filter initialization in the WRF model has shown considerable improvements in detecting the cyclone characteristics compared to other experiments.”
The issue is that DFI has one single member while DFI+SSPT is an ensemble of 10 members. The reduction of error in DFI+SSPT could result from a better sampling of possible outcomes. A possible evidence of that is that the “best member” of DFI+SSPT has a comparable score to DFI for the intensity metrics and is often in lesser agreement with the ensemble mean of DFI+SSPT, although it is one of its member!
A more rigorous assessment should compare two ensembles of similar sizes for DFI and DFI+SSPT.
[2] There is no mention of an ocean model in the model setup, so I assume that all experiments are atmosphere-only. It seems necessary to describe the SST product. In particular, because the surface latent heat flux is argued to be the cause of the errors in track of CNTL.
[3] It is likely that there is no or a strong underestimation of the cold wake feedback, and as such, it is not surprising that the experiments tend to overestimate the intensity of the two TCs. The fact that DFI+SSPT captures the peak intensity of Nivar, while the other experiments overestimate it, is probably not a good thing, as the cyclone would have been weaker with a SST cooling. The authors should discuss that.
[4] In their analysis of Fig 11 and Fig. 12, the authors analyse the fact that the precipitation intensity is reduced in DFI+SSPT and in closer agreement with the observations as an improvement due the SPPT scheme. But it is most likely the result of averaging the ensemble. Again, an evidence of that is that the best member of DFI+SSPT has more intense precipitation than DFI+SSPT ensemble mean.
[5] A retuning of DFI and DFI+SPPT has been performed. Which parameters have been retuned? This retuning is as likely to explain the differences in track between CNTL and the other experiments, as the introduction of a DFI in the initial state. It should be described.
[6] The authors suggest on the contrary that the lesser surface turbulent heat flux is the cause of the difference in CNTL track for Tauktae : could they test their hypothesis? I believe that the different sets of parameters would cause CNTL to track more west than the retuned experiments, which would cause lesser surface turbulent heat flux, rather than the contrary.
Minor revisions
line 33 : “frequent and intense TCs” is a confusing statement. Does it mean “more frequent and more intense TCs” : there is certainly no consensus on an increase of TC frequency! Or do they mean “more frequent intense TCs”?
Technical issues
line 52 : influence -> influences
line 125 : Why were two different convection schemes used?
line 198 : from the Best Member -> for the Best Member.
line 384 : which was occured -> which occured
Citation: https://doi.org/10.5194/wcd-2022-10-RC1 - AC1: 'Reply on RC1', Pankaj Kumar, 15 Aug 2022
-
RC2: 'Comment on wcd-2022-10', Anonymous Referee #2, 19 Jul 2022
The paper mainly describes results from WRF ensemble model simulations at 9 km grid spacing using stochastically perturbed physics-tendencies. The authors use the ensemble-mean approach along with digital filter initialization to the initial and boundary conditions to analyse two TCs that made landfall in India. The experiments showed that the ensemble-mean approach, the digital filter initialization approach, and the combination of both methods lead to improvements in the track forecast. Only the combined methods showed an improvement in intensity forecast.
Major comments
- I’m struggling to see how the paper explains processes (dynamics) of TCs and what we learn about TCs.
- From reading the introduction the main aim of the paper is to compare different WRF model configurations and their ability to forecast TCs. I can see the rationale behind that, but the model experiments do not give us new insight into understanding TCs. Maybe the benefits and links to operational forecasting could be made clearer.
- At 9-km grid spacing convection has to be parameterised. Several studies have shown that going to higher resolutions of 4 km and below, where the convective parametrisation needs to be switched off, the intensity of TCs is captured better than in coarser model runs. Have you tested the impact of grid spacing on the model results?
- Are you using fixed SST values for the simulations?
- The ensembles seem not to have a lot of spread, even when you are randomly perturbing the initial conditions and perturbing the stochastic physics. Might your ensemble be underrepresenting the uncertainties?
- Do you have any observations (maybe microwave satellite data/imagery) that gives insights into the observed structure of both storms? I like the plots that show rainfall accumulation in comparison to the observed rainfall. How good is ASCAT at representing the windspeed in TCs?
- I find that parts of the paper are quite descriptive.
Minor comments
- The manuscript (including the abstract) contains too many acronyms, which reduces its readability.
- 49-50: The text could be worded better. I noticed also in other places that the text could be more precise. I suggest going through the text carefully again.
- 67, L. 73: Remove “very”.
- 122: “Horizontal grid spacing of 5 minutes” – It seems something has gone wrong here.
- 143: Dot instead of colon after Table 1.
- “IMD” is undefined. The definition only comes in l. 395, which is in the conclusions.
- Is the “best member” defined only based on the track?
- 185: “relatively some more error” – Please be more specific.
- 234: What does “making the results more perspective” mean?
- It seems nothing from Section 5 has made it into the abstract. What are the key results from this section?
- 362: “the tuned experiments” – Are you referring to all the other runs discussed in the manuscript?
- 363-364: “a hot spot in the central-east ARB pulled the cyclonic circulation towards it” – Not sure what you mean. Are you referring to a maximum upward latent heat flux? The whole section 5.2 could be written in a more precise manner.
- 415: You have discussed the wind structure but not any storm dynamics.
Citation: https://doi.org/10.5194/wcd-2022-10-RC2 - AC2: 'Reply on RC2', Pankaj Kumar, 15 Aug 2022
Status: closed
-
RC1: 'Comment on wcd-2022-10', Anonymous Referee #1, 11 Jul 2022
General assessment
The authors analyse the ensemble forecast performance of tropical cyclones when a stochastic disturbance is added to the physics trends (SPPT) in the WRF model. They also add a numerical filter initialization (DFI) to the initial state of the forecast. The performance of the DFI+SSPT ensemble forecast is evaluated against a control simulation and against a simulation that uses only DFI. These various model set ups are use to forcast two tropical cyclones which developed in the North Indian Ocean. The authors show that for a number of parameters evaluating the transverse and longitudinal position error, intensity, precipitation associated with CT, DFI+SSPT performs better. Therefore, they recommend that it be used in future forecasts made with WRF.
The paper and especially the result section describe the figures (which are of good quality) in great detail: there are many numbers and acronyms which sometimes make the reading difficult. The text could sometimes be simplified. While the conclusion merely summarises the main results, what is striking is the absence of any discussion of the results in the paper, which I think is necessary.
From a scientific point of view, the study of the impact of the SPPT parameterisation is relevant, since its usefulness for forecasting is debated. On the other hand, the evaluation made here is questionable: the DFI+SSPT ensemble is compared to a deterministic forecast of the CNTL and DFI, which should not allow to conclude (I develop this concern below). The difference between DFI+SSPT and the other experiments is always interpreted as due to the introduction of SPPT but could be equally due the size of the ensemble. Similarly, the difference between CNTL and the other experiments is always interpreted as the introduction of DFI, but could be equally due to the retuning that has been performed.
I therefore recommend that the paper be reconsidered for publication after a major revision.
Major concerns
[1] I have a major concern about the comparison between DFI and DFI+SSPT. This comparison is essential because it supports the only major conclusion of the paper : “The SPPT based ensemble mean approach with digital filter initialization in the WRF model has shown considerable improvements in detecting the cyclone characteristics compared to other experiments.”
The issue is that DFI has one single member while DFI+SSPT is an ensemble of 10 members. The reduction of error in DFI+SSPT could result from a better sampling of possible outcomes. A possible evidence of that is that the “best member” of DFI+SSPT has a comparable score to DFI for the intensity metrics and is often in lesser agreement with the ensemble mean of DFI+SSPT, although it is one of its member!
A more rigorous assessment should compare two ensembles of similar sizes for DFI and DFI+SSPT.
[2] There is no mention of an ocean model in the model setup, so I assume that all experiments are atmosphere-only. It seems necessary to describe the SST product. In particular, because the surface latent heat flux is argued to be the cause of the errors in track of CNTL.
[3] It is likely that there is no or a strong underestimation of the cold wake feedback, and as such, it is not surprising that the experiments tend to overestimate the intensity of the two TCs. The fact that DFI+SSPT captures the peak intensity of Nivar, while the other experiments overestimate it, is probably not a good thing, as the cyclone would have been weaker with a SST cooling. The authors should discuss that.
[4] In their analysis of Fig 11 and Fig. 12, the authors analyse the fact that the precipitation intensity is reduced in DFI+SSPT and in closer agreement with the observations as an improvement due the SPPT scheme. But it is most likely the result of averaging the ensemble. Again, an evidence of that is that the best member of DFI+SSPT has more intense precipitation than DFI+SSPT ensemble mean.
[5] A retuning of DFI and DFI+SPPT has been performed. Which parameters have been retuned? This retuning is as likely to explain the differences in track between CNTL and the other experiments, as the introduction of a DFI in the initial state. It should be described.
[6] The authors suggest on the contrary that the lesser surface turbulent heat flux is the cause of the difference in CNTL track for Tauktae : could they test their hypothesis? I believe that the different sets of parameters would cause CNTL to track more west than the retuned experiments, which would cause lesser surface turbulent heat flux, rather than the contrary.
Minor revisions
line 33 : “frequent and intense TCs” is a confusing statement. Does it mean “more frequent and more intense TCs” : there is certainly no consensus on an increase of TC frequency! Or do they mean “more frequent intense TCs”?
Technical issues
line 52 : influence -> influences
line 125 : Why were two different convection schemes used?
line 198 : from the Best Member -> for the Best Member.
line 384 : which was occured -> which occured
Citation: https://doi.org/10.5194/wcd-2022-10-RC1 - AC1: 'Reply on RC1', Pankaj Kumar, 15 Aug 2022
-
RC2: 'Comment on wcd-2022-10', Anonymous Referee #2, 19 Jul 2022
The paper mainly describes results from WRF ensemble model simulations at 9 km grid spacing using stochastically perturbed physics-tendencies. The authors use the ensemble-mean approach along with digital filter initialization to the initial and boundary conditions to analyse two TCs that made landfall in India. The experiments showed that the ensemble-mean approach, the digital filter initialization approach, and the combination of both methods lead to improvements in the track forecast. Only the combined methods showed an improvement in intensity forecast.
Major comments
- I’m struggling to see how the paper explains processes (dynamics) of TCs and what we learn about TCs.
- From reading the introduction the main aim of the paper is to compare different WRF model configurations and their ability to forecast TCs. I can see the rationale behind that, but the model experiments do not give us new insight into understanding TCs. Maybe the benefits and links to operational forecasting could be made clearer.
- At 9-km grid spacing convection has to be parameterised. Several studies have shown that going to higher resolutions of 4 km and below, where the convective parametrisation needs to be switched off, the intensity of TCs is captured better than in coarser model runs. Have you tested the impact of grid spacing on the model results?
- Are you using fixed SST values for the simulations?
- The ensembles seem not to have a lot of spread, even when you are randomly perturbing the initial conditions and perturbing the stochastic physics. Might your ensemble be underrepresenting the uncertainties?
- Do you have any observations (maybe microwave satellite data/imagery) that gives insights into the observed structure of both storms? I like the plots that show rainfall accumulation in comparison to the observed rainfall. How good is ASCAT at representing the windspeed in TCs?
- I find that parts of the paper are quite descriptive.
Minor comments
- The manuscript (including the abstract) contains too many acronyms, which reduces its readability.
- 49-50: The text could be worded better. I noticed also in other places that the text could be more precise. I suggest going through the text carefully again.
- 67, L. 73: Remove “very”.
- 122: “Horizontal grid spacing of 5 minutes” – It seems something has gone wrong here.
- 143: Dot instead of colon after Table 1.
- “IMD” is undefined. The definition only comes in l. 395, which is in the conclusions.
- Is the “best member” defined only based on the track?
- 185: “relatively some more error” – Please be more specific.
- 234: What does “making the results more perspective” mean?
- It seems nothing from Section 5 has made it into the abstract. What are the key results from this section?
- 362: “the tuned experiments” – Are you referring to all the other runs discussed in the manuscript?
- 363-364: “a hot spot in the central-east ARB pulled the cyclonic circulation towards it” – Not sure what you mean. Are you referring to a maximum upward latent heat flux? The whole section 5.2 could be written in a more precise manner.
- 415: You have discussed the wind structure but not any storm dynamics.
Citation: https://doi.org/10.5194/wcd-2022-10-RC2 - AC2: 'Reply on RC2', Pankaj Kumar, 15 Aug 2022
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
637 | 289 | 57 | 983 | 75 | 50 | 30 |
- HTML: 637
- PDF: 289
- XML: 57
- Total: 983
- Supplement: 75
- BibTeX: 50
- EndNote: 30
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1