Adjusted spectral correction method for calculating extreme winds in tropical cyclone affected water areas
 Wind Energy Department, Technical University of Denmark
 Wind Energy Department, Technical University of Denmark
Abstract. A method is developed to calculate the extreme wind for tropical cyclone affected water areas. The method is based on the spectral correction method by Larsén et al. (2012) in connection with the use of numerically modeled data, where an enhancement coefficient is derived as a function of wind speed to reflect the large wind fluctuation during tropical cyclones. This is done through calibration with the estimates from Ott (2005) who used the best track data and Holland model to estimate the extreme wind over the Typhoon affected area in the western North Pacific. The method is applied in the current study to three regions where the 50year winds with an effective temporal resolution of 10 minutes are obtained at 10 m, 50 m, 100 m and 150 m. The results are in agreement with Ott (2005) over their study domain, though with much more spatial details of the extreme wind distribution.
Xiaoli Larsén and Søren Ott
Status: closed

RC1: 'Comment on wes202264', Anonymous Referee #1, 25 Jul 2022
The comment was uploaded in the form of a supplement: https://wes.copernicus.org/preprints/wes202264/wes202264RC1supplement.pdf

RC2: 'Encouraging initial steps towards the generalization of the U50 spectral correction method for its use in cyclonic offshore wind climates.', Javier Sanz Rodrigo, 05 Aug 2022
Interesting method extending to tropical cyclone areas the use of the spectral correction method for mesoscale or reanalysis data. The study is limited to data from a rather old study from Ott (2005) which highlights the lack of benchmark data in these conditions. This motivates the authors to stay on the conservative side and adopt a simple correction factor (n) on the previous method for good reasons. The approach is robust and practical. The only thing I'm missing is some quantification of uncertainty, at least with regards to the spatial variability which is something that Ott data can provide. This would give more confidence when you extrapolate to the other large areas without validation data.
Indeed, when deriving the relationships between r, n and u (eqs. 5 and 6), you mention that this could be site dependent but you end up using the location with the stronger winds to be on the conservative side. However, when you extrapolate to other cyclone areas you end up applying the same relationship at every grid point over very large areas so we may anticipate large errors normally biased to overprediction of the extreme winds. It would be interesting to apply the regression technique locally to each of the Ott (2005) grid points to compare with your results using eqs 5 and 6. This way you can provide some quantification of the potential bias/uncertainty introduced by your conservative approach at least for the validation area.
Given the high uncertainty of the vertical extrapolation methods why not sticking to the more conservative and simpler Charnock method, which has confluence with Andreas equation up to 40 m/s? In fact, seeing the parabolic dependency of n with wind speed in Figure 4, one could argue that the n coefficient may be a generalization of the Charnock constant for cyclone wind events. Maybe you can try to relate the two?
The paper has value as it is so it could be published after some editing but it would be more solid if these suggestions are addressed.
The new recipe is described in the paragraph of p5.128. I think it would be easier to follow if you use a numbered list with all the steps from input to output.
Some editorial corrections/suggestions:
P1.21: particularly for over water > particularly over water
P2.36: for the assessing > for assessing
P2.37: are forecast > are forecasts
P2.38: improvement > improvements
P2.44: are of spatial > come with a spatial
P2.45: has used > have used; extreme wind > extreme winds; data validation > validation
P2.48: suffers from the smoothing effect…> suffer from the smoothing effect, introduced by a coarse grid that facilitates the convergence of the model.
P2.53: wind variability to… > wind variability from modeled time series through a spectral model. Thus, the corrected time series will follow the power spectrum down to the temporal resolution of the measurements.
P2.56: create extreme wind > predict extreme winds
P3.59: same way as for the midlatitude > same way for midlatitude
P3.62: the strenth of the methods from Ott (2005) and Larsen et al (2012).
P3.75: which were from the best > which is derived from best track data, and Holland model, for an area over…
P3.76: in contour lines > as contour lines
P3.80: and, accordingly, translates into low spectral energy at high frequencies when compared to measurements.
P3.85: zeroth and secondorder
P4.110: from Gage and Nastrom (1986) for the same area? (please specify)
P4.118: WRFSWAN > have you coupled the SWAN model? This is not described in the Annex
P4.119: the higher resolution of the WRF model simulation, at 2 km, produced a spatial wind variability that is 3 to 5 times larger than that of the CFSv2 data at 25 km resolution.
P5.128: To define n in Eq.4, firstly, annual wind maxima u_max are extracted for a period of 32 years from 1979 to 2010 from CFSR1 data. Then, the 50year…
P6.Fig2: calculated, in the SN and WE directions, from the WRF
P6.131: to train the CFSR1 data > to derive a regression model for the CFSR1 data. (avoid machine learning jargon)
P6.133: in grey dots, covering a range from 1 to 2.6.
P7.Eq(6): replace u by u_max.
P7.149: merges > coincides
P7.154: a couple of hundreds of meters. We > 200 m. Therefore, we
P7.157: to Sonde … conditions, as shown in e.g. Powell et al. (2003) and Giammanco et al. (2013)
P8.Fig4: (a) Distribution… with respect to U_50ott; (b) Derived…
P9.167: algorithm
P9.181: the difference > a difference
P10:190: move the link to the references section
P10.193: move the zenodo link to the references section using doi citation (thanks for sharing!)
P10.203: Imberger and Larsén (2022) show
P10.206: This suggest an important source of uncertainty associated to the input reanalysis data.
P10.214: on only one parameter, the wind speed, allows to use the the SCTC method in areas of strong winds that are not necessarily affected by tropical cyclones. (Is this what you mean?)
P12.234: two purposes for using
P12.240: from the surface to a pressure level
P12.243: as initial and boundary conditions for WRF.
P12.244: OISST data were used to define the sea surface temperature conditions.
P12.244: started… and ended at
P12.245: The model outputs are recorded every 10 min.

RC3: 'Comment on wes202264', Anonymous Referee #3, 24 Aug 2022
Adjusted spectral correction method for calculating extreme winds in tropical cyclone affected water areas – REVIEW
The article of Larsen and Ott addresses an interesting and highly relevant topic in the field of site assessment for wind energy.The authors present a promising approach towards a reliable extreme wind estimation from reanalysis data in regions affected by tropical cyclones.
However, a few points described in the following definitely need to be addressed before publication.
The method development in section 2 particularly 2.2 lacks clarity and needs to be revised taking into account the following comments and questions.
Line 84: “The maximum wind that occurs once a year …” is not precise in my opinion. In the framework of a Poisson process of wind velocities exceeding a threshold, Eq. (1) gives the velocity which is on average exceeded once in a period of T0. At this point there is no maximum estimation involved… However, there is a relation to the annual maximum but to be precise, it is not the same thing.
Generally, the description of the SC method is pretty confusing to me. As far as I remember, the SC method calculates a correction factor sometime called smoothing effect as the ratio of Eq (1) for corrected and uncorrected spectra. This Factor is used to correct the the 50year wind estimates of the reanalysis data, esimated by e.g. annual maximum method. Eq(1) is thus not directly used for estimating U50. O am I wrong?
In my experience a direct estimation with Eq (1) fails due to the various unfulfilled assumptions (Gaussian wind, …).
Line 101119: This is an interesting paragraph illustrating the effect of a typhoon on the spectra. But as far as I can see, it does not directly contribute to the SCTC method. Thus the paragraph could be put into an extra section?!
Line 125: Do you really need to mention version 2 of the enhanced spectrum?
Line 131: “We regrid…” How do you regrid exactly? By bilinear interpolation?
Line 134 and following: From here on, I get pretty confused. This might be also due to a lack of my expertise since some years have passed since I last worked at similar topics. However, since I am not completely new to the topic I should be able to follow you explanation of the method more easily.
Do you match the SCTC results by tuning n to match the U50,Ott wind?
Is there on r for every grid point? If so, how could you get a relation for like EQ 5 for every grid point? How exactly do you estimat alpha and beta? Why do you choose a quadratic dependence of n on r?
What is u exactly in Eq(6)?
I get that you show r dependent on U50,Ott in figure 4a. But how do you get a dependence on the annual maxima? Does this lead to a different r and n every year?
I thank the authors in advance for clarifications.
Last but not least, the conclusions in section 4 are no conclusions in my opinion. The section is just a slightly rephrased version of the abstract. However, the authors do offer some conclusions in the discussion. This might also be a matter of style nowadays. But I like conclusions to be conclusions not abstracts :)
Other minor corrections have been mentioned in other comments already.

AC1: 'Comment on wes202264', Xiaoli Larsén, 23 Sep 2022
Dear Editor, Reviewers
We have prepared our replies to the three reviewers' comments combined in one document "Reply_to_Reviewerswes202264 20220923.pdf" and uploaded here.
We sincerely thank the reviewers for their effort, which has helped improving the paper greatly.
Best, Xiaoli G. Larsén and Søren Ott
Status: closed

RC1: 'Comment on wes202264', Anonymous Referee #1, 25 Jul 2022
The comment was uploaded in the form of a supplement: https://wes.copernicus.org/preprints/wes202264/wes202264RC1supplement.pdf

RC2: 'Encouraging initial steps towards the generalization of the U50 spectral correction method for its use in cyclonic offshore wind climates.', Javier Sanz Rodrigo, 05 Aug 2022
Interesting method extending to tropical cyclone areas the use of the spectral correction method for mesoscale or reanalysis data. The study is limited to data from a rather old study from Ott (2005) which highlights the lack of benchmark data in these conditions. This motivates the authors to stay on the conservative side and adopt a simple correction factor (n) on the previous method for good reasons. The approach is robust and practical. The only thing I'm missing is some quantification of uncertainty, at least with regards to the spatial variability which is something that Ott data can provide. This would give more confidence when you extrapolate to the other large areas without validation data.
Indeed, when deriving the relationships between r, n and u (eqs. 5 and 6), you mention that this could be site dependent but you end up using the location with the stronger winds to be on the conservative side. However, when you extrapolate to other cyclone areas you end up applying the same relationship at every grid point over very large areas so we may anticipate large errors normally biased to overprediction of the extreme winds. It would be interesting to apply the regression technique locally to each of the Ott (2005) grid points to compare with your results using eqs 5 and 6. This way you can provide some quantification of the potential bias/uncertainty introduced by your conservative approach at least for the validation area.
Given the high uncertainty of the vertical extrapolation methods why not sticking to the more conservative and simpler Charnock method, which has confluence with Andreas equation up to 40 m/s? In fact, seeing the parabolic dependency of n with wind speed in Figure 4, one could argue that the n coefficient may be a generalization of the Charnock constant for cyclone wind events. Maybe you can try to relate the two?
The paper has value as it is so it could be published after some editing but it would be more solid if these suggestions are addressed.
The new recipe is described in the paragraph of p5.128. I think it would be easier to follow if you use a numbered list with all the steps from input to output.
Some editorial corrections/suggestions:
P1.21: particularly for over water > particularly over water
P2.36: for the assessing > for assessing
P2.37: are forecast > are forecasts
P2.38: improvement > improvements
P2.44: are of spatial > come with a spatial
P2.45: has used > have used; extreme wind > extreme winds; data validation > validation
P2.48: suffers from the smoothing effect…> suffer from the smoothing effect, introduced by a coarse grid that facilitates the convergence of the model.
P2.53: wind variability to… > wind variability from modeled time series through a spectral model. Thus, the corrected time series will follow the power spectrum down to the temporal resolution of the measurements.
P2.56: create extreme wind > predict extreme winds
P3.59: same way as for the midlatitude > same way for midlatitude
P3.62: the strenth of the methods from Ott (2005) and Larsen et al (2012).
P3.75: which were from the best > which is derived from best track data, and Holland model, for an area over…
P3.76: in contour lines > as contour lines
P3.80: and, accordingly, translates into low spectral energy at high frequencies when compared to measurements.
P3.85: zeroth and secondorder
P4.110: from Gage and Nastrom (1986) for the same area? (please specify)
P4.118: WRFSWAN > have you coupled the SWAN model? This is not described in the Annex
P4.119: the higher resolution of the WRF model simulation, at 2 km, produced a spatial wind variability that is 3 to 5 times larger than that of the CFSv2 data at 25 km resolution.
P5.128: To define n in Eq.4, firstly, annual wind maxima u_max are extracted for a period of 32 years from 1979 to 2010 from CFSR1 data. Then, the 50year…
P6.Fig2: calculated, in the SN and WE directions, from the WRF
P6.131: to train the CFSR1 data > to derive a regression model for the CFSR1 data. (avoid machine learning jargon)
P6.133: in grey dots, covering a range from 1 to 2.6.
P7.Eq(6): replace u by u_max.
P7.149: merges > coincides
P7.154: a couple of hundreds of meters. We > 200 m. Therefore, we
P7.157: to Sonde … conditions, as shown in e.g. Powell et al. (2003) and Giammanco et al. (2013)
P8.Fig4: (a) Distribution… with respect to U_50ott; (b) Derived…
P9.167: algorithm
P9.181: the difference > a difference
P10:190: move the link to the references section
P10.193: move the zenodo link to the references section using doi citation (thanks for sharing!)
P10.203: Imberger and Larsén (2022) show
P10.206: This suggest an important source of uncertainty associated to the input reanalysis data.
P10.214: on only one parameter, the wind speed, allows to use the the SCTC method in areas of strong winds that are not necessarily affected by tropical cyclones. (Is this what you mean?)
P12.234: two purposes for using
P12.240: from the surface to a pressure level
P12.243: as initial and boundary conditions for WRF.
P12.244: OISST data were used to define the sea surface temperature conditions.
P12.244: started… and ended at
P12.245: The model outputs are recorded every 10 min.

RC3: 'Comment on wes202264', Anonymous Referee #3, 24 Aug 2022
Adjusted spectral correction method for calculating extreme winds in tropical cyclone affected water areas – REVIEW
The article of Larsen and Ott addresses an interesting and highly relevant topic in the field of site assessment for wind energy.The authors present a promising approach towards a reliable extreme wind estimation from reanalysis data in regions affected by tropical cyclones.
However, a few points described in the following definitely need to be addressed before publication.
The method development in section 2 particularly 2.2 lacks clarity and needs to be revised taking into account the following comments and questions.
Line 84: “The maximum wind that occurs once a year …” is not precise in my opinion. In the framework of a Poisson process of wind velocities exceeding a threshold, Eq. (1) gives the velocity which is on average exceeded once in a period of T0. At this point there is no maximum estimation involved… However, there is a relation to the annual maximum but to be precise, it is not the same thing.
Generally, the description of the SC method is pretty confusing to me. As far as I remember, the SC method calculates a correction factor sometime called smoothing effect as the ratio of Eq (1) for corrected and uncorrected spectra. This Factor is used to correct the the 50year wind estimates of the reanalysis data, esimated by e.g. annual maximum method. Eq(1) is thus not directly used for estimating U50. O am I wrong?
In my experience a direct estimation with Eq (1) fails due to the various unfulfilled assumptions (Gaussian wind, …).
Line 101119: This is an interesting paragraph illustrating the effect of a typhoon on the spectra. But as far as I can see, it does not directly contribute to the SCTC method. Thus the paragraph could be put into an extra section?!
Line 125: Do you really need to mention version 2 of the enhanced spectrum?
Line 131: “We regrid…” How do you regrid exactly? By bilinear interpolation?
Line 134 and following: From here on, I get pretty confused. This might be also due to a lack of my expertise since some years have passed since I last worked at similar topics. However, since I am not completely new to the topic I should be able to follow you explanation of the method more easily.
Do you match the SCTC results by tuning n to match the U50,Ott wind?
Is there on r for every grid point? If so, how could you get a relation for like EQ 5 for every grid point? How exactly do you estimat alpha and beta? Why do you choose a quadratic dependence of n on r?
What is u exactly in Eq(6)?
I get that you show r dependent on U50,Ott in figure 4a. But how do you get a dependence on the annual maxima? Does this lead to a different r and n every year?
I thank the authors in advance for clarifications.
Last but not least, the conclusions in section 4 are no conclusions in my opinion. The section is just a slightly rephrased version of the abstract. However, the authors do offer some conclusions in the discussion. This might also be a matter of style nowadays. But I like conclusions to be conclusions not abstracts :)
Other minor corrections have been mentioned in other comments already.

AC1: 'Comment on wes202264', Xiaoli Larsén, 23 Sep 2022
Dear Editor, Reviewers
We have prepared our replies to the three reviewers' comments combined in one document "Reply_to_Reviewerswes202264 20220923.pdf" and uploaded here.
We sincerely thank the reviewers for their effort, which has helped improving the paper greatly.
Best, Xiaoli G. Larsén and Søren Ott
Xiaoli Larsén and Søren Ott
Xiaoli Larsén and Søren Ott
Viewed
HTML  XML  Total  BibTeX  EndNote  

178  69  14  261  2  2 
 HTML: 178
 PDF: 69
 XML: 14
 Total: 261
 BibTeX: 2
 EndNote: 2
Viewed (geographical distribution)
Country  #  Views  % 

Total:  0 
HTML:  0 
PDF:  0 
XML:  0 
 1