Next Article in Journal
Sensitivity of Summertime Convection to Aerosol Loading and Properties in the United Arab Emirates
Next Article in Special Issue
Evaluation Schoolyard Wind Comfort Changes Due to Rapid Developments: Case Study of Nanjing, China
Previous Article in Journal
Online Chemical Characterization and Sources of Submicron Aerosol in the Major Mediterranean Port City of Piraeus, Greece
Previous Article in Special Issue
Planetary Boundary Layer Heights from Cruises in Spring to Autumn Chukchi-Beaufort Sea Compared with ERA5
 
 
Article
Peer-Review Record

Application of Bias Correction to Improve WRF Ensemble Wind Speed Forecast

Atmosphere 2021, 12(12), 1688; https://doi.org/10.3390/atmos12121688
by Chin-Cheng Tsai 1,2,*, Jing-Shan Hong 3, Pao-Liang Chang 1, Yi-Ru Chen 1, Yi-Jui Su 1 and Chih-Hsin Li 1
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Reviewer 4: Anonymous
Atmosphere 2021, 12(12), 1688; https://doi.org/10.3390/atmos12121688
Submission received: 25 October 2021 / Revised: 13 December 2021 / Accepted: 14 December 2021 / Published: 16 December 2021
(This article belongs to the Special Issue Atmospheric Boundary Layer: Observation and Simulation)

Round 1

Reviewer 1 Report

The authors present a well-conceived analysis for implementing a bias correction to operationally-forecasted surface wind speeds. Two main artifacts that need correction are the seasonal and diurnal wind-speed oscillations, where the diurnal oscillations are in focus here. The authors presented a clear improvement in the ensemble mean forecast after implementing the proposed strategy. I can recommend that this manuscript be published after minor improvements are made and the English is properly edited.

Some important suggestions for improvement are:

1) Better describe Figure. 2 and its significance. I was particularly confused on the PM column in the figure where I could not understand how those values were obtained.
2) Extensive editing of the English language are required. There are too many corrections required for me to edit here.
3) I would recommend to replace sentences starting with a numeric citation (e.g. [32] on line 67) with the in-text citation format.

Author Response

Response to Reviewer 1 Comments

 

 

Point 1: Better describe Figure. 2 and its significance. I was particularly confused on the PM column in the figure where I could not understand how those values were obtained.

 

Response 1:

The authors appreciate for the Reviewer’s comments. In Figure 2, the value of the PM column was calculated from the average of the first quartile member to the median member in the same subgroup in this study. For instance, we averaged the first quartile (9 from M02) to the median (11 from M03) at the maximum rank, then the maximum rank of the PM column will be 10. The improved description of the proposed PM method was added in the revised manuscript in red color.

 

Point 2: Extensive editing of the English language are required. There are too many corrections required for me to edit here.

 

Response 2:

The authors appreciate for the Reviewer’s comments and the manuscript was revised and corrected in red color.

 

 

Point 3: I would recommend to replace sentences starting with a numeric citation (e.g. [32] on line 67) with the in-text citation format.

 

Response 3:

The authors appreciate for the Reviewer’s comments and the citation format was improved in the revised paper in red color.

Reviewer 2 Report

This paper addresses a very interesting research topic, about 3 bias correction methods applied to WRF ensemble wind speed forecasts. The paper is clear and well written but some more detail information of the simulations, the model and methodology is missing. At the end, the paper could also benefit from a deeper discussion explaining the reasons of the achieved performances. A list of suggestions and corrections are:



Abstract

ok

1. Introduction

55 “and outperform than any single” “and outperform any single”



2. Methods

Fig1: Table 150?

Fig1: Is it a nearly 5-month forecast? From the initial reading I had the idea this study addresses a “regular” widely-used time scale such as 10-days forecast. This time-scale question should be explained.

So far it is not described the boundary/forcing conditions for WRF simulations.



3. Experimental design and materials

Information of vertical and horizontal resolutions, as well as physics option selected in WRF should be included.



4. Results

From Fig3 I see the forecast horizon of interest is 72hours. This information should be included in the text since the beginning of the manuscript.



Besides ME and RMSE, other error metrics could be included, such as correlation coefficient, scatter index etc.



Fig11. The RMSE values should increase with the forecast lead time but in this plot it remains nearly constant. What is the explanation?



Page 15, “4. Discussion” should be “5. Discussion”



318 . “Among the three PMM products, BC01PM performs best” Authors could discuss why this method performs best.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

It is an interesting study and one of its kind. I congratulate authors for taking such important topics to examine and to make an effort to understand the process. I have few comments for authors, which I believe will be helpful for further improving the manuscript.

 

  1. The introduction section needs a modification. Authors have reported relevant studies, but the connectivity between reported studies is not very well. Also, the objective statement needs to be rephrased by highlighting the importance of present work in a better way.
  2. [31] had demonstrated, [32] had shown….It is better to give the name initial before number, i.e., Chen et al. [31], or Tsia et al. [32]. Please do that in whole manuscript.
  3. No site map for performing the experiment? A map of domain with latitude and longitude in the axis with the source of map will improve the manuscript.
  4. The analysis is totally depending on Mean Error and RMSE. Don’t the authors think that other statistical estimates as MBE, Akaike information criterion, or Bias computation will more legitimately enlighten your results? The ME and RMSE seems to dilute the findings here.
  5. Discussion (and Conclusion) section: Authors have talked about the numbers. They should also address why this study is important in terms of adding new knowledge? Also, they should explain the physical meaning of their findings. The manuscript is just reporting a number and equation fit in its present form.
  6. Conclusions of the present study in their present form are not convincing. I request authors to improve the conclusion section with a separate heading “Conclusion”.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Mayor issues:

line 41:
as operated from 2011 in ...

for the year 2011? What are the forecast periods?

line 43
"combining multi-physics perturbation with random initial condition."

presumably random perturbations to the initial condition, not random initial condition? How was the random perturbation generated?

line 64-66
"In Taiwan, the capacity of planed wind power will be enlarged about 6 times before year 2025 that was proclaim by the Bureau of Energy in Taiwan. Therefore, the calibration strategy for ensemble wind speed forecast should be studied and localized."

Nobody will base wind energy forecasts on 10m model wind speeds.


Please write text in between 2. and 2.1 to introduce and prepare the section.

Figure 1: Why does it say "Table 150." in the caption?

Please write text in between 3. and 3.1 to introduce and prepare the section.


line 151 to 153: "BC03 will take 3 model errors independently accumulated from each mean of the three PBL groups and BC20 will take 20 model errors from independent member."

Please define model errors and show how there can be multiple model errors. Ideally with equations.

Line 162-164: "A five-month numerical wind speed forecast of the WEPS with horizontal grid size been equal to 3-km was evaluated." Did the forecast really run continuously for 5 months? Without DA and nudging? It is surprising that at the end of that resembled anything of the observation. Probably you ran several forecasts of, say, 3 days each, over that period? Please rewrite this paragraph and indicate clearly which forecasts were used for what purpose, with forecast periods, initial dates, and a clear definition of the sampling and error computations.

Please write text in between 4. and 4.1 to introduce and prepare the section.


Fig 3: Okay, here I seem to be able to deduce that you are running for 72 hours. And you start your forecasts every day? Or every third day?

Fig 4: These pdfs look strange. Can you tell us why? Usually, the pdf follows a Weibull distribution. But these look nothing like Weibull. Is it just the log scale that is distorting them?  Don't think so, but maybe. Is the log scale really necessary? Your mean seems to be too low, below or around 2m/s? You have errors larger than that, how can that be? Please explain. Maybe if you can plot the raw wind speed data as well it becomes more obvious. Please also provide the raw time-series data as netcdf file, attached to the paper. To me, it looks more like you are showing here a pdf of the error, but not of the wind speed. But then you could not have obs.


You have not presented enough data on the model setup. Domain size, nesting, vertical resolution, driving data, etc.

3km is a difficult resolution for cumulus parameterizations. But as this study was on wind speed I guess it is not that important. You should, however, comment on it, the scale awareness of the chosen schemes, the occurrence of deep convection over the study period, and the likely impact on the wind speeds. If there were a large tropical storm, it would surely impact the windspeeds.


"In addition, the renewable energy policy in Taiwan was officially announced to phase out nuclear power in 2025, the application of the proposed bias correction strategy combined with bias adjustment will be valuable to be applied to correct the ensemble wind speed forecast for wind turbine." No, not with 10m wind speeds.

 

Suggested english corrections in attached pdf.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 4 Report

Still, many issues are not resolved/unclear/confusing.

It is very disappointing to see that the 213 correctness issues in English were not at all followed up, including leaving typos and blatantly grammatically wrong sections in the manuscript. Not to speak of improving unclear sections. This makes it much harder to understand what you are trying to present. Ofen it is not clear if you are talking about your or another study.

 

L 29 Has the error really a “error characteristic has seasonal to diurnal evolution”? What doe that even mean? Does the error evolve?

 

L35: What is the “as long as the chaos nature of atmosphere”? Maybe you mean “as well as”?? .. and chaotic, not chaos...

 

L36 How can it be “limited from hours to weeks.” From hours to weeks would be extended! Who even can do a week wind speed forecast and claim it will match the observation??

 

L42: Really only “with 72 forecast hours since 2011.” ONLY 72? Is that only one 72hour forecast?

L167-170: This is not really any clearer. I appreciate that English is not the author's strength. That does not help. I really recommend expressing it in mathematics. Starting with defining the wind field (v) depending on ensemble member (m) and time (t), v(m,t). Then you can write how you define the error using obs(t) and any v or combination/averaging thereof.

 

L182 “The model is initialized 4 times a day and integrated with a time step of 60s for a total of 72 h.” How is one model initialized 4 times a day? Do you mean you run four individual forecasts with four different initial times?

 

L183: You have to be more specific about the “regional deterministic forecast from CWB” Is it also 3km? I guess, but you should please tell us. Is it also WRF? There must be a paper detailing it?

 

L185- : Now you say that “The lateral boundary conditions are taken from the 10 members of the NCEP Global Ensemble Forecast System (GEFS; [39]).” Now, are you talking about the regional deterministic forecast from CWB or the modeling for this paper? How can you choose LBCs from 10 members for one deterministic model? Or do use one model for initial data and another for LBCs? Do you choose a different member for each member of your ensemble? Confusing …..!

 

Section 4.2: I still don't see how you perform a monthly verification but only show errors over 72 hours. Are averaging all 30(?) days for each forecast hour? What does that mean? Please try to answer this clearly in math, not in English. I.e something like:

 

ME(fh)=1/30 (sum_{day=1}^{30}(ME_day(fh))

 

But probably absolute maximum errors of your final chosen forecast would be more informative.

 

L266-269: “...was calibrated with 3 model biases to represent the system bias ... BC03 with 3-kind model biases...” What are the 3-kind model biases? Please provide math. You can make up your own nomenclature, but you have to explain it, clearly (!). And then you have to use it consistently.

 

L289: “...expereiment….” Really? Now can choose to ignore all the corrections we sent you. It's your choice, albeit not one I would recommend. You are free to assume that they are not necessary, but then add even more blatant spelling mistakes? No.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Back to TopTop