the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Sensitivity of the Weather Research and Forecasting model (WRF) to downscaling extreme events over Northern Tunisia
Abstract. Rainfall is one of the most important variables for water and flood management. We investigate the capacity of the Weather Research and Forecasting model (WRF) to dynamically downscale the ECMWF Re-Analysis data for Northern Tunisia. This study aims to examine the sensitivity of WRF rainfall estimates to different Planetary Boundary Layer (PBL) and Cumulus Physics (Cu) schemes. The verification scheme consists of three statistical criteria (Root Mean Square Error (RMSE), Pearson correlation, and the ratio bias coefficient). Moreover, the FSS coefficient (fraction skill score) and the quality coefficient SAL (structure amplitude latitude) are calculated. The database is composed of four heavy events covering an average of 318 rainfall stations. We mean by heavy event, each event occurred a rainfall of more than 50 mm per observed day at least in one rainfall station. The sensitivity study showed that there is not a best common combination scheme (PBL and Cu) for all the events. The average of the best 10 combinations for each event is adopted to get the ensemble map. We conclude that some schemes are sensitive and others less sensitive. The best three performing schemes for PBL and Cu parametrizations are selected for future rainfall estimation by WRF over Northern Tunisia.
- Preprint
(1895 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on nhess-2020-376', Anonymous Referee #1, 25 Jan 2021
Title: Sensitivity of the Weather Research and Forecasting model (WRF) to downscaling extreme events over Northern Tunisia
Authors: Saoussen Dhib, Víctor Homar, Zoubeida Bargaoui, Mariadelmar VichThe aim of this study is to examine the sensitivity of WRF rainfall estimates for different Planetary Boundary Layer (PBL) and Cumulus Physics (Cu) schemes. Sensitivity studies have shown that there is no best common combination scheme (PBL and Cu) for all events. The average of the 10 best combinations for each event is used to map the ensemble. The authors conclude that some schemes are more sensitive and others less sensitive.
The abstract lacks important information. A few sentences about the setup, time simulation, and significant results are desirable. Provide the readers with this information here so they can decide if the paper is useful to their needs and so they can determine where to find the information they need within the main text.
The introduction also lacks the study purpose. In addition, the scientific contributions are not clear in the current form of the manuscript and should be further elaborated. What the gap will be filled?
From the description of the methods given, it is not clear what advancements have been made over other designs already described in the literature. Has there been any previous research using this method? What are the superior methods used in this study?
The results from the study are scientifically interesting and may represent a good examination to get the benefits of processing the entire simulations processes. However, there are several important concerns with the manuscript, such as the presence of a large number of vague descriptions, questionable arguments, and the lack of in-depth discussions. To resolve the fundamental flaws, an essential re-work on discussions is believed to be required, and the resultant manuscript would be so different from this one that it would be considered a new piece of research work.
The conclusion does not have significant finding according to the important results. Which scheme should be recommendation by authors?Line 32: Please check the correct format of abbreviation! It should be …blab la bla (MSG MPE) …
Line 68: … Radar Topography Mission (SRTM)… Is this correct abbreviation?
Line 73: What is mean ‘Heavy event’? Please elaborated it!
Line 78: Why the authors select 2 stations? Where is this location? Please explain it more clearly!
Figure 4 is not clear information! The authors employ both outer and inner domains, where is the boundaries? Give the clear information in your figure!
Line 159: It is difficult to understand this sentence.
Line 162 – 164: Please give the clear explanation for quantile quantile comparison?
Line 173: Why the authors are choosing a threshold of 0.1 mm? What is the reason?
Line 182: The blank space should be removed!Citation: https://doi.org/10.5194/nhess-2020-376-RC1 -
AC1: 'Reply on RC1', Saoussen Dhib, 31 Mar 2021
Response to reviewers’ comments
Title: Sensitivity of the Weather Research and Forecasting model (WRF) to downscaling extreme events over Northern Tunisia Authors: Saoussen Dhib, Víctor Homar, Zoubeida Bargaoui, Mariadelmar Vich
We thank the reviewers for their time and helpful comments. We have addressed each point below. Reviewer comments are shown in bold, while author responses are shown in unformatted text.
The aim of this study is to examine the sensitivity of WRF rainfall estimates for different Planetary Boundary Layer (PBL) and Cumulus Physics (Cu) schemes. Sensitivity studies have shown that there is no best common combination scheme (PBL and Cu) for all events. The average of the 10 best combinations for each event is used to map the ensemble. The authors conclude that some schemes are more sensitive and others less sensitive.
Response: We thank the reviewer for this feedback. We have ensured that the revised manuscript addresses the main issues that were brought up in the review.
The abstract lacks important information. A few sentences about the setup, time simulation, and significant results are desirable. Provide the readers with this information here so they can decide if the paper is useful to their needs and so they can determine where to find the information they need within the main text.
Response: Thank you for critical feedback. We have ensured that the overall purpose of the study is more clearly defined in the revised abstract. These paragraphs will be added.
The period January 2007- August 2009 is investigated based on a daily rainfall network composed by 318 rain gauges covering 36000 km². We focus on heavy rainfall situations composed by days when a threshold of 50 mm/day is exceeded at least one recording location. Thus, a total of 77 heavy rainfall days are identified. Inverse distance is adopted to elaborate rain maps which are compared with maps obtained using WRF estimations (10 km resolution). The calibrated power coefficient is found 1.2 using cross-validation approach. MSGMPE failed to detect as rainy 11 days out of 77. So we propose to use WRF too to predict them. To run the comparison only 4 representative days out 11 are studied because the important time of simulations: the day with highest average spatial rainfall, the day with highest spatial standard deviation, one typical day with spatial average and standard deviations represented the most common situation and the day ranked 2 with respect to spatial average.
For each studied day, we begin the simulation at 12pm the day before that means 6 hours to attend some stability of the model before the studied day beginning. The end of the simulation is one day after. As a first test of the performance of the various parameters schemes, we did a quantile quantile comparison for the 12/01/2009 using the 3 parameters (PBL, Cu, Mp) schemes. The results show that there is not an important variability in the performance of the various Mp schemes. Then, we decided to continue the sensitivity study only with PBL and Cu parameters. Using WRF, simulation of 10 combination takes in average 4 days. It depends on the UIB department server availability. Also, usually we should run again about 20% of the simulations each time because of WRF crash or internet interruption. Then, for the 99 simulations of the different Cu and PBL for all the rainy days take about 190 days.
The introduction also lacks the study purpose. In addition, the scientific contributions are not clear in the current form of the manuscript and should be further elaborated. What the gap will be filled?
Response: WRF has many parametrizations the most common in the literature sensitive parametrizations for rainfall localization and intensity, are Cu, and much less PBL and Mp (Hewitson et al.2004; Tadross et al.2006). In this study, for each day, we will need to do 792 simulations (Cu with 11 schemes, PBL with 9 schemes and Mp with 8 schemes) to get a satisfactory configurations which can be find only with testing numerous physical parameterizations. 10 simulations takes in average 4 days which depends on the UIB department server availability. Also, usually we should run again about 20% of the simulations each time because of WRF crash or internet interruption. Then, for the 792 simulations we will need 380 days for each event. It is clear that the simulations duration is very long and it should be reduced with conservation of the performance of the WRF rainfall estimation. The aim of this study is to choose the best representing schemes of extreme rainfall by WRF over Northern Tunisia which will make the use of WRF more efficient for users in short time.
From the description of the methods given, it is not clear what advancements have been made over other designs already described in the literature. Has there been any previous research using this method? What are the superior methods used in this study?
Response: All previous mentioned research, examine few schemes for each parameter and aim to choose one best combination which is not representative for the different climate variability and the rainfall intensity. It is the first sensitivity study for rainfall estimation over Tunisia. The originality of this research appear in two components. Firstly, we will test all the schemes of each parameter. Secondly, we will choose not one best combination but the 10 best combinations which will be averaged later to give an ensemble map. This ensemble map will give the best estimation in comparison with all the other individual combinations. Secondly, based on the sensitivity study we selected 3 best schemes for each parameter (PBL, Cu) which have the ability to give a good results for the various extreme event types.
The results from the study are scientifically interesting and may represent a good examination to get the benefits of processing the entire simulations processes. However, there are several important concerns with the manuscript, such as the presence of a large number of vague descriptions, questionable arguments, and the lack of in-depth discussions. To resolve the fundamental flaws, an essential re-work on discussions is believed to be required, and the resultant manuscript would be so different from this one that it would be considered a new piece of research work.
Response
Thank you for the remarks about the discussion. We are working, now, to make it deeper and to highlight the influence of schemes with the location (topographic area, coastal area,…). Also, we are trying to highlight the influence of convective cumulus schemes in rainfall amounts within the different type of the studied rainy days.
The conclusion does not have significant finding according to the important results. Which scheme should be recommendation by authors?
Response: Thank you for the encouragement. Actually in the conclusion, we summarized the main useful tools of the sensitivity study which are: using different evaluation coefficients performance, the ensemble map of the best 10 combinations of each studied day, and the identification of three best schemes of each parameter (Cu and PBL) for the whole study area. However, we didn’t mentioned the real names of these schemes. In the revised manuscript, we will add the names of the best schemes. Furthermore, we will highlight the amount of wined time by using the 9 combinations of the best 3 selected schemes of each parameter (Cu and PBL) at the place of using an ensemble map of the combination of all the schemes (99 combinations).
Line 32: Please check the correct format of abbreviation! It should be …blab la bla (MSG MPE) …
Response: We corrected it. Meteosat Second Generation Multi-sensor Precipitation Estimate (MSG MPE) Line 68: … Radar Topography Mission (SRTM)… Is this correct abbreviation? Response: We corrected it. Shuttle Radar Topography Mission Line 73: What is mean ‘Heavy event’? Please elaborated it! Response: Heavy events are defined as those rainy days exceeding 50 mm/day for at least one station. Line 78: Why the authors select 2 stations? Where is this location? Please explain it more clearly!
Response: The heavy days evaluated by MSGMPE are selected based on a threshold of 50 mm/day for at least one station out of 318 stations. For this study we selected the rainy days with a threshold of at least 2 stations surpassing 50 mm/day which avoid very spatial localized extreme events. There is not a specified location of these stations. They can be in urban zone, forest, orographic area,... Figure 4 is not clear information! The authors employ both outer and inner domains, where is the boundaries? Give the clear information in your figure!
Response: We will change Figure 4 with another figure containing the two domains with all the coordinates information. Line 159: It is difficult to understand this sentence.
Response: For the PBL schemes simulation, the Cu scheme was fixed to 2 and Mp scheme to 6 (Fig.5a). Line 162 – 164: Please give the clear explanation for quantile quantile comparison?
Response:
In statistics, a quantile-quantile (Q-Q) plot is a probability plot which associates the frequency of non-overshoot. It is a robust estimator, which doesn’t compare pixel by pixel but compares the shapes of two probability distributions by plotting their quantiles against each other. If the two distributions being compared are similar, the points in the Q–Q plot will approximately lie on the line y = x. We compared the quantile distrubition of WRF estimation (spatial resolution of 10 km) against the quantile distribution of the correspondent ground map. Firstly The Q-Q graph shows if there is an overestimation or an underestimation by WRF. Figure 5 plots have WRF as abscissa and ground rainfall as ordinate. If WRF curve is over the X=Y line that means WRF underestimate rainfall and if WRF curve is below x=y line that means WRF overestimate rainfall. Secondly, Q-Q plot highlight the capacity of WRF estimation of the different type of rainfall: low (less than 20 mm/day/pixel), medium (between 20 and 60 mm/day/pixel) and maximum values (more than 60 mm/day/pixel).
Line 173: Why the authors are choosing a threshold of 0.1 mm? What is the reason?
Response: A threshold of 0.1 mm/pixel is used in SAL and FSS verification to distinguish between rainy and no rainy pixels. We chose 0.1 threshold to detect the rainy pixels. Line 182: The blank space should be removed!
Response: We corrected it. Thank you.
Thank you for the feedback. In the updated manuscript, we will ameliorate the redaction quality of the whole paper, add deeper discussion, compare our results with many previous finds as suggested.
We would again like to thank the reviewers for their time and helpful comments.
Citation: https://doi.org/10.5194/nhess-2020-376-AC1
-
AC1: 'Reply on RC1', Saoussen Dhib, 31 Mar 2021
-
RC2: 'Comment on nhess-2020-376', Anonymous Referee #2, 02 Feb 2021
This work presents a sensitivity analysis to the modification of the cumulus and boundary layer parameterizations in WRF in extreme precipitation events. The objective of this study seems to be focused on obtaining a better prediction of extreme events in the study area based on an optimal ensemble.
Although the underlying idea of the work is interesting, the structure of the work, the exposition of the methodology and the discussion of the results is poor. The objectives of the work are not clearly established, and the low quality writing makes very difficult to find a storyline. In general, the article is written in a fuzzy and lazy way. This makes very difficult to follow it and understand the message that the authors try to send. For all these reasons, I cannot recommend this paper for publication.
The authors should rewrite the article again, trying to give it a rational structure, a more complete and rational exposition of the configuration of the experiments carried out, and a more detailed and clarifying analysis of the results.
Below I list as an example some of the points on which I base my review
l-15. latitude should be location
l-22. "What schemes?"-After line 58. What are your objetives? What do you can provide as new knowledge?
l-63. In situ data....Observations??
l-68. spatial resolution?
l-72,94. Where do you interpolate precipitation data? Some information about the mesh must be provided.
l-94. Why interpolate data?? In fact, the results of the crossvalidation show not very good results.
l-112 are these the variables you use to build initial and boundary conditions. This has no sense.
l- 119. (Figure 4) What domain is that? How many domains do you use? Which resolucion?
** Here, a complete model configuration should be exposed (radiation, LSM, etc), as well as spatial configuration (vertical levels, soil ... etc) In addtion how simulations were done (simulated time, spin-up period,
l-125 The description of Cumulus schemes is not usefull at all. If authors try to explain something they have to clasify the schemes used, for example if they are or not flux mass schmes, trigger mecanish . etc.
-134 The same as before
- About the metrics.. It is really necessary to use all these metrics? Each metric focus on different aspects of the skill. What is the sense of use a metric that is sum of all?
l-244 You inverse the metric (X) ... what do you mean?? 1/X???
**Some more comments.
-Figure captions should be improved. All information needed to interpret the figures must be in the caption.
- You dont argue why the selection of these 4 cases. I do not understand why undetected evens by satellite are candidates for the case selection.
- FIgures 8, 9, 10, 11. why they have different styles??
Citation: https://doi.org/10.5194/nhess-2020-376-RC2 -
AC2: 'Reply on RC2', Saoussen Dhib, 31 Mar 2021
Response to reviewer comments
Title: Sensitivity of the Weather Research and Forecasting model (WRF) to downscaling extreme events over Northern Tunisia Authors: Saoussen Dhib, Víctor Homar, Zoubeida Bargaoui, Mariadelmar Vich
We thank the reviewers for their time and helpful comments. We have addressed each point below. Reviewer comments are shown in bold, while author responses are shown in unformatted text.
This work presents a sensitivity analysis to the modification of the cumulus and boundary layer parameterizations in WRF in extreme precipitation events. The objective of this study seems to be focused on obtaining a better prediction of extreme events in the study area based on an optimal ensemble.
Response:
Yes, this is the main objective. Thank you.
Although the underlying idea of the work is interesting, the structure of the work, the exposition of the methodology and the discussion of the results is poor.
Response: the structure of the work needs to be improved. Authors recognize this fact. In the next proposal, section 2 will be dedicated to present data and section 2 for the methodology. In section 2 the reference to our previous work using MSG-MPE will be added. Paragraph 81-91 will be reformulated. the subsection 2.2 will be removed (as title). The word "event" will be replaced by "day" in all the manuscript. Also Section 2.4 will be suppressed (as title).
-The methodology is based on the use of conventional performance criteria such as RMSE, Pearson correlation coefficient and ratio test in the comparison of the maps obtained by interpolating ground observations and those obtained using WRF predictions. In addition more sophisticated performance criteria such as SAL and FSS are used because they consider the internal spatial variability of the rainfall fields. Moreover, the methodology adopts the combination of PBL and Cu schemes assuming a given MP scheme. The criteria as rescaled and summed in order to base the comparison on a single composed criterion.
-the analysis is applied to the study of 4 days representing different conditions with respect to rainfall amounts and spatial variability. Results show that no single scheme can be recommended for the 4 studied days. The more difficult to be predicted are the day displaying the greatest rainfall amounts (12/1/2009) and that with the highest spatial variability (13/9/2008).
The objectives of the work are not clearly established, and the low quality writing makes very difficult to find a storyline.
Response: Thank you for critical feedback. We ensure that the overall purpose of the study will be more clearly defined.
WRF has many parametrizations the most common in the literature sensitive parametrizations for rainfall localization and intensity, are Cu, and much less PBL and Mp (Hewitson et al.2004; Tadross et al.2006). In this study, for each day, we will need to do 792 simulations (Cu with 11 schemes, PBL with 9 schemes and Mp with 8 schemes) to get a satisfactory configurations which can be find only with testing numerous physical parameterizations. 10 simulations takes in average 4 days which depends on the UIB department server availability. Also, usually we should run again about 20% of the simulations each time because of WRF crash or internet interruption. Then, for the 792 simulations we will need 380 days for each event. It is clear that the simulations duration is very long and it should be reduced with conservation of the performance of the WRF rainfall estimation. The aim of this study is to choose the best representing schemes of extreme rainfall by WRF over Northern Tunisia which will make the use of WRF more efficient for users in short time.
In general, the article is written in a fuzzy and lazy way. This makes very difficult to follow it and understand the message that the authors try to send. For all these reasons, I cannot recommend this paper for publication.
Response: We ensure that the revised manuscript will be edited by a native English speaker to improve comprehension and quality.
The authors should rewrite the article again, trying to give it a rational structure, a more complete and rational exposition of the configuration of the experiments carried out, and a more detailed and clarifying analysis of the results.
Response:
Thank you for the suggestion. In the updated manuscript, we will ameliorate the redaction quality of the whole paper, add deeper discussion, compare our results with many previous founds as suggested.
Below I list as an example some of the points on which I base my review l-15. latitude should be location
Response: Yes, thank you. l-22. "What schemes?"
Response: WRF is based on parameters representing the various physical processs. We consider particularly Cumulus (Cu), Planetary boundary level (PBL) and micro physics (MP).These parameters have many options which depend of the physical process complexity. These options are what we call "schemes". The schemes description is presented in Table 1. They are 9 different schemes for representing Cu and 11 for representing PBL.
-After line 58. What are your objectives? What do you can provide as new knowledge?
Response: the reviewer is right. The literature review shows that the performance of the parameterization of WRF is not known a priori. No best parameterization can be recommended for a given case study. It depends on the metrics used for evaluation, and on the case study itself (the geographical region and the type of rainfall event). All previous mentioned research examines few schemes from each parameter. It is the first sensitivity study for rainfall estimation over Tunisia. The originality of this research appears in two components. Firstly, we will test all the schemes of each parameter. Secondly, we will choose not one best combination but the 10 best combinations which will be averaged later to give an ensemble map. This ensemble map will give the best estimation in comparison with all the other individual combinations. Secondly, based on the sensitivity study we selected 3 best schemes for each parameter (PBL, Cu) which have the ability to give a good results for the various extreme event types.
l-63. In situ data....Observations??
Response: In situ data are observed rainfall amounts using the national rainfall network of Tunisia. Observations are daily rainfall.
l-68. spatial resolution?
Response: interpolation of in situ data is achieved using a 10 km spatial resolution
l-72,94. Where do you interpolate precipitation data? Some information about the mesh must be provided.
Response: the mesh is 10 km. It is represented in figure 1.
l-94. Why interpolate data?? In fact, the results of the crossvalidation show not very good results.
Response: The in-situ stations are not well scattered in the 10 km resolution. Some pixels have 5 stations and many others pixel without any information. That is why we interpolated data. Other approaches may be used such as comparing WRF grid nodes with the nearest observed locations.
l-112 are these the variables you use to build initial and boundary conditions. This has no sense.
122 (ERA) is dynamically downscaled using WRF to obtain downscaled reanalysis at 10 km resolution.
Response: paragraph 108-122 is aimed to briefly describe the ERA-Interim global atmospheric reanalysis variables and WRF model. We will rewrite it in clearer way.
l- 119. (Figure 4) What domain is that? How many domains do you use? Which resolucion?
Response: We will change the Figure 4 with a figure showing the two used domains with their coordinates. The resolutions of the two domains are 30 km and 10 km.
** Here, a complete model configuration should be exposed (radiation, LSM, etc), as well as spatial configuration (vertical levels, soil ... etc) In addition how simulations were done (simulated time, spin-up period,
Response: Yes thank you. We will add all the setup details.
l-125 The description of Cumulus schemes is not usefull at all. If authors try to explain something they have to clasify the schemes used, for example if they are or not flux mass schmes, trigger mecanish . etc.
Response: the classification is presented in Table 1. The text will be reformulated in order to be more informative and in relation with the rainfall estimation. You are right. Thank you.
-134 The same as before
Response: In the updated manuscript, we will highlight the difference of the various schemes in Table1 and how it will influence the rainfall forecast.
- About the metrics.. It is really necessary to use all these metrics? Each metric focus on different aspects of the skill. What is the sense of use a metric that is sum of all?
Response: Each metric gives an evaluation aspect and has drawbacks and good points. Before doing the sum we rescale the metrics. This is a way to weight to metrics in order to consider one single evaluation score.
Yes the reviewer has right. One can consider one single criterion in a time, rank the schemes and identify the best in the light of each criterion separately. This will be done.
l-244 You inverse the metric (X) ... what do you mean?? 1/X???
Response: Yes. in order to allow it the give the best result correspond to the low values of the metric.
**Some more comments.
-Figure captions should be improved. All information needed to interpret the figures must be in the caption.
Response: We will do it.
- You dont argue why the selection of these 4 cases. I do not understand why undetected evens by satellite are candidates for the case selection.
Response: This research is a continuity of previous evaluation of extreme events over Tunisia using satellite data (Dhib et al. 2017). These 11 cases remained without response when using satellite information. We looked for other means to predict them. These 4 cases are well selected to represent the 11 cases. We took the day displaying the greatest average rainfall, greatest spatial variability represented by spatial standard deviation, the day which is ranked second for the average, and an average day for both spatial average and spatial variability. They are shown in Figure 2a.
- FIgures 8, 9, 10, 11. why they have different styles??
Response: This will be homogenized. Thank you.
We would again like to thank the reviewers for their time and helpful comments.
Citation: https://doi.org/10.5194/nhess-2020-376-AC2
-
AC2: 'Reply on RC2', Saoussen Dhib, 31 Mar 2021
Status: closed
-
RC1: 'Comment on nhess-2020-376', Anonymous Referee #1, 25 Jan 2021
Title: Sensitivity of the Weather Research and Forecasting model (WRF) to downscaling extreme events over Northern Tunisia
Authors: Saoussen Dhib, Víctor Homar, Zoubeida Bargaoui, Mariadelmar VichThe aim of this study is to examine the sensitivity of WRF rainfall estimates for different Planetary Boundary Layer (PBL) and Cumulus Physics (Cu) schemes. Sensitivity studies have shown that there is no best common combination scheme (PBL and Cu) for all events. The average of the 10 best combinations for each event is used to map the ensemble. The authors conclude that some schemes are more sensitive and others less sensitive.
The abstract lacks important information. A few sentences about the setup, time simulation, and significant results are desirable. Provide the readers with this information here so they can decide if the paper is useful to their needs and so they can determine where to find the information they need within the main text.
The introduction also lacks the study purpose. In addition, the scientific contributions are not clear in the current form of the manuscript and should be further elaborated. What the gap will be filled?
From the description of the methods given, it is not clear what advancements have been made over other designs already described in the literature. Has there been any previous research using this method? What are the superior methods used in this study?
The results from the study are scientifically interesting and may represent a good examination to get the benefits of processing the entire simulations processes. However, there are several important concerns with the manuscript, such as the presence of a large number of vague descriptions, questionable arguments, and the lack of in-depth discussions. To resolve the fundamental flaws, an essential re-work on discussions is believed to be required, and the resultant manuscript would be so different from this one that it would be considered a new piece of research work.
The conclusion does not have significant finding according to the important results. Which scheme should be recommendation by authors?Line 32: Please check the correct format of abbreviation! It should be …blab la bla (MSG MPE) …
Line 68: … Radar Topography Mission (SRTM)… Is this correct abbreviation?
Line 73: What is mean ‘Heavy event’? Please elaborated it!
Line 78: Why the authors select 2 stations? Where is this location? Please explain it more clearly!
Figure 4 is not clear information! The authors employ both outer and inner domains, where is the boundaries? Give the clear information in your figure!
Line 159: It is difficult to understand this sentence.
Line 162 – 164: Please give the clear explanation for quantile quantile comparison?
Line 173: Why the authors are choosing a threshold of 0.1 mm? What is the reason?
Line 182: The blank space should be removed!Citation: https://doi.org/10.5194/nhess-2020-376-RC1 -
AC1: 'Reply on RC1', Saoussen Dhib, 31 Mar 2021
Response to reviewers’ comments
Title: Sensitivity of the Weather Research and Forecasting model (WRF) to downscaling extreme events over Northern Tunisia Authors: Saoussen Dhib, Víctor Homar, Zoubeida Bargaoui, Mariadelmar Vich
We thank the reviewers for their time and helpful comments. We have addressed each point below. Reviewer comments are shown in bold, while author responses are shown in unformatted text.
The aim of this study is to examine the sensitivity of WRF rainfall estimates for different Planetary Boundary Layer (PBL) and Cumulus Physics (Cu) schemes. Sensitivity studies have shown that there is no best common combination scheme (PBL and Cu) for all events. The average of the 10 best combinations for each event is used to map the ensemble. The authors conclude that some schemes are more sensitive and others less sensitive.
Response: We thank the reviewer for this feedback. We have ensured that the revised manuscript addresses the main issues that were brought up in the review.
The abstract lacks important information. A few sentences about the setup, time simulation, and significant results are desirable. Provide the readers with this information here so they can decide if the paper is useful to their needs and so they can determine where to find the information they need within the main text.
Response: Thank you for critical feedback. We have ensured that the overall purpose of the study is more clearly defined in the revised abstract. These paragraphs will be added.
The period January 2007- August 2009 is investigated based on a daily rainfall network composed by 318 rain gauges covering 36000 km². We focus on heavy rainfall situations composed by days when a threshold of 50 mm/day is exceeded at least one recording location. Thus, a total of 77 heavy rainfall days are identified. Inverse distance is adopted to elaborate rain maps which are compared with maps obtained using WRF estimations (10 km resolution). The calibrated power coefficient is found 1.2 using cross-validation approach. MSGMPE failed to detect as rainy 11 days out of 77. So we propose to use WRF too to predict them. To run the comparison only 4 representative days out 11 are studied because the important time of simulations: the day with highest average spatial rainfall, the day with highest spatial standard deviation, one typical day with spatial average and standard deviations represented the most common situation and the day ranked 2 with respect to spatial average.
For each studied day, we begin the simulation at 12pm the day before that means 6 hours to attend some stability of the model before the studied day beginning. The end of the simulation is one day after. As a first test of the performance of the various parameters schemes, we did a quantile quantile comparison for the 12/01/2009 using the 3 parameters (PBL, Cu, Mp) schemes. The results show that there is not an important variability in the performance of the various Mp schemes. Then, we decided to continue the sensitivity study only with PBL and Cu parameters. Using WRF, simulation of 10 combination takes in average 4 days. It depends on the UIB department server availability. Also, usually we should run again about 20% of the simulations each time because of WRF crash or internet interruption. Then, for the 99 simulations of the different Cu and PBL for all the rainy days take about 190 days.
The introduction also lacks the study purpose. In addition, the scientific contributions are not clear in the current form of the manuscript and should be further elaborated. What the gap will be filled?
Response: WRF has many parametrizations the most common in the literature sensitive parametrizations for rainfall localization and intensity, are Cu, and much less PBL and Mp (Hewitson et al.2004; Tadross et al.2006). In this study, for each day, we will need to do 792 simulations (Cu with 11 schemes, PBL with 9 schemes and Mp with 8 schemes) to get a satisfactory configurations which can be find only with testing numerous physical parameterizations. 10 simulations takes in average 4 days which depends on the UIB department server availability. Also, usually we should run again about 20% of the simulations each time because of WRF crash or internet interruption. Then, for the 792 simulations we will need 380 days for each event. It is clear that the simulations duration is very long and it should be reduced with conservation of the performance of the WRF rainfall estimation. The aim of this study is to choose the best representing schemes of extreme rainfall by WRF over Northern Tunisia which will make the use of WRF more efficient for users in short time.
From the description of the methods given, it is not clear what advancements have been made over other designs already described in the literature. Has there been any previous research using this method? What are the superior methods used in this study?
Response: All previous mentioned research, examine few schemes for each parameter and aim to choose one best combination which is not representative for the different climate variability and the rainfall intensity. It is the first sensitivity study for rainfall estimation over Tunisia. The originality of this research appear in two components. Firstly, we will test all the schemes of each parameter. Secondly, we will choose not one best combination but the 10 best combinations which will be averaged later to give an ensemble map. This ensemble map will give the best estimation in comparison with all the other individual combinations. Secondly, based on the sensitivity study we selected 3 best schemes for each parameter (PBL, Cu) which have the ability to give a good results for the various extreme event types.
The results from the study are scientifically interesting and may represent a good examination to get the benefits of processing the entire simulations processes. However, there are several important concerns with the manuscript, such as the presence of a large number of vague descriptions, questionable arguments, and the lack of in-depth discussions. To resolve the fundamental flaws, an essential re-work on discussions is believed to be required, and the resultant manuscript would be so different from this one that it would be considered a new piece of research work.
Response
Thank you for the remarks about the discussion. We are working, now, to make it deeper and to highlight the influence of schemes with the location (topographic area, coastal area,…). Also, we are trying to highlight the influence of convective cumulus schemes in rainfall amounts within the different type of the studied rainy days.
The conclusion does not have significant finding according to the important results. Which scheme should be recommendation by authors?
Response: Thank you for the encouragement. Actually in the conclusion, we summarized the main useful tools of the sensitivity study which are: using different evaluation coefficients performance, the ensemble map of the best 10 combinations of each studied day, and the identification of three best schemes of each parameter (Cu and PBL) for the whole study area. However, we didn’t mentioned the real names of these schemes. In the revised manuscript, we will add the names of the best schemes. Furthermore, we will highlight the amount of wined time by using the 9 combinations of the best 3 selected schemes of each parameter (Cu and PBL) at the place of using an ensemble map of the combination of all the schemes (99 combinations).
Line 32: Please check the correct format of abbreviation! It should be …blab la bla (MSG MPE) …
Response: We corrected it. Meteosat Second Generation Multi-sensor Precipitation Estimate (MSG MPE) Line 68: … Radar Topography Mission (SRTM)… Is this correct abbreviation? Response: We corrected it. Shuttle Radar Topography Mission Line 73: What is mean ‘Heavy event’? Please elaborated it! Response: Heavy events are defined as those rainy days exceeding 50 mm/day for at least one station. Line 78: Why the authors select 2 stations? Where is this location? Please explain it more clearly!
Response: The heavy days evaluated by MSGMPE are selected based on a threshold of 50 mm/day for at least one station out of 318 stations. For this study we selected the rainy days with a threshold of at least 2 stations surpassing 50 mm/day which avoid very spatial localized extreme events. There is not a specified location of these stations. They can be in urban zone, forest, orographic area,... Figure 4 is not clear information! The authors employ both outer and inner domains, where is the boundaries? Give the clear information in your figure!
Response: We will change Figure 4 with another figure containing the two domains with all the coordinates information. Line 159: It is difficult to understand this sentence.
Response: For the PBL schemes simulation, the Cu scheme was fixed to 2 and Mp scheme to 6 (Fig.5a). Line 162 – 164: Please give the clear explanation for quantile quantile comparison?
Response:
In statistics, a quantile-quantile (Q-Q) plot is a probability plot which associates the frequency of non-overshoot. It is a robust estimator, which doesn’t compare pixel by pixel but compares the shapes of two probability distributions by plotting their quantiles against each other. If the two distributions being compared are similar, the points in the Q–Q plot will approximately lie on the line y = x. We compared the quantile distrubition of WRF estimation (spatial resolution of 10 km) against the quantile distribution of the correspondent ground map. Firstly The Q-Q graph shows if there is an overestimation or an underestimation by WRF. Figure 5 plots have WRF as abscissa and ground rainfall as ordinate. If WRF curve is over the X=Y line that means WRF underestimate rainfall and if WRF curve is below x=y line that means WRF overestimate rainfall. Secondly, Q-Q plot highlight the capacity of WRF estimation of the different type of rainfall: low (less than 20 mm/day/pixel), medium (between 20 and 60 mm/day/pixel) and maximum values (more than 60 mm/day/pixel).
Line 173: Why the authors are choosing a threshold of 0.1 mm? What is the reason?
Response: A threshold of 0.1 mm/pixel is used in SAL and FSS verification to distinguish between rainy and no rainy pixels. We chose 0.1 threshold to detect the rainy pixels. Line 182: The blank space should be removed!
Response: We corrected it. Thank you.
Thank you for the feedback. In the updated manuscript, we will ameliorate the redaction quality of the whole paper, add deeper discussion, compare our results with many previous finds as suggested.
We would again like to thank the reviewers for their time and helpful comments.
Citation: https://doi.org/10.5194/nhess-2020-376-AC1
-
AC1: 'Reply on RC1', Saoussen Dhib, 31 Mar 2021
-
RC2: 'Comment on nhess-2020-376', Anonymous Referee #2, 02 Feb 2021
This work presents a sensitivity analysis to the modification of the cumulus and boundary layer parameterizations in WRF in extreme precipitation events. The objective of this study seems to be focused on obtaining a better prediction of extreme events in the study area based on an optimal ensemble.
Although the underlying idea of the work is interesting, the structure of the work, the exposition of the methodology and the discussion of the results is poor. The objectives of the work are not clearly established, and the low quality writing makes very difficult to find a storyline. In general, the article is written in a fuzzy and lazy way. This makes very difficult to follow it and understand the message that the authors try to send. For all these reasons, I cannot recommend this paper for publication.
The authors should rewrite the article again, trying to give it a rational structure, a more complete and rational exposition of the configuration of the experiments carried out, and a more detailed and clarifying analysis of the results.
Below I list as an example some of the points on which I base my review
l-15. latitude should be location
l-22. "What schemes?"-After line 58. What are your objetives? What do you can provide as new knowledge?
l-63. In situ data....Observations??
l-68. spatial resolution?
l-72,94. Where do you interpolate precipitation data? Some information about the mesh must be provided.
l-94. Why interpolate data?? In fact, the results of the crossvalidation show not very good results.
l-112 are these the variables you use to build initial and boundary conditions. This has no sense.
l- 119. (Figure 4) What domain is that? How many domains do you use? Which resolucion?
** Here, a complete model configuration should be exposed (radiation, LSM, etc), as well as spatial configuration (vertical levels, soil ... etc) In addtion how simulations were done (simulated time, spin-up period,
l-125 The description of Cumulus schemes is not usefull at all. If authors try to explain something they have to clasify the schemes used, for example if they are or not flux mass schmes, trigger mecanish . etc.
-134 The same as before
- About the metrics.. It is really necessary to use all these metrics? Each metric focus on different aspects of the skill. What is the sense of use a metric that is sum of all?
l-244 You inverse the metric (X) ... what do you mean?? 1/X???
**Some more comments.
-Figure captions should be improved. All information needed to interpret the figures must be in the caption.
- You dont argue why the selection of these 4 cases. I do not understand why undetected evens by satellite are candidates for the case selection.
- FIgures 8, 9, 10, 11. why they have different styles??
Citation: https://doi.org/10.5194/nhess-2020-376-RC2 -
AC2: 'Reply on RC2', Saoussen Dhib, 31 Mar 2021
Response to reviewer comments
Title: Sensitivity of the Weather Research and Forecasting model (WRF) to downscaling extreme events over Northern Tunisia Authors: Saoussen Dhib, Víctor Homar, Zoubeida Bargaoui, Mariadelmar Vich
We thank the reviewers for their time and helpful comments. We have addressed each point below. Reviewer comments are shown in bold, while author responses are shown in unformatted text.
This work presents a sensitivity analysis to the modification of the cumulus and boundary layer parameterizations in WRF in extreme precipitation events. The objective of this study seems to be focused on obtaining a better prediction of extreme events in the study area based on an optimal ensemble.
Response:
Yes, this is the main objective. Thank you.
Although the underlying idea of the work is interesting, the structure of the work, the exposition of the methodology and the discussion of the results is poor.
Response: the structure of the work needs to be improved. Authors recognize this fact. In the next proposal, section 2 will be dedicated to present data and section 2 for the methodology. In section 2 the reference to our previous work using MSG-MPE will be added. Paragraph 81-91 will be reformulated. the subsection 2.2 will be removed (as title). The word "event" will be replaced by "day" in all the manuscript. Also Section 2.4 will be suppressed (as title).
-The methodology is based on the use of conventional performance criteria such as RMSE, Pearson correlation coefficient and ratio test in the comparison of the maps obtained by interpolating ground observations and those obtained using WRF predictions. In addition more sophisticated performance criteria such as SAL and FSS are used because they consider the internal spatial variability of the rainfall fields. Moreover, the methodology adopts the combination of PBL and Cu schemes assuming a given MP scheme. The criteria as rescaled and summed in order to base the comparison on a single composed criterion.
-the analysis is applied to the study of 4 days representing different conditions with respect to rainfall amounts and spatial variability. Results show that no single scheme can be recommended for the 4 studied days. The more difficult to be predicted are the day displaying the greatest rainfall amounts (12/1/2009) and that with the highest spatial variability (13/9/2008).
The objectives of the work are not clearly established, and the low quality writing makes very difficult to find a storyline.
Response: Thank you for critical feedback. We ensure that the overall purpose of the study will be more clearly defined.
WRF has many parametrizations the most common in the literature sensitive parametrizations for rainfall localization and intensity, are Cu, and much less PBL and Mp (Hewitson et al.2004; Tadross et al.2006). In this study, for each day, we will need to do 792 simulations (Cu with 11 schemes, PBL with 9 schemes and Mp with 8 schemes) to get a satisfactory configurations which can be find only with testing numerous physical parameterizations. 10 simulations takes in average 4 days which depends on the UIB department server availability. Also, usually we should run again about 20% of the simulations each time because of WRF crash or internet interruption. Then, for the 792 simulations we will need 380 days for each event. It is clear that the simulations duration is very long and it should be reduced with conservation of the performance of the WRF rainfall estimation. The aim of this study is to choose the best representing schemes of extreme rainfall by WRF over Northern Tunisia which will make the use of WRF more efficient for users in short time.
In general, the article is written in a fuzzy and lazy way. This makes very difficult to follow it and understand the message that the authors try to send. For all these reasons, I cannot recommend this paper for publication.
Response: We ensure that the revised manuscript will be edited by a native English speaker to improve comprehension and quality.
The authors should rewrite the article again, trying to give it a rational structure, a more complete and rational exposition of the configuration of the experiments carried out, and a more detailed and clarifying analysis of the results.
Response:
Thank you for the suggestion. In the updated manuscript, we will ameliorate the redaction quality of the whole paper, add deeper discussion, compare our results with many previous founds as suggested.
Below I list as an example some of the points on which I base my review l-15. latitude should be location
Response: Yes, thank you. l-22. "What schemes?"
Response: WRF is based on parameters representing the various physical processs. We consider particularly Cumulus (Cu), Planetary boundary level (PBL) and micro physics (MP).These parameters have many options which depend of the physical process complexity. These options are what we call "schemes". The schemes description is presented in Table 1. They are 9 different schemes for representing Cu and 11 for representing PBL.
-After line 58. What are your objectives? What do you can provide as new knowledge?
Response: the reviewer is right. The literature review shows that the performance of the parameterization of WRF is not known a priori. No best parameterization can be recommended for a given case study. It depends on the metrics used for evaluation, and on the case study itself (the geographical region and the type of rainfall event). All previous mentioned research examines few schemes from each parameter. It is the first sensitivity study for rainfall estimation over Tunisia. The originality of this research appears in two components. Firstly, we will test all the schemes of each parameter. Secondly, we will choose not one best combination but the 10 best combinations which will be averaged later to give an ensemble map. This ensemble map will give the best estimation in comparison with all the other individual combinations. Secondly, based on the sensitivity study we selected 3 best schemes for each parameter (PBL, Cu) which have the ability to give a good results for the various extreme event types.
l-63. In situ data....Observations??
Response: In situ data are observed rainfall amounts using the national rainfall network of Tunisia. Observations are daily rainfall.
l-68. spatial resolution?
Response: interpolation of in situ data is achieved using a 10 km spatial resolution
l-72,94. Where do you interpolate precipitation data? Some information about the mesh must be provided.
Response: the mesh is 10 km. It is represented in figure 1.
l-94. Why interpolate data?? In fact, the results of the crossvalidation show not very good results.
Response: The in-situ stations are not well scattered in the 10 km resolution. Some pixels have 5 stations and many others pixel without any information. That is why we interpolated data. Other approaches may be used such as comparing WRF grid nodes with the nearest observed locations.
l-112 are these the variables you use to build initial and boundary conditions. This has no sense.
122 (ERA) is dynamically downscaled using WRF to obtain downscaled reanalysis at 10 km resolution.
Response: paragraph 108-122 is aimed to briefly describe the ERA-Interim global atmospheric reanalysis variables and WRF model. We will rewrite it in clearer way.
l- 119. (Figure 4) What domain is that? How many domains do you use? Which resolucion?
Response: We will change the Figure 4 with a figure showing the two used domains with their coordinates. The resolutions of the two domains are 30 km and 10 km.
** Here, a complete model configuration should be exposed (radiation, LSM, etc), as well as spatial configuration (vertical levels, soil ... etc) In addition how simulations were done (simulated time, spin-up period,
Response: Yes thank you. We will add all the setup details.
l-125 The description of Cumulus schemes is not usefull at all. If authors try to explain something they have to clasify the schemes used, for example if they are or not flux mass schmes, trigger mecanish . etc.
Response: the classification is presented in Table 1. The text will be reformulated in order to be more informative and in relation with the rainfall estimation. You are right. Thank you.
-134 The same as before
Response: In the updated manuscript, we will highlight the difference of the various schemes in Table1 and how it will influence the rainfall forecast.
- About the metrics.. It is really necessary to use all these metrics? Each metric focus on different aspects of the skill. What is the sense of use a metric that is sum of all?
Response: Each metric gives an evaluation aspect and has drawbacks and good points. Before doing the sum we rescale the metrics. This is a way to weight to metrics in order to consider one single evaluation score.
Yes the reviewer has right. One can consider one single criterion in a time, rank the schemes and identify the best in the light of each criterion separately. This will be done.
l-244 You inverse the metric (X) ... what do you mean?? 1/X???
Response: Yes. in order to allow it the give the best result correspond to the low values of the metric.
**Some more comments.
-Figure captions should be improved. All information needed to interpret the figures must be in the caption.
Response: We will do it.
- You dont argue why the selection of these 4 cases. I do not understand why undetected evens by satellite are candidates for the case selection.
Response: This research is a continuity of previous evaluation of extreme events over Tunisia using satellite data (Dhib et al. 2017). These 11 cases remained without response when using satellite information. We looked for other means to predict them. These 4 cases are well selected to represent the 11 cases. We took the day displaying the greatest average rainfall, greatest spatial variability represented by spatial standard deviation, the day which is ranked second for the average, and an average day for both spatial average and spatial variability. They are shown in Figure 2a.
- FIgures 8, 9, 10, 11. why they have different styles??
Response: This will be homogenized. Thank you.
We would again like to thank the reviewers for their time and helpful comments.
Citation: https://doi.org/10.5194/nhess-2020-376-AC2
-
AC2: 'Reply on RC2', Saoussen Dhib, 31 Mar 2021
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
714 | 401 | 56 | 1,171 | 51 | 51 |
- HTML: 714
- PDF: 401
- XML: 56
- Total: 1,171
- BibTeX: 51
- EndNote: 51
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1