Evaluation of the WRF Model to Simulate a High-Intensity Rainfall Event over Kampala, Uganda

Round 1

Reviewer 1 Report

I think that this manuscript could be accepted after some minor revisions, consisting in integrations in the introduction part.

In details, in the final part of the introduction, authors should focus on the important future developments of their analysis (extended to a consistent number of events) regarding definition of Design Hyetographs with assigned Joint Return Period (JRP), involving (as example) rain peak, total rain volume and areal extension. Moreover, in this context of future developments, it should be mention that different types of JRP could be considered (Shiau, 2003; Serinaldi, 2015; De Luca and Biondi, 2017): i) all the variables must exceed a certain magnitude to achieve critical conditions; or (ii) at least one variable must be greater than a threshold; or (iii) critical conditions are induced by all the events with a joint Cumulative Density Function (CDF) overcoming an assigned probability threshold.

References to be cited:

Return period of bivariate distributed extreme hydrological events. Stoch. Environ. Res. Risk Assess. 2003, 17, 42–57.

Dismissing return periods. Stoch. Environ. Res. Risk Assess. 2015, 29, 1179–1189.

Return Period for Design Hyetograph and Relationship with T-Year Design Flood Peak. Water 2017, 9, 673; https://doi.org/10.3390/w9090673  

Author Response

Reviewer #1

Request for minor revisions, consisting of integrations in the introduction part:

“In details, in the final part of the introduction, authors should focus on the important future developments of their analysis (extended to a consistent number of events) regarding definition of Design Hyetographs with assigned Joint Return Period (JRP), involving (as example) rain peak, total rain volume and areal extension. Moreover, in this context of future developments, it should be mention that different types of JRP could be considered (Shiau, 2003; Serinaldi, 2015; De Luca and Biondi, 2017): i) all the variables must exceed a certain magnitude to achieve critical conditions; or (ii) at least one variable must be greater than a threshold; or (iii) critical conditions are induced by all the events with a joint Cumulative Density Function (CDF) overcoming an assigned probability threshold.”

References to be cited:

• Shiau, 2003: Return period of bivariate distributed extreme hydrological events. Stoch. Environ. Res. Risk Assess. 2003, 17, 42–57.
• Serinaldi, 2015: Dismissing return periods. Stoch. Environ. Res. Risk Assess. 2015, 29, 1179–1189.
• De Luca and Biondi, 2017: Return Period for Design Hyetograph and Relationship with T-Year Design Flood Peak. Water 2017, 9, 673; https://doi.org/10.3390/w9090673

Response: In the introduction the suggestion to include the importance of this work for future developments has been incorporated in the text as well as in the main objective. In the conclusion the consideration for developing design storms is put.

Reviewer 2 Report

The authors of this manuscript present the comparison of WRF performances based on different parameterization combinations and intend to reveal the spatiotemporal features of urban high intensity rainfall events. The model setup was proper. While, beside the strength, there are something could be considered as well before resubmission.

1. The abstract seems focus on the urban regional weather simulation, while, the contents doesn’t show the consideration of urbanization.
2. The simulation results are compared with auto weather station data and CHIRPS satellite produces. The WRF simulation time step is 6 mins, AWS has 10 mins recodes. It would be nice to see the comparison between WRF results and AWS recode directly. General readers have no clear idea about the evaluation scores. The direct overlapping would make it more obvious. On the aspect of model calibration and validation, one event is too less to conclude the goodness of model setup. Authors could include few more events. my opinion is that different strong rainfall processes could use different parameter combinations.
3. The high intensity rainfall events do not always result urban floods. The rain duration, rain total amount, initial soil moisture, detention ponds, ditches, sewer system, terrain of city, buildings, land cover and many other factors will interact with rainfall. Current manuscript didn’t involve these factors, it would be nice to revise the text and focus on the points that authors want to express.
4. The innovation of paper is not clear. The parameterization comparisons of WRF are well studied worldwide. This manuscript did present its highlight points clearly. Please notice, the new location (study area) are not considered as innovation.

Author Response

Reviewer #2

1. The abstract seems focus on the urban regional weather simulation, while, the contents doesn’t show the consideration of urbanization.

Response: The abstract is adjusted accordingly

2. The simulation results are compared with auto weather station data and CHIRPS satellite produces. The WRF simulation time step is 6 mins, AWS has 10 mins recodes. It would be nice to see the comparison between WRF results and AWS recode directly. General readers have no clear idea about the evaluation scores. The direct overlapping would make it more obvious. On the aspect of model calibration and validation, one event is too less to conclude the goodness of model setup.

Response: For clarification, the WRF simulation time step is not 6 minutes. A numerical time step of 60 seconds with adaptive time step is used to ensure numerical stability. The output frequency of the model is 10 minutes and these are compared with the observations, which results is a fair model evaluation

The suggested high-temporal comparison is considered, but not included for the following reasons: 1) the used evaluation scores are a quantitative summary of the comparison, including spatial and temporal comparison, 2) a direct comparison would not do justice to the quality of WRF to simulate high intensity rainfall within the hydrological catchment considered for flood modelling because we are more interested in the model capacity to generate peak events for flood modelling than in the precise timing on a 10-min time scale.

3. Authors could include few more events. my opinion is that different strong rainfall processes could use different parameter combinations.

Response: We agree that one event is not enough to conclude on the goodness of the model setup for this region. Although we consider this work as very valuable first step towards the use of WRF in flood hazard modelling, especially in regions where the standard procedures of using IDF curves is not sufficiently accurate for decision making with respect to flood prevention planning and protection. We emphasized this consideration in the revised discussion and conclusion.

4. The high intensity rainfall events do not always result urban floods. The rain duration, rain total amount, initial soil moisture, detention ponds, ditches, sewer system, terrain of city, buildings, land cover and many other factors will interact with rainfall. Current manuscript didn’t involve these factors, it would be nice to revise the text and focus on the points that authors want to express.

Response: The actual flood hazard modelling is beyond the scope of this manuscript. Here we express the potential to use WRF model results as driver for flood hazard modelling.

5. The innovation of paper is not clear. The parameterization comparisons of WRF are well studied worldwide. This manuscript did present its highlight points clearly. Please notice, the new location (study area) are not considered as innovation.

Response: The objective has been  rephrased to emphasize the future developments needed to utilize WRF model results for flood hazard modelling in a region where high quality direct and remotely-sensed observations of precipitation are absent. The sentence now reads as “This study's main objective is to analyze the performance of parametrization combinations in WRF to simulate the 25 June 2012 HIRE in order to evaluate the applicability of WRF for urban flood modelling in Kampala.”

In this study we emphasized that it is an overall very thorough study in which AND many parameterizations are tested AND CP on / off is tested AND TOPSIS is performed AND all this against 2 types of observations and that in our opinion there are few studies that already exist. These aspects combine, and that it is therefore a valuable contribution to the research community.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This manuscript discusses the study in which the WRF model is run with a variety of combinations of parameterization schemes, in order to simulate one particular rain event in Kampala. This was a storm that led to flash flooding and the desire is to test the efficacy of WRF in correctly simulating location, intensity and timing of the event.

The manuscript is well written and - to me - clearly explained the study and its results.

I suspect the authors are non-"English as a first language", and as a result there are a number of words & combinations that could be improved (tenses, singular/plural etc.) I have attached a copy of the manuscript with things highlighted in yellow. Almost all of these can be fixed by the technical editors, unless they are directed to the authors. Examples: line 16, "that has triggered" tense wrong...should be "that triggered". line 23 "outperformed combinations" could be "most successful combination".

Could be published as is, but would read better if the highlighted items are fixed!

Comments for author File: Comments.pdf

Reviewer 2 Report

The authors use WRF to try to determine the best combination of microphysics, cumulus, and planetary boundary layer parameterization schemes for the prediction of an intense rainfall event, which occurred in Kampala, Uganda. Using a plethora of skill scores, they found the M2-GF-ACM2 combination performed the best overall with the highest SES and US values. However, due to the major issues listed below along with other issues, I recommend rejection for this paper, but I encourage the authors to resubmit the work once the issues are addressed.

Major issues:

1. While the authors provide an argument for including CP schemes at 1 km, the justification is poor. At convection-permitting scales (i.e., < ~4-km grid spacing), the use of CP schemes is highly discouraged and has been found to sometimes have a negative effect on model forecast skill (e.g., Jeworrek et al. 2019). The authors should rerun the experiments with the CP schemes turned off at the 3- and 1-km scales. Or at the very least, the authors should rerun the double-moment microphysics schemes experiments with the CP schemes turned off at those two smaller scales and compare to the experiments in the paper.

2. Even if you rerun all or some of the experiments with the CP schemes turned off, how does this study improve forecasts for the Kampala area? Experiments were run for only one case, and while you recognize this issue in the discussion section, the results presented cannot be generalized to other cases. Also, we already know NWP models, specifically WRF, can successfully predict intense rainfall events in non-dense remote sensing areas. Therefore, I’m having trouble finding what’s novel about this study. If the authors ran these experiments for a season’s worth of cases, you the results could be better generalized.

Minor issues:

1. Line 146: What was the shortest time step that was used by the adaptive method?

2. Lines 159-160: “In the WRF model, non-convective and convective rainfall at the surface is produced by MP and CP, respectively…” Microphysics parameterization schemes also produce convective rainfall, so this statement is false.

3. More details are needed about the WRF simulation. For example, did you run WRF for 3 days straight as a free forecast from the ECMWF ICs and BCs? Did you do any data assimilation?

4. How big are the four domains (in km or grid points)? Those sizes should be mentioned. Also, the 1-km domain appears to be rather small (~16 km x 16 km?), so storms/showers would only be in the domain for a short period of time unless storm motions are nearly 0 km/h.

5. While the skill scores are appropriate, plots of accumulated precipitation for subjective evaluations could be useful.

6. In Figure 6d, the x-axis labeling is wrong. I recommend switching the x-axis values to actual minutes from analysis time (not 10-min time steps) or time (UTC).

Jeworrek, J., G. West, and R. Stull, 2019: Evaluation of Cumulus and Microphysics Parameterizations in WRF across the Convective Gray Zone. Wea. Forecasting, 34, 1097–1115, https://doi.org/10.1175/WAF-D-18-0178.1.