WRF Model Sensitivity to Spatial Resolution in Singapore: Analysis for a Heavy Rain Event and General Suitability
Round 1
Reviewer 1 Report
The manuscript needs major revision. The attached file has comments.
Comments for author File: 
Comments.pdf
Author Response
Reviewer #1:
The study aims to explore the influence of spatial resolution on rainfall modeling over Singapore and test the hypothesis that higher resolution WRF grids will also achieve greater skill when forecasting rainfall. The experiment involves assessment of the performance of WRF model runs using 1km, 3km, 9 km and 12km grid simulations. The analysis is based on (a) three case studies (event forecast verification) (b) a 15-day testing set. The verification is based on standard categorical verification methods. The results of the study fail to bring out conclusive findings.
Over all, the attempt is interesting. However, the approach/methodology gives rise to serious questions. The paper needs major revision before it can be considered for publication.
Recommendation: Major revision
Major Comments:
- The write-up needs serious improvement in terms of expressing the aim, objectives, procedure, steps followed etc. The authors need to clearly mention where the observations are used. Figure 1 shows locations of the observations. It is not clear if the verification of gridded rainfall forecasts is carried out against the point observations. In that case, the authors should indicate how grid-station matching is done. The verification against Radar rainfall is not elaborately mentioned, thus it appears it is not used for scores. It appears, Radar rainfall data is used just for visual comparison in Figs 5,6,7 and 9. This is grossly underutilization of the available data.
Response: While the comparison between observed values and gridded forecast values was mentioned in the methods section with the statement on lines 163-165, this has been further emphasised in the results section with the new statement “Comparison between observed point values and gridded forecast values was performed by utilising the nearest gridded value (spatial distance) for each observed point.”
The reviewer is correct to point out that comparisons against radar were conducted graphically only in the original manuscript. We have now added a section to compare gridded radar rainfall estimates with the gridded forecasts.
- It is important to understand that prediction of high rainfall amounts is critical to assess high resolution modelling and forecasting. No point in suggestion “1km grid model shows highest skill in predicting 0.5mm/h rain”. Verification of gridded rainfall should be carried out against the Radar gridded observations since they are available since, the station point observations under sample the rainfall field.
Response: This has now been done.
- Verification of high resolution rainfall forecasts is a challenge and it is important to account for the double penalty. There are various new methods that are recommended for verification of high resolution forecasts. These methods bring out the benefits of high resolution forecasts and the same is reflected in the metrics as against the traditional verification scores. The authors need to use some of these methods to obtain more conclusive results. A detailed review of spatial methods is compiled in https://journals.ametsoc.org/collection/ICP.
Response: The authors welcome this suggestion from the reviewer and have now added the fractions skill score to the results section for the highlighted days to add more depth to the results and provide more context.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This is a modeling study with WRF on the influence of model spatial resolution on the simulated rain events in Singapore. The quality of the paper fits the magazine and I only have a few minor comments.
Page 1, second paragraph, Why introduce so many details of CNN? It is not directly related to this paper.
Page 5. The meaning of the 3 measures should be introduced, i.e., what they exactly measure? For example, it is not clear to me why it is “a+b+c” in the denominator of Eq.(3).
Fig.5-7 and Fig.9. To make the comparisons of the patterns easier, I suggest to enclose the region of d02 in all the panels and mark the location of Singapore.
Author Response
Reviewer #2:
This is a modeling study with WRF on the influence of model spatial resolution on the simulated rain events in Singapore. The quality of the paper fits the magazine and I only have a few minor comments.
- Page 1, second paragraph, Why introduce so many details of CNN? It is not directly related to this paper.
Response: Indeed it was too lengthy and has now been trimmed to match the point of the paragraph, which is simply to outline methods for forecasting at very short time horizons and juxtapose to the longer time horizons where NWP models are valuable.
- Page 5. The meaning of the 3 measures should be introduced, i.e., what they exactly measure? For example, it is not clear to me why it is “a+b+c” in the denominator of Eq.(3).
Response: We have now added extra information on how to interpret these values. The following has been added to the text: “The HR describes the proportion of hits () in relation to the total hits and misses (), the FAR describes the proportion of false-alarms () to the total hits and false-alarms () and the CSI described the proportion of hits () to the total hits, misses and false-alarms ().”
- 5-7 and Fig.9. To make the comparisons of the patterns easier, I suggest to enclose the region of d02 in all the panels and mark the location of Singapore.
Response: The authors agree and have limited the plotting area to the domain of d02 and have also included a dotted line around Singapore to assist with identification on the maps.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The revised manuscript and responses to reviewer comments are now satisfactory. Just a minor suggestion. The FSS plots in figures 9 to 12 and 14 can have common y axis of 0-1 and indicate FSSuseful with a dashed line. FSSuseful>= 0.5+(f/2) where f is wet area ratio (fractional rainfall coverage over domain). For small domain where f does not change much FSSuseful >=0.5 (Mittermaier and Roberts, 2010).
Faggian referred in line 198 is present in references. Other three are missing.
Authors are advised to check for all references, typos and spellings.
Author Response
Reviewer #1:
The revised manuscript and responses to reviewer comments are now satisfactory.
Just a minor suggestion. The FSS plots in figures 9 to 12 and 14 can have common y axis of 0-1 and indicate FSSuseful with a dashed line. FSSuseful>= 0.5+(f/2) where f is wet area ratio (fractional rainfall coverage over domain). For small domain where f does not change much FSSuseful >=0.5 (Mittermaier and Roberts, 2010).
Response: This has now been done.
Faggian referred in line 198 is present in references. Other three are missing.
Response: The authors are not sure which others the reviewer is referring to. Nevertheless we have added references to the original FSS paper by Roberts and Lean 2008, and now the Mittermaier and Roberts, 2010 paper to introduce the FSS_useful.
Authors are advised to check for all references, typos and spellings.
Response: Noted.
Author Response File: Author Response.docx
