The Role of Water Vapor Observations in Satellite Rainfall Detection Highlighted by a Deep Learning Approach

Round 1

Reviewer 1 Report

The authors proposed a DL model for satellite rainfall detection which based on WV and TIR channels of Meteosat Second Generation and temporal information. The developed DL model shows strong performance in rainfall binary classification. Basically the paper was well written, but I would like to ask some questions.

 

1.       The article only analyzes the accuracy of self-developed DL models. Please compare the proposed model with the existing ones.

 

2.       There are many other models which can handle satellite rainfall detection. Please provide why DL is used in the study. What are the pros and cons of DL.

 

3.       Please put the figure near the corresponding content.

 

4.       Please explain the meaning of the different variables expressed in the formula.

 

5.       The authors should change the way they express the serial number of the subgraph. For example, The serial number of the subgraph should be "(a)" instead of " a) ".

 

6.       Line 85-89: passive voice is recommended.

 

7.       Line 170: Why was 1mm/3h chosen as the threshold for rainfall and no rainfall? Could the authors please provide the basis for the choice or explain the reasons for the choice?

 

8.       line 271: The first line of any paragraph needs to be indented by two characters.

 

 

9.       Line 490: change "with the use of " into " by ".

Comments for author File: Comments.pdf

Author Response

Please find attached a response to the reviewer's comments point by point.

Author Response File: Author Response.docx

Reviewer 2 Report

Review

 

 

In the presented article, the actual problem of satellite rainfall detection based on WV and TIR channels of Meteosat Second Generation and temporal information is considered. The advantage of algorithms using MSG data is a high update rate compared to PMW data, which should be emphasized in the introduction. In general, the article has a finished look with good explanations and illustrations.

Disadvantages:

1.    Not enough explanation for Figure 3e. What is the frequency of precipitation events? Number of events per year? Then this should be noted in the figure. What does it mean in fig. 3e width of the figures shown.

 

2. Line 177. Formulas (1) and (2). The value of max(X) should be explained with an example.    (max(X)=24 h?)

 

3.    Line 198. Relation (3) does not follow from Planck's formula. It is necessary to clarify the notation for formula (3).

 

4.    Line 210. The algorithm for selecting significant events in part (1) misclassified probabilistic output values of the averaged models

should be explained in more detail.

 

5. Line 246 and formulas (5)-(10). Formally, it is necessary to explain the quantities TP, TN, FP, FN.

 

The article is recommended for publication with minor corrections.

Author Response

We thank the reviewer for the detail feedback. Please find attached our response to the reviewer's comments point by point.

Author Response File: Author Response.docx

Reviewer 3 Report

The manuscript deals with the role of water vapor in West African rainfall dynamics. It investigates whether water vapor observations together with temporal information can complement thermal infrared data for satellite rainfall retrieval in a Deep Learning framework.

The article is an original contribution and the topic is of interest for the readership of the Atmosphere journal. It addresses the aims for the special issue “Precipitation in Africa”.

English language is clear, and the presentation is good; anyway, I have detected some criticisms in the text that should be properly addressed.

The Authors can benefit from the comments below to improve their paper. These have to be accomplished before manuscript acceptance.

 

 

Title

Title is appropriate.

 

 

Abstract

The abstract is concise and reflects the content of the article. It includes the main findings of the study.

Some acronyms are included in the abstract even if they are defined in the following sections. I suggest avoiding acronyms before their definition in the text.

 

 

Keyworks: the provided keywords are effective.

                 

 

Introduction

This section is clear and fluent. Aims of the study are properly clarified and relevant references are included.

Lines 20-55: Concerning the problem of insufficient coverage of the rain gauge network and limited amount of data, Authors are recommended to include, among others, the following reference in order to enrich the introductory discussion:

-        Barbero G., Moisello U., Todeschini S. (2014). Evaluation of the areal reduction factor in an urban area through rainfall records of limited length: a case study. ASCE’s Journal of Hydrologic Engineering, 19(11): 05014016-1-10, DOI: 10.1061/(ASCE)HE.1943-5584.0001022.

This study addresses the estimation of the areal rainfall based on point rainfall depth quantiles and shows inconsistencies due to insufficient coverage of the rain gauge network and to the limited amount of data.

 

Materials and methods

This section is clear, adequately detailed and supported by relevant references. The provided figures and table are clear and necessary.

Lines 104-106: The inability of water vapor imagery to detect low-level clouds like stratocumulus or nimbostratus clouds in moist environments is a critical aspect for the study and deserves further discussion.

Lines 132-133: Authors are encouraged to add more information or a reference in order to justify the assumption that the study area of northern Ghana is representative of the wider Sudanian savanna of West Africa.

Line 133: In figure 2, please add [m] after elevation.

Line 164-165: Authors should briefly discuss on the choice of the pixel size. Some rainfall events are localised in space and time and the adopted scale is too wide for their detection.

Line 170-172: The adopted threshold of  1mm/3h to discriminate between rain and no-rain sequences is reasonable, nevertheless it should be discussed since in the technical literature 1mm/1d is usually adopted to discriminate between rain and no-rain day.

Line 188: In table 1, I suggest to indicate only one decimal for the ratio dry/rain.

Line 234: Figure 4 is clear but it should be improved since smaller fonts are difficult to read.

Lines 241-242: A reference is recommended for the adopted set of categorical metrics.

Line 250: please check that all the symbols of eqs. (5)-(10) are specified in the text.

Line 260: the rain category “heavy rain” is very broad.

 

Results

This section is clear and presented in a logical sequence. The provided figures and table are clear and necessary for the presentation of the results. It should be improved based on the following comments.

Line 274: It should be specified the meaning of red and green colors in the contingency tables of figure 5.

Lines 281-282: “The small performance increase between IMERG Early Run and IMERG Final Run does not seem to justify the great difference in latency time”. Authors should discuss on possible reasons for this results and for the great difference in latency time.

Line 306-308: I agree with the Authors that the addition of the time of the day is particularly positive during the early rainy season when rainfall is still occurring during late afternoon hours.

Line 335: In the caption of figure 8 it should be specified the meaning of acronyms adopted for images (a) and (b), i.e., DJF and JAS.

 

Discussion

This section is logical, and the analysis on misclassification and missed events is very interesting. The strong limitation due to scarce West African gauge coverage is properly outlined. The spatial and temporal resolution of the model is a critical challenge for the future and the of new MTG-I1 launched in December 2022 is properly mentioned.

 

 

Conclusions

Conclusion seems reasonable and are supported by the results.

 

 

Appendix

The provided information is clear.

 

References

Relevant references are included in the paper. One references is recommended on the problem of insufficient coverage of the rain gauge network and limited amount of data. Apart from this reference, based on my knowledge, no important reference is missing.

 

Author Response

Please find attached our response to the reviewer's comments point by point.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The manuscript has been significantly improved following the recommendations of the Reviewers; all my concerns have been addressed and convincingly justified.