Severe Precipitation Recognition Using Attention-UNet of Multichannel Doppler Radar

Round 1

Reviewer 1 Report

Please, see the attached file.

Comments for author File: Comments.pdf

Author Response

We really appreciate all your comments and suggestions.  Our responses are in the attached file.

Author Response File: Author Response.docx

Reviewer 2 Report

This work presents a fairly straightforward demonstration of a machine learning algorithm used to map radar reflectivities to surface precipitation estimates, having been trained on gauge data. The work is interesting and ML is certainly a hot topic at the moment. I recommend publication following some minor revisions.

General comments:

The paper could use a full grammatical review.

The paper makes frequent reference to Doppler radar but it's not clear to me that the radial velocities are used in any way. I would replace these instances with weather radar or S-band radar as appropriate.

Figure captions should be expanded to include more specific information and plots labeled with titles.

Specific comments:

Line 11-12: please add quantitative information to abstract demonstrating improvement (HSS improved by x, bias decreased by y, etc.)

Line 16-17: please reword these two sentences for grammar

Line 23: suggest "precipitation type" rather than "cloud type" here or replace with regime, atmospheric conditions, or similar

Line 26: "and research"

Line 29-32: please add additional references/examples here if possible

Line 31: I struggled with the use of "label" throughout the paper to refer to gauge data. Suggest "using the rain gauges as precipitation training data" or similar

Line 37: As mentioned above, are the authors using Doppler information? If not use "weather radar", "S-band radar", etc.

Line 47: reword "which is resemble" - not sure what is being said here

line 50: "will not get satisfactory results"

Line 75-76: Can the authors expand on this? What might be some effects of diverging from the natural frequency of precipitation occurrence?

Line 99: Here and elsewhere the authors refer to "multichannel Doppler radar" - is the Doppler information being used? Is more than one frequency or polarization being used? If not I suggest removing "multichannel Doppler"

Line 115-116: Please reword - not sure what this sentence is saying.

Line 121: Assuming this is all single frequency/pol?

Figure 3: please expand on what is included in "C" - is this dual pol?

Figure 6: please add more information to caption - "for x area over y period"

Line 178: "Figure 6 demonstrates"

Line 179: please add more details on "our dataset" here to remind the reader - area + time period

Line 203: In this section please provide a detailed description of the gauges (type) and radars (freq, pol, type, etc.) used 

Figure 7: Change "Label" to "Gauge" or similar. Please describe area in caption and refer to Table 1. Change "channels" to products or similar

Line 217 + *general comment" - the authors refer to radar "features" "channels and "products" at various points - channels is confusing as that typically refers to frequency and polarization channels. Please use the same terminology throughout - I suggest "products" 

Section 3: there needs to be a discussion somewhere in this section of the type of Z-R relationship used, and what assumptions are made therein about DSD, etc.

Line 224: please specify origin of 2D precipitation images (gauges?) here

Table 3: How is VIL calculated?

Table 4: "radar product"

Tables 5-6: I find the POD/FAR/TS organization difficult and suggest separate boxes for these or another way of reporting the statistics - I realuze this makes for a large table but it would be easier to interpret the results

Table 6: Rather than A,B,C please spell out each combination

Line 292: Reduction is extremely small and seems fairly insignificant here

Line 296: "performance of 4 models"

Figure 8: please add plot titles across the top of the figure (CR, gauge, RRED-Net). Please replace "Label" with "gauge data"

Line 311: please discuss the resolution and smoothing issues evident here.

Line 313: Assume "ground truth" is the gauge data - please use consistent terminology throughout the paper

Table 8: Please add over what time period, area, number of pixels, etc. to caption

Line 423-434: not sure how novel this approach is - ML is being applied universally at the moment, to everything. 

Please comment on how specific these findings are to the area - geophysical conditions, DSD type, etc.

Line 434 - please expand on auxiliary information - I agree that adding information on conditions could increase information content and improve the retrievals. What types of parameters do the authors suggest?

Author Response

We really appreciate all your comments and suggestions.  Our responses are in the attached file.

Author Response File: Author Response.docx