AMT-2017-464

Re; Preliminary verification for application of a support vector machine based cloud detection method to GOSAT-2 CAI-2
Yu Oishi et al.

Dear Authors,

In your response to Reviewer 1, where you have modified section 3.1, this is rather difficult to read. The added text would be much easier for the reader to assimilate if it were presented as a table. Further the remainder of this section is not in an acceptable format/style, where nearly every sentence is its own paragraph.

You have added the following sentence to the abstract;
"Although the use of visual inspection of clouds as a standard for judging is not practical for screening a full satellite data set, it has the advantage of allowing for locally optimized thresholds, while CLAUDIA1+3 use common global thresholds."
The first half, up to the comma, is a reasonable response to reviewer 1. But I don't know what you mean after the comma. The whole paper appears to be a comparison of CLAUDIA1 with CLAUDIA3, using visual inspection of a subset pixels to decide on the relative amounts of overlook and overestimate. If it is about using the the the visual inspection to set local thresholds then the added sentence seems self-contradictory. I'm sure that you didn't mean this to be contradictory, but then why add the second half of this sentence?

Reviewer 2 made the comment that the language of the paper needs to be tightened up. I agree with this and the above comments are examples of why I think so. You have not actually addressed this reviewer comment.

Thus I regret to say that I cannot accept this paper in its present form for publication in AMT. The science in this paper is very good and I recommend that the paper be rewritten, with careful attention to style, and especially to the meaning of the words used. By and large the grammar in the paper is excellent, though there are one or two obviously missing words that a careful proof reading by a native English speaker would pick up.

Sincerely
Murray Hamilton
(Assoc. Ed.)

Dear Authors,

thankyou for your revised manuscript "Preliminary verification for application of a support vector machine based cloud detection method to GOSAT-2 CAI-2".

To indicate what I meant by missing words, I attach a copy of part of version 4 of the manuscript where the missing words that I spotted in a cursory reading are indicated with blue marks, along with one or two other grammatical errors. These are relatively minor, but ought to have been picked up by the company that you employ to do the proofreading. In any case, the copy editors of the publisher would also pick these up, so I am not really very concerned with them. (Please ignore the green marks, and the yellow marks pertain to what I've written below.)

However I should add that I would not expect a company that provides proof reading services to employ people who can pick up parts of a text that are technically correct English, but fail to convey the intended meaning. Guidance on writing style, combined with a detailed knowledge of the subject, to help eliminate ambiguity becomes very useful, but unfortunately is difficult to find.

In your latest response you state

"The human eyes can recognize the objects by optimized thresholds for every attention area. On the other hand, CLAUDIA1+3 use common global thresholds. Therefore, cloud discrimination accuracy of visual inspection is greater than that of CLAUDIA1+3. In other words, visual inspection results can be used as truth data for accuracy evaluation of CLAUDIA1+3."

The last sentence says it all! This statement actually says explicitly what you are trying to achieve in this work. However I cannot find a clear, explicit statement like this in the manuscript itself. This is why the first referee, formed the impression that you were advocating to inspect visually each pixel of every image. This is really the critical failing of the manuscript.

Be careful though, because I think that the first referee is correct is questioning the actual accuracy of visual inspection. Statements like "optimised thresholds" and "discrimination accuracy of visual inspection" imply something quantitative about the visual inspection process that is unwarranted, at least without backup from the psycho-physics literature on human vision. To simply say that you visually inspect a subset of pixels to see if they are in fact cloudy is reasonable, however. While people would rightly question your ability to quantitatively judge the amount of cloud by eye, it is perfectly valid and useful to say that, visually, there is cloud (or is not).

There are three places that it seems to me that you can make this aspect of the work clearer; first is in the abstract, second at the end of section 1, and third in section 2.4, where you should emphasise that you only inspect 1600 pixels of each image.

Your new table 4 certainly improves the readability, but I'm at a loss to understand why you have kept all that highlighted text in section 3.1, much of which just repeats the information that is in table 4. To be sure, some of the sentences there summarise parts of the table, but the numbers in the text are just repetitious and continue to make the section hard to read.
The second half of section 3.1 is greatly improved however.

sincerely,
Murray Hamilton
(Assoc. Ed.)