Automated Aerosol Classification from Spectral UV Measurements Using Machine Learning Clustering
Round 1
Reviewer 1 Report
despite the efforts of the aauthors the literarture is poor. It is under the authority of the Editor to accept the paper.
Author Response
We cannot find any more relevant papers to include. If you have any specific pieces of literature in mind please let us know.
Reviewer 2 Report
Thank you for including the time series! I recommend this manuscript for publication.
Author Response
Thank you for your comments.
Reviewer 3 Report
See attached file.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
I appreciate the authors' work to address my review comments.
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
Review
This manuscript presents a method for aerosol classification with a double monochromator Brewer spectrometer. The method is based on a comparison of Brewer measurements with collocated CIMEL measurements. From the CIMEL, fine mode fraction (FMF, 550nm) and single scattering albedo (SSA, 440nm) are used to classify hourly measurements into six categories (sea salt, dust, BC low, BC med, BC high, fine non-absorbing) with similar threshold approach as in Lee et al. (2010). Next, the distributions of Brewer extinction Angstrom exponent (EAE, 320-360nm) and SSA (340nm) are analysed for the CIMEL categories. These distributions in the EAE-SSA space are then used as reference for Brewer classification.
Not surprisingly, the Brewer-based method classifies some cases into neighbouring categories. However, the Brewer classification may still be useful, as it enables creating long (20 years) time series of aerosol types at the study location. However, this time series is not presented. So it remains unclear if the Brewer classification can reproduce same temporal trends as the CIMEL, or if the uncertainty in the classification remains too large. Including the time series would increase the value of the manuscript substantially. Therefore, I suggest a major revision before this study can be published in Remote Sensing.
Major comment
As stated above, my main concern is that the Brewer time series may be too noisy for trend analysis. Therefore, I suggest to present the time series of Brewer and CIMEL classification side-by-side. Do the two instruments find similar trends?
Minor comments
Please check that all abbreviations are defined at first use, e.g. AAOD on line 52 and SZA on line 141.
L88-91 Please give the fraction of data available with these criteria.
Fig. 1 caption: “The uncertainties for both products are included” The error bars are so small that they are hardly visible. Please consider increasing line width, at least for some representative example points.
Fig. 2 please indicate the region for sea salt aerosol.
L414 Please update the reference to the final version of the paper.
Reference
Lee, J.; Kim, J.; Song, C.; Kim, S.; Chun, Y.; Sohn, B.; Holben, B. Characteristics of aerosol types from AERONET sunphotometer measurements. Atmospheric Environment 2010, 44, 3110 – 3117.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The paper "Automated aerosol classification from spectral UV
measurements using machine learning techniques" is really valuable for this domain, extending the automated aerosol classification to other instruments and data.
However a more deep study it is necessary to validate the method. The ~1000 CIMEL data are not sufficient to do the training proper. I recommend the authors to include the confidence of the typing process and the confidence of classification process of the sun photometer data, which represent the base of the classification algorithm.
Also it is mandatory to show how the algorithm have been tested/compared with manual classification and which are the results
Comments for author File: Comments.docx
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors present an aerosol classification technique based on machine learning analyses of spectrophotometric observations and measurements. The paper is an interesting and valuable contribution to the field of research with a method that can be used in many projects. As such, I recommend publication after some relatively minor revisions are made.
These revisions should not take too much time.
General comments:
Consider explaining the effects of clouds in the introduction, particularly of cloud screening algorithms (the latter mentioned in line 91).
Also consider using providing examples of in-situ and remote sensing measurements mentioned in lines 26 and 28 respectively.
Suggestion: even though the acronyms are defined as part of the paper, I would suggest defining them on their first use in the text (except the Abstract). Then keep consistent with the use of acronyms.
Specific comments:
line 80: capitalise 'l' in CIMEL
line 131: consider briefly explaining the quality control and assurance procedure.
line 338: consider rewording the sentence, particularly the part ending in "...as they appear seldom or never/"
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The manuscript in general is well written and the results are clearly presented.
The main problem of this manuscript is the extremely poor literature. The references are really outdated even though during the last years many aerosol studies for the region were published.
The authors have to enrich the bibliography. Also, they must compare their results with others similar recent studies for the region.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
I’d like to thank the authors for their reply.
To some degree, I understand the plan to prepare a separate publication focusing on climatological analysis. Yet, without that analysis this paper is of rather limited interest in my opinion. However, if the editor considers this method interesting enough without the time series analysis, it is acceptable for me.
Reviewer 2 Report
The manuscript has been improved from the last version, the reference clusters have been extended and the algorithm compared against manual classification.
Reviewer 4 Report
The authors avoid answering my comments. They did not refresh the bibliography and the conclusions. Please take into account my previous comments.