Mid-Infrared Laser Spectroscopy Detection and Quantification of Explosives in Soils Using Multivariate Analysis and Artificial Intelligence

Reviewer # 2:

 Action

1.    The reviewed paper presents results of the characterization of explosive gases. This is a often discussed technique, where other recent authors refer to approaches for rapid identification of such particular gases in view of routine requirements of security checks.

R1. The reviewer has mistaken the results presented. At no time the investigation focused on the detection of explosives in the gas phase.

2.    The paper describes useful results. Unfortunately, some recent papers of relevance are neglected, such as:

V.P. Sirkeli, et all, Proposal for a Monolithic  Broadband Terahertz Quantum Cascade Laser Array Tailored to Detection of Explosive  Materials, Sens. Lett. 16 (2018) 1–7. https://doi.org/10.1166/sl.2018.3919.   The manuscript should also refer to such authors, too
Otherwise, the text is acceptable for publication.

R2.

        Thanks. Recommendation accepted. The reference was added

1.    Further information on the collection optics would be useful. What was the power at the detector for the 3 lasers?

R1. The detector was a thermoelectrically cooled mercury–cadmium–telluride (MCT). This detector has detectivity  in the range used of 10¹⁰ D*cm Hz^1/2/W

2.    Do you get a difference in sensitivity with different laser scan speeds? So the dataset is only valid for identical measurement conditions.

 R2. Yes, the identical measurement conditions were used. The scan time was approximately 1.5 s for all the spectral ranges.

3.    How was the difference in power across the full spectral range treated in the data analysis, whereby the signal to noise ratios are quite different? An average noise (eqn 2) was used but could a weighting method give better results?

R3. Noise averaging was used for the entire spectral range. If the weighting method is used, it is most likely that of lower noise level could be obtained. However, the difference was considered non-significant.

4.    Could using a selection of narrower spectral windows give better results?

R4.   Although the spectral window in which the QCL works is already narrow, various spectral ranges were tested, and there were no significant differences in the statistical parameters observed. In some cases, these values got worse. The spectra were divided into 20 windows, and all possible combinations were tested in an optimization process.

5.    It would be worthwhile comparing the detection sensitivities attained with some of the other methods described in the introduction. Are the LODs achieved of practical value for the applications proposed in the Intro?

R5.   The limits of detection that we found using this proposed methodology were adequate for field applications. However, if we compare the proposed method with others, such as chromatography based methods, the later are far superior, i.e., they are more sensitive and have lower LOD by about two orders of magnitude. However, these conventional methods cannot be used in the field easily because require multiple steps for analysis and pre-treatment of the samples.

There are no comments on how the system could be improved

 Minor errors:

486 detection of explosives

491 involving should be involved

492 KBr matrix does not have

done

1.    In the introduction, the authors should choose if citing other works by names or by numbers. See Line 72 and 81.

R1. Changed; used numbers only.

2.    How the results of the paper are influenced by the training period? How the data have been divided in training and validation sets? Please explain why you choose 53% and 47%.

R2. Typically, 30% is used for the validation set. We decided to increase this value because there was sufficient data in the training set to justify the increase in the data allocated in the validation set.