A Machine-Learning-Assisted Classification Algorithm for the Detection of Archaeological Proxies (Cropmarks) Based on Reflectance Signatures

Round 1

Reviewer 1 Report

Summary

The manuscript: “A machine learning-assisted classification algorithm for the detection  of  archaeological  proxies  (cropmarks)  based  on  reflecatance signatures.” assesses the role of machine learning in the detection of cropmarks. The study introduces and analyzes classification indices and criteria of the spectral signatures taken over crops during phonological cycle in a controlled environment.

Introduction

Overall, the authors should do a grammar check to the introduction to improve flow and readability.

Line 38:

English is not correct “long practiced”, rephrase to “long been used”

Line 47-55:

Could benefit by having a short summary of the advantages and disadvantages of the remote sensors used for crop marks in terms of spatial, spectral and temporal resolutions. Additionally add a justification as to why the study uses hyperspectral data.

Line 61-69:

The gap in knowledge and research gap covered by the study in terms of previous research is not phrased clearly

Line 74:

Remove”, as part of the PhD thesis of the first author” a reference will suffice

Materials and methods

Methods are reproducible and the experimental design is appropriate.

Results

The contrasts between healthy and stressed crops are clearly presented. Machine learning methods: (logistic regression and decision tree classifiers) and noise simulations are done for Index 570. An index based on the physical properties of the leaf and canopy is introduced by analyzing the estimated parameters of the PROSAIL model. The statistical tests validate the results accordingly.

Line 270:

Weakly correlated not weekly correlated

Line 591:

It should be recommended that future studies focus on real world applications as the study was based on a controlled environment.

References

The study cites mostly current references within the field of cropmark detection.

Minor English edits required

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

A brief summary:

The paper focused on detecting stressed reflectance signatures in ground spectroradiometric data to indicate the presence of buried archaeological remains. Three detection methods and quantitative metrics are presented. The paper focuses more on analyzing the dataset obtained from previous studies and touches a little on the machine learning side. The research gap and research question lack clarity and need to be clarified, as this creates confusion between the stated title and the presented analysis.

General comments:

The related works are briefly mentioned in the introduction section; however, they lack a clear correlation with the research conducted. Some of these should be rewritten to better reflect their relevance to the current study. Furthermore, considering the depth of related literature, it might be beneficial to dedicate a separate section solely to reviewing related works.

For example, after reading the introduction section, it is not clear why the author decided to use machine learning compared to traditional approaches and, most importantly, why the decision tree classifier was selected among other various available supervised methods.

The Materials and Methods section is inadequately detailed and does not provide sufficient information to ensure the reproducibility of the study. Thorough explanations are important to ensure that the paper can stand alone. e.g., the dataset used is referenced appropriately; however, a more comprehensive explanation of how the spectral signatures were obtained would be helpful.

Several sections of the ‘Results’ could be integrated into the ‘Materials and Methods’ section. Authors are suggested to have subsections named "Data Preprocessing" within the Materials and Methods section.

Given the multitude of observations and analyses in the results section, it is suggested to have a workflow diagram in the materials and methods section, it would provide readers with a visual representation of the processes and analyses conducted.

Specific comments:

In lines 39-41, the authors mention several references that focus on the detection of archaeological proxies; it would be good to have a table showing the properties of datasets used in these studies, what methods they used, their detection success, etc., so that their relevance can be compared with the study. 

Line 49-50: “For instance, in [18-25] different remote sensors have been proposed to detect archaeological proxies, based on integrated signals from optical and radar sensors”. I don’t see any relevance of this sentence to the research work, as no analysis is linked with the sensor characteristics.

It would be good to provide a reference for the GER device (if available). In line 90, instead of saying "range around 1.5nm", it's better to use "FWHM.” Furthermore, it is advisable to incorporate a reference to the spectralon panel employed.

In line 105, "The above-mentioned ground spectroradiometric datasets were further elaborated in the Python programming language [30]", it is not clear what the author is trying to say. Please rephrase.

Line 145: it is suggested to have both A and H spectral curve to see the distinction between the curves (similar to Figure 2(a)). If both spectral signatures are very similar, one might expect to see a straight line passing through the origin when plotting the spectral curves against each other. Given the obviousness of this observation, it may not be necessary to provide detailed explanation.

Moreover, it would be beneficial to incorporate more quantitative analyses for the A and H curves. For instance, compute some distance function or spectral correlation coefficient etc.

It needs to be clarified how the analysis made in different sections (e.g., 3.1 and 3.2) is linked to the application of machine learning to this dataset.

Some recommended papers:

Aqdus, Syed Ali, Jane Drummond, and William S. Hanson. "Discovering archaeological cropmarks: a hyperspectral approach." Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci 37 (2008): 361-365

Masini, Nicola, et al. "On the characterization of temporal and spatial patterns of archaeological crop-marks." Journal of Cultural Heritage 32 (2018): 124-132.

Kotsiantis, Sotiris B. "Decision trees: a recent overview." Artificial Intelligence Review 39 (2013): 261-283.

Aqdus, Syed Ali, William S. Hanson, and Jane Drummond. "The potential of hyperspectral and multi-spectral imagery to enhance archaeological cropmark detection: A comparative study." Journal of Archaeological Science 39.7 (2012): 1915-1924.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Dear Authors,

Congrats on a very good manuscript on a very actual and not so advanced issue. The identification of archaeological remains with the help of AI. The method presented advances the knowledge in the field and is replicable. I have added a few comments in the attached .pdf file to include in your final manuscript. At this stage, I think that the manuscript could be accepted with some very minor reviews.

Kind regards!

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

--