More than XRF Mapping: STEAM (Statistically Tailored Elemental Angle Mapper) a Pioneering Analysis Protocol for Pigment Studies^†

Round 1

Reviewer 1 Report

I read the paper titled "More than XRF mapping: STEAM (Statistically Tailored Elemental Angle Mapper) a pioneering analysis protocol for pigment studies"
I found the research topic really interesting, the paper is well written and the goal is focused and well-realized.
The paper not only provides a validation of the proposed approach but also drive researcher to new possibility about the application of the XRF mapping to the investigation of painting.

These motivations justify my recommendation

Author Response

Thank you for your comments.

Reviewer 2 Report

This paper present a new statistical method for mapping and interpreting XRF scanning data on works of art, using a modification of the SAM method to compare XRF measurements acquired on “unknown” areas paintings to either previously measured and interpreted areas of the same painting (a well characterized part of the gable of San Diego in this case), or to references sets consisting either of other art works, or in this case, deliberately prepared panels producing using well understood recipes and methods of canvas preparation, painting, and restoration. The method consists of using a PCA analysis on measurements and reference data to defined “end members,” characteristic photon peak signatures for particular compositions. The particularly novel feature consists in filtering out the signatures for particular elements or combinations of elements based on their contribution to spectra, which allows filtering out the impacts of highly abundant or ubiquitous elements (like Pb, or Ca/Sr used to prep canvases), thereby sequentially focusing on the impacts of particular lower abundance or less universally present elements or combinations of elements. These are then displayed as a type of heatmap showing the relative similarity (vector angle) between reference data and new data, which as I understand it, essentially produces a false color image of the scanned area but highlighting the contribution of only particular components of the XRF spectrum associated with particular elements or elemental combinations (or phrased another way, how similar each pixel is to a particular reference set). With some interpretation, this allows the analyst to tease out the contribution of particular recipe and pigment components, but also to interpret different sequential layers of application, for instance separating out painting or restoration from underlying canvas preparation, or even uncovering earlier painting that has been subsequently covered. Furthermore, the method allows identification of areas of a painting that are not similar to any reference sequences, which can then be further examined.

It is not totally clear to me how the procedure allows for comparisons between different instruments, given that the spectral heights are dependent on a number of instrument and analysis specific factors, including beam current, detector type, X-ray tube and target composition, geometry, and so forth. I assume that the STEAM procedure deals with these factors, but as written, and after reading the manuscript several times, I confess that I don’t fully understand how the procedure is instrument independent given that the angle between any two analysis vectors in equation (3) is dependent on vector length (i.e., peak heights).

While I appreciate that English is probably not the primary language of the authors there are places in which the writing is hard to follow to a level that makes comprehension difficult, so some additional editing would help clarify. It took me several reads to understand he pattern of producing the reference canvas, for instance. It eventually was clear that the canvas consists of a number of separate sequences of pigments, but at first, it seemed that the canvas contained a uniform layered set of all utilized pigments. The section “2.3.2 STEAM and GS”

This paper present a new statistical method for mapping and interpreting XRF scanning data on works of art, using a modification of the SAM method to compare XRF measurements acquired on “unknown” areas paintings to either previously measured and interpreted areas of the same painting (a well characterized part of the gable of San Diego in this case), or to references sets consisting either of other art works, or in this case, deliberately prepared panels producing using well understood recipes and methods of canvas preparation, painting, and restoration. The method consists of using a PCA analysis on measurements and reference data to defined “end members,” characteristic photon peak signatures for particular compositions. The particularly novel feature consists in filtering out the signatures for particular elements or combinations of elements based on their contribution to spectra, which allows filtering out the impacts of highly abundant or ubiquitous elements (like Pb, or Ca/Sr used to prep canvases), thereby sequentially focusing on the impacts of particular lower abundance or less universally present elements or combinations of elements. These are then displayed as a type of heatmap showing the relative similarity (vector angle) between reference data and new data, which as I understand it, essentially produces a false color image of the scanned area but highlighting the contribution of only particular components of the XRF spectrum associated with particular elements or elemental combinations (or phrased another way, how similar each pixel is to a particular reference set). With some interpretation, this allows the analyst to tease out the contribution of particular recipe and pigment components, but also to interpret different sequential layers of application, for instance separating out painting or restoration from underlying canvas preparation, or even uncovering earlier painting that has been subsequently covered. Furthermore, the method allows identification of areas of a painting that are not similar to any reference sequences, which can then be further examined.

It is not totally clear to me how the procedure allows for comparisons between different instruments, given that the spectral heights are dependent on a number of instrument and analysis specific factors, including beam current, detector type, X-ray tube and target composition, geometry, and so forth. I assume that the STEAM procedure deals with these factors, but as written, and after reading the manuscript several times, I confess that I don’t fully understand how the procedure is instrument independent given that the angle between any two analysis vectors in equation (3) is dependent on vector length (i.e., peak heights).

While I appreciate that English is probably not the primary language of the authors there are places in which the writing is hard to follow to a level that makes comprehension difficult, so some additional editing would help clarify. It took me several reads to understand he pattern of producing the reference canvas, for instance. It eventually was clear that the canvas consists of a number of separate sequences of pigments, but at first, it seemed that the canvas contained a uniform layered set of all utilized pigments. The section “2.3.2 STEAM and GS” is also very difficult to follow without reading it several times, particularly the introductory sentence, “The GS resumes the frequencies with which a significant signal has been detected at a specific channel…” Despite several reads, I cannot understand this sentence, and therefore, the rest of the section (a critical part of the method description) is not clear. What is meant by “resumes” here, and how is “significant signal” determined?

Given the amount of data involved, I assume that the authors have developed software or code to perform STEAM, but this is never mentioned nor is any code provided. Minimally, the authors should mention how or in what software/statistical package they have conducted their analysis and mapping, and better yet, should provide code so that other analysts can implement their method.

is also very difficult to follow without reading it several times, particularly the introductory sentence, “The GS resumes the frequencies with which a significant signal has been detected at a specific channel…” Despite several reads, I cannot understand this sentence, and therefore, the rest of the section (a critical part of the method description) is not clear. What is meant by “resumes” here, and how is “significant signal” determined?

Given the amount of data involved, I assume that the authors have developed software or code to perform STEAM, but this is never mentioned nor is any code provided. Minimally, the authors should mention how or in what software/statistical package they have conducted their analysis and mapping, and better yet, should provide code so that other analysts can implement their method.

Author Response

Dear colleague,

please find below our point-by-point answer to all your comments

Author Response File: Author Response.docx

Reviewer 3 Report

In the paper “ More than XRF mapping: STEAM (Statistically Tailored Elemental Angle Mapper) a pioneering analysis protocol for pigment studies” the authors apply an analysis protocol based on the Spectral Angle Mapper (SAM) algorithm to interpret MA-XRF and single spot spectra in a unified way.  The method is based on the Cosine Similarity approach and it is applied to experimental data from reference canvas and a painting by Giotto to demonstrate its validation. 

The paper is of high interest to the rapidly expanding research area of MA-XRF, and I recommend its publication. However, the manuscript does not always contain a consistent description of parameters, some repetitive statements appear, and stronger arguments for presenting the results in few cases are required. 

Comments for author File: Comments.pdf

Author Response

Dear colleague,

please find below our point-by-point answer to all your comments

Author Response File: Author Response.docx