Multi-Sensor Approach for Tropical Soil Fertility Analysis: Comparison of Individual and Combined Performance of VNIR, XRF, and LIBS Spectroscopies

Round 1

Reviewer 1 Report

The paper is a very valuable, original paper presenting the latest research in development of proximal agricultural soils sensors. It was prepared meticulously. However, a few critical errors/shortcomings have not been avoided. In my opinion once these errors will been corrected, the paper should be published.

Errors and suggestions for change:

Line: 55

In my opinion the correct form is „base saturation”not „percent base saturation”, in the sense used in this sentence.

Line 89:

“…in tropical soils.”. The study was conducted on soils from the Brazilian area (probably). They were therefore only conducted for selected tropical, Brazilian soils.

Material and Methods:

Line: 96:

It is necessary to determine the location of the fields from which soil samples were taken. Although it is pointed out that “soil samples (…) representing common types of soil in Brazilian tropical areas”. We (readers) do not know if the samples were taken from South American, Africa or India? Ferralsols and Lixisols can also be found in other regions of the world. I only guess that it was a tropical zone. If the soils are from Brazilian territory, an approximation of the location of their collection is necessary too. I think an accuracy (+/-) 100 km will be sufficient.

Line: 96 & 97:

Instead of “These samples were from two agricultural fields with Lixisol and Ferralsols “would not a better sentence be " These samples were from two agricultural fields with soils characterized as Lixisol and Ferralsols”?

In relation to line 89, what was the key to selecting these tropical soils for study, Lixisol and Ferralsol? Are theyeconomically significant, in Brazil or globaly?

Line 111

“The data acquisition with XRF and VNIR sensors was performed on dry and sieved samples with 2 mm particle size.” Line 98-99 already describes that the samples were sieved < 2mm. Perhaps a description that it was air dry (or absolute dry, don't now) would be sufficient. ...Unless fractions > 2mm were again added to the scanned samples. This needs clarification and clearer description.

Line 126:

"The spectrometer was calibrated using four reference materials."

What was the reference materials? This is relevant to the measurement method used i.e. (NIR).

Line 121-194.

If specifying that it was a Veris manufactured spectrometer, and the Veris portfolio has several field spectrometers, it would be advisable to specify which one was used. For XRF technology, the exact models are given. If it is not possible to give the model name for VNIR, it will be necessary to add a detailed description of the spectrometer optics.

Question: Were the measurements has been made? Under laboratory conditions or in the open air? If in lab, what were the room conditions: temp, illumination, air humidity like in lab or more like in research hall? Applies to NIR, XRF and LIBS.

What software was used in the XRF, and LIBS spectra pre-procesing/spectra analysis? In the VNIR method software is described in detail (line 144), in others it is not. If software was common, same for to the methods (Unscrabler) the description should be elsewhere. Reader knows only that stage of model development and validation for the three methods was done with Unscrambler (line 217), the pre-processing stage needs to be clarified.

Line 188:

It is worth considering whether if in diagrams B and C accuracy (on the X scale) with two decimal places is necessary?

Line 190:

“VNIR spectra are presented with their standard deviation above and below the mean curve.”

Suggestion changes to: “VNIR spectra are presented with their standard deviation (dash line) above and below the mean curve.”

Linie 244

Calibration was performed on 68 samples and validation on 34 samples. To add value to the  paper it would be necessary to indicate why a small number of samples is acceptable and what are the limitations of results passage.

 Line 289:

Table 1. The colour scale in the table is misleading. Row one (VIR): why is 2.61 darker than 3.37? The scales are arranged in columns, but the reader has to guess. I propose to dispense with the colour scale or present as in other tables.

Line 434:

Is: "VNIR + LIBS and VNIR + XRF". Sugestion: "VNIR + LIBS > VNIR + XRF"

Line 454:

Probably incorrect citation. Literature position 61 does not refer to on-board agricultural machinery. This was a laboratory LIBS study.

Author Response

Dear Reviewer,

Thank you for carefully reading the paper and providing us insightful comments to improve the quality of the paper. The authors tried their best to improve to the original manuscript and fulfil the reviewer suggestions. Based on your comments, the manuscript was revised using the track changes function activated in Word.

Below you can find our responses to each of your comments.

1. Comments: “Line 55 - In my opinion the correct form is „base saturation ”not „percent base saturation”, in the sense used in this sentence.”

Answer: Thanks for this comment, we agree. In the entire text, it was removed the "percent" before base saturation (e.g., L23, L55, L488).  

2. Comments – “Line 89 - “…in tropical soils.”. The study was conducted on soils from the Brazilian area (probably). They were therefore only conducted for selected tropical, Brazilian soils. In relation to line 89, what was the key to selecting these tropical soils for study, Lixisol and Ferralsol? Are they economically significant, in Brazil or globaly?”.

Line 96 - It is necessary to determine the location of the fields from which soil samples were taken. Although it is pointed out that “soil samples (…) representing common types of soil in Brazilian tropical areas”. We (readers) do not know if the samples were taken from South American, Africa or India? Ferralsols and Lixisols can also be found in other regions of the world. I only guess that it was a tropical zone. If the soils are from Brazilian territory, an approximation of the location of their collection is necessary too. I think an accuracy (+/-) 100 km will be sufficient.”

Answer: Thank you for this comment. Lixisols and Ferralsols are representative and common types of soil in Brazilian agricultural fields. Both are characterized by poor natural fertility and low cation retention, which makes recurrent inputs of fertilizers a precondition for continuous cultivation. A proper soil fertility management in Brazil is particularly important because we are the fourth largest consumer of fertilizers in the world (FAO, 2017). Hence, they are interesting types of soils for this pioneering evaluation.

Furthermore, as reported by IUSS (2015), Ferralsols are present in the humid tropics on the continental shields of South America (especially Brazil) and Africa (especially Congo, Democratic Republic of the Congo, southern Central African Republic, Angola, Guinea and eastern Madagascar). In addition, Lixisols are found in tropical, subtropical and warm temperate regions, occurring in (sub-) Sahelian and East Africa, South and Central America, Indian subcontinent, and in Southeast Asia and Australia.

The central location of the fields where the samples were collected has been added to the manuscript, as can be seen in L102.

References:

FAO - Food and Agriculture Organization of the United Nations. World Fertilizer Trends and Outlook to 2020. Food and Agriculture Organization of the United Nations (FAO), Rome, Italy, 2017.

IUSS Working Group WRB World reference base for soil resources 2014, update 2015: International soil classification system for naming soils and creating legends for soil maps; Schad, P., van Huyssteen, C., Micheli, E., Eds.; FAO: Rome, Italy, 2015; ISBN 978-92-5-108369-7.

3. Comments – “Line: 96 & 97- Instead of “These samples were from two agricultural fields with Lixisol and Ferralsols “would not a better sentence be " These samples were from two agricultural fields with soils characterized as Lixisol and Ferralsols”?”.

Answer: Thanks for the comment, we have modified this sentence (L100).

4. Comments – “ Line 111 - The data acquisition with XRF and VNIR sensors was performed on dry and sieved samples with 2 mm particle size.” Line 98-99 already describes that the samples were sieved < 2mm. Perhaps a description that it was air dry (or absolute dry, don't now) would be sufficient. ...Unless fractions > 2mm were again added to the scanned samples. This needs clarification and clearer description. ”.

Answer: Thank you for this comment that allowed us to clarify our methods further. We added to L116 that the samples used for the XRF and VNIR scanning were the same dried and sieved samples described previously in 103. For convenience, we copied the sentence of L103 and L116 below:

In L103 - “The samples were collected from 0–20 cm depth, and stored after air-drying (for 48 h) and sieving (≤ 2 mm).”

In L116 - The data acquisition with XRF and VNIR sensors was performed on the dry and sieved samples.”

5. Comments – “ Line 126: "The spectrometer was calibrated using four reference materials."

What was the reference materials? This is relevant to the measurement method used i.e. (NIR). ”.

Answer: They are factory reference materials with known reflectance . The spectrometer system, during its initialization, uses these standards to check the stability of the readings. We described this procedure in the text because it is an extra precaution to ensure analytical precision, not common in other VNIR spectrometers. To make this information clearer in the text we have added the following sentence in L131.

“As a quality control, the instrument automatically checks its spectrum intensity using four factory reference materials during its initialization”

6. Comments – “ Line 121-194 -If specifying that it was a Veris manufactured spectrometer, and the Veris portfolio has several field spectrometers, it would be advisable to specify which one was used. For XRF technology, the exact models are given. If it is not possible to give the model name for VNIR, it will be necessary to add a detailed description of the spectrometer optics.

Answer: We have added the model of the Veris VNIR spectrometer (L125) and also more information about the equipment (L130-137).

7. Comments – “Question: Were the measurements has been made? Under laboratory conditions or in the open air? If in lab, what were the room conditions: temp, illumination, air humidity like in lab or more like in research hall? Applies to NIR, XRF and LIBS.”

Answer: The VNIR, XRF, and LIBS analyses were performed in a research laboratory, with a temperature of approximately 21 °C, ambient lighting, without air humidity control. For the LIBS measurements an argon flow rate of 5 L/min was used tangentially to the pellet surface, reducing the presence of atmospheric air in the sampling environment. This information is detailed in the experimental part (L173) of the article. For XRF, the measurements were made under atmospheric pressure (no vacuum was used) using a portable equipment. For VNIR, using the Veris equipament there is no need for a dark room, since the samples are placed in a sample holder that isolates the entrance of external light (information added in L134).

8. Comments – “What software was used in the XRF, and LIBS spectra pre-procesing/spectra analysis? In the VNIR method software is described in detail (line 144), in others it is not. If software was common, same for to the methods (Unscrabler) the description should be elsewhere. Reader knows only that stage of model development and validation for the three methods was done with Unscrambler (line 217), the pre-processing stage needs to be clarified.”.

Answer: Further details on the pre-processing of XRF and LIBS data were provided in L164 and L193. The added sentences were also copied below.

In L164: “XRF emission lines were obtained using the Artax® software (Bruker AXS, Madison, USA), whose pre-processing was conducted in Excel 2016 (Microsoft Corporation, Redmond, USA).”

In L193: “The LIBS spectra were pre-processed in MATLAB (MathWorks Inc., Natick, USA) using the script developed by Gomes et al. 2013.”

Reference:

Gomes, A.D.A.; Galvão, R.K.H.; De Araújo, M.C.U.; Véras, G.; Da Silva, E.C. The successive projections algorithm for interval selection in PLS. Microchem. J. 2013, 110, 202–208. 10.1016/j.microc.2013.03.015.

9. Comments – “ Line 188- It is worth considering whether if in diagrams B and C accuracy (on the X scale) with two decimal places is necessary? ”.

Answer: Thank you for your consideration. Due to the high spectral resolution of XRF and LIBS sensors, their emission lines are very close to each other and their information is commonly presented with two (or three) decimal digits. So, we would prefer to keep the plot with two decimal digits.

10. Comments – “Line 190- VNIR spectra are presented with their standard deviation above and below the mean curve.”

Suggestion changes to: “VNIR spectra are presented with their standard deviation (dash line) above and below the mean curve.”  ”.

Answer: Thank you for your comment. The suggested change was made (L200).

11. Comments – “ Line 244 - Calibration was performed on 68 samples and validation on 34 samples. To add value to the paper it would be necessary to indicate why a small number of samples is acceptable and what are the limitations of results passage. ”.

Answer: We appreciate this comment that alerted us to this important issue.

Although the number of samples could be greater, they were carefully selected to represent a wide range of variation in the studied soil fertility attributes. Once a considerable variation in fertility attributes was guaranteed, this limited number of samples was selected due to the time-consuming work of preparing pellets. However, to make clear the limitation of the number of samples and need for further research in larger databases, we have added a final paragraph to the discussion addressing these issues (in L455).

Starting in L455: “This study was conducted using 102 soil samples with broad variation of fertility attributes. They belong to Lixisols and Ferralsols, which are representative and common types of soil in Brazilian tropical agricultural lands. In addition, this pioneering evaluation provided useful information to help PSS users to understand the advantages and drawbacks of the combined use of VNIR, XRF, and LIBS sensors for soil fertility analysis. Nevertheless, future researches should consider larger datasets that accounts for other types of soils, soil mineralogy, textural classes, concentration range, and different agricultural practices.”

12. Comments – “ Line 289 - Table 1. The colour scale in the table is misleading. Row one (VIR): why is 2.61 darker than 3.37? The scales are arranged in columns, but the reader has to guess. I propose to dispense with the colour scale or present as in other tables. ”.

Answer: Thank you for your comment. We agree that the Table failed to inform the readers that the comparison of the values was made per column, i.e., comparing the different approaches for each attribute. As a solution we indicate the following information at each table footnote: “The values for the same soil attribute were compared and presented on green scale, highlighting the highest values within each soil attribute” (please see L304, L338, L539). We believe that the color scale helps the readers understand the performance of each approach, since this intuition would not be possible with a table of numbers only.

13. Comments – “ Line 434- Is: "VNIR + LIBS and VNIR + XRF". Sugestion: "VNIR + LIBS > VNIR + XRF" ”.

Answer: The suggested change was made (L447).

14. Comments – “ Probably incorrect citation. Literature position 61 does not refer to on-board agricultural machinery. This was a laboratory LIBS study. ”

Answer: Thank you for this comment. Indeed Knadel et al. did not address in situ applications of LIBS. Our citation refers to the fact that the authors mentioned that in situ applications with LIBS is challenging and need to be further investigated (see excerpt pasted below). This is exactly the point we are highlighting in our final discussion. To avoid confusion we have removed the example in L474.

Knadel et al. (2017) excerpt - “Moreover, since the presented results summarise work performed on air-dried soil samples, the applicability of the techniques for in situ field studies would be an interesting future study. The vis–NIRS technique has already been successfully used on Danish soils directly in the field. However, measurements on field-moist samples with the LIBS system are not possible at this stage of development. The water content in the samples would absorb all the laser energy, preventing plasma creation. It should be emphasised that the LIBS system employed in this study is still under development, thus, further improvements to the performance and feasibility for field application are expected.”

L474: “These extra sample preparation steps to ensure the collection of reliable data make it challenging to apply LIBS to run analyses directly in the field (Knadel et al., 2017).”

Reference:

Knadel, M.; Gislum, R.; Hermansen, C.; Peng, Y.; Moldrup, P.; de Jonge, L.W.; Greve, M.H. Comparing predictive ability of laser-induced breakdown spectroscopy to visible near-infrared spectroscopy for soil property determination. Biosyst. Eng. 2017, 156, 157–172. 10.1016/j.biosystemseng.2017.01.007

Best regards,

Authors

Reviewer 2 Report

This study compared three major spectroscopy-based sensing approaches for their ability to estimate soil fertility parameters. The manuscript is well-organized and written. Some more details on the soil samples (soil type, geographic area, land use) can be helpful.

Specific comments:

Line 19: ‘Rapid, cost-effective and environment friendly’

25: Replace ‘was’ with ‘were’, ‘has’ with ‘have’

30: What is ‘V’?

108: Replace ‘relationship’ with ‘relationships’, delete ‘the’

109: Replace ‘was’ with ‘were’

111: Air dry or oven dry? What temperature?

201: add ‘which’ after ‘samples’

203: Ratio of Performance to Deviation? Check name.

228: Is this K-fold or Leave-one-out?

238: ‘minimum squares’ or ‘least squares?

365: Delete ‘of’ after ‘via’

Author Response

Dear Reviewer,

We thank you for your suggestions. Based on your comments, the manuscript was revised using the track changes function activated in Word. Regarding soil sample information, the soil sample location and their land use (agriculture) has been added to the text. The soil types had already been informed.

Below you can follow our responses to each of your comments.

1. Comments: “30: What is ‘V’?”

Answer: V was the abbreviation used for base saturation, as indicated in L23, L55, and L486.

2. Comments: “111: Air dry or oven dry? What temperature?”

Answer: The samples used for the XRF and VNIR scanning were air-dried and sieved. We indicated this information in L102 and L115, as you can also see below:

In L102 - “The samples were collected from 0–20 cm depth, and stored after air-drying (for 48 h) and sieving (≤ 2 mm).”

In L115 - The data acquisition with XRF and VNIR sensors was performed on the dry and sieved samples.”

3. Comments: “203: Ratio of Performance to Deviation? Check name.”

Answer: Thanks for this comment. We agree that there are some variations in the RPD term designations, such as Residual Prediction Deviation, Ratio of Performance to Deviation, and Ratio of Standard Deviation to RMSECV, all referring to the same quality index. We decided to use Residual Prediction Deviation to agree with some important publications in the field of proximal soil sensing, such as Mouazen et al. (2010), Kuang et al. (2012), Kodaira and Shibusawa (2013), and Wang et al. (2015).

References:

Kodaira, M., & Shibusawa, S. (2013). Using a mobile real-time soil visible-near infrared sensor for high resolution soil property mapping. Geoderma, 199, 64-79.

Kuang, B., Mahmood, H. S., Quraishi, M. Z., Hoogmoed, W. B., Mouazen, A. M., & van Henten, E. J. (2012). Sensing soil properties in the laboratory, in situ, and on-line: a review. Advances in agronomy, 114, 155-223.

Mouazen, A. M., Kuang, B., De Baerdemaeker, J., & Ramon, H. (2010). Comparison among principal component, partial least squares and back propagation neural network analyses for accuracy of measurement of selected soil properties with visible and near infrared spectroscopy. Geoderma, 158(1-2), 23-31.

Wang, D., Chakraborty, S., Weindorf, D. C., Li, B., Sharma, A., Paul, S., & Ali, M. N. (2015). Synthesized use of VisNIR DRS and PXRF for soil characterization: Total carbon and total nitrogen. Geoderma, 243, 157-167.

4. Comments: “228: Is this K-fold or Leave-one-out?”

Answer: We used leave-one-out cross-validation to define the number of latent variables. We added this information to L239, as shown below:

“The number of latent variables for each PLS model was determined according to the leave-one-out cross-validation that resulted in the lowest RMSE.”

5. Comments: “238: ‘minimum squares’ or ‘least squares?”

Answer: Thanks for this comment. We used the least squares method. This was better expressed in L249.

6. Comments: “(i) Line 19: ‘Rapid, cost-effective and environment friendly’; (ii) L25: Replace ‘was’ with ‘were’, ‘has’ with ‘have’; (iii) L108: Replace ‘relationship’ with ‘relationships’, delete ‘the’; (iv) L109: Replace ‘was’ with ‘were’; (v) L201: add ‘which’ after ‘samples’; (vi) L365: Delete ‘of’ after ‘via’;

Answer: Thanks for these comments. All requested changes have been made.

Best regards,

Authors

Reviewer 3 Report

A particularly valuable manuscript. A necessary and modern research subject that answers the latest scientific questions. A perfectly developed methodology (although the number of samples for the implementation of this research goal could be many times greater - these are the results of pioneering research, however). Statistical methods allow definitive conclusions based on the obtained research results. I congratulate the researchers on the idea and its implementation due to the need for this type of research to revolutionize the concept of soil fertility assessment in the near future.

Author Response

Dear Reviewer,

We thank you for your comments and for your time required for the review. We made minor changes to the paper (using the track changes function) to meet the other reviewers' suggestions.

If necessary, we continue available to make further alterations.

Kind regards,

Authors

Round 2

Reviewer 1 Report

Authors responded to all comments and made necessary improvements. The paper is highly valuable.