Combination of Hyperspectral and Machine Learning to Invert Soil Electrical Conductivity
, Keli Jia 2 and Xiaoning Zhao 1,*
Round 1
Reviewer 1 Report
Use of hyperspectral remote sensing (RS) for analysis of soil properties has been a history. This study focuses on use of hyperspectral with machine learning to estimate soil electrical conductivity (SEC) in comparing with measured in field. For this study related literature has been completely surveyed for grouping the techniques and algorithms to model the relationships for the optimal ones. The results were reasonably accepted and discussed to conclude the feasibility of the methods for further study and use for this purpose.
However, it is not clear that for SEC estimation compared to multispectral RS how much improvement hyperspectral RS can bring and at the same time how many more computing burdens it causes. Also, same with machine learning how much improvement it can bring compared with other statistical pattern recognition approaches and at the same time how many more computing burdens it causes.
Also, in the paper there are quite number of approaches introduced? Why are they? Be focusing! Don't list the not directly related approaches.
Some other issues but not limited to:
- What is "invert"? Is it "retrieve"?
- In abstract what are "The 5 ( ..." and "The 13 ( ..."?
- Fig 8 what is the x-axis?
- Fig 10 what is the x-axis and y-axis for a and b respectively?
- ...
Author Response
Use of hyperspectral remote sensing (RS) for analysis of soil properties has been a history. This study focuses on use of hyperspectral with machine learning to estimate soil electrical conductivity (SEC) in comparing with measured in field. For this study related literature has been completely surveyed for grouping the techniques and algorithms to model the relationships for the optimal ones. The results were reasonably accepted and discussed to conclude the feasibility of the methods for further study and use for this purpose.
Response: Thank you so much for your affirmation of our research.
Point 1: it is not clear that for SEC estimation compared to multispectral RS how much improvement hyperspectral RS can bring and at the same time how many more computing burdens it causes. Also, same with machine learning how much improvement it can bring compared with other statistical pattern recognition approaches and at the same time how many more computing burdens it causes.
Response 1: Hyperspectral data has high spectral resolution, which can fully explore the spectral information and accurately reflect the subtle characteristics of the ground object spectrum. It is one of the ideal means for soil salinization monitoring. Remote sensing image can obtain soil spectral characteristics quickly, macroscopically and timely, and is widely used in dynamic monitoring and evaluation of large area soil salinization, but the inversion accuracy needs to be improved. Machine learning inversion model has strong nonlinear fitting ability and excellent data mining ability, which will increase the use of spectral reflection information. When the number of variables involved is large and the correlation with soil salt content is low, the mining ability of machine learning algorithm can be fully demonstrated. Compared with other statistical pattern recognition approaches, the machine model will involve the problem of parameter adjustment, but the accuracy will be improved, and the research of LightGBM and ERT models on soil electrical conductivity is not enough. Therefore, we choose the machine model to carry out the inversion of soil conductivity.
Point 2: in the paper there are quite number of approaches introduced? Why are they? Be focusing! Don't list the not directly related approaches.
Response 2: Yes, you are right. In order to eliminate hyperspectral noise, the paper uses 13 traditional hyperspectral pre-processing methods to process the data, and then analyzes the data with the conductivity. According to the correlation analysis results, 5 transformation methods with better effects were selected to participate in the follow-up research. According to the suggestions, only 5 spectral processing methods selected are retained in this paper.
Point 3: What is "invert"? Is it "retrieve"?
Response 3: Compared with retrieve, inverse can better reflect the research work in soil conductivity and hyperspectrum, including prediction and mapping of salinity.
Point 4: In abstract what are "The 5 ( ..." and "The 13 ( ..."?
Response 4: Changes have been made in the abstract to make it clearer to the reader.
“The 5 spectral pre-processing methods of standard normal variate (SNV), standard normal variate and detrend (SNV-DT), inverse (1/OR) (OR is original spectrum), inverse-log (Log(1/OR) and fractional order derivative (FOD) (range 0-2, with intervals of 0.25) were performed.” “The 13 characteristic factors of Slope, NDI, SI-T, RI, Profile curvature, DOA, Plane curvature, SI (conventional), Elevation, Int2, Aspect, S1 and TWI had provided 90% of the cumulative importance for EC using GBM.”
Point 5: Fig 8 what is the x-axis?
Response 5: This has been added in Figure 8 to make it clearer to the reader
Figure 8. The training set fitting effect diagram and score of six machine learning methods (extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), random forest (RF), extremely randomized trees (ERT), classification and regression tree (CART), and ridge regression (RR)).
Point 6: Fig 10 what is the x-axis and y-axis for a and b respectively?
Response 6: This has been added in Figure 10 to make it clearer to the reader
Figure 10. Fitting effect graph of ERT for EC (a) and model using validation correlation diagram of ERT for EC (b) in July 2018.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Combination of hyperspectral and machine learning to invert soil electrical conductivity
The work is intersting, however there ia a major issue that has to be considered by the authors: as reported by the Encyclopedia of Soils, saline soils have many physical changes in them as compared with normal healthy soils. These soils have low biodiversity, have nutrient deficiencies and toxicities because of boron, carbonate, and aluminate ions present. These soils are observed to have very low organic matter and biological activity; increase in soil salinity results in progressively smaller and more stressed microbial community, which is also seen to be least efficient toward metabolism. Also a different behaviour on water is additionally recognizable.
The author should consider this: are they really studying direct correlations betweeen salt and reflectance? or maybe correlations are due to other secondary characteristics of the soil following salinity (as e.g. organic matter, water,...)?
This is an important issue that is influencing the overall validity of the work, so the authors should add a deidicated paragraph/section on this.
As recognizable in Figure 2 there is not a trend between spectra and salinity. This is du to the fact that other factors (beside salinity) are affecting soil. Please discuss which other characteristics you think are speifically affecting the curves.
There is a peak at 1850 nm, that seems to be an instrument discontinuity/artefact. rather than an actual soil reesponse. Please better discuss on that.
In the discussion schapter, there is a new literature review on models (lines 430-450 and Table 5). The literature review should be don at the beginning: so remove ths part (or move it to the introduction) and improve the discussion.
Other
AUthors mention that "The spectral resolution was set at 3.5 nm from 350 to 1000 nm, 10 nm from 1000 to 1500 nm, and 7 nm from 1500 to 2100 nm.": please explain why such spectral resolutions have been decided/set.
Fonts within some figures are unreadable or too small: please keep the size of the figure, but increase the font size (in particular figures 3, 6, 7, 8, 11)
Figure 1: the legend is unreadable: please make it moree understandable
Figure 2: please organize the legend in terms of growing salinity (what is more: very or strong salinity?)
Figure5: 2D correlations are difficult to compare becaue they are all using different scale. Please use always thee same colour scale (ee.g. -0.7+0.7) at least with maps with the same degre of correlation.
Authors report R squared value with six decimal digits: this is unreasonable! In scientific writing numbers are reported based on their significance! 3 decimal digits are in this case enough!
In case the paper will be accepted for revision, please address above comments and correct accordingly the paper,
- giving your pertinent comments in the “Response to reviewer” document
- reporting in the “Response to reviewer” document also the paragraph with amended text highlighted with yellow colour or the new amended figure.
Author Response
The work is interesting. However, there is a major issue that has to be considered by the authors: as reported by the Encyclopedia of Soils, saline soils have many physical changes in them as compared with normal healthy soils. These soils have low biodiversity, have nutrient deficiencies and toxicities because of boron, carbonate, and aluminate ions present. These soils are observed to have very low organic matter and biological activity; increase in soil salinity results in progressively smaller and more stressed microbial community, which is also seen to be least efficient toward metabolism. Also a different behavior on water is additionally recognizable.
Point 1: The author should consider this: are they really studying direct correlations between salt and reflectance? or maybe correlations are due to other secondary characteristics of the soil following salinity (as e.g. organic matter, water...)? This is an important issue that is influencing the overall validity of the work, so the authors should add a dedicated paragraph/section on this.
Response 1: The spectral characteristics of salinized soil are significantly affected by soil texture, organic matter content, soil moisture and soil salt content. At present, the inversion is based on some soil properties, such as soil moisture content inversion [1], soil organic matter inversion [2], soil pH inversion [3], etc. It is of great significance to comprehensively consider the inversion of salt, organic matter and water. The existing research has a certain guiding role for the future research, which is one aspect of the current research and can serve as a basis for the comprehensive consideration of these factors in the future.
In the inversion study of soil electrical conductivity, the correlations were mainly between the reflectance and the salinity degree. Although properties of saline soils are comparatively changing and so, may affect the reflectance but such changes or the intensity of changes and consequences for the reflectance are also because of or depending on the salinity levels. Such effects, although present, should not generally change the overall finding of this manuscript, which was providing a basis for simulation of soil salinization at larger scales. Nevertheless, the effects of the mentioned factors will be deeply and more systematically studied in our future research.
Point 2: As recognizable in Figure 2 there is not a trend between spectra and salinity. This is due to the fact that other factors (beside salinity) are affecting soil. Please discuss which other characteristics you think are specifically affecting the curves.
Response 2: The pattern of spectra curves of salinized soil between 400 and 1400 nm show regular changes with the increase of salinization, that is, the soil reflectance increases with the increase of salinization. However, this rule is not obvious after 1400 nm. This may be because there are a variety of salinization types in this study. This regularity can distinguish different salinization degree, based on this we can accurately distinguish different salinization soil through certain treatment.
Point 3: There is a peak at 1850 nm, that seems to be an instrument discontinuity/artefact. rather than an actual soil response. Please better discuss on that.
Response 3: Our spectrometer uses SR3500, the detection band of SR3500 spectrometer is 350-2500 nm, in which the resolution of 350 ~ 1000 nm is 3.5 nm, the resolution of 1000 ~ 1500 nm is 10 nm, and the resolution of 1500 ~ 2100 nm is 7 nm. The inconsistency of detection elements leads to a little deviation at the junction of spectra (It makes no difference to the overall result). The change rule of the overall soil spectral curve is that the spectral reflectance increases with the increase of the band, while 1400 and 1900 nm are the absorption zone of water vapor, and the spectral curve shows the absorption valley, so it will show the peak near 1850.
Point 4: In the discussion chapter, there is a new literature review on models (lines 430-450 and Table 5). The literature review should be done at the beginning: so remove this part (or move it to the introduction) and improve the discussion.
Response 4: Yes, you are right. The table has been deleted. This part of the content has been rewritten in discussion.” At present, scholars have done relevant studies on the inversion of soil EC, but the models have mixed results [69-73].”
Point 5: Authors mention that "The spectral resolution was set at 3.5 nm from 350 to 1000 nm, 10 nm from 1000 to 1500 nm, and 7 nm from 1500 to 2100 nm.": please explain why such spectral resolutions have been decided/set.
Response 5: These are the spectrometer's own parameter attributes, which can better explain the spectral reflection characteristic curve.
Point 6: Fonts within some figures are unreadable or too small: please keep the size of the figure, but increase the font size (in particular figures 3, 6, 7, 8, 11)
Response 6: Yes, you are right. We increased the font size to preserve the figure, as shown below:
Figure 3. Preprocessing of mean spectral curves including standard normal variate (SNV), standard normal variate and detrend (SNV-DT), inverse (1/OR) (OR is original spectrum), inverse-log (Log(1/OR) and fractional order derivative (FOD) (range 0–2, with intervals of 0.25) (green areas represent the standard deviations of the spectra) collected from soil measured on the ground.
Figure 6. Maximum absolute correlation coefficient (MACC) of Visible-near-infrared (Vis-NIR) (a) and two-band index (b) under different reflectance conversion modes.
Figure 7. Feature importance of spectral index and topographic factors ranking using gradient boosting machine (GBM). Dotted vertical line represent cumulative feature importance reached 0.9 (90%).
Figure 8. The training set fitting effect diagram and score of six machine learning methods (extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), random forest (RF), extremely randomized trees (ERT), classification and regression tree (CART), and ridge regression (RR)).
Figure 11. Spatial distribution of soil EC (a: Measured value, b: ERT of study area, c: Sample point in Yinchuan Plain, d: Measured value in Yinchuan Plain)
Point 7: Figure 1: the legend is unreadable: please make it more understandable
Response 7: Yes, you are right. The figure 1 legend has been modified readable.
Point 8: Figure 2: please organize the legend in terms of growing salinity (what is more: very or strong salinity?)
Response 8: Yes, you are right. Figure 2 legend has been reformatted, and changed Non saline, Very slightly saline, slightly saline, Moderately saline, Strongly saline to Non saline, slightly saline, Moderately saline, Strongly saline, Extremely saline.
Point 9: Figure5: 2D correlations are difficult to compare because they are all using different scale. Please use always the same color scale (e.g. -0.7+0.7) at least with maps with the same degree of correlation.
Response 9: The two-dimensional correlation diagram is mainly to show the correlation between different two-dimensional salinity indices and soil EC. In order to compare the maximum absolute correlation coefficient of different two-dimensional salt indices, the maximum and minimum values are mainly regarded. Some scholars [1-2] have done relevant studies and obtained good results as followed 2 figures.
Point 10: Authors report R squared value with six decimal digits: this is unreasonable! In scientific writing numbers are reported based on their significance! 3 decimal digits are in this case enough!
Response 10: Yes, you are right. The R2 decimal values in this article have been changed to 3 decimal digits (0.849,0.552,0.869,0.981,0.838,0.251).
References
- Wang, X.P., Zhang, F., Ding, J.L., Kung, H.T., Latif, A., Johnson, V.C. Estimation of soil salt content (SSC) in the Ebinur Lake Wetland National Nature Reserve (ELWNNR), Northwest China, based on a Bootstrap-BP neural network model and optimal spectral indices. Science of the Total Environment. 2017, 20, 25
- Hong, Y.S., Yu, L., Chen, Y.Y., Liu, Y.F., Liu, Y.L., Liu, Y., Cheng, H. Prediction of Soil Organic Matter by VIS–NIR Spectroscopy Using Normalized Soil Moisture Index as a Proxy of Soil Moisture. Remote Sensing. 2017.10. 28. 10.3390/rs10010028.
- Ainiwaer, M., Ding, J.L., Kasim, N., Wang J.Z., Wang, J.Z. Regional scale soil moisture content estimation based on multi-source remote sensing parameters, International Journal of Remote Sensing, 2020, 41, 9, 3346-3367.
- Wang, J.Z., Ding, J.L., Yu D.L., Ma, X.K., Zhang, Z.P., Ge, X.Y., Teng, D.X., Li, X.H., Liang, J., Lizaga, I., Chen, X.Y., Yuan, L., Guo, Y.H. Capability of Sentinel-2 MSI data for monitoring and mapping of soil salinity in dry and wet seasons in the Ebinur Lake region. Xinjiang, China. Geoderma. 2019, 353, 172–187.
- Hong, Y.S., Chen, S.C., Zhang, Y., Chen, Y.Y., Lei, Y., Liu, Y.F., Liu, Y.L., Cheng, H., Liu, Y. Rapid identification of soil organic matter level via visible and near-infrared spectroscopy: Effects of two-dimensional correlation coefficient and extreme learning machine. Science of the Total Environment. 2018, 644, 1232–1243.
Author Response File: 
Author Response.pdf
Round 2
Reviewer 1 Report
The authors acceptably address the concerns I have. So, this manuscript can be proceeded for publication.
Author Response
May 19th, 2022
Dear Reviewer,
Thank you so much for the review of our manuscript (remotesensing-1692806) entitled “Combination of hyperspectral and machine learning to invert soil electrical conductivity".
The detailed responses to comments are included in the following pages.
Yours sincerely
Xiaoning Zhao
The School of Geographical Sciences
Nanjing University of Information Science and Technology
Ningliu Road 219, Nanjing, China
Tel: +86 17351789670
Email: [email protected]
Response to Reviewer 1 Comments
The authors acceptably address the concerns I have. So, this manuscript can be proceeded for publication.
Response: Thank you so much for your affirmation of our research. We are grateful to you for reviewing our paper and your positive feedback.
Author Response File: Author Response.docx
Reviewer 2 Report
The paper has been somehow improved, however some of the comments have not been integrated into the paper.
Point 1: The author should consider this: are they really studying direct correlations between salt and reflectance? or maybe correlations are due to other secondary characteristics of the soil following salinity (as e.g. organic matter, water...)? ...
Response 1: The spectral characteristics of salinized soil are significantly affected by soil texture, organic matter content, soil moisture and soil salt content. At present, the inversion is based on some soil properties, such as soil moisture content inversion [1], soil organic matter inversion [2], soil pH inversion [3], etc. It is of great significance to comprehensively consider the inversion of salt, organic matter and water. The existing research has a certain guiding role for the future research, which is one aspect of the current research and can serve as a basis for the comprehensive consideration of these factors in the future.
In the inversion study of soil electrical conductivity, the correlations were mainly between the reflectance and the salinity degree. Although properties of saline soils are comparatively changing and so, may affect the reflectance but such changes or the intensity of changes and consequences for the reflectance are also because of or depending on the salinity levels. Such effects, although present, should not generally change the overall finding of this manuscript, which was providing a basis for simulation of soil salinization at larger scales. Nevertheless, the effects of the mentioned factors will be deeply and more systematically studied in our future research.
Referee: please add a few lines in the better discussing the above mentioned issue.
Point 2: As recognizable in Figure 2 there is not a trend between spectra and salinity. This is due to the fact that other factors (beside salinity) are affecting soil. Please discuss which other characteristics you think are specifically affecting the curves.
Response 2: The pattern of spectra curves of salinized soil between 400 and 1400 nm show regular changes with the increase of salinization, that is, the soil reflectance increases with the increase of salinization. However, this rule is not obvious after 1400 nm. This may be because there are a variety of salinization types in this study. This regularity can distinguish different salinization degree, based on this we can accurately distinguish different salinization soil through certain treatment.
Referee: please report this comment in the paper.
Point 6: Fonts within some figures are unreadable or too small: please keep the size of the figure, but increase the font size (in particular figures 3, 6, 7, 8, 11)
Response 6: Yes, you are right. We increased the font size to preserve the figure, as shown below:
Referee: Authors have NOT improved the figures (or just in a negligible way). Please keep the size of the graphs/figures but increase the size of the fonts within the figures.
Point 7: Figure 1: the legend is unreadable: please make it more understandable
Response 7: Yes, you are right. The figure 1 legend has been modified readable.
Referee: Authors have NOT improved the figure: please make the legend (within Figure 1b) more readable
Point 9: Figure5: 2D correlations are difficult to compare because they are all using different scale. Please use always the same color scale (e.g. -0.7+0.7) at least with maps with the same degree of correlation.
Response 9: The two-dimensional correlation diagram is mainly to show the correlation between different two-dimensional salinity indices and soil EC. In order to compare the maximum absolute correlation coefficient of different two-dimensional salt indices, the maximum and minimum values are mainly regarded. Some scholars [1-2] have done relevant studies and obtained good results as followed 2 figures.
Referee: I understand the usefulness of the figure. Just please use a constant to color scale, from -0.7 to 0.7.
Author Response
May 19th, 2022
Dear Reviewer,
Thank you so much for the review of our manuscript (remotesensing-1692806) entitled “Combination of hyperspectral and machine learning to invert soil electrical conductivity".
The detailed responses comments are included in the following pages.
Yours sincerely
Xiaoning Zhao
The School of Geographical Sciences
Nanjing University of Information Science and Technology
Ningliu Road 219, Nanjing, China
Tel: +86 17351789670
Email: [email protected]
Response to Reviewer 2 Comments
The paper has been somehow improved, however some of the comments have not been integrated into the paper.
Point 1: The author should consider this: are they really studying direct correlations between salt and reflectance? or maybe correlations are due to other secondary characteristics of the soil following salinity (as e.g. organic matter, water...)? ...
Response 1: The spectral characteristics of salinized soil are significantly affected by soil texture, organic matter content, soil moisture and soil salt content. At present, the inversion is based on some soil properties, such as soil moisture content inversion [1], soil organic matter inversion [2], soil pH inversion [3], etc. It is of great significance to comprehensively consider the inversion of salt, organic matter and water. The existing research has a certain guiding role for the future research, which is one aspect of the current research and can serve as a basis for the comprehensive consideration of these factors in the future.
In the inversion study of soil electrical conductivity, the correlations were mainly between the reflectance and the salinity degree. Although properties of saline soils are comparatively changing and so, may affect the reflectance but such changes or the intensity of changes and consequences for the reflectance are also because of or depending on the salinity levels. Such effects, although present, should not generally change the overall finding of this manuscript, which was providing a basis for simulation of soil salinization at larger scales. Nevertheless, the effects of the mentioned factors will be deeply and more systematically studied in our future research.
Referee: please add a few lines in the better discussing the above mentioned issue.
Response 1: We have added and yellowed them in the discussing part of paper.
Point 2: As recognizable in Figure 2 there is not a trend between spectra and salinity. This is due to the fact that other factors (beside salinity) are affecting soil. Please discuss which other characteristics you think are specifically affecting the curves.
Response 2: The pattern of spectra curves of salinized soil between 400 and 1400 nm show regular changes with the increase of salinization, that is, the soil reflectance increases with the increase of salinization. However, this rule is not obvious after 1400 nm. This may be because there are a variety of salinization types in this study. This regularity can distinguish different salinization degree, based on this we can accurately distinguish different salinization soil through certain treatment.
Referee: please report this comment in the paper.
Response 2: We have added and yellowed them in appropriate places in the paper.
Point 6: Fonts within some figures are unreadable or too small: please keep the size of the figure, but increase the font size (in particular figures 3, 6, 7, 8, 11)
Response 6: Yes, you are right. We increased the font size to preserve the figure, as shown below:
Referee: Authors have NOT improved the figures (or just in a negligible way). Please keep the size of the graphs/figures but increase the size of the fonts within the figures.
Response 6: We increased the font size to preserve the figure, as shown below:
Figure 3. Preprocessing of mean spectral curves including standard normal variate (SNV), standard normal variate and detrend (SNV-DT), inverse (1/OR) (OR is original spectrum), inverse-log (Log(1/OR) and fractional order derivative (FOD) (range 0–2, with intervals of 0.25) (green areas represent the standard deviations of the spectra) collected from soil measured on the ground.
Figure 6. Maximum absolute correlation coefficient (MACC) of Visible-near-infrared (Vis-NIR) (a) and two-band index (b) under different reflectance conversion modes.
Figure 7. Feature importance of spectral index and topographic factors ranking using gradient boosting machine (GBM). Dotted vertical line represent cumulative feature importance reached 0.9 (90%).
Figure 8. The training set fitting effect diagram and score of six machine learning methods (extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), random forest (RF), extremely randomized trees (ERT), classification and regression tree (CART), and ridge regression (RR)).
Figure 11. Spatial distribution of soil EC (a: Measured value, b: ERT of study area, c: Sample point in Yinchuan Plain, d: Measured value in Yinchuan Plain)
Point 7: Figure 1: the legend is unreadable: please make it more understandable
Response 7: Yes, you are right. The figure 1 legend has been modified readable.
Referee: Authors have NOT improved the figure: please make the legend (within Figure 1b) more readable
Response 7: The figure 1 legend has been modified readable (We made the legend font color bold and enlarged the icon).
Point 9: Figure5: 2D correlations are difficult to compare because they are all using different scale. Please use always the same color scale (e.g. -0.7+0.7) at least with maps with the same degree of correlation.
Response 9: The two-dimensional correlation diagram is mainly to show the correlation between different two-dimensional salinity indices and soil EC. In order to compare the maximum absolute correlation coefficient of different two-dimensional salt indices, the maximum and minimum values are mainly regarded. Some scholars [1-2] have done relevant studies and obtained good results as followed 2 figures.
Referee: I understand the usefulness of the figure. Just please use a constant to color scale, from -0.7 to 0.7.
Response 9: The two dimensional correlation coefficients (r) is different according to different transform reflectance. The highest r could show the best transform reflectance. If the r was set to the same threshold, the best wavelength could not be calculated. The best wavelength was used for the machine learning in the next step. We analyzed the best wavelength in table 4 based on the result of figure 5. Just like the published results [1-5], the maximum absolute value of the correlation coefficient was different.
- Wang, J.Z., Ding, J.L., Yu D.L., Ma, X.K., Zhang, Z.P., Ge, X.Y., Teng, D.X., Li, X.H., Liang, J., Lizaga, I., Chen, X.Y., Yuan, L., Guo, Y.H. Capability of Sentinel-2 MSI data for monitoring and mapping of soil salinity in dry and wet seasons in the Ebinur Lake region. Xinjiang, China. Geoderma. 2019, 353, 172–187.
- Hong, Y.S., Chen, S.C., Zhang, Y., Chen, Y.Y., Lei, Y., Liu, Y.F., Liu, Y.L., Cheng, H., Liu, Y. Rapid identification of soil organic matter level via visible and near-infrared spectroscopy: Effects of two-dimensional correlation coefficient and extreme learning machine. Science of the Total Environment. 2018, 644, 1232–1243.
- Zhang, Z.P., Ding, J.L., Wang, J.Z., Ge, X.Y., Prediction of soil organic matter in northwestern China using fractional-order derivative spectroscopy and modified normalized difference indices, Catena, 2020, 185, 104257,
- Zhu, C.M, Zhang, Z.P., Wang, H.W., Wang, J.Z., Yang, S.T., Assessing Soil Organic Matter Content in a Coal Mining Area through Spectral Variables of Different Numbers of Dimensions. Sensors,2020. 20(6), 1795.
- Hong, Y.S., Shen, R.L., Cheng, H., Chen, S.C., Chen, Y.Y., Guo, L., He, J.H., Liu, Y.L., Yu, L., Liu, Y., Cadmium concentration estimation in peri-urban agricultural soils: Using reflectance spectroscopy, soil auxiliary information, or a combination of both?, Geoderma, 2019, 354, 113875.
Author Response File: Author Response.pdf
