Soil Salinity Estimation in Cotton Fields in Arid Regions Based on Multi-Granularity Spectral Segmentation (MGSS)

Round 1

Reviewer 1 Report

In this study, MGSS method was used to extract key information of soil salinity. The prediction accuracy was significantly improved compared to the traditional method. The research method is relatively novel.
The contents of the abstract section cannot help us to determine at what level the method proposed in this paper is. The results obtained by the traditional method need to be shown together.

Soil spectra are affected by many elements, and the method proposed in this study, MGSS and PLSR combined, can regress the soil salinity model close to 100%, can we understand that this method can completely replace the traditional method? If the method is applied to soil organic carbon or other soil properties is it still possible to get high accuracy?

The future outlook section needs to be refined as to what the results will be if the method is applied to satellite data.

The conclusion section is not well described, and the specific results obtained need to be highlighted in detail.

no

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments to the authors:

1.      In study site section: Geological formations, the elevation of the region and the names of the soils of the region should be given.

2.      Provide valid references for working methods and measurement methods

3.      In model evaluation criteria, it is better to use NRMSE because RMSE does not have good accuracy.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Soil salinization is an important issue in agricultural production in Xinjiang. The current work is a really intriguing scientific endeavor. This study focuses on using VNIR spectra to retrieve soil salinity in cotton fields. The authors introduced a multi-feature extraction technology named MGSS to extract the potential information from spectra and explore its performance in soil salinity retrieval. I simply have a few minor concerns and suggestions and I'm hoping for a response.

-1, First of all, it would be better to give the line numbers throughout the manuscript.

-2, The coordinate system is missing from Figure 1. Please add a coordinate system and update the map of China, including the South China Sea.

-3, The collection of samples is very important and the most crucial step for the success of the experiment. However, this article introduces the selection of five points for sampling on each piece of farmland. How are these five points arranged specifically? Please provide a detailed explanation.

-4, The collection of indoor spectral data was not found throughout the article, and there is no content that matches Figure 2. Please add a detailed process for collecting indoor spectral data.

-5, The number of modeling and validation sets is a key factor affecting model accuracy. This article introduces the division of modeling and validation sets, but it is still relatively vague. "Arrange soil samples based on EC content from minimum to maximum, select one sample for validation every three samples, and use the remaining samples for calibration". Is the selection of every three samples in order here? Or randomly select one of these three samples for validation?

-6, This article adds drones and satellites to the discussion, which is very meaningful. This article tests the effectiveness of the MGSS method in extracting feature spectral information through spectral bands. In the discussion, spectral indices can also be added because they are constructed based on their sensitive bands. Therefore, considering the combination of spectral indices with MGSS, will the model perform better?

-7, MGSS can highlight the spectral features of soil EC at different granularity levels, effectively excavate some hidden spectral information, and expand the spectral utilization range, thus improving model accuracy. These are the advantages of the MGSS method, but what are the limitations of the MGSS method? Please explain in the discussion section.

Soil salinization is an important issue in agricultural production in Xinjiang. The current work is a really intriguing scientific endeavor. This study focuses on using VNIR spectra to retrieve soil salinity in cotton fields. The authors introduced a multi-feature extraction technology named MGSS to extract the potential information from spectra and explore its performance in soil salinity retrieval. I simply have a few minor concerns and suggestions and I'm hoping for a response.

-1, First of all, it would be better to give the line numbers throughout the manuscript.

-2, The coordinate system is missing from Figure 1. Please add a coordinate system and update the map of China, including the South China Sea.

-3, The collection of samples is very important and the most crucial step for the success of the experiment. However, this article introduces the selection of five points for sampling on each piece of farmland. How are these five points arranged specifically? Please provide a detailed explanation.

-4, The collection of indoor spectral data was not found throughout the article, and there is no content that matches Figure 2. Please add a detailed process for collecting indoor spectral data.

-5, The number of modeling and validation sets is a key factor affecting model accuracy. This article introduces the division of modeling and validation sets, but it is still relatively vague. "Arrange soil samples based on EC content from minimum to maximum, select one sample for validation every three samples, and use the remaining samples for calibration". Is the selection of every three samples in order here? Or randomly select one of these three samples for validation?

-6, This article adds drones and satellites to the discussion, which is very meaningful. This article tests the effectiveness of the MGSS method in extracting feature spectral information through spectral bands. In the discussion, spectral indices can also be added because they are constructed based on their sensitive bands. Therefore, considering the combination of spectral indices with MGSS, will the model perform better?

-7, MGSS can highlight the spectral features of soil EC at different granularity levels, effectively excavate some hidden spectral information, and expand the spectral utilization range, thus improving model accuracy. These are the advantages of the MGSS method, but what are the limitations of the MGSS method? Please explain in the discussion section.

Author Response

Responses to Reviewer #3

Soil salinization is an important issue in agricultural production in Xinjiang. The current work is a really intriguing scientific endeavor. This study focuses on using VNIR spectra to retrieve soil salinity in cotton fields. The authors introduced a multi-feature extraction technology named MGSS to extract the potential information from spectra and explore its performance in soil salinity retrieval. I simply have a few minor concerns and suggestions and I'm hoping for a response.

-1, First of all, it would be better to give the line numbers throughout the manuscript.

Response: Many thanks for your suggestion. We have added line number in the manuscript.

-2, The coordinate system is missing from Figure 1. Please add a coordinate system and update the map of China, including the South China Sea.

Response: Thank you for your interest in this critical issue. We have added the coordinate system in Figure 1 and also updated the map of China(Please see lines 129-130 of the revised manuscript). Thanks.

Figure 1. Distribution of cotton fields for soil sampling in southern Xinjiang, China

-3, The collection of samples is very important and the most crucial step for the success of the experiment. However, this article introduces the selection of five points for sampling on each piece of farmland. How are these five points arranged specifically? Please provide a detailed explanation.

Response: Many thanks for your suggestion. Firstly, the two diagonal lines of a cotton field were drawn. Then, the intersection (center point) of the two diagonal lines was selected as a sampling point. After that, four points were selected as the other four sampling points at half the distance from the four corners of the cotton field to the center point (Figure 2) ( Please see lines 118-121and 131-132 of the revised manuscript).

Figure 2. Selection of sampling points

-4, The collection of indoor spectral data was not found throughout the article, and there is no content that matches Figure 2. Please add a detailed process for collecting indoor spectral data.

Response: Sorry for this mistake. We have added the content of the collection of indoor spectral data, and cited Figure 2.

“2.2.3 Spectral acquisition

Due to soil salinity and soil EC have a positive correlation, and soil EC has a more prominent spectral response than soil salinity, the spectral information of soil EC has been widely used to estimate soil salinity [1]. In this study, the ASD Field Spec Pro FR spectrometer (Boulder, CO, USA) was used to collect the spectral data of soil samples in the lab. The wavelength range was 350 ~ 2500 nm, and the spectral resolution at 350 ~ 1000 nm and 1000 ~ 2500 nm were 3 and 10 nm, respectively. The spectral sampling interval was 1 nm.

The spectral acquisition details were as follows: Firstly, soil sample was placed in an aluminum box (5 cm in radius and 1.5 cm in depth). Then, the optical fiber was connected to the handle. After that, the switch on the handle and the APP configured for the instrument were connected through Bluetooth, and the number of spectral curves was set to 5. After half an hour, the four walls of the luminous port at the end of the handle came into direct contact with the soil surface, forming a confined space where all the light hit the soil sample. The reflectance spectra of the soil samples were received by the probe (Figure 3). The device was calibrated every ten minutes during measurement to prevent sensor drift and the change of incidence angle. The spectral acquisition was performed three times for each soil sample, and the average value was used for analysis”, (Please see lines 139-156 of the revised manuscript).

Figure 3. Soil spectral data acquisition

-5, The number of modeling and validation sets is a key factor affecting model accuracy. This article introduces the division of modeling and validation sets, but it is still relatively vague. "Arrange soil samples based on EC content from minimum to maximum, select one sample for validation every three samples, and use the remaining samples for calibration". Is the selection of every three samples in order here? Or randomly select one of these three samples for validation?

Response: Thanks for your comment. We have revised the content related to the number of calibration and validation sets in the manuscript. In this study, the entire dataset (191) was divided into calibration set (115) and validation set (76) in a 3:2 ratio (Table 1) by the Kennard-Stone (K-S) method [2] to ensure generalization and robustness of the model [3] (Please see lines 211-213 of the revised manuscript).

-6, This article adds drones and satellites to the discussion, which is very meaningful. This article tests the effectiveness of the MGSS method in extracting feature spectral information through spectral bands. In the discussion, spectral indices can also be added because they are constructed based on their sensitive bands. Therefore, considering the combination of spectral indices with MGSS, will the model perform better?

Response: Thank you for your advice. Spectral indices constructed based on multiple spectral features have been widely used in the monitoring of plant growth and environment. Since MGSS has the potential to improve model accuracy by extracting weak spectral information and expanding the spectral utilization range, Pang et al. [4] have combined MGSS and spectral indices to improve the accuracy of estimation of grassland aboveground biomass (AGB) based on satellite remote sensing. According to the report of Pang et al., the R² and the prediction accuracy (EA) of the PLSR model constructed based on the vegetation index constructed with the spectral features extracted by MGSS increased by 0.2 and 22.26%, respectively, and the RMSE decreased by 14.08 g/m², compared with those of the model constructed based on the vegetation index constructed with the raw spectrum. It can be seen that the combination of MGSS and spectral index has the potential to improve the accuracy of estimation of vegetation growth attributes based on satellite remote sensing, but its performance in the estimation of soil properties need to be further studied (Please see lines 391-396 of the revised manuscript).

-7, MGSS can highlight the spectral features of soil EC at different granularity levels, effectively excavate some hidden spectral information, and expand the spectral utilization range, thus improving model accuracy. These are the advantages of the MGSS method, but what are the limitations of the MGSS method? Please explain in the discussion section.

Response: Many thanks for your suggestion. We have added the limitations of the MGSS method in the Discussion following your suggestion.

However, in the continuous segmentation of the raw spectrum by MGSS, while extracting effective spectral information and weak spectral information, it also produces some independent spectral information that is not related to soil EC. Besides, with the increase of granularity level, the number of spectra produced increases exponentially, which may cause data redundancy. Therefore, how to use MGSS to extract effective spectral information while eliminating irrelevant spectral information and reducing data redundancy will be one of our future research priorities (Please see lines 400-406 of the revised manuscript).

References

1. Ren, J. H.; Chen, Q.; Ma, D. L.; Xie, R. F.; Zhu, H. L.; Zang, S. Y. Study on a fast EC measurement method of soda sa-line-alkali soil based on wavelet decomposition texture feature. Catena. 2021, 203, 105272. https://doi.org/10.1016/j.catena.2021.105272.
2. Wang, J. Z.; Ding, J. L.; Yu, D. L.; Teng, D. X.; He, B.; Chen, X. Y.; Ge, X. Y.; Zhang, Z. P.; Wang, Y.; Yang, X. D.; Shi, T. Z.; Su, F. Z. Machine learning-based detection of soil salinity in an arid desert region, Northwest China: A comparison between Landsat-8 OLI and Sentinel-2 MSI. Science of The Total Environment. 2020, 707, 136092. https://doi.org/10.1016/j.scitotenv.2019.136092.
3. Chen, W. H.; Chen, H. Z.; Feng, Q. X.; Mo, L. N.; Hong, S. Y.; A hybrid optimization method for sample partitioning in near-infrared analysis. Spectrochim Acta A. 2021, 248, 119182. https://doi.org/10.1016/j.saa.2020.119182.
4. Pang, H. Y.; Zhang, A. W.; Kang, X. Y.; He, N. P.; Dong, G. Estimation of the grassland aboveground biomass of the inner mongolia plateau using the simulated spectra of sentinel-2 Images. Remote. Sens-basel. 2020, 12(24). http//:doi.org/10.3390/rs12244155.

Author Response File: Author Response.docx