Satellite-Based Estimation of Soil Moisture Content in Croplands: A Case Study in Golestan Province, North of Iran

Round 1

Reviewer 1 Report (New Reviewer)

This study uses a number of machine learning algorithms to estimate soil moisture content from optical satellite imagery.

I have gone through the manuscript to try to understand the added value of the work here. More especially, how useful the study is to soil moisture monitoring, but I confess that it has been difficult to find very convincing ones.

One the side of data production, I don’t think this study provides much for rigorous applications. But I think some more discussions can be added to improve the scientific relevance of the soil moisture estimates. Hopefully, my comments can help the authors improve this aspect.

A challenge of using these so-called machine learning algorithms is how it is mostly difficult to explain relative importance of input factors. Thus there are some comments I would like to make:

1.      What informed the authors’ decision to choose the factors? Why were these specific factors chosen? I would like the authors to link this the characteristics of the satellite source. Different satellite observations (e.g. microwave based or thermal infrared, etc.) have different challenges which inform how we study the factors that affect quality. Thus it will be good to have some more discussions on this. For instance, how does salinity improve soil moisture estimation?

2.      In Figs 4 and 5, I think more scientific discussions should be given on why we have these orders (rankings) of features? Why would vegetation be more important than another feature? Discussions should be based on how this impact the regional and temporal qualities of soil moisture.

3.      Figure 6 is not clear. Additionally, I don’t see figure 7. I think the discussions of the SMC estimates should be done based on a land features data.

4.      Finally, I think the relevance of the study is not clear. This should be highlighted more especially in the abstract, introduction and conclusions.

5.      What are the limitations of the study? Please add this to the discussions.

6.      Can the obtained results be compared with existing soil moisture data?

7.      What is scantime (of the day) of the satellite observations? And how does this affect the results?

Author Response

Dear Editor,

Thank you for giving us the opportunity to submit our manuscript titled “Satellite-based estimation of soil moisture content in croplands: a case study in Golestan province, north of Iran” to the Remote Sensing Journal. We appreciate the time and effort that you and the reviewers have dedicated to providing your valuable feedback on our manuscript. We have been able to incorporate changes to reflect most of the suggestions provided by the reviewers. These changes have been highlighted within the manuscript. However, for cases where we have not been able to apply the suggestions of the reviewers, it is necessary to state the following explanations:

Dear reviewers, Thank you for your valuable comments. We have tried to respond to your points by improving the introduction, literature review, and the study area and materials. The methodology has been further described to address your points. In the following your points are listed and responses are described in Bold. The modified parts have been highlighted in the manuscript.

Comments from Reviewer 1

This study uses a number of machine learning algorithms to estimate soil moisture content from optical satellite imagery.

I have gone through the manuscript to try to understand the added value of the work here. More especially, how useful the study is to soil moisture monitoring, but I confess that it has been difficult to find very convincing ones.

One the side of data production, I don’t think this study provides much for rigorous applications. But I think some more discussions can be added to improve the scientific relevance of the soil moisture estimates. Hopefully, my comments can help the authors improve this aspect.

A challenge of using these so-called machine learning algorithms is how it is mostly difficult to explain relative importance of input factors. Thus there are some comments I would like to make:

Thank you for your insightful comments on the manuscript. I understand your concerns regarding the added value of the study and the scientific relevance of the soil moisture estimates. Allow me to address your comments one by one.

1. What informed the authors’ decision to choose the factors? Why were these specific factors chosen? I would like the authors to link this the characteristics of the satellite source. Different satellite observations (e.g. microwave based or thermal infrared, etc.) have different challenges which inform how we study the factors that affect quality. Thus it will be good to have some more discussions on this. For instance, how does salinity improve soil moisture estimation?

Point1: Regarding the data production, I would like to clarify that the primary goal of this study is to explore the potential of machine learning algorithms in estimating soil moisture content from optical satellite imagery. While we acknowledge that microwave-based or thermal infrared observations may offer different challenges, our focus was on the analysis of optical satellite imagery. Nonetheless, we appreciate your suggestion, and we can include a brief discussion on how other types of satellite observations could impact the factors affecting soil moisture quality.

Salinity has a significant influence on the estimation of soil moisture in remote sensing due to its effect on the dielectric constant of the soil. The presence of dissolved salts in saline soils can increase the electrical conductivity of the soil, which in turn reduces the effective dielectric constant of the soil, resulting in different relationships between soil moisture and dielectric constant in saline soils compared to non-saline soils. To account for this influence, various approaches have been proposed, including the use of a soil salinity correction factor and the development of empirical models that combine remote sensing data with ground-based measurements of soil moisture and salinity. Studies have shown that accounting for the influence of salinity on soil moisture estimation can significantly improve the accuracy of soil moisture estimates in saline soils. Therefore, it is crucial to consider the influence of salinity when estimating soil moisture using remote sensing techniques.

2. In Figs 4 and 5, I think more scientific discussions should be given on why we have these orders (rankings) of features? Why would vegetation be more important than another feature? Discussions should be based on how this impact the regional and temporal qualities of soil moisture.

Point02: Thank you for your comment on the manuscript. I agree with you that it is essential to provide more scientific discussions on the ranking of features and their relative importance in soil moisture estimation.

In our study, we used machine learning algorithms to estimate soil moisture content from optical satellite imagery, and we applied feature selection techniques to identify the most relevant features. The ranking of features was based on their contribution to the predictive performance of the machine learning models. The most important features were the ones that provided the most significant information gain in predicting soil moisture content. Regarding the importance of vegetation, it has been shown that vegetation indices can provide valuable information on the soil moisture content due to the relationship between plant water content and soil moisture. Vegetation can also influence the surface energy balance and the partitioning of energy between sensible and latent heat fluxes, which in turn affect soil moisture dynamics. Therefore, vegetation can be a critical factor in soil moisture estimation, and this may explain why it was ranked as an essential feature in our study.

3. Figure 6 is not clear. Additionally, I don’t see figure 7. I think the discussions of the SMC estimates should be done based on a land features data.

Point 3-Thank you for your valuable comment, To answer this point, we corrected map figure 6,7

4. Finally, I think the relevance of the study is not clear. This should be highlighted more especially in the abstract, introduction and conclusions.

Point 4- Thanks for your reviewing this work. It has been tried to fully edit parts especially in the abstract, introduction and conclusions.

5. What are the limitations of the study? Please add this to the discussions.

It has several limitations, including:

1. The study only focuses on a specific region, i.e., Golestan province, north of Iran, which may limit the generalizability of the findings to other regions with different soil and vegetation characteristics.
2. The study used only optical satellite imagery, which may not be optimal for estimating soil moisture content, especially in areas with dense vegetation cover or cloud cover. Other types of satellite imagery, such as microwave or thermal infrared, could provide complementary information and improve the accuracy of soil moisture estimates.
3. The study used a limited number of machine learning algorithms and feature selection techniques, which may not capture the full complexity of the soil moisture estimation problem.
4. The study did not consider the irrigation practices, or other factors that could affect soil moisture dynamics in the region.

Overall, while the study provides useful insights into the potential of satellite-based soil moisture estimation in croplands, it is important to consider the limitations and uncertainties associated with the findings. Further research is needed to validate the results and explore the potential of other satellite imagery and machine learning techniques in improving the accuracy and generalizability of soil moisture estimates.

6. Can the obtained results be compared with existing soil moisture data?

It is important to note that a comparison between the two products should take into account their differences in spatial resolution and ground survey sampling technique. However, it could be a valuable future direction to explore the accuracy and reliability of satellite-based soil moisture estimates in the study area.

7. What is scantime (of the day) of the satellite observations? And how does this affect the results?

The scan time of the day can affect the results of satellite-based studies in several ways. For example, the vegetation water content and soil moisture can vary throughout the day due to changes in temperature, solar radiation, and atmospheric conditions. Therefore, the scan time of the day may impact the accuracy of the soil moisture estimates. However, in this study we utilized Landsat-8 data which in 10:25 am in the region of interest and is does not change between different acquisitions. Therefore, the machine learning itself will ignore this factor.

Author Response File: Author Response.pdf

Reviewer 2 Report (New Reviewer)

Soil moisture level influences crops growth and thus is an important parameter across croplands. This study used optical satellite remote sensing data to estimate soil moisture in croplands, which is an interesting research topic. The overall structure of this manuscript is reasonable, and the developed empirical models show acceptable accuracy in deriving soil moisture. However, there are some concerns that should be addressed.

(1) Optical remote sensing can only capture canopy reflectance and has limited ability in penetrating crops. It means that optical remote sensing obtains information on crops rather than soil in croplands. Microwave remote sensing can penetrate vegetation and capture information on soli, which is the reason why microwave remote sensing usually be used to derive soil moisture at regional and global scales. Please explain the physical mechanism of soil moisture estimation from optical satellite.

(2) It is stated that “variety of crop types are cultivated”. Please investigate, compare, and analyze the model performance under different crops.

(3) There are many open soil moisture products at regional and global scales. To prove the feasibility of the developed estimation model, please compare this study mapped spatial distribution of soil moisture with current existing products.

(4) Two figure 6?

(5) Comparison results among different machine learning methods and different satellites data should be shown using scatter figures. If only statistical indicators are listed in tables, the readers cannot clearly understand the detailed models’ performance.

Author Response

Comments from Reviewer 2

Soil moisture level influences crops growth and thus is an important parameter across croplands. This study used optical satellite remote sensing data to estimate soil moisture in croplands, which is an interesting research topic. The overall structure of this manuscript is reasonable, and the developed empirical models show acceptable accuracy in deriving soil moisture. However, there are some concerns that should be addressed.

(1) Optical remote sensing can only capture canopy reflectance and has limited ability in penetrating crops. It means that optical remote sensing obtains information on crops rather than soil in croplands. Microwave remote sensing can penetrate vegetation and capture information on soli, which is the reason why microwave remote sensing usually be used to derive soil moisture at regional and global scales. Please explain the physical mechanism of soil moisture estimation from optical satellite.

Point 1-You are correct that optical remote sensing can only capture canopy reflectance and has limited ability to penetrate crops, which means that it obtains information on crops rather than soil in croplands. However, it is possible to estimate soil moisture from optical satellite imagery through the physical mechanism of vegetation response to soil moisture.

Vegetation responds to changes in soil moisture by altering its biophysical properties, such as leaf area index, canopy cover, and vegetation water content. This leads to changes in the reflectance of different spectral bands captured by optical sensors. For instance, in the near-infrared (NIR) spectral region, vegetation reflectance decreases as soil moisture increases due to increased canopy cover and water absorption. Conversely, in the shortwave infrared (SWIR) spectral region, vegetation reflectance increases as soil moisture increases due to increased water content in the leaves.

By using machine learning algorithms that take into account the relationships between vegetation biophysical properties and soil moisture, it is possible to estimate soil moisture from optical satellite imagery. However, it is important to note that the accuracy of these estimates may be affected by factors such as scan time, atmospheric conditions, and the type of vegetation and soil present in the study area. It is also worth mentioning the high impact of vegetation index from feature importance analysis.

(2) It is stated that “variety of crop types are cultivated”. Please investigate, compare, and analyze the model performance under different crops.

Point 2-Thank you for your comment. We appreciate your suggestion to investigate and compare the model performance under different crops. However, due to the limitations of our data and resources, it is not possible to compare the performance of our machine learning models under different crops. The study area has a large number of crop types cultivated, and collecting and analyzing soil moisture data for each crop type would require a significant amount of additional resources and time. Therefore, our analysis focuses on estimating soil moisture in croplands in general, rather than for specific crop types. We acknowledge that this may limit the generalizability of our findings to specific crops, but we believe that our approach provides valuable insights into the potential of using satellite-based soil moisture estimation for cropland management.

(3) There are many open soil moisture products at regional and global scales. To prove the feasibility of the developed estimation model, please compare this study mapped spatial distribution of soil moisture with current existing products.

Point (3) The resolution of existing products is coarse, and the entire area is covered with a few pixels. Therefore, this comparison with our production with a resolution of 30 meters is not logical and has no results.

(4) Two figure 6?

Point 4- Thank you for your comment. This problem is solved

(5) Comparison results among different machine learning methods and different satellites data should be shown using scatter figures. If only statistical indicators are listed in tables, the readers cannot clearly understand the detailed models’ performance.

Point 5- Thanks for your opinion about the research. It has been tried to fully answer your questions to improve the article Considering that in the beginning it is assumed that I drove scatter plot but I have drawn a scatter diagram, but due to the repeated content, we have avoided it because it is the same as the table

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (New Reviewer)

I thank the authors for addressing the points I raised. 

Just a follow up on a few points:

1. The authors will provide a more convincing argument for the relevance of their work if they could outline how the factors they consider hamper the qualities in other satellite sources like TIRs and microwave based soil moisture datasets as well as their retrieval models.

2. The authors rightly noted in their response #2 that vegetation can impact the energy balance, but don't translate some of these lessons into how this affects the time of observation which the authors noted as 10:25am. Studies have shown that even at that time (such as with ASCAT and FengYun3C), we find active evapotranspiration going on which reduces thermal equilibrium and makes soil moisture retrieval more difficult. Please include some impacts of acquisition time on the retrievals here.

Author Response

Point1-Thank you for your comment. While we agree that exploring how the factors we consider may impact the qualities of other satellite sources and retrieval models could be informative, it may not be feasible due to the scope and focus of our study. Our study primarily focuses on developing a machine learning model to estimate soil moisture content from optical satellite imagery in a specific region, which was also mentioned in the limitation part of the revised version.

Point2-Thank you for your comment. You raise a valid point regarding the impact of acquisition time on soil moisture retrieval from optical satellite imagery. However, by considering the impact of this factor we tried to decrease its effect. As mentioned in the paper, section 2.2, the fields had sparse crop cover at the time of sampling, which might have lessened the impact of active evapotranspiration on the soil moisture retrievals. However, it is still important to note that acquisition time can affect the thermal equilibrium and therefore impact the accuracy of soil moisture retrievals, especially in areas with denser vegetation cover. So, we added the following to the limitation part:

“To minimize the impact of active evapotranspiration on the soil moisture retrievals, this study conducted sampling in cropland areas when fields had sparse crop cover, and utilized the relevant satellite data. While this approach was suitable for the goal of studying the impact of drainage, it may not be recommended for temporal studies, particularly when the crop cover is dense.

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

There are still two issues that should be addressed. The explanation of question 1 should also be added in the main manuscript.

In addtion, I donot think scatter figures have repeated information on model evaluation compared to table. Because indicators in table are just numbers, while readers can examine samples distribution via scatter figures. It means that scatter figures have more infromation. It is suggested that evaluation results of the optimal model should be shown in scatter figure.

Author Response

Point1-Thank you for the comment. To answer this point, we added the following parts to the manuscript, Page 26, lines 832-850: Section 1 Highlighted in red.

Point 2-Thank you for the comment. To answer this point, we added the following parts to the manuscript, Page 16, 17 lines 535-598: Section3-1 result Highlighted in red.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Before publishing, the criteria listed below must be taken into account.

I feel authors may significantly improve the quality of their paper to make it more insightful after properly addressing the suggestions.

In my opinion, the present MS hasn't completely explored certain intriguing features.

In particular, the authors do not present results in form of the seasonal soil moisture variations, however it would be interesting to plot corresponding CV for each mean soil moisture intervals. This analysis method may describe some interesting patterns. In terms of spatial variation, the relation between CV and soil moisture often show a hysteresis pattern. It will be interesting to see whether the hysteresis is also observed in this study and how that might difference between the different rainfall events. The authors could also use coefficient of variation and standard deviation and relative difference as used in a very recent studies (https://doi.org/10.36334/modsim.2019.K6.srivastava; https://doi.org/10.1002/2014WR016102; https://doi.org/10.1029/2009WR008611).

To show the seasonal fluctuations for each component, perhaps the authors might also plot the frequency distribution of CV and mean temporal soil moisture? In order to give their views more scientific weight, authors may choose to identify research that support their claims. In the introduction, a thorough discussion of soil moisture might be included. The first and second paragraphs from the above studies as well as any relevant research may be added by the authors. I strongly advise authors to incorporate these significant research since they are unquestionably significant and will enhance the credibility of the work.

 Please include a description of the findings that contrasts them with the study's original purpose. Have you addressed each of the introduction-listed research questions? Yes, but how? If not, what measures must be taken in order to respond to them?

Reviewer 2 Report

The present study has evaluated several machine learning methods for estimating soil moisture over cropland. To my opinion, there is no fault with the proposed methods and results, also no novelty can be found in this kind of studies. The manuscript is overall clear written, but the authors should be more serious. The reference number in the context is not in order.

Reviewer 3 Report

The study by Bandak et al. assesses the soil moisture variations using in-situ measurement and satellite imaging. The topic is interesting and within the scope of the journal. But there are several crucial things that need to be resolved before proceeding. Please find the detailed comments and suggestions below.  

There are too many typos (e.g., Lines 4, 5, among several others). Please read through the manuscript again for such typos and punctuation errors.

In the introduction, before directly coming to soil moisture, general background about terrestrial systems (e.g., terrestrial water storage, https://doi.org/10.1016/j.jhydrol.2021.126868) and then highlighting the importance of soil moisture should be better.

Lines 9-62. Also crucial for the propagation behavior of the drought, and subsequently for water availability. Please see the above reference for such details.  

Lines 80-83. The research gap is not clear and therefore the motivation of this study is weak. The authors need to reformulate their objectives and highlight the novelty more convincingly. 

Lines 93-94. What is the temperature range? 

What does the ‘point’ (in legends) mean in Fig. 1? Also, the caption should be revised to include all the information presented in the figure. Same comment for all other figures too. 

Line 107. Any information about the exact depth associated with each sample?

Line 123. What are S2 and L8? Not defined yet. 

Possible uncertainties and errors associated with human as well as instrument errors while collection, transportation, and laboratory tests should be discussed. And uncertainty bounds should be provided to have an idea of the accuracy of the results. 

Table 1. Please also mention the reference of the dataset. And how did the authors deal with the varying spatial resolution of various datasets? 

Table 2. Authors should provide reasoning about the selection of these indices.

Lines 285-305. Such ML-based application has also been employed for leveraging dwindling groundwater in India and for data gap filling in GRACE-based TWS data and streamflow, among others.

Fig. 3. Please revise for better comprehension. 

The discussion section is too short to infer meaningful takeaways. To name a few, a more rigorous sensitivity analysis should be performed for the three methods, a detailed discussion about the observed match and mismatch between these methods should be discussed, a comparison with other studies in this region and other regions should be provided (e.g., https://doi.org/10.1038/s41598-019-52650-3), implications of the current findings should be discussed with future research directions. 

Lines 604-605. How about the applicability of these findings to other regions where field sampling is challenging? Authors should provide such a generalized application of the methods employed.