Soil Organic Carbon Mapping Using Multispectral Remote Sensing Data: Prediction Ability of Data with Different Spatial and Spectral Resolutions

Round 1

Reviewer 1 Report

This paper requires additional analysis and some restructuring before it is acceptable for publication. First, a strict comparison of sensors and models is not particularly novel without sufficient accompanying interpretation. My main concern, then, is that the hyperspectral data is underutilized. The primary objective outlined in the paper’s introduction states that the study is a comparison of multispectral datasets of varying parameters for SOC estimation, and that the hyperspectral data is merely a reference. However, much of the discussion is devoted to the hyperspectral model’s superior performance and largely sidelines the multispectral models. I believe the author’s should make their objectives twofold and explicitly stated in the introduction, with one objective being to quantify the improvement offered by hyperspectral data and the second being to quantify the impact of spatial resolution and different bands in the multispectral models. I’ve suggested below further steps that should be done with the CASI-SASI data. The authors should conduct additional analysis with the PLSR approach and propose feasible explanations for why other multispectral instruments perform relatively poorer, as per the other models. The discussion section should then reflect this and be restructured with clearer paragraph topics that are distinct and connect to each other.

 

Currently, the CASI-SASI dataset is only used as a reference for the multispectral data and predictions, but it should be investigated in greater depth to help explain SOC’s spectral characteristics and interpret why there are differences in retrieval quality from the various sensors. The PLSR portion can be greatly expanded to help with this. To start, statistics/figures for determining the number of latent variables should be reported. A figure showing your PLSR coefficients is also very important. Finally, I would suggest calculating the Variable Importance in the Projection for your PLSR model and displaying the results. See the papers below for some examples of studies that employed PLSR with hyperspectral data and how to depict and report your results. The PLSR coefficients and VIP values will highlight the particular spectral features that are most associated with SOC, allowing you to make more scientific interpretation of the remote sensing data.

 

Singh, A., S. P. Serbin, B. E. McNeil, C. C. Kingdon, and P. A. Townsend. 2015. Imaging spectroscopy algorithms for mapping canopy foliar chemical and morphological traits and their uncertainties. Ecological Applications 25 (8):2180–2197.

 

Jensen, D., K. C. Cavanaugh, M. Simard, G. S. Oking, E. Castañeda-moya, A. McCall, and R. R. Twilley. 2019. Integrating Imaging Spectrometer and Synthetic Aperture Radar Data for Estimating Wetland Vegetation Aboveground Biomass in Coastal Louisiana. Remote Sensing 11 (21):2533. https://www.mdpi.com/2072-4292/11/21/2533.

 

Specific points:

Line 68: There are VNIR hyperspectral UAS sensors available. The sentence should not make UAS and hyperspectral systems exclusive from each other. Line 133-135: The objective statement does not mention the hyperspectral data. It is thus unclear exactly how the CASI-SASI dataset fits in the scope of your outlined objectives. It states later that it is just used as reference data, but that is also unclear and much more can be derived from the data as previously mentioned. Line 143-145: It is unclear what is meant by “there are not enough data inputs available for robust analysis…”. This statement is right by the close of the introduction and undercuts the take-home message. Line 174: Add spatial resolutions to Table 1 Line 279: Make explicit that the spectral data are image-based (i.e. not from a field spectrometer). Line 312: What are hyperparameters and how are they derived? Citations for PLS and machine learning methods would be useful here to reference protocols. Line 421: The opening statement of your discussion is ancillary to your stated objective in the introduction. You should lead with the primary finding vis a vis your objective/hypothesis, then work from there to address the other facets in detail. Each of the first three paragraphs in the discussion start with a similar sentence about the hyperspectral data. Line 485-487: Why can this not be tested in this study?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors:
The title of the manuscript (MS) deals with "Soil Organic Mapping Carbon Using Multispectral Remote Sensing Data: Prediction Ability of Data with Different Spatial, Spectral and Temporal Resolutions". While the topic is very interesting and important, the MS needs more work to be considered for publication.

I suggest the authors consider my comments in the attached PDF. Thanks.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments

Manuscript Number: remotesensing-644420

Title: Mapping Soil Organic Carbon using Multispectral Remote Sensing data: Prediction Ability of Data with Different Spatial, Spectral and Temporal Resolutions

Žížala et al.

 

1.Recommendation

Major Revision

 

Overview and general recommendation:

Soil carbon mapping and carbon stocks have called a lot of attention. Remote and Proximal Sensing techniques have been tested to collect data through cheaper and faster approaches. Nevertheless, despite important achievements it is still an area of research.   

Manuscript remotesensing-644420 seeks to compare the prediction ability of different sources of remote sensing data applied to soil carbon mapping.

In its current content, the paper presents some important limitations. Especially if we consider that there was a time laps between soil sampling in the field and the passage of the satellites. And, added to it, the reference for soil organic carbon prediction was not established with laboratory spectra from a spectrometer oven dried soil samples. In this respect we consider that the paper has to go through major revision.  

 

Major comments (by each paper section)

Title: Soil Organic Mapping Carbon Using Multispectral Remote Sensing Data: Prediction Ability of Data with Different Spatial, Spectral and Temporal Resolutions

Too many words and an awkward phrasing of ‘Soil Organic Mapping Carbon’.

Highlights:

NA

Graphical Abstract:

NA

Abstract:

No comments.

 

Introduction:

Line 42 The manuscript mentioned “soil carbon stocks in the soil”. But, how a 10 cm sampling will contribute to our knowledge of soil carbon stocks, since it usually considers a larger volume of soil up to 1m deep?

Looking at some papers already published at the Remote Sensing journal, we could find Introductions with around 830 to 1170 words, in four to five paragraphs. This manuscript (remotesensing-644420) has some 1500 words in nine paragraphs. Authors should make an effort to cut around 30% of the Introduction, bringing the introduction to 1000 words distributed within five paragraphs maximum.

The focus should be given to similar works, their results and limitations, the scientific gap in the field, and how it links to the hypothesis to be evaluated in this work.    

Line 133: Authors have claimed that “…capability of easily accessible data…”.  We can not say that PlanetScope data is ‘easily available’ since it is not free like Landsat 8 or Sentinel 2. Please rephrase it.

 

Materials and methods:

Line 150 “with an area of 1.45 km2 ”  versus abstract with “on a study plot (100 ha)”. There is near a 50% difference between them.

Line 226 Regarding the sensors CASI and SASI and “with first derivative were used as reference data”. How this choice can affect the results of the study? A more ‘neutral’ reference would be the use of laboratory spectra from spectrometer (e.g. ASD), and collected spectra on oven dried soil samples.  

Table 1. The acquisition date for orbital remote sensing data was 2018, but for the ‘reference’ was 2015. Since the goal is to map soil organic carbon (SOC), the manuscript has to demonstrate how this time difference will affect results. From 2015 to 2018 there could have been a minimum of three cultivating seasons. So, SOC amounts could have increased (or decreased) if we consider the agriculture activities in the area during this period.  We have also to consider that SOC content is not static, and it changes with time. In this context, authors need to present solid arguments considering SOC data and how it is influenced by external factors. Besides, soil sampling were carried out during the field campaign on 6 April 2016, once more another source of variation is introduced since remote sensing and soil sampling were all conducted in different dates.   

 

Line 283 What is the meaning of “composite samples (covering 1 m2)”? How many samples?

 

There is no information regarding soil types in their distribution in the research area. A map with soil types could be presented.  

 

Line 323 Among each best model “the spatial prediction of soil attributes was performed using a selected model with the best predictive ability”. The manuscript should present the most important spectral regions for every sensor. Besides, is there a pattern among sensors and the spectral regions and their contribution to the predictive models?

 

Results:

Table2. Quite often papers related to soil research present a broader picture regarding soil and samples characteristics. The only information presented so far is SOC.  

 

Figure 2. It shows that every sensor collects different information. Since the match between all sensors is not 1:1, how this difference will affect the paper results and comparison?  

 

Discussion:

Line 433 Considering “SOC concentration can have high dynamics in both space and time, and differences in sampling and acquisition time can negatively affect results”. And “…it is necessary to take into account the possible influence on the results.” How those effects could be quantified? What can be found in the literature? How important it would be to have a more reliable reference like a laboratory spectra and oven dry graded soil samples?

 

Line 462 to 471 No need to repeat all RMSE, RPD and RPIQ values from the Results section.

 

Line 505 Regarding “The prediction of soil properties using RS data requires the presence of bare soil in the images.” Not a single information was given in the M&M section regarding the field conditions, any soil cover, the degree of soil purity (straws and residues), or soil wetness at the moment of the Remote Sensing data acquisition.  

 

 

Conclusions:

How the findings will contribute to soil carbon stocks in the soil (from line 42)?

 

References:

No comments.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors have added valuable new content and discussion to this paper, and adequately addressed my concerns.

Reviewer 2 Report

Thank you for the thorough consideration of my comments, and the excellent additions to the manuscript.  

 

Reviewer 3 Report

Manuscript Number: remotesensing-644420 R1

Title: Mapping Soil Organic Carbon using Multispectral Remote Sensing data: Prediction Ability of Data with Different Spatial, Spectral and Temporal Resolutions

Žížala et al.

General recommendation to version R1:

I’ll just add that authors should clearly state that their conclusions are limited by the complexity of each remote sensing acquisition calendar and the number of soil samples (only 50).  

Finally, authors have answered our demands regarding some topics in which the paper could be improved. In my opinion this second version is suitable for publishing.