the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Disentangling the effects of vegetation and water on the satellite observations of soil organic carbon stocks in western European topsoils
Abstract. The performance of models based on satellite observations of soil organic carbon (SOC) stock in European soils is seriously limited by the complexity of natural land surfaces. Therefore, disentangling the SOC stock from other natural land surfaces including vegetation and water bodies has become a rather difficult but necessary task. This study proposed a novel and promising approach intended to resolve this frustrating problem. Based on a series of spectral narrowing, unchanging, and enlarging processes, 23,914,845 sets of SOC models were developed both for vegetation fuzzy disentangling and water fuzzy disentangling. The optimal model was obtained through comparison and was determined as the model that ultimately performed obviously better than the unfuzzified model. This model simulated the per-unit and total SOC stocks in western European topsoils as 99.742 t C ha−1 and 9.373 Pg, respectively. In comparison with the results of previous studies, the gaps in the simulated per-unit SOC stocks across the western European countries were considerably narrower (83.673–104.334 t C ha−1). The outstanding model performance and stable simulated per-unit values are the result of disentangling of the vegetation and water cover. This study proposed a valuable reference solution for disentangling the SOC stock from complex natural land cover.
- Preprint
(847 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on bg-2023-170', Anonymous Referee #1, 21 Nov 2023
General comments:
This study by Lixin Lin and coworkers is focused on disentangling the soil organic carbon (SOC) stock from complex vegetation and water cover using fuzzy deep learning, employing satellite data (Landsat-8) and spectral indices, like NDVI and NDWI. The model modifies the spectral reflectance according to the vegetation and water cover so that it is optimized for SOC stock estimations.
The whole paper could benefit a lot from better writing and would improve content understanding. Work that has been done should be explained better and the scope of this work should be made clear:
Vegetation, and water can affect the spectral reflectance in satellite imagery, and a fuzzy formula to modify the reflectance according to vegetation and water indices is used. Models that use the fuzzified spectra perform better than the unfuzzyfied.Specific comments and Technical Corrections:
Line 33 - 38: The references and their description from line 33 to line 38 should be made clearer:
Ward et al. tested SOC ML models using airborne hyperspectral remote sensing data and simulated satellite EnMAP data as input.
Hutengs et al. examined in-situ spectroscopy and SOC estimation models using in-situ MIR spectra.
Yigini and Panagos performed a digital soil mapping for SOC, using climate, land cover, terrain, and soil covariates.
Lugato et al. used an agro-ecosystem SOC model to calculate SOC stocks using soil/climate/land-use/management drivers, among others.The sequence should also make sense to the reader and address the challenges related to vegetation and water faced by SOC estimation models using satellite remote sensing data. Subsequently, the rationale for this study should be elaborated.
Line 36: Such studies usually predict SOC content, that can be used as a proxy for SOC stock, but not stock directly.
Line 48: The following should be a new paragraph "Therefore, the focus of this study..."
Line 109: "inverse" does not seem to be the correct word: "modify", for example, seems more appropriate.
Line 180: Please check whether they used satellite imagery in this study.
Citation: https://doi.org/10.5194/bg-2023-170-RC1 -
CC1: 'Response to Editor Reviewers Comments-bg-2023-170', Lixin Lin, 26 Nov 2023
Dear Editor of the BG:
We would like to express our heartfelt gratitude to you for giving us this chance to revise our manuscript entitled "Disentangling the effects of vegetation and water on the satellite observations of soil organic carbon stocks in western European topsoils" (bg-2023-170). Thanks for your and reviewers’ help, we have carefully reviewed all comments and revised the manuscript accordingly. Once again, we really appreciate all your help and wish you and your family happy everyday! The revised manuscript and response to editor reviewers’ comments see the zip-based supplement.
Corresponding author: Lixin Lin
Address: School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing, 210044, China
Email: llxshxdhl8@nuist.edu.cn; llxshxdhl8@gmail.com. (My gmail was added as spare)
Thanks.
Wishes.
-
AC2: 'Reply on RC1', Lixin Lin, 22 Dec 2023
Dear Editor of the BG:
Dear Editor of the BG:
We would like to express our heartfelt gratitude to you for giving us this chance to revise our manuscript entitled "Disentangling the effects of vegetation and water on the satellite observations of soil organic carbon stocks in western European topsoils" (bg-2023-170). Thanks for your and reviewers’ help, we have carefully reviewed all comments and revised the manuscript accordingly ,and the language quality of this manuscript has been improved by James Buxton, Ph.D. from Liwen Bianji (Edanz) (www.liwenbianji.cn).
Once again, we really appreciate all your help and wish you and your family happy everyday! The revised manuscript and response to editor reviewers’ comments see the zip-based supplement.
Corresponding author: Lixin Lin
Address: School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing, 210044, China
Email: llxshxdhl8@nuist.edu.cn; llxshxdhl8@gmail.com.
-
CC1: 'Response to Editor Reviewers Comments-bg-2023-170', Lixin Lin, 26 Nov 2023
-
RC2: 'Comment on bg-2023-170', Anonymous Referee #2, 06 Dec 2023
This manuscript bg-2023-170 is entitled “Disentangling the effects of vegetation and water on the satellite observations of soil carbon stocks in western European topsoils”. The authors state that the performance of satellite-derived models of soil organic carbon (SOC) stocks is “seriously limited by the complexity of natural land surfaces”. To address such limitations, they propose to develop a so-called “fuzzy disentangling” which multiplies the number of satellite-derived SOC stocks predicting models, then select the best predicting models.
Overall the article is confusing. It is based on massive calculation (~24 millions of models…calculation cost and duration?) but without sound justification and link to observed spectral behaviour of soil organic carbon.
The soil stocks are computed from the European soil reference LUCAS database of 2018 over six EU-member countries (France, Belgium, Luxembourg, Netherlands, Ireland) in addition to United Kingdom. Why were these countries chosen and not countries of Southern Europe?
Introduction: lines 30-33: the authors refer to the “spectral response of SOC” citing Thaler et al (2019), who derived a spectral index for SOC content prediction from bare soil reflectance. It seems that the authors confuse SOC contents and SOC stocks, and the predictions that can be derived from bare soil (for instance at European scale, Castaldi et al 2021) from those which can be indirectly derived from vegetated surfaces. The authors ignore a number of studies that have derived SOC content from satellite, and notably from bare soil (see review paper doi.org/10.3390/rs14122917).
The SOC stock estimation from bare soil only is tricky, as neither the bulk density nor the coarse fragment content is easy to measure and model. aAccuracy and uncertainty for the formula used to compute the bulk density that relies on a Hollis et al 2012’s pedotransfer function should be discussed. The gain yielded by fuzzification is unlikely to compensate for very uncertain estimated stocks at LUCAS locations.
The concept of disentangling is not well explained and motivated. What does it aim to exactly, compared to another algorithm such as random forest, which provides the variable importance?
The fuzzy model underlying the study notably relies on the following assumption “if spectral reflectance is independent on vegetation, then the two pixels will have similar spectral reflectance”: does it means that all bare soils have similar reflectance?
Moreover, why water should be disentangled, while surface waters i) are not included in the SOC stock assessment; ii) can easily be discriminated from other surfaces, for instance through a mask of landuse map.
Other methodological details are as fuzzy as the methodological section: is the fuzzy model applied band per band or over the seven Landsat bands as a whole? Are the reflectance values derived from the whole time series from March to November 2015 (if so, how?) or from one single date (which one?)? the accuracy assessment procedure is not explained. This is a very important issue given the importance of monitoring changes (see De Rosa et al 2023 doi.org/10.1111/gcb.16992).
Despite the potential interest of this topic, this manuscript needs substantial reworking hence the recommendation of rejection and encouragement to resubmit.
In detail:
Line 36: the cited reference Ward et al 2020 shows tremendous potential for SOC content (not stock) prediction from bare soil (not vegetated) and this, at local scale (not European)
Lines 37-38: please better justify what spectral effects vegetation and water may have on SOC content/stock prediction. On what previous references does this assertion rely?
Line 40: the authors cite lab/field spectroscopy studies but there is a corpus of remote sensing- studies based on airborne or satellite imagery that should be referred to.
Line 42: please specify what covariates were used; it is unclear whether this study used satellite remote sensing or not. Same comment for the other studies in the remaining of the introduction.
Line 54: this study uses two indices only, it did not investigate the issue of “whether the satellite spectral indices…” but two of them;
Lines 57-58: please provide references for NDVI and NDWI, and the reason for using these indices
Line 60: mainland France (excluding Corsica), unless overseas France was included in the study? You can drop Monaco at this scale.
Lines 75-77: please specify accuracy and application domain
Line 94: many studies, only one reference is cited to justify these many studies, please cite more references or else, rephrase.
Line 103: reflectance in general or in a specific band?
Line 105: spectral reflectance independent of vegetation…then specify what it corresponds to (bare soil or water surface or impervious surface?) Is it relevant to consider two pixels with same stock amount having the same reflectance spectra?
Line 107: reflectance in general or in a specific band?
Line 117: reflectance of a given band? Fuzzification, please explain; should NDVI/NDWI be understood here as “one or the other” or as the ratio of the two?
Line 132: Wadoux et al 2019 used RF to optimize sampling design and used SOC concentration as target property, but did not indicate that RF is a useful tool for SOC specifically.
Lines 133-136: please chose reference other than Ballabio et al 2019 (who used a Gaussian process regression). Please better elaborate on what PLS-RF means: does it mean the averaging of random forest and partial least square (Cardelli et al, 2017) or “random frog partial least square”, which seems to have nothing to do with the previous one (Wang et al 2019).
Line 137: please explain the accuracy assessment procedure: how were the results validated? Please add RPIQ and specify whether variables have normal distribution
Line 161: what is the “unit” meant in “per-unit”? per country?
Citation: https://doi.org/10.5194/bg-2023-170-RC2 -
AC1: 'Reply on RC2', Lixin Lin, 21 Dec 2023
Dear Editor of the BG:
We would like to express our heartfelt gratitude to you for giving us this chance to revise our manuscript entitled "Disentangling the effects of vegetation and water on the satellite observations of soil organic carbon stocks in western European topsoils" (bg-2023-170). Thanks for your and reviewers’ help, we have carefully reviewed all comments and revised the manuscript accordingly. Once again, we really appreciate all your help and wish you and your family happy everyday!
The revised manuscript and response to editor reviewers’ comments see the zip-based supplement.
Corresponding author: Lixin Lin
Address: School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing, 210044, China
Email: llxshxdhl8@nuist.edu.cn; llxshxdhl8@gmail.com
-
AC1: 'Reply on RC2', Lixin Lin, 21 Dec 2023
Status: closed
-
RC1: 'Comment on bg-2023-170', Anonymous Referee #1, 21 Nov 2023
General comments:
This study by Lixin Lin and coworkers is focused on disentangling the soil organic carbon (SOC) stock from complex vegetation and water cover using fuzzy deep learning, employing satellite data (Landsat-8) and spectral indices, like NDVI and NDWI. The model modifies the spectral reflectance according to the vegetation and water cover so that it is optimized for SOC stock estimations.
The whole paper could benefit a lot from better writing and would improve content understanding. Work that has been done should be explained better and the scope of this work should be made clear:
Vegetation, and water can affect the spectral reflectance in satellite imagery, and a fuzzy formula to modify the reflectance according to vegetation and water indices is used. Models that use the fuzzified spectra perform better than the unfuzzyfied.Specific comments and Technical Corrections:
Line 33 - 38: The references and their description from line 33 to line 38 should be made clearer:
Ward et al. tested SOC ML models using airborne hyperspectral remote sensing data and simulated satellite EnMAP data as input.
Hutengs et al. examined in-situ spectroscopy and SOC estimation models using in-situ MIR spectra.
Yigini and Panagos performed a digital soil mapping for SOC, using climate, land cover, terrain, and soil covariates.
Lugato et al. used an agro-ecosystem SOC model to calculate SOC stocks using soil/climate/land-use/management drivers, among others.The sequence should also make sense to the reader and address the challenges related to vegetation and water faced by SOC estimation models using satellite remote sensing data. Subsequently, the rationale for this study should be elaborated.
Line 36: Such studies usually predict SOC content, that can be used as a proxy for SOC stock, but not stock directly.
Line 48: The following should be a new paragraph "Therefore, the focus of this study..."
Line 109: "inverse" does not seem to be the correct word: "modify", for example, seems more appropriate.
Line 180: Please check whether they used satellite imagery in this study.
Citation: https://doi.org/10.5194/bg-2023-170-RC1 -
CC1: 'Response to Editor Reviewers Comments-bg-2023-170', Lixin Lin, 26 Nov 2023
Dear Editor of the BG:
We would like to express our heartfelt gratitude to you for giving us this chance to revise our manuscript entitled "Disentangling the effects of vegetation and water on the satellite observations of soil organic carbon stocks in western European topsoils" (bg-2023-170). Thanks for your and reviewers’ help, we have carefully reviewed all comments and revised the manuscript accordingly. Once again, we really appreciate all your help and wish you and your family happy everyday! The revised manuscript and response to editor reviewers’ comments see the zip-based supplement.
Corresponding author: Lixin Lin
Address: School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing, 210044, China
Email: llxshxdhl8@nuist.edu.cn; llxshxdhl8@gmail.com. (My gmail was added as spare)
Thanks.
Wishes.
-
AC2: 'Reply on RC1', Lixin Lin, 22 Dec 2023
Dear Editor of the BG:
Dear Editor of the BG:
We would like to express our heartfelt gratitude to you for giving us this chance to revise our manuscript entitled "Disentangling the effects of vegetation and water on the satellite observations of soil organic carbon stocks in western European topsoils" (bg-2023-170). Thanks for your and reviewers’ help, we have carefully reviewed all comments and revised the manuscript accordingly ,and the language quality of this manuscript has been improved by James Buxton, Ph.D. from Liwen Bianji (Edanz) (www.liwenbianji.cn).
Once again, we really appreciate all your help and wish you and your family happy everyday! The revised manuscript and response to editor reviewers’ comments see the zip-based supplement.
Corresponding author: Lixin Lin
Address: School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing, 210044, China
Email: llxshxdhl8@nuist.edu.cn; llxshxdhl8@gmail.com.
-
CC1: 'Response to Editor Reviewers Comments-bg-2023-170', Lixin Lin, 26 Nov 2023
-
RC2: 'Comment on bg-2023-170', Anonymous Referee #2, 06 Dec 2023
This manuscript bg-2023-170 is entitled “Disentangling the effects of vegetation and water on the satellite observations of soil carbon stocks in western European topsoils”. The authors state that the performance of satellite-derived models of soil organic carbon (SOC) stocks is “seriously limited by the complexity of natural land surfaces”. To address such limitations, they propose to develop a so-called “fuzzy disentangling” which multiplies the number of satellite-derived SOC stocks predicting models, then select the best predicting models.
Overall the article is confusing. It is based on massive calculation (~24 millions of models…calculation cost and duration?) but without sound justification and link to observed spectral behaviour of soil organic carbon.
The soil stocks are computed from the European soil reference LUCAS database of 2018 over six EU-member countries (France, Belgium, Luxembourg, Netherlands, Ireland) in addition to United Kingdom. Why were these countries chosen and not countries of Southern Europe?
Introduction: lines 30-33: the authors refer to the “spectral response of SOC” citing Thaler et al (2019), who derived a spectral index for SOC content prediction from bare soil reflectance. It seems that the authors confuse SOC contents and SOC stocks, and the predictions that can be derived from bare soil (for instance at European scale, Castaldi et al 2021) from those which can be indirectly derived from vegetated surfaces. The authors ignore a number of studies that have derived SOC content from satellite, and notably from bare soil (see review paper doi.org/10.3390/rs14122917).
The SOC stock estimation from bare soil only is tricky, as neither the bulk density nor the coarse fragment content is easy to measure and model. aAccuracy and uncertainty for the formula used to compute the bulk density that relies on a Hollis et al 2012’s pedotransfer function should be discussed. The gain yielded by fuzzification is unlikely to compensate for very uncertain estimated stocks at LUCAS locations.
The concept of disentangling is not well explained and motivated. What does it aim to exactly, compared to another algorithm such as random forest, which provides the variable importance?
The fuzzy model underlying the study notably relies on the following assumption “if spectral reflectance is independent on vegetation, then the two pixels will have similar spectral reflectance”: does it means that all bare soils have similar reflectance?
Moreover, why water should be disentangled, while surface waters i) are not included in the SOC stock assessment; ii) can easily be discriminated from other surfaces, for instance through a mask of landuse map.
Other methodological details are as fuzzy as the methodological section: is the fuzzy model applied band per band or over the seven Landsat bands as a whole? Are the reflectance values derived from the whole time series from March to November 2015 (if so, how?) or from one single date (which one?)? the accuracy assessment procedure is not explained. This is a very important issue given the importance of monitoring changes (see De Rosa et al 2023 doi.org/10.1111/gcb.16992).
Despite the potential interest of this topic, this manuscript needs substantial reworking hence the recommendation of rejection and encouragement to resubmit.
In detail:
Line 36: the cited reference Ward et al 2020 shows tremendous potential for SOC content (not stock) prediction from bare soil (not vegetated) and this, at local scale (not European)
Lines 37-38: please better justify what spectral effects vegetation and water may have on SOC content/stock prediction. On what previous references does this assertion rely?
Line 40: the authors cite lab/field spectroscopy studies but there is a corpus of remote sensing- studies based on airborne or satellite imagery that should be referred to.
Line 42: please specify what covariates were used; it is unclear whether this study used satellite remote sensing or not. Same comment for the other studies in the remaining of the introduction.
Line 54: this study uses two indices only, it did not investigate the issue of “whether the satellite spectral indices…” but two of them;
Lines 57-58: please provide references for NDVI and NDWI, and the reason for using these indices
Line 60: mainland France (excluding Corsica), unless overseas France was included in the study? You can drop Monaco at this scale.
Lines 75-77: please specify accuracy and application domain
Line 94: many studies, only one reference is cited to justify these many studies, please cite more references or else, rephrase.
Line 103: reflectance in general or in a specific band?
Line 105: spectral reflectance independent of vegetation…then specify what it corresponds to (bare soil or water surface or impervious surface?) Is it relevant to consider two pixels with same stock amount having the same reflectance spectra?
Line 107: reflectance in general or in a specific band?
Line 117: reflectance of a given band? Fuzzification, please explain; should NDVI/NDWI be understood here as “one or the other” or as the ratio of the two?
Line 132: Wadoux et al 2019 used RF to optimize sampling design and used SOC concentration as target property, but did not indicate that RF is a useful tool for SOC specifically.
Lines 133-136: please chose reference other than Ballabio et al 2019 (who used a Gaussian process regression). Please better elaborate on what PLS-RF means: does it mean the averaging of random forest and partial least square (Cardelli et al, 2017) or “random frog partial least square”, which seems to have nothing to do with the previous one (Wang et al 2019).
Line 137: please explain the accuracy assessment procedure: how were the results validated? Please add RPIQ and specify whether variables have normal distribution
Line 161: what is the “unit” meant in “per-unit”? per country?
Citation: https://doi.org/10.5194/bg-2023-170-RC2 -
AC1: 'Reply on RC2', Lixin Lin, 21 Dec 2023
Dear Editor of the BG:
We would like to express our heartfelt gratitude to you for giving us this chance to revise our manuscript entitled "Disentangling the effects of vegetation and water on the satellite observations of soil organic carbon stocks in western European topsoils" (bg-2023-170). Thanks for your and reviewers’ help, we have carefully reviewed all comments and revised the manuscript accordingly. Once again, we really appreciate all your help and wish you and your family happy everyday!
The revised manuscript and response to editor reviewers’ comments see the zip-based supplement.
Corresponding author: Lixin Lin
Address: School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing, 210044, China
Email: llxshxdhl8@nuist.edu.cn; llxshxdhl8@gmail.com
-
AC1: 'Reply on RC2', Lixin Lin, 21 Dec 2023
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
377 | 98 | 36 | 511 | 46 | 28 |
- HTML: 377
- PDF: 98
- XML: 36
- Total: 511
- BibTeX: 46
- EndNote: 28
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1