Evaluation of Sentinel-2 Vegetation Indices for Estimating Leaf Area Index in Cassava Plots
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsReview of the article: “Sentinel-2 Vegetation Indices for Estimating Cassava Leaf Area Index across Growth Stages in Thailand”
In brief, the article is well written and poses a relevant issue closely related to the crop remote sensing. In this paper, thorough assessment on the usefulness of sentinel2 derived vegetation indices for prediction of cassava leaf area index at different phenological stages was performed over extensive ground reference measurements. The method is an appropriate and well explained research design. The results provide useful insights into the performance of the index and stage specific dynamics that is valuable for both research and operational applications. The manuscript can however be improved by some clarifications and slight tightening to make it more smooth to the reader, if statements are more strongly interpreted in some cases. My fuller comments are below.
1. Introduction
The introduction gives sufficient background on LAI and the use of satellite data for crop monitoring. However, the research gap is not stated clearly enough. The authors should explain more directly why previous LAI studies are still limited for cassava, particularly with respect to differences across growth stages. A short comparison with LAI studies in other crops would help clarify what is missing and improve the overall context of the study. The objectives should also be stated more clearly at the end of the introduction.
2. Materials and Method
2.4.1. Regression analysis between VIs and ground–LAI
The statistical approach is appropriate. However, the manuscript should clarify how repeated monthly LAI observations from the same plots were treated in the regression analysis, particularly with respect to potential temporal dependence in the data. This can be addressed by explaining whether monthly observations were treated as independent or whether temporal dependence was considered negligible due to the stage-based analysis. From a reviewer’s perspective, re-analysis is not considered necessary.
3. Results
3.1. Temporal Patterns of Ground–LAI and Vegetation indices
The temporal patterns of LAI and vegetation indices are clearly presented. I believe that linking these patterns more explicitly to crop growth stages or seasonal dynamics would improve the interpretation of the results. This can be addressed by adding an explanation that relates key changes in LAI and vegetation indices to specific growth stages or seasonal conditions.
3.2 Stage-Specific Performance of Vegetation Indices
The predictive performance was evaluated using aggregated data across all growth stages. Stage-wise analyses are recommended to better capture phenological differences.
3.3. Stage-Specific Performance of Vegetation Indices
While stage-wise results are presented, additional statistical comparisons (e.g., significance testing or error analysis) would further strengthen the evaluation.
3.4. Model Performance Comparison by Plot Types
In my view, using different validation procedures between plot types may bias the comparison. A consistent validation method and comparable confidence intervals for both groups would improve fairness.
4. Discussion
Expanding the discussion to include broader comparisons with existing literature would help clarify the general applicability of the findings.
5. Conclusions
The practical implications of the proposed models are only briefly mentioned. Expanding on how the approach can be applied in real-world crop monitoring or decision-making would strengthen the overall impact of the study.
Author Response
In brief, the article is well written and poses a relevant issue closely related to the crop remote sensing. In this paper, thorough assessment on the usefulness of sentinel2 derived vegetation indices for prediction of cassava leaf area index at different phenological stages was performed over extensive ground reference measurements. The method is an appropriate and well explained research design. The results provide useful insights into the performance of the index and stage specific dynamics that is valuable for both research and operational applications. The manuscript can however be improved by some clarifications and slight tightening to make it more smooth to the reader, if statements are more strongly interpreted in some cases. My fuller comments are below.
> Thank you for your thoughtful consideration of our manuscript. We have addressed all of the issues that you have raised in this manuscript. You might find our responses after “>” below each comment.
- Introduction
The introduction gives sufficient background on LAI and the use of satellite data for crop monitoring. However, the research gap is not stated clearly enough. The authors should explain more directly why previous LAI studies are still limited for cassava, particularly with respect to differences across growth stages.
A short comparison with LAI studies in other crops would help clarify what is missing and improve the overall context of the study.
> We added a short review of LAI studies in other crops in the introduction.
The objectives should also be stated more clearly at the end of the introduction.
> We modified the last paragraph on the introduction to make our objectives more explicit.
- Materials and Method
2.4.1. Regression analysis between VIs and ground–LAI
The statistical approach is appropriate. However, the manuscript should clarify how repeated monthly LAI observations from the same plots were treated in the regression analysis, particularly with respect to potential temporal dependence in the data. This can be addressed by explaining whether monthly observations were treated as independent or whether temporal dependence was considered negligible due to the stage-based analysis. From a reviewer’s perspective, re-analysis is not considered necessary.
> Temporal dependence was not directly assessed in this model, because each plot only had one value for each month and the effect of time was examined separate stage-specific analyses below, as you correctly pointed out.
- Results
3.1. Temporal Patterns of Ground–LAI and Vegetation indices
The temporal patterns of LAI and vegetation indices are clearly presented. I believe that linking these patterns more explicitly to crop growth stages or seasonal dynamics would improve the interpretation of the results. This can be addressed by adding an explanation that relates key changes in LAI and vegetation indices to specific growth stages or seasonal conditions.
> We have the explanation for the temporal patterns related to crop growth stage in the discussion section 4.1. We have revised the text to provide the clearer links to the growth stages in Section 3.1 as well.
3.2 Performance of Vegetation Indices
The predictive performance was evaluated using aggregated data across all growth stages. Stage-wise analyses are recommended to better capture phenological differences.
> In the section 3.2, we wanted to determine the general applicability of each VI in predicting ground LAI regardless of growth stages. The effect of stages was not the primary objective of this particular analysis. Estimation of individual slopes for each month is not possible with the current model configuration with the plot as a random factor. We believe that figures 2,3 in the section 3.1 have already captured phenological patterns from both ground truth LAI and other VIs. Stage-specific analysis was performed in section 3.3.
3.3. Stage-Specific Performance of Vegetation Indices
While stage-wise results are presented, additional statistical comparisons (e.g., significance testing or error analysis) would further strengthen the evaluation.
> For the stage-specific performance, we had only one RMSE value for the model from each month, making it difficult to produce significant testing between different VIs and months without further subsampling within our dataset.
3.4. Model Performance Comparison by Plot Types
In my view, using different validation procedures between plot types may bias the comparison. A consistent validation method and comparable confidence intervals for both groups would improve fairness.
> We acknowledge your concern about our current approach to compare the empirical RMSE from the plots without trees with the bootstrapping-derived RMSE distribution of the plots with trees. However, we believe that this approach allows us to use the maximum number of plots without trees (n = 19) for comparisons with the other type of plots. If we try to use the bootstrapping approach for both types of plots, we will have to sample for a smaller number of plots, resulting in even lower statistical power in this already relatively small sample size.
- Discussion
Expanding the discussion to include broader comparisons with existing literature would help clarify the general applicability of the findings.
> We have added comparisons of our results with other crop plants in the discussion.
- Conclusions
The practical implications of the proposed models are only briefly mentioned. Expanding on how the approach can be applied in real-world crop monitoring or decision-making would strengthen the overall impact of the study.
> we added more detailed practical implications in the conclusion.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe introduction provides a good overview of the topic; however, the authors are encouraged to clearly highlight the specific research gap and explicitly state the novelty of the proposed approach.
Recent and closely related studies in agricultural remote sensing should be discussed in more depth to better position the contribution of this work within existing literature.
The methodology section would benefit from additional details on data acquisition, including sensor/platform specifications, spatial and temporal resolution, and study area characteristics.
Please clarify the data preprocessing steps and parameter settings to improve reproducibility of the proposed approach.
If machine learning or statistical models are employed, the training, validation, and testing procedures should be described more clearly, including dataset split ratios and evaluation protocols.
The results are clearly presented; however, a more in-depth discussion comparing the obtained results with existing methods or benchmarks would strengthen the manuscript.
The practical implications of the findings for agricultural monitoring, management, or decision-making should be emphasized in the discussion section.
Some figures and tables require improved clarity (resolution, labeling, and captions) to ensure they are fully self-explanatory.
Minor grammatical issues and sentence-level clarity problems are present throughout the manuscript; careful language editing is recommended to enhance readability.
The conclusion section could be strengthened by briefly summarizing the main contributions and outlining future research directions.
Comments on the Quality of English LanguageThe manuscript is generally understandable; however, there are several grammatical issues, awkward sentence constructions, and inconsistencies in technical terminology. The authors are encouraged to carefully revise the manuscript to improve clarity, coherence, and readability. A thorough language editing by a fluent English speaker or a professional editing service is recommended to ensure that the scientific contributions are communicated more clearly.
Author Response
The introduction provides a good overview of the topic; however, the authors are encouraged to clearly highlight the specific research gap and explicitly state the novelty of the proposed approach.
> Thank you for your consideration of our manuscript. We have addressed all of the issues that you have raised in this manuscript. You might find our responses after “>” below each comment.
Recent and closely related studies in agricultural remote sensing should be discussed in more depth to better position the contribution of this work within existing literature.
> We added a short review of LAI studies in other crops in the introduction.
The methodology section would benefit from additional details on data acquisition, including sensor/platform specifications, spatial and temporal resolution, and study area characteristics.
> We have provided the relevant information in these sections: sensor/platform specifications, spatial and temporal resolution are detailed in the section 2.3, and study area characteristics are described in the section 2.1. We have gone through these sections again to ensure that we specified platform (Sentinel-2 data through GoogleEarth Engine), spatial resolution (10-m to 20-m depending on VIs), temporal resolution (Growing season of 2022), and details on soil and growing practices in study area. We welcome guidance on any aspects that the reviewers would like to see elaborated.
Please clarify the data preprocessing steps and parameter settings to improve reproducibility of the proposed approach.
> We have provided the JavaScript and R code for data acquisition and analysis here for the full reproducibility here https://github.com/fonnknp/Sentinel-2-derived-VIs-LAI-on-cassava-plots.
If machine learning or statistical models are employed, the training, validation, and testing procedures should be described more clearly, including dataset split ratios and evaluation protocols.
> In this particular study, we did not employ a machine learning method that requires splitting into testing and training dataset. Other testing procedures are already described in the section 2.4. All the codes used in this study are available in our GitHub repository here above.
The results are clearly presented; however, a more in-depth discussion comparing the obtained results with existing methods or benchmarks would strengthen the manuscript.
> We have added the comparison with cases from other crop plants in the manuscript.
The practical implications of the findings for agricultural monitoring, management, or decision-making should be emphasized in the discussion section.
> We have added more practical implication in our discussion and conclusion.
Some figures and tables require improved clarity (resolution, labeling, and captions) to ensure they are fully self-explanatory.
> We have gone through the manuscript to ensure that they are all self-explanatory. We would appreciate specific comments on any figures and tables that require further improvement.
Minor grammatical issues and sentence-level clarity problems are present throughout the manuscript; careful language editing is recommended to enhance readability.
> We have gone through the manuscript carefully to improve the clarity and resolve minor grammatical issues.
The conclusion section could be strengthened by briefly summarizing the main contributions and outlining future research directions.
> We have modified the conclusion accordingly.
Reviewer 3 Report
Comments and Suggestions for AuthorsI have no major concerns. This is a well-crafted manuscript.
Author Response
I have no major concerns. This is a well-crafted manuscript.
> Thank you very much for your positive feedback on our manuscript.
Reviewer 4 Report
Comments and Suggestions for AuthorsThe study evaluated Sentinel-2 Vegetation Indices for Estimating Leaf Area Index in Cassava Plots. The study seems interesting. Below are the comments:
- Why were 13 VIs estimated/calculated? What is the background for selecting particular VI. Justify the selection.
- How did you convert Sentinel-2 data into VIs. Give details about the software and process involved in the process.
- How does the Leaf Area Index were measured actually in the field? Explain in detail about the measurement procedure. Give details of instruments used. How was ground-level Leaf Area Index (ground-LAI) measured in the study using the SunScan Canopy Analyzer.
- What is the comparative performance of conventional and red-edge-based spectral indices from Sentinel-2 for monitoring cassava canopy structure?
- Plz check the range of red-edged bands used in the study.
- The soil texture is sandy loam with poor water retention. The soil pH ranges from strongly acidic to moderately acidic with low fertility. Give numerical values to all the factors.
- What were the steps involved in converting solar radiation measurements into LAI values?
- What preprocessing steps were applied to Sentinel-2 Level-2A data before analysis? Why were selected two Red Ege bands (RE1 and RE2).
- Which metrics were used to evaluate the performance of the LAI prediction models?
- Which vegetation indices showed the most consistent pattern with ground–LAI changes
- How do ground–LAI and vegetation indices relate to cassava growth stages during the growing period.
- Conclusion is not supported with numerical results. Please provide numerical value to support the conclusion derived in this study.
Author Response
The study evaluated Sentinel-2 Vegetation Indices for Estimating Leaf Area Index in Cassava Plots. The study seems interesting. Below are the comments:
> Thank you for your constructive feedback from our manuscript. We have now provided point-by-point responses from your comments below.
Comment 1: Why were 13 VIs estimated/calculated? What is the background for selecting particular VI. Justify the selection.
> The 13 VIs were selected to cover five key optical parameters relevant to LAI estimation in cassava: canopy greenness (NDVI, DVI, GNDVI, and GRVI), soil background correction (SAVI and EVI), chlorophyll absorption (CIG, TCARI, BNDVI, and VIG), canopy water content (NDWI), and Sentinel-2 red-edge sensitivity (SeLI and RVI). This multi-mechanism approach ensures robustness across cassava's full growth cycle. The additional justification has been added to the first paragraph of the Section 2.3.3.
Comment 2: How did you convert Sentinel-2 data into VIs. Give details about the software and process involved in the process.
> We have provided a more detailed explanation of the VI computation process in the second paragraph of Section 2.3.3.
Comment 3: How does the Leaf Area Index were measured actually in the field? Explain in detail about the measurement procedure. Give details of instruments used. How was ground-level Leaf Area Index (ground-LAI) measured in the study using the SunScan Canopy Analyzer.
> The detailed explanation of ground-LAI measurement using the SunScan Canopy Analyzer was described as in the Section 2.2.
Comment 4: What is the comparative performance of conventional and red-edge-based spectral indices from Sentinel-2 for monitoring cassava canopy structure?
> This study compared the performance of convention (NIR-based indices including, NDVI, DVI, GNDVI, GRVI, SAVI, EVI, BNDVI, CIG, NDWI, and VIG), against red-edge-based indices, including SeLI, RVI, and TCARI), for monitoring cassava canopy structure. Evidence from analogous broadleaf tropical crops suggests that Sentinel-2 red-edge indices generally demonstrate superior or at least more robust performance than conventional VIS–NIR indices for canopy structure variables such as LAI. However, direct cassava-specific studies remain limited, highlighting the importance of local calibration and validation as conducted in this study to confirm the applicability of these indices under cassava growing conditions.
Comment 5: Plz check the range of red-edged bands used in the study.
> The wavelenghs of the red-edge and other spectral bands used for the VIs computation have been verified against the Sentinel-2 Level-2A spectral band specifications. The Table2 has been update to reflect the accurate band designations.
Comment 6: The soil texture is sandy loam with poor water retention. The soil pH ranges from strongly acidic to moderately acidic with low fertility. Give numerical values to all the factors.
> We have already added numerical values for the soil data as highlighted in the red in the Section 2.1 Study site.
Comment 7: What were the steps involved in converting solar radiation measurements into LAI values?
> The SunScan Canopy Analyzer converted solar radiation measurements into LAI values through the following steps. First, radiation levels above and below the canopy were recode using photodiodes along a one-meter wand. Second, canopy transmittance was calculated as the ratio of below- to above-canopy radiation. Third, gap fraction, defined as the probability of light non-interception through the canopy, was derived from transmittance values. Finally, LAI was back-calculated from gap fraction using Wood’s (1996) SunScan equations which integrate Campbell’s (1989) ellipsoidal leaf angle distribution (ELADP) over the full sky hemisphere to account for both direct beam and diffuse radiation componets.
Campbell, G. S. 1989. Extinction coefficients for radiation in plant canopies using an ellipsoidal inclination angle distribution. Agricultural and Forest Meteorology, 36 (4): 317-321.
Wood, J. 1996. SunData software and canopy theory.
Comment 8: What preprocessing steps were applied to Sentinel-2 Level-2A data before analysis? Why were selected two Red Ege bands (RE1 and RE2).
> The preprocessing of Sentinel-2 Level-2A data involved selecting images with cloud coverage below 20%, using the Scene Classification Layer (SCL) for cloud masking to exclude cloud-contaminated pixels before computing the vegetation indices.
> Regarding the selection of both Red Ege bands (RE1 and RE2), as shown in Table 2, RVI uses RE2 (B6; 740 nm) and SeLI uses RE1 (B5; 704 nm), each index requires a different Red Edge band. Therefore, both RE1 and RE2 were necessary to compute these two indices. Furthermore, RE1 and RE2 capture complementary biophysical information throughout the canopy development, especially canopy becomes dense. RE1 is sensitive to chlorophyll content while RE2 is sensitive to canopy structure, both of which are directly related to LAI estimation.
Comment 9: Which metrics were used to evaluate the performance of the LAI prediction models?
> The model performance was evaluated using R2 and RMSE, which are described in the Section 2.4.
Comment 10: Which vegetation indices showed the most consistent pattern with ground–LAI changes
> We stated that GNDVI and NDWI had the most consistent pattern with ground-LAI changes. Based on your concern, we therefore chose only GNDVI as it had a pattern that is consistent in the same direction as ground-LAI.
Comment 11: How do ground–LAI and vegetation indices relate to cassava growth stages during the growing period.
> This has been addressed in the Section 4.1. The ground-LAI and vegetation indices showed similar temporal patterns corresponding to cassava growth stages, with rapid increases during active leaf development (2-3 MAP) and intensive root storage formation (5-7 MAP).
Comment 12: Conclusion is not supported with numerical results. Please provide numerical value to support the conclusion derived in this study.
> The conclusion has been updated to include numerical values to support all findings presented in this study.
Round 2
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors have done a good job revising and editing the manuscript.
