Three-Dimensional Spectral Index-Driven Nondestructive Quantification of Chlorophyll in Winter Wheat: Cross-Phenology Extrapolation and Independent Validation

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Editors and Authors,

 

I read with interest the manuscript entitled "Three-dimensional spectral index-Driven Nondestructive Quantification of Chlorophyll in Winter Wheat: Cross-Phenology Extrapolation and Independent Validation". This work aims to deliver innovative, efficient, and nondestructive methods for chlorophyll monitoring in winter wheat. Therefore, the manuscript needs some adjustments so that it can then be forwarded to the publication process. The manuscript needs the following adjustments:

 

ABSTRACT

 

- Lines 13–18: Three families of spectral indices and the use of three ML algorithms were presented, but the main theoretical contribution is unclear. Was it the creation of 3D OSI indices or their fusion with empirical indices? I suggest clarifying this here.

- Lines 27–29: The abstract concludes that 3D indices offer a new path, but it should inform practical innovation. How can this be applied to agricultural management? Add a practical conclusion. This was not done.

- Line 31: Replace keywords that are repeated in the title.

 

INTRODUCTION

 

- Lines 41–48: The justification for replacing destructive methods is reasonable, but could include other nondestructive sensors, such as fluorescence, RGB images, and others, as a comparison, showing why hyperspectral is more advantageous. - Lines 63–69: Regarding ML, SVM, BPNN, and RF were mentioned, but recent models, such as XGBoost, LightGBM, and deep learning CNNs/RNNs, which have already been applied in hyperspectral, were not mentioned. I suggest adding this to improve the review.

- Lines 77–82: The transition to proposing 3D indices is interesting, but it does not present a clear hypothesis. I suggest adding the scientific hypothesis. This should be added before mentioning the objectives, in the last paragraph.

- Lines 83–89: The justification for the elongation stage is interesting, but it could be better associated with agricultural management, such as nitrogen fertilisation decisions, among others.

 

MATERIALS AND METHODS

 

- Lines 100–105: The experimental design (5 N rates x 4 fertilisation methods) is complex, but the article does not discuss whether these differences were actually explored or whether they merely served to generate variability in the data. Was the design, therefore, a 5x4 factorial scheme? This should be clarified. Were the results interpreted by comparing the means as a factorial scheme?

- Lines 110–113: The ASD FieldSpec covers up to 2500 nm, but only up to 1830 nm was used. Why was 1830–2500 nm discarded? Regarding noise, there are relevant water and lignin bands in this range. Check information.

- Lines 114–115: I suggest specifying the spectral resolution, for example, 3 nm in VNIR, 10 nm in SWIR. This is important in hyperspectral analysis.

- Lines 119–127: The alcohol extraction method is standard, but the number of replicates or whether the spectrophotometric reading was taken in replicates was not mentioned. Add information.

- Lines 128–132: Only Savitzky–Golay smoothing was applied. Was there no atmospheric noise removal or additional corrections, such as normalisation or transformations?

- Lines 136–138: Only 10 classical indices were used as a baseline. I suggest further justifying this choice. There are many other indices, such as MCARI, PRI, and OSAVI, which are known to be related to chlorophyll.

- Lines 144–149: The explanation for the search for 3D combinations is interesting, but it would be important to specify how many indices were tested in total and whether overfitting was controlled.

- Lines 150–168: Only three models were tested. I suggest including at least one linear reference model, such as PLSR or LASSO, to demonstrate that the gain truly comes from the complexity of the methods.

- Lines 169–174: The sample size (n=66) is minimal for machine learning. The article partially acknowledges this, but I suggest reinforcing this limitation right here.

Was there no more robust cross-validation?

 

RESULTS

 

- Lines 192-198: Table 2 only contains specific empirical indices. Others known to have a strong relationship with chlorophyll were left out.

- Lines 205-207: The 3D index (DTSI) correlated 0.703, only slightly better than the 2D (DI, r = 0.686). This improvement needs to be discussed more carefully. The practical benefit may not be as significant. Improvement is needed.

- Lines 225-229: The RF+Input 5 model performed better (R² = 0.816), but it would be important to apply a statistical test to verify whether the difference with other models is significant. How can we confirm whether it was better or not if there was no comparison with other models? Consider the possibility of doing so.

- Lines 243–247: The model had R² = 0.897 but MRE = 21.8%, which is contradictory, meaning high fit and significant relative error. This needs to be discussed.

External validation was performed at the same experimental site, only at a different phenological stage. Does this guarantee spatial robustness? Suggest future evaluation in other environments and cultivars.

 

DISCUSSION

 

- Lines 272–288: The biophysical explanation of wavelengths is good, but could be more critical. Is the gain compared to 2D OSI truly substantial?

- Lines 295–299: Good discussion of spectral redundancy, but it fails to mention which automatic variable selection methods could address this.

- Lines 300–311: The superiority of RF is well explained, but the lack of more modern methods weakens the conclusion. I suggest mentioning this limitation. - Lines 313–323: The limitation of representation is correctly highlighted, but it would be helpful to discuss the practice.

 

CONCLUSIONS

I suggest including information such as applicability in agricultural monitoring systems, the need for validation across multiple regions and years, and the potential for incorporating 3D indices into drone/satellite sensors.

Author Response

Plants
Title: Three-dimensional spectral index-Driven Nondestructive Quantification of Chlorophyll in Winter Wheat: Cross-Phenology Extrapolation and Independent Validation

Dear Editor,

Response: Thank you very much for offering us an opportunity to revise the manuscript (agronomy-3903027), and we also thank the reviewers for giving us constructive comments and suggestions which have helped us improve the quality of the manuscript. We have now modified the manuscript according to the reviewer’s comments and suggestions. All modifications will not influence the content and framework of the manuscript. The following are point-to-point responses to the reviewers’ comments. We sincerely hope this manuscript will be acceptable and look forward to hearing from you soon.

 

 

Reviewer 1

 

I read with interest the manuscript entitled "Three-dimensional spectral index-Driven Nondestructive Quantification of Chlorophyll in Winter Wheat: Cross-Phenology Extrapolation and Independent Validation". This work aims to deliver innovative, efficient, and nondestructive methods for chlorophyll monitoring in winter wheat. Therefore, the manuscript needs some adjustments so that it can then be forwarded to the publication process. The manuscript needs the following adjustments:

 

Response: Thank you to Reviewer 1 for your detailed and constructive feedback on our revised manuscript. In response to your comments, we have addressed each point individually as follows.

 

 

ABSTRACT

 

- Lines 13–18: Three families of spectral indices and the use of three ML algorithms were presented, but the main theoretical contribution is unclear. Was it the creation of 3D OSI indices or their fusion with empirical indices? I suggest clarifying this here.

Response: Thank you for pointing this out. We have revised the relevant sections of the abstract as requested, explicitly stating that the core innovation lies in developing novel three-dimensional optimal spectral indices (3D OSIs) and integrating them with traditional vegetation indices. The abstract has been updated to: “The main contribution lies in devising novel 3D OSIs that combine three spectral bands and demonstrating how their fusion with classic two-band indices can improve chlorophyll quantification.”

 

 

- Lines 27–29: The abstract concludes that 3D indices offer a new path, but it should inform practical innovation. How can this be applied to agricultural management? Add a practical conclusion. This was not done.

Response: Thank you for pointing this out. We have added a final sentence to the abstract to meet the practical-conclusion requirement: “Altogether, integrating three-dimensional (3D) spectral indices with classical vegetation indices and deploying random forest (RF) enables accurate, nondestructive estimation of winter wheat chlorophyll, offering a new hyperspectral pathway for monitoring crop physiological status and strengthening precision agricultural management and fertilization, and guiding in-season fertilization to optimize nitrogen use, thereby advancing precision agriculture.”

 

- Line 31: Replace keywords that are repeated in the title.

Response: Thank you for pointing this out. We have made the requested revisions accordingly. (Keywords: Triticum aestivum L.; hyperspectral; machine learning; correlation matrix)

INTRODUCTION

 

- Lines 41–48: The justification for replacing destructive methods is reasonable, but could include other nondestructive sensors, such as fluorescence, RGB images, and others, as a comparison, showing why hyperspectral is more advantageous.

Response: Thank you for pointing this out. We have expanded our discussion of other nondestructive sensors as follows: “Previous studies indicate that, while fluorescence and RGB imaging are useful, they have limitations relative to hyperspectral approaches. Hyperspectral imaging provides much richer spectral information—spanning the visible, near-infrared (NIR), and short-wave infrared (SWIR)—than simple RGB cameras. RGB imagery is easier to acquire but captures only broad color bands, whereas hyperspectral sensors resolve narrowband features linked to pigments, leaf structure, and water absorption. Likewise, fluorescence sensors (e.g., chlorophyll fluorometers) quantify photosynthetic activity but typically require special conditions (e.g., dark adaptation) and measure fundamentally different signals. By contrast, hyperspectral reflectance naturally covers the key absorption and scattering regions most directly related to chlorophyll content (visible for pigments, NIR for leaf structure, SWIR for water), which explains why hyperspectral methods can achieve higher accuracy.”

 

- Lines 63–69: Regarding ML, SVM, BPNN, and RF were mentioned, but recent models, such as XGBoost, LightGBM, and deep learning CNNs/RNNs, which have already been applied in hyperspectral, were not mentioned. I suggest adding this to improve the review.

Response: Thank you for pointing this out. We have expanded our discussion of recent machine learning models as follows: “Moreover, support vector machines (SVM), back-propagation neural networks (BPNN), random forests (RF), eXtreme Gradient Boosting (XGBoost), convolutional neural networks (CNN), long short-term memory (LSTM) networks, and extreme learning machines (ELM) have been widely employed to estimate crop parameters because they can accommodate the high dimensionality and nonlinearity of spectral data [13,14]. For example, Ju et al. (2024) [48] identified an optimal pipeline for predicting rice leaf chlorophyll content from fluorescence spectral data—CNN+LSTM combined with IVSO–IVISSA. These and related studies indicate that ML methods—most commonly XGBoost, SVM, and RF—can effectively predict chlorophyll content. However, the constraints imposed by the choice and quality of input variables in these ML-based studies should not be overlooked.”

 

- Lines 77–82: The transition to proposing 3D indices is interesting, but it does not present a clear hypothesis. I suggest adding the scientific hypothesis. This should be added before mentioning the objectives, in the last paragraph.

Response: Thank you for pointing this out. We have added the requested content: “We hypothesize that integrating spectral information from three wavelengths into a three-dimensional index can capture physiologically relevant features that two-dimensional indices miss, thereby improving predictive accuracy. Accordingly, during 2018–2020 we conducted two consecutive growing-season experiments in typical winter wheat fields on the Loess Plateau, systematically collecting canopy hyperspectral reflectance at the jointing stage together with destructive chlorophyll measurements. We first computed classical empirical vegetation indices, then used a correlation-matrix approach to construct two- and three-dimensional spectral indices and quantify the sensitivity of all spectral parameters to chlorophyll. Next, we combined these spectral parameters—with the three-dimensional indices as the primary predictors—to build chlorophyll retrieval models (random forest, RF; back-propagation neural network, BPNN; and support vector machine, SVM) and identified the optimal monitoring scheme. Finally, we assessed feasibility using observations from the grain-filling stage in 2020. This study aims to provide an innovative, efficient, and nondestructive method for monitoring winter wheat chlorophyll.”

 

- Lines 83–89: The justification for the elongation stage is interesting, but it could be better associated with agricultural management, such as nitrogen fertilisation decisions, among others.

Response: Thank you for pointing this out. We have added the requested content: “As a critical phase in the winter wheat growth cycle, the jointing stage is characterized by rapid internode elongation and the initiation of floral primordia. Concurrently, crop demand for water, nutrients, and light surges; thus even small fluctuations in chlorophyll closely reflect physiological status and nutritional level [17]. Monitoring chlorophyll during jointing enables in-season fertilization adjustments [ref]. Therefore, accurate and timely acquisition of chlorophyll content at jointing is essential for assessing growth status and provides a scientific basis for refined field management and precision fertilization.”

 

 

 

MATERIALS AND METHODS

 

- Lines 100–105: The experimental design (5 N rates x 4 fertilisation methods) is complex, but the article does not discuss whether these differences were actually explored or whether they merely served to generate variability in the data. Was the design, therefore, a 5x4 factorial scheme? This should be clarified. Were the results interpreted by comparing the means as a factorial scheme?

Response: Thank you for pointing this out. This was a 5×4 factorial experiment (five nitrogen rates × four application methods) with multiple replicates. We explicitly state that the factorial design was used to induce broad variation in canopy chlorophyll, rather than to analyze treatment effects. This clarification has been added to the main text: “The factorial design—five N levels crossed with four application methods, with multiple replicates—was chosen to create broad chlorophyll variation. We verified that treatments produced significantly different chlorophyll means (ANOVA; results not shown), but we did not focus on comparing treatment means; instead, all observations were pooled for model development.”

 

- Lines 110–113: The ASD FieldSpec covers up to 2500 nm, but only up to 1830 nm was used. Why was 1830–2500 nm discarded? Regarding noise, there are relevant water and lignin bands in this range. Check information.

Response: Thank you for pointing this out. We have added the following explanation in the main text for this choice: “An ASD FieldSpec™ 3 spectroradiometer (Analytical Spectral Devices, Boulder, CO, USA) was used to collect reflectance spectra from 350–1830 nm. Although the instrument spans 350–2500 nm, the 1830–2500 nm region was excluded because pronounced atmospheric water-vapor absorption—particularly around 1800–1900 nm—reduces the signal-to-noise ratio and yields noisy, unreliable reflectance. Accordingly, SWIR wavelengths beyond 1830 nm were discarded to improve data quality.”

 

 

- Lines 114–115: I suggest specifying the spectral resolution, for example, 3 nm in VNIR, 10 nm in SWIR. This is important in hyperspectral analysis.

Response: Thank you for pointing this out. We have added the following specification in the main text: “The ASD FieldSpec™ 3 provides ~3 nm spectral resolution (full width at half maximum, FWHM) in the VNIR and ~10 nm in the SWIR.”

 

- Lines 119–127: The alcohol extraction method is standard, but the number of replicates or whether the spectrophotometric reading was taken in replicates was not mentioned. Add information.

Response: Thank you for pointing this out. We have added the following to the main text: “Each canopy chlorophyll determination was performed in triplicate, and the mean value was used for analysis. For the spectrophotometric assay, absorbance was measured three times per extract and the readings were averaged to reduce analytical error.”

 

 

- Lines 128–132: Only Savitzky–Golay smoothing was applied. Was there no atmospheric noise removal or additional corrections, such as normalisation or transformations?

Response: Thank you for pointing this out. We have added the following to the main text: “On each sampling day, we calibrated the reflectance spectra using a Spectralon® white reference panel to correct for illumination changes. Raw radiance measurements were converted to reflectance with the Spectralon reference before any smoothing. The reflectance spectra were then processed with Savitzky–Golay (SG) smoothing (second-order polynomial, 9-band moving window) to suppress high-frequency noise and minor instrumental drift while preserving key spectral features, thereby improving the signal-to-noise ratio (SNR) for parameter extraction. No additional normalization (e.g., continuum removal or standard normal variate, SNV) was applied; future work may explore these preprocessing options.”

 

 

- Lines 136–138: Only 10 classical indices were used as a baseline. I suggest further justifying this choice. There are many other indices, such as MCARI, PRI, and OSAVI, which are known to be related to chlorophyll.

Response: Thank you for pointing this out. Our empirical index set intentionally focuses on broadly validated, crop- and context-robust generic vegetation indices. The aim is not to “exhaust every chlorophyll index,” but to provide a comparable, reusable baseline against which to assess the incremental value of our new 2D/3D optimal spectral indices (OSIs). Given the current sample size and experimental scale, indiscriminately expanding the index library could introduce multiple-comparisons bias and redundant multicollinearity. We have clarified this in the Discussion (Limitations and Outlook): “Using robust, comparable generic empirical indices as a baseline, this study evaluates the incremental value of 3D OSIs under limited-sample conditions. At the current scale, indiscriminate expansion of the index library risks multiple comparisons and redundant collinearity; therefore, not all chlorophyll indices were included. In subsequent work with larger samples and multiple sites, we will systematically add extended indices (e.g., MCARI, OSAVI, PRI) and assess their relative contributions using feature-selection methods with FDR correction.”

 

 

- Lines 144–149: The explanation for the search for 3D combinations is interesting, but it would be important to specify how many indices were tested in total and whether overfitting was controlled.

Response: Thank you for pointing this out. In the Methods we now quantify the search space and clarify why this process does not constitute model overfitting: “The discrete wavelength set spans 350–1830 nm with 1,481 wavelength points; an exhaustive scan of all three-band combinations therefore comprises 1,481³ (≈3.25×10⁹) triplets. For each candidate triplet, we computed its Pearson correlation with chlorophyll and used parallel computing for acceleration. This step is a univariate sensitivity enumeration that involves no model parameter fitting; selection bias was subsequently controlled via independent validation and cross-validation.”

 

 

- Lines 150–168: Only three models were tested. I suggest including at least one linear reference model, such as PLSR or LASSO, to demonstrate that the gain truly comes from the complexity of the methods.

Response: Thank you for pointing this out. We have incorporated partial least squares regression (PLSR) as a linear baseline, specified its configuration and evaluation in the Methods, and added a comparative discussion. Specifically, the number of latent variables (LVs) was selected via 10-fold cross-validation using the minimum root-mean-square error of cross-validation (RMSECV) criterion. A search over 1–10 components yielded three LVs, which explained 87% of the spectral variance and mitigated multicollinearity. In each fold, predictive R² and bias were monitored to ensure a stable fit without over-simplification.

Due to its linearity assumption and the construction of latent variables by maximizing covariance, PLSR struggles to capture the nonlinear/saturating behavior and high-er-order interactions in the chlorophyll–spectral relationship; under strong collinearity and noise it is sensitive to the number of components and prone to underfitting, there-by yielding lower overall fitting accuracy.

 

 

- Lines 169–174: The sample size (n=66) is minimal for machine learning. The article partially acknowledges this, but I suggest reinforcing this limitation right here.

Was there no more robust cross-validation?

Response: Thank you for pointing this out. We acknowledge this limitation and have added remedies in the Discussion: “Regarding sample size and external validation, our current sample size is limited, and same-site, cross-phenology external tests only suggest transferability across certain periods, which is insufficient to demonstrate spatial robustness. Owing to the small training set, we did not implement leave-one-year-out (LOYO) validation. To preserve training stability, we instead used stratified random splits, random forest out-of-bag (OOB) estimates, and phenology-specific external validation to curb overfitting. Going forward, we will expand the number of years and plots, include multiple cultivars/soils/climates, and conduct cross-year and cross-site external validation to systematically assess spatiotemporal generalization.”

 

 

 

RESULTS

 

- Lines 192-198: Table 2 only contains specific empirical indices. Others known to have a strong relationship with chlorophyll were left out.

Response: Thank you for pointing this out. Our empirical spectral index set intentionally focuses on broadly validated, cross-crop, and cross-scenario generic vegetation indices. The goal is not to “exhaust every chlorophyll index,” but to provide a comparable, reusable baseline for assessing the incremental value of the new 2D/3D optimal spectral indices (OSIs). Given the current sample size and experimental scale, indiscriminately expanding the index library could introduce multiple-comparisons bias and redundant multicollinearity. We have clarified this in the Discussion (Limitations and Outlook): “Using robust, comparable generic empirical indices as a baseline, this study aims to evaluate the incremental value of 3D OSIs under limited-sample conditions. At the current scale, indiscriminate expansion of the index library risks multiple comparisons and redundant collinearity; therefore, not all chlorophyll indices were included. In subsequent work with larger samples and multiple sites, we will systematically add extended indices (e.g., MCARI, OSAVI, PRI) and assess their relative contributions using feature-selection methods with FDR correction.”

 

- Lines 205-207: The 3D index (DTSI) correlated 0.703, only slightly better than the 2D (DI, r = 0.686). This improvement needs to be discussed more carefully. The practical benefit may not be as significant. Improvement is needed.

Response: Thank you for pointing this out. We agree that improvements in correlation from any single index are modest. Accordingly, we have softened the wording in the Results and Discussion and now emphasize that the key benefit of our approach stems from the complementary information provided by combining 3D OSIs with empirical indices in machine-learning models: “Although a standalone 3D index yields only limited correlation gains over 2D indices, fusing it with empirical indices within ML models produces a substantially larger improvement in out-of-sample generalization, indicating that the advantages of the 3D OSI are realized primarily at the model level through information complementarity.”

 

 

 

- Lines 225-229: The RF+Input 5 model performed better (R² = 0.816), but it would be important to apply a statistical test to verify whether the difference with other models is significant. How can we confirm whether it was better or not if there was no comparison with other models? Consider the possibility of doing so.

Response: Thank you for pointing this out. We have added a comparison table in the revised manuscript to assess the statistical significance of the differences (Table 4).

 

- Lines 243–247: The model had R² = 0.897 but MRE = 21.8%, which is contradictory, meaning high fit and significant relative error. This needs to be discussed. External validation was performed at the same experimental site, only at a different phenological stage. Does this guarantee spatial robustness? Suggest future evaluation in other environments and cultivars.

Response: Thank you for pointing this out. R² and mean relative error (MRE) are not inherently negatively correlated; when the response range is narrow or errors are predominantly proportional, one may observe a relatively high R² alongside comparatively large relative errors. This is especially likely when the sample size is small and covers only a single growth stage. In this study, the purpose of the external validation was to assess feasibility across phenological stages, which is not equivalent to spatial extrapolation. Regarding “spatial robustness,” we will state in the Discussion (Outlook) our plan for follow-up evaluations across regions, cultivars, and years.

 

 

 

 

DISCUSSION

 

- Lines 272–288: The biophysical explanation of wavelengths is good, but could be more critical. Is the gain compared to 2D OSI truly substantial?

Response: Thank you for pointing this out. We agree that correlation improvements from any single index are limited. Accordingly, we have tempered the language in the Results and Discussion and emphasize that the key benefit comes from combining 3D OSIs with empirical indices within machine-learning models: “Although a standalone 3D index yields only limited correlation improvements over 2D indices, fusing it with empirical indices within ML models produces a substantially larger gain in out-of-sample generalization, indicating that the advantages of the 3D OSI are realized primarily at the model level through complementary information.”

 

 

- Lines 295–299: Good discussion of spectral redundancy, but it fails to mention which automatic variable selection methods could address this.

Response: Thank you for pointing this out. We have added the following content to the Discussion (Outlook): “Regarding sample size and external validation, our current sample size is small, and same-site, cross-phenology external tests only suggest transferability across certain periods, which is insufficient to demonstrate spatial robustness. Owing to the limited training set, we did not adopt LOYO validation; to maintain training stability, we instead used stratified random splits, random forest (RF) out-of-bag (OOB) estimates, and phenology-specific external validation to curb overfitting. Going forward, we will expand the number of years and plots, include multiple cultivars/soils/climates, and conduct cross-year and cross-site external validation to systematically assess spatiotemporal generalization. To alleviate spectral redundancy, we will explore LASSO/Elastic Net, recursive feature elimination (RFE)/Boruta, mutual-information screening, and automated feature-selection approaches such as the successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), and Monte Carlo uninformative variable elimination (MC-UVE). On the sensing and application side, the sensitivity advantage of 3D OSIs entails certain computational demands and requirements for spectral bandwidth/precision; large-area deployment must account for viewing/illumination geometry, soil background, and atmospheric conditions.”

 

- Lines 300–311: The superiority of RF is well explained, but the lack of more modern methods weakens the conclusion. I suggest mentioning this limitation. - Lines 313–323: The limitation of representation is correctly highlighted, but it would be helpful to discuss the practice.

Response: Thank you for pointing this out. We have added the following to the Discussion (Outlook): “Future work will primarily leverage UAV hyperspectral data, with satellite multispectral imagery providing spatiotemporal supplementation. We will generate spatiotemporal chlorophyll maps during the critical jointing–booting window and integrate them with field crop models to support experiments and variable-rate nitrogen application. In addition, we will explore automated index generation or moderately deep feature extraction, coupled with physical models (e.g., PROSPECT/PROSAIL), to reduce reliance on manual screening and improve transferability.”

 

CONCLUSIONS

I suggest including information such as applicability in agricultural monitoring systems, the need for validation across multiple regions and years, and the potential for incorporating 3D indices into drone/satellite sensors.

Response: Thank you for pointing this out. We have added the following to the Conclusions: “The proposed ‘empirical indices + 3D OSI + RF’ workflow is highly feasible and well suited for integration into routine UAV field-scouting workflows to generate chlorophyll maps and support in-season nitrogen optimization and precision management. Future work will conduct systematic external validation across multiple regions, years, cultivars, and management scenarios to comprehensively assess spatiotemporal generalization and robustness. At the sensor level, UAV hyperspectral platforms can implement the 3D indices directly; for satellite multispectral data, progressive operationalization can be achieved via approximate bandwidth combinations or customized narrow bands.”

 

 

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Three-dimensional spectral index-Driven Nondestructive Quantification of Chlorophyll in Winter Wheat: Cross-Phenology Extrapolation and Independent Validation

2025/09/29

General Comments:

In this study, canopy hyperspectral reflectance together with destructive chlorophyll assays were systematically acquired from Yangling field trials conducted during 8–0. Three families of spectral indices were devised: classical empirical indices; two-dimensional optimal spectral indices (2D      OSI) selected by correlation-matrix screening; and novel three-dimensional optimal spectral indices (3D OSI). Altogether, integrating 3D spectral indices with classical vegetation indices and deploying   RF enabled accurate, nondestructive estimation of winter wheat chlorophyll, offering a new hyperspectral pathway for monitoring crop physiological status and advancing precision agricultural management and fertilization. Following considerable revision, the journal may consider the manuscript. Please find below some comments that will improve the paper. Addressing the following points will significantly enhance the overall quality and impact of the paper.

Comments and questions to improve:

(1) Introduction can be further improved—for example, previous most recent studies that compare similar methods, their features and their outcomes.

(2) Figures appear to be of quite bad quality. The text font size in some figures (Figures 3, 4, and 6) is too small to be read clearly.  Clarity to be improved in the figures also (Figures 3, 4, and 6).

(3) Please explain the innovation of this article, especially the integration of three-dimensional spectral indices with classical empirical vegetation indices, as the author states that RF is employed to achieve high-accuracy, non-destructive estimation of winter wheat chlorophyll content.  The authors should describe this in detail.

(4) What are the development prospects of the proposed method ( Three-dimensional spectral index-driven Nondestructive Quantification of Chlorophyll in Winter Wheat)? How is this technique advancing the yield estimation compared to others?

(5) The author should provide a more detailed explanation in the conclusion section, the directions of further research should be mentioned in the conclusion section.

Author Response

Plants
Title: Three-dimensional spectral index-Driven Nondestructive Quantification of Chlorophyll in Winter Wheat: Cross-Phenology Extrapolation and Independent Validation

Dear Editor,

Response: Thank you very much for offering us an opportunity to revise the manuscript (agronomy-3903027), and we also thank the reviewers for giving us constructive comments and suggestions which have helped us improve the quality of the manuscript. We have now modified the manuscript according to the reviewer’s comments and suggestions. All modifications will not influence the content and framework of the manuscript. The following are point-to-point responses to the reviewers’ comments. We sincerely hope this manuscript will be acceptable and look forward to hearing from you soon.

 

 

Reviewer2

Comments and Suggestions for Authors

Three-dimensional spectral index-Driven Nondestructive Quantification of Chlorophyll in Winter Wheat: Cross-Phenology Extrapolation and Independent Validation

2025/09/29

General Comments:

In this study, canopy hyperspectral reflectance together with destructive chlorophyll assays were systematically acquired from Yangling field trials conducted during 8–0. Three families of spectral indices were devised: classical empirical indices; two-dimensional optimal spectral indices (2D      OSI) selected by correlation-matrix screening; and novel three-dimensional optimal spectral indices (3D OSI). Altogether, integrating 3D spectral indices with classical vegetation indices and deploying   RF enabled accurate, nondestructive estimation of winter wheat chlorophyll, offering a new hyperspectral pathway for monitoring crop physiological status and advancing precision agricultural management and fertilization. Following considerable revision, the journal may consider the manuscript. Please find below some comments that will improve the paper. Addressing the following points will significantly enhance the overall quality and impact of the paper.

Response: Thank you to Reviewer 1 for your detailed and constructive feedback on our revised manuscript. In response to your comments, we have addressed each point individually as follows.

 

 

Comments and questions to improve:

• Introduction can be further improved—for example, previous most recent studies that compare similar methods, their features and their outcomes.

Response: Thank you for the suggestion. We have added a comparative review paragraph to the Introduction that contrasts recent approaches, their features, and reported results: “Previous studies indicate that, while fluorescence and RGB imaging are useful, they have limitations relative to hyperspectral approaches. Hyperspectral imaging provides richer spectral information—covering the visible, near-infrared (NIR), and short-wave infrared (SWIR)—than simple RGB cameras. RGB imagery is easier to acquire but captures only broad color bands, whereas hyperspectral sensors resolve narrowband features related to pigments, leaf structure, and water absorption. Likewise, fluorescence sensors (e.g., chlorophyll fluorometers) quantify photosynthetic activity but typically require special conditions (e.g., dark adaptation) and measure different signals. Hyperspectral reflectance inherently covers the key absorption and scattering regions most directly related to chlorophyll content (visible for pigments, NIR for leaf structure, SWIR for water), which helps explain the higher accuracy often achieved by hyperspectral methods. Moreover, support vector machines (SVM), back-propagation neural networks (BPNN), random forests (RF), eXtreme Gradient Boosting (XGBoost), convolutional neural networks (CNN), long short-term memory (LSTM) networks, and extreme learning machines (ELM) have been widely employed to estimate crop parameters because they can accommodate the high dimensionality and nonlinearity of spectral data [13,14]. For example, Ju et al. (2024) [48] identified an optimal pipeline for predicting rice leaf chlorophyll content from fluorescence spectral data—CNN+LSTM combined with IVSO–IVISSA. Collectively, these studies show that ML methods—most commonly XGBoost, SVM, and RF—can effectively predict chlorophyll content.”

 

• Figures appear to be of quite bad quality. The text font size in some figures (Figures 3, 4, and 6) is too small to be read clearly.  Clarity to be improved in the figures also (Figures 3, 4, and 6).

Response: Thank you for the suggestion. We have redrawn Figures 3, 4, and 6 to ensure a consistent style and to improve their resolution and overall figure quality.

 

• Please explain the innovation of this article, especially the integration of three-dimensional spectral indices with classical empirical vegetation indices, as the author states that RF is employed to achieve high-accuracy, non-destructive estimation of winter wheat chlorophyll content.  The authors should describe this in detail.

Response: Thank you for the suggestion. We have rewritten the final paragraph of the Introduction to better highlight the innovation (novel spectral indices + input fusion + machine learning): “We hypothesize that integrating spectral information from three wavelengths into a three-dimensional index can capture physiologically relevant features that two-dimensional indices miss, thereby improving predictive accuracy. Accordingly, during 2018–2020 we conducted two consecutive growing-season experiments in typical winter wheat fields on the Loess Plateau, systematically collecting canopy hyperspectral reflectance at the jointing stage together with destructive chlorophyll measurements. We first computed classical empirical vegetation indices, then used a correlation-matrix approach to construct two- and three-dimensional spectral indices and quantify the sensitivity of all spectral parameters to chlorophyll. Next, we combined these spectral parameters—with the three-dimensional indices as the primary predictors—to build chlorophyll retrieval models (random forest, RF; back-propagation neural network, BPNN; and support vector machine, SVM) and identified the optimal monitoring scheme. Finally, we assessed feasibility using observations from the grain-filling stage in 2020. This study aims to provide an innovative, efficient, and nondestructive method for monitoring winter wheat chlorophyll.”

 

• What are the development prospects of the proposed method (Three-dimensional spectral index-driven Nondestructive Quantification of Chlorophyll in Winter Wheat)? How is this technique advancing the yield estimation compared to others?

Response: Thank you for the suggestion. We have added a brief application outlook to the Discussion: “The relationships among chlorophyll, plant nitrogen status, photosynthetic activity, and yield are generally stable, and our results indicate that 3D OSIs provide complementary spectral information to classical empirical indices. From an applications standpoint, a prudent strategy is to rely primarily on UAV hyperspectral sensing, with satellite multispectral imagery supplying spatiotemporal coverage. Future work will primarily leverage UAV hyperspectral data, with satellite multispectral imagery providing spatiotemporal supplementation. We will generate spatiotemporal chlorophyll maps during the critical jointing–booting window and integrate them with field crop models to support experiments and variable-rate nitrogen application. In addition, we will explore automated index generation or moderately deep feature extraction, coupled with physical models (e.g., PROSPECT/PROSAIL), to reduce reliance on manual screening and improve transferability.”

 

 

• The author should provide a more detailed explanation in the conclusion section, the directions of further research should be mentioned in the conclusion section.

Response: Thank you for the suggestion. We have added the requested content to the Conclusions: “The proposed ‘empirical indices + 3D OSI + RF’ workflow is highly feasible and well-suited for integration into routine UAV field-scouting workflows to generate chlorophyll maps and support in-season nitrogen optimization and precision management. Future work will conduct systematic external validation across multiple regions, years, cultivars, and management scenarios to comprehensively assess spatiotemporal generalization and robustness. At the sensor level, UAV hyperspectral platforms can implement the 3D indices directly; for satellite multispectral data, progressive operationalization can be achieved via approximate bandwidth combinations or customized narrow bands.”

 

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript has been corrected. Organize the numbering of the new references inserted.

Author Response

Thank you for your suggestion. We have changed the order of references as required.

Reviewer 2 Report

Comments and Suggestions for Authors

The paper contains some errors, detracting from its readability and professionalism. For example, the reference list is not ordered numerically in the sequence they appear in the text. Hence,  I suggest some minor adjustments to enhance the paper's quality further.

Author Response

Thank you for your suggestion. We have changed the order of references as required.