Response to Reviewer 1 Comments

1. Summary

Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files. [This is only a recommended summary. Please feel free to adjust it. We do suggest maintaining a neutral tone and thanking the reviewers for their contribution although the comments may be negative or off-target. If you disagree with the reviewer’s comments please include any concerns you may have in the letter to the Academic Editor.]

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Yes

The background was refined to emphasize potato canopy vertical LNC heterogeneity and the study scope.

Are all the cited references relevant to the research?

Can be improved

Relevant references and crop-comparison context were checked and clarified.

Is the research design appropriate?

Yes

No fundamental design change was required; the sampling strategy and study boundary were clarified.

Are the methods adequately described?

Can be improved

Section 2.2 was expanded to define Top, Middle, and Bottom leaves in multi-stem potato plants.

Are the results clearly presented?

Can be improved

The Results were clarified to focus on layer-specific LNC differences and related spectral responses.

Are the conclusions supported by the results?

Can be improved

The Conclusions were revised to include limitations, potential significance, and application boundaries.

3. Point-by-point response to Comments and Suggestions for Authors

The manuscript "Hyperspectral Estimation of Layer-Specific Leaf Nitrogen Content in Potato Canopy by Integrating Fractional-Order Derivatives and Three-Dimensional Spectral Indices" by Ming Jin, Liaoyuan Ma, Liang Cheng, Zhiying Liu, Zijun Tang, Wangyang Li, Ruiqi Du, Tao Sun, Youzhen Xiang, and Fucang Zhang discusses an original approach to assessing different layers of potato plants using hyperspectral imaging to determine nitrogen distribution within the plant, with potential applications to agronomy.
The manuscript is formatted according to the rules and contains all required sections.
Minor comments that should be addressed during revision.
Potato plants often have multiple stems and their development is uneven. Please provide a detailed description of the relationship of leaves to a particular layer. I find that the description of physiological differences presented in the article does not fully reflect the actual physiology of nitrogen distribution in aging and shaded leaves. Furthermore, I don't see any comparisons between plants in the vegetative growth stage, those entering flowering, and those after fruiting, when maximum tuber growth is observed, which is crucial for this crop, as we consume the tubers, not the tops. Please expand.

Response: Thank you for these constructive comments and for recognizing the potential of this study. We agree that potato plants commonly have multiple stems and uneven shoot development, and that the relationship between leaf position and canopy layer should be described more clearly. Therefore, we revised Section 2.2 to clarify that the Top, Middle, and Bottom layers were defined according to the relative vertical position of leaves within the whole plant canopy, rather than according to the leaf order of a single stem. We also clarified the physiological characteristics of leaves in each layer, including illumination condition, leaf age, shading, and early senescence.

We also agree that the physiological interpretation of nitrogen distribution in aging and shaded leaves needed to be expanded. Therefore, we revised the Discussion to explain that the lower LNC in Bottom leaves may result from the combined effects of reduced light interception, leaf aging, early senescence, and nitrogen remobilization during tuber formation and early tuber bulking. We further clarified that the spectral information related to lower-layer LNC should be interpreted as indirect information contained in mixed top-of-canopy reflectance, rather than as a direct optical signal from lower leaves.

Regarding growth stages, this study focused on the tuber formation to early tuber bulking period, when canopy closure, source-sink regulation, and tuber development are important for potato production. We did not compare vegetative growth, flowering, tuber bulking, and senescence stages in the present dataset. Therefore, we added this point as a limitation and future research direction in the Conclusion. We also expanded the Conclusion to discuss the potential of the proposed framework for understanding canopy nitrogen allocation, lower-leaf physiological status, and vertical canopy function, while emphasizing that further validation across growth stages, cultivars, production regions, and UAV or satellite hyperspectral platforms is still required.

Revision made: We revised Section 2.2 to clarify the relationship between leaf position and canopy layer in multi-stem potato plants. We expanded the Discussion to explain the physiological meaning of nitrogen distribution in aging and shaded lower leaves. We also revised the Conclusion to explicitly address the limitations of the present study and the potential application of the proposed framework.

Location: Section 2.2, Discussion, and Conclusions.

4. Response to Comments on the Quality of English Language

Point 1:The English is fine and does not require any improvement.

Response 1: Thank you for the evaluation. We checked the manuscript and made minor language and terminology edits where necessary.

Response to Reviewer 2 Comments

1. Summary

Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files. [This is only a recommended summary. Please feel free to adjust it. We do suggest maintaining a neutral tone and thanking the reviewers for their contribution although the comments may be negative or off-target. If you disagree with the reviewer’s comments please include any concerns you may have in the letter to the Academic Editor.]

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Yes

The Introduction was revised to clarify that the study is a methodological and canopy-physiological assessment rather than an immediate fertilization-management tool.

Are all the cited references relevant to the research?

Can be improved

All in-text citations and reference numbering were checked and corrected where necessary.

Is the research design appropriate?

Can be improved

The experimental unit, plot replicates, sampling window, and year-independent validation were clarified.

Are the methods adequately described?

Must be improved

Sections 2.1-2.3 and 2.9 were expanded with details on the field experiment, equipment, sampling, preprocessing, and validation.

Are the results clearly presented?

Can be improved

Sections 3.1-3.4 were condensed; figure numbering and model-performance descriptions were revised.

Are the conclusions supported by the results?

Can be improved

The Conclusion was rewritten to explain the significance, limitations, and operational boundary of the findings.

3. Point-by-point response to Comments and Suggestions for Authors

The manuscript describes the results of two field experiments with one time of measurement each. The scientific topic of characterizing the vertical gradient in the leaf nitrogen content of potato canopies is interesting and relevant for the journal, however, there are some problems which have to be solved before publication of the manuscript is possible. It is not clear whether the authors want to make a contribution to basic sciences – studying the nitrogen distribution within potato canopies depending on the N level available – or whether they want to develop a management tool – hyperspectral reflectance for measuring the N status – for potato production.

Response: Thank you for this important comment. We have revised the Abstract, Introduction, and Conclusions to clarify that the study is primarily a methodological and canopy-physiological assessment rather than an immediately applicable fertilization-decision tool. Specifically, we clarified that the objective is to evaluate whether canopy-level hyperspectral reflectance contains sufficient information to indirectly estimate layer-specific LNC in potato canopies. We also emphasized the methodological reference of the framework and added limitations regarding single-site, single-cultivar, and single-measurement-period design, noting the need for further validation across growth stages, cultivars, ecological regions, fertilization regimes, and remote-sensing platforms.

L36-41: These results suggest that FOD1.5-integrated three-band optimized spectral indices can improve the indirect estimation of layer-specific LNC from canopy reflectance, particularly for lower canopy leaves whose spectral information is attenuated and mixed within the top-of-canopy signal. The findings provide a methodological reference for canopy vertical nitrogen diagnosis and functional assessment, while their operational use in fertilization management requires further validation across growth stages, cultivars, sites, and remote-sensing platforms.

L205-210: This study provides methodological support for the refined diagnosis of vertical canopy nitrogen heterogeneity and offers a reference for canopy functional assessment in semi-arid potato systems. However, its application to operational fertilization management requires further validation under different growth stages, cultivars, ecological regions, and fertilization regimes.

L889-896: Therefore, the proposed framework should be regarded as a methodological reference for vertical canopy nitrogen diagnosis and canopy functional assessment, rather than as an immediately applicable fertilization-management tool. This is because this study was conducted at one site, with one cultivar, and with one measurement period per year under a pre-planting in-furrow fertilization system, the results should not yet be generalized as a direct tool for in-season fertilization decisions. Further validation across growth stages, cultivars, ecological regions, fertilization regimes, and UAV or satellite hyperspectral platforms is required before operational application.

The chapter Materials and Methods is incomplete – important information on the field experiments and technical equipment is missing. The growth stage ‘tuber formation’, e.g. covers a period of several weeks (ranging from growth stage GS 40 to GS 49); more detailed information is needed. The informative value of the two-years study suffers from the lack of treatment replicates per experiment.

Response: Thank you for this helpful comment. We agree that the original Materials and Methods section did not provide sufficient detail on the field experiment, growth stage, technical equipment, and experimental unit. We have revised Sections 2.1, 2.2, and 2.3 accordingly.

First, we clarified in Section 2.1 that the same treatment layout and three plot replicates were maintained in both experimental years, and that the plot was used as the experimental unit for subsequent LNC measurement and canopy spectral analysis. This revision was added to clarify that the field experiment included replicated plots for each treatment combination.

Second, we revised the growth-stage description in Section 2.2. Instead of only using the broad term “tuber formation stage”, we now specify that leaf sampling was conducted during the tuber formation to early tuber bulking period. We also added the sampling dates and corresponding days after planting: 7 July 2022 and 8 July 2023, corresponding to approximately 63 and 68 days after planting, respectively. We further clarified that the measurements represented a defined phenological window in each year rather than the entire tuber formation stage.

Third, we added more technical details on canopy hyperspectral data acquisition in Section 2.3. Specifically, we identified the instrument as a portable ASD FieldSpec 3 field spectroradiometer, added the probe height above the canopy, and described the plot-level spectral acquisition procedure. We clarified that three representative measurement positions were selected in each plot, three spectral curves were collected at each position, abnormal curves were removed, and the remaining spectra were averaged first at the position level and then at the plot level. We also clarified that within-plot spectral measurements were treated as technical replicates rather than independent biological samples. In addition, we added the smoothing method used for spectral preprocessing.

L224-227: The same treatment layout and three plot replicates were maintained in both experimental years, and the plot was used as the experimental unit for subsequent LNC measurement and canopy spectral analysis.

L241-245: To characterize the vertical heterogeneity of nitrogen within the potato canopy, stratified leaf sampling was conducted during the tuber formation to early tuber bulking period. Sampling was performed on 7 July 2022 and 8 July 2023, corresponding to approximately 63 and 68 days after planting, respectively, and was synchronized with canopy spectral data acquisition.

L279-282: A portable ASD FieldSpec 3 field spectroradiometer (Analytical Spectral Devices, Inc., Boulder, CO, USA) was used to collect potato canopy reflectance data, with the spectral range of 350–1830 nm used for subsequent analysis [19].

L284-288: The spectral probe was held approximately 50–80 cm above the top of the potato canopy and positioned vertically downward toward the plot canopy, avoiding plot edges and areas with obvious missing plants, to ensure that the observed area represented the overall canopy condition of each plot.

L288-295: In each plot, three representative measurement positions were selected, and three spectral curves were collected at each position. Abnormal curves were first removed. The remaining curves were smoothed using the Savitzky–Golay method with a second-order polynomial and an 11-point moving window to reduce high-frequency spectral noise [21]. The smoothed curves were then averaged first at the position level and then at the plot level to obtain plot-scale canopy reflectance. These within-plot spectral measurements were treated as technical replicates and were not used as independent biological samples.

The text in the chapter Results often includes repetitions and conclusions on the ‘optimal’ combination of wavebands and mathematical data processing. This should be avoided by condensing the text to the essence of experimental results. It seems logical that (PLSR) prediction models derived from indices with higher correlations between spectral information and LNC are superior to models using indices with lower correlation coefficients. Similarly, graphs (Figure 7) on the model performance – R², RMSE, MRE – give information very similar to the scattered plots on the relationship between observed and predicted LNC values (Figure 8).

Response: Thank you for this helpful comment. We agree that the Results section contained repeated descriptions and overly conclusive statements regarding the “optimal” waveband combinations and mathematical preprocessing. We have revised Sections 3.1–3.4 to condense the text and focus on the essential experimental results.

In Section 3.1, we shortened the description of the vertical LNC gradient and removed repeated interpretive statements on the response of traditional vegetation indices. In Section 3.2, we condensed the description of the two-band spectral indices by retaining the key correlation values under FOD1.5 and removing repeated descriptions of individual index types that were already listed in Table 2. We also replaced “spatial distribution” with “spectral distribution” when referring to the heatmaps.

In Section 3.3, we similarly condensed the description of three-band spectral indices. Detailed band combinations are retained in Table 3, while the main text now reports only the key maximum correlation values and the main layer-dependent band characteristics. We also replaced “optimal band combinations” with “selected band combinations” where appropriate to avoid overly strong wording.

In Section 3.4, we combined the repeated Top, Middle, and Bottom LNC model-performance descriptions into one concise paragraph. The key testing-set R² and RMSE results were retained, but repeated interpretations for each canopy layer were removed. We also clarified that the measured-versus-predicted scatter plots are retained as a visual supplement to the quantitative metrics, thereby reducing overlap between the model-performance graph and the scatter plots. Finally, we replaced the statement that FOD1.5 combined with three-band spectral indices was the “optimal strategy” with a more cautious statement limited to the present dataset.

L454-461: One-way ANOVA followed by Tukey’s multiple comparison test showed that LNC differed significantly among the three canopy layers (F = 1587.146, P < 0.001). The compact letter display in Figure 2 indicates that Top, Middle, and Bottom LNC were significantly different from each other. The LNC of potato leaves decreased vertically from the Top to Middle and Bottom canopy layers, following the order Top LNC > Middle LNC > Bottom LNC (Figure 2). This vertical gradient indicated that layer-specific LNC measurements provided additional information beyond a single canopy-averaged nitrogen indicator.

L484-488: Overall, traditional vegetation indices reflected LNC variation to some extent, but their correlations decreased from the upper to lower canopy layers. The correlation coefficients between traditional vegetation indices and layer-specific LNC are summarized in Table 1.

L510-511: These results showed that full-band two-band combinations had stronger correlations with LNC than traditional vegetation indices.

L512-515: After FOD transformation, the correlations between two-band indices and layered LNC increased, with FOD1.5 showing the strongest responses among the tested orders. Under FOD1.5, the highest |r| values reached 0.855, 0.849, and 0.814 for Top, Middle, and Bottom LNC, respectively.

L517-518: The correlation heatmaps further showed that FOD transformation changed the spectral distribution of sensitive band combinations.

L548-549: Three-band spectral indices showed stronger overall correlations with layered LNC than two-band indices, particularly under FOD1.5.

L550-552: Under the original reflectance condition, three-band indices already showed relatively high correlations with layered LNC. At FOD0, the highest |r| values reached 0.856, 0.708, and 0.631 for Top, Middle, and Bottom LNC, respectively.

L553-555: After FOD transformation, FOD1.5 produced the highest correlations across all three canopy layers. Under FOD1.5, the highest |r| values reached 0.893, 0.885, and 0.852 for Top, Middle, and Bottom LNC, respectively.

L556-563: The spectral distribution of correlation coefficients for the three-band spectral indices is shown in Figure 5. For visualization of the three-band correlation space, the displayed plane in each heatmap was selected by fixing the k-band at the wavelength included in the selected three-band combination with the highest absolute Pearson correlation coefficient for the corresponding canopy layer, FOD order, and index type. The heatmap therefore represents the slice-based correlation distribution over band i and band j under the selected k-band slice. The heatmaps showed layer-dependent correlation patterns, with clearer high-correlation regions under FOD1.5.

L564-566: The selected band combinations and correlation coefficients of the three-band spectral indices are summarized in Table 3. The selected bands differed among canopy layers; for Bottom LNC, wavelengths around 1142 and 1307 nm were frequently involved.

L589-598: As shown in Figure 6, the testing-set R² generally increased from the traditional vegetation index models to the 2D-FOD1.5 and 3D-FOD1.5 models. Among the three index systems, the 3D-FOD1.5 model achieved the highest testing-set R² for all three canopy layers, although the error-based metrics did not always show the same ranking. For Top, Middle, and Bottom LNC, the testing-set R² values increased from 0.615, 0.491, and 0.402 in the traditional vegetation index models to 0.803, 0.706, and 0.681 in the 3D-FOD1.5 models, respectively. The largest increase was observed for Bottom LNC, with R² increasing by 0.279 and RMSE decreasing from 0.159 to 0.113. However, RMSE and MRE did not always follow the same ranking as R², indicating that model performance should be evaluated using multiple metrics.

L599-600: Figure 7 shows the agreement between measured and predicted LNC values and is retained as a visual supplement to the quantitative metrics in Figure 6.

L601-603: Overall, within the present dataset, the 3D-FOD1.5 indices produced the highest testing-set R² values among the compared index systems, with the largest improvement observed for Bottom LNC.

The penultimate paragraph of the Discussion gives an overview on possible future research areas – spectral sensing of more field experiments using different cultivars, ecological regions, years, different growth stages of potato crops, data processing, etc. However, more detailed studies of ground truth, i.e. the relationship between LNC values of different leaf layers and their composition is missing. Knowledge on the within-canopy distribution patterns of LNC – depending on N fertilizer level? - allows modelling of the layer-specific LNC from remote sensing data of the top leaf layer.

Response: Thank you for this important comment. We agree that more detailed ground-truth information is needed to fully explain the relationship between layer-specific LNC, leaf composition, canopy structure, and top-of-canopy reflectance. We revised the Discussion to clarify that canopy spectra measured from above the canopy represent mixed signals from different leaf layers, leaf angles, shadows, and canopy gaps. Therefore, the present dataset cannot quantitatively separate the individual contributions of Top, Middle, and Bottom leaves to the total canopy reflectance.

We also clarified that the ground-truth information in this study was mainly limited to layer-specific LNC, although this allowed the vertical nitrogen gradient within the potato canopy to be directly characterized. Future studies should include simultaneous measurements of layer-specific chlorophyll content, leaf water content, dry matter, protein content, leaf age, LAI, biomass distribution, and within-canopy light distribution. In addition, we added that the present study did not explicitly model how nitrogen application level affected the relative distribution of LNC among canopy layers, and that such treatment-level analysis would be useful for linking within-canopy LNC patterns with top-of-canopy spectral observations.

At the same time, we clarified that the mixed canopy reflectance still contains useful integrated information related to canopy structure, nitrogen redistribution, and lower-leaf physiological status, which supports the methodological value of layer-specific LNC estimation from canopy hyperspectral data.

L796-802: However, because the canopy spectrum was measured from above the canopy in the nadir direction, it represents a mixed signal from different leaf layers, leaf angles, shadows, and canopy gaps. The present dataset cannot quantitatively separate the individual contributions of Top, Middle, and Bottom leaves to the total canopy reflectance. Nevertheless, the improved estimation of layer-specific LNC suggests that the mixed canopy reflectance still contains useful integrated information related to canopy structure, nitrogen redistribution, and lower-leaf physiological status.

L806-820: This study did not explicitly model how nitrogen application level affected the relative distribution of LNC among the Top, Middle, and Bottom layers. Such treatment-level analysis would be useful for linking ground-truth within-canopy LNC patterns with top-of-canopy spectral observations. In addition, the transferability of the selected FOD order, selected wavelength combinations, and PLSR models still needs to be validated across more years, cultivars, ecological regions, and management conditions. A year-independent validation was further conducted as an initial test of temporal transferability. The positive but moderate R² values suggest that the selected 3D-FOD1.5 indices captured some temporally transferable spectral information. However, the layer-dependent performance also indicates that the selected wavelength combinations may be affected by year-specific canopy structure, growth status, illumination conditions, and nitrogen distribution patterns. Therefore, these indices should be regarded as dataset-supported candidate indices rather than universally applicable indices. Broader validation across more years, cultivars, sites, growth stages, and fertilization regimes remains necessary before these indices can be generalized.

L825-831: Third, the ground-truth information used in this study was limited to layer-specific LNC. The physiological interpretation of the selected wavelength combinations was therefore mainly based on spectral knowledge and model performance. Future studies should simultaneously measure layer-specific chlorophyll content, leaf water content, dry matter, protein content, leaf age, LAI, biomass distribution, and within-canopy light distribution to clarify how leaf composition and canopy structure link Top, Middle, and Bottom LNC to the mixed top-of-canopy reflectance signal.

The chapter Conclusion gives an extended summary (Abstract) of the manuscript, but does not include neither explanations for the findings, nor ideas on their significance. Do the authors want to explore the relationship between top canopy reflectance and physiological status of lower leaf layers of potato? Or do they think that their results may be useful for potato fertilization under production conditions despite of only one time of measurement per growth season in their experiments and the fact that potatoes receive in-furrow fertilization before planting? Using pre-planting in-furrow fertilization, potato growers do not need LNC assessment for management decisions.

Response: Thank you for this important comment. We agree that the original Conclusion was too similar to an extended summary of the Abstract and did not sufficiently explain the significance and application boundary of the findings. We have therefore rewritten the Conclusion to reduce repeated numerical summaries and to better explain the meaning of the results.

In the revised Conclusion, we first emphasize that the vertical LNC gradient indicates that canopy-averaged LNC may obscure layer-specific nitrogen differences, particularly in middle and lower leaves. We then clarify that the improvement in Bottom LNC estimation should not be interpreted as direct optical measurement of lower canopy leaves. Rather, it suggests that mixed top-of-canopy reflectance contains indirect but useful information related to canopy structure, nitrogen redistribution, senescence, and the physiological status of lower leaves.

We also revised the statement on practical application. The revised Conclusion now states that the proposed framework should be regarded as a methodological reference for vertical canopy nitrogen diagnosis and canopy functional assessment, rather than as an immediately applicable fertilization-management tool. We further clarify that, because this study was conducted at one site, with one cultivar, and with one measurement period per year under a pre-planting in-furrow fertilization system, the results should not yet be generalized as a direct tool for in-season fertilization decisions. Further validation across growth stages, cultivars, ecological regions, fertilization regimes, and UAV or satellite hyperspectral platforms is required before operational application.

L868-874: This study demonstrated that potato canopy LNC showed clear vertical heterogeneity during the tuber formation to early tuber bulking period, with LNC decreasing from the Top to Middle and Bottom canopy layers. This vertical pattern indicates that canopy-averaged LNC may obscure layer-specific nitrogen differences, particularly in middle and lower leaves. Therefore, layer-specific LNC estimation provides a more refined description of canopy nitrogen allocation and functional status than conventional canopy-averaged approaches.

L875-881: Traditional vegetation indices were able to reflect LNC variation to some extent, but their response capacity decreased with increasing canopy depth. In contrast, FOD-transformed two-band and three-band spectral indices improved the spectral representation of layer-specific LNC. Among the tested FOD orders, FOD1.5 showed the strongest overall response. The combination of FOD1.5 and three-band spectral indices produced the highest testing-set R² values among the compared index systems, with the greatest improvement observed for Bottom LNC.

L882-896: The improvement in Bottom LNC estimation should not be interpreted as direct optical measurement of lower canopy leaves. Rather, it suggests that mixed top-of-canopy reflectance contains indirect but useful information related to canopy structure, nitrogen redistribution, senescence, and the physiological status of lower leaves. This finding highlights the methodological value of integrating FOD transformation and three-band spectral indices for extracting layer-specific nitrogen information from canopy-scale hyperspectral data. Therefore, the proposed framework should be regarded as a methodological reference for vertical canopy nitrogen diagnosis and canopy functional assessment, rather than as an immediately applicable fertilization-management tool. This is because this study was conducted at one site, with one cultivar, and with one measurement period per year under a pre-planting in-furrow fertilization system, the results should not yet be generalized as a direct tool for in-season fertilization decisions. Further validation across growth stages, cultivars, ecological regions, fertilization regimes, and UAV or satellite hyperspectral platforms is required before operational application.

The numbering of references used in the text should be checked again, as, for example, text references [20] on page 7 and [21] on page 8 are not correct.

Response: Thank you for pointing this out. We carefully checked the numbering of all in-text citations and the corresponding entries in the reference list throughout the manuscript. The mismatched citations mentioned by the reviewer, including the citations around the technical equipment, white-panel calibration, and spectral preprocessing descriptions, were corrected where necessary. We also rechecked the full reference list to ensure consistency between the in-text citations and the reference entries.

Revision made: The numbering of in-text citations and the reference list was checked throughout the manuscript. In particular, the citations related to the ASD FieldSpec 3 spectroradiometer, white-panel calibration, and Savitzky-Golay smoothing in Section 2.3 were checked and corrected where necessary.

Location: Section 2.3, Lines 280–292, and the References section.

The authors often use the term ‘weak spectral signal’ when referring to nitrogen effects on spectral information. Do they mean small spectral differences? Small spectral differences indicate small physiological differences among leaf levels – why should it be of interest to increase (bloat) small differences – only because it is possible by computation? The key question is, what is the contribution of the LNC of middle and bottom leaf layers of potato canopies to the overall spectral reflectance signal? Is it possible to quantify the LNC of lower leaf layers from spectral information recorded by a sensor above the canopy providing mixed information from a vertical gradient? In contrast, it seems logical that the creation of hundreds of new waveband combinations results in correlations wit LNC which are higher than those for spectral vegetation indices developed for other purposes. The authors have not shown that the new indices may be useful in different experiments. Testing of the new indices developed from data recorded in the year 2022 on data recorded in the year 2023 would be a first step.

Response: Thank you for this important comment. We agree that the term “weak spectral signal” was not sufficiently precise and could be interpreted as artificially amplifying small spectral differences. We have therefore revised the wording throughout the manuscript. Expressions such as “weak spectral signal”, “weak nitrogen signals”, and “weak spectral information” were replaced with more specific terms such as “attenuated and mixed nitrogen-related spectral information”, “subtle but systematic spectral variations”, or “indirect information related to lower-layer nitrogen status”.

We also clarified the interpretation of lower-layer LNC estimation. In the revised Discussion, we state that canopy spectra measured from above the canopy represent mixed signals from different leaf layers, leaf angles, shadows, and canopy gaps. Therefore, the present dataset cannot quantitatively separate the individual contributions of Top, Middle, and Bottom leaves to the total canopy reflectance. The proposed approach should therefore be interpreted as indirect estimation from mixed canopy reflectance, rather than direct optical measurement of lower leaves.

To address the concern regarding possible overfitting and the generalization ability of the newly selected indices, we added a year-independent validation. The 2022 dataset was used for sensitive-band selection, 3D-DI, 3D-NDI, and 3D-RI construction, and PLSR calibration, and the independent 2023 dataset was used for testing. The reverse validation, with 2023 used for index selection and model calibration and 2022 used for testing, was also conducted as a sensitivity analysis. The key validation results have been summarized in Section 3.4, and the detailed year-independent validation results are provided in the table below. The year-independent validation showed partial cross-year transferability. In the 2022-to-2023 validation, the R² values for Top, Middle, and Bottom LNC were 0.419, 0.326, and 0.237, respectively. In the reverse validation, the corresponding R² values were 0.367, 0.421, and 0.257.

We have added the validation procedure to Section 2.9, summarized the validation results in Section 3.4, and revised the Discussion to state that the selected wavelength combinations should be regarded as dataset-supported candidate indices rather than universally applicable indices. Broader validation across more years, cultivars, sites, growth stages, and fertilization regimes is still required.

L134-136: As a result, traditional fixed-band indices may not fully capture such attenuated and mixed nitrogen-related spectral information from internal canopy layers.

L161-164: For layer-specific LNC estimation in potato, three-band spectral indices are expected to improve the representation of attenuated nitrogen-related information in middle and lower canopy layers by combining visible, red-edge, near-infrared, and shortwave infrared bands.

L430-438: In addition to the random calibration/testing split, a year-independent validation was conducted to evaluate the temporal transferability of the selected 3D-FOD1.5 indices. The 2022 dataset was used for sensitive-band selection, 3D-DI, 3D-NDI, and 3D-RI construction, and PLSR calibration, and the independent 2023 dataset was used for testing. The reverse validation, with the 2023 dataset used for index selection and model calibration and the 2022 dataset used for testing, was also performed as a sensitivity analysis. For each PLSR model, the number of latent variables was selected using leave-one-out cross-validation within the calibration dataset.

L603-610: Year-independent validation was further conducted to evaluate the temporal transferability of the selected 3D-FOD1.5 indices. When the 2022 dataset was used for calibration and the 2023 dataset was used for testing, the models achieved R² values of 0.419, 0.326, and 0.237 for Top, Middle, and Bottom LNC, respectively. In the reverse validation, with 2023 used for calibration and 2022 used for testing, the corresponding R² values were 0.367, 0.421, and 0.257. These results indicate that the selected indices retained partial cross-year transferability under a stricter year-independent validation, although the validation accuracy varied among canopy layers.

L660-662: As a result, traditional fixed-band indices are insufficient for extracting attenuated and mixed nitrogen-related information from internal canopy layers.

L690-692: The key scientific question is whether attenuated and indirect nitrogen-related information from lower canopy layers can be extracted from mixed canopy hyperspectral signals using more flexible spectral transformations and multi-band index construction.

L740-744: Original reflectance spectra contain abundant physiological information, but nitrogen-related spectral variations can be attenuated or masked by illumination conditions, background variation, canopy structure, and band redundancy.

L757-758: Therefore, FOD1.5 likely provided an effective compromise between preserving biologically meaningful spectral patterns and enhancing subtle local spectral features.

L800-802: Nevertheless, the improved estimation of layer-specific LNC suggests that the mixed canopy reflectance still contains useful integrated information related to canopy structure, nitrogen redistribution, and lower-leaf physiological status.

L789-795: Traditional vegetation indices rely on fixed two-band formulas and are mainly dominated by upper-canopy greenness, making it difficult for them to capture attenuated and indirect information related to lower-layer nitrogen status. In contrast, three-band optimized spectral indices integrate three wavelengths from different spectral regions, while FOD1.5 enhances subtle local spectral differences. Their combination therefore improves the extraction of lower-layer nitrogen information from canopy-scale reflectance. This is the main reason why the improvement was most evident for Bottom LNC.

L812-820: A year-independent validation was further conducted as an initial test of temporal transferability. The positive but moderate R² values suggest that the selected 3D-FOD1.5 indices captured some temporally transferable spectral information. However, the layer-dependent performance also indicates that the selected wavelength combinations may be affected by year-specific canopy structure, growth status, illumination conditions, and nitrogen distribution patterns. Therefore, these indices should be regarded as dataset-supported candidate indices rather than universally applicable indices. Broader validation across more years, cultivars, sites, growth stages, and fertilization regimes remains necessary before these indices can be generalized.

L856-859: The combination of FOD1.5 and three-band optimized spectral indices improved the representation of multi-band spectral information and the estimation accuracy of layered LNC, especially for the lower canopy, where layer-related information was attenuated and mixed within the top-of-canopy reflectance.

Table 1. Year-independent validation of the selected 3D-FOD1.5-PLSR models.

Spectral indices from literature have not been developed for assessment of the leaf nitrogen content, but for estimation of biomass, chlorophyll content, greenness, etc.. Other waveband combinations (two or three wavebands), therefore, are very likely to give better results. However, they have to be tested for generalization under various cropping conditions.

Response: Thank you for this important comment. We agree that most spectral vegetation indices from the literature were not specifically developed for layer-specific LNC estimation. Many of them were originally designed for canopy greenness, chlorophyll content, biomass, or general crop nitrogen status. Therefore, the better performance of newly selected two- or three-band combinations should be interpreted cautiously.

To address this issue, we revised the Introduction and Discussion. In the revised manuscript, traditional vegetation indices are described as baseline references rather than as indices specifically designed for layer-specific LNC estimation. We also clarified that the selected two- and three-band indices should be regarded as data-supported candidate indices for the present layer-specific LNC estimation problem, rather than universally applicable spectral indices.

We also agree that different selected wavelengths among Top, Middle, and Bottom layers indicate interactions among LNC, leaf position, biomass distribution, canopy structure, and optical mixing. We have therefore added a discussion stating that these wavelength combinations should not be generalized across crops, cultivars, canopy architectures, layer definitions, growth stages, or environmental conditions without further validation.

In addition, we clarified that canopy spectra measured from above the canopy represent mixed signals from different leaf layers, leaf angles, shadows, and canopy gaps. The present dataset cannot quantitatively separate the individual contributions of Top, Middle, and Bottom leaves to the total canopy reflectance. Therefore, the proposed approach should be interpreted as indirect estimation from mixed canopy reflectance, rather than as a computational replacement for the biological and technical limitations of lower-canopy sensing.

Finally, to provide an initial test of generalization, we added a year-independent validation. The 2022 dataset was used for sensitive-band selection, index construction, and PLSR calibration, and the independent 2023 dataset was used for testing. The reverse validation was also conducted. The validation results showed partial cross-year transferability, indicating that the selected indices captured some temporally transferable spectral information, although their performance varied among canopy layers. These results have been summarized in Section 3.4.

L125-128: Traditional vegetation indices have been widely used to estimate crop nitrogen-, chlorophyll-, greenness-, and biomass-related traits because of their simple structure, clear physical meaning, and ease of calculation. Common vegetation indices are usually constructed using visible, red-edge, and near-infrared bands.

L667-670: It should also be noted that most traditional vegetation indices used in this study were originally developed for canopy greenness, chlorophyll content, biomass, or general crop nitrogen status, rather than for layer-specific LNC estimation.

L726-738: These layer-dependent selected wavelength combinations suggest that the spectral response of LNC is influenced by interactions among leaf nitrogen status, leaf position, biomass distribution, canopy structure, and optical mixing. Therefore, these wavelength combinations should not be regarded as universally applicable spectral indices for LNC quantification across crops or cultivars. Instead, they represent candidate band combinations identified under the present potato cultivar, canopy structure, growth stage, and field conditions. Their broader applicability requires further validation under different cultivars, canopy architectures, layer definitions, growth stages, and environmental conditions. Therefore, Bottom LNC estimation cannot rely solely on greenness or chlorophyll absorption, but requires the integration of spectral signals from red-edge, near-infrared, and shortwave infrared regions. This also explains why three-band spectral indices showed a more pronounced advantage for lower-canopy LNC estimation.

If different wavebands of the full spectrum are necessary to quantify the LNC of different leaf layers, there is no possibility of generalization of the application of spectral indices for LNC quantification – different crops, different crop cultivars (?), number and position of layers, environmental conditions, etc.. The lack of robustness of spectral vegetation indices for lower leaf areas because of biological– overlapping of various leaf layers, differences in biomass and its representation – and technical – a sensor above the canopy records the sum of reflectance values from various horizontal levels with unknown contribution - reasons cannot be replaced by computation of spectral data taken from Nadir position. The fact, that different wavebands (and computation procedures) gave the best results for the three leaf layer levels investigated, demonstrates that there are interactions between LNC and leaf layer / biomass composition. It is not clear I) what is the contribution of different leaf layers to the overall canopy signal, and II) what is the spectral information characteristic for the (LNC status of) the middle and bottom leaf layer, respectively.

Response: Thank you for this important comment. We agree that the use of different wavelength combinations for different canopy layers raises important questions regarding generalization and interpretation. We have revised the manuscript to address this issue more cautiously.

First, we clarified in the Introduction and Discussion that most traditional vegetation indices used in this study were originally developed for canopy greenness, chlorophyll content, biomass, or general crop nitrogen status, rather than specifically for layer-specific LNC estimation. Therefore, their lower performance for Middle and Bottom LNC should not be interpreted as a failure of these indices, but rather as evidence that fixed-band empirical indices are not specifically designed to resolve vertically mixed LNC information within a closed potato canopy. In the revised manuscript, traditional vegetation indices are treated as baseline references, whereas the selected two- and three-band indices are described as data-supported candidate indices for the present layer-specific LNC estimation problem.

Second, we added a discussion explaining that the layer-dependent selected wavelength combinations indicate interactions among leaf nitrogen status, leaf position, biomass distribution, canopy structure, and optical mixing. Therefore, these wavelength combinations should not be regarded as universally applicable spectral indices for LNC quantification across crops or cultivars. Instead, they represent candidate band combinations identified under the present potato cultivar, canopy structure, growth stage, and field conditions.

Third, we clarified that canopy spectra measured from above the canopy represent mixed signals from different leaf layers, leaf angles, shadows, and canopy gaps. The present dataset cannot quantitatively separate the individual contributions of Top, Middle, and Bottom leaves to the total canopy reflectance. Therefore, the proposed approach should be interpreted as indirect estimation from mixed canopy reflectance, rather than as a computational replacement for the biological and technical limitations of lower-canopy sensing.

Finally, to provide an initial test of generalization, we added a year-independent validation. The 2022 dataset was used for sensitive-band selection, index construction, and PLSR calibration, and the independent 2023 dataset was used for testing; the reverse validation was also conducted. The results showed partial cross-year transferability, indicating that the selected indices captured some temporally transferable spectral information, although their performance varied among canopy layers. The validation procedure has been added to Section 2.9, and the key results have been summarized in Section 3.4.

L125-137: Traditional vegetation indices have been widely used to estimate crop nitrogen-, chlorophyll-, greenness-, and biomass-related traits because of their simple structure, clear physical meaning, and ease of calculation. Common vegetation indices are usually constructed using visible, red-edge, and near-infrared bands, and can characterize canopy greenness, pigment absorption, and structural scattering features [10]. However, these indices generally rely on a small number of fixed bands, and their sensitivity can be affected by crop type, growth stage, canopy structure, soil background, and environmental conditions [11]. For layer-specific LNC, especially LNC in lower leaves, the target signal is often weakened by upper-leaf occlusion, leaf overlap, shadow effects, and mixed canopy reflectance. As a result, traditional fixed-band indices may not fully capture such attenuated and mixed nitrogen-related spectral information from internal canopy layers. Therefore, it is necessary to expand the sensitive wavelength search space to improve the ability of spectral features to characterize layer-specific LNC.

L430-438: In addition to the random calibration/testing split, a year-independent validation was conducted to evaluate the temporal transferability of the selected 3D-FOD1.5 indices. The 2022 dataset was used for sensitive-band selection, 3D-DI, 3D-NDI, and 3D-RI construction, and PLSR calibration, and the independent 2023 dataset was used for testing. The reverse validation, with the 2023 dataset used for index selection and model calibration and the 2022 dataset used for testing, was also performed as a sensitivity analysis. For each PLSR model, the number of latent variables was selected using leave-one-out cross-validation within the calibration dataset.

L601-610: Overall, within the present dataset, the 3D-FOD1.5 indices produced the highest testing-set R² values among the compared index systems, with the largest improvement observed for Bottom LNC. Year-independent validation was further conducted to evaluate the temporal transferability of the selected 3D-FOD1.5 indices. When the 2022 dataset was used for calibration and the 2023 dataset was used for testing, the models achieved R² values of 0.419, 0.326, and 0.237 for Top, Middle, and Bottom LNC, respectively. In the reverse validation, with 2023 used for calibration and 2022 used for testing, the corresponding R² values were 0.367, 0.421, and 0.257. These results indicate that the selected indices retained partial cross-year transferability under a stricter year-independent validation, although the validation accuracy varied among canopy layers.

L667-670: It should also be noted that most traditional vegetation indices used in this study were originally developed for canopy greenness, chlorophyll content, biomass, or general crop nitrogen status, rather than for layer-specific LNC estimation.

L726-738: These layer-dependent selected wavelength combinations suggest that the spectral response of LNC is influenced by interactions among leaf nitrogen status, leaf position, biomass distribution, canopy structure, and optical mixing. Therefore, these wavelength combinations should not be regarded as universally applicable spectral indices for LNC quantification across crops or cultivars. Instead, they represent candidate band combinations identified under the present potato cultivar, canopy structure, growth stage, and field conditions. Their broader applicability requires further validation under different cultivars, canopy architectures, layer definitions, growth stages, and environmental conditions. Therefore, Bottom LNC estimation cannot rely solely on greenness or chlorophyll absorption, but requires the integration of spectral signals from red-edge, near-infrared, and shortwave infrared regions. This also explains why three-band spectral indices showed a more pronounced advantage for lower-canopy LNC estimation.

L796-802: However, because the canopy spectrum was measured from above the canopy in the nadir direction, it represents a mixed signal from different leaf layers, leaf angles, shadows, and canopy gaps. The present dataset cannot quantitatively separate the individual contributions of Top, Middle, and Bottom leaves to the total canopy reflectance. Nevertheless, the improved estimation of layer-specific LNC suggests that the mixed canopy reflectance still contains useful integrated information related to canopy structure, nitrogen redistribution, and lower-leaf physiological status.

L809-820: In addition, the transferability of the selected FOD order, selected wavelength combinations, and PLSR models still needs to be validated across more years, cultivars, ecological regions, and management conditions. A year-independent validation was further conducted as an initial test of temporal transferability. The positive but moderate R² values suggest that the selected 3D-FOD1.5 indices captured some temporally transferable spectral information. However, the layer-dependent performance also indicates that the selected wavelength combinations may be affected by year-specific canopy structure, growth status, illumination conditions, and nitrogen distribution patterns. Therefore, these indices should be regarded as dataset-supported candidate indices rather than universally applicable indices. Broader validation across more years, cultivars, sites, growth stages, and fertilization regimes remains necessary before these indices can be generalized.

Specific comments:

Please see the PDF with more specific comments in the text.

Annotated comment 1:“These methods are not of similar importance - PLSR is used for testing the suitability of reflectance-based indices.”

Response: Thank you for this helpful comment. We agree that PLSR should not be presented as having the same methodological role as hyperspectral reflectance preprocessing and spectral-index construction. In the revised Abstract, we separated the construction of the spectral-index framework from the modeling step. Specifically, we now state that the layer-specific LNC estimation framework was based on canopy hyperspectral reflectance, FOD transformation, and two-band and three-band optimized spectral indices, whereas PLSR was used to evaluate the predictive ability of the selected spectral indices for Top, Middle, and Bottom LNC.

Original text: L17-21:To address the insufficient characterization of vertical heterogeneity in potato canopy leaf nitrogen content (LNC), this study developed a layer-specific LNC estimation framework by integrating hyperspectral reflectance, fractional-order derivative (FOD) transformation, two-dimensional spectral indices, three-dimensional optimal spectral indices (3D-OSIs), and partial least squares regression (PLSR).

Revised text: L17-21: To address the insufficient characterization of vertical heterogeneity in potato canopy leaf nitrogen content (LNC), this study developed a layer-specific LNC estimation framework based on canopy hyperspectral reflectance, fractional-order derivative (FOD) transformation, and two-band and three-band optimized spectral indices. Partial least squares regression (PLSR) was then used to evaluate the predictive ability of the selected spectral indices for Top, Middle, and Bottom LNC.

Annotated comment 2:“not clear - correlation (coefficient) used previously calculated in another way?”

Response: Thank you for pointing this out. We agree that the original wording did not clearly specify how the correlation coefficients were calculated. In the revised Abstract, we clarified that the reported values refer to the maximum absolute Pearson correlation coefficients between the selected two-band indices and layer-specific LNC. We also replaced “two-band indices” with “two-band indices” in this sentence to make the terminology clearer.

Original text: L28-32:Compared with traditional vegetation indices, FOD-based two-dimensional indices improved the spectral response to layer-specific LNC. Under FOD1.5, the maximum absolute correlation coefficients (∣r∣) of the optimal two-dimensional indices reached 0.855, 0.849, and 0.814 for Top, Middle, and Bottom LNC, respectively.”

Revised text: L28-32: Compared with traditional vegetation indices, FOD-based two-band indices showed stronger Pearson correlations with layer-specific LNC. Under FOD1.5, the maximum absolute Pearson correlation coefficients (|r|) between the selected two-band indices and LNC reached 0.855, 0.849, and 0.814 for Top, Middle, and Bottom LNC, respectively. The three-band optimized spectral indices further enhanced spectral information extraction, with maximum |r| values of 0.893, 0.885, and 0.852, respectively.

Annotated comment 3:may be deleted - this is well-known from worldwide statistics, not studies.

Response: Thank you for this comment. We agree that the original sentence was unnecessary and that the global importance of potato can be stated more concisely. We deleted the sentence beginning with “Previous studies have indicated...” and retained only a concise background statement on the importance of potato.

Original text: L48-50:Potato (Solanum tuberosum L.) is an important food and economic crop worldwide, playing a critical role in ensuring food security, increasing farmers’ income, and supporting dryland agricultural production. Previous studies have indicated that potato is a major non-cereal food crop globally and has an irreplaceable role in food security and livelihood support, especially in developing countries [1].”

Revised text: L47-49: Potato (Solanum tuberosum L.) is an important food and economic crop worldwide, playing a critical role in ensuring food security, increasing farmers’ income, and supporting dryland agricultural production [1].

Annotated comment 4:The sentence referring to Figure 1 in Section 2.1 was highlighted in the annotated PDF.

Response: Thank you for highlighting this sentence. We revised it to make the figure reference more concise and corrected the spacing error in the text.

Original text: L218-220:The soil type at the experimental site is dominated by sandy soil, representing a typical potato production region in northwestern China.The geographical location and elevation background of the study area are shown in Figure 1.

Revised text: L218-220:The soil type at the experimental site is dominated by sandy soil, representing a typical potato production region in northwestern China. The location of the study area is shown in Figure 1.

Annotated comment 5:chlorophyll is a prerequisite for photosynthesis and CO2 fixation.

Response: Thank you for this helpful clarification. We agree that the relationship among nitrogen, chlorophyll, photosynthesis, and carbon assimilation should be expressed more accurately. We revised the sentence to clarify that nitrogen affects photosynthetic capacity and CO₂ fixation through its role as a major component of chlorophyll, photosynthetic proteins, and enzymes.

Original text: L65-70:Leaf nitrogen content (LNC) is a direct indicator of crop nitrogen nutritional status and is closely related to chlorophyll content, photosynthetic enzyme activity, leaf physiological function, and biomass accumulation. Nitrogen is involved not only in the synthesis of chlorophyll and proteins, but also in plant photosynthetic capacity and carbon assimilation; therefore, LNC is commonly used to evaluate crop nitrogen supply level and physiological activity [3][4].

Revised text: L63-69: Leaf nitrogen content (LNC) is a direct indicator of crop nitrogen nutritional status and is closely related to chlorophyll content, photosynthetic enzyme activity, leaf physiological function, and biomass accumulation. As a major component of chlorophyll, photosynthetic proteins, and enzymes, nitrogen affects light absorption, photosynthetic capacity, CO₂ fixation, and carbon assimilation; therefore, LNC is commonly used to evaluate crop nitrogen supply level and physiological activity [3][4].

Annotated comment 6:SPAD value is a reflectance-based method for estimating the chlorophyll content and greenness - much simpler and cheaper than hyperspectral measurements.

Response: Thank you for this clarification. We agree that SPAD readings provide a simple and low-cost optical indicator of leaf chlorophyll status and greenness. We revised the sentence to acknowledge this advantage, while also clarifying that SPAD readings remain indirect proxies for nitrogen status, whereas LNC directly reflects nitrogen accumulation and allocation within plants.

Original text: L70-74:Compared with indirect indicators such as canopy greenness and soil plant analysis development (SPAD) values, LNC can more directly reflect nitrogen accumulation and allocation within plants.

Revised text: L69-73: Although canopy greenness and soil plant analysis development (SPAD) chlorophyll meter readings provide simple and low-cost optical indicators of leaf chlorophyll status and greenness, they remain indirect proxies for nitrogen status. In contrast, LNC more directly reflects nitrogen accumulation and allocation within plants.

Annotated comment 7:The morphology and structure of grass-like crops (rice, wheat) differs significantly from the broad-leaf potato crop. The senescence of dicotyledonous crops and cereals also differ.

Response: Thank you for this important clarification. We agree that results from rice and wheat should not be directly generalized to potato because cereal crops differ from broad-leaf potato in canopy morphology, leaf arrangement, and senescence patterns. We revised this part of the Introduction to clarify that the rice and wheat studies were cited only as evidence that vertical nitrogen heterogeneity is a common canopy phenomenon, whereas the layer-specific spectral response of potato needs to be examined separately.

Original text:He et al. investigated the vertical distribution of leaf nitrogen within rice canopies using hyperspectral data and found significant differences in LNC among different functional leaf layers, suggesting that accurate estimation of vertical nitrogen distribution is helpful for understanding canopy nutrient supply-demand relationships and yield formation [7]. Ma et al. also reported differences in spectral responses and nitrogen estimation accuracy among different leaf positions in wheat, indicating that leaf position or canopy layer is an important factor affecting nitrogen remote sensing modeling [8]. For potato, differences among canopy layers may be particularly important.”

Revised text: L102-107: Previous studies in rice and wheat have shown that leaf position or canopy layer can influence LNC distribution, spectral responses, and nitrogen estimation accuracy [7][8]. However, rice and wheat are grass-like cereal crops, and their canopy morphology, leaf arrangement, and senescence patterns differ from those of broad-leaf potato. Therefore, these studies are cited only to indicate that vertical nitrogen heterogeneity is a common canopy phenomenon, while the layer-specific spectral response of potato needs to be examined separately.

Annotated comment 8:As the LNC gradient from top to bottom may be modelled, it is sufficient to measure the LNC of the upper leaves in order to know the within-canopy level distribution. A LNC gradient from bottom to top is very unlikely - except for physiological damage.

Response: Thank you for this important clarification. We agree that potato LNC generally follows a top-to-bottom vertical gradient and that such a gradient may be described empirically. We revised the Introduction to avoid implying that layer-specific measurements are needed simply because the existence of a vertical gradient is unknown. Instead, we clarified that the magnitude and shape of the gradient may vary with nitrogen supply, leaf age, shading, canopy structure, and senescence status. Therefore, layer-specific LNC measurements are needed to establish ground-truth information on within-canopy nitrogen distribution and to evaluate whether top-of-canopy hyperspectral reflectance contains indirect information related to Middle and Bottom LNC.

Original text:Therefore, potato LNC often shows a vertical gradient from the upper to lower canopy layers. Estimating only canopy-averaged LNC may obscure nitrogen deficiency or early senescence information in middle and lower leaves, and may be particularly insufficient for revealing attenuated but physiologically meaningful nitrogen-related information in the lower canopy. Therefore, estimating LNC separately for upper, middle, and lower potato canopy leaves is important for refined nitrogen diagnosis and canopy functional assessment.

Revised text: L117-124: Therefore, potato LNC often shows a general vertical gradient from the upper to lower canopy layers. Although such a gradient can be described empirically, its magnitude and shape may vary with nitrogen supply, leaf age, shading, canopy structure, and senescence status. Therefore, layer-specific LNC measurements are still necessary to establish ground-truth information on within-canopy nitrogen distribution and to evaluate whether top-of-canopy hyperspectral reflectance contains indirect information related to Middle and Bottom LNC. This information is important for refined canopy nitrogen diagnosis and canopy functional assessment.

Annotated comment 9:The reviewer questioned the use of the term “weak spectral signal” and related expressions.

Response: Thank you for this comment. We agree that the term “weak spectral signal” was not sufficiently precise and could be misinterpreted as artificially amplifying small spectral differences. We therefore further checked the revised manuscript and replaced the remaining “weak” expressions with more specific wording. In the revised text, “weak spectral information”, “weak spectral signals”, and “weak spectral differences” were replaced with “subtle spectral variations”, “attenuated and mixed spectral information”, or “subtle spectral differences”, depending on the context. We also corrected a repeated phrase in the Discussion.

Original text 1:These results indicate that FOD has potential for extracting weak spectral information and improving the estimation accuracy of nitrogen-related parameters.

Revised text 1: L177-181: These results indicate that FOD has potential for enhancing subtle spectral variations and improving the estimation accuracy of nitrogen-related parameters. However, the optimal fractional order may differ among crops, target traits, and canopy layers. At present, the applicability of FOD to layer-specific LNC estimation in potato and its potential gain when combined with three-band spectral indices remain insufficiently understood.

Original text 2:...especially for lower canopy leaves with weak spectral signals.

Revised text 2: L204-206: ...especially for lower canopy leaves whose spectral information is attenuated and mixed within the top-of-canopy reflectance.

Original text 3:Fractional-order derivative transformation can enhance local absorption features, slope variations, and weak spectral differences while preserving the overall variation pattern of the original spectra, thereby improving the sensitivity of spectral information to changes in leaf nitrogen content.

Revised text 3: L300-316: Fractional-order derivative transformation was applied to the preprocessed canopy reflectance. In this study, multiple fractional orders were used to transform canopy reflectance. The FOD orders were set to 0, 0.5, 1.0, 1.5, 2.0, and 2.5. Through FOD treatments at different orders, multiple fractional-order derivative spectra were obtained and used for subsequent construction of two-band and three-band spectral indices.

Original text 4:Original reflectance spectra contain abundant physiological information, Original reflectance spectra contain abundant physiological information, but many nitrogen-related signals are weak and can be masked by illumination conditions, background variation, canopy structure, and band redundancy.

Revised text 4: L740-744: Original reflectance spectra contain abundant physiological information, but nitrogen-related spectral variations can be attenuated or masked by illumination conditions, background variation, canopy structure, and band redundancy.

Location: Introduction, Section 2.4, and Discussion, Lines 134–136, 177–181, 204–206, 300–316, and 740–744.

Annotated comment 10:This information and the maps of Figure 1 should not be relevant for potato physiological and potato reflectance. It should be deleted.

Response: Thank you for this comment. We agree that the geographical and elevation maps were not essential for understanding potato physiological status or canopy reflectance in this study. Therefore, we deleted the original Figure 1 and removed the corresponding sentence referring to this figure. After deleting the original Figure 1, the subsequent figure numbers were updated accordingly.

Original text:The location of the study area is shown in Figure 1.

Revised text:This sentence was deleted.

Annotated comment 11:The reviewer noted that the growth stage “tuber formation” covers a relatively long period and that more detailed sampling information was needed.

Response: Thank you for this comment. We agree that the sampling period should be described more precisely. In the revised manuscript, we clarified that stratified leaf sampling was conducted during the tuber formation to early tuber bulking period. We also added the exact sampling dates and the corresponding days after planting for both experimental years, and stated that leaf sampling was synchronized with canopy spectral data acquisition.

Original text:To characterize the vertical heterogeneity of nitrogen within the potato canopy, stratified leaf sampling was conducted during the tuber formation stage. Sampling was performed on 7 July 2022 and 8 July 2023, synchronized with canopy spectral data acquisition.

Revised text: L241-245: To characterize the vertical heterogeneity of nitrogen within the potato canopy, stratified leaf sampling was conducted during the tuber formation to early tuber bulking period. Sampling was performed on 7 July 2022 and 8 July 2023, corresponding to approximately 63 and 68 days after planting, respectively, and was synchronized with canopy spectral data acquisition.

Annotated comment 12:The reviewer requested a more detailed description of the canopy hyperspectral data acquisition procedure.

Response: Thank you for this comment. We have revised Section 2.3 to provide a more detailed description of the canopy hyperspectral measurement procedure. Specifically, we added the instrument model and manufacturer, the spectral range used for analysis, white-reference calibration, probe height, nadir-viewing geometry, plot-position selection, the number of spectral curves collected per plot, the removal of abnormal curves, and the averaging procedure used to obtain plot-scale canopy reflectance. We also clarified that within-plot spectral measurements were treated as technical replicates and were not used as independent biological samples.

Original text:A portable field spectroradiometer was used to collect potato canopy reflectance data, with a spectral range of 350–1830 nm. The spectral probe was positioned vertically downward toward the plot canopy, avoiding plot edges and areas with obvious missing plants, to ensure that the observed area represented the overall canopy condition of each plot. Multiple canopy spectral curves were collected for each plot, and abnormal curves were removed. The remaining curves were averaged to obtain the canopy reflectance of each plot.

Revised text: L279-295: A portable ASD FieldSpec 3 field spectroradiometer (Analytical Spectral Devices, Inc., Boulder, CO, USA) was used to collect potato canopy reflectance data, with the spectral range of 350–1830 nm used for subsequent analysis [19]. Before each measurement, the instrument was calibrated using a standard white reference panel, and white-panel calibration was repeated during data acquisition when illumination conditions changed [19][20]. The spectral probe was held approximately 50–80 cm above the top of the potato canopy and positioned vertically downward toward the plot canopy, avoiding plot edges and areas with obvious missing plants, to ensure that the observed area represented the overall canopy condition of each plot. In each plot, three representative measurement positions were selected, and three spectral curves were collected at each position. Abnormal curves were first removed. The remaining curves were smoothed using the Savitzky–Golay method with a second-order polynomial and an 11-point moving window to reduce high-frequency spectral noise [21]. The smoothed curves were then averaged first at the position level and then at the plot level to obtain plot-scale canopy reflectance. These within-plot spectral measurements were treated as technical replicates and were not used as independent biological samples.

Annotated comment 13:Please give methodological details on smoothing.

Response: Thank you for this comment. We agree that the smoothing procedure should be described in more detail. In the revised manuscript, we specified the Savitzky–Golay smoothing parameters and clarified the preprocessing sequence from abnormal-curve removal to smoothing and plot-level averaging.

Original text:Abnormal curves were removed, and the remaining curves were averaged first at the position level and then at the plot level to obtain the plot-scale canopy reflectance. These within-plot spectral measurements were treated as technical replicates and were not used as independent biological samples. Raw spectral data were first subjected to outlier removal and smoothing to reduce the effects of instrumental noise and environmental disturbance on the reflectance curves. Spectral smoothing was performed using the Savitzky–Golay method [21]. Subsequently, multiple spectral observations within each plot were averaged to obtain plot-scale mean canopy reflectance.

Revised text: L289-295: Abnormal curves were first removed. The remaining curves were smoothed using the Savitzky–Golay method with a second-order polynomial and an 11-point moving window to reduce high-frequency spectral noise [21]. The smoothed curves were then averaged first at the position level and then at the plot level to obtain plot-scale canopy reflectance. These within-plot spectral measurements were treated as technical replicates and were not used as independent biological samples.

Annotated comment 14:No replicates per treatment, but average for one plot per treatment per year!?

Response: Thank you for pointing this out. We clarified the distinction between field replicates and within-plot spectral technical replicates. In the experimental design, each treatment had three plot replicates, resulting in 30 experimental plots per year. The plot was used as the experimental unit for LNC measurement and canopy spectral analysis. Within each plot, multiple spectral curves were collected only to obtain a representative plot-scale canopy reflectance. These within-plot spectral measurements were treated as technical replicates and were averaged to one plot-level spectral observation; they were not used as independent biological replicates.

Original text:These within-plot spectral measurements were treated as technical replicates and were not used as independent biological samples.

Revised text: L292-295: The smoothed curves were then averaged first at the position level and then at the plot level to obtain plot-scale canopy reflectance. These within-plot spectral measurements were treated as technical replicates and were not used as independent biological samples.

Annotated comment 15:The reviewer questioned the use of “weak spectral differences” in the description of the FOD transformation.

Response: Thank you for this comment. We agree that the term “weak spectral differences” was not sufficiently precise and could be misinterpreted as artificially amplifying small spectral differences. This expression has been revised to “subtle spectral differences” to make the meaning clearer and more appropriate.

Original text:Fractional-order derivative transformation can enhance local absorption features, slope variations, and weak spectral differences while preserving the overall variation pattern of the original spectra, thereby improving the sensitivity of spectral information to changes in leaf nitrogen content.

Revised text: L300-316: Fractional-order derivative transformation was applied to the preprocessed canopy reflectance. In this study, multiple fractional orders were used to transform canopy reflectance. The FOD orders were set to 0, 0.5, 1.0, 1.5, 2.0, and 2.5. Through FOD treatments at different orders, multiple fractional-order derivative spectra were obtained and used for subsequent construction of two-band and three-band spectral indices.

Annotated comment 16:has been already mentioned in the Introduction. Not part of MaM - should be deleted here.

Response: Thank you for this comment. We agree that the explanatory statement on the role of FOD was more suitable for the Introduction than for the Materials and Methods section. Therefore, we revised the beginning of Section 2.4 to focus on the actual processing procedure and removed the repeated explanatory description.

Original text:To enhance subtle spectral signals related to layer-specific LNC in the potato canopy, fractional-order derivative transformation was applied to the preprocessed canopy reflectance. Fractional-order derivative transformation can enhance local absorption features, slope variations, and weak spectral differences while preserving the overall variation pattern of the original spectra, thereby improving the sensitivity of spectral information to changes in leaf nitrogen content.

Revised text: L300-316: Fractional-order derivative transformation was applied to the preprocessed canopy reflectance. In this study, multiple fractional orders were used to transform canopy reflectance. The FOD orders were set to 0, 0.5, 1.0, 1.5, 2.0, and 2.5. Through FOD treatments at different orders, multiple fractional-order derivative spectra were obtained and used for subsequent construction of two-band and three-band spectral indices.

Location: Section 2.4, L300-316.

Annotated comment 17:meaning of k?

Response: Thank you for pointing this out. We agree that the meaning of (k) in the fractional-order derivative formula should be defined explicitly. We added the definition of (k) as the summation index ranging from 0 to (m) in the explanation following Equation (1).

Original text:where Dαf(λi) is the fractional-order derivative value at λi; α is the fractional order; h is the spectral sampling interval; m is the number of preceding bands involved in the calculation; Γ(·) is the Gamma function; and f(λi−kh) represents the spectral reflectance at the corresponding wavelength.

Revised text: L308-313: where Dαf(λi) is the fractional-order derivative value at λi; α is the fractional order; h is the spectral sampling interval; k is the summation index, ranging from 0 to m; m is the number of preceding bands involved in the calculation; Γ(·) is the Gamma function; and f(λi−kh) represents the spectral reflectance at the corresponding wavelength.

Location: Section 2.4, L308-313.

Annotated comment 18:give names of indices in full at first mentioning

Response: Thank you for this comment. We agree that the abbreviations of the traditional vegetation indices should be defined when they first appear. We revised Section 2.5 by adding the full names of the selected traditional vegetation indices at their first mention.

Original text:Based on previous studies and the requirements of potato canopy nitrogen diagnosis, the selected indices included NDVI, NDRE, OSAVI, GNDVI, CI, CCI1, CCI2, PRI1, SR1, SR3, SR705, and SR680, which are closely related to chlorophyll or nitrogen status [23][24][25][26][27][28].

Revised text: L324-330: Based on previous studies and the requirements of potato canopy nitrogen diagnosis, the selected indices included the normalized difference vegetation index (NDVI), normalized difference red-edge index (NDRE), optimized soil-adjusted vegetation index (OSAVI), green normalized difference vegetation index (GNDVI), chlorophyll index (CI), chlorophyll content indices (CCI1 and CCI2), photochemical reflectance index (PRI1), simple ratio indices (SR1 and SR3), and simple ratio indices centered at 705 and 680 nm (SR705 and SR680), which are closely related to chlorophyll or nitrogen status [23][24][25][26][27][28].

Location: Section 2.5, L324-330.

Annotated comment 19:The term ‘two-band’ is clearer than ‘two-dimensional’; NDVI and other ‘traditional’ indices use two or more wavebands as well as normalization.

Response: Thank you for this helpful comment. We agree that “two-band” is clearer than “two-dimensional” in this context. To avoid ambiguity, we revised the terminology in Section 2.6 and replaced “two-dimensional spectral indices” with “two-band spectral indices”. We also clarified that these indices were constructed by traversing all possible two-band combinations across the full spectral range using difference, normalized difference, and ratio operations.

Original text:2.6. Construction of Two-dimensional Spectral Indices

Revised text:2.6. Construction of Two-Band Spectral Indices

Original text:To overcome the dependence of traditional vegetation indices on fixed bands, two-dimensional spectral indices were constructed using full-band combinations. These indices expand the sensitive wavelength search space by applying difference, normalized difference, and ratio operations to arbitrary two-band combinations, thereby allowing band pairs more closely related to layer-specific LNC to be identified [29].

Revised text: L337-343: To overcome the dependence of traditional vegetation indices on fixed bands, two-band spectral indices were constructed by traversing all possible two-band combinations across the full spectral range. These indices expanded the sensitive wavelength search space by applying difference, normalized difference, and ratio operations to arbitrary band pairs, thereby allowing band combinations more closely related to layer-specific LNC to be identified [29].

Location: Section 2.6, L337-343.

Annotated comment 20:“Three-bandis” clearer thanThree-dimensional”

Response: Thank you for this helpful comment. We agree that “three-band” is clearer than “three-dimensional” in this context, because the indices were constructed by introducing a third wavelength rather than by using spatial three-dimensional information. Therefore, we revised the terminology in Section 2.7 and replaced “three-dimensional spectral indices” with “three-band spectral indices” where the index construction method was described. We also clarified that the method screened candidate three-band combinations from the selected sensitive bands.

Original text:2.7. Construction of Three-dimensional Spectral Indices

Revised text:2.7. Construction of Three-Band Spectral Indices

Original text:On the basis of the two-dimensional spectral indices, a third band was introduced to construct three-dimensional spectral indices. The purpose was to enhance the multi-band representation of spectral information related to layer-specific LNC. Compared with two-band combinations, three-dimensional spectral indices can integrate information from three wavelengths and may better characterize the combined effects of pigment absorption, red-edge variation, near-infrared structural scattering, and shortwave-infrared biochemical features [29][30].

Revised text: L358-365: On the basis of the two-band spectral indices, a third band was introduced to construct three-band spectral indices. The purpose was to enhance the multi-band representation of spectral information related to layer-specific LNC. Compared with two-band combinations, three-band spectral indices can integrate information from three wavelengths and may better characterize the combined effects of pigment absorption, red-edge variation, near-infrared structural scattering, and shortwave-infrared biochemical features [29][30].

Location: Section 2.7, L358-365.

Annotated comment 21:The reviewer highlighted the paragraph describing the vertical distribution pattern of LNC in Section 3.1.

Response: Thank you for highlighting this paragraph. We agree that the original description was repetitive and overemphasized the vertical gradient of LNC. Therefore, we shortened this part of the Results section and retained only the key result that LNC decreased from the Top to the Middle and Bottom canopy layers. We also revised the interpretation to state more cautiously that layer-specific LNC measurements provide additional information beyond a single canopy-averaged nitrogen indicator.

Original text:The distribution pattern of LNC among the three canopy layers suggested that the upper leaves maintained a higher nitrogen status, whereas the middle and lower leaves exhibited progressively lower LNC values. This vertical difference provided the physiological basis for separately estimating LNC at different canopy layers rather than using a single canopy-averaged nitrogen indicator. As shown in Figure 3, LNC decreased from the top canopy layer to the middle and bottom layers, confirming the vertical heterogeneity of nitrogen distribution within the potato canopy.

Revised text: L457-461: The LNC of potato leaves decreased vertically from the Top to Middle and Bottom canopy layers, following the order Top LNC > Middle LNC > Bottom LNC (Figure 2). This vertical gradient indicated that layer-specific LNC measurements provided additional information beyond a single canopy-averaged nitrogen indicator.

Location: Section 3.1, L454-461.

Annotated comment 22:The reviewer highlighted the paragraph describing the response of traditional vegetation indices to LNC at different canopy layers.

Response: Thank you for highlighting this paragraph. This issue has been addressed in the revised manuscript during the previous revision of Section 3.1. The original repetitive description was shortened, and the revised Results section now focuses on the key layer-dependent response pattern of traditional vegetation indices. Specifically, the revised text reports that the correlations between traditional vegetation indices and LNC were strongest for Top LNC, decreased for Middle LNC, and were lowest for Bottom LNC. The final summary sentence was also simplified to state that traditional vegetation indices reflected LNC variation to some extent, but their correlations decreased from the upper to lower canopy layers.

Location: Section 3.1

Annotated comment 23:The reviewer highlighted the paragraph describing the correlation response of two-dimensional spectral indices under different FOD orders.

Response: Thank you for highlighting this paragraph. Following the reviewer’s previous suggestion that “two-band” is clearer than “two-dimensional”, we revised the terminology in this part of the Results section. Specifically, “two-dimensional spectral indices” and “two-dimensional indices” were replaced with “two-band spectral indices” and “two-band indices”. We also replaced “optimal” with “selected” where the text referred to data-driven band combinations, to avoid implying universal optimality.

Original text:Compared with traditional vegetation indices, two-dimensional spectral indices expanded the search space of sensitive bands by using all possible two-band combinations. The correlations between DI, NDI, RI, and LNC varied substantially among FOD orders and canopy layers, indicating that fractional-order derivative transformation changed the spectral response of LNC. Figure 4 shows the mean spectral curves after different fractional-order derivative treatments, indicating that FOD transformation altered the spectral amplitude and local variation patterns of the original reflectance spectra. Under the original reflectance condition, two-dimensional indices already showed stronger correlations with layered LNC than traditional vegetation indices. At FOD0, the optimal two-dimensional index for Top LNC was RI, with the optimal band combination of 979 and 996 nm and an r value of 0.814. For Middle LNC, the optimal two-dimensional index was NDI or RI, with the optimal band combination of 977 and 976 nm and an r value of 0.778. For Bottom LNC, the optimal two-dimensional index was RI, with the optimal band combination of 374 and 350 nm and an r value of -0.618. These results showed that full-band two-dimensional combinations had stronger correlations with LNC than traditional vegetation indices.

Revised text: L497-504: Compared with traditional vegetation indices, two-band spectral indices expanded the search space of sensitive bands by using all possible two-band combinations. The correlations between DI, NDI, RI, and LNC varied substantially among FOD orders and canopy layers, indicating that fractional-order derivative transformation changed the spectral response of LNC. Figure 3 shows the mean spectral curves after different fractional-order derivative treatments, indicating that FOD transformation altered the spectral amplitude and local variation patterns of the original reflectance spectra.

Location: Section 3.2, L497-504.

Annotated comment 24:The reviewer highlighted the statement identifying FOD1.5 as the optimal order for two-dimensional spectral indices.

Response: Thank you for highlighting this statement. This issue has been addressed in the revised manuscript. We revised the original wording to avoid overemphasizing universal optimality and now state more cautiously that FOD1.5 showed the strongest overall correlations among the tested FOD orders. Following the reviewer’s terminology suggestion, we also replaced “two-band indices” with “two-band indices” in this paragraph.

Original text:Therefore, FOD1.5 was identified as the optimal order for enhancing the response of two-dimensional spectral indices to layered LNC.

Revised text: L512-515: Among the tested FOD orders, FOD1.5 showed the strongest overall correlations for the two-band spectral indices. Under FOD1.5, the highest |r| values reached 0.855, 0.849, and 0.814 for Top, Middle, and Bottom LNC, respectively.

Location: Section 3.2, L512-515.

Annotated comment 25:The reviewer highlighted the paragraph describing the results of three-dimensional spectral indices in Section 3.3.

Response: Thank you for highlighting this paragraph. Following the reviewer’s terminology suggestion, we revised this part of the Results section to make the number of bands clearer. Specifically, “three-dimensional spectral indices” and “two-dimensional indices” were replaced with “three-band spectral indices” and “two-band indices”, respectively. We also revised the section title and figure captions accordingly, and replaced “optimal” with “selected” where appropriate to avoid implying universal optimality.

Original text:3.3. Optimal band combinations of three-dimensional spectral indices under different FOD orders

Revised text: L548-549: Three-band spectral indices showed stronger overall correlations with layered LNC than two-band indices, particularly under FOD1.5.

Original text:Three-dimensional spectral indices showed stronger overall correlations with layered LNC than two-dimensional indices, particularly under FOD1.5.

Revised text: L548-549: Three-band spectral indices showed stronger overall correlations with layered LNC than two-band indices, particularly under FOD1.5.

Location: Section 3.3 and Figure 5 caption.

Annotated comment 26:The reviewer highlighted the terminology used for the PLSR model comparison and Figure 7.

Response: Thank you for highlighting this point. Following the reviewer’s terminology suggestions, we revised the wording in Section 3.4 and the caption of the model-performance figure. Specifically, “optimal two-dimensional FOD indices” and “optimal three-dimensional FOD indices” were replaced with “selected two-band FOD indices” and “selected three-band FOD indices”, respectively. The model-performance figure caption, now Figure 6 after renumbering, was also revised accordingly to keep the terminology consistent throughout the manuscript.

Original text 1:To further evaluate the predictive ability of different spectral index systems, PLSR models were established for Top, Middle, and Bottom LNC using traditional vegetation indices, optimal two-dimensional FOD indices, and optimal three-dimensional FOD indices as input variables.

Revised text 1: L584-587: To further evaluate the predictive ability of different spectral index systems, PLSR models were established for Top, Middle, and Bottom LNC using traditional vegetation indices, selected two-band FOD indices, and selected three-band FOD indices as input variables.

Original text 2:Figure 7. Comparison of model performance for estimating layer-specific LNC using traditional vegetation indices, FOD1.5-based optimal two-dimensional spectral indices, and FOD1.5-based optimal three-dimensional spectral indices.

Revised text 2: L612-615: Figure 6. Comparison of model performance for estimating layer-specific LNC using traditional vegetation indices, FOD1.5-based selected two-band spectral indices, and FOD1.5-based selected three-band spectral indices.

Location: Section 3.4 and Figure 6 caption.

Annotated comment 27:The reviewer noted that Figure 8 provides information similar to the model-performance metrics shown in Figure 7.

Response: Thank you for this comment. We agree that the original description of Figure 8 repeated information already presented in the quantitative model-performance comparison. Therefore, we shortened the corresponding Results text and avoided repeated interpretation of the scatter plots. In the revised manuscript, Figure 7 is retained only as a visual supplement to the quantitative model-performance metrics shown in Figure 6.

Original text:As shown in Figure 8, the scatter plots between measured and predicted LNC further confirmed the superior performance of the 3D-FOD1.5 model. Compared with the traditional vegetation index and 2D-FOD1.5 models, the predicted values of the 3D-FOD1.5 model were more closely distributed around the 1:1 line, especially for Top and Bottom LNC.

Revised text: L599-600: Figure 7 is retained only as a visual supplement to the quantitative model-performance metrics shown in Figure 6.

Location: Section 3.4, L599-600.

4. Response to Comments on the Quality of English Language

Point 1:The English is fine and does not require any improvement.

Response 1: Thank you for the evaluation. We checked the manuscript and made minor language and terminology edits where necessary.

Response to Reviewer 3 Comments

1. Summary

Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files. [This is only a recommended summary. Please feel free to adjust it. We do suggest maintaining a neutral tone and thanking the reviewers for their contribution although the comments may be negative or off-target. If you disagree with the reviewer’s comments please include any concerns you may have in the letter to the Academic Editor.]

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Yes

The background and terminology were revised, including replacement of two-dimensional/three-dimensional wording with two-band/three-band terminology where appropriate.

Are all the cited references relevant to the research?

Can be improved

Reference numbering was checked and corrected throughout the manuscript.

Is the research design appropriate?

Yes

The index-selection procedure, PLSR inputs, and statistical analysis strategy were clarified.

Are the methods adequately described?

Can be improved

Methods were expanded to define sensitive bands, PLSR input variables, ANOVA/Tukey testing, and year-independent validation.

Are the results clearly presented?

Can be improved

Figures 4-7, Tables 1-3, heatmap captions, and statistical annotations were revised for clarity and consistency.

Are the conclusions supported by the results?

Yes

The Discussion and Conclusions were revised to strengthen limitations and transferability to UAV/satellite remote sensing.

3. Point-by-point response to Comments and Suggestions for Authors

This manuscript is devoted to the development of tools for remote estimation of leaf nitrogen content (LNC) in potato leaves under varying light conditions, which represents a timely and practically relevant task. The authors employed detailed field measurements in combination with hyperspectral observations. The novelty of the study lies in the integration of Fractional-order Derivative Spectral Transformation with two- and three-dimensional spectral indices, complemented by multivariate modeling using machine learning techniques. The authors achieved strong estimation accuracy, particularly for the most shaded leaves. The paper is well-structured and the overall approach is sound.

I have only a few comments that require further clarification and correction:

Throughout the manuscript, the numbering of references does not match the reference list.

Response: Thank you for pointing this out. We carefully checked the numbering of all in-text citations and the corresponding entries in the reference list throughout the manuscript. The mismatched reference numbers were corrected, and the reference list was updated to ensure that each citation number in the text corresponds to the correct source in the reference list. Particular attention was paid to the references cited in the Materials and Methods section, including those related to field spectroscopy, reflectance measurement, spectral smoothing, and fractional-order derivative calculation.

Section 2.7. Please clarify what is meant by “sensitive bands” for which the spectral indices were constructed.

Response: Thank you for this helpful comment. We have clarified the meaning of “sensitive bands” in Section 2.7. In the revised manuscript, sensitive bands are defined as wavelengths whose reflectance or fractional-order derivative values showed relatively strong Pearson correlations with layer-specific LNC under a given FOD order and canopy layer. We also clarified that the top K sensitive bands ranked by the absolute correlation coefficient were retained to generate candidate three-band combinations.

L376-380: In this study, sensitive bands refer to the wavelengths whose reflectance or fractional-order derivative values showed relatively strong Pearson correlations with layer-specific LNC under a given FOD order and canopy layer. For each FOD order, candidate three-band combinations were screened from the top K sensitive bands (K = 200).

“Section 2.9. The procedure for selecting two- and three-band spectral indices for the PLSR models needs to be detailed. How many indices were included in each model? Were indices selected based on the highest correlation in univariate relationships, or were metrics such as variance inflation factor (VIF) used to account for multicollinearity among variables?”

Response: Thank you for this helpful comment. We have clarified the procedure for selecting the two-band and three-band spectral indices used in the PLSR models in Section 2.9. In the revised manuscript, we now state that each two-band or three-band PLSR model included three selected spectral index variables, namely DI, NDI, and RI. For each canopy layer and FOD order, the selected DI, NDI, and RI were determined according to the highest absolute Pearson correlation coefficient with the corresponding layer-specific LNC. Therefore, the index selection was based on univariate Pearson correlation analysis. VIF-based variable screening was not applied, because the purpose of this study was to compare predefined spectral-index systems, and PLSR is suitable for handling collinearity among predictor variables through latent-variable extraction.

L417-424: For the two-band and three-band index models, three index variables were included for each canopy layer, namely DI, NDI, and RI. Specifically, for each canopy layer and FOD order, the two-band DI, NDI, and RI with the highest absolute Pearson correlation coefficients with LNC were retained as the selected two-band indices. Similarly, the three-band DI, NDI, and RI with the highest absolute Pearson correlation coefficients with LNC were retained as the selected three-band indices. Therefore, each two-band PLSR model and each three-band PLSR model used three selected spectral indices as input variables.

Section 3.1. Although the differences in LNC among canopy layers appear substantial, it would be beneficial to supplement this section with a formal statistical analysis of these differences (e.g., ANOVA with compact letter display on Figure 3).

Response: Thank you for this helpful suggestion. We performed a one-way ANOVA to test the differences in LNC among the Top, Middle, and Bottom canopy layers, followed by Tukey’s multiple comparison test. The results showed that LNC differed significantly among the three canopy layers (F = 1587.146, P < 0.001), and the multiple comparison results indicated that Top, Middle, and Bottom LNC were significantly different from each other. We have added the statistical analysis method to the Materials and Methods section, added the ANOVA result to Section 3.1, and added compact letter displays to Figure 2.

L437-441: Differences in LNC among the Top, Middle, and Bottom canopy layers were tested using one-way analysis of variance (ANOVA), followed by Tukey’s multiple comparison test. Significant differences were indicated using compact letter displays at P < 0.05.

L454-458: One-way ANOVA followed by Tukey’s multiple comparison test showed that LNC differed significantly among the three canopy layers (F = 1587.146, P < 0.001). The compact letter display in Figure 2 indicates that Top, Middle, and Bottom LNC were significantly different from each other.

L489-491: Figure 2. Distribution of leaf nitrogen content in different potato canopy layers. Different lowercase letters indicate significant differences among canopy layers at P < 0.05.

Tables 1–3. Since all presented correlation coefficients are statistically significant, the use of double asterisks seems redundant. The significance of all coefficients can simply be noted in the table captions.

Response: Thank you for this helpful suggestion. We agree that the use of double asterisks is redundant because all correlation coefficients presented in Tables 1–3 are statistically significant. Therefore, we removed the double asterisks from all correlation coefficients in Tables 1–3 and revised the table notes to state that all correlation coefficients are statistically significant at P < 0.05.

For Table 1, the note was revised to: Note: All correlation coefficients are statistically significant at P < 0.05. Top LNC, Middle LNC, and Bottom LNC represent the upper, middle, and lower canopy layers, respectively.

For Table 2, the note was revised to: Note: Values in parentheses represent the selected band combination (i, j). All correlation coefficients are statistically significant at P < 0.05.

For Table 3, the note was revised to: Note: Values in parentheses represent the selected three-band combination (i, j, k). All correlation coefficients are statistically significant at P < 0.05.

Sections 3.2–3.3. Avoid using the term “spatial” to describe the heatmaps. These represent distributions in spectral space, not in physical space.

Response: Thank you for this helpful comment. We agree that the heatmaps represent distributions in spectral space rather than physical space. Therefore, we revised the wording in Sections 3.2 and 3.3 by replacing “spatial distribution” with “spectral distribution” when describing the correlation heatmaps.

The correlation heatmaps further showed that FOD transformation changed the spectral distribution of sensitive band combinations.

L518-519: The spectral distribution of correlation coefficients between the two-band spectral indices and layer-specific LNC is shown in Figure 4.

L556-557: The spectral distribution of correlation coefficients for the three-band spectral indices is shown in Figure 5.

Location: Sections 3.2–3.3.

Figures 5–6. The rationale for separating panels (a) and (b) is unclear; it would be preferable to combine them into a single figure. This would also eliminate the complete duplication of panel descriptions in the caption.

Response: Thank you for this helpful suggestion. We agree that separating the heatmaps into panels (a) and (b) was not necessary and resulted in repeated caption descriptions. Therefore, we combined the original two-band heatmaps into a single Figure 4 and the original three-band heatmaps into a single Figure 5. The captions were revised accordingly to provide one unified description for each figure, thereby avoiding duplicated panel descriptions.

Figure 6. Please explain in the text and the caption how the planes displayed in the 3D plot were selected.

Response: Thank you for this helpful comment. We have clarified how the displayed planes in Figure 5 were selected. Because the three-band spectral indices form a three-dimensional correlation space involving band i, band j, and band k, only selected slice planes can be visualized in the heatmaps. In the revised manuscript, we explain that, for each canopy layer, FOD order, and index type, the displayed plane was selected by fixing band k at the wavelength included in the selected three-band combination with the highest absolute Pearson correlation coefficient. The heatmap then shows the correlation distribution over band i and band j under the selected k-band slice. This explanation has been added to both Section 3.3 and the caption of Figure 5.

Revision made: L556-563: For visualization of the three-band correlation space, the displayed plane in each heatmap was selected by fixing the k-band at the wavelength included in the selected three-band combination with the highest absolute Pearson correlation coefficient for the corresponding canopy layer, FOD order, and index type. The heatmap therefore represents the slice-based correlation distribution over band i and band j under the selected k-band slice.

L571-579: Figure 5. Correlation heatmaps between layer-specific LNC and selected three-band spectral indices under different FOD treatments. The columns represent the Top, Middle, and Bottom canopy layers, respectively, and the rows represent different FOD treatments from FOD0 to FOD2.5. For each heatmap, the displayed slice planes were selected according to the selected three-band combination with the highest absolute Pearson correlation coefficient for the corresponding canopy layer, FOD order, and index type. The axes indicate bands i, j, and k, and the color scale represents Pearson’s correlation coefficient (r). DI, NDI, and RI denote the selected three-band index types for each canopy layer and FOD treatment.

Location: Section 3.3 and Figure 5 caption.

Figure 7. I suspect that the metrics for the training set and the test set may have been swapped. I have never encountered a situation where test set metrics outperform training set metrics. The values mentioned in the text also point to this error. Please verify this.

Response: Thank you for pointing this out. We carefully rechecked the original model-output table and the data used to generate the model-performance figure. The calibration/training-set and testing-set labels were verified and were not swapped. The testing-set R² values reported in the text, namely 0.615, 0.491, and 0.402 for the traditional vegetation index models and 0.803, 0.706, and 0.681 for the 3D-FOD1.5 models, correspond to the testing-set results in the original model output. The Bottom LNC RMSE values reported in the text, decreasing from 0.159 to 0.113, also correspond to the testing-set results. In the revised manuscript, the model-performance figure is Figure 6 after figure renumbering.

In several cases, some testing-set metrics were slightly better than the corresponding training-set metrics. This was mainly because the dataset was relatively small and the testing subset had a different sample distribution from the training subset under the random 2:1 partition. To avoid possible misunderstanding, we rechecked Figure 7 and the corresponding text in Section 3.4 and ensured that the training-set and testing-set metrics are correctly labeled and consistently reported.

Revision made: The data used for Figure 6 and the corresponding descriptions in Section 3.4 were rechecked. The calibration/training-set and testing-set labels were confirmed to be correctly assigned.

Discussion. The authors mention that “selected wavelength combinations and modeling framework should be further evaluated using UAV hyperspectral platforms or satellite sensors at operational production scales.” I recommend expanding on this point, because in potential practical applications it is remote sensing, rather than field measurements, that would be used. The transferability of the present findings to remote sensing data therefore requires a more thorough discussion.

Response: Thank you for this important suggestion. We agree that the practical application of the proposed framework would rely more on UAV- or satellite-based remote sensing than on plot-level ground measurements. Therefore, we expanded the Discussion to clarify the limitations of transferring the present ground-based findings to operational remote-sensing platforms. In the revised manuscript, we now discuss that UAV and satellite observations may be affected by spatial resolution, spectral bandwidth, atmospheric correction, illumination geometry, viewing angle, canopy background, mixed pixels, and sensor noise. We also emphasize that the selected wavelengths may not be directly available in current operational sensors, and that band resampling may influence index performance. Therefore, further validation using UAV hyperspectral data, sensor-band matching, and multi-date, multi-cultivar, and multi-management field tests is required before the proposed framework can be used for operational potato nitrogen diagnosis.

L835-851: Revision made: Finally, this study was based on ground-based canopy hyperspectral observations obtained under relatively controlled field measurement conditions. For practical nitrogen diagnosis, however, UAV- or satellite-based remote sensing would be more relevant than plot-level ground measurements. Therefore, the transferability of the selected wavelength combinations and modeling framework to airborne or spaceborne observations requires further evaluation. Compared with ground-based spectra, UAV and satellite data are affected by differences in spatial resolution, spectral bandwidth, atmospheric correction, illumination geometry, viewing angle, canopy background, mixed pixels, and sensor noise. In addition, not all selected wavelengths in this study may be available in current multispectral or hyperspectral sensors, and band resampling may change the performance of the selected indices. Future studies should therefore test the proposed framework using UAV hyperspectral data first, evaluate whether the selected bands can be matched or approximated by operational sensor bands, and validate the models across different flight heights, imaging dates, cultivars, growth stages, and nitrogen management conditions. Such validation is necessary before the present ground-based findings can be transferred to operational remote-sensing applications for potato nitrogen diagnosis.

4. Response to Comments on the Quality of English Language

Point 1:The English is fine and does not require any improvement.

Response 1: Thank you for the evaluation. We checked the manuscript and made minor language and terminology edits where necessary.

Response to Reviewer 2 Comments

1. Summary

Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files. We have carefully addressed the remaining concerns regarding within-canopy LNC distribution, mixed top-of-canopy reflectance, cross-year validation, terminology, the scope of application, and the interpretation of layer-specific LNC estimation.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Quality of English Language

The English is fine and does not require any improvement.

No major language revision was required. We made targeted wording changes to improve clarity and avoid overstatement.

Does the introduction provide sufficient background and include all relevant references?

Yes

The Introduction was shortened and refocused on vertical LNC heterogeneity, integrated top-of-canopy reflectance, and statistical estimation of layer-specific LNC.

Is the research design appropriate?

Can be improved

The objectives, sampling window, experimental unit, layer-specific LNC sampling, and cross-year validation were clarified.

Are the methods adequately described?

Can be improved

Sections 2.2, 2.3, and 2.9 were expanded with details on stratified sampling, canopy spectral acquisition, technical replicates, ANOVA/Tukey’s test, LNC ratios, and the empirical attenuation coefficient.

Are the results clearly presented?

Can be improved

The Results were condensed, Table 1 was added, Figure 7 was removed, Figure 4 visibility was improved, and cross-year validation was described more cautiously.

Are the conclusions supported by the results?

Can be improved

The Conclusions were shortened and restricted to the main findings, cautious interpretation of Bottom LNC estimation, and limited temporal generalization.

Are all figures and tables clear and well-presented?

Yes

The graphical abstract/workflow was removed, Figure 4 line visibility was improved, Figure 5 was described as heat maps, and redundant figures or repeated descriptions were removed.

3. Point-by-point response to Comments and Suggestions for Authors

General Comment 1: There are no details on LNC of the three leaf layers (depending on N fertilizer level), no approach to model the vertical LNC distribution among leaf levels.

Response: Thank you for this important comment. We agree that the previous version did not provide sufficient details on the LNC of the three canopy layers under different nitrogen fertilizer levels, nor did it include a quantitative approach to describe the vertical LNC distribution among leaf layers.

To address this issue, we revised the objective statement in the Introduction to clarify that one objective of this study was to analyze the vertical distribution characteristics of LNC in upper, middle, and lower potato canopy leaves under different nitrogen fertilizer levels and to quantify this vertical pattern using relative LNC ratios and an empirical attenuation coefficient.

In the Materials and Methods section, we added the calculation of Middle/Top (%) and Bottom/Top (%) LNC ratios for each nitrogen fertilizer level. We also introduced an empirical log-linear model, ln(LNC_layer) = a - k x Layer, where Layer was coded as 0, 1, and 2 for Top, Middle, and Bottom leaves, respectively. The coefficient k was used as a descriptive indicator of the steepness of the vertical LNC decline.

In the Results section, we added Table 1 to present Top, Middle, and Bottom LNC under N0-N4, together with Middle/Top, Bottom/Top, and k. The results showed that Top LNC was consistently higher than Middle and Bottom LNC under all nitrogen levels. Middle LNC accounted for 70.3%-73.8% of Top LNC, and Bottom LNC accounted for 51.9%-52.9% of Top LNC.

L206-217: The specific objectives were revised to include the analysis of LNC vertical distribution under different nitrogen fertilizer levels and the quantification of this pattern using relative LNC ratios and an empirical attenuation coefficient.

L459-471: Middle/Top (%), Bottom/Top (%), and the empirical vertical attenuation coefficient k were added in Section 2.9.

L493-504: The LNC values of Top, Middle, and Bottom leaves were summarized separately under each nitrogen treatment, and the relative LNC ratios and k values were reported in Table 1.

General Comment 2: The question on the contribution of middle and lower leaf levels to the overall reflectance signal measured for the canopy by a non-imaging spectrometer operated above the canopy remains untangled.

Response: Thank you for this important comment. We agree that the canopy reflectance measured by a non-imaging spectrometer positioned above the canopy represents a mixed top-of-canopy signal, and that the present dataset cannot quantitatively separate the individual reflectance contributions of the Top, Middle, and Bottom leaf layers.

To avoid overinterpretation, we revised the manuscript to clarify this point explicitly. The Methods now state that the measured reflectance was a plot-scale top-of-canopy mixed signal integrating reflectance components from different leaf layers, leaf angles, shadowed leaves, canopy gaps, and background effects.

We further clarified that the spectral models were used to evaluate statistical relationships between mixed canopy reflectance and measured layer-specific LNC, rather than to physically or quantitatively separate the reflectance contributions of different canopy layers.

L307-315: The measured reflectance was defined as a plot-scale top-of-canopy mixed signal, and the models were described as statistical rather than physically separating layer-specific reflectance contributions.

L813-823: The improved Bottom LNC prediction was interpreted as enhanced statistical sensitivity to canopy-scale spectral variations rather than quantitative separation of lower-leaf optical contribution.

General Comment 3: The authors have added a test on the transfer of their most-promising spectral indices from the year 2022 to 2023 and vice versa. The low correlation coefficients indicate a substantial problem in generalizing the authors’ approach to use spectral information from nadir recordings of canopy reflectance for the assessment of the vertical LNC distribution within crop canopies.

Response: Thank you for this important comment. We agree that the cross-year validation results revealed an important limitation of the proposed approach. Although the selected FOD1.5-based three-band indices showed relatively high testing-set performance under random sample division, their performance decreased substantially when transferred between years.

To address this issue, we replaced the term “year-independent validation” with “cross-year validation” throughout the manuscript. We also revised the Results to explicitly compare the cross-year validation results with the random calibration/testing split.

In the revised manuscript, we state that the mean R2 decreased from 0.730 under the random calibration/testing split to 0.327 in the 2022-to-2023 validation and 0.348 in the 2023-to-2022 validation. This reduction indicates limited temporal transferability of the selected indices for layer-specific LNC estimation.

L451-457: “Cross-year validation” was used to describe the validation strategy in which one year was used for calibration and the other year for testing.

L650-665: Cross-year validation results were reported and interpreted as evidence of limited temporal transferability; the selected FOD1.5-based three-band indices were described as data-supported spectral features requiring further validation.

General Comment 4: The revised Discussion includes some of the comments made by the review(s) and, therefore, has increased in length. Instead, the text has to be condensed and focused.

Response: Thank you for this helpful suggestion. We agree that the Discussion became too long after incorporating explanatory statements in response to previous comments. In the revised manuscript, we condensed and refocused the Discussion to avoid repetition and overextended interpretation.

Specifically, we shortened the discussion of vertical LNC gradients, mixed canopy reflectance, and Bottom LNC estimation. The revised Discussion now focuses on the central issues: vertical LNC heterogeneity, statistical estimation from mixed top-of-canopy reflectance, limited temporal transferability, and the need for additional physiological and structural information.

L696-723: The Discussion was refocused on the statistical association between integrated top-of-canopy reflectance and measured layer-specific LNC, rather than physical separation of layer-specific spectral components.

L824-837: The limitations were condensed to limited experimental scope, limited temporal transferability, and the need for additional physiological, structural, and radiative-transfer information.

General Comment 5: The chapter Conclusions has been revised but it is still a summary of the previous chapters. It may be deleted or the sentences on potential perspectives should be divided strictly between Discussion and Conclusions.

Response: Thank you for this helpful comment. We agree that the previous Conclusions section still contained repeated summary statements and future perspectives overlapping with the Discussion. We substantially revised and shortened the Conclusions section.

In the revised manuscript, the Conclusions focus only on the main findings: the vertical gradient of potato canopy LNC, improved response of FOD-based two-band indices, superior testing-set performance of the FOD1.5-based three-band model, cautious interpretation of Bottom LNC estimation, and limited temporal generalization indicated by cross-year validation. Broader future perspectives were kept in the Discussion.

L847-868: The Conclusions were shortened and restricted to the main findings, cautious interpretation of Bottom LNC estimation, and limited temporal generalization.

Specific Comment 1: L53 Graphical abstract / work flow (should be part of MaM) following the abstract is not necessary (was not included in the original submission) and is rather confusing because of the multitude of small graphs / details.

Response: Thank you for this comment. We agree that the graphical abstract/workflow placed after the Abstract was not necessary and may have caused confusion. We removed this graphical abstract/workflow from the revised manuscript. The methodological procedures are now described only in the Materials and Methods section.

The graphical abstract/workflow following the Abstract was deleted.

Specific Comment 2: L89-93 Being also a spectral method, hyperspectral information is an indirect proxy of LNC as well.

Response: Thank you for this helpful comment. We agree that hyperspectral reflectance does not directly measure LNC and should be interpreted as an indirect optical proxy of crop nitrogen status. We revised the Introduction accordingly and changed wording such as “reflect crop nitrogen status” to “indirectly estimate crop nitrogen status”.

L79-93: Hyperspectral reflectance was described as an indirect optical proxy related to nitrogen status through pigment absorption, red-edge variation, near-infrared structural scattering, and shortwave infrared biochemical absorption.

Specific Comment 3: L127-131 The reviewer’s comment has been completely integrated into the text - please shorten and refer to the problem itself.

Response: Thank you for this suggestion. We shortened this part of the Introduction and refocused it on the main research problem rather than integrating a long explanatory response into the manuscript.

L103-112: The Introduction now focuses on whether vertical LNC distribution should be examined statistically because canopy reflectance measured above the canopy represents an integrated mixed signal rather than layer-specific leaf reflectance.

Specific Comment 4: L148-155 LNC distribution among different leaf layers - depending on the N amount available? - has to be studied in detail before studying spectral options for indirect LNC assessment by layers. Is there a fixed ratio for LNC values among layers (as supposed by the data in Figure 2)?

Response: Thank you for this important comment. We agree that the vertical distribution of LNC among canopy layers under different nitrogen fertilizer levels should be clarified before evaluating spectral approaches for layer-specific LNC estimation.

We added Table 1 and calculated Middle/Top and Bottom/Top ratios under each nitrogen level. These ratios were used as descriptive indicators, not as an assumed fixed ratio. We also added the empirical attenuation coefficient k to describe the vertical LNC decline.

L493-504: Middle LNC accounted for 70.3%-73.8% of Top LNC, and Bottom LNC accounted for 51.9%-52.9% of Top LNC; k ranged from 0.318 to 0.328.

Specific Comment 5: L175 Similarly, new indices are not suitable to quantify the contribution of different leaf layers to the mixed signal recorded at the top of the canopy.

Response: Thank you for this important clarification. We agree that the newly constructed spectral indices cannot quantify the individual optical contributions of different canopy layers to the mixed top-of-canopy reflectance signal.

We revised the Introduction to describe three-band spectral indices as empirical tools for linking canopy-scale spectral variation with measured layer-specific LNC, rather than tools for physically separating the reflectance contributions of individual canopy layers.

L165-172: Three-band spectral indices were described as empirical tools for statistical estimation of Top, Middle, and Bottom LNC, but not for physically separating reflectance contributions of individual canopy layers.

Specific Comment 6: L204 Do you have studied the question what is the layer-specific spectral information in the mixed signal of canopy reflectance?

Response: Thank you for this important comment. We clarify that the present study did not physically identify or separate layer-specific spectral components within the mixed canopy signal.

The revised objective states that the study evaluated whether integrated top-of-canopy hyperspectral reflectance contains statistical information associated with measured Top, Middle, and Bottom LNC, and compared different spectral index systems for layer-specific LNC estimation.

L211-217: The objective was revised to evaluate statistical information associated with measured Top, Middle, and Bottom LNC and to assess Bottom LNC under canopy signal attenuation and mixing.

Specific Comment 7: L297 approximately (?) 63 and 69 days after planting. Why you use “approximately” here?

Response: Thank you for pointing this out. The word “approximately” was unnecessary because the days after planting were calculated from the planting dates. We deleted “approximately” to provide a more precise description of sampling time.

L253-256: Sampling was performed on 7 July 2022 and 8 July 2023, corresponding to 63 and 68 days after planting, respectively.

Specific Comment 8: L549 Why the authors call this validation “year-independent” when testing 2022 results on 2023 data and vice versa? It should be named “year-dependent” or “by year” instead.

Response: Thank you for this helpful comment. We agree that “year-independent validation” was not appropriate. Since the model was calibrated using data from one year and tested using data from the other year, we replaced this term with “cross-year validation” throughout the manuscript.

L451-457: “Cross-year validation” was used to describe the validation strategy.

Specific Comment 9: L573 Where information is given on ANOVA and Tukey’s test in MaM?

Response: Thank you for pointing this out. We added the information on ANOVA and Tukey’s multiple comparison test in Section 2.9 “Model Construction and Evaluation”.

L459-461: Differences in LNC among the Top, Middle, and Bottom canopy layers were tested using one-way analysis of variance (ANOVA), followed by Tukey’s multiple comparison test. Significant differences were indicated using compact letter displays at P < 0.05.

Specific Comment 10: L573-579 Please give the percentage values of LNC of middle and bottom leaf layer as compared to the top layer. The strong differences among layers result in a mixed reflectance signal with unknown contribution of reflectance values from the three layers.

Response: Thank you for this important comment. We added the Middle/Top and Bottom/Top LNC ratios in the Results and Table 1. We also clarified that the canopy spectrum measured from above the canopy represented a mixed top-of-canopy signal and that individual reflectance contributions of the three leaf layers could not be quantified from the present data.

L493-504: Middle/Top and Bottom/Top ratios were added, and the mixed top-of-canopy signal with unknown individual reflectance contributions was explicitly stated.

Specific Comment 11: L621 should be deleted as well.

Response: Thank you for this suggestion. We deleted this sentence from the Discussion section as requested.

The indicated sentence was deleted from the revised Discussion.

Specific Comment 12: L674-679 The sentence “The spectral distribution ... in Figure 4” may be deleted. The next sentence includes the same information.

Response: Thank you for this suggestion. We agree that this sentence was redundant because the following sentence already introduced the information shown in Figure 4. We deleted the repeated sentence.

L573-574: Figure 4 presents the correlation patterns under FOD0, FOD0.5, FOD1.0, FOD1.5, FOD2.0, and FOD2.5.

Specific Comment 13: L690 The green line (FOD2.0) and the red line (FOD1.5) are not visible in the figure.

Response: Thank you for pointing this out. We revised Figure 4 by increasing the line widths of the curves to improve their visibility. In particular, the FOD1.5 and FOD2.0 curves are now clearer in the revised figure.

Figure 4 was revised by increasing the curve line widths.

Specific Comment 14: L752 “The spectral distribution of correlation coefficients for the three-band spectral indices is shown in Figure 5”. Please rephrase: “Heat maps of ...”.

Response: Thank you for this suggestion. We rephrased the sentence as suggested to more accurately describe the presentation format of Figure 5.

L605-606: Heat maps of correlation coefficients between the three-band spectral indices and layer-specific LNC are shown in Figure 5.

Specific Comment 15: L823 The acronym 3D-FOD1.5 is no longer appropriate referring to the change from “three-dimensional indices” to “three-band(s) indices”; please revise throughout the manuscript.

Response: Thank you for this helpful comment. We agree that “3D-FOD1.5” was no longer appropriate after revising the terminology from “three-dimensional indices” to “three-band indices”.

We revised this terminology throughout the manuscript. “3D-FOD1.5 model”, “3D-FOD1.5 indices”, and “3D-FOD1.5 framework” were replaced with “FOD1.5-based three-band model”, “FOD1.5-based three-band indices”, and “FOD1.5-based three-band framework”, respectively.

L637-665: The model and index descriptions were revised to “FOD1.5-based two-band” and “FOD1.5-based three-band”.

Specific Comment 16: L846 The authors should decide whether to use Figure 6 or Figure 7 - both figures include similar information.

Response: Thank you for this suggestion. We agree that Figure 6 and Figure 7 contained partly overlapping information. To avoid redundancy and make the Results more concise, we retained Figure 6 because it directly compares model performance among spectral index systems, and removed the original Figure 7 and its corresponding description.

L637-649: Figure 6 is used to present the model-performance comparison; the redundant scatter-plot figure was removed.

Specific Comment 17: L857 “Year-independent” ??

Response: Thank you for pointing this out. We replaced “year-independent validation” with “cross-year validation” throughout the manuscript.

L451-457 and L652-665: The validation strategy is now consistently called cross-year validation.

Specific Comment 18: L858-864 The rather low R2 values for the cross-year validation (0.327 vs. 0.730 for the random division) of spectral indices underlines the problem in generalization of spectral indices for LNC assessment by leaf layers.

Response: Thank you for this important comment. We agree that the low R2 values obtained from cross-year validation indicate limited temporal generalization.

We revised the Results to explicitly compare cross-year validation with the random calibration/testing split. The selected FOD1.5-based three-band indices are described as data-supported spectral features under the present experimental conditions, while their stability requires further validation.

L658-665: The mean R2 decreased from 0.730 under the random calibration/testing split to 0.327 in the 2022-to-2023 validation and 0.348 in the 2023-to-2022 validation; the selected indices were interpreted cautiously.

Specific Comment 19: L930-939 Why spectral indices derived from nadir recording of top canopy reflectance should be suitable for LNC assessment of different leaf layers? This question is valid for “traditional” indices as well as for the indices introduced in this study.

Response: Thank you for this important comment. We agree that spectral indices derived from nadir top-of-canopy reflectance cannot directly measure the LNC of different leaf layers or quantify the individual optical contributions of Top, Middle, and Bottom leaves. This limitation applies to both traditional vegetation indices and the newly constructed FOD-based indices.

We revised the Discussion to state that all spectral indices used in this study were derived from an integrated top-of-canopy reflectance signal. Their use for layer-specific LNC estimation is based on statistical relationships between mixed canopy reflectance and measured layer-specific LNC, rather than direct measurement of layer-specific leaf reflectance.

L712-723: The Discussion now states that the improved performance of FOD-based two-band and three-band indices indicates enhanced representation of LNC-related spectral variation within the mixed canopy signal, rather than physical separation of layer-specific spectral components.

Specific Comment 20: L997-1004 This statement minimizes the value of your study to random.

Response: Thank you for this valuable comment. We agree that the previous wording was too negative and may have underestimated the value of the selected wavelength combinations.

We revised this part to provide a more balanced interpretation. The revised manuscript clarifies that the wavelength combinations were not randomly selected, but were identified through systematic correlation screening across FOD orders, index types, and canopy layers. We also emphasize that these bands are located in physiologically meaningful spectral regions.

L759-768: The selected wavelength combinations were described as systematically screened, data-supported, and physiologically interpretable spectral features, while their stability and broader applicability still require validation.

Specific Comment 21: L1067 “indirect” information ??

Response: Thank you for pointing this out. We agree that the phrase “indirect information” was ambiguous and could lead to misunderstanding.

In the revised manuscript, we replaced the expression “whether top-of-canopy hyperspectral reflectance contains indirect information related to Middle and Bottom LNC” with “evaluate the statistical association between integrated top-of-canopy hyperspectral reflectance and measured Middle and Bottom LNC”. This revision clarifies that the canopy hyperspectral signal was used to assess statistical relationships with measured layer-specific LNC, rather than to imply direct extraction of layer-specific optical information.

L125-129: Therefore, layer-specific LNC measurements are still necessary to establish ground-truth information on within-canopy nitrogen distribution and to evaluate the statistical association between integrated top-of-canopy hyperspectral reflectance and measured Middle and Bottom LNC.

Specific Comment 22: L1073-1079 In which way, the mixed signal of spectral reflectance of the canopy may be separated into layer-specific information without additional ground truth information or models on vertical N allocation within canopies?

Response: Thank you for this important comment. We agree that the mixed top-of-canopy reflectance signal cannot be physically separated into layer-specific spectral components without additional measurements or models, such as layer-specific reflectance, within-canopy light interception, vertical leaf area distribution, or radiative transfer/nitrogen allocation models.

To avoid overinterpretation, we clarified that this study did not attempt to separate the mixed canopy reflectance signal into layer-specific optical contributions. The spectral models were used to evaluate statistical relationships between mixed canopy reflectance and measured layer-specific LNC.

L307-315: The Methods explicitly state that the models were not intended to quantitatively separate the individual reflectance contributions of different canopy layers.

L813-823: The Discussion states that resolving individual contributions of different canopy layers would require additional measurements such as layer-specific reflectance, within-canopy light interception, leaf area distribution, or radiative transfer modeling.

Specific Comment 23: L1091- “year-independent” ?, 3D-FOD1.5 ? “candidate indices” ? See my comments above.

Response: Thank you for this comment. We revised these terms according to the reviewer’s previous comments. “Year-independent validation” was replaced with “cross-year validation”, and “3D-FOD1.5” was replaced with “FOD1.5-based three-band” to match the revised terminology of “three-band spectral indices”.

We also revised the interpretation of the selected indices to avoid underestimating their value. In the Results, the selected FOD1.5-based three-band indices are described as data-supported spectral features for layer-specific LNC estimation under the present experimental conditions, while their stability requires further validation.

L650-665: The revised Results use “cross-year validation”, “FOD1.5-based three-band indices”, and “data-supported spectral features”.

Specific Comment 24: L1118-1134 The authors speculate on the use of UAV recording without having solved the problem on vertical N distribution in plant canopies. The recording platform and the sensor type - whether non-imaging or imaging - are low interest as long as the biological background of within-canopy LNC distribution is not known. Furthermore, what does the UAV-based spectral information on LNC differences between leaf layers help farmers to optimize potato fertilization which is done as in-furrow fertilization before planting?

Response: Thank you for this important comment. We agree that the Discussion should not overemphasize UAV-based recording, sensor platforms, or operational fertilization management before the biological basis of within-canopy LNC distribution is better understood.

To address this issue, we removed the speculative statements about UAV or satellite hyperspectral applications and operational fertilization management. The revised Discussion was refocused on vertical canopy nitrogen status, functional heterogeneity, and the biological basis of within-canopy LNC distribution.

L838-846: These findings provide a methodological reference for assessing vertical canopy nitrogen status and functional heterogeneity in potato. Further studies should clarify the biological basis of within-canopy LNC distribution across different years, cultivars, growth stages, ecological regions, and nitrogen supply conditions.

Specific Comment 25: L1173 Please explain the meaning of “indirect information”.

Response: Thank you for this comment. We agree that “indirect information” was ambiguous and could be misunderstood. Our intention was to indicate that top-of-canopy hyperspectral reflectance does not directly represent the optical signal of Middle and Bottom leaves, but may be statistically associated with measured Middle and Bottom LNC because the canopy reflectance signal is affected by canopy structure, leaf overlap, shading, and within-canopy nitrogen distribution.

To avoid ambiguity, we removed the phrase “indirect information” and revised the sentence to describe the statistical association between integrated top-of-canopy hyperspectral reflectance and measured Middle and Bottom LNC.

L125-129: The phrase “indirect information” was replaced with “the statistical association between integrated top-of-canopy hyperspectral reflectance and measured Middle and Bottom LNC”.

Response to Reviewer 2 Comments

1. Summary

Thank you very much for taking the time to review this manuscript again. We sincerely appreciate the reviewer’s positive evaluation that the manuscript has been revised according to the previous comments and that our responses were adequate. In this revision, we made one targeted change in the Conclusion according to the reviewer’s suggestion. Specifically, we added a sentence emphasizing that combining spectral assessment of Top LNC with the observed relative LNC ratios among canopy layers (Top = 100, Middle ≈ 72.5, and Bottom ≈ 52.5) may be a promising approach for describing the vertical LNC distribution in potato canopies. The corresponding revision is highlighted in the re-submitted manuscript.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Yes

All evaluation items have been addressed. We checked the Introduction, Methods, Results, figures and tables, references, and Conclusion. The revised manuscript clarifies the study scope, experimental details, sampling and validation procedures, figure/table presentation, citation numbering, and adds the reviewer-recommended Conclusion sentence on using Top LNC with layer LNC ratios to describe vertical LNC distribution.

Are the methods adequately described?

Can be improved

Is the research design appropriate?

Can be improved

Are the results clearly presented?

Can be improved

Are all figures and tables clear and well-presented?

Yes

Are the conclusions supported by the results?

Can be improved

3. Point-by-point response to Comments and Suggestions for Authors

Comment 1: The manuscript has been revised according to the reviewer(s) comments and the authors have responded to the comments adequately. Although the correlation between spectral information (from the canopy top) and leaf layer-specific LNC values is rather low, the vertical within-canopy gradient (as shown in Table 1) may be used to quantify the LNC value of different leaf layers of potato canopies. The authors may add a sentence to the Conclusion, that the combination of spectral assessment of top canopy LNC and the known LNC ratio among leaf levels – top 100 – middle 72.5 – bottom 52.5 – may be the most promising approach to describe the vertical LNC distribution in a potato crop.

Response 1: Thank you very much for the positive evaluation and constructive suggestion. We agree that the observed vertical LNC gradient and the relative LNC ratios among canopy layers provide useful information for describing the vertical LNC distribution in potato canopies. According to the reviewer’s suggestion, we added a sentence to the Conclusion to emphasize that combining spectral assessment of Top LNC with the observed relative LNC ratios among canopy layers may be a promising approach for describing vertical LNC distribution in potato.

Added sentence: Section 5, Conclusions, after L849–851: These results suggest that combining spectral assessment of Top LNC with the observed relative LNC ratios among canopy layers, namely Top = 100, Middle ≈ 72.5, and Bottom ≈ 52.5, may be a promising approach for describing the vertical LNC distribution in potato canopies.

Location: Section 5, Conclusions; added after the sentence reporting the vertical LNC gradient (Top LNC > Middle LNC > Bottom LNC).

4. Response to Comments on the Quality of English Language

Point 1:The English is fine and does not require any improvement.

Response 1: Thank you for the evaluation. We checked the manuscript and made minor language and terminology edits where necessary.