Spectral Phenology, Climate, and Topography as Determinants of Vigor, Yield, and Fruit Quality in Avocado (cv. Semil-34)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Summary

The manuscript entitled “Spectral phenology, climate, and topography as determinants of vigor, yield, and fruit quality in avocado (Persea americana cv. Semil-34) on tropical hillsides” presents a comprehensive analysis integrating multi-year Sentinel-2 spectral indices, ERA5-Land climate reanalysis data, and geomorphometric variables to explain yield and fruit quality variability in a tropical hillside avocado system.

The study addresses an important gap by focusing on a tropical cultivar (cv. Semil-34) grown outside the well-studied ‘Hass’ systems and under complex topographic conditions. The identification of a cultivar-specific “vigor paradox” and the demonstrated hierarchical role of topography over spectral vigor signals represent clear scientific novelties. The methodology is generally robust, and the results are clearly structured.

Overall, the manuscript is of good scientific quality and is suitable for publication after minor to moderate revisions aimed at improving clarity, formatting consistency, and the depth of physiological interpretation.

General Comments

The manuscript is well written and technically sound, with a strong integration of remote sensing, climate, and topographic analyses. The experimental design, multivariate statistical framework, and interpretation of the main findings are appropriate for the aims of the study.

However, several aspects require improvement before publication:

• Stronger conceptual framing of phenology–environment–physiology interactions;
• Improved numerical precision and consistency in statistical reporting;
• Clearer figure and table annotations (especially colors, abbreviations, and grouping letters);

These revisions will substantially improve readability and scientific impact without requiring additional analyses.

 

 Specific Comments

Major Comments

Phenology–Environment Framework
The Introduction should further elaborate on the physiological significance of phenology and its interaction with environmental drivers (temperature, radiation, water availability) in perennial fruit crops. A clearer conceptual link between phenological phases and spectral sensitivity would strengthen the scientific framing of the study.

Justification of Methodological Choices

• Please justify the selected sample size in the context of multivariate analyses (PCA, PLSR).

• Briefly explain why Partial Least Squares Regression (PLSR) was chosen over alternative regression or machine-learning approaches.

• Clarify the type of cross-validation used (leave-one-out or k-fold) and specify the number of folds where applicable.

 Statistical Reporting and Decimal Precision
Excessive decimal precision is used throughout the manuscript. Decimal places should generally be limited to two digits for means, standard deviations, correlation coefficients, and model performance metrics (R², Q², RMSE) to improve clarity and consistency.

Figures and Visual Clarity

• clearly define what different colors represent (e.g., environments A1–A5).

• Verify whether color differentiation is necessary in all figures; simplify where possible to enhance interpretability.

• Ensure uniform formatting across all tables in line with MDPI guidelines.

Discussion
The inverse relationship between spectral vigor and yield is a key contribution of this study. The Discussion should further explore this phenomenon in the context of source–sink competition, carbon allocation, and drainage-related constraints in hillside systems, and clarify whether the findings are cultivar-specific or potentially generalizable.

Minor Comments

1. Ensure consistent terminology across the manuscript
2. Define all abbreviations at first mention and again in figure and table captions where appropriate.

For specific comments see attached pdf file.

 

Recommendation:      Moderate Revision

 

The article is well written. However, some changes are needed to refine your findings. I suggest reducing the length of the title, as you are discussing details in methodology. 

You need to improve your hypothesis, introduction and discussion on the interaction of phenology, physiology, and fruit quality, and how your indices are related to these physiological processes. 

Limit decimal places where possible

Define all abbreviations and colors in figure captions. 

Follow uniform citations in the text and recheck the citation numbering throughout the draft.

For further details, see the attached file. I hope these suggestions will help you revise your draft

Comments for author File: Comments.pdf

Author Response

We sincerely thank the reviewer for the careful reading of our manuscript and for the constructive and detailed comments. The suggestions significantly improved the clarity, methodological transparency, and interpretation of the results. Below we address each comment individually and describe the modifications incorporated in the revised manuscript.

 

Specific Comments

Major Comments

Phenology–Environment Framework
The Introduction should further elaborate on the physiological significance of phenology and its interaction with environmental drivers (temperature, radiation, water availability) in perennial fruit crops. A clearer conceptual link between phenological phases and spectral sensitivity would strengthen the scientific framing of the study.

Response: We appreciate the reviewer's insight regarding the physiological framing. We have revised the Introduction to explicitly define phenology as a critical metabolic window where environmental drivers (specifically radiation and water status) modulate the tree's carbon allocation. We now establish a clearer conceptual link between phenological transitions—such as the flowering-to-fruit-set shift—and their manifestation in spectral sensitivity, where indices like NDRE and NDMI capture the physiological cost of reproductive flushes versus vegetative growth.

“In perennial fruit crops, phenological phases function as metabolic windows where environmental drivers—specifically temperature, radiation, and water availability—modulate canopy reflectance, enabling spectral indices to capture the underlying source–sink competition that determines final yield [14, 50, 56]. Consequently, this metabolic complexity challenges the direct interpretation of canopy “greenness” as a proxy for productive performance, necessitating analytical frameworks, such as Partial Least Squares Regression (PLSR), that explicitly integrate the temporal (e.g., 5-year phenological time series) and functional (e.g., source–sink competition and biomass-yield trade-offs) dimensions of the crop [1,2,49].”

 

 

Justification of Methodological Choices

• Please justify the selected sample size in the context of multivariate analyses (PCA, PLSR).

• Briefly explain why Partial Least Squares Regression (PLSR) was chosen over alternative regression or machine-learning approaches.

• Clarify the type of cross-validation used (leave-one-out or k-fold) and specify the number of folds where applicable.

Response: We appreciate the reviewer’s request for methodological clarification. (i) Sample Size: While n = 25 is relatively small for standard machine learning, it is statistically robust for PLSR when each experimental unit (tree) is characterized with high precision through multi-temporal spectral series and subsampling (fruit-level). In precision horticulture, this "quality over quantity" approach allows for capturing micro-topographic variance that would be lost in larger, less-controlled populations. (ii) PLSR vs. Alternatives: PLSR was selected over Multiple Linear Regression (MLR) because it effectively handles the high multicollinearity between spectral indices and topographic variables. Unlike "black-box" machine learning (e.g., Random Forest), PLSR provides VIP scores, which allow for a physiological interpretation of the predictors. (iii) Cross-Validation: We have clarified in Section 2.6.2 that Leave-One-Out Cross-Validation (LOO-CV) was employed, as it is the most stable approach for maximizing the training set in studies with finite sample sizes.

 

“To synthesize and rank the relative contribution of the predictors, Partial Least Squares Regression (PLSR) models were developed. The models were fitted using individual tree-level observations (n = 25) to preserve the high-resolution variance necessary to identify micro-topographic and spectral determinants. PLSR was chosen over alternative machine-learning approaches for its robustness against multicollinearity and its suitability for finite sample sizes [49, 56]. Model stability and predictive capacity were strictly assessed through Leave-One-Out (LOO) cross-validation, thereby ensuring an iterative procedure that maximizes the training set for each experimental unit, ensuring reliable metrics (Q² and RMSE) within this exploratory design [44]”

 

 Statistical Reporting and Decimal Precision
Excessive decimal precision is used throughout the manuscript. Decimal places should generally be limited to two digits for means, standard deviations, correlation coefficients, and model performance metrics (R², Q², RMSE) to improve clarity and consistency.

Response: We sincerely thank the reviewer for this observation. We agree that excessive decimal precision can hinder clarity. Accordingly, we have standardized all statistical reporting throughout the manuscript. Means, standard deviations (SD), correlation coefficients (r), and model performance metrics (R², Q², RMSE) have been limited to two decimal places. Regarding $p$-values, we have maintained three decimal places where necessary to denote significance levels (p < 0.05, 0.01, 0.001). These changes have been applied to Tables 2, 3, 4, and 5, as well as to the Results section.

 

 

Figures and Visual Clarity

• clearly define what different colors represent (e.g., environments A1–A5).

• Verify whether color differentiation is necessary in all figures; simplify where possible to enhance interpretability.

• Ensure uniform formatting across all tables in line with MDPI guidelines.

 

Response: We agree with the reviewer that visual consistency is paramount for interpretability. (i) We have updated all figure captions to clearly define the color coding for the five environments (A1–A5), ensuring that the legend is consistent across the entire manuscript. (ii) We have reviewed the necessity of color in each figure. (iii) Finally, we have standardized all tables (Tables 1–5) according to MDPI's "minimalist" formatting guidelines, ensuring a uniform academic style.

 

Discussion
The inverse relationship between spectral vigor and yield is a key contribution of this study. The Discussion should further explore this phenomenon in the context of source–sink competition, carbon allocation, and drainage-related constraints in hillside systems, and clarify whether the findings are cultivar-specific or potentially generalizable.

Response: We sincerely thank the reviewer for highlighting the importance of the inverse relationship between spectral vigor and yield. We have expanded the Discussion (Section 4) to provide a deeper physiological interpretation of this "vigor paradox." We now explicitly discuss: (i) source–sink competition, framing excessive vegetative flushes as metabolic sinks that penalize fruit set; (ii) carbon allocation, explaining how high NDVI/NDRE values during reproductive stages signal a shift toward biomass at the expense of fruit; and (iii) drainage constraints, where topographic positions with high water retention (concave toe-slopes) promote luxurious vegetative growth but compromise productive stability. Finally, we clarify that while this phenomenon is particularly evident in the vigorous cv. Semil-34, the underlying principles are potentially generalizable to other perennial crops in heterogeneous tropical hillsides.

“Within this framework, excessive vegetative growth during flowering acted as a competitive sink, reducing fruit set and penalizing the yield of cv. Semil-34, a finding consistent with source–sink theory in perennial fruit trees. This indicates that productivity is governed by how hydroclimatic forcing modulates vigor, potentially leading to the source–sink imbalances identified in this mountain system [56].”
´ Despite this spectral mixing, the observed phenological trajectories remain robust, as the perennial biomass constitutes the most stable radiometric component within the analytical footprint, effectively neutralizing ephemeral signals from the managed understory. Topography introduced a transversal gradient by modulating incident radiation and water redistribution, thereby reinforcing spatial vigor patterns even under homogeneous regional climatic conditions [56].”

“This observation underscores a critical physiological trade-off in cv. Semil-34: the tree must balance the carbon demands of simultaneous vegetative and reproductive flushes. When environmental conditions (e.g., nitrogen availability or water retention in specific topographic positions) favor excessive canopy growth, the resulting source–sink imbalance diverts assimilates away from developing fruits. Consequently, the high vegetation index values recorded in these environments do not reflect productive potential but rather a metabolic prioritization of vegetative biomass over reproductive stability [56].”

“Spatial yield variability was structured by topographic gradients that regulated the microclimate and soil water dynamics, thereby conditioning the expression of productive potential [5,58]. Higher yields were associated with intermediate topographic positions, where the balance between drainage and water retention prevented root hypoxia, a critical constraint in sectors with persistent saturation [59]. TWI operated primarily as a risk indicator by identifying zones with higher susceptibility to root restrictions and pathogens such as Phytophthora cinnamomic [60].  These gradients validate the inclusion of geomorphometric variables to delimit productive environments and support differentiated management strategies in heterogeneous agricultural landscapes [6,43,56]. “

 

 

Minor Comments

1. Ensure consistent terminology across the manuscript
2. Define all abbreviations at first mention and again in figure and table captions where appropriate.

Response: We appreciate the reviewer’s attention to detail. (i) Terminology: We have conducted a full-text audit to ensure consistent terminology. Specifically, the term "altitudinal gradient" has been systematically replaced with "elevational gradient" to align with the use of Digital Elevation Models (DEM). (ii) Abbreviations: We have verified that all abbreviations (e.g., NDVI, NDRE, TWI, PLSR) are defined at their first mention in the text. Furthermore, following the reviewer’s suggestion, we have included the full definitions for each abbreviation in all figure and table captions to ensure they are self-explanatory.

 

We sincerely thank the reviewer for the valuable suggestions, which significantly improved the methodological clarity and scientific rigor of the manuscript.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study focuses on the relationships between spectral phenology, climate, topography and yield as well as fruit quality of Semil-34 avocado in tropical mountainous areas. A model was constructed by integrating Sentinel-2 remote sensing, climate reanalysis and topographic data. The research topic has important practical and theoretical value, the methodological system is relatively complete, and the results provide guiding significance for the precision management of avocados in tropical mountainous areas. It is recommended for publication after minor revisions.

 

Specific comments:

1. The abscissa and coordinate axes should be added to Figure 2 for greater rigor, even though the authors’ intended content is currently very easy to understand. In addition, it is suggested to consider adding error bars to Figure 2e if feasible.

 

2. The principal component analysis (PCA) only explained 52.0% of the total variance (PC1 = 36.2%, PC2 = 15.8%), representing a relatively low variance explanation rate. The contributions of the third and higher principal components were not analyzed, nor was the reason for the low explanation rate clarified. It is necessary to supplement the variance contribution spectrum of principal components and provide a corresponding explanation.

 

3. Section 3.7 identifies NDMI and NDRE as the main predictors of yield, yet the discussion fails to compare and explain the reasons for the differences in the predictive ability of different spectral indices (e.g., why NDMI is more important than NDVI). It is required to analyze the intrinsic causes of these differences in combination with the biological implications of the indices.

Author Response

 

We sincerely thank the reviewer for the careful reading of our manuscript and for the constructive and detailed comments. The suggestions significantly improved the clarity, methodological transparency, and interpretation of the results. Below we address each comment individually and describe the modifications incorporated in the revised manuscript.

 

Specific comments:

1. The abscissa and coordinate axes should be added to Figure 2 for greater rigor, even though the authors’ intended content is currently very easy to understand. In addition, it is suggested to consider adding error bars to Figure 2e if feasible.

Response: We agree with the reviewer’s call for greater graphic rigor. We have updated Figure 3 to include explicit labels for the abscissa (Months) and ordinate axes (Index units, °C, and mm) across all subplots. Furthermore, we have incorporated Standard Error (SE) bars into Figure 3e to represent the inter-annual climatic variability (2020–2025), providing a clearer view of the environmental stability of the study site.

 

 

1. The principal component analysis (PCA) only explained 52.0% of the total variance (PC1 = 36.2%, PC2 = 15.8%), representing a relatively low variance explanation rate. The contributions of the third and higher principal components were not analyzed, nor was the reason for the low explanation rate clarified. It is necessary to supplement the variance contribution spectrum of principal components and provide a corresponding explanation.

Response to Comment 2: We appreciate the reviewer's comment regarding the explained variance. A cumulative variance of $52.0\%$ for the first two principal components is characteristic of multi-source field studies that integrate high-dimensional datasets (spectral phenology, climate, and topography).

Our internal analysis of the Scree Plot confirmed that the 'elbow' of the distribution occurs at PC2. While PC3 (11.5%) and PC4 (8.2%) were examined, they were found to represent localized stochastic noise, likely derived from micro-site soil heterogeneity and intra-tree genetic plasticity, rather than structured agro-environmental drivers. Since these higher components do not alter the interpretation of the dominant relationships between topography and spectral vigor, we have focused the biplot on the first two axes to maintain clarity and focus on the primary research objectives. (Line 671-674)

“The analysis accounted for 52.0% of the total variance, primarily concentrated in the first two principal components (PC1 = 36.2%; PC2 = 15.8%). Although higher-order components (PC3 and PC4) were evaluated, they contributed an additional 11.5% and 8.2% to the total variance.”

 

.

 

1. Section 3.7 identifies NDMI and NDRE as the main predictors of yield, yet the discussion fails to compare and explain the reasons for the differences in the predictive ability of different spectral indices (e.g., why NDMI is more important than NDVI). It is required to analyze the intrinsic causes of these differences in combination with the biological implications of the indices.

Response: We agree that a deeper comparison between the indices is necessary to clarify their predictive roles. We have expanded the Discussion to analyze the intrinsic causes of these differences. Specifically, we explain that (i) NDRE outperforms NDVI because the red-edge band penetrates deeper into the multi-layered avocado canopy and avoids the radiometric saturation common in high-biomass systems, thus better capturing the nitrogen status necessary for fruit load. (ii) The superior importance of NDMI is attributed to its sensitivity to leaf turgor and water status (SWIR region), which in tropical hillside systems, where drainage is the primary structural filter, is a more direct driver of fruit expansion and yield stability than mere vegetative vigor. (Line 933-940)

“The superior predictive performance of NDRE and NDMI over NDVI is attributed to their ability to overcome the radiometric saturation typical of dense, multi-layered avocado canopies [3,4]. While NDVI loses sensitivity at high biomass levels, the red-edge-based NDRE provides deeper canopy penetration, effectively capturing the nitrogen and chlorophyll status required to sustain fruit load [53]. Furthermore, NDMI monitors leaf turgor and hydric status via the SWIR region, a factor that, in topographically-governed hillside systems, is more decisive for fruit expansion and productive stability than mere structural vigor [39,40,56].”

 

We sincerely thank the reviewer for the valuable suggestions, which significantly improved the methodological clarity and scientific rigor of the manuscript.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript presents integrate spectral phenology, climatic variables, and geomorphometric predictors to explain variability in avocado yield and fruit quality in a tropical mountain production system. The integration of Sentinel-2 time series, ERA5-Land climatic data, and terrain variables represents a framework for precision agriculture and landscape-scale orchard monitoring. However, several methodological and conceptual issues limit the robustness of the conclusions. Many of these limitations arise from inconsistencies between the hypothesis proposed in the introduction and the statistical framework implemented in the methods.

Introduction

1. The introduction proposes that “Normalized Difference Red Edge (NDRE) has shown superior potential for discriminating functional foliage changes over (NDVI) and the Enhanced Vegetation Index (EVI),” which suggests that this will be the objective of the study, but the hypothesis proposes another topic.
2. Line 93, 123, 131 The term altitude must be changed for elevation, since it was derived from a digital elevation model (line 253).
3. The first and second objectives were not included in the general objective
4. The hypothesis must be developed in introduction section. It is proposed that avocado productivity is governed by an optimal balance of spectrally-captured vigor, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief. In this regard, the introduction suggests that NDRE is better than NDVI and EVI, but this is not stated in the hypothesis. The effect of geomorphometric variables has already been tested, as reported in lines 75-78. Define what production is, the kilograms produced, number of fruits, etc.

Methods

5. The hypothesis is that “avocado productivity is governed by an optimal balance of spectrally-captured vigor, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief,” but there is no index or model that quantifies this balance, no physiological or statistical criterion that represents this equilibrium is defined, nor is an explicit interaction between vigor and topography established.
6. Include the scale bar in Figure 1, as it is necessary to identify the distance between samples.
7. Report the distance between sampling sites and between trees in the same site, given that the hydroclimatic data was obtained from a source with a spatial resolution of 9 km.
8. Specify the average canopy diameter of the trees, since the discussion interprets that the greenness index values in the 10*10m pixel correspond to the tree (lines 594-596), but commercial orchards usually maintain a controlled diameter of 4 to 6 m. Therefore, the sensor will be recording the tree and associated vegetation.
9. Five samples per group leads to low statistical power, in which there is a high risk of committing a type II error, since a very large effect is required to detect significant differences and there is a risk of not identifying real differences between treatments. there is a high risk of the outlier effect; statistical analyses (ANOVA, regression, GLM, etc.) depend on a reliable estimate of residual variance, and with small n, the estimated variance fluctuates greatly, confidence intervals are very wide, and inference becomes unstable. On the other hand, five fruits per tree have five independent biological replicates per treatment, because the five fruits from the same tree are subsamples or technical replicates within the tree.
10. The dependent variables should be presented in the introduction so that the reader knows that the study analyzed the morphological variations of the fruits and the variation in the number of fruits per tree.
11. In the case of section 2.4, it is necessary to report all the variables that were measured for the fruits. A table would be useful to identify the identity and units of measurement of the variables recorded for the fruits and trees.
12. The study measured many dependent and independent variables, but strictly speaking, these were obtained from five samples.
13. Specify the instrument used to measure hourly variables, which were aggregated to a daily scale.
14. The Partial Least Squares should be based on the average per group, since the five samples per group are not independent of each other.
15. In section 2.6.2, it is advisable to specify how many independent variables were analyzed with the PCA, how many topographic variables, how many spectral indices, and how many yield/quality parameters.
16. PCA is a multivariate analysis that orders elements according to their values in a series of variables and identifies which variables are most closely related to the canonical values (principal components). However, the author could use Discriminant Function Analysis to estimate differences between the five groups based on the variables that measured. The analysis will avoid having to perform univariate tests for all the variables measured. The results of the DFA could be more conclusive than those reported in section 3.1, as it would summarize the text in lines 312 to 341, table 2, section 3.3, part of 3.4, and 3.7.
17. The partial least squares analysis is wasted, since the correlation can be presented within and between groups of variables. In this case, the set of environmental variables with the set of fruit variables. The two-dimensional graph of this analysis would show the relationship between both groups of variables, but the univariate form presented by the authors is more difficult for readers to understand, as they would have to interpret Figures 3 and 7 separately, when they could have them in a single graph.
18. In the introduction, they propose that Normalized Difference Red Edge (NDRE) has shown superior potential for discriminating functional foliage changes over (NDVI) and the Enhanced Vegetation Index (EVI), but in their methods they omitted comparisons of these indices.

Results

19. Throughout the section, they refer to correlation as association, but it should be relationship, since the value of r is obtained from regression.
20. The study's hypothesis is “avocado productivity is governed by an optimal balance of spectrally-captured vigor, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief.” However, in methods, it is not reported what they mean by governed, nor how they determined the optimal balance of spectrally-captured vigor, so in the results section it is difficult to determine whether their hypothesis was fulfilled. In addition, the section is very saturated with analyses that could be simplified, which could be achieved with the analysis of discriminant functions and if they present the results of Partial Least Squares.

Discussion

21. The first four paragraphs compare the results between different vegetation indices, which would be correct if the study sought to find the best index to reflect the phenological changes in the tree, but the hypothesis was “avocado productivity is governed by an optimal balance of spectrally-captured vigor, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief.” Therefore, priority should be given to discussing how vigor and topography relate to productivity.
22. The sentence “Excessive vegetative vigor during flowering acted as a competitive sink that reduced fruit set, whereas its moderation favored both fruit caliber and commercial homogeneity (lines 642-643)” could be interpreted as a trade-off by the plant between generating vegetative and reproductive structures, which would be interesting for the authors to explore further.
23. The section interprets the results based on the theoretical basis that precedes them, but it is necessary to discuss the factors that could compromise their results. In this regard, the biggest risk is having a small sample size, as the replicates were taken at a short distance, which increases the risk of spatial autocorrelation, as the samples could have been distributed throughout the study area to include greater variability.
24. It is necessary to discuss the effect of spatial autocorrelation, environmental multicollinearity, and different time scales.
25. Sentinel-2 has a spatial resolution of 10 m, but avocado trees typically have a canopy of 6–10 m. Therefore, a pixel may contain canopy, shade, soil, and grass, which introduces pixel mixing. Although the article mentions that the pixel closest to the tree is used, spectral mixing effects or spatial error are not discussed.
26. Finally, it is recommended that you include a section proposing future studies or changes that should be made to strengthen the results that are reported.

Author Response

Response to Reviewer 3

 

We sincerely thank the reviewer for the careful reading of our manuscript and for the constructive and detailed comments. The suggestions significantly improved the clarity, methodological transparency, and interpretation of the results. Below we address each comment individually and describe the modifications incorporated in the revised manuscript.

 

Introduction

Comment 1: The introduction proposes that “Normalized Difference Red Edge (NDRE) has shown superior potential for discriminating functional foliage changes over NDVI and EVI,” which suggests that this will be the objective of the study, but the hypothesis proposes another topic.

Response 1: We appreciate the reviewer’s observation regarding the conceptual focus. The introduction has been revised to clarify that the study does not aim to establish a ranking of indices, but rather to utilize a multidimensional spectral suite. By presenting NDVI, EVI, NDRE, and NDMI as synergistic indicators, capturing structural vigor, canopy architecture, chlorophyll dynamics, and water status, respectively, we ensure full alignment with the hypothesis that productivity is governed by a complex functional balance rather than a single spectral proxy.

New text has been rewritten in line: (85-89) “Within this framework, red-edge-based indices, such as the Normalized Difference Red Edge (NDRE), provide complementary sensitivity for discriminating functional foliage changes associated with nutritional status and photosynthetic efficiency, acting in synergy with structural metrics to capture the physiological complexity of cv. Semil-34 [22,23]”.

 

Comment 2: Line 93, 123, 131 The term altitude must be changed for elevation, since it was derived from a digital elevation model.

Response 2: We thank the reviewer for pointing out this terminology issue. The term altitude has been replaced with “elevation” throughout the manuscript when referring to values derived from the Digital Elevation Model (DEM). Changes implemented in lines (85)

 

Comment 3: The first and second objectives were not included in the general objective.

Response 3: We agree with the reviewer’s assessment regarding the logical hierarchy of the objectives. The general objective has been reformulated to act as a conceptual umbrella that explicitly incorporates the components of spectral phenology and climatic sensitivity. By defining the main goal as the characterization of the interactions between phenological trajectories, hydroclimatic forcing, and geomorphometry, the specific objectives (i, ii, and iii) are now logically subsumed and aligned with the overall aim of the study. Line (107-108)

“Against this backdrop, the primary objective of this study is to analyze the interaction between spectral phenology, hydroclimatic forcing, and geomorphometry to elucidate the determinants of yield and quality variability in avocado (cv. Semil-34) across a mountain elevational gradient in the Dominican Republic.”

 

Comment 4: The hypothesis must be developed in introduction section. It is proposed that avocado productivity is governed by an optimal balance of spectrally-captured vigor, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief. In this regard, the introduction suggests that NDRE is better than NDVI and EVI, but this is not stated in the hypothesis. The effect of geomorphometric variables has already been tested, as reported in lines 75-78. Define what production is, the kilograms produced, number of fruits, etc.

Response 4: We agree with the reviewer. The hypothesis has been explicitly reformulated in the final paragraph of the Introduction to ensure logical flow. We have removed any implication of "superiority" of NDRE; instead, it is now presented as a functional component that, alongside NDVI, EVI, and NDMI, allows for a multi-faceted characterization of canopy vigor. Furthermore, "productivity" is now clearly defined through its components: fruit count, total weight, and average fruit weight. These operational definitions have been integrated into Section 2.4.1. Text in line (107-120)

“Against this backdrop, the primary objective of this study is to analyze the interaction between spectral phenology, hydroclimatic forcing, and geomorphometry to elucidate the determinants of yield and quality variability in avocado (cv. Semil-34) across a mountain elevational gradient in the Dominican Republic. This integrative framework specifically aims to: (i) characterize multitemporal phenological dynamics using multispectral indices; (ii) analyze the sensitivity of these indices to climatic forcing as a function of the phenological phase; and (iii) quantify the relative contribution of topography and spectral response to yield components -specifically fruit count and cumulative weight per tree- and fruit quality, encompassing morphological variations (diameters and mass) and dry matter content. We hypothesize that avocado productivity (defined by fruit count and total yield per tree) is governed by an optimal balance of spectrally-captured vigor. This balance is characterized through the integration of structural, functional, and hydric spectral indicators, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief.”

 

 

Methods

Comment 5: The hypothesis is that “avocado productivity is governed by an optimal balance of spectrally-captured vigor, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief,” but there is no index or model that quantifies this balance, no physiological or statistical criterion that represents this equilibrium is defined, nor is an explicit interaction between vigor and topography established.

Response 5: We appreciate the reviewer's precision. We have clarified that the "optimal balance" is not expressed as a single algebraic index, but is quantified through the Partial Least Squares Regression (PLSR) framework. This model explicitly integrates spectral (vigor) and geomorphometric (filter) variables, where the Variable Importance in Projection (VIP) scores serve as the statistical criterion to weight their relative contribution to productivity. Following established protocols, variables with VIP > 1.0 were identified as primary determinants, while those between 0.8 and 1.0 were considered to have moderate influence [49]. Furthermore, the 'vigor paradox’  identified in the results provides the physiological criterion for this equilibrium, defining the threshold where excessive canopy growth compromises yield. We have updated Section 2.6.2 to clarify this modeling approach. Text in line (434-438, 448-450)

 

“The multivariate architecture of the system, comprising 17 variables including 5 topography covariates (elevation, slope, aspect, TWI, and TPI), 4 spectral indices (NDVI, NDRE, EVI, and NDMI), and 8 yield and fruit quality parameters (fruit count, total weight, average weight, diameters, pulp and seed mass, and dry matter content), was examined using Principal Component Analysis (PCA) on z-score standardized variables.

 

“This unsupervised ordination approach was selected over classification-based methods (e.g., DFA) to prioritize the exploration of continuous bio-physical gradients and to identify non-linear trade-offs between canopy vigor and yield, which are the core focus of this investigation. This approach facilitated the interpretation of covariation patterns and environmental differentiation through the use of biplots and cluster análisis [46].”

 

 

Comment 6: Include the scale bar in Figure 1, as it is necessary to identify the distance between samples.

Response 6: A scale bar has been added to Figure 1 to facilitate spatial interpretation of the sampling locations and study area. Text in line ( )

 

Comment 7: Report the distance between sampling sites and between trees in the same site, given that the hydroclimatic data was obtained from a source with a spatial resolution of 9 km.

Response 7: We have updated the manuscript to specify the spatial scales of our sampling design. To ensure independence and minimize spatial autocorrelation, the selected trees within each environment were separated by a distance greater than 12 m, while the five sampling environments (A1–A5) were distributed with inter-strata distances exceeding 100 m. While these scales fall within a single ERA5-Land pixel (9 km), we clarify that the climatic data was used exclusively to identify temporal phenological trajectories (Objectives i and ii). The spatial variability in yield and quality was addressed using high-resolution topographic (5 m) and spectral (10 m) covariates, which are the appropriate scales for orchard-level heterogeneity. Text in line (181-185)

 

“To ensure spatial representativeness and minimize local dependencies, sampling trees within each environment were separated by at least 12 m, preserving the high-resolution variance necessary to identify micro-topographic and spectral determinants, while the five environments (A1–A5) were distributed across the landscape with inter-strata distances exceeding 100 m.”

 

Comment 8: Specify the average canopy diameter of the trees, since the discussion interprets that the greenness index values in the 10*10m pixel correspond to the tree (lines 594-596), but commercial orchards usually maintain a controlled diameter of 4 to 6 m. Therefore, the sensor will be recording the tree and associated vegetation.

Response 8: We appreciate the reviewer’s technical observation regarding the spatial scales of our study. The manuscript has been updated to report an average canopy diameter of 4–6 m within a 6 × 6 m planting grid. We explicitly acknowledge that the 10 m Sentinel-2 pixel introduces sub-pixel heterogeneity (spectral mixing), incorporating signals from the target tree, adjacent canopies, and the managed understory. However, we argue that the 10 m signal effectively represents a spatially discrete functional unit that captures the integrated physiological status of the productive microsite. To ensure statistical independence and avoid pixel overlap, sampled trees were separated by at least 12 m, ensuring that each analytical footprint was geographically unique and dominated by the target individual’s functional response. This operational approach is common in satellite-based monitoring where the perennial canopy remains the most stable and significant biomass component within the pixel. Text in line (323-328 )

 

“Spectral analysis was conducted at the individual tree level. Precise GPS coordinates for the 25 experimental units were acquired using a DJI RTK3 antenna and subsequently reprojected to the EPSG:32619 coordinate system. For each tree, the value from the nearest 10-m pixel was extracted. Given the average canopy diameter of 4–6 m, this 10-m analytical footprint captures a spatially discrete functional unit centered on the target tree. While this resolution introduces sub-pixel heterogeneity, integrating soil and undergrowth signals, the sampling design ensures that each pixel represents a unique response dominated by the perennial canopy, as supported by the minimum 12-m inter-tree spacing established in the experimental design. Bottom-of-Atmosphere (BOA) reflectance values were rescaled by a factor of 0.0001, adhering to Copernicus data standards.”

 

 

Comment 9: Five samples per group leads to low statistical power, in which there is a high risk of committing a type II error, since a very large effect is required to detect significant differences and there is a risk of not identifying real differences between treatments. there is a high risk of the outlier effect; statistical analyses (ANOVA, regression, GLM, etc.) depend on a reliable estimate of residual variance, and with small n, the estimated variance fluctuates greatly, confidence intervals are very wide, and inference becomes unstable. On the other hand, five fruits per tree have five independent biological replicates per treatment, because the five fruits from the same tree are subsamples or technical replicates within the tree.

Response 9: We acknowledge the reviewer’s concern regarding statistical power and the risk of Type II errors. In response, we have incorporated a dedicated paragraph in the Discussion (Section 4) addressing these limitations. However, we emphasize that the large effect sizes observed—evidenced by highly significant differences (p < 0.001) across environments—suggest that the environmental drivers in these tropical hillsides are sufficiently robust to be captured by our design. Furthermore, we explicitly clarify in Section 2.2 and 2.4 that the tree (n=25) is the experimental unit, while the fruits are treated as subsamples to estimate the mean response, thus avoiding pseudoreplication. This exploratory approach is justified by the logistical complexity of high-altitude tropical research and the need for high-frequency field measurements. Text in line (177-181 )

 

“This controlled selection was implemented to isolate the influence of the environmental gradient by minimizing biological variance associated with tree structural size. These trees constitute the independent experimental units, whereas individual fruits are treated as subsamples to estimate the mean response of each unit and mitigate intra-tree variability.”

 

Comment 10: The dependent variables should be presented in the introduction so that the reader knows that the study analyzed the morphological variations of the fruits and the variation in the number of fruits per tree.

Response 10: We agree with the reviewer’s suggestion for improved clarity. The final paragraph of the Introduction has been revised to explicitly present the dependent variables. The text now specifies that productivity is evaluated through yield components (fruit count and total weight per tree), while fruit quality is assessed through morphometric variations (diameters and weight) and internal composition (dry matter content). This ensures the reader has a clear overview of the parameters analyzed before entering the Methods section. Text in line (110-119 )

 

“This integrative framework specifically aims to: (i) characterize multitemporal phenological dynamics using multispectral indices; (ii) analyze the sensitivity of these indices to climatic forcing as a function of the phenological phase; and (iii) quantify the relative contribution of topography and spectral response to yield components -specifically fruit count and cumulative weight per tree- and fruit quality, encompassing morphological variations (diameters and mass) and dry matter content.”

 

 

Comment 11: In the case of section 2.4, it is necessary to report all the variables that were measured for the fruits. A table would be useful to identify the identity and units of measurement of the variables recorded for the fruits and trees.

Response 11: A new table has been added in the Methods section summarizing all dependent variables, their definitions, and measurement units. Text in line (123 )

 

“The complete set of evaluated variables, including their operational definitions, units, and scales of measurement, is summarized in Table 1.

 

Table 1. Summary of the dependent yield and quality variables evaluated in avocado (cv. Semil-34).

Category	Variable	Operational Definition	Units	Measurement Scale
Productivity	Yield Fruit Count	Total number of fruits harvested per tree	fruits tree⁻¹	Tree
	Yield Total Weight	Cumulative mass of export-grade fruit	kg tree⁻¹	Tree
	Yield Avg. Weight	Mean calculated mass per fruit	g fruit⁻¹	Tree (Mean)
Morphometry	Long. Diameter	Distance along the apical-peduncular axis	mm	Fruit (Subsample)
	Equat. Diameter	Maximum diameter in the equatorial plane	mm	Fruit (Subsample)
Biomass	Pulp Mass	Mass of the edible portion (mesocarp)	g	Fruit (Subsample)
	Seed Mass	Mass of the seed (endocarp/testa)	g	Fruit (Subsample)
Quality	Dry Matter (DM)	Percentage of dry matter (NIR/Gravimetric)	%	Fruit (Subsample)
	Firmness	Pulp penetration resistance	N	Fruit (Subsample)

 

 

 

Comment 12: The study measured many dependent and independent variables, but strictly speaking, these were obtained from five samples.

Response 12: We acknowledge the reviewer’s point regarding the sample size per environment. To address this, we have explicitly clarified in Sections 2.2 and 2.4 that the tree is the true experimental unit (n=25), and that the high number of variables measured is supported by a robust sampling protocol where individual fruits act as subsamples to stabilize the tree-level mean. This design aims to prioritize the depth of the characterization per unit over the total number of units, which is a standard approach in precision fruit-growing studies in complex terrains. Text in line (223-225 )

“By utilizing the tree as the primary analytical unit (n=25), we ensured that the multiple dependent variables recorded, despite the high number of parameters, reflect stabilized averages rather than isolated observations.”

 

Comment 13: Specify the instrument used to measure hourly variables, which were aggregated to a daily scale.

Response 13: We have clarified in the manuscript that the hourly variables were not obtained from a single local instrument, but from the ERA5-Land reanalysis dataset. This dataset is produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) through the integration of global meteorological observations with advanced numerical weather prediction models via data assimilation. We justified its use given the low density of reliable in-situ stations in the study area, as ERA5-Land provides a consistent and validated multi-decadal record for tropical agroclimatic research. Text in line ( 283-289)

“A continuous hydroclimatic data extraction workflow for the 2020–2025 period was implemented in Google Earth Engine (GEE(https://earthengine.google.com/), utilizing ERA5-Land dataset,  produced by the European Centre for Medium-Range Weather Forecasts (ECMWF), providing a synthesized estimate of land-surface variables at a (~9 km resolution). Unlike direct local instrumentation, this dataset provides a synthesized estimate by combining global observation data with numerical weather models through 4D-Var data assimilation. Climate variables were extracted for the orchard’s central coordinates (18.472630° N, −70.199002° W).”

 

Comment 14: The Partial Least Squares should be based on the average per group, since the five samples per group are not independent of each other.

Response 14: We respectfully disagree with the suggestion to average the data per group. In the context of precision agriculture and high-resolution ecophysiological modeling, the individual tree (n=25) is the fundamental experimental unit. Each tree integrates a unique combination of micro-topography, soil-water dynamics, and canopy architecture that would be masked by environmental averaging.

To ensure the validity of this approach:

1. Independence: As detailed in Section 2.2, trees were separated by a minimum of 12 m to prevent spectral overlap and root competition.
2. Control of Confounding Variables: The basal morphological homogeneity achieved through the dendrometric census was a deliberate strategy to minimize "biological noise" (differences in age or structural development). This ensures that the observed variance is driven by the environmental gradient rather than intrinsic tree differences.
3. Statistical Robustness: Partial Least Squares Regression (PLSR) was specifically selected because it is mathematically designed to handle the multicollinearity inherent in these predictors without requiring the aggregation of data. Averaging by group would reduce the statistical power to a degree that would invalidate the multivariate modeling of these complex interactions. Text in line ( 460-463)

“To synthesize and rank the relative contribution of the predictors, Partial Least Squares Regression (PLSR) models were developed. The models were fitted using individual tree-level observations (n = 25) to preserve the high-resolution variance necessary to identify micro-topographic and spectral determinants. This method is particularly suitable for addressing highly correlated variables and finite sample sizes [49].”

 

Comment 15: In section 2.6.2, it is advisable to specify how many independent variables were analyzed with the PCA, how many topographic variables, how many spectral indices, and how many yield/quality parameters.

Response 15: We agree that specifying the dimensionality of the multivariate analysis improves the clarity of the study. Section 2.6.2 has been revised to detail the 17 variables integrated into the PCA: 5 topographic covariates (elevation, slope, aspect, TWI, and TPI), 4 spectral vegetation indices (NDVI, EVI, NDRE, and NDMI), and 8 yield and fruit quality parameters (fruit count, total weight, average weight, longitudinal and equatorial diameters, pulp and seed mass, and dry matter content). This explicit breakdown ensures full alignment between the methodological description and the results presented in the PCA biplots. Text in line (434-437 )

“The multivariate architecture of the system, comprising 17 variables including 5 topography covariates (elevation, slope, aspect, TWI, and TPI), 4 spectral indices (NDVI, NDRE, EVI, and NDMI), and 8 yield and fruit quality parameters (fruit count, total weight, average weight, diameters, pulp and seed mass, and dry matter content), was examined using Principal Component Analysis (PCA) on z-score standardized variables.”

 

Comment 16: PCA is a multivariate analysis that orders elements according to their values in a series of variables and identifies which variables are most closely related to the canonical values (principal components). However, the author could use Discriminant Function Analysis to estimate differences between the five groups based on the variables that measured. The analysis will avoid having to perform univariate tests for all the variables measured. The results of the DFA could be more conclusive than those reported in section 3.1, as it would summarize the text in lines 312 to 341, table 2, section 3.3, part of 3.4, and 3.7.

Response 16: We thank the reviewer for this insightful statistical suggestion. While Discriminant Function Analysis (DFA) is an excellent tool for group separation, we opted to retain the PCA and PLSR framework for three fundamental reasons: (i) our primary objective was to elucidate functional gradients and covariation patterns (such as the "vigor paradox"), which are more effectively visualized through the eigenvector-based biplots of PCA; (ii) the elevational environments represent a continuous environmental gradient rather than discrete, independent categories, making an unsupervised ordination more ecologically sound; and (iii) the use of PLSR with VIP scores already provides the "supervised" analytical power the reviewer seeks, identifying which variables best explain the system's variance. We have added a clarifying sentence in Section 2.6.2 to justify this choice and have tightened the text in the Results section to improve conciseness as suggested. Text in line (445-448 )

 

“This unsupervised ordination approach was selected over classification-based methods (e.g., DFA) to prioritize the exploration of continuous bio-physical gradients and to identify non-linear trade-offs between canopy vigor and yield, which are the core focus of this investigation. This approach facilitated the interpretation of covariation patterns and environmental differentiation through the use of biplots and cluster análisis [46].”

 

Comment 17: The partial least squares analysis is wasted, since the correlation can be presented within and between groups of variables. In this case, the set of environmental variables with the set of fruit variables. The two-dimensional graph of this analysis would show the relationship between both groups of variables, but the univariate form presented by the authors is more difficult for readers to understand, as they would have to interpret Figures 3 and 7 separately, when they could have them in a single graph.

Response 17: We appreciate the reviewer’s perspective on the integration of variables. We have revised the Results section (3.6 and 3.7) to explicitly state that the PCA and PLSR act as the unifying framework for the climatic and bivariate data presented earlier. By adding clear cross-references, we ensure the reader perceives the transition from univariate details (Figures 3 and 7) to the integrated two-dimensional synthesis (Figure 8) and the final quantitative ranking (Figure 9). This structure preserves the necessary detail while providing the "all-in-one" perspective requested. Text in line (755-757 and 808-813

 

“To integrate the temporal climatic sensitivities (Section 3.2) and the bivariate associations (Section 3.5) into a single analytical framework, a multivariate synthesis was performed. The PCA integrated topographic features, spectral indices, and yield/quality parameters, revealing a systematic differentiation of environments within the factorial space, characterized by distinct clusters and partial overlaps between contiguous sites (Figure 8).”

“PLSR models constructed using the four predictors with the highest importance (VIP ≥ 1) demonstrated moderate explanatory power for quantitative yield variables (Table 5). This modeling approach builds upon the previously described multivariate ordination by quantifying the relative weight (VIP) of each predictor within the system, identifying which specific interactions between vigor and topography are most decisive for avocado yield.”

 

Comment 18: In the introduction, they propose that Normalized Difference Red Edge (NDRE) has shown superior potential for discriminating functional foliage changes over (NDVI) and the Enhanced Vegetation Index (EVI), but in their methods they omitted comparisons of these indices.

Response 18: We appreciate the reviewer's observation. We have clarified the Introduction and Methods to reflect that the objective of this study was not to perform a comparative performance ranking (or "superiority" test) of the indices, but rather to exploit their functional complementarity. While NDVI is a robust proxy for structural biomass, NDRE provides specific sensitivity to chlorophyll fluctuations and nitrogen status in high-biomass canopies where NDVI typically saturates. The revised manuscript now explicitly describes this synergistic approach, where each index contributes a unique dimension (structural, nutritional, or hydric) to the multivariate PLSR model, rather than competing for statistical superiority. Text in line (85-89 )

 

“Within this framework, red-edge-based indices, such as the Normalized Difference Red Edge (NDRE), provide complementary sensitivity for discriminating functional foliage changes associated with nutritional status and photosynthetic efficiency, acting in synergy with structural metrics to capture the physiological complexity of cv. Semil-34 [4]”.

 

Results

Comment 19: Throughout the section, they refer to correlation as association, but it should be relationship, since the value of r is obtained from regression.

Response 19: We appreciate the reviewer's guidance on terminology. Following your suggestion, we have systematically replaced the terms "association" and "correlation" with "relationship" throughout the Results section when referring to Pearson’s r values. This adjustment ensures terminological consistency with the regression-based framework utilized in the study. Text in line ( 570)

 

“Pearson correlation analysis revealed statistically significant relationships (p < 0.001) between spectral indices and climatic variables across all phenological phases (Figure 3), with sample sizes ranging from 2,750 to 5,750 observations per phase.”

 

“Bivariate relationships between Environmental Predictors and Productive Performance

Bivariate correlations revealed predominantly negative, moderate-magnitude, and statistically significant relationships between spectral indices and productive variables (Figure 7).

.Text in line ( 723-739)

“Total fruit number per tree (TFN) also exhibited moderately and negatively relationships with NDMI (r = −0.514) and weakly with NDRE (r = −0.424). Regarding morphological parameters, longitudinal fruit diameter (LFD) showed weak negative relationships with NDMI (r = −0.481), NDVI (r = −0.435), and NDRE (r = −0.419), while equatorial fruit diameter (EFD) presented negative relationships with NDVI (r = −0.497) and NDMI (r = −0.418). Dry matter (DM) displayed low spectral dependence, showing only a weak negative relationships with topographic aspect (r = −0.442). The highest magnitude relationships corresponded to NDVI–AFW (r = −0.609) and NDMI–TY (r = −0.585), both demonstrating moderate significance.”

 

 

Comment 20: The study's hypothesis is “avocado productivity is governed by an optimal balance of spectrally-captured vigor, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief.” However, in methods, it is not reported what they mean by governed, nor how they determined the optimal balance of spectrally-captured vigor, so in the results section it is difficult to determine whether their hypothesis was fulfilled. In addition, the section is very saturated with analyses that could be simplified, which could be achieved with the analysis of discriminant functions and if they present the results of Partial Least Squares.

Response 20: We appreciate the reviewer’s call for conceptual precision. We have revised the Methods (Section 2.6.2) and Discussion (Section 5) to explicitly define the operational terms of our hypothesis:

1. "Governed" refers to the hierarchical influence of environmental predictors on yield, quantified through the VIP scores in the PLSR model; a variable "governs" the system when its VIP > 1.0.
2. "Optimal Balance" is operationally defined as the range of spectral vigor (NDVI/NDRE) that maximizes yield before the source-sink competition (the "vigor paradox") triggers a decline in productivity.
3. "Landscape Filter" refers to the role of topographic variables (TWI and Slope) in modulating water availability, as shown by their high importance in the PLSR synthesis.

We have also streamlined the Results section by prioritizing the PLSR and PCA as the primary integrative tools, as suggested, ensuring the text directly addresses the fulfillment of the hypothesis. Text in line (468-479)

“This multivariate framework was employed to quantify the 'optimal balance' proposed in our hypothesis, effectively integrating spectral indices and geomorphometric covariates into a unified explanatory model. In this context, the term 'governed' refers to the statistical dominance of predictors with Variable Importance in Projection (VIP) scores > 1.0, which provides a statistical weight for the interaction between canopy status and topographic constraints [8,60]. Following established chemometric protocols, variables with VIP scores between 0.8 and 1.0 were considered to have moderate influence, while those with VIP < 0.8 were categorized as low-contribution predictors [16,55]. Furthermore, the 'optimal balance' is operationally identified as the range of spectral vigor where the relationship with yield remains stable before the onset of source–sink competition. By integrating these distinct blocks of variables, the PLSR serves as the final synthetic step to bridge the gap between regional climatic forcing and site-specific productive responses [8,60]. ”

 

 

Discussion

Comment 21: The first four paragraphs compare the results between different vegetation indices, which would be correct if the study sought to find the best index to reflect the phenological changes in the tree, but the hypothesis was “avocado productivity is governed by an optimal balance of spectrally-captured vigor, which is structurally filtered by the drainage capacity and water retention patterns imposed by the landscape relief.” Therefore, priority should be given to discussing how vigor and topography relate to productivity.

Response 21: We appreciate this critical structural observation. The Discussion section has been thoroughly reorganized to shift the focus from a comparison of indices to a functional interpretation of the vigor–topography–yield interaction. The initial paragraphs now lead with the ecophysiological implications of the "vigor paradox" and the role of topography as a structural filter of productive stability. Vegetation indices are now discussed as integrated functional indicators rather than individual metrics, ensuring the discussion directly supports the evaluation of our central hypothesis regarding the optimal balance of spectrally-captured vigor. Text in line ( 839-842)

“Collectively, the observed spectral phenology represents an integrated expression of regional climatic control modulated by local soil-plant conditions, establishing the functional foundation for the spectrally-captured vigor proposed in our hypothesis [10,61]. “

)

 

Comment 22: The sentence “Excessive vegetative vigor during flowering acted as a competitive sink that reduced fruit set, whereas its moderation favored both fruit caliber and commercial homogeneity (lines 642-643)” could be interpreted as a trade-off by the plant between generating vegetative and reproductive structures, which would be interesting for the authors to explore further.

Response 22: We thank the reviewer for highlighting this fundamental physiological mechanism. We have expanded the discussion (Section 4, paragraph 4) to explicitly explore the vegetative–reproductive trade-off. We now frame the "vigor paradox" as a competition for limited photoassimilates, where excessive vegetative flushes during flowering and fruit set act as a dominant metabolic sink. This competition prioritizes canopy expansion over reproductive success, particularly in high-vigor environments. By incorporating this ecophysiological lens, we clarify how spectral indices (NDVI and NDRE) function as proxies for this internal resource partitioning, where high values during critical reproductive windows indicate a shift toward vegetative dominance that penalizes final yield. Text in line ( 886-895)

 

“Excessive vegetative vigor during flowering acted as a competitive sink that reduced fruit set, whereas its moderation favored both fruit caliber and commercial homogeneity [2,34]. This observation underscores a critical physiological trade-off in cv. Semil-34: the tree must balance the carbon demands of simultaneous vegetative and reproductive flushes. When environmental conditions (e.g., nitrogen availability or water retention in specific topographic positions) favor excessive canopy growth, the resulting source–sink imbalance diverts assimilates away from developing fruits. Consequently, the high vegetation index values recorded in these environments do not reflect productive potential but rather a metabolic prioritization of vegetative biomass over reproductive stability.”

 

 

Comment 23: The section interprets the results based on the theoretical basis that precedes them, but it is necessary to discuss the factors that could compromise their results. In this regard, the biggest risk is having a small sample size, as the replicates were taken at a short distance, which increases the risk of spatial autocorrelation, as the samples could have been distributed throughout the study area to include greater variability.

Response 23: We appreciate this constructive critique regarding the spatial design. While the 12-m inter-tree spacing was selected to ensure unique Sentinel-2 pixel footprints (10 m), we acknowledge that a small sample size in a localized area may increase the risk of spatial autocorrelation. We have revised the Discussion (Section 4, final paragraph) to explicitly address this limitation. We argue that while the detected environmental effects are highly significant ($p < 0.001$), they should be interpreted as a robust exploratory characterization of the Semil-34 hillside system. We now suggest that future research should implement larger-scale sampling schemes and spatially explicit modeling to further decouple local spatial dependence from broad-scale environmental drivers. Text in line (958-960)

 

“Furthermore, the 12-m proximity between sampling trees, although necessary to align the experimental units with the 10-m spectral footprint of Sentinel-2, may introduce a risk of spatial autocorrelation.”

 

 

Comment 24: It is necessary to discuss the effect of spatial autocorrelation, environmental multicollinearity, and different time scales.

Response 24: We appreciate this comprehensive technical suggestion. The Discussion has been expanded to address these three critical factors. We now clarify that: (i) Environmental multicollinearity was a primary driver for selecting PLSR, which effectively handles correlated predictors through latent variable projection; (ii) Spatial autocorrelation is acknowledged as a localized risk due to tree proximity, framing the study as a robust exploratory analysis; and (iii) Time scales were addressed by integrating multi-year spectral phenology (2020–2025) with high-frequency climatic data, ensuring that the detected sensitivities are not merely seasonal artifacts but consistent ecophysiological patterns. Text in line (951-960 )

 

“Finally, the limitations associated with the sample size (n=5 per environment) must be acknowledged. While a small n may increase the risk of Type II errors and sensitivity to outliers, the detection of highly significant differences (p < 0.001) in yield and spectral variables suggests that the environmental gradients in these tropical hillsides exert a dominant influence that overcomes power constraints. The inherent environmental multicollinearity was a fundamental justification for employing Partial Least Squares Regression (PLSR), which effectively handled correlated predictors through latent internal quality [59]. Furthermore, the 12-m proximity between sampling trees, although necessary to align the experimental units with the 10-m spectral footprint of Sentinel-2, may introduce a risk of spatial autocorrelation.”

 

Comment 25: Sentinel-2 has a spatial resolution of 10 m, but avocado trees typically have a canopy of 6–10 m. Therefore, a pixel may contain canopy, shade, soil, and grass, which introduces pixel mixing. Although the article mentions that the pixel closest to the tree is used, spectral mixing effects or spatial error are not discussed.

Response 25: We appreciate the reviewer's technical observation regarding spectral mixing. We have revised the Discussion (Section 4) to explicitly address this limitation. We clarify that while the 10-m resolution of Sentinel-2 introduces sub-pixel heterogeneity (integrating signals from the canopy, shadows, and understory), the 12-m inter-tree spacing was specifically designed to ensure that each pixel remains dominated by a unique experimental unit. We argue that for regional phenological monitoring, this "analytical footprint" serves as a robust operational proxy, as the perennial canopy constitutes the most stable radiometric component. However, we acknowledge that spatial uncertainty and spectral mixing are inherent to orbital sensors and suggest that future studies should incorporate UAV-based object-oriented analysis to further decouple these signals. Text in line (960-970 )

“Regarding spatial resolution, the 10-m Sentinel-2 pixel inevitably introduces spectral mixing effects, integrating the canopy signal with soil background, shadows, and understory vegetation. Although the 12-m inter-tree spacing ensured that each pixel was dominated by a unique experimental unit, this sub-pixel heterogeneity remains a source of spatial uncertainty. This implies that the observed patterns should be interpreted as a robust exploratory framework rather than a final generalization. By integrating different time scales, we minimized seasonal bias and confirmed that the 'vigor paradox' is a stable physiological trait of cv. Semil-34. This study should be interpreted as a robust exploratory analysis of cv. Semil-34 in under-documented mountain systems, providing a foundation for future larger-scale spatial assessments utilizing higher-resolution sensors or object-oriented analysis to further mitigate pixel mixing.”

 

 

Comment 26: Finally, it is recommended that you include a section proposing future studies or changes that should be made to strengthen the results that are reported.

Response 26: We agree with the reviewer’s suggestion to provide a forward-looking perspective. The Conclusions section has been expanded to explicitly propose three strategic research directions that build upon our current findings. These include (i) expanding spatial sampling to further decouple landscape-scale dependencies, (ii) integrating sub-decametric sensors (UAVs) to refine spectral purity, and (iii) establishing precise physiological thresholds for the "optimal vigor balance." This addition ensures the study serves as a robust foundation for the next generation of predictive models in tropical mountain agroecosystems. Text in line (998-1003 )

“Building upon these findings, future research should focus on: (i) implementing larger-scale sampling designs to further mitigate spatial dependencies and enhance the generalization of the 'vigor paradox' (ii) integrating sub-decametric UAV imagery to resolve the spectral mixing effects inherent in orbital sensors; and (iii) defining precise functional thresholds for the optimal source–sink balance through in-situ physiological monitoring. These advancements will facilitate the transition from this exploratory framework to multiscale predictive models that fully integrate topography as a structural determinant.”

 

 

We sincerely thank the reviewer for the valuable suggestions, which significantly improved the methodological clarity and scientific rigor of the manuscript.

 

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The article entitled “Spectral phenology, climate, and topography as determinants of vigor, yield, and fruit quality in avocado (Persea americana cv. Semil-34) on tropical hillsides” presents a relevant theme and is in accordance with the scope of the Horticultura journal. However, for the manuscript to achieve the necessary rigor for publication, the following revisions are requested: In the abstract, some questions arose and need to be detailed (see specific comments). In the Introduction, the author should expand the regional context, including a paragraph addressing the economic and social importance of avocado for the Dominican Republic, specifying the volume of the annual harvest in this location. It is necessary to deepen the discussion on the water and climate needs of the crop, highlighting the risks that avocado production faces in the face of climate change and clarifying whether the best plant development occurs at low or high altitudes. Also in this section, the use of the ERA5-Land dataset should be discussed, pointing out its advantages, disadvantages, and justifying its specific application in this study. In the Materials and Methods section, it is suggested that visual resources be enhanced by including a map of the local altitude and a flowchart that clearly demonstrates the data processing steps. Regarding remote sensing, it is essential to formally introduce the spectral, temporal, and radiometric resolutions of the Sentinel-2 sensor, providing a technical interpretation of the adopted spectral indices. Additionally, statistical rigor should be increased through the description and presentation of the formulas for the metrics used, specifically: Coefficient of Determination (R²), Predictive Ability (Q²), Root Mean Square Error (RMSE), and the Variance Inflation Factor (VIF), the latter being essential for multicollinearity analysis. Finally, a thorough review of the references should be carried out, as several citations throughout the text do not follow the bibliographic standard required by this journal.

Specific Comments
12 ... Monitoring avocado (Persea americana Mill., cv. Semil-34) in tropical mountain (? Specify location) landscapes is inherently complex due to the pronounced topographical
14. ...modulates the crop’s ecophysiological responses (?). What are these responses? List them.

This study integrates multi-year (? Specify the time series period) Sentinel-2 time series, ERA5-Land climatic variables (Which variables?), and geomorphometric covariates to explain variability in yield and fruit quality. What errors are associated with the ERA5-Land variables? Did the authors seek to validate ERA5-Land before using it in the research?

23 ...were associated with lower production (r = −0.58, p < 0.0021 ?), reflecting a critical source... Although p is significant, R² does not explain the relationship. How do you justify this?
25... While yield variability was partially explainable (R² = 0.38?), internal fruit quality,... In fact, the topographic variable only managed to explain 38% of productivity. Is this a good result for your research, although not one of the best?
40-41... where topographic and climatic heterogeneity imposes structural constraints (Which ones? Cite them) on both agronomic management (What type? Cite them) and yield stability (What does this mean?).
48... particularly during the flowering stage (? How many days does the flowering phase last?),...
51-52... necessitating analytical frameworks (What are these structures? Give an example!) that explicitly integrate the temporal (How long? Give an example.) and functional (What type? Cite them.) dimensions of the crop [1,2]
56-57... spatial resolution ( ? Cite them.), temporal frequency (? Cite it), and specific spectral bands, most notably the red edge ( ? Cite the interval!), ...
Add more articles on the application of NDRE. Write about ERA5_Land: a brief review of the state of the art, its limitations and errors when using this type of reanalysis. Write about the ERA5_Land variable that was used.

103. Figure 1 was not cited in the text. Please do so!

107. ... 6? × 6 m planting grid (277 trees ha⁻¹) and is managed under rainfed conditions (?). Write 6 m x 6 m! Describe the climatic conditions of the region. The rainy season, the dry season.
112. Figure 1.... Improve Figure 1. Add the geographic coordinates, figure scale, and north indicator. The source from which the maps were obtained. What type of color composition is it? Natural color? From which satellite? Wouldn't you have a better resolution image to show the avocado orchards? Figure 1a doesn't clearly show Figure 1b. Figure 1c doesn't clearly show it in Figure 1b. Why isn't Figure 1C a Sentinel-2 image? This image
118... specific climatic components (What are these components? List them) are subsequently integrated as explanatory covariates for the ...
120... The production system follows intensive agronomic management aligned with GlobalG.A.P. certification standards(?)... What standards are these? List them.

122... and foliar fertilization synchronized with phenological stages (Which ones? List them.), and manual weed control.

Create a flowchart of the methodology's processing steps.

128... An observational-correlational design (?) with an explanatory scope was adopted to... Show a Figure with this design with the five areas. Figure 1C is not clear! Improve it!

131... five environments (A1–A5) (Figure 1c) distributed across an altitudinal gradient ranging... What are the differences between the five orchards that deserve highlighting?
132... Following a comprehensive dendrometric census (What criteria were used to conduct this census? Explain this in the text!), five trees per...
135... was verified using variables derived from a Digital Elevation Model (DEM) (What is the source of the DEM? Cite it.),... It would be interesting to show a map of the local topography.
136-137... slope, Topographic Wetness Index (TWI)(?), Topographic Position Index (TPI) (?), and aspect (?), which confirmed the absence of significant geomorphological differences other than those associated with the altitudinal gradient [14]. Show the formulas and interpretation of the variables.
139-147. Improve the writing of the sentence.
140... Commercial harvesting was conducted across three distinct events (What events are these? Cite them) during the 2025 production cycle, adhering to local producer operational criteria (What criteria?) and international export market (What standard is this?) standards [15]. Please explain the sentence in more detail, showing details. Show a figure showing the phenological phases of the avocado.

142... Commercial maturity was determined by fruit caliber (?) What is the value of this caliber? How is it measured?

162... based on commercial caliber (?) and mass thresholds (?). State these mass and caliber limits.

163... A size-fractionated harvest strategy (?) was implemented... Explain this strategy in more detail. Give examples of fruits that followed this strategy. Show a figure with different fruit sizes.

169... recorded across C1, C2, and C3, allowing for... Explain what C1, C2, and C3 are?

191... processed per tree according to AOAC Method (?) 934.06, where DM was determined by... Explain this method in more detail! What is AOAC?

198... implemented in Google Earth Engine (GEE), utilizing the ECMWF/ERA5-LAND/HOURLY reanalysis (~9 km resolution). Provide the GEE website! What does ECMWF/ERA5-LAND mean?

209... ported in CSV (What is CSV?) format for integration with spectral, yield, and quality variables. 210. What are the spatial, radiometric, and spectral resolutions of the Sentinel-2 bands used in the research?
223-225... Precise GPS coordinates for the 25 experimental units were acquired using a DJI RTK3 antenna and subsequently reprojected to the EPSG:32619... What are GPS, DJI RTK3, and EPSG?
229... What are the variables of the indices? Explain the interpretation of each index.
232-233... In this study, specific spectral indicators (? Cite them) were selected for their proven robustness in complex agricultural landscapes.
236... As detailed in Table 1, this methodological synergy allows for the... Table 1 should be below your citation in the text. Place the table below this paragraph.

244... compromising the integrity of critical phenological events (What events are these? Cite them) (Guo 244 et al., 2023; Rahman et al., 2022) ? Outside the norms. Use [41, 42?]
245... All data processing was executed in a Python (Which version?) environment using Google Colab (Which website? Cite it).
247-249 ... For the comparative analysis, individual observations were assigned to operative phenological phases (?), vegetative growth, flowering, fruit development/filling, and harvest, based on the calendar month. Show a Figure with the phenological phases and their time periods. Is the crop annual?
265-266... ​​of all records and outlier detection using robust criteria based on the interquartile range (?). Detail the interquartile range further.

276... Normality (?) and homoscedasticity (?) assumptions were evaluated using Shapiro–Wilk and Levene’s tests, respectively, performed on the model residuals. It is important to briefly describe how normality and homoscedasticity are verified by the Shapiro-Wilk and Levene’s tests. Please include bibliographic citations.

277... When these assumptions were satisfied, Analysis of Variance (ANOVA) was applied; otherwise, the non-parametric Kruskal–Wallis (?) test was employed. Was the non-parametric test also applied? Explain it. For what situation is the non-parametric test used?

279-280... Post-hoc comparisons were facilitated through Tukey’s Honest Significant Difference (HSD) (?) for parametric data or Dunn’s test (?) with a Bonferroni adjustment for non-parametric analyses [44]. It is extremely crucial to make clear how these tests are interpreted! 291... was examined using Principal Component Analysis (PCA) on z-score (?) standardized variables. Please clarify the interpretation of PCA and z-score.
296... with effect sizes reported and significance levels adjusted using the False Discovery Rate (FDR) (?).... Explain what FDR is and what it is used for? It is not clear in the text.
300... The importance of each predictor was determined via Variable Importance in Projection (VIP) scores, with a threshold (?) of VIP > 1 established to identify relevant determinants [47,48]. And if the threshold is less than or equal to 1, how do I interpret VIP?
303. List the formulas for the statistical metrics: R², Q², the REQM, and Variance Inflation Factor.
308... During this process, multicollinearity was strictly controlled by ensuring a Variance Inflation Factor (VIF) (?) of less than 10 [49]. What is the VIF factor and how to interpret it?
343. Figure 2. The colors of the phenological phases within the graphs are imperceptible. Please increase the shades and add the names of the phenological phases within the graph. In the title of Figure 2, indicate the monthly time period of each phenological phase. In the graphs, change accumulated precipitation to rainfall and average temperature to average air temperature.

355... 3.2. Phase-Specific Spectral Responses: Index–Climate Associations... The associations between index and climate are practically nonexistent or very weak. How to explain this?

430... Table 3. What is EE in Table 3?

437.., Cite Figure 5 in the text.

574. What is (a) at the beginning of Figure 9? Where is (b)?
580... the model achieved an R² (?) of 0.346 and a Q² of 0.173, suggesting acceptable... The R² remains very low, indicating that the explanatory power of one variable in relation to another is very low, or not?
641-659... defined thresholds (Lahack et al., 2025 ?). Excessive vegetative vigor during flow...
..both fruit caliber and commercial homogeneity (Cano-Gallego et al., 2023?).

In contrast to... Citations outside the journal's standard.

Author Response

Reviewer 4

We sincerely thank the reviewer for the careful reading of our manuscript and for the constructive and detailed comments. The suggestions significantly improved the clarity, methodological transparency, and interpretation of the results. Below we address each comment individually and describe the modifications incorporated in the revised manuscript.

 

Comment 1: 12 ... Monitoring avocado (Persea americana Mill., cv. Semil-34) in tropical mountain (? Specify location) landscapes is inherently complex due to the pronounced topographical

Response 1: We agree with the reviewer's observation regarding the importance of geographic context. The specific location of the study has been incorporated into the abstract to provide immediate spatial clarity.

“The text in line 12 has been updated as follows: "...monitoring avocado (Persea americana Mill., cv. Semil-34) in tropical mountain landscapes of Cambita, San Cristóbal, Dominican Republic, is inherently complex..."

 

Comment 2; 14. ...modulates the crop’s ecophysiological responses (?). What are these responses? List them.

Response 2: We appreciate the reviewer's request for precision. The abstract has been revised to explicitly list the specific ecophysiological responses investigated in this study, ensuring a clearer technical overview of the parameters addressed and avoiding generic terminology.

The text in line 14 has been updated as follows: "...modulates the crop’s ecophysiological responses, specifically vegetative vigor, carbon allocation, and the synchronization of reproductive flushes."

 

 

Comment 3; This study integrates multi-year (? Specify the time series period) Sentinel-2 time series, ERA5-Land climatic variables (Which variables?), and geomorphometric covariates to explain variability in yield and fruit quality. What errors are associated with the ERA5-Land variables? Did the authors seek to validate ERA5-Land before using it in the research?

Response 3: We appreciate the reviewer's request for technical precision. Following this suggestion, the specific timeframe (2020–2025) and the primary climatic variables have been explicitly stated in the abstract. To ensure a concise abstract while maintaining scientific rigor, the detailed justification for the ERA5-Land dataset—addressing its advantages in regions with sparse local weather stations, associated errors, and validated performance in similar tropical environments—has been incorporated into the Introduction and Methodology sections.

The text in lines 15–18 of the Abstract has been updated as follows: "...integrates a 5-year (2020–2025) Sentinel-2 time series, ERA5-Land climatic variables (air temperature, total precipitation, and radiation), and geomorphometric covariates to explain variability in yield and fruit quality."

Furthermore, the following technical details have been integrated into the Introduction (Section 1) and Methodology (Section 2.5.1):

• Introduction: Expanded discussion on the ecophysiological requirements of avocado and the risks posed by climate change, justifying the need for high-frequency reanalysis data.
• Methodology: Specified the ERA5-Land variables (T2m, Tskin, TP, PAR), the GEE source, and a brief review of the state of the art, including the role of 4D-Var data assimilation in mitigating errors in mountain landscapes.

 

Comment 4; 23 ...were associated with lower production (r = −0.58, p < 0.0021 ?), reflecting a critical source... Although p is significant, R² does not explain the relationship. How do you justify this?

Response 4: We agree that in complex hillside environments, external factors contribute to residual variance. However, the high statistical significance (p < 0.01) confirms that the identified "vigor paradox" is a robust and non-random biological trend. We have surgically refined the abstract to better reflect the significance of this association within the context of mountain agroecosystems.

The text in line 25-26 has been updated as follows: “High vegetation index values were significantly and negatively associated with lower production (r = −0.58, p < 0.0021), reflecting a critical source–sink imbalance.”

 

Comment 5; 25... While yield variability was partially explainable (R² = 0.38?), internal fruit quality,... In fact, the topographic variable only managed to explain 38% of productivity. Is this a good result for your research, although not one of the best?

Response 5: We respectfully argue that an R² = 0.38 is a robust and scientifically significant result given the study's context. Modeling tree-level productivity in complex tropical mountain landscapes using exclusively indirect signals (satellite and topography) involves high environmental stochasticity. Capturing nearly 40% of the variance provides a highly valuable operational framework for site-specific management in systems where no such predictive tools previously existed. We have refined the abstract and discussion to better contextualize the significance of this explanatory power.

 

The text in line 27 has been updated as follows: “Topography functioned as a structural filter, regulating root drainage and productive stability across the landscape. While yield variability was significantly, albeit partially explainable (R² = 0.38), internal fruit quality, measured as dry matter content, exhibited high environmental stability.”

Comment 6; 40-41... where topographic and climatic heterogeneity imposes structural constraints (Which ones? Cite them) on both agronomic management (What type? Cite them) and yield stability (What does this mean?).
Response 6: We appreciate the reviewer's suggestion to ground the introductory framework in specific operational realities. The text has been expanded to explicitly list the biophysical constraints and management types affected by hillside topography. Additionally, a definition of yield stability in the context of mountain landscapes has been provided to ensure conceptual clarity and scientific rigor.

 

The text in line 42-46 has been updated as follows: “In tropical and subtropical regions, production has increasingly expanded into mountain landscapes, where topographic and climatic heterogeneity imposes structural constraints specifically steep gradients, restricted soil drainage, and soil erosion risk, on both agronomic management (primarily mechanized fertilization, weed control, and harvest logistics) and yield stability, referring to the spatial and temporal consistency of productive performance across the complex landscape.”

 

Comment 7; 48... particularly during the flowering stage (? How many days does the flowering phase last?),...

Response 7: We agree that specifying the biological timing of phenological events strengthens the manuscript's precision. The estimated duration of the flowering phase for cv. Semil-34 under the studied tropical conditions has been incorporated based on regional phenological records, providing a clearer temporal context for the reproductive period.

 

The text in line 59-61 has been updated as follows: “Research evidence suggests that excessive vegetative vigor, particularly during the flowering stage (typically spanning 30 to 45 days in these tropical hillside environments), can function as a competitive sink that compromises fruit set and filling, thereby inducing nonlinear relationships between growth and yield.”

 

Comment 8; 51-52... necessitating analytical frameworks (What are these structures? Give an example!) that explicitly integrate the temporal (How long? Give an example.) and functional (What type? Cite them.) dimensions of the crop.

Response 8: We appreciate the reviewer's request for conceptual clarity. The manuscript has been revised to provide specific examples of the analytical frameworks and dimensions addressed. We have explicitly mentioned Partial Least Squares Regression (PLSR) as the integrative framework, along with the 5-year multi-temporal series and the source–sink dynamics as the functional components.

The text in line 65-67 has been updated as follows: “Consequently, this metabolic complexity challenges the direct interpretation of canopy “greenness” as a proxy for productive performance, necessitating analytical frameworks, such as Partial Least Squares Regression (PLSR), that explicitly integrate the temporal (e.g., 5-year phenological time series) and functional (e.g., source–sink competition and biomass-yield trade-offs) dimensions of the crop.”

 

 

Comment 9; 56-57... spatial resolution ( ? Cite them.), temporal frequency (? Cite it), and specific spectral bands, most notably the red edge ( ? Cite the interval!), ...

Response 9: We agree that providing precise technical specifications for the satellite sensor is essential for methodological transparency. The spatial resolution, temporal revisit frequency, and the specific wavelength interval for the red-edge region have been incorporated into the text, supported by relevant technical citations.

The text in line 72-75 has been updated as follows: “In particular, the European Space Agency's Sentinel-2 constellation offers a unique synergy of spatial resolution (10-20 m), temporal frequency (5 days), and specific spectral bands, most notably the red edge (705–783 nm), which is exceptionally well-suited for capturing variations in chlorophyll content and photosynthetic activity .”

 

Comment 10; Add more articles on the application of NDRE. Write about ERA5_Land: a brief review of the state of the art, its limitations and errors when using this type of reanalysis. Write about the ERA5_Land variable that was used.

Response 10: We appreciate this constructive observation, which coincides with the interests expressed by other reviewers regarding the robustness of our data. We have significantly expanded the Methodology section (Section 2.5.1) to include a comprehensive review of the state of the art for both the NDRE index and the ERA5-Land reanalysis. We have added specific citations regarding the application of NDRE in high-biomass perennial crops, highlighting its capability to avoid the saturation issues common with NDVI. Furthermore, we have provided a critical assessment of ERA5-Land, detailing its advantages for regions with limited ground-based stations, while explicitly acknowledging its limitations in complex terrains—such as topographic smoothing and potential thermal biases (RMSE of 1.5–2.5 °C). The specific variables used and their acquisition via Google Earth Engine are now clearly listed.

The text in the Introduction and Section 2.5.1 has been updated as follows: "The NDRE index provides superior sensitivity in dense perennial canopies by utilizing the red-edge transition zone, effectively capturing physiological changes that are often masked by NDVI saturation. Regarding climatic data, ERA5-Land provides a global 9-km resolution framework using 4D-Var data assimilation. While it is the most robust reanalysis for tropical areas with sparse weather stations, we have detailed its structural limitations in hillside environments, such as the smoothing of extreme precipitation events. In this study, we specifically integrated 2-m air temperature, skin temperature, total precipitation, and surface solar radiation (converted to PAR) to identify the environmental drivers of avocado phenology."

 

Comment 11; 103. Figure 1 was not cited in the text. Please do so!

Response 11: We thank the reviewer for identifying this technical omission. The citation for Figure 1 has been properly integrated into the Study Area description to ensure that the text and the visual evidence are correctly linked, improving the overall readability of the section.

The text in line 128 has been updated as follows: “The research was conducted in a commercial avocado orchard (Persea americana Mill., cv. Semil-34) located in Cambita, San Cristóbal Province, Dominican Republic (Figure 1) (18.472630° N, −70.199002° W; WGS84).”

 

Comment 12; 107. ... 6? × 6 m planting grid (277 trees ha⁻¹) and is managed under rainfed conditions (?). Write 6 m x 6 m! Describe the climatic conditions of the region. The rainy season, the dry season.

Response 12: We appreciate the reviewer's suggestion to standardize technical notations and provide a more detailed climatic characterization. The planting grid notation has been corrected to "6 m x 6 m." Furthermore, Section 2.1 has been expanded to include a description of the local tropical sub-humid climate, explicitly detailing the temporal distribution of the rainy and dry seasons, which is fundamental to understanding the water dynamics of this rainfed production system.

The text in line 132-135 has been updated as follows: “The orchard was established with a 6 m × 6 m planting grid (277 trees ha⁻¹) and is managed under rainfed conditions. The region is characterized by a tropical sub-humid climate with two distinct periods: a rainy season from May to November, influenced by tropical waves and convective activity, and a dry season from December to April. This seasonality dictates the moisture availability for the phenological stages of the Semil-34 cultivar.”

Comment 13; 112. Figure 1.... Improve Figure 1. Add the geographic coordinates, figure scale, and north indicator. The source from which the maps were obtained. What type of color composition is it? Natural color? From which satellite? Wouldn't you have a better resolution image to show the avocado orchards? Figure 1a doesn't clearly show Figure 1b. Figure 1c doesn't clearly show it in Figure 1b. Why isn't Figure 1C a Sentinel-2 image? This image

Response 13: "We appreciate the reviewer's detailed feedback. Figure 1 has been completely redesigned. We have integrated a coordinate grid (WGS84), graphic scales, and north indicators in all panels. Regarding Figure 1b, a high-resolution RGB orthophoto from a DJI Mavic Dron, was selected instead of Sentinel-2 to allow for the clear visual identification of individual tree crowns and environment boundaries, which would be impossible at a 10-meter resolution. However, all subsequent spectral analyses (Section 3) are strictly based on Sentinel-2 imagery as described in the methodology."

 

“Figure 1 has been updated to include a coordinate system (WGS84), a graphic scale bar, and a north indicator to ensure professional cartographic quality and spatial accuracy.”

Comment 14;118... specific climatic components (What are these components? List them) are subsequently integrated as explanatory covariates for the ...

Response 14: We appreciate the reviewer's request for clarity. To improve the traceability of our methodology, we have explicitly listed the hydroclimatic variables derived from the ERA5-Land reanalysis that were utilized as covariates in our multivariate modeling and phenological analysis.

The text in line 158 has been updated as follows: “This characterization is provided in this section solely as a descriptive site framework, while specific climatic components, including air temperature (mean, maximum, and minimum), skin temperature, cumulative precipitation, and photosynthetically active radiation (PAR), are subsequently integrated as explanatory covariates for the crop's spectral, productive, and quality responses.”

Comment 15; 120... The production system follows intensive agronomic management aligned with GlobalG.A.P. certification standards(?)... What standards are these? List them.

Response 15: We agree that detailing the specific management standards is crucial for establishing the uniformity of agronomic practices across the environmental gradient. We have revised the text to explicitly list the key technical pillars of the GlobalG.A.P. certification implemented in the orchard, ensuring that the observed variations are attributable to the environmental drivers studied rather than inconsistent management practices.

 

The text in line 163-167 has been updated as follows:  “The production system follows intensive agronomic management aligned with Global G.A.P. certification standards, specifically encompassing Integrated Pest Management (IPM), documented nutrient management plans based on soil and foliar analysis, standardized pre-harvest intervals for phytosanitary applications, and rigorous traceability and worker safety protocols.”

 

Comment 16; 122... and foliar fertilization synchronized with phenological stages (Which ones? List them.), and manual weed control.

Response 16: We have integrated the strategic standards of the Global G.A.P. certification with the specific operational actions of the orchard. This ensures a comprehensive description of the intensive management system, highlighting that nutrient and phytosanitary applications were not only documented but also strictly synchronized with the biological demands of the tree's phenological cycle.

The text in line 167-169 has been updated as follows: “This framework encompassing scheduled phytosanitary control, edaphic and foliar fertilization synchronized with phenological stages, specifically flowering, fruit set, and filling, complemented by manual weed control. The orchard spans an operational elevational gradient of 250–450 m above sea level (m a.s.l.), featuring high topographic heterogeneity and local slopes exceeding 40%. These conditions justify its selection as a representative model for tropical mountain environments.”

 

 

Comment 17; Create a flowchart of the methodology's processing steps.

Response 17: We agree that a visual synthesis of the methodology significantly enhances the clarity and transparency of our research workflow. Following the reviewer’s suggestion, a new figure (Figure 2) has been integrated into the manuscript. This flowchart provides a structured overview of the sequential stages of the study, from multi-source data acquisition and satellite-climatic processing to tree-level spatial integration and the final multivariate statistical modeling framework.

“A new flowchart (Figure 2) has been added to the Methodology section.”

 

Comment 18; 128... An observational-correlational design (?) with an explanatory scope was adopted to... Show a Figure with this design with the five areas. Figure 1C is not clear! Improve it!

Response 18: We appreciate the reviewer's call for clarity regarding the experimental framework. The term "observational-correlational" reflects our approach of monitoring existing environmental gradients without experimental manipulation to identify statistical associations. The "explanatory scope" is justified by our use of multivariate modeling (PLSR) to identify the drivers of yield variability. To address the visual concerns, Figure 1b has been completely redesigned using a higher-resolution RGB orthophoto and clear, high-contrast delineations for each of the five environmental strata (A1–A5). This ensures that the spatial stratification and the distribution of the individual experimental units (trees) are immediately apparent to the reader.

“Figure 1b has been updated with improved color-coded polygons for environments A1 through A5 and high-visibility markers for each of the 25 sampled trees. Additionally, a hierarchical diagram has been integrated into the methodology to explicitly show the nesting of trees within the environmental strata, reinforcing the understanding of the observational-correlational design.”

 

Comment 19; 131... five environments (A1–A5) (Figure 1c) distributed across an altitudinal gradient ranging... What are the differences between the five orchards that deserve highlighting?

Response 19: We appreciate the reviewer's request to characterize the environmental strata. The selection of the five environments (A1–A5) was not arbitrary; they were strategically chosen to represent a topoclimatic gradient. The primary differences lie in the topographic position index (TPI) and slope dynamics, which directly influence water residence time and soil moisture availability. For instance, A1 represents the upper-slope positions with higher solar exposure and rapid drainage, whereas A5 corresponds to lower-slope positions with higher soil depth and accumulation of runoff. These physical contrasts are the drivers behind the observed variability in vegetation indices and fruit quality. We have added a summary table in Section 2.1 to explicitly detail these differences.

The text in line 193-201 has been updated as follows: “Topographic comparability across environments was verified using variables derived from a Digital Elevation Model (DEM), including slope, Topographic Wetness Index (TWI), Topographic Position Index (TPI), and aspect, which confirmed the absence of significant geomorphological differences other than those associated with the altitudinal gradient. A1 and A2 are located in high-slope areas (> 25%) with convex landforms, promoting high runoff; A3 serves as a transitional mid-slope zone; and A4 and A5 are situated in concave toe-slope positions with gentler gradients (< 10%), characterized by higher soil moisture retention and cumulative radiation.”

Comment 20;132... Following a comprehensive dendrometric census (What criteria were used to conduct this census? Explain this in the text!), five trees per...

Response 20: We appreciate the reviewer's request for a detailed description of the sampling criteria. To ensure the reliability of the spectral-yield relationships, the 25 experimental units were selected using a rigorous three-step filtering process: (1) Genetic uniformity, selecting only cv. Semil-34 individuals; (2) Structural homogeneity, choosing trees whose Diameter at Breast Height (DBH) fell within the population median to minimize age or vigor bias; and (3) Spatial independence, applying a minimum distance filter of >10 meters between selected trees. This spatial constraint ensures that each tree represents a distinct sampling unit, avoiding spatial autocorrelation and ensuring that each unit corresponds to independent pixels in the Sentinel-2 imagery (10 m resolution).

The text in line 180-184 has been updated as follows: “Following a comprehensive dendrometric census, five trees per environment were selected (n = 25) based on trunk diameters near the stratum median to ensure basal morphological homogeneity. To further ensure the robustness of the spectral-yield relationships, the selection was limited to the cv. Semil-34 variety, and a minimum inter-tree distance of 10 m was strictly maintained. This spatial buffer ensures statistical independence and prevents spectral signal overlap between pixels, matching the native resolution of the Sentinel-2 bands.”

Comment 21; 135... was verified using variables derived from a Digital Elevation Model (DEM) (What is the source of the DEM? Cite it.),... It would be interesting to show a map of the local topography.

Response 21: We appreciate the reviewer's emphasis on topographic precision. The Digital Elevation Model (DEM) used for this study was acquired via LandViewer (EOS Data Analytics), with a native spatial resolution of 4.7 meters. This high-resolution dataset allowed for the precise calculation of primary and secondary topographic indices, ensuring that the environmental characterization of the hillside gradient was conducted at a finer scale than the 10-meter spectral data. As requested, we have incorporated a new map (Figure 1) showcasing the local topography, specifically the elevation, to visually demonstrate the environmental complexity of the study site.

The text in line 193-196 has been updated as follows: “Topographic comparability across environments was verified using variables derived from a high-resolution Digital Elevation Model (DEM) (4.7 m native resolution) acquired through LandViewer (EOS Data Analytics), including slope, Topographic Wetness Index (TWI), Topographic Position Index (TPI), and aspect, which confirmed the absence of significant geomorphological differences other than those associated with the altitudinal gradient.”

 

Comment 22; 136-137... slope, Topographic Wetness Index (TWI)(?), Topographic Position Index (TPI) (?), and aspect (?), which confirmed the absence of significant geomorphological differences other than those associated with the altitudinal gradient. Show the formulas and interpretation of the variables.

Response 22: We appreciate the reviewer's request for mathematical and conceptual precision. We have incorporated the formulas for the topographic indices used and provided their ecophysiological interpretation within the context of hillside avocado production. The inclusion of these variables allowed us to confirm that the environmental strata (A1–A5) are primarily differentiated by their hydrological behavior and energy balance, while maintaining geomorphological stability in terms of parent material and soil classification.

 

The text in line 399-415 has been updated as follows:

“Using this high-resolution topographic data, the structural filters of the landscape were quantified and interpreted as follows:

• Slope (β): Represents the local rate of change in elevation, measured in degrees (°).
• Aspect: Defined as the azimuthal orientation of the slope (0°–360°).
• Topographic Wetness Index (TWI): Quantifies the potential for water accumulation based on the upslope contributing area (α) and the local slope (β):

where β is the slope angle in degrees (°). Higher values indicate a greater potential for soil saturation

Topographic Position Index (TPI): Discerns landforms by comparing the elevation of a specific point  to the mean elevation () of its surrounding neighborhood:

Positive values represent ridges or convex landforms (potential for high drainage), while negative values indicate valleys or concave positions (potential for water accumulation).”

Comment 23; 139-147. Improve the writing of the sentence. “Commercial harvesting was conducted across three distinct events during the 2025 production cycle, adhering to local producer operational criteria and international export market standards.”

Response 23: We agree that the description of the harvesting process required more technical precision. The sentence has been restructured to emphasize the systematic nature of the harvest cycles and their alignment with maturity indicators and international quality protocols.

The text in line 208-211 has been updated as follows:  “Commercial harvesting was executed in three discrete cycles during the 2025 production season, strategically timed to align with fruit physiological maturity and strictly adhering to both regional operational protocols and international export quality standards”

Comment 24; 140... Commercial harvesting was conducted across three distinct events (What events are these? Cite them) during the 2025 production cycle, adhering to local producer operational criteria (What criteria?) and international export market (What standard is this?) standards. Please explain the sentence in more detail, showing details. Show a figure showing the phenological phases of the avocado.

Response 24: We agree that the description of the harvesting process required more technical precision. The sentence has been restructured to emphasize the systematic nature of the harvest cycles and their alignment with maturity indicators and international quality protocols.

The text in line 211-217 has been updated as follows:  “Commercial maturity was executed in three discrete cycles (September, October, and November 2025), corresponding to the early, mid, and late maturity windows. These events were governed by local operational criteria, primarily fruit caliber (sizes 24–18) and minimum dry matter thresholds (> 18% ) to ensure post-harvest ripening quality.”

 

Comment 25; 142... Commercial maturity was determined by fruit caliber (?) What is the value of this caliber? How is it measured?

Comment 26; 162... based on commercial caliber (?) and mass thresholds (?). State these mass and caliber limits.

Response 25-26: We appreciate the reviewer's query regarding the harvest selection process. To maintain the study's alignment with real-world commercial conditions, the harvest was executed by the farm's highly experienced technical staff. These individuals acted as expert evaluators, selecting fruits based on standardized commercial maturity indicators—primarily fruit caliber (weight and equatorial diameter) and skin texture/coloration—without interference from the research team. This observational approach ensures that the data reflects the actual performance of the production system under standard export-oriented management. The selected fruits consistently fell within the 12–18 caliber range (approx. 360–600 g), a benchmark for the international green-skin avocado market.

The text in line 219-227 has been updated as follows:  “During each event, the total volume of commercial-grade fruit from the 25 selected trees was harvested. To ensure the research captures the authentic dynamics of a commercial operation, harvesting was performed by the farm's specialized crew following established export-quality protocols. The research team maintained a non-interventional role during selection, allowing expert personnel to identify fruits at commercial maturity based on phenotypic indicators such as caliber and skin lenticel density. This ensured that the sampled units represented the true commercial output of each environmental stratum, with selected fruits strictly conforming to international market standards.”

 

Comment 27; 163... A size-fractionated harvest strategy (?) was implemented... Explain this strategy in more detail. Give examples of fruits that followed this strategy. Show a figure with different fruit sizes.

Response 27: We appreciate the reviewer's request for detail. The "size-fractionated harvest strategy" is evidenced by the weight distribution and quality metrics presented in the new Figure 5 - 6 and Table 4. These results show how the selective harvest maintained the fruit parameters within the commercial export window (calibers 12–18). Figure 5 specifically illustrates the dispersion and stability of fruit weight and dry matter across the environmental gradient (A1–A5), while Table 4 provides the ANOVA results confirming that while the strategy was uniform, the environment significantly influenced the final yield and morphometric traits.

The text in line 229-236 has been updated as follows:  “For the assessment of internal quality and morphometry, five median-mass fruits per tree were selected to form composite samples per plant and harvest event. This protocol minimizes intra-tree variance, ensuring a representative estimation of the mean physiological status of the experimental unit”

 

Comment 28; 169... recorded across C1, C2, and C3, allowing for... Explain what C1, C2, and C3 are?

Response 28: We appreciate the reviewer's point regarding the need for explicit definitions of abbreviations. C1, C2, and C3 refer to the three discrete harvest cycles (Collections) conducted during the 2025 season. These cycles were established to capture the temporal variability of fruit maturity and quality under the size-fractionated harvest strategy. The manuscript has been updated to define these terms upon their first mention.

The text in line 239-241 has been updated as follows: “Response variables were defined at the tree level (experimental unit) for each harvest cycles: C1 (October), C2 (November), and C3 (December), ensuring hierarchical traceability across the Environment–Tree–Harvest levels.”

 

Comment 29; 191... processed per tree according to AOAC Method (?) 934.06, where DM was determined by... Explain this method in more detail! What is AOAC?

Response 29: We appreciate the reviewer’s request for analytical detail. AOAC International (originally the Association of Official Analytical Chemists) is a globally recognized organization that develops and validates standardized analytical methods for food and agricultural sciences. While Method 934.06 provides the general framework for gravimetric moisture determination, we optimized the drying temperature to 65 °C. This specific adjustment was implemented to preserve the integrity of the avocado’s lipid-rich matrix, preventing potential oxidative degradation or the loss of volatile compounds that can occur at higher temperatures, thereby ensuring a more precise and stable dry mass measurement for cv. Semil-34.

 

The text in line 282-290 has been updated as follows: “For analytical reference, composite samples were processed per tree according to AOAC International (Association of Official Analytical Chemists) Method 934.06.  Specifically, 20 g of homogenized pulp was placed in aluminum dishes and dehydrated in a forced-air oven at 65 °C until a constant weight was achieved (approximately 48–72 h). This temperature was strategically selected to minimize lipid oxidation and the thermal degradation of non-structural carbohydrates, which are critical in high-oil matrices such as avocado. The DM percentage was calculated using the gravimetric formula: DM (%) = (W_dry / W_fresh)*100, where W_fresh is the initial mass and W_dry is the final mass after dehydration.”

 

Comment 30; 198... implemented in Google Earth Engine (GEE), utilizing the ECMWF/ERA5-LAND/HOURLY reanalysis (~9 km resolution). Provide the GEE website! What does ECMWF/ERA5-LAND mean?

Response 30: We appreciate the reviewer's suggestion to formalize the technical nomenclature and data access points. Following this recommendation, we have included the official URL for the Google Earth Engine (GEE) platform (https://earthengine.google.com/) and provided the full institutional name for the ECMWF (European Centre for Medium-Range Weather Forecasts). We have also clarified that ERA5-Land was specifically selected as the state-of-the-art land-surface reanalysis due to its optimized resolution for agricultural applications in complex terrains.

The text in line 294-300 has been updated as follows: “A continuous hydroclimatic data extraction workflow for the 2020–2025 period was implemented in Google Earth Engine (GEE) (https://earthengine.google.com/), utilizing ERA5-Land dataset,  produced by the European Centre for Medium-Range Weather Forecasts (ECMWF), providing a synthesized estimate of land-surface variables at a (~9 km resolution).”

 

Comment 31; 209... ported in CSV (What is CSV?) format for integration with spectral, yield, and quality variables. 210. What are the spatial, radiometric, and spectral resolutions of the Sentinel-2 bands used in the research?

Response31: We appreciate the reviewer's request for technical formalization. Comma-Separated Values (CSV) was utilized as the universal, non-proprietary standard for structured data interchange, ensuring seamless interoperability between the climatic reanalysis, the spectral time-series, and the field-level productive datasets. Regarding the Sentinel-2 constellation, we have updated the manuscript to explicitly detail the resolutions of the bands used in the vegetation indices (NDVI, EVI, NDRE, and NDMI). Specifically, the bands operate at a 12-bit radiometric resolution, with spatial resolutions of 10 m and 20 m depending on the spectral region (Visible/NIR vs. Red Edge/SWIR).

“The complete daily time series (2005–2025) was exported in Comma-Separated Values (CSV) format for integration with spectral, yield, and quality variables.”

 

The text in line 318-323 has been updated as follows:  “The image collection was filtered according to the orchard polygon, a 2020–2025 temporal window, and a cloud cover threshold of <10%. To characterize the canopy status, specific Sentinel-2 bands were processed at their native resolutions: spatial resolution (10 m for B4 and B8; 20 m for B5, B6, B7, and B11), radiometric resolution (12-bit, stored as 16-bit integers), and spectral resolution (e.g., B8 NIR at 842 nm and B5 Red Edge at 705 nm).”

Comment 32; 223-225... Precise GPS coordinates for the 25 experimental units were acquired using a DJI RTK3 antenna and subsequently reprojected to the EPSG:32619... What are GPS, DJI RTK3, and EPSG?

Response 32: We appreciate the reviewer's suggestion to provide a more rigorous description of the geolocation and spatial reference protocols. We have updated the manuscript to clarify that the Global Positioning System (GPS) was utilized. The DJI RTK module (Real-Time Kinematic) was employed to ensure centimeter-level horizontal and vertical accuracy, which is critical for tree-level spectral extraction. Furthermore, EPSG:32619 was specified as the standardized Coordinate Reference System (CRS) code corresponding to WGS 84 / UTM zone 19N, the official projection for the study region in the Dominican Republic.

The text in line 330-333 has been updated as follows: “Spectral analysis was conducted at the individual tree level. Precise Global Positioning System (GPS) coordinates for the 25 experimental units were acquired using a DJI (Real-Time Kinematic (RTK) RTK positioning antenna and subsequently reprojected to the EPSG:32619 reference code (WGS 84 / UTM zone 19N), which is the standardized Coordinate Reference System (CRS) for the Dominican Republic. For each tree, the value from the nearest 10-m pixel was extracted.”

Comment 33; 229... What are the variables of the indices? Explain the interpretation of each index.

Response 33: We appreciate the reviewer's request for an explicit definition of the spectral variables and their physiological interpretations. To ensure maximum clarity and technical rigor, we have synthesized this information into Table 2. This table now details the mathematical formulations, the specific Sentinel-2 bands involved, and the biophysical proxy each index represents (structural, nutritional, or hydric). This integrated view allows for a transparent interpretation of the canopy’s ecophysiological status throughout the study.

 

The text in line 355-337 has been updated as follows:

Index	Proxy	Equation	Biophysical Proxy & Interpretation	Reference
NDVI	Vegetative Vigor	(NIR − Red) / (NIR + Red)	Vegetative vigor and green biomass.
EVI	Canopy Structure	2.5 × (NIR − Red) / (NIR + 6·Red − 7.5·Blue + 1)	Structural descriptor
NDRE	Chlorophyll	(NIR − RedEdge) / (NIR + RedEdge)	Chlorophyll density and nitrogen status
NDMI	Canopy Water Status	(NIR − SWIR) / (NIR + SWIR)	Canopy hydric status

The mathematical equations, specific bands involved, and the biophysical interpretation for each index (NDVI, EVI, NDRE, and NDMI) are systematically detailed in Table 2 “

Comment 34; 232-233... In this study, specific spectral indicators (? Cite them) were selected for their proven robustness in complex agricultural landscapes.

Response 34: We appreciate the reviewer's suggestion to explicitly support our selection of spectral indicators with authoritative literature. The manuscript has been updated to include key citations demonstrating the performance and reliability of NDVI, EVI, NDRE, and NDMI for monitoring high-biomass perennial systems and heterogeneous terrains. These references justify the use of our multi-index approach for identifying ecophysiological transitions in avocado orchards under mountain conditions.

The text in line 344-347 has been updated as follows: “In this study, specific spectral indicators (NDVI, EVI, NDRE, and NDMI) were selected for their proven robustness in complex agricultural landscapes.”

 

Comment 35; 236... As detailed in Table 1, this methodological synergy allows for the... Table 1 should be below your citation in the text. Place the table below this paragraph.

Response 35: We appreciate the reviewer's guidance on the manuscript's structural flow. Following the journal's editorial standards, we have ensured that all tables are formally introduced and cited within the narrative before their physical placement. Accordingly, we have moved the table (now renumbered as Table 2) to appear immediately after the paragraph that justifies its inclusion. This adjustment improves the document's readability and maintains a logical sequence between the methodological description and the technical data.

 

Comment 36; 244... compromising the integrity of critical phenological events (What events are these? Cite them) (Guo 244 et al., 2023; Rahman et al., 2022) ? Outside the norms. Use [41, 42?]

Response36: We appreciate the reviewer's correction regarding the citation format and the need for phenological specificity. The "critical phenological events" protected by the Savitzky–Golay filter settings correspond to flowering, fruit set, and the intensive fruit-filling phase, which are characterized by rapid shifts in canopy reflectance. We have updated the manuscript to explicitly name these stages and have standardized the citations to the required numerical format, utilizing [1, 37, 42] to support both the time-series smoothing methodology and the phenological scale applied to avocado.

The text in line 373-378 has been updated as follows: “Multi-year time-series of the selected spectral indices (2020–2025) were processed using a Savitzky–Golay filter, configured with a seven-observation window and a second-order polynomial, to preserve the underlying seasonal structure while mitigating high-frequency noise without compromising the integrity of critical phenological events specifically flowering, fruit set, and rapid fruit growth [1, 37, 42]. These stages represent the highest metabolic activity and spectral variability in cv. Semil-34. All data processing was executed in a Python environment using Google Colab.”

Comment 37; 245... All data processing was executed in a Python (Which version?) environment using Google Colab (Which website? Cite it).

Response: We appreciate the reviewer's request for technical precision to ensure the reproducibility of our computational workflow. We have updated the manuscript to specify that Python v3.10 was the environment used for all data integration and statistical modeling. Additionally, we have provided the official URL for Google Colab (https://colab.research.google.com/), which served as the cloud-based platform for executing the Python scripts.

 

The text in line 379-381 has been updated as follows: “All data processing was executed in a Python v3.10 environment using Google Colab (https://colab.research.google.com/), ensuring a reproducible and traceable workflow for the multi-source data integration.”

 

Comment 38; 247-249 ... For the comparative analysis, individual observations were assigned to operative phenological phases (?), vegetative growth, flowering, fruit development/filling, and harvest, based on the calendar month. Show a Figure with the phenological phases and their time periods. Is the crop annual?

Response 38: We appreciate the reviewer's request for biological and structural clarification. To clarify, avocado (Persea americana cv. Semil-34) is a perennial fruit tree; however, it follows a well-defined annual reproductive cycle under the tropical conditions of the Dominican Republic. We have refined the manuscript to better describe our "dual approach," which integrates discrete chronological windows with continuous spectral data.

The text in line 382-390 has been updated as follows: For the comparative analysis, spectral observations were assigned to operative phenological phases: vegetative growth, flowering, fruit development/filling, and harvest, based on the calendar month. This classification was utilized as an auxiliary categorical variable to provide a chronological baseline, which was effectively integrated with the high-resolution continuous dynamics derived from the Savitzky–Golay smoothed series [1, 42]. This dual approach allows for the capturing of both broad developmental windows and the precise, fine-scale ecophysiological transitions of cv. Semil-34 across the landscape.”

Comment 39; 265-266... ​​of all records and outlier detection using robust criteria based on the interquartile range (?). Detail the interquartile range further.

Response 39: We appreciate the reviewer's request for statistical detail. Outlier detection was performed using the Interquartile Range (IQR) method, a robust non-parametric approach that is resistant to the influence of extreme values. The IQR was calculated as the difference between the third quartile (Q₃) and the first quartile (Q₁). Following Tukey’s fences criteria, observations falling outside the range [Q₁ - 1.5 x IQR, Q₃ + 1.5 x IQR] were flagged as potential outliers. This threshold is particularly suitable for agricultural datasets with high environmental heterogeneity, as it provides a stable boundary for identifying anomalous data without assuming a normal distribution [44]

The text in line 422-429 has been updated as follows: “Quality control protocols involved double-entry verification of all records and outlier detection using robust criteria based on the interquartile range (IQR). The IQR was calculated as Q₃ - Q₁. Specifically, the Tukey’s fences method was applied, where values outside the boundaries defined by [Q₁ - 1.5 x IQR, Q₃ + 1.5 x IQR] were identified as potential outliers. This approach is particularly suited for agricultural datasets characterized by environmental heterogeneity, as it provides a stable threshold that is not influenced by the extreme values it aims to detect. Extreme values were only corrected upon confirmation of measurement or data-entry errors; otherwise, they were retained to preserve inherent natural biological variability [44].”

 

 

Comment 40; 276... Normality (?) and homoscedasticity (?) assumptions were evaluated using Shapiro–Wilk and Levene’s tests, respectively, performed on the model residuals. It is important to briefly describe how normality and homoscedasticity are verified by the Shapiro-Wilk and Levene’s tests. Please include bibliographic citations.

Response 40: We appreciate the reviewer's suggestion. We have updated the Statistical Analysis section to include a brief technical description of how the Shapiro–Wilk and Levene’s tests verify the model assumptions. Specifically, we clarified that these tests evaluate the null hypotheses of normal distribution and equal variances, respectively, ensuring the validity of the subsequent parametric or non-parametric inferences. Citations [44] and [45] have been included to support these procedures.

The text in line 438-441 has been updated as follows: “Normality and homoscedasticity assumptions were evaluated using Shapiro–Wilk and Levene’s tests, respectively, performed on the model residuals. The Shapiro–Wilk test was used to verify the null hypothesis that the residuals follow a normal distribution (p > 0.05), while Levene’s test assessed the equality of variances across environmental strata [44].”

 

Comment 41; 277... When these assumptions were satisfied, Analysis of Variance (ANOVA) was applied; otherwise, the non-parametric Kruskal–Wallis (?) test was employed. Was the non-parametric test also applied? Explain it. For what situation is the non-parametric test used?

Response 41: We appreciate the reviewer's request for clarification. The Kruskal–Wallis test was indeed applied to variables that did not meet the parametric assumptions. This non-parametric test is a rank-based alternative to the one-way ANOVA; it is specifically used in situations where model residuals exhibit significant skewness or heteroscedasticity, as it does not assume a normal distribution of errors and is less sensitive to outliers in agricultural field datasets

The text in line 442-446 has been updated as follows:  “When these assumptions were satisfied, Analysis of Variance (ANOVA) was applied; otherwise, the non-parametric Kruskal–Wallis test was employed to compare the distribution of ranks between environmental strata without assuming a normal distribution of errors. This test was specifically utilized for variables where the stochastic component of the model exhibited significant skewness or non-constant variance.”

 

Comment 42; 279-280... Post-hoc comparisons were facilitated through Tukey’s Honest Significant Difference (HSD) (?) for parametric data or Dunn’s test (?) with a Bonferroni adjustment for non-parametric analyses [44]. It is extremely crucial to make clear how these tests are interpreted!

Response 42: We appreciate the reviewer's emphasis on the clarity of the post-hoc interpretation. We have expanded the text to explicitly state that these tests were interpreted by comparing pairwise p-values against a significance threshold of alpha = 0.05. This process allowed for the assignment of a letter-based grouping system (Compact Letter Display), where groups sharing a common letter are considered statistically equivalent, while distinct letters indicate significant differences in means (for Tukey’s HSD) or mean ranks (for Dunn’s test).

 

The text in line 446-452 has been updated as follows: “Post-hoc comparisons were facilitated through Tukey’s Honest Significant Difference (HSD) for parametric data or Dunn’s test with a Bonferroni adjustment for non-parametric analyses . These tests were interpreted based on pairwise p-values using a significance level of 0.05; results are reported using a letter-based grouping system, where the absence of shared letters between environmental strata denotes a statistically significant difference in their respective means or mean ranks.”

 

Comment 43: 291... was examined using Principal Component Analysis (PCA) on z-score (?) standardized variables. Please clarify the interpretation of PCA and z-score.

Response 43: We appreciate the reviewer's request for technical clarification. We have updated the manuscript to explicitly describe the rationale for using z-score standardization, which ensures that all variables contribute equally to the variance analysis regardless of their original units. Additionally, we clarified that PCA is interpreted as a dimensionality reduction technique used to identify the dominant agro-environmental gradients by projecting the data onto orthogonal axes that capture the maximum variance [46].

The text in line 462-473 has been updated as follows: “The multivariate architecture of the system, comprising 17 variables including 5 topography covariates (elevation, slope, aspect, TWI, and TPI), 4 spectral indices (NDVI, NDRE, EVI, and NDMI), and 8 yield and fruit quality parameters (fruit count, total weight, average weight, diameters, pulp and seed mass, and dry matter content), was examined using Principal Component Analysis (PCA) on z-score standardized variables, a procedure that centers each variable at a mean of zero and scales it to unit variance. This transformation is critical to ensure that variables measured in different units—such as yield (kg), elevation (m), and spectral reflectance—contribute equally to the variance calculation, preventing variables with larger numerical ranges from disproportionately influencing the model. PCA was interpreted as a dimensionality reduction technique where the resulting principal components represent orthogonal axes of maximum variability, providing a synthetic view of the dominant agro-environmental drivers.”

 

Comment 44; 296... with effect sizes reported and significance levels adjusted using the False Discovery Rate (FDR) (?).... Explain what FDR is and what it is used for? It is not clear in the text.

Response: We appreciate the reviewer's observation. We have expanded the text to define the False Discovery Rate (FDR) as a multiple-testing correction procedure designed to control the expected proportion of Type I errors (false positives) among the rejected null hypotheses. This adjustment is crucial when evaluating extensive correlation matrices, as it maintains high statistical power while significantly reducing the probability of identifying spurious relationships by chance, a common risk in multi-source agricultural datasets.

The text in line 479-486 has been updated as follows: “Bivariate relationships between predictors and response variables were quantified using Pearson correlation coefficients (r), with effect sizes reported and significance levels adjusted using the False Discovery Rate (FDR). This procedure was employed to control the expected proportion of Type I errors (false discoveries) that can arise when performing multiple simultaneous hypothesis tests across a large set of variables. Unlike more conservative methods, FDR provides a robust balance between statistical power and the prevention of spurious correlations, ensuring that the identified relationships between topography, spectral indices, and fruit quality are statistically sound.”

 

Comment 45; 300... The importance of each predictor was determined via Variable Importance in Projection (VIP) scores, with a threshold (?) of VIP > 1 established to identify relevant determinants [47,48]. And if the threshold is less than or equal to 1, how do I interpret VIP?

Response 45: We appreciate the reviewer’s request for a more nuanced interpretation of the VIP scores. In PLSR, the VIP represents a weighted sum of squares of the PLS loadings, where the average of the squared VIP scores is equal to 1. Consequently, a threshold of VIP > 1.0 is widely accepted as the primary criterion to identify "highly influential" predictors. For values ≤ 1, we have followed the hierarchical classification established in chemometric literature: predictors with 0.8 < VIP ≤ 1.0 are considered to have "moderate influence," providing secondary explanatory power, while variables with VIP ≤ 0.8 are interpreted as "low-influence" or redundant. We have updated Section 2.6.2 to explicitly define these interpretation tiers and their role in identifying the dominant determinants of yield.

 

The text in line 502-504 has been updated as follows:  “Following established chemometric protocols, variables with VIP scores between 0.8 and 1.0 were considered to have moderate influence, while those with VIP < 0.8 were categorized as low-contribution predictors.”

 

Comment 46; 303. List the formulas for the statistical metrics: R², Q², the REQM, and Variance Inflation Factor.

Response46: We appreciate the reviewer’s request for mathematical transparency. We have updated the Statistical Analysis section (Section 2.6.2) to include the formal definitions and formulas for the coefficient of determination (R²), the predictive squared correlation (Q²), the Root Mean Square Error (RMSE), and the Variance Inflation Factor (VIF). These metrics provide a comprehensive evaluation of model fit, predictive stability, and multi-collinearity control.

The text in line 518-538 has been updated as follows:

These parameters were calculated based on the following formulations [44, 49]:

 

• The coefficient of determination, which quantifies the proportion of variance explained by the model:

 

•  

The Root Mean Square Error (RMSE), representing the average deviation between observed and predicted values:

• The predictive squared correlation (Q²), derived from leave-one-out cross-validation to assess model stability:

where PRESS is the Predicted Residual Error Sum of Squares and TSS is the Total Sum of Squares.

 

• Finally, multicollinearity in MLR models was monitored using the Variance Inflation Factor (VIF):

where  is the coefficient of determination of a regression of predictor j on all other predictors [44, 49].

 

Comment 47; 308... During this process, multicollinearity was strictly controlled by ensuring a Variance Inflation Factor (VIF) (?) of less than 10 [49]. What is the VIF factor and how to interpret it?

Response 47: We appreciate the reviewer's request for clarification on the collinearity diagnostics. We have updated the manuscript to define the Variance Inflation Factor (VIF) as a metric that quantifies how much the variance of an estimated regression coefficient is increased due to collinearity with other predictors in the model. Furthermore, we have included the interpretation criteria: a VIF = 1 indicates no correlation, values between 1 and 5 suggest moderate correlation, and a VIF > 10 serves as a critical threshold indicating high multicollinearity that could destabilize the model's coefficients and reduce predictive reliability

The text in line 539-547 has been updated as follows: “Furthermore, multiple linear regression (MLR) models with forward stepwise selection were fitted to derive parsimonious and interpretable explanatory models. During this process, multicollinearity was strictly controlled by ensuring a Variance Inflation Factor (VIF) of less than 10 [49]. The VIF was interpreted as a diagnostic tool to quantify the inflation of parameter variance due to linear dependencies among predictors; whereas a VIF = 1 denotes a complete absence of correlation, the threshold of 10 was strictly enforced to identify and exclude redundant variables that could otherwise compromise the numerical stability of the Beta coefficients and the overall inferential validity of the models [44, 49].”

 

Comment 48; 343. Figure 2. The colors of the phenological phases within the graphs are imperceptible. Please increase the shades and add the names of the phenological phases within the graph. In the title of Figure 2, indicate the monthly time period of each phenological phase. In the graphs, change accumulated precipitation to rainfall and average temperature to average air temperature.

Response 48: We appreciate the reviewer’s feedback on the figure's readability. Figure 2 has been updated to increase the contrast of the phenological background shading and to include the names of each phase directly within the panels. In the caption, we have now specified the monthly time periods for each phenological stage. Furthermore, axis labels have been standardized to "average air temperature" and "rainfall" to ensure technical precision and clarity as requested.

The text in line 585-596 has been updated as follows: “Figure 2. Spatio-temporal dynamics of spectral indices and climatic variables in avocado (Persea americana cv. Semil-34) across environmental gradients (2020–2025). Shaded regions denote the annual operative phenological phases: Vegetative (V, December–January), Flowering (F, February–March), Fruit Growth and Maturation (M, April–August), and Harvest (H, September–November). Average monthly time-series are shown for (a) NDVI, (b) NDRE, (c) EVI, and (d) NDMI (e) average air temperature and (f) rainfall. Data represent monthly smoothed means for each environment (A1–A5).”

 

 

Comment 49; 355... 3.2. Phase-Specific Spectral Responses: Index–Climate Associations... The associations between index and climate are practically nonexistent or very weak. How to explain this?

 

Response: We appreciate the reviewer’s critical assessment of the correlation magnitudes. While the Pearson coefficients (r) range from low to moderate (0.05 to 0.31), they are highly significant (p < 0.001) due to the large sample size (n > 2,750 per phase). The relatively low magnitude is expected in tropical hillside agroecosystems for three reasons:

1. Physiological Buffering: Avocado is a perennial tree with significant starch reserves (Non-Structural Carbohydrates), which decouple immediate spectral responses from short-term climatic fluctuations compared to annual crops.
2. Environmental Stability: The tropical climate of the Dominican Republic exhibits low seasonal thermal amplitude, reducing the "signal-to-noise" ratio of climatic forcing on canopy reflectance.
3. Topographic Masking: Local microclimates created by slope and aspect can partially mask the regional climate signals provided by ERA5-Land.

We have updated the Discussion (Section 4) to explicitly address this "weak but significant" association as a reflection of the system's complexity.

The text in line 891-869 has been updated as follows: “The relationships between spectral indices and climatic variables underwent systematic reorganization throughout the phenological cycle, reflecting fundamental shifts in the crop's physiological priorities (Figure 3). While these associations were statistically significant, their low-to-moderate magnitudes reflect the high physiological buffering of perennial avocado trees, where internal carbon reserves partially decouple canopy reflectance from immediate hydroclimatic forcing [1, 51].”

 

Comment 50; 430... Table 3. What is EE in Table 3?

Response50: We apologize for the lack of clarity in the table header. EE refers to the Standard Error (SE) of the mean. To adhere to international scientific nomenclature, we have updated the column header from "EE" to "SE" in Tables 3 and 4.

 

Comment 51; 437.., Cite Figure 5 in the text.

Response 51: We appreciate the reviewer's correction. An explicit citation to Figure 5 has been integrated into Section 3.3.1 to support the presentation of the environmental effect on yield and fruit quality parameters during the 2025 cycle.

 

Comment 52; 574. What is (a) at the beginning of Figure 9? Where is (b)?

Response 52: We appreciate the reviewer's observation. Figure 9 is a standalone, single-panel plot. The panel identifier (a) was a clerical remnant from a previous version of the manuscript and has been removed from both the figure and the caption to ensure consistency and clarity.                                                                        

Comment 53; 580... the model achieved an R² (?) of 0.346 and a Q² of 0.173, suggesting acceptable... The R² remains very low, indicating that the explanatory power of one variable in relation to another is very low, or not?

Response 53: We acknowledge the reviewer’s point regarding the magnitude of the R² values. While R² 0.35 might be considered low in controlled laboratory settings, in agricultural field studies—particularly those involving remote sensing and complex tropical mountain topography—R²  values between 0.3 and 0.5 are considered moderate and biologically significant. This reflects the inherent "stochastic noise" of open-field systems, where uncontrolled variables (e.g., individual tree genetics, micro-management, and soil microbiome) contribute to the unexplained variance. Crucially, the Q² > 0 (0.173) confirms that the model possesses genuine predictive power and is not a result of overfitting, which is the gold standard for PLSR validation in perennial crops

The text in line 975-984 has been updated as follows: “The PLSR models demonstrated moderate explanatory power, with R² values ranging from 0.35 to 0.38 for yield components (Table 5). Although these values indicate that a significant portion of the variance remains unexplained, such magnitudes are typical for multi-source studies in heterogeneous mountain agroecosystems, where uncontrolled environmental and biological factors—such as intra-tree genetic variability and localized soil conditions—contribute to the total variance. Most importantly, the positive Q² values (0.17–0.22) confirm the model's predictive stability, demonstrating that the identified relationship between topography, spectral indices, and yield is not an artifact of the sample size but a stable physiological trend in cv. Semil-34.”

 

Comment 54; 641-659... defined thresholds (Lahack et al., 2025 ?). Excessive vegetative vigor during flow...
..both fruit caliber and commercial homogeneity (Cano-Gallego et al., 2023?).

Response 54: We appreciate the reviewer's attention to bibliographic detail. We have verified both citations: (i) Lahack et al. (2025) is a recent study (Reference [2] in our list) that establishes physiological thresholds for avocado productivity under Mediterranean conditions, which we have extrapolated to our tropical mountain context. (ii) Regarding the second point, we have replaced the citation of Cano-Gallego with Salameh et al. (2022) [17] and Mwelase et al. (2022) [15], which are already indexed in our references and specifically address fruit caliber and commercial homogeneity in avocado environmental gradients.

The text in line 933-939 has been updated as follows:  “An increased fruit load intensified the competition for assimilates, thereby restricting individual fruit expansion, while the altitudinal environment modulated this process by conditioning temperature and water availability during early developmental stages within genetically defined thresholds [12]. Excessive vegetative vigor during flowering acted as a competitive sink that reduced fruit set, whereas its moderation favored both fruit caliber and commercial homogeneity [2, 34].”

 

We sincerely thank the reviewer for the valuable suggestions, which significantly improved the methodological clarity and scientific rigor of the manuscript.

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The draft has significantly improved. Thank you for considering our suggestions. Only a few minor edits in the discussion and conclusion. The conclusion is still a bit wordier. See attached file for suggestions   

Comments for author File: Comments.pdf

Author Response

We sincerely thank for the positive evaluation of the revised manuscript and for the constructive suggestions that have significantly enhanced the clarity and flow of the paper.

 

We have addressed the remaining minor points as follows:

 

Optimization of the Conclusion and Discussion Sections

Reviewer Comment: The conclusion is still a bit wordier. Suggest moving the first paragraph of the conclusion to the discussion.

Response: We have followed this excellent suggestion. The conceptual synthesis regarding the "vigor paradox" and the source-sink equilibrium has been moved to the Discussion section (Section 4). This adjustment has allowed us to provide a deeper interpretive context before the final closing remarks. Consequently, the Conclusions (Section 5) have been significantly streamlined, focusing now on the core findings, practical applications, and future research directions. This change has improved the overall conciseness and impact of the final section.

Line (1045-1052)

“In synthesis, the evidence suggests that avocado productivity in tropical hillsides is not a linear function of canopy vigor. High vegetation indices, rather than indicating productivity, reflect a source–sink imbalance that penalizes fruit set and development. This 'vigor paradox' constitutes the primary conceptual contribution of this work, shifting the analytical focus from maximum greenness toward the identification of optimal functional ranges. Within this framework, topography acts as the structural regulator where landscape-scale water redistribution delineates the environmental thresholds that govern this physiological equilibrium.”

 

Line (1088-1093)

“A central finding of this study is the identification of a 'vigor paradox', where excessive canopy growth during critical phenological stages acts as a competitive sink that penalizes yield. This discovery redefines the interpretation of spectral signals in avocado, demonstrating that productivity is not a linear function of greenness but is governed by an optimal functional balance between vegetative and reproductive structures.”

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have improved transparency and acknowledge the study’s shortcomings, but in terms of methodological robustness, their response remains incomplete and does not address the study’s primary risk. The authors made several useful adjustments. They now clarify that the tree is the experimental unit, that the fruits are subsamples, that there is a minimum spacing between trees, that the PLSR was fitted using tree-level observations, and that in the discussion they explicitly acknowledge the small sample size and the potential risk of spatial autocorrelation. This improves the manuscript’s transparency and alleviates some of the original confusion. However, acknowledging a limitation is not the same as resolving it.

In their discussion, they note that the small sample size may increase Type II error and sensitivity to outliers, and that the 12-meter proximity “may introduce a risk of spatial autocorrelation.” However, they do not perform any corrective or sensitivity analyses to quantify how much this affects their inferences. There is no spatial diagnosis, no residual dependence analysis, no comparison between tree level results and average results per environment, nor an explicit reformulation of the inferential scope. In essence, the limitation is acknowledged, but the central statistical inference remains almost intact. On the other hand, their defense of independence is more assertive than demonstrative. In the response, they argue that 12 m between trees “ensures independence and minimizes spatial autocorrelation,” and in the manuscript they even state that this buffer “ensures statistical independence.” That claim is too strong for the presented design.

Spacing trees 12 m apart may help prevent spectral overlap of Sentinel-2 pixels, but it does not demonstrate ecological statistical independence between trees within the same environment. Shared topography, management practices, soil moisture, and local spatial context remain perfectly compatible with autocorrelation. In other words, they addressed the spectral mixing problem more effectively than the problem of statistical independence.

The authors maintain an analytical framework in which univariate tests continue to carry significant interpretive weight, despite suggestions for a more robust multivariate synthesis. In their response, they justify retaining PCA + PLSR and rejecting DFA because they seek continuous gradients and because PLSR already provides the necessary supervised power. However, in the revised manuscript, they still retain bivariate correlations and univariate comparisons as an important part of the inferential analyses, even after acknowledging finite sample size, multicollinearity, and spatial risk; yet this does not address the comment that the reader would end up interpreting various figures and blocks of results separately when there could be a more integrated multivariate synthesis between environmental variables and fruit variables. The problem is not using PCA or PLSR; the problem is that they do not truly replace the weight of univariate tests but rather coexist with them.

The current analytical approach relies heavily on multiple univariate comparisons and pairwise correlations to describe differences among environments (A1–A5). While these analyses provide useful descriptive insights, they fragment the inferential framework and increase the risk of both Type I and Type II errors, particularly given the limited sample size. Discriminant Function Analysis would allow the authors to compare all variables simultaneously (thereby reducing the problem of multiple comparisons), maximize the variance between groups and reduce the variation within groups (unlike PCA, which ignores which group the observations belong to); it would identify the variables related to these differences (factor structure), the variability within each group (the percentage of trees that resemble the group more than they do another group), the capacity to discriminate between groups (Wilks’ Lambda), identify the environmental gradient, and propose an equation that would allow them to predict the group to which the remaining trees in the parcel belong based on the environmental variables that authors analyzed. PCA is not designed to compare, only to ordinate observation in function of “n” environmental variables, and PLSR does not assess discrimination between groups.

The authors argue that averaging by environment would result in the loss of microtopographic variation and reduce statistical power. This may be a reasonable approach for exploratory purposes, or when the scale under analysis is a microhabitat level, but the study is conducted at landscape level. In fact, with 5 trees per plot and a strong internal spatial structure, a complementary comparison at the plot level or a sensitivity analysis would have been precisely the most convincing way to show that their conclusions do not depend on treating all trees as fully independent observations.

This manuscript has the potential to become a valuable reference for applied studies in agroecological systems. In this sense, reinforcing the use of integrative statistical approaches would not only improve the internal consistency of the results but also contribute to promoting good methodological practices among students and researchers who may use this work as a guide

The authors have improved clarity, but they have not yet analytically addressed the two central risks of the study: potential pseudoreplication/spatial autocorrelation and inferential fragility stemming from the small sample size. The revision improves the manuscript but does not fully resolve the major methodological objection. The limitation is discussed but is not sufficiently supported analytically. Therefore, I request that:

• the conclusions be moderated and explicitly presented as exploratory
• the inferential weight of the univariate tests be reduced,
• at least one complementary analysis evaluating robustness with respect to spatial structure or the average per environment would be desirable.

Author Response

We would like to express our sincere gratitude for the rigorous and constructive methodological critique. We agree that acknowledging a limitation is not equivalent to resolving it; therefore, following your suggestions, we have implemented a series of diagnostic and corrective analyses to analytically validate the robustness of our inferential framework.

 

 

The specific adjustments are detailed below:

1. Spatial Diagnosis and Statistical Independence

Reviewer Comment: The claim that 12 m ensures independence is too strong; shared topography and management remain compatible with autocorrelation.

Response: We have performed a formal spatial diagnosis to resolve this concern. A Moran’s I index was calculated on the PLSR model residuals to detect potential spatial dependencies. The results (I = -0.1390, p = 0.2450). Instead of claiming that the distance 'ensures' decoupling, we now state that our spatial diagnosis (Moran's I) confirmed the absence of significant autocorrelation in the residuals, providing empirical support for the independence of the observations at this analytical scale..

Line (194-195);

“The effectiveness of this spacing was formally evaluated through a spatial autocorrelation diagnosis.”

Line (465-470);

“Furthermore, the spatial independence of the observations was formally assessed by calculating Moran’s I on the residuals of the primary yield models. This analysis was conducted to determine whether the 12-m minimum inter-tree spacing was sufficient to mitigate potential spatial dependencies associated with shared microtopographic conditions or management practices. The results were consistent with the assumption of statistical independence at the analyzed scale”

 

Line (483-489)

“While PCA was primarily used to explore continuous biophysical gradients and identify non-linear trade-offs, a Discriminant Function Analysis (DFA) was complementarily implemented as a supervised validation step. The DFA was employed to to evaluate the statistical distinctness of the environmental strata (A1–A5) and to identify the variables that maximize inter-group variance, thereby reducing the risk of Type I errors associated with multiple univariate comparisons.”

Line (505-511)

“To analytically address the risk of pseudoreplication and ensure the assumption of sta-tistical independence, two diagnostic steps were performed: (i) spatial autocorrelation was assessed using Moran’s I index on the PLSR residuals to assess whether significant spatial dependency, providing operational support for the 12-m inter-tree spacing at the analyzed scale; and (ii) a sensitivity analysis was conducted by comparing the di-rectional consistency of PLSR coefficients at the tree-level (n=25) and environ-ment-level (n=5) scales..”

2. Integrated Multivariate Synthesis (DFA)

Reviewer Comment: The analytical framework relies on multiple univariate comparisons; DFA would allow for a more integrated multivariate synthesis.

Response: We have incorporated a Discriminant Function Analysis (DFA) as a supervised validation step. By focusing on the primary environmental drivers (VIP > 1.0), the DFA achieved a 92.0% classification accuracy (Wilks’ Lambda= 0.0216, p < 0.001). This integrative approach confirms that the environmental strata (A1–A5) represent statistically distinct functional units, significantly reducing the reliance on fragmented univariate tests.

Manuscript Changes: See Section 3.8 (Results) and Section 4 (Discussion).

Line (894-900)

“The Discriminant Function Analysis (DFA), integrated by the primary environ-mental drivers (VIP > 1.0) to ensure model parsimony, yielded a global Wilks’ Lambda of 0.0216 (p < 0.001). The classification accuracy under Leave-One-Out Cross-Validation (LOO-CV) was 92.0%, with 23 out of 25 experimental units correctly assigned to their respective environmental strata (A1–A5). Misclassifications were limited to exchanges between intermediate strata A2 and A4. Structural loadings for the first canonical function (LD1 loadings > 90) identified NDVI and NDMI as the dominant environmental discriminants.”

 

Line(964-970)

“The statistical distinctness of these strata was further validated by Discriminant Function Analysis (DFA). This integrative approach confirms that the defined environments represent consistent functional units, providing a robust multivariate synthesis that effectively aligns the topographic gradients with the observed biophysical responses.”

 

3. Inferential Fragility and Sensitivity Analysis

Reviewer Comment: A complementary comparison at the plot level or a sensitivity analysis is needed to show that conclusions do not depend on tree-level independence.

Response: We conducted a sensitivity analysis comparing the directional consistency of the PLSR coefficients at two scales: tree-level (n = 25) and environment-level means (n = 5). The results revealed a 77.8% directional consistency, with the primary drivers of the 'vigor paradox' (NDVI, NDRE, NDMI, and TWI) maintaining identical signs across both aggregation scales. This proves that the identified productive trends are stable landscape-level processes and not artifacts of the sampling density.

Line (901-907)

“Spatial autocorrelation analysis of the PLSR residuals yielded a Moran’s I value of -0.1390 (p = 0.2450), indicating no significant spatial autocorrelation and supporting the validity of the inferences at the analyzed scale. Additionally, the sensitivity analy-sis indicated a 77.8% directional consistency rate between the regression coefficients estimated at the individual tree level (n = 25) and those based on environment-level means (n = 5). Specifically, NDVI, NDRE, NDMI, and TWI showed identical negative signs across both levels of aggregation.”

 

Line (1061-1068)

“Furthermore, the 12-m proximity sampling trees, although necessary to align the experimental units with the 10-m spectral footprint of Sentinel-2, was statistically evaluated to ensure it did not bias the inferences. Our Moran’s I diagnosis suggested that this spacing, combined with the topographic stratification, appears to have reduced the risk of spatial autocorrelation. Furthermore, a sensitivity analysis demonstrated a directional consistency between tree-level and environment-level PLSR coefficients. This consistency across scales suggests that the identified “vigor paradox” and its as-sociated drivers (NDVI, NDRE, NDMI, and TWI) may reflect landscape-level processes rather than being explained solely by sampling density.”

 

4. Reformulation of the Inferential Scope

Reviewer Comment: Conclusions should be explicitly presented as exploratory.

Response: We have adjusted the tone of the final conclusions and the discussion to present this work as a robust exploratory framework. We have explicitly acknowledged that these findings serve as a foundation for future larger-scale assessments in under-documented mountain systems.

Line (1073-1079)

“This implies that the observed patterns should be interpreted as a robust exploratory framework rather than a final generalization. By integrating different time scales, we reduced the influence of seasonal bias and found consistent evidence of the 'vigor paradox' in cv. Semil-34. This study should be interpreted as a robust exploratory analysis of cv. Semil-34 in under-documented mountain systems, providing a founda-tion for future larger-scale spatial assessments utilizing higher-resolution sensors or object-oriented analysis to further mitigate pixel mixing.”

 

Line (1106)

“Collectively, this research provides a preliminary conceptual and methodological foundation for advancing toward precision management of avocado on tropical slopes, facilitating the transition from this exploratory framework toward multiscale predictive models that integrate topography as a structural determinant, with high potential for transferability to other perennial crops grown in complex agricultural landscapes.”

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The article titled 'Spectral phenology, climate, and topography as determinants of vigor, yield, and fruit quality in avocado (Persea americana cv. Semil-34) on tropical hillsides' is in full compliance with the scope of the journal Horticultura. The suggested revisions have been properly incorporated into the manuscript; however, the location of the Dominican Republic relative to its continent is still not clear in the presented map (Figure 1). Despite this specific point, I recommend the article for publication.

Author Response

We sincerely thank for the positive recommendation for publication and for the constructive observation regarding the geographic contextualization of the study area.

 

Geographic Context of the Dominican Republic

Reviewer Comment: "...the location of the Dominican Republic relative to its continent is still not clear in the presented map (Figure 1)."

Response: We agree that providing a clear macro-regional context is essential for international readers. In the revised manuscript, Figure 1 has been updated (see Panel a). We have incorporated a global/continental inset that explicitly shows the position of the Dominican Republic and the Caribbean region within the American continent.

Manuscript Changes: Figure 1 has been replaced with an optimized version that includes the continental reference frame to ensure immediate geographic orientation.

Figure 1. Geographic location and experimental layout of the study site in Cambita, San Cristóbal, Dominican Republic. (a) Global and continental context showing the location of the Dominican Republic within the Caribbean and the American continent. (b) National context indicating the location of the study region. (c) Regional topographic context showing the municipality of Cambita Garabitos; the map integrates a Digital Elevation Model (SRTM, 30 m resolution) visualized with a Viridis color ramp and a multidirectional hillshade to highlight the rugged terrain of the tropical hillsides. (d) High-resolution natural color (RGB) orthomosaic of the avocado (Persea americana cv. Semil-34) orchard obtained via an Unmanned Aerial Vehicle (UAV). This image provides sub-decimetric resolution, far exceeding Sentinel-2 capabilities, allowing for the precise identification of individual tree crowns and orchard architecture. Red dots indicate the georeferenced sampling points (Trees), and yellow polygons delimit the experimental plots (Environment: A1, A2, A3, A4, A5). Geographic coordinates are expressed in decimal degrees (WGS 84), and scale bars are provided in both maps to identify the distances between sampled units.

 

 

Author Response File: Author Response.pdf