Early Prediction of Coffee Yield in the Central Highlands of Vietnam Using a Statistical Approach and Satellite Remote Sensing Vegetation Biophysical Variables

Round 1

Reviewer 1 Report

The introduction section could be improved presenting crucial agronomic and phenology events that are important for the coffee development and crop yield output.

Line 174.  It is not mentioned if the coffee planted in Dak Lak province is under natural shadow or if it is planted under the sun.  Both cases will have different interpretation of the remote sensing indices used in the study. The author in case of natural shadow should explain how the vegetation indices have been impacted and how the solved the noise produced by natural shadow.

Figure 1. The authors should explain why was not possible to obtain a pure coffee crop mask?

Line 222.  The authors should better explain the impact of water during the dry months (Feb-May).  Have all farmers access to irrigation?  In other part of the world, coffee plantation needs an isolated rainfall (more than 50 mm) to provoke. the blossom.  This event it is very important because it will determine the organoleptic quality of the coffee.

Line 223.  The selection of the second period is not well justified by the authors.  The period overlap dry months with the start of the rainy period (May-June).  From June to October, there is many phenological events that can explain the crop yield variation of coffee such grain abortion due to water stress or the grow period of the grain due to water availability.   I suggest reviewing the justification of the second period selection. 

Line 321.  The authors should explain if they removed the trend before doing the regressions. Or how the CST software will deal with the trend of the depend variable and the impact on the interpretation of the results. What are the assumptions of the trend?    

Table 3.  I recommend doing a test of autocorrelation between the independent variables.  NDVI, LAI and FAPAR has a degree of autocorrelation.

Line 446.  My previous remark regarding autocorrelation between those vegetation indices could explain why there is not previous research on combining these three variables in a model. Normally the research try to probe the value added when one of these variables are used instead other, for instances FAPAR instead NDVI.

Line 449.  The spatial resolution of the data should not increase the results of the regression because the crop yield data is aggregated at the administrative unit.  A better crop mask of coffee could made the difference. 

 

Author Response

Many thanks to your comments enabled us to significantly enhance the quality of this paper. We bring answers to your comments below and please see the attachment.

Point 1: Line 174.  It is not mentioned if the coffee planted in Dak Lak province is under natural shadow or if it is planted under the sun.  Both cases will have different interpretation of the remote sensing indices used in the study. The author in case of natural shadow should explain how the vegetation indices have been impacted and how the solved the noise produced by natural shadow.

Thank you for your comments.

The coffee planted in Dak Lak province is in the vast majority planted under the sun in coffee tree plantations. There is no natural shadow due to other higher trees for example. In order to clarify this point, the following sentence was added at the end of the “study area” section at lines 182-183: “The vast majority of coffee trees are part of coffee tree plantations where coffee trees are the main vegetation story”.

Point 2: Figure 1. The authors should explain why was not possible to obtain a pure coffee crop mask?

It was not possible to obtain a pure coffee crop mask because such data is not available and it was not possible for the authors to produce such data.

We explain in section “Processing of satellite images in SPIRITS software” in part 2.2.1  what the mask is corresponding to :“Third, zonal statistics were extracted for these phenological variable images for the perennial agricultural vegetation zone of the Dak Lak province thanks to an extraction mask coming from the official 2015 land use map of Dak Lak province collected from the Department of Agriculture and Rural Development of Dak Lak province (Figure 1).”

We have now added after this section, at lines 258-260, a sentence saying: “No pure coffee crop mask was available and it was not possible for the authors to produce such mask in the framework if this study.”

Point 3: Line 222.  The authors should better explain the impact of water during the dry months (Feb-May).  Have all farmers acscess to irrigation?  In other part of the world, coffee plantation needs an isolated rainfall (more than 50 mm) to provoke. the blossom.  This event it is very important because it will determine the organoleptic quality of the coffee.

We have added the following information in the paper about the irrigation practices in Dak Lak province, in relation with the rainfall, at line 183-186:

“Irrigation was applied one to four times per year from 2008 across robusta coffee in Dak Lak province (on average 1345 litre/tree/year, i.e. 148 mm/year). The stated irrigation quantities varied based on rainfall patterns during the coffee growing season. [40].”

Irrigation is a common practice in the area but we don’t know if absolutely all farmers have access to irrigation. We don’t have this information.

And regarding the rainfall, we don’t have information about isolated rainfall. We don’t have such information and so we cannot take it into account. But in this study we don’t take into consideration the organoleptic properties of the coffee, we only focus on coffee yield (quantity and not quality).

Point 4: Line 223.  The selection of the second period is not well justified by the authors.  The period overlaps dry months with the start of the rainy period (May-June).  From June to October, there is many phenological events that can explain the crop yield variation of coffee such grain abortion due to water stress or the grow period of the grain due to water availability.   I suggest reviewing the justification of the second period selection. 

The justification of the choice of the second period (decades 1 to 18; January to June) is justified in the paper as follows, at lines 231-240:

“The second period corresponds to 18 decades, from January to June (decades 1 to 18). This period was considered because a longer period may be more representative of the global coffee development conditions and consequently result in variables that have a higher explanatory power. Additionally, using the first six months of the year to predict coffee yield will give planners time to consider or find solutions before the end of the coffee season.”

We have now reformulated a little bit this last sentence, at lines 235-240, in order to reinforce this justification, as follows:

“Additionally, the objective of the methodology developed in this research being to produce models that enable to forecast coffee yield well in advance compared to the harvest period of October to December, it was decided to make the coffee yield forecast at the end of June at the latest. Indeed, using the first six months of the year to predict coffee yield will give planners sufficient time to consider or find solutions before the end of the coffee season.”

Note that the justification related to the potentially better representativeness of this longer period (1-18 versus 5-15) was revealed to be particularly relevant since better models were observed for this longer period compared to the shorter period of decades 5 to 15, as discussed in the discussion at lines 492-499.

Point 5: Line 321.  The authors should explain if they removed the trend before doing the regressions. Or how the CST software will deal with the trend of the dependent variable and the impact on the interpretation of the results. What are the assumptions of the trend?    

There is indeed a linear time trend for yield as presented in the figure 3, and this trend was taken into account by CST software during the regression process. CST takes this time trend into account simply by adding a term in the model that corresponds to the time trend as it can be seen in table 3. The variable corresponding to “Time trend linear” in table 3 corresponds to the years minus an offset that is set in CST by convention to 1965 (i.e. “year-1965”). This offset is used in CST in order to increase numerical precision.

We added this explanation at lines 293-297 in the paper as follow:

“CST takes the potential time trend into account by adding a term in the model that corresponds to that time trend, if applicable. To increase numerical precision, the regression coefficient for the linear time trend is for “year – offset” rather than “year” itself. The offset is fixed at 1965 by default in CST. Likewise, the regression coefficient for the quadratic time trend is for (year - offset)² [28].”

Point 6: Table 3.  I recommend doing a test of autocorrelation between the independent variables.  NDVI, LAI and FAPAR has a degree of autocorrelation.

For sure, there is a certain level of autocorrelation between the independent variables, and in particular between the same phenological variables (e.g: rsd) derived from different biophysical variable (NDVI, LAI, FAPAR) (eg: rsd LAI and rsd NDVI with a R of 0.62 in the model 8 considering decades from 5 to 15).

In the submitted paper we already mention the general auto-correlation between the independent variables selected in a given model, at lines 386-389 as follow :“A relatively high negative or positive correlations was observed between some variables selected in some of the best models (R varies in the range from -0.863 (for Adn_LAI and Dmn_LAI in model 5) to 0.795 (for Vmn_LAI and Dmx_LAI in model 1)).” (this sentence is a slight reformulation and edition of the submitted version) but these results were not shown.

We have now added the presentation of a systematic test of auto-correlation between all independent variables that are kept in each of the 8 best models, as presented in the figure 5, at line 417 in the file attachment.

Besides, in order to better answer to the reviewer’s suggestion, we have now also presented an additional figure showing the correlation of the 11 phenological variables between the three biophysical satellite products LAI, NDVI and FAPAR, as presented in the new figure 6 in the file attachment. In this way we can now clearly see the level of correlation between a given phenological variable for the 3 biophysical variables.

We have added a short analysis of this figure in the text at lines 389- 402, as follow:

“When considering the period of decade 1 (start of January) to decade 18 (end of June) of the years 2000 to 2019, the analysis of the Pearson correlation coefficient of the 11 phenological variables between the three biophysical satellite products LAI, NDVI and FAPAR (figure 6) shows a highly variable level of correlation between these phenological variables, from, in absolute value, 0.00 to 0.97, i.e. from no correlation to a very high level of correlation. For this period, the phenological variables derived from FAPAR and NDVI are the most correlated (average absolute correlation of 0.59, 3^rd column of figure 6) while those derived from LAI are less correlated to NDVI and FAPAR variables, especially for FAPAR (average absolute correlation of 0.30, 1^st column of figure 6). The low correlation values observed for at least some phenological variables in each pair of biophysical products (LAI and FAPAR, LAI and NDVI, FAPAR and NDVI) suggests that these three products may bring some non-redundant (uncorrelated) information, and thus be complementary at some point, and consequently that it is relevant to consider the three of them in the search for the best coffee yield prediction models.”

Point 7: Line 446.  My previous remark regarding autocorrelation between those vegetation indices could explain why there is not previous research on combining these three variables in a model. Normally the research try to probe the value added when one of these variables are used instead other, for instances FAPAR instead NDVI.

We agree with the reviewer’s comments regarding the fact that it is interesting to analyze the added value of a biophysical variable compared to the other (LAI, NDVI, FAPAR). And that’s what we did in this paper by analyzing what biophysical variables is the most present in the selected models, as commented for example at line 504-506 in the discussion “Furthermore, the variables derived from the LAI product were shown as more efficient for coffee yield forecast model than those derived from NDVI and FAPAR, though some complementarity was observed between these products for some models”. 

We now also added a comment going in the same direction in the result section in order to emphasize this observation, at line 382-386: “When considering the 8 best models, LAI derived variables occur 18 times, while NDVI derived variables 6 times and FAPAR derived variables 4 times only. This observation suggests that LAI derived variables are more efficient than NDVI and FAPAR ones for coffee yield forecasting.”

This said, the auto-correlation between LAI, FAPAR and NDVI derived variables seems not to be a reason to not consider these 3 biophysical variables as candidate variables when searching for the best possible models. Even if a high auto-correlation may occur between these variables, it is worth to test them all simultaneously in the multiple linear regression model in order to assess and find what variables will be selected and how these 3 biophysical variables may be complementary at some points.

We see the fact of considering these 3 biophysical variables simultaneously for models building as a real originality/novelty and interest of this research as it was never done before.

Point 8: Line 449.  The spatial resolution of the data should not increase the results of the regression because the crop yield data is aggregated at the administrative unit.  A better crop mask of coffee could made the difference. 

We mention in the paper that ‘Therefore, future studies should consider using the more recent and similar products derived at 300m spatial resolution.’

We agreed with your comments about the fact that indeed the crop yield data is aggregated at the administrative unit.

However, regarding the resolution of the data, both the resolution (or accuracy) of the crop mask and of the satellite images may have a positive impact on the models accuracy. Indeed, a better crop mask will simply better delimitate the region of interest, so the advantage of such enhancement is obvious. And regarding the satellite images, the higher the resolution, the smaller are the image pixels, and so the pixels may represent with more accuracy which regions are of interest and which ones are not. For a given crop mask, the smaller is the image pixels, the more these pixels will fit and be representative of the area delimitated by the mask, and inversely.

Reviewer 2 Report

This study uses multiple linear regression to predict coffee yield from satellite-derived bio-physical variables. In general the paper is well-structured and well-written. The methodology and language are sound. It impressed me that the authors have a good understanding of the coffee industry and agricultural practices in the study area. However, there are a few places where improvements could be made as listed below:

L170: “below 300m it is hot all year round” – doesn’t the area have an elevation range of 400-800m as stated in L168?

L202: Is the 1km resolution potentially too coarse if multiple types of crops are covered by a pixel? What is the implication of the 1km resolution to the modelling accuracy? I would suggest including this into discussion.

L203-215: These indices are familiar to most remote sensing researchers. I would encourage the authors to rephrase these definitions in a way that explains how these indices are potentially helpful in coffee yield prediction, rather than directly quoting them from the Copernicus website.

L250: What is the spatial resolution of the yield statistics? Are they reported at provincial or districtual level? How were they aligned with the gridded remote sensing products in this work?

L259: This is a linear model. I would imagine that some biophysical variables would not strictly follow a linear correlation with yield. This could add as a discussion.

L295: “* Note: in Eq.4” - This note is for Eq. 3 if I understand correctly.

L395: This figure is potentially misleading as years 2000 to 2019 have been used for model calibration/selection. Only year 2020 has been kept independent. R-squared values reported on training set may overstate the model performance.

Author Response

Many thanks to your comments enabled us to significantly enhance the quality of this paper. We bring answers to your comments in red below in this document and please see the attachment.

Point 1: L170: “below 300m it is hot all year round” – doesn’t the area have an elevation range of 400-800m as stated in L168?

Thank you for your comments.

We made a mistake of inaccuracy in the first statement at line 169. We have now corrected it at line 170 as follow:

“…with an average elevation range of 400–800m”, meaning that some areas are below 400 m and some over 800 m.

We have also slightly reformulated the text following that sentence, at line 171-172, as follow:

“below 300m it is hot all year round,  in the range of 400–800 m it is hot and humid, and it is cool over 800m[36].”

Point 2: L202: Is the 1km resolution potentially too coarse if multiple types of crops are covered by a pixel? What is the implication of the 1km resolution to the modelling accuracy? I would suggest including this into discussion.

When considering 1km resolution image, it is unavoidable, in the real world, that the pixels considered for the analysis will in fact correspond to a mixture of various land covers: coffee trees of course, but also other vegetation, paths, roads, rural houses, water bodies, etc (at the rare exception where there might be zones that are corresponding to a unique land cover over very big surface of more than 1 km²). In addition, as presented in the paper at lines 252-260, in this study, the accurate delimitation of pure coffee zones was not available for the extraction of statistics from the satellites images and the area we considered for the extraction of statistics from the satellite images corresponds to the land cover class “perennial agricultural vegetation” that contains approximately 62.5 to 68.2% of coffee only, as explained in that same section. So in this research the statistics extracted from satellite images correspond to a mixture of land covers (containing a relatively high percentage of coffee), and this is now clearly presented in the paper. Regarding the resolution, accuracy and representativeness of the input data used in this study, both the one of the satellite images (higher spatial resolution) and the one of the coffee area delimitation (pure coffee area mapping) are sources of potential improvement of the method.

We already mention in the paper that the use of more accurate input data could be a source of enhancement of the method, for example in the conclusion at lines 553-555 as follow:

“…with enhanced input data (finer spatial resolution for satellite images and more accurate coffee maps)”

or in the discussion at lines 511-512 as follow:

“Therefore, future studies should consider using the more recent and similar products derived at 300m spatial resolution”

Point 3: L203-215: These indices are familiar to most remote sensing researchers. I would encourage the authors to rephrase these definitions in a way that explains how these indices are potentially helpful in coffee yield prediction, rather than directly quoting them from the Copernicus website.

At this place of the text, in the method section, where we describe the data we work with, we think that it is relevant to define sufficiently accurately the three indices that are at the base of this work, even if, it is true, we agree with you, these three indices are very familiar to remote sensing researchers. That’s also why their description is quite short. On the other side, this description will be helpful for people that might not be remote sensing experts and that might read this open source paper. We think that citing the official data provider was also a relevant way to proceed.

Regarding the fact of explaining how these indices are potentially helpful in coffee yield prediction, we presented some elements going in that direction in the introduction section where the literature review is made. We think it is the most appropriate place where such considerations should be done. These are mainly the following:

At lines 94-96: “Kouadio et al. (2021) also successfully tested a process-based model using satellite remote sensing data (LAI) and model-based gridded climate data for predicting robusta coffee yield in the Central Highlands of Viet Nam [3]”.

At lines 129-132: “Fall et al. (2021) also used the CST to predict millet yield at a regional scale in Senegal with input data containing weather data combined with variables derived from remote sensing indicators (NDVI) [23].”

And especially at lines 141-152: “With the development of satellite imagery, agricultural monitoring systems have been using agro-meteorological indices coming from the spectral reflectance of the vegetation to provide timely and concise information about seasonal vegetative growing [31]. Remote sensing derived vegetation indices (e.g., the Normalized difference vegetation index, NDVI) and biophysical variables (e.g., the Fraction of Absorbed Photosynthetically Active Radiation, FAPAR; the Leaf Area Index, LAI) can be used to predict crop yield, either directly or indirectly [32,33]. In addition, remote sensing vegetation variables enable estimating crop growth variability to quantify the crops’ relative development and health conditions [34]. Such vegetation indices and biophysical variables are the most common satellite products utilized for these purposes [31]. At the national and regional levels, satellite systems can contribute effectively to early warning of crop stress during the growing period and in forecasting harvest yields [31,35].”

Point 4: L250: What is the spatial resolution of the yield statistics? Are they reported at provincial or districtual level? How were they aligned with the gridded remote sensing products in this work?

The official coffee statistics consist of coffee area and coffee dry grain production, available at the district and provincial levels. There are 15 districts in the Dak Lak province.

In this work, only the provincial coffee yield was considered for building models. This provincial coffee yield (ton/ha) was computed by dividing the official provincial coffee production (tons) by the official provincial coffee area (ha).

In order to clarify this point in the paper, we have slightly modified the section “2.2.2. Official coffee yield datasets.” at lines 267-271 by detailing a bit more the origin of the yield data, as follow:

“The coffee yields considered in this study are  provincial coffee yields and were computed by dividing the official provincial coffee production by the official provincial coffee area coming from the Dak Lak Statistical Yearbook 2009, 2014, 2018, and 2020 [38,44–46]. The period from 2000 to 2020 was considered. These coffee yields correspond to coffee dry grain yield.”

The provincial coffee yield was related to the remote sensing variables extracted on an area corresponding to “the perennial agricultural vegetation zone of the Dak Lak province thanks to an extraction mask coming from the official 2015 land use map of Dak Lak province” as explained in the paper at lines 252-254. So in this study both the yields and the remote sensing derived variables are considered at the provincial level only, never at the district level.

Point 5: L259: This is a linear model. I would imagine that some biophysical variables would not strictly follow a linear correlation with yield. This could add as a discussion.

Yes indeed, we are using here a multiple linear regression technique that is suited to identify and use the linear relationships between the predictors and the dependent variable to produce prediction models.

It may indeed be possible that some biophysical variables would not strictly follow a linear correlation with coffee yield. This was however not assessed. And in that potential case, other modelling approaches than the one used would be more suited or complementary.

As suggested by the reviewers, we have added the following sentences in the discussion at lines 518-525 in order to mention this interesting point:

“In this research, we used a multiple linear regression technique in order to produce coffee yield prediction models. Such technique is particularly suited to identify and use the linear relationships between the predictors and the dependent variable. However the linearity of the relationship between coffee yield and the phenological variables was not assessed in this study and it may be possible that some variables present a nonlinear relationship with yield. Consequently, further research might be interested in testing other nonlinear modelling approaches for predicting coffee yield from biophysical variables such those used in this study.”

Point 6: L295: “* Note: in Eq.4” - This note is for Eq. 3 if I understand correctly.

Yes indeed. The correction was done. Note that now, as the order of presentation of the equation has been changed a little bit (inversion of equation 2 and 3), this note now refers to equation 2.

Point 7: L395: This figure is potentially misleading as years 2000 to 2019 have been used for model calibration/selection. Only year 2020 has been kept independent. R-squared values reported on training set may overstate the model performance.

We paid a particular attention in the structure of the paper to separate in different sections the 2 following things:

• The results concerning the calibration of the models on the 2000-2019 period and the related models’ performances. This is presented in the section “3.1. Model performance” where models’ performances are presented in details in table 3.
• The results concerning the application of the selected models on the year 2020, in the section “3.2. Coffee yield predictions for 2020”.

The figure you mention refers specifically to and is included in this second section. The R-squared and p-values reported on this figure are those of the relation between observed versus predicted yield for the full period 2000-2020. These R-squared are slightly different than those of the model calibration (2000-2019) presented in table 3.

Following your comment and in order to avoid any potential confusion, we have complemented the legend of this figure with the following text: “R-squared and p-values reported on this figure are those of the relation between observed and predicted yield for the full period 2000-2020.”

Reviewer 3 Report

In Lines 216 to 227. The authors should explain more about the coffee. When it ripes, when it is harvested, and how long it takes to harvest it. The yields you are predicting it is as cherry or just the grain?

Lines 233 to 244. About zonal statistics. For me, it is unclear how many farms you analyzed?. What size were the farms? The zonal statistics tool was used on each farm?. The vegetation biophysical products were obtained for each farm? This part is not clear.

Lines 261-272. Please, explain how you screen the variables in Table 2. For me, it is not clear how you selected just three in each model.  

Author Response

Many thanks to your comments enabled us to significantly enhance the quality of this paper. We bring answers to your comments in red below in this document and please see the attachment.

Point 1: In Lines 216 to 227. The authors should explain more about the coffee. When it ripes, when it is harvested, and how long it takes to harvest it. The yields you are predicting it is as cherry or just the grain?

Thank you for your comments.

We added some information related to the coffee chronology at lines 222– 226 as follow:

“In Vietnam, the coffee phenology can be presented in five periods: (i) the flower-bud initiation and blooming season is from January to March; (ii) the fruit setting period is from April to May; (iii) the cherry development period is from May to August; (iv) the maturity stage is from September to October; (v) the ripening/harvest  period is organized from October to December [3].”

The considered yield is the dry grain yield.

This information was added in the text at lines 270-271 as follow: “These coffee yields correspond to coffee dry grain yield.”

Point 2: Lines 233 to 244. About zonal statistics. For me, it is unclear how many farms you analyzed?. What size were the farms? The zonal statistics tool was used on each farm?. The vegetation biophysical products were obtained for each farm? This part is not clear.

In this study we did not focused on specific farms. We worked at a much bigger scale: at the scale of the Dak Lak province and by considering only areas corresponding to “perennial agricultural vegetation zone” as specified in the paper at lines 252-255 as follow:

“Third, zonal statistics were extracted for these phenological variable images for the perennial agricultural vegetation zone of the Dak Lak province thanks to an extraction mask coming from the official 2015 land use map of Dak Lak province collected from the Department of Agriculture and Rural Development of Dak Lak province (Figure 1).”

We also now added the following sentence just after this section, in order to clarify this point:

“This land use map did not contain a class specific to coffee plants but only a broad class relative to agricultural perennial plants containing approximately 62.5 to 68.2% of coffee only, from 2015 to 2018 [38,44–46]. No pure coffee crop mask was available and it was not possible for the authors to produce such mask in the framework if this study.”

Point 3: Lines 261-272. Please, explain how you screen the variables in Table 2. For me, it is not clear how you selected just three in each model.  

The selection of the variables is done automatically by CST software through multiple linear regression technique. The best model, and consequently the variables composing this best model, is selected based on a parameter that can be chosen in CST among a list of parameters such as RMSEP, Adj R², etc. In our case, the RMSEP was always used for the automatic best model selection by CST.

A sentence has been added at lines 305-307 in order to better explain that, as follow: “The automatic selection and ordering of the best models by CST at each CST iteration for a given set of candidate variables was based on the root mean square error of prediction (RMSEp) (equation 2).”

For more clarity, the section concerned by this point has been a bit reorganized as you will see in revision mode in the document.

Complementary explanations about the way CST works for the best models selection are presented in the section “2.2.3. Crop yield forecasting model in the CST software” of the paper.

Note that some of the selected models in this study are composed of 4 variables, some of 3 variables. The maximum number of variables CST allows is 4.

Round 2

Reviewer 1 Report

Thank you for the revisions done based on my previous comments.  I would suggest to add more physiological explanations with the vegetation indices.  For instance you can review the classical results in coffee yield such Valencia 1973  regarding the relationship between LAI and crop yield. (Valencia, 1973. Relación entre el índice foliar y la productividad del cafeto. CENICAFE, 24, 79-89.)  Or the studies on physiology of coffee flowering and explain how the vegetation indices can explain (no only statistical point of view) the impact on the crop yield.  

 

Author Response

Many thanks to your comments enabled me to significantly enhance the quality of this paper. We bring answers to your comments below:

- We have now added information in the part introduction, on lines 153-167, a sentence saying: "Bernardes et al. (2012)[36] observed, in the Brazilian largest coffee-exporting province and from a dataset covering the 2002-2009 period, correlations between variations of the yield of coffee plots and variations of MODIS derived EVI and NDVI vegetation indices computed from pure coffee crop 250 m pixels overlapping the same coffee plots. The vegetation index metrics best correlated to yield were the amplitude and the minimum values over the growing season. The best correlations were obtained between the variation of yield and variation of vegetation indices of the previous year (R² = 0.55). In another study, Nogueira (2018) [37] evaluated the relationships between coffee productivity of some coffee plantations in Brazil and values of NDVI, SAVI and NDWI vegetation indices derived from LANDSAT-8-OLI sensor for different coffee phenological phases. They concluded that the best phenological phases of coffee to determine coffee productivity from spectral indices were the stages of dormancy and flowering. The results also indicated that the NDVI was the best index to estimate the productivity of coffee trees, with the coefficient of determination (R²) ranging from 0.58 to 0.90. "

 

Author Response File: Author Response.docx

Reviewer 2 Report

The authors have addressed my comments satisfactorily. I recommend this paper to be published.

Author Response

Dear revier,

I have already checked English by MDPI service, I send you the English-Editing-Certificate.

Many thanks to your comments enabled significantly enhanced the quality of this paper.

Best regards

Author Response File: Author Response.pdf