Round 1

Reviewer 1 Report

The submitted paper presents an interesting methodology for yield mapping in mango orchards. Good results have been achieved, and, although the methodology requires some fieldwork, I think it can be applied without problems in a high range of situations. My only methodological concern is that the authors always select the same amount of trees in all the orchards and, considering that there are small and large orchards, this fact introduces a high variability in the sampling intensity. If the authors can justify this election. Besides this concern, I have the following doubts/comments:  

L38: “incomes, food security”. There are only two items, I think that “incomes and food security” is better.

L105: “mapped” or map?

Figure 1: The canopy height model is also used in the GEOBIA classification according to table 1. Consequently, a line to the classification process should link CHM.

L150-152: Knowing the proportion of commercial and family-based orchards could be interesting for the readers.

L161: “In each orchard, ten trees were selected…” Why did the authors select the same amount of trees in orchards with a high range of areas? Selecting ten trees in an orchard with around 1.000 trees (orchard 15 according to fig 5a) is not the same that selecting 10 trees in a small family-based orchard.

Section 2.3.1: Which were the phenological stages of the trees during the flights? The presence of flowers in the mango trees would affect the classification algorithm performance.

L192: “leading to captured areas ranging from 4.6 to 20.5 ha”. I am sure that some of these areas required several flights for being mapped flying at 40 m. Please, comment this point in the manuscript.

L202: Which is the methodology for DTM creation used by Pix4D?

L223-224: In these lines, the authors say that they classified buildings using thresholds, but in L233-234 they say that the buildings are classified in “level 1 classification” with random forest. Clarify this point.

L245: The authors should include the bands used in the NDI calculation. The reference included beside the index (11) does not report the formula neither.

L269-270: This classification performance evaluation (using single points) does not guarantee that the tree limits are correct.

Table 2: Why are there no data for the training rate in orchard 6? Some data from this table should be included in results section (the accuracy columns).

Section 2.3.4: the extraction of the parameter proposed in this section can be done in eCognition. Why did the authors use another software?

L331: The author have selected a fixed amount of trees per orchard again. It supposes introducing a high variability in the sampling intensity.  

Table 4: The authors established three levels of reliability, but they have used only two (low and high).

Author Response

Response to Reviewer 1 Comments

General comments:

Point 1: The submitted paper presents an interesting methodology for yield mapping in mango orchards. Good results have been achieved, and, although the methodology requires some fieldwork, I think it can be applied without problems in a high range of situations. My only methodological concern is that the authors always select the same amount of trees in all the orchards and, considering that there are small and large orchards, this fact introduces a high variability in the sampling intensity. If the authors can justify this election.

Response 1: The reviewer 1 highlighted an unclear point in our work. The 10 sampled trees per orchard are not made for depicting the tree variability within the orchard itself, but rather to be part of a larger calibration sampling (150 trees) that represents the mango tree variability in the entire study area. Therefore, various t-test comparison of means have been added to our results to show the representativeness of the calibration trees regarding structural parameters (tree height, crown volume, and crown area). These trees were used to calibrate predictive models. However, a second sampling was performed by selecting 50 trees on a cross transect in each orchard. This sampling was used to weight the load index variable for each cultivar in the orchard. We assumed that 50 sampled trees were enough to capture the variability within small orchards (less than 2 ha, the majority in the study area). Moreover, in the study area, large orchards are conventional monospecific systems in where tree structure and production are homogeneous, resulting in a lower sampling intensity needed to capture the orchard variability. Moreover, for the largest studied orchards (#15 of 14 ha), we performed 2 cross transects to ensure an equal sampling intensity between orchards. Under these conditions, we believe that our sampling methods provide a balanced sampling intensity among orchards.

Modified L214-222: “In each orchard, ten trees were selected among the three main cultivars found in the study area –'Kent', 'Keitt' and 'Boucodiékhal' (BDH) to depict the entire variability of tree structure (trunk perimeter, height, and crown volume), age, and fruit load found in the study area. Thus, this set of 150 mango trees made of 89 'Kent', 39 'Keitt' and 22 'BDH' were used to calibrate the predictive models by measuring their production using the machine vision system describe below. Then, the representativeness of the calibration trees was tested by comparing their structure assessed by UAV with the structure of all mango trees within the 15 orchards. For this purpose, independent comparisons of means between the two groups were conducted using Student t-test at a 95% confidence level.”

Modified L507-510: “The transect passed through the two orchard diagonals, with 25 trees sampled on each, ensuring the spatial distribution of the sampling. As orchard 15 was large (13.8 ha), two transects were performed resulting in a total of 100 trees sampled.”

Modified L947-952: “We assumed that 50 sampled trees were enough to capture the variability within small orchards (less than 2 ha, the majority in the study area). Moreover, in the study area, large orchards are conventional monospecific systems in where tree structure and production are homogeneous, resulting in a lower sampling intensity needed to capture the orchard variability. Under these conditions, we believe that our sampling methods provide a balanced sampling intensity among orchards.”

Specific comments:

Point 2: L38: “incomes, food security”. There are only two items, I think that “incomes and food security” is better.

Response 2: As suggested, we modified the text.

Point 3: L105: “mapped” or map?

Response 3: “Map” is indeed the correct word, we modified the text.

Point 4: Figure 1: The canopy height model is also used in the GEOBIA classification according to table 1. Consequently, a line to the classification process should link CHM.

Response 4: The reviewer is true, we modified the figure (Box UAV process).

Point 5: L150-152: Knowing the proportion of commercial and family-based orchards could be interesting for the readers.

Response 5: We agreed and added the information in the main text. Modified L204: “from small family-based diversified orchards, including other tree species (citrus, cashew, etc.) (n = 9), to large commercial-based monospecific orchards (n = 6)”

Point 6: L161: “In each orchard, ten trees were selected…” Why did the authors select the same amount of trees in orchards with a high range of areas? Selecting ten trees in an orchard with around 1.000 trees (orchard 15 according to fig 5a) is not the same that selecting 10 trees in a small family-based orchard.

Response 6: See response and modified text to Point 1 of reviewer #1

Point 7: Section 2.3.1: Which were the phenological stages of the trees during the flights? The presence of flowers in the mango trees would affect the classification algorithm performance.

Response 7: We agreed and clarified period and mango phenologies during image acquisitions. And we added text in the paragraph 2.3.3 to specify that a classifier was train for each classification level in each orchard to address this issue.

Modified L292-293: “Due to the large period of UAV images acquisition, mango tree phenological stage varied between vegetative growth, flowering, and fruit set.”

Modified L363: “Because of variations in mango tree phenological stages between orchards and in environmental conditions during UAV flights, a RF classifier was train for each classification in each orchard.”

Point 8: L192: “leading to captured areas ranging from 4.6 to 20.5 ha”. I am sure that some of these areas required several flights for being mapped flying at 40 m. Please, comment this point in the manuscript.

Response 8: As suggested, we added that one orchard (the largest) was mapped with two successive flights at 40 m.

Modified L281: “Because of its large area (20.5 ha), the orchard #15 was mapped using two successive flights.”

Point 9: L202: Which is the methodology for DTM creation used by Pix4D?

Response 9: The DTM was automatically generated by the Pix4D software that filtered out the non-ground points in the point cloud before a smoothing step. We added this precision in the Materials and Methods section.

Modified L304-305: “DTM was automatically generated by the software that filtered out non-ground points from the point cloud before a smoothing operation.”

Point 10: L223-224: In these lines, the authors say that they classified buildings using thresholds, but in L233-234 they say that the buildings are classified in “level 1 classification” with random forest. Clarify this point.

Response 10: This is due to the first thresholding step that classify only the large and homogeneous objects (bare soils, buildings). However, the level 1 classification permitted to classify the small buildings and soils. As suggested by the reviewer, we gave some precision in the manuscript.

Modified L344-346: “This classification included also small heterogeneous “building” and “soil” objects that were not removed by the thresholding after the first segmentation.”

Point 11: L245: The authors should include the bands used in the NDI calculation. The reference included beside the index (11) does not report the formula neither.

Response 11: We added the information on the bands used. Additionally, we corrected a mistake in the references that gave the formula (Payne et al., 2014 instead of Payne et al., 2013).

Modified L357: “NDI, computed on R and G bands” and Reference [11]

Point 12: L269-270: This classification performance evaluation (using single points) does not guarantee that the tree limits are correct.

Response 12: We totally agree with this comment. In our work, we distinguished the classification of objects (which has been evaluated in our study), and the tree crown delineation. The latter was not evaluated as our method was not improving the state-of-the-art algorithms used in the current literature. We clarified this point in the discussion section.

Modified L734: “In addition, testing and evaluating new tree crown delineation methods should be complete.”

Point 13: Table 2: Why are there no data for the training rate in orchard 6? Some data from this table should be included in results section (the accuracy columns).

Response 13: Training rate in orchard 6 was missing due to typing error. We also removed the Overall accuracy and Mango accuracy to a new Table 3 in the results.

Point 14: Section 2.3.4: the extraction of the parameter proposed in this section can be done in eCognition. Why did the authors use another software?

Response 14: We agreed that we could have done that. However, for the good completion of our toolbox, we have to use a GIS software (crown delineation checking and potential manual tree reshaping); consequently, performing the structural calculations under the GIS was straightforward.

Point 15: L331: The author have selected a fixed amount of trees per orchard again. It supposes introducing a high variability in the sampling intensity.  

Response 15: See response and modified text to Point 1 of reviewer #1

Point 16: Table 4: The authors established three levels of reliability, but they have used only two (low and high).

Response 16: The two stars indicating the “medium” reliability level was missing line #3 in Table 5 due to a layout error. Modified in the Table 5.

Author Response File: Author Response.docx

Reviewer 2 Report

Dear author,

Your manuscript is really in good shape- it is concise and clear. The discoveries in manuscript are novel and will interest a lot of people in the relative fields for the yield estimation and mapping at orchard level using UAV. Definitely, the use RGB is a low-cost technology in UAV platform.

After working on a few minor comments (listed below), this manuscript is suitable to publish in the Remote Sensing journal.

1.       Abstract: study year is missing.

2.       Introduction: First paragraph needs to divide into 2-3 paragraphs. It’s a long paragraph right now!

3.       Materials and Methods: Schematic workflow- please write down full form of Av and Nb.

Good luck!

Author Response

Response to Reviewer 2 Comments

Your manuscript is really in good shape- it is concise and clear. The discoveries in manuscript are novel and will interest a lot of people in the relative fields for the yield estimation and mapping at orchard level using UAV. Definitely, the use RGB is a low-cost technology in UAV platform.

After working on a few minor comments (listed below), this manuscript is suitable to publish in the Remote Sensing journal.

Point 1:  Abstract: study year is missing.

Response 1: As suggested, we modified the abstract.
Modified L26: “In 2017,”

Point 2: Introduction: First paragraph needs to divide into 2-3 paragraphs. It’s a long paragraph right now!

Response 2: We are not sure to understand this comment as we already made 5 distinct paragraphs in the Introduction.

Point 3: Materials and Methods: Schematic workflow- please write down full form of Av and Nb.

Response 3: As suggested, we modified the Figure 1.

Author Response File: Author Response.docx

Reviewer 3 Report

General Comments

The paper is very well written and I have only few main comments on the content. There are numerous English language mistakes though throughout the entire manuscript which are distracting from the content. I listed most of them in the detailed comments section below.

The authors seem be unaware about the difference between land cover and land use. Land cover are the actual features on the ground (e.g. grass, trees, buildings), while land use is the usage class of the land. For example, an entire orchard would be a single land use class “orchard”, comprised of several land cover classes “trees”, “grass”, “bare ground”, etc. Please replace all occurrences of “land use” with “land cover”.

The use of the term “yield” should be clarified and defined. The standard definition of yield is production per unit of land area. It could also be defined as production per tree, as it is used mostly in this paper.

It is also not clear from the abstract and not really explained anywhere in the paper that the authors are not assessing yield per tree of the current season, but really average mango production per tree, calibrated with mango production measurements of the year 2017.

In the discussion the authors should discuss how applicable their model, which was based on the production and climate conditions of 2017, is likely to be applicable to other years as well as to author orchards in other areas.

Specific Comments

·       L 38: Add “mango orchard” or “mango yield” to the list of keywords

·       L 38: replace “incomes, food security” with “income and food security”

·       L 41: Add a sentence explaining how many different mango cultivars are grown in Senegal and how different they are.

·       L 43: How long prior to harvest? 4 weeks? 2 months? More? What is needed/required ideally?

·       L 59: Replace “tool” with “tools”

·       L 65: Replace “Mango trees is” with “Mango tree are”

·       L 87:  “the LiDAR sensor”

·       L 88: “such tools”

·       L 89: “high accuracies”

·       L 90: “(i.e. orchards with”

·       L 96: “forest stands”

·       L 105: “map land uses”

·       L 105: “delineate”

·       L 109: “olive orchards”

·       L 111: “95% accuracy”

·       L 112: “tree structure measurements”

·       L 116: replace “never” with “rarely” or add “to our knowledge”. Somebody might have tried it and you just didn’t find the paper or report and could easily be proven wrong

·       L 122: You mean “Land cover maps” rather than use maps”

·       L 144: “proximity to the Atlantic Ocean”

·       L 144: “temperatures ranging from”

·       L 144: Why are you providing only the temperatures for a single year (2017)? In addition please also provide average temperatures for the last 10 to 30 years (long-term average). Values from a single year are can be exceptional and are of limited use.

·       L 149: Please provide a list of coordinates for all 15 orchards

·       L 157: “from orchards with” and “to orchards with”

·       L 158: “by the grower”

·       L 166: “referred to as”

·       L 174: “The estimated fruit number”

·       L 178: You mean “Orchard land cover”. Land cover is the actual stuff that’s on the ground, for example tree or bare ground, while land use is the usage type of the land, for example use as an orchard.

·       L 179: “were key steps”

·       L 202: Please explain how the DTM was derived. I assume by classifying the point cloud into ground and non-ground and interpolation between the ground points. Please elaborate.

·       L 210: “a four band multi-layer”

·       L 215: “land cover”

·       L 216: “The segmentation scale parameter”

·       L 217: “the final size”

·       L 233: “by the segmentations”

·       L 245: Which bands were used to calculate the NDI?

·       L 255: “For the level 1 classification”

·       L 258: “For the level 2 classification”

·       L 272: “The confusion matrix” or “Confusion matrices”

·       L 280: “tree crowns”

·       L 284: “mango tree object”

·       L284: “computed for each”

·       L285: “land cover map”

·       L286: The crown volume can’t be an area. Please rephrase the definition.

·       L 297: “mango trees display

·       L 303: “in the field”

·       L 319: “cultivars”

·       L 322: “in the field”

·       L 336: The values of the parameters a, b, c, and d should be provided, so that the reader can apply the same model with the same parameters to other areas.

·       L 349: “land cover maps”

·       L 355: “Land cover mapping”

·       L 357: “in the number of classes”

·       L 369: I wouldn’t call the UAV-derived tree height “estimated” and the terrestrial measurement “measured”. Both are equally measurements or estimates of the same parameter. Consider calling them Terrestrial measured height and Aerial measured height or UAV measured height.

·       L 380 / Figure 4: Yield is defined as production per unit land area and the unit of yield cannot be kg. Production has the unit kg, yield would be kg/ha or kg/tree. Please correct.

·       Line 380/Figure 4: “Actual yield” is not really actual yield, but also just an estimation of the total yield per tree (or production). I would call it “Terrestrial measured yield per tree” or “Terrestrial measured production” or something similar.

·       L 429: “land cover maps”

·       L 432: Provide more information of how the Simpson index was calculated. Diversity of land cover classes? Of tree species? Of mango cultivars?

·       L 433: What yield are you talking about here? Production per tree? Production per hectare? Production per orchard? Please define yield correctly and use consistently throughout the paper.

·       L 448: Please provide the number of hectares of an average-sized orchard.

·       L 475: “land cover”

·       L 498: “other studies”

·       L 511: “to fuse objects”

·       L 550: “of each tree”

·       L 598: “land cover”

References

·       L 647: Make title of Normand et al. lower case

Author Response

Response to Reviewer 3 Comments

The paper is very well written and I have only few main comments on the content. There are numerous English language mistakes though throughout the entire manuscript which are distracting from the content. I listed most of them in the detailed comments section below.

Point 1: The authors seem be unaware about the difference between land cover and land use. Land cover are the actual features on the ground (e.g. grass, trees, buildings), while land use is the usage class of the land. For example, an entire orchard would be a single land use class “orchard”, comprised of several land cover classes “trees”, “grass”, “bare ground”, etc. Please replace all occurrences of “land use” with “land cover”.

Response 1: We are grateful to Reviewer 3 for this instructive comment. Indeed, a confusion was made between land use and land cover as the studied landscape comprised different usages of lands between natural area and the different type of agricultural system. The mistake was corrected throughout the entire manuscript (including title).

Modified text throughout.

Point 2: The use of the term “yield” should be clarified and defined. The standard definition of yield is production per unit of land area. It could also be defined as production per tree, as it is used mostly in this paper.

Response 2: We agreed with the reviewer comment. So we clarify in the main text the definitions used for “orchard yield” as production per unit of area, and for “tree production” as kg of fruit per tree. We then used both terms throughout the entire manuscript.

Point 3: It is also not clear from the abstract and not really explained anywhere in the paper that the authors are not assessing yield per tree of the current season, but really average mango production per tree, calibrated with mango production measurements of the year 2017.

Response 3: As suggested by the reviewer, we clarify this point by specifying that both calibration and estimation of tree production has been performed the same year 2017.

Modified L26: “In 2017,…”

Modified L253-254: “During June 2017, individual tree production from one month to two weeks before harvest was measured on the 150 calibration trees by the mean of RGB ground image analysis”

Modified L273: “During 2017, the 15 orchards were overflown using a DJI Mavic Pro quadricopter”

Point 4: In the discussion the authors should discuss how applicable their model, which was based on the production and climate conditions of 2017, is likely to be applicable to other years as well as to author orchards in other areas.

Response 4: Additionally, to discuss the load index issues in paragraph 4.4 (index encompassing climate, site, practice effects), as suggested we added a sentence discussing the robustness of the production models in different conditions (in time and space). 

Modified L944-945: “Further studies should test the robustness of the tree production models under various conditions (different years and other study areas).”

Specific Comments

Point 5: L 38: Add “mango orchard” or “mango yield” to the list of keywords

Response 5: As suggested, we added “mango orchard” in the keywords.

Point 6: L 38: replace “incomes, food security” with “income and food security”

Response 6: As suggested, we modified the text.

Point 7:  L 41: Add a sentence explaining how many different mango cultivars are grown in Senegal and how different they are.

Response 7: We added a sentence and a reference addressing this point in the introduction.

Modified L40-41: “In this region, more than 20 polyembryonic and monoembryonic cultivars were featured by Rey et al.”

Point 8:  L 43: How long prior to harvest? 4 weeks? 2 months? More? What is needed/required ideally?

Response 8: As suggested, we added the information in the main text.

Modified L54: “Mango yield, defined here as the production per unit of area (e.g. t.ha-1), estimation as early as possible before harvest is key to inform growers”

Modified L61-62: “Depending on the growers, this estimation is proceeded between two months and two weeks before harvest.”

Point 9: L 59: Replace “tool” with “tools”

Response 9: As suggested we modified the text.

Point 10:  L 65: Replace “Mango trees is” with “Mango tree are”

Response 10: As suggested we modified the text.

Point 11:  L 87:  “the LiDAR sensor”

Response 11: As suggested we modified the text.

Point 12: L 88: “such tools”

Response 12: As suggested we modified the text.

Point 13:   L 89: “high accuracies”

Response 13: As suggested we modified the text.

Point 14:  L 90: “(i.e. orchards with”

Response 14: As suggested we modified the text.

Point 15:  L 96: “forest stands”

Response 15: As suggested we modified the text.

Point 16: L 105: “map land uses”

Response 16: As suggested we modified the text.

Point 17: L 105: “delineate”

Response 17: As suggested we modified the text.

Point 18:  L 109: “olive orchards”

Response 18: As suggested we modified the text.

Point 19: L 111: “95% accuracy”

Response 19: As suggested we modified the text.

Point 20:  L 112: “tree structure measurements”

Response 20: As suggested we modified the text.

Point 21:   L 116: replace “never” with “rarely” or add “to our knowledge”. Somebody might have tried it and you just didn’t find the paper or report and could easily be proven wrong

Response 21: As suggested we modified the text.

Point 22:   L 122: You mean “Land cover maps” rather than use maps”

Response 22: See Point 1 Reviewer #3

Point 23:    L 144: “proximity to the Atlantic Ocean”

Response 23: As suggested we modified the text.

Point 24:   L 144: “temperatures ranging from”

Response 24: As suggested we modified the text.

Point 25:   L 144: Why are you providing only the temperatures for a single year (2017)? In addition please also provide average temperatures for the last 10 to 30 years (long-term average). Values from a single year are can be exceptional and are of limited use.

Response 25: We agreed that providing long-term daily mean temperature gives a better overview of the actual climate. We changed the text accordingly.

Point 26:    L 149: Please provide a list of coordinates for all 15 orchards

Response 26: Added text in Table 1.

Point 27:   L 157: “from orchards with” and “to orchards with”

Response 27: As suggested we modified the text.

Point 28:   L 158: “by the grower”

Response 28: We change the phrase to: “the actual yield of each orchard was asked to the growers” as we interviewed the grower to know the actual yield of the year.

Point 29:   L 166: “referred to as”

Response 29: As suggested we modified the text.

Point 30:    L 174: “The estimated fruit number”

Response 30: As suggested we modified the text.

Point 31:   L 178: You mean “Orchard land cover”. Land cover is the actual stuff that’s on the ground, for example tree or bare ground, while land use is the usage type of the land, for example use as an orchard.

Response 31: See Point 1 Reviewer #3.

Point 32:   L 179: “were key steps”

Response 32: As suggested we modified the text.

Point 33:   L 202: Please explain how the DTM was derived. I assume by classifying the point cloud into ground and non-ground and interpolation between the ground points. Please elaborate.

Response 33: The DTM was automatically generated by the Pix4D software that filtered out the non-ground points in the point cloud before a smoothing step. We added this precision in the Materials and Methods section.

Modified L304-305: “DTM was automatically generated by the software that filtered out non-ground points from the point cloud before a smoothing operation”

Point 34:    L 210: “a four band multi-layer”

Response 34: As suggested we modified the text.

Point 35:    L 215: “land cover”

Response 35: As suggested we modified the text.

Point 36:    L 216: “The segmentation scale parameter”

Response 36: As suggested we modified the text.

Point 37:    L 217: “the final size”

Response 37: As suggested we modified the text.

Point 38:    L 233: “by the segmentations”

Response 38: As suggested we modified the text.

Point 39:    L 245: Which bands were used to calculate the NDI?

Response 39: We added the information on the bands used.

Modified L357: “NDI, computed on R and G bands”

Point 40:    L 255: “For the level 1 classification”

Response 40: As suggested we modified the text.

Point 41:    L 258: “For the level 2 classification”

Response 41: As suggested we modified the text.

Point 42:    L 272: “The confusion matrix” or “Confusion matrices”

Response 42: As suggested we modified the text.

Point 43:     L 280: “tree crowns”

Response 43: As suggested we modified the text.

Point 44:    L 284: “mango tree object”

Response 44: As suggested we modified the text.

Point 45:    L284: “computed for each”

Response 45: As suggested we modified the text.

Point 46:    L285: “land cover map”

Response 46: As suggested we modified the text.

Point 47:    L286: The crown volume can’t be an area. Please rephrase the definition.

Response 47: We removed the definition in brackets as the volume is better define in the sentence.

Point 48:    L 297: “mango trees display

Response 48: As suggested we modified the text.

Point 49:    L 303: “in the field”

Response 49: As suggested we modified the text.

Point 50:    L 319: “cultivars”

Response 50: As suggested we modified the text.

Point 51:    L 322: “in the field”

Response 51: As suggested we modified the text.

Point 52:    L 336: The values of the parameters a, b, c, and d should be provided, so that the reader can apply the same model with the same parameters to other areas.

Response 52: These parameters correspond to the weighted values of the different load index categories (in %) assessed in each orchard using the cross transects. These values, that depend on each orchard, can be found in Fig. 5c.

Point 53:    L 349: “land cover maps”

Response 53: As suggested we modified the text.

Point 54:    L 355: “Land cover mapping”

Response 54: As suggested we modified the text.

Point 55:    L 357: “in the number of classes”

Response 55: As suggested we modified the text.

Point 56:    L 369: I wouldn’t call the UAV-derived tree height “estimated” and the terrestrial measurement “measured”. Both are equally measurements or estimates of the same parameter. Consider calling them Terrestrial measured height and Aerial measured height or UAV measured height.

Response 56: Based on the state-of the art literature on drone photogrammetry, we believe that the correct terms are “estimated” for UAV and “measured” for terrestrial. See Diaz Varela et al. 2015, Birdal et al. 2017, Torres-Sanchez et al. 2015.

Point 57:    L 380 / Figure 4: Yield is defined as production per unit land area and the unit of yield cannot be kg. Production has the unit kg, yield would be kg/ha or kg/tree. Please correct.

Response 57: As suggested we modified the axes and caption of the graphs.

Point 58:  Line 380/Figure 4: “Actual yield” is not really actual yield, but also just an estimation of the total yield per tree (or production). I would call it “Terrestrial measured yield per tree” or “Terrestrial measured production” or something similar.

Response 58: We change “actual yield” for “ground-measured tree production”

Point 59:  L 429: “land cover maps”

Response 59: As suggested we modified the text.

Point 60:  L 432: Provide more information of how the Simpson index was calculated. Diversity of land cover classes? Of tree species? Of mango cultivars?

Response 60: The Simpson index was computed on object classes as the probability that two pixels selected at random would be of different object classes. The sentence has been modified.

Modified L528-529: “Simpson diversity index which represent the probability that two pixels selected at random would be of different object classes.”

Point 61: L 433: What yield are you talking about here? Production per tree? Production per hectare? Production per orchard? Please define yield correctly and use consistently throughout the paper.

Response 61: See response to point 2 of reviewer #3.

Point 62: L 448: Please provide the number of hectares of an average-sized orchard.

Response 62: As suggested the information was added in the text.

Modified text L813: “(around 2 ha)”

Point 63: L 475: “land cover”

Response 63: As suggested we modified the text.

Point 64:  L 498: “other studies”

Response 64: As suggested we modified the text.

Point 65: L 511: “to fuse objects”

Response 65: As suggested we modified the text.

Point 66: L 550: “of each tree”

Response 66: As suggested we modified the text.

Point 67: L 598: “land cover”

Response 67: As suggested we modified the text.

Point 68: L 647: Make title of Normand et al. lower case

Response 68: As suggested we modified the reference title.

Author Response File: Author Response.docx