UAV-Based Mapping of Banana Land Area for Village-Level Decision-Support in Rwanda

Round 1

Reviewer 1 Report

This is, overall, a well written and interesting manuscript that directly compares various classification schemes using UAV data for agricultural mapping (bananas in this case). The authors clearly are aiming this at end users who might want to use these methods to improve agricultural mapping, and they do a good job of explaining why the reader should care about the relative accuracy of the methods, how to apply it, and (most importantly for this paper) how the data can be used to inform decision making.

Having said that, I have some overall comments and some more specific line items that need to be addressed before considering publication. First, the intro is well written but somewhat unbalanced compared to the rest of the paper (I think it could be cut down some). For example the intro discusses smallholder farms in general for Africa, and then specifically for Rwanda--some of this information seems a bit redundant (statements such as smallholder farms being important in Rwanda follow from a more general statement about smallholder farms dominating agricultural crops in Africa--so unless Rwanda is anomalous, it seems kind of obvious that it would be important after reading the more general info about Africa). 

It also feels unbalanced because after the fairly extensive introduction, the actual description of the remote sensing methods is rather short and not very detailed. 

Author Response

Reviewer #1

Thank you for the overall positive impression of the paper and we’d like to provide response to each comment, with action taken.

First, the intro is well written but somewhat unbalanced compared to the rest of the paper (I think it could be cut down some). For example the intro discusses smallholder farms in general for Africa, and then specifically for Rwanda--some of this information seems a bit redundant (statements such as smallholder farms being important in Rwanda follow from a more general statement about smallholder farms dominating agricultural crops in Africa--so unless Rwanda is anomalous, it seems kind of obvious that it would be important after reading the more general info about Africa).

We’re happy to receive the positive comments regarding the intro section and we have revised it a bit as well. Yet, we opine that it’s relevant to set the scene relative to the Rwandan national context since the analysis was implemented at village level (4 villages), with focus on supporting national baseline for banana cropland. Also, we think that readers may be interested in the local importance of this research to farmers in-country, rather than general global view.

It also feels unbalanced because after the fairly extensive introduction, the actual description of the remote sensing methods is rather short and not very detailed.

We have attempted to balance out the content by fleshing out the explanation under the methods section. We are also cautious to avoid lengthiness of the methods section.

Author Response File: Author Response.docx

Reviewer 2 Report

In the study, authors used RF model to map the banana plantation based on UAV-based remote sensing in four villages in Rwanda. The whole paper was generally clearly written and the results were principally sound. However, to have it published on the Remote Sensing, substantial improvements are still needed. Below please find some questions and suggestions that are either minor or major.

Main comments:

1. Should Table 2 be Table 1? Is the total number of samples of each category in Table 2 not the sum of the samples of the three villages (Karambo, Rubira and Rubira)?
2. It is suggested that the geographic coordinate frame in Figure 3 should be standardized and beautiful.
3. The author collected 605 sample points in line 153, but the sample points in Table 2 are 750. Please confirm.
4. What conditions are level 2 based on in Table 2?
5. Does the author use RF model based on Level 3 to classify in Section 3.4?
6. Lines 183-184, the authors use UAV images for classification and manually selects sample points from the UAV images. Is this classification appropriate?

Author Response

Reviewer #2

Thank you for the general positive note

Main comments:

1. Should Table 2 be Table 1? Is the total number of samples of each category in Table 2 not the sum of the samples of the three villages (Karambo, Rubira and Rubira)?

Thanks for flagging this. The Table numbering has been updated for correctness; There was an erroneous omission of the 4^th row for Murambo village and this has been corrected as well

2. It is suggested that the geographic coordinate frame in Figure 3should be standardized and beautiful.

Thank you, and Done

3. The author collected 605 sample points in line 153, but the sample points in Table 2 are 750. Please confirm.

Good note - Now revised for clarity; We collected 605 banana location points, and other landcover class points made the rest, adding up to 750. additional data points were collected to form the “1^st level” of data richness which represents minimum “ground-acquired” data. In total, 2306 points were pooled from all classes (i.e. banana + built-up + bareland + other vegetation + water) after utilizing the UAV imageries to digitize more reference points for the classification. 

4.What conditions are level 2 based on in Table 2?

Our focus was on the volume/ robustness of ground datapoints, so level 2 indicate a scenario where the initial (ground-acquired) datapoints were doubled by digitizing additional 750 points from the high-resn UAV imageries

5. Does the author use RF model based on Level 3 to classify in Section 3.4?

Yes, RF model is used based on the level 3 which has all data samples.

Updated. Lines 286

6. Lines 183-184, the authors use UAV images for classification and manually selects sample points from the UAV images. Is this classification appropriate?

Thanks for asking. We consider this approach appropriate for the intended purpose. The points that were sampled from the UAV imageries were used to enrich the available ground survey sample points. Since the UAV imageries were acquired with real-time kinematic (RTK) GPS which has centimeter-level accuracy, the selection of points within each vegetation/land-cover class is considered as accurate as the ground-level survey which were conducted with hand-held GPS unit.

 

 

Author Response File: Author Response.docx

Reviewer 3 Report

Dear authors,

I have with great interest read and commented your manuscript and the following overall and more specific comment:

I would like you to further expalin line 19-20, why is it critical?

line 28 - 'areal'?

Line 28-29: further explain 'for extension-delivery efforts towards banana farmers' what kind of extension delivery do you mean?

Introduction

Line 39, Error!..

line 50 - areal

Line 68 and 69: It is critical and extremely important to cite others work correctly in international articles. Your reference 15 is not the right one for NDVI and there is no reference for GNDVI and it is not your reference#16

Line 119-120: dividing the dataset into training and testset is also important. You use random which is fine but you could also have used a more stratified method eg. using one of the villages as testset.

Material and methods:

Line 170-171. sometimes free spaces before and after + and - and sometimes not.

Line181- 184: Level 2 is not mentioned

 

Discussion:

Line 351: explain why it can support the mitigation of BXW risk.

 

Author Response

I would like you to further explain line 19-20, why is it critical?

This verbiage was used in the abstract to stress the importance of understanding the distribution/extent of banana crop lands; however, based on the reviewer’s note, we have change the wording to “important”

line 28 - 'areal'?

This is a standard use of the word “areal” to describe area covered – see https://www.merriam-webster.com/dictionary/areal

Line 28-29: further explain 'for extension-delivery efforts towards banana farmers' what kind of extension delivery do you mean?

Additional explanation regarding “extension delivery efforts” has been added to the conclusion, however, we’re constrained by the brevity of space on the abstract.

Introduction

Line 39, Error!..

Corrected

line 50 – areal

Please see note above

Line 68 and 69: It is critical and extremely important to cite others work correctly in international articles. Your reference 15 is not the right one for NDVI and there is no reference for GNDVI and it is not your reference#16

Thank you for flagging the inadvertent misplacement of references; the cited references have been duly updated. We confirmed that reference 16 is correct for GNDVI

Line 119-120: dividing the dataset into training and test set is also important. You use random which is fine but you could also have used a more stratified method e.g. using one of the villages as test set.

We thank the reviewer for this note. The random split of the data point was implemented as a generally acceptable approach for partitioning of training and test datasets. The reviewer's suggestion is noted for anticipated follow-up analyses for future publication

Material and methods:

Line 170-171. sometimes free spaces before and after + and – and sometimes not.

The irregular spacing has been corrected

Line181- 184: Level 2 is not mentioned

This was an omission error – this is now mentioned

Discussion:

Line 351: explain why it can support the mitigation of BXW risk.

Additional texts have been added to explain this point

Author Response File: Author Response.docx

Reviewer 4 Report

The article presents an application of UAV multi-spectral in mapping using machine learning approaches. The article needs to be improved in format and content. Following I add my major and minor comments: 

Major comments:

•  Expose better what are the goal or goals of the article. 
• The format of the article needs to be improved.
• One flight is enough? Why not temporal series? Why not use or combine it with Sentinel-2 (for example)? Discuss it is needed. 
• Define better the area of interest: what is the surface flighted?

Minor comments: 
Line 38: “The world’s population is gradually 38 increasing and projected to reach 9 billion by the year 2050” needs to be referenced.
Error in line 39: “Error! Reference source not found. [2]”
Line 52: could you provide examples of "conventional methods"?
Line 62: a space is missing after “[10].”
Line 160: this table is Table 1, not Table 2. 
Line 247: Explain better the description of this figure. What values are used to build the chart?

Decision: 
Reconsider after major revision (control missing in some experiments)

Author Response

Reviewer #4

Major comments:

Expose better what are the goal or goals of the article.

Noted. As suggested, we have revised the goal for better clarity in the article

The format of the article needs to be improved.

Formatting has been revised, and will be further revised prior to final publishing

One flight is enough? Why not temporal series? Why not use or combine it with Sentinel-2 (for example)? Discuss it is needed.

We implemented 4 flight missions, and over 20 flights in total, to acquire imageries that cover each entire village. We did not consider temporal series because our objective was to set the baseline. We would like to clarify that temporal series will be relevant for future assessment of temporal changes, but not for our current objective for this article. More texts are now added to explain this in the article.

Define better the area of interest: what is the surface flighted?

Well noted. Additional texts have been added to further describe the area of interest and to indicate the area of land covered by flights

Minor comments:

Line 38: “The world’s population is gradually 38 increasing and projected to reach 9 billion by the year 2050” needs to be referenced.

Reference included and updated

Error in line 39: “Error! Reference source not found. [2]”

Corrected

Line 52: could you provide examples of "conventional methods"?

Yes; this is now included in-text

Line 62: a space is missing after “[10].”

Space added

Line 160: this table is Table 1, not Table 2.

Inadvertent error in Table numbering has been corrected

Line 247: Explain better the description of this figure. What values are used to build the chart?

Thank you. Additional texts are now added to explain the chart (now lines 243-249) and the title. The calculated indices values, based on band reflectance values, were used to plot the charts

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The revised manuscript meet the requirements of publication

Reviewer 4 Report

The response of the authors to my comments is enough, and the manuscript has been improved. 

In my opinion, this paper could be published.