Seeing the Forest for the Trees: Mapping Cover and Counting Trees from Aerial Images of a Mangrove Forest Using Artificial Intelligence

Round 1

Reviewer 1 Report

Dear authors,

After carefully reviewing your article “Seeing the forest for the trees: mapping cover and counting trees from aerial images of a mangrove forest using artificial intelligence” I may say that the proposed methodology is an interesting tool for processing and interpreting very high-resolution images obtained using commercial grade UAV over dense mangrove forest. The drawbacks of the study consist of the local level application with no easy upscaling of the results to other areas having in mind the variability of illumination conditions, vegetation phenological state, etc.. Another issue that authors have to explain is regarding spatial accuracy, keeping in mind the use of commercial grade UAS that relies on absolute GPS positioning meaning that is around metric accuracy depending on the local condition.

 

However, the clear description of the methodology, as well as the results obtained, make this work a useful mangrove monitoring tool for protected area managers. That is why I am recommending this article be accepted for publication after minor revision.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Mangrove is an important ecosystem in the intertidal zone and the inventory in it is always time-cost and difficult due to the mud and stem structure. This study provided a new method to extract the individual mangrove tree and derive it’s attribute, such as the canopy size and tree height through combining deep learning and UAV imagery. The manuscript is well organized, and I think it reach the journal’s quality after a major revision. Following are my specific comments:

 

1)     Line 11, what’s meaning of AP ?

2)     Line 18 “ the digital elevation and terrain models”, the digital elevation model is the digital terrain model, are you meaning the “digital elevation and surface models”

3)     Line 42 “plots plots ”

4)     In the introduction, authors fails to tell readers the reason to use instance segmentation and what is the contribution of this study in this field comparing the previous studies?

5)     Line 275 the name of some variables are mistake, such as “temp_tile”

6)     In section 2.6, please keep consistence while using DTM or DEM? The initial digital model generated by UAV SfM software is digital surface model. Moreover, the description in the Figure 1 is not appropriate.

7)     Table1 please provide spatial resolution and the UAV flight parameters in capturing images.

8)     In this study, both instance segmentation and semantic segmentation are used, but the authors do not provide how many samples use to train and validate. A comparison of other approaches needs to be added to the discussion. Moreover, what’s the benefit of combining instance segmentation and semantic segmentation in this study?

 

9)     In this study, tree height is validated by comparing two different DTM generating approaches. It cannot tell readers the accuracy of this method to extract tree height. The difference in tree heights from these two methods come from the interpolate error of DTM which is caused by different ground point. I think it should add some ground observation data of tree height to validate the result.

Minor editing of English language required

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have well addressed my concerns in the previous version and this version has largely improved.  I also understand that the ground truth of mangrove tree height is hard to get and the authors declare this in the discussion. I think the paper can be accepted after minor revision. Following are my specific comments.

1. Line 328-330 What interpolation method is used to create DTM? The authors said using Geospatial Data Abstraction Library’s (GDAL) fill_no_data34 function to interpolate and smooth-out the DTM into a DEM. but we did not know what is the exacted method, bilinear or IDW or other methods.

2. Line 350-352 the description of extracting tree height can be more compact

3. line 376 what does the area height distributions meaning ?

4.  line 399 please provide the score of mud

5. line 446 what is the unit for the -0.72 and -0.18 ?

Minor editing of English language required.

Line 229 there are two "has"

Line 251-252: this sentence needs to be revised since it is a little difficult to understand.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf